diff --git a/packages/gcc/10.5.0/0024-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch b/packages/gcc/10.5.0/0024-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch new file mode 100644 index 00000000..25049ef3 --- /dev/null +++ b/packages/gcc/10.5.0/0024-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch @@ -0,0 +1,349 @@ +From 153060b97c667bbd06caad9c3fc1f9f6fe7266df Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 16:43:41 +0100 +Subject: [PATCH 24/42] aarch64: Use local frame vars in shrink-wrapping code + +aarch64_layout_frame uses a shorthand for referring to +cfun->machine->frame: + + aarch64_frame &frame = cfun->machine->frame; + +This patch does the same for some other heavy users of the structure. +No functional change intended. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_callee_saves): Use + a local shorthand for cfun->machine->frame. + (aarch64_restore_callee_saves, aarch64_get_separate_components): + (aarch64_process_components): Likewise. + (aarch64_allocate_and_probe_stack_space): Likewise. + (aarch64_expand_prologue, aarch64_expand_epilogue): Likewise. + (aarch64_layout_frame): Use existing shorthand for one more case. +--- + gcc/config/aarch64/aarch64.c | 115 ++++++++++++++++++----------------- + 1 file changed, 60 insertions(+), 55 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index e6bf5ae38085..368bb394105e 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7403,6 +7403,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx_insn *insn; + unsigned regno; + unsigned regno2; +@@ -7417,8 +7418,8 @@ aarch64_save_callee_saves (poly_int64 start_offset, + bool frame_related_p = aarch64_emit_cfi_for_reg_p (regno); + + if (skip_wb +- && (regno == cfun->machine->frame.wb_candidate1 +- || regno == cfun->machine->frame.wb_candidate2)) ++ && (regno == frame.wb_candidate1 ++ || regno == frame.wb_candidate2)) + continue; + + if (cfun->machine->reg_is_wrapped_separately[regno]) +@@ -7426,7 +7427,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -7439,7 +7440,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + { + gcc_assert (known_eq (start_offset, 0)); + poly_int64 fp_offset +- = cfun->machine->frame.below_hard_fp_saved_regs_size; ++ = frame.below_hard_fp_saved_regs_size; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -7461,8 +7462,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -7512,6 +7512,7 @@ static void + aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { ++ aarch64_frame &frame = cfun->machine->frame; + unsigned regno; + unsigned regno2; + poly_int64 offset; +@@ -7528,13 +7529,13 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + rtx reg, mem; + + if (skip_wb +- && (regno == cfun->machine->frame.wb_candidate1 +- || regno == cfun->machine->frame.wb_candidate2)) ++ && (regno == frame.wb_candidate1 ++ || regno == frame.wb_candidate2)) + continue; + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -7545,8 +7546,7 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -7640,6 +7640,7 @@ offset_12bit_unsigned_scaled_p (machine_mode mode, poly_int64 offset) + static sbitmap + aarch64_get_separate_components (void) + { ++ aarch64_frame &frame = cfun->machine->frame; + sbitmap components = sbitmap_alloc (LAST_SAVED_REGNUM + 1); + bitmap_clear (components); + +@@ -7656,18 +7657,18 @@ aarch64_get_separate_components (void) + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + continue; + +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + + /* If the register is saved in the first SVE save slot, we use + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection +- && maybe_ne (cfun->machine->frame.below_hard_fp_saved_regs_size, 0) ++ && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) + && known_eq (offset, 0)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -7686,11 +7687,11 @@ aarch64_get_separate_components (void) + /* If the spare predicate register used by big-endian SVE code + is call-preserved, it must be saved in the main prologue + before any saves that use it. */ +- if (cfun->machine->frame.spare_pred_reg != INVALID_REGNUM) +- bitmap_clear_bit (components, cfun->machine->frame.spare_pred_reg); ++ if (frame.spare_pred_reg != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.spare_pred_reg); + +- unsigned reg1 = cfun->machine->frame.wb_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_candidate2; ++ unsigned reg1 = frame.wb_candidate1; ++ unsigned reg2 = frame.wb_candidate2; + /* If registers have been chosen to be stored/restored with + writeback don't interfere with them to avoid having to output explicit + stack adjustment instructions. */ +@@ -7799,6 +7800,7 @@ aarch64_get_next_set_bit (sbitmap bmp, unsigned int start) + static void + aarch64_process_components (sbitmap components, bool prologue_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx ptr_reg = gen_rtx_REG (Pmode, frame_pointer_needed + ? HARD_FRAME_POINTER_REGNUM + : STACK_POINTER_REGNUM); +@@ -7813,9 +7815,9 @@ aarch64_process_components (sbitmap components, bool prologue_p) + machine_mode mode = aarch64_reg_save_mode (regno); + + rtx reg = gen_rtx_REG (mode, regno); +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -7840,14 +7842,14 @@ aarch64_process_components (sbitmap components, bool prologue_p) + break; + } + +- poly_int64 offset2 = cfun->machine->frame.reg_offset[regno2]; ++ poly_int64 offset2 = frame.reg_offset[regno2]; + /* The next register is not of the same class or its offset is not + mergeable with the current one into a pair. */ + if (aarch64_sve_mode_p (mode) + || !satisfies_constraint_Ump (mem) + || GP_REGNUM_P (regno) != GP_REGNUM_P (regno2) + || (crtl->abi->id () == ARM_PCS_SIMD && FP_REGNUM_P (regno)) +- || maybe_ne ((offset2 - cfun->machine->frame.reg_offset[regno]), ++ || maybe_ne ((offset2 - frame.reg_offset[regno]), + GET_MODE_SIZE (mode))) + { + insn = emit_insn (set); +@@ -7869,7 +7871,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset2 -= frame.below_hard_fp_saved_regs_size; + else + offset2 += crtl->outgoing_args_size; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); +@@ -7964,6 +7966,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + bool frame_related_p, + bool final_adjustment_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; +@@ -7984,25 +7987,25 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + register as a probe. We can't assume that LR was saved at position 0 + though, so treat any space below it as unprobed. */ + if (final_adjustment_p +- && known_eq (cfun->machine->frame.below_hard_fp_saved_regs_size, 0)) ++ && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = cfun->machine->frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else + gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); + } + +- poly_int64 frame_size = cfun->machine->frame.frame_size; ++ poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ + gcc_assert (min_probe_threshold > 0); + + if (flag_stack_clash_protection && !final_adjustment_p) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; + + if (known_eq (frame_size, 0)) + { +@@ -8291,17 +8294,18 @@ aarch64_epilogue_uses (int regno) + void + aarch64_expand_prologue (void) + { +- poly_int64 frame_size = cfun->machine->frame.frame_size; +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 frame_size = frame.frame_size; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_candidate2; +- bool emit_frame_chain = cfun->machine->frame.emit_frame_chain; ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_candidate1; ++ unsigned reg2 = frame.wb_candidate2; ++ bool emit_frame_chain = frame.emit_frame_chain; + rtx_insn *insn; + + if (flag_stack_clash_protection && known_eq (callee_adjust, 0)) +@@ -8367,7 +8371,7 @@ aarch64_expand_prologue (void) + + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - cfun->machine->frame.hard_fp_offset); ++ - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + + /* The offset of the bottom of the save area from the current SP. */ +@@ -8470,15 +8474,16 @@ aarch64_use_return_insn_p (void) + void + aarch64_expand_epilogue (bool for_sibcall) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_candidate2; ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_candidate1; ++ unsigned reg2 = frame.wb_candidate2; + rtx cfi_ops = NULL; + rtx_insn *insn; + /* A stack clash protection prologue may not have left EP0_REGNUM or +@@ -8511,7 +8516,7 @@ aarch64_expand_epilogue (bool for_sibcall) + /* We need to add memory barrier to prevent read from deallocated stack. */ + bool need_barrier_p + = maybe_ne (get_frame_size () +- + cfun->machine->frame.saved_varargs_size, 0); ++ + frame.saved_varargs_size, 0); + + /* Emit a barrier to prevent loads from a deallocated stack. */ + if (maybe_gt (final_adjust, crtl->outgoing_args_size) +@@ -11170,24 +11175,24 @@ aarch64_can_eliminate (const int from ATTRIBUTE_UNUSED, const int to) + poly_int64 + aarch64_initial_elimination_offset (unsigned from, unsigned to) + { ++ aarch64_frame &frame = cfun->machine->frame; ++ + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ return frame.hard_fp_offset - frame.locals_offset; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return frame.frame_size - frame.locals_offset; + } + +- return cfun->machine->frame.frame_size; ++ return frame.frame_size; + } + + +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0025-aarch64-Avoid-a-use-of-callee_offset.patch b/packages/gcc/10.5.0/0025-aarch64-Avoid-a-use-of-callee_offset.patch new file mode 100644 index 00000000..398f7c94 --- /dev/null +++ b/packages/gcc/10.5.0/0025-aarch64-Avoid-a-use-of-callee_offset.patch @@ -0,0 +1,73 @@ +From 901269e14884a86c532b99fb41d5f46b519344f0 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:12:21 +0100 +Subject: [PATCH 25/42] aarch64: Avoid a use of callee_offset + +When we emit the frame chain, i.e. when we reach Here in this statement +of aarch64_expand_prologue: + + if (emit_frame_chain) + { + // Here + ... + } + +the stack is in one of two states: + +- We've allocated up to the frame chain, but no more. + +- We've allocated the whole frame, and the frame chain is within easy + reach of the new SP. + +The offset of the frame chain from the current SP is available +in aarch64_frame as callee_offset. It is also available as the +chain_offset local variable, where the latter is calculated from other +data. (However, chain_offset is not always equal to callee_offset when +!emit_frame_chain, so chain_offset isn't redundant.) + +In c600df9a4060da3c6121ff4d0b93f179eafd69d1 I switched to using +chain_offset for the initialisation of the hard frame pointer: + + aarch64_add_offset (Pmode, hard_frame_pointer_rtx, +- stack_pointer_rtx, callee_offset, ++ stack_pointer_rtx, chain_offset, + tmp1_rtx, tmp0_rtx, frame_pointer_needed); + +But the later REG_CFA_ADJUST_CFA handling still used callee_offset. + +I think the difference is harmless, but it's more logical for the +CFA note to be in sync, and it's more convenient for later patches +if it uses chain_offset. + +gcc/ + * config/aarch64/aarch64.c (aarch64_expand_prologue): Use + chain_offset rather than callee_offset. +--- + gcc/config/aarch64/aarch64.c | 4 +--- + 1 file changed, 1 insertion(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 368bb394105e..2aaf187bcc3f 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8299,7 +8299,6 @@ aarch64_expand_prologue (void) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size + = frame.below_hard_fp_saved_regs_size; +@@ -8408,8 +8407,7 @@ aarch64_expand_prologue (void) + implicit. */ + if (!find_reg_note (insn, REG_CFA_ADJUST_CFA, NULL_RTX)) + { +- rtx src = plus_constant (Pmode, stack_pointer_rtx, +- callee_offset); ++ rtx src = plus_constant (Pmode, stack_pointer_rtx, chain_offset); + add_reg_note (insn, REG_CFA_ADJUST_CFA, + gen_rtx_SET (hard_frame_pointer_rtx, src)); + } +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0026-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/10.5.0/0026-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..30142714 --- /dev/null +++ b/packages/gcc/10.5.0/0026-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,49 @@ +From c8ffe35ca907e0042524a5369158fcb5631d5108 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 17:00:51 +0100 +Subject: [PATCH 26/42] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +And if we allocate the frame in one go, it should be allocated +as the initial_adjust rather than the final_adjust. This allows the +frame size to grow to guard_size - guard_used_by_caller before a stack +probe is needed. (A frame with no register saves is necessarily a +leaf frame.) + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.c | 8 +++++--- + 1 file changed, 5 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 2aaf187bcc3f..81537c019511 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7035,9 +7035,11 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; +- if (frame.frame_size.is_constant (&const_size) +- && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ if (known_eq (frame.saved_regs_size, 0)) ++ frame.initial_adjust = frame.frame_size; ++ else if (frame.frame_size.is_constant (&const_size) ++ && const_size < max_push_offset ++ && known_eq (frame.hard_fp_offset, const_size)) + { + /* Simple, small frame with no outgoing arguments: + +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0027-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch b/packages/gcc/10.5.0/0027-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch new file mode 100644 index 00000000..1a069d3f --- /dev/null +++ b/packages/gcc/10.5.0/0027-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch @@ -0,0 +1,233 @@ +From 781490f3e1bd3789048485981e5039300abf41cb Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 16:55:12 +0100 +Subject: [PATCH 27/42] aarch64: Add bytes_below_saved_regs to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_saved_regs): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_saved_regs instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 71 ++++++++++++++++++------------------ + gcc/config/aarch64/aarch64.h | 5 +++ + 2 files changed, 41 insertions(+), 35 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 81537c019511..6109518189fb 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6902,6 +6902,8 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + ++ frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small + offset range. These saves happen below the hard frame pointer. */ +@@ -7006,18 +7008,18 @@ aarch64_layout_frame (void) + + poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; + +- poly_int64 above_outgoing_args ++ poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + + frame.hard_fp_offset +- = above_outgoing_args - frame.below_hard_fp_saved_regs_size; ++ = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +- gcc_assert (multiple_p (crtl->outgoing_args_size, ++ gcc_assert (multiple_p (frame.bytes_below_saved_regs, + STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = above_outgoing_args + crtl->outgoing_args_size; ++ frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + + frame.locals_offset = frame.saved_varargs_size; + +@@ -7033,7 +7035,7 @@ aarch64_layout_frame (void) + else if (frame.wb_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; +@@ -7041,31 +7043,31 @@ aarch64_layout_frame (void) + && const_size < max_push_offset + && known_eq (frame.hard_fp_offset, const_size)) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + frame.callee_adjust = const_size; + } +- else if (crtl->outgoing_args_size.is_constant (&const_outgoing_args_size) ++ else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) + && frame.saved_regs_size.is_constant (&const_saved_regs_size) +- && const_outgoing_args_size + const_saved_regs_size < 512 +- /* We could handle this case even with outgoing args, provided +- that the number of args left us with valid offsets for all +- predicate and vector save slots. It's such a rare case that +- it hardly seems worth the effort though. */ +- && (!saves_below_hard_fp_p || const_outgoing_args_size == 0) ++ && const_below_saved_regs + const_saved_regs_size < 512 ++ /* We could handle this case even with data below the saved ++ registers, provided that that data left us with valid offsets ++ for all predicate and vector save slots. It's such a rare ++ case that it hardly seems worth the effort though. */ ++ && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca + && frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset)) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_saved_regs] ++ stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_outgoing_args_size; ++ frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -7075,30 +7077,29 @@ aarch64_layout_frame (void) + + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = (frame.hard_fp_offset + + frame.below_hard_fp_saved_regs_size); +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments or SVE saves, but with +- a small local area: ++ /* Frame with large area below the saved registers, or with SVE saves, ++ but with a small area above: + + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else + { +- /* Frame with large local area and outgoing arguments or SVE saves, +- using frame pointer: ++ /* General case: + + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] +@@ -7106,10 +7107,10 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.hard_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + + /* Make sure the individual adjustments add up to the full frame size. */ +@@ -7672,7 +7673,7 @@ aarch64_get_separate_components (void) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -7821,7 +7822,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -7875,7 +7876,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset2 -= frame.below_hard_fp_saved_regs_size; + else +- offset2 += crtl->outgoing_args_size; ++ offset2 += frame.bytes_below_saved_regs; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -7949,10 +7950,10 @@ aarch64_stack_clash_protection_alloca_probe_range (void) + registers. If POLY_SIZE is not large enough to require a probe this function + will only adjust the stack. When allocating the stack space + FRAME_RELATED_P is then used to indicate if the allocation is frame related. +- FINAL_ADJUSTMENT_P indicates whether we are allocating the outgoing +- arguments. If we are then we ensure that any allocation larger than the ABI +- defined buffer needs a probe so that the invariant of having a 1KB buffer is +- maintained. ++ FINAL_ADJUSTMENT_P indicates whether we are allocating the area below ++ the saved registers. If we are then we ensure that any allocation ++ larger than the ABI defined buffer needs a probe so that the ++ invariant of having a 1KB buffer is maintained. + + We emit barriers after each stack adjustment to prevent optimizations from + breaking the invariant that we never drop the stack more than a page. This +@@ -8161,7 +8162,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + /* Handle any residuals. Residuals of at least MIN_PROBE_THRESHOLD have to + be probed. This maintains the requirement that each page is probed at + least once. For initial probing we probe only if the allocation is +- more than GUARD_SIZE - buffer, and for the outgoing arguments we probe ++ more than GUARD_SIZE - buffer, and below the saved registers we probe + if the amount is larger than buffer. GUARD_SIZE - buffer + buffer == + GUARD_SIZE. This works that for any allocation that is large enough to + trigger a probe here, we'll have at least one, and if they're not large +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 9be4ec3d280e..083ffbfbc530 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -820,6 +820,11 @@ struct GTY (()) aarch64_frame + /* The size of the callee-save registers with a slot in REG_OFFSET. */ + poly_int64 saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the bottom of the register save area. ++ This value is always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_saved_regs; ++ + /* The size of the callee-save registers with a slot in REG_OFFSET that + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0028-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/10.5.0/0028-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..7aeb8b13 --- /dev/null +++ b/packages/gcc/10.5.0/0028-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,84 @@ +From a519149de34facf645064c1ea5748aeb7d72aca4 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:17:33 +0100 +Subject: [PATCH 28/42] aarch64: Add bytes_below_hard_fp to frame info + +Following on from the previous bytes_below_saved_regs patch, this one +records the number of bytes that are below the hard frame pointer. +This eventually replaces below_hard_fp_saved_regs_size. + +If a frame pointer is not needed, the epilogue adds final_adjust +to the stack pointer before restoring registers: + + aarch64_add_sp (tmp1_rtx, tmp0_rtx, final_adjust, true); + +Therefore, if the epilogue needs to restore the stack pointer from +the hard frame pointer, the directly corresponding offset is: + + -bytes_below_hard_fp + final_adjust + +i.e. go from the hard frame pointer to the bottom of the frame, +then add the same amount as if we were using the stack pointer +from the outset. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it. + (aarch64_expand_epilogue): Use it instead of + below_hard_fp_saved_regs_size. +--- + gcc/config/aarch64/aarch64.c | 6 +++--- + gcc/config/aarch64/aarch64.h | 5 +++++ + 2 files changed, 8 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 6109518189fb..2652515b3615 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6954,6 +6954,7 @@ aarch64_layout_frame (void) + of the callee save area. */ + bool saves_below_hard_fp_p = maybe_ne (offset, 0); + frame.below_hard_fp_saved_regs_size = offset; ++ frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -8481,8 +8482,7 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 final_adjust = frame.final_adjust; + poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_candidate1; + unsigned reg2 = frame.wb_candidate2; + rtx cfi_ops = NULL; +@@ -8538,7 +8538,7 @@ aarch64_expand_epilogue (bool for_sibcall) + is restored on the instruction doing the writeback. */ + aarch64_add_offset (Pmode, stack_pointer_rtx, + hard_frame_pointer_rtx, +- -callee_offset - below_hard_fp_saved_regs_size, ++ -bytes_below_hard_fp + final_adjust, + tmp1_rtx, tmp0_rtx, callee_adjust == 0); + else + /* The case where we need to re-use the register here is very rare, so +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 083ffbfbc530..a4cbc678cef3 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -829,6 +829,11 @@ struct GTY (()) aarch64_frame + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0029-aarch64-Tweak-aarch64_save-restore_callee_saves.patch b/packages/gcc/10.5.0/0029-aarch64-Tweak-aarch64_save-restore_callee_saves.patch new file mode 100644 index 00000000..52eaf798 --- /dev/null +++ b/packages/gcc/10.5.0/0029-aarch64-Tweak-aarch64_save-restore_callee_saves.patch @@ -0,0 +1,221 @@ +From 6ef2e97b16bea4d4ce2cd52de444a3a2595504d6 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:21:41 +0100 +Subject: [PATCH 29/42] aarch64: Tweak aarch64_save/restore_callee_saves + +aarch64_save_callee_saves and aarch64_restore_callee_saves took +a parameter called start_offset that gives the offset of the +bottom of the saved register area from the current stack pointer. +However, it's more convenient for later patches if we use the +bottom of the entire frame as the reference point, rather than +the bottom of the saved registers. + +Doing that removes the need for the callee_offset field. +Other than that, this is not a win on its own. It only really +makes sense in combination with the follow-on patches. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::callee_offset): Delete. + * config/aarch64/aarch64.c (aarch64_layout_frame): Remove + callee_offset handling. + (aarch64_save_callee_saves): Replace the start_offset parameter + with a bytes_below_sp parameter. + (aarch64_restore_callee_saves): Likewise. + (aarch64_expand_prologue): Update accordingly. + (aarch64_expand_epilogue): Likewise. +--- + gcc/config/aarch64/aarch64.c | 56 ++++++++++++++++++------------------ + gcc/config/aarch64/aarch64.h | 4 --- + 2 files changed, 28 insertions(+), 32 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 2652515b3615..884002e3b7b5 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7028,7 +7028,6 @@ aarch64_layout_frame (void) + frame.final_adjust = 0; + frame.callee_adjust = 0; + frame.sve_callee_adjust = 0; +- frame.callee_offset = 0; + + HOST_WIDE_INT max_push_offset = 0; + if (frame.wb_candidate2 != INVALID_REGNUM) +@@ -7068,7 +7067,6 @@ aarch64_layout_frame (void) + stp reg1, reg2, [sp, bytes_below_saved_regs] + stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -7398,12 +7396,13 @@ aarch64_add_cfa_expression (rtx_insn *insn, rtx reg, + } + + /* Emit code to save the callee-saved registers from register number START +- to LIMIT to the stack at the location starting at offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. HARD_FP_VALID_P +- is true if the hard frame pointer has been set up. */ ++ to LIMIT to the stack. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. HARD_FP_VALID_P is true if the hard ++ frame pointer has been set up. */ + + static void +-aarch64_save_callee_saves (poly_int64 start_offset, ++aarch64_save_callee_saves (poly_int64 bytes_below_sp, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { +@@ -7431,7 +7430,9 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -7442,9 +7443,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + else if (GP_REGNUM_P (regno) + && (!offset.is_constant (&const_offset) || const_offset >= 512)) + { +- gcc_assert (known_eq (start_offset, 0)); +- poly_int64 fp_offset +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 fp_offset = frame.bytes_below_hard_fp - bytes_below_sp; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -7508,12 +7507,13 @@ aarch64_save_callee_saves (poly_int64 start_offset, + } + + /* Emit code to restore the callee registers from register number START +- up to and including LIMIT. Restore from the stack offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. Write the +- appropriate REG_CFA_RESTORE notes into CFI_OPS. */ ++ up to and including LIMIT. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. Write the appropriate REG_CFA_RESTORE ++ notes into CFI_OPS. */ + + static void +-aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, ++aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { + aarch64_frame &frame = cfun->machine->frame; +@@ -7539,7 +7539,9 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -8304,8 +8306,6 @@ aarch64_expand_prologue (void) + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; + unsigned reg1 = frame.wb_candidate1; + unsigned reg2 = frame.wb_candidate2; + bool emit_frame_chain = frame.emit_frame_chain; +@@ -8377,8 +8377,8 @@ aarch64_expand_prologue (void) + - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + +- /* The offset of the bottom of the save area from the current SP. */ +- poly_int64 saved_regs_offset = chain_offset - below_hard_fp_saved_regs_size; ++ /* The offset of the current SP from the bottom of the static frame. */ ++ poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { +@@ -8386,7 +8386,7 @@ aarch64_expand_prologue (void) + { + reg1 = R29_REGNUM; + reg2 = R30_REGNUM; +- aarch64_save_callee_saves (saved_regs_offset, reg1, reg2, ++ aarch64_save_callee_saves (bytes_below_sp, reg1, reg2, + false, false); + } + else +@@ -8426,7 +8426,7 @@ aarch64_expand_prologue (void) + emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + +- aarch64_save_callee_saves (saved_regs_offset, R0_REGNUM, R30_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, R0_REGNUM, R30_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + if (maybe_ne (sve_callee_adjust, 0)) +@@ -8436,16 +8436,17 @@ aarch64_expand_prologue (void) + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, + sve_callee_adjust, + !frame_pointer_needed, false); +- saved_regs_offset += sve_callee_adjust; ++ bytes_below_sp -= sve_callee_adjust; + } +- aarch64_save_callee_saves (saved_regs_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, P0_REGNUM, P15_REGNUM, + false, emit_frame_chain); +- aarch64_save_callee_saves (saved_regs_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, V0_REGNUM, V31_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + + /* We may need to probe the final adjustment if it is larger than the guard + that is assumed by the called. */ ++ gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); + } +@@ -8480,7 +8481,6 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_candidate1; +@@ -8548,13 +8548,13 @@ aarch64_expand_epilogue (bool for_sibcall) + + /* Restore the vector registers before the predicate registers, + so that we can use P4 as a temporary for big-endian SVE frames. */ +- aarch64_restore_callee_saves (callee_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, V0_REGNUM, V31_REGNUM, + callee_adjust != 0, &cfi_ops); +- aarch64_restore_callee_saves (callee_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, P0_REGNUM, P15_REGNUM, + false, &cfi_ops); + if (maybe_ne (sve_callee_adjust, 0)) + aarch64_add_sp (NULL_RTX, NULL_RTX, sve_callee_adjust, true); +- aarch64_restore_callee_saves (callee_offset - sve_callee_adjust, ++ aarch64_restore_callee_saves (final_adjust + sve_callee_adjust, + R0_REGNUM, R30_REGNUM, + callee_adjust != 0, &cfi_ops); + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index a4cbc678cef3..405b7545913d 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -856,10 +856,6 @@ struct GTY (()) aarch64_frame + It is zero when no push is used. */ + HOST_WIDE_INT callee_adjust; + +- /* The offset from SP to the callee-save registers after initial_adjust. +- It may be non-zero if no push is used (ie. callee_adjust == 0). */ +- poly_int64 callee_offset; +- + /* The size of the stack adjustment before saving or after restoring + SVE registers. */ + poly_int64 sve_callee_adjust; +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0030-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch b/packages/gcc/10.5.0/0030-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch new file mode 100644 index 00000000..cc451bf6 --- /dev/null +++ b/packages/gcc/10.5.0/0030-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch @@ -0,0 +1,45 @@ +From 176001612860c2b04133c1de26eb673860800865 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:24:27 +0100 +Subject: [PATCH 30/42] aarch64: Only calculate chain_offset if there is a + chain + +After previous patches, it is no longer necessary to calculate +a chain_offset in cases where there is no chain record. + +gcc/ + * config/aarch64/aarch64.c (aarch64_expand_prologue): Move the + calculation of chain_offset into the emit_frame_chain block. +--- + gcc/config/aarch64/aarch64.c | 10 +++++----- + 1 file changed, 5 insertions(+), 5 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 884002e3b7b5..69cfa4b86a6e 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8372,16 +8372,16 @@ aarch64_expand_prologue (void) + if (callee_adjust != 0) + aarch64_push_regs (reg1, reg2, callee_adjust); + +- /* The offset of the frame chain record (if any) from the current SP. */ +- poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); +- gcc_assert (known_ge (chain_offset, 0)); +- + /* The offset of the current SP from the bottom of the static frame. */ + poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { ++ /* The offset of the frame chain record (if any) from the current SP. */ ++ poly_int64 chain_offset = (initial_adjust + callee_adjust ++ - frame.hard_fp_offset); ++ gcc_assert (known_ge (chain_offset, 0)); ++ + if (callee_adjust == 0) + { + reg1 = R29_REGNUM; +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0031-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/10.5.0/0031-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..6878e6f8 --- /dev/null +++ b/packages/gcc/10.5.0/0031-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,91 @@ +From 45365e7aeb03a56ac4ec0781764017952645348f Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:25:40 +0100 +Subject: [PATCH 31/42] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 6 +++--- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 6 insertions(+), 6 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 69cfa4b86a6e..456b8b3308c1 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7022,7 +7022,7 @@ aarch64_layout_frame (void) + STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + +- frame.locals_offset = frame.saved_varargs_size; ++ frame.bytes_above_locals = frame.saved_varargs_size; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +@@ -11184,13 +11184,13 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.locals_offset; ++ return frame.hard_fp_offset - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return frame.frame_size - frame.locals_offset; ++ return frame.frame_size - frame.bytes_above_locals; + } + + return frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 405b7545913d..41a847991754 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -834,10 +834,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 locals_offset; ++ poly_int64 bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0032-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/10.5.0/0032-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..ff16c5df --- /dev/null +++ b/packages/gcc/10.5.0/0032-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,148 @@ +From ba6a8c0b2cb27af94913f6dea26e10106e38ec3e Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:28:11 +0100 +Subject: [PATCH 32/42] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.c | 26 +++++++++++++------------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 16 insertions(+), 16 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 456b8b3308c1..81ddb99c1651 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7014,7 +7014,7 @@ aarch64_layout_frame (void) + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.hard_fp_offset ++ frame.bytes_above_hard_fp + = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +@@ -7035,13 +7035,13 @@ aarch64_layout_frame (void) + else if (frame.wb_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ && known_eq (frame.bytes_above_hard_fp, const_size)) + { + /* Simple, small frame with no data below the saved registers. + +@@ -7058,8 +7058,8 @@ aarch64_layout_frame (void) + case that it hardly seems worth the effort though. */ + && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca +- && frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset)) ++ && frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset)) + { + /* Frame with small area below the saved registers: + +@@ -7077,12 +7077,12 @@ aarch64_layout_frame (void) + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.hard_fp_offset ++ frame.initial_adjust = (frame.bytes_above_hard_fp + + frame.below_hard_fp_saved_regs_size); + frame.final_adjust = frame.bytes_below_saved_regs; + } +- else if (frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset) ++ else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset) + { + /* Frame with large area below the saved registers, or with SVE saves, + but with a small area above: +@@ -7092,7 +7092,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.callee_adjust = const_fp_offset; ++ frame.callee_adjust = const_above_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -7107,7 +7107,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = frame.hard_fp_offset; ++ frame.initial_adjust = frame.bytes_above_hard_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -8379,7 +8379,7 @@ aarch64_expand_prologue (void) + { + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); ++ - frame.bytes_above_hard_fp); + gcc_assert (known_ge (chain_offset, 0)); + + if (callee_adjust == 0) +@@ -11181,10 +11181,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return frame.hard_fp_offset; ++ return frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.bytes_above_locals; ++ return frame.bytes_above_hard_fp - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 41a847991754..9839e18a27ef 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -839,10 +839,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 hard_fp_offset; ++ poly_int64 bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0033-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/10.5.0/0033-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..b60f91dd --- /dev/null +++ b/packages/gcc/10.5.0/0033-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,35 @@ +From b5e2fcfa7a1c722fe426132562af4a96fff9fb5c Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 22 Jun 2023 22:26:30 +0100 +Subject: [PATCH 33/42] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 9839e18a27ef..8d8916c7227e 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -844,8 +844,8 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- frame (incomming SP) to the stack_pointer. This value is always ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always + a multiple of STACK_BOUNDARY. */ + poly_int64 frame_size; + +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0034-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch b/packages/gcc/10.5.0/0034-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch new file mode 100644 index 00000000..4ad67cdf --- /dev/null +++ b/packages/gcc/10.5.0/0034-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch @@ -0,0 +1,196 @@ +From ae7192e24b8fd6b3f14b44de03bebc6af92c13ed Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:36:08 +0100 +Subject: [PATCH 34/42] aarch64: Measure reg_offset from the bottom of the + frame + +reg_offset was measured from the bottom of the saved register area. +This made perfect sense with the original layout, since the bottom +of the saved register area was also the hard frame pointer address. +It became slightly less obvious with SVE, since we save SVE +registers below the hard frame pointer, but it still made sense. + +However, if we want to allow different frame layouts, it's more +convenient and obvious to measure reg_offset from the bottom of +the frame. After previous patches, it's also a slight simplification +in its own right. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame): Add comment above + reg_offset. + * config/aarch64/aarch64.c (aarch64_layout_frame): Walk offsets + from the bottom of the frame, rather than the bottom of the saved + register area. Measure reg_offset from the bottom of the frame + rather than the bottom of the saved register area. + (aarch64_save_callee_saves): Update accordingly. + (aarch64_restore_callee_saves): Likewise. + (aarch64_get_separate_components): Likewise. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 53 ++++++++++++++++-------------------- + gcc/config/aarch64/aarch64.h | 3 ++ + 2 files changed, 27 insertions(+), 29 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 81ddb99c1651..1ed335f54879 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6826,7 +6826,6 @@ aarch64_needs_frame_chain (void) + static void + aarch64_layout_frame (void) + { +- poly_int64 offset = 0; + int regno, last_fp_reg = INVALID_REGNUM; + machine_mode vector_save_mode = aarch64_reg_save_mode (V8_REGNUM); + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); +@@ -6902,7 +6901,9 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + +- frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ poly_int64 offset = crtl->outgoing_args_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.bytes_below_saved_regs = offset; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -6914,7 +6915,8 @@ aarch64_layout_frame (void) + offset += BYTES_PER_SVE_PRED; + } + +- if (maybe_ne (offset, 0)) ++ poly_int64 saved_prs_size = offset - frame.bytes_below_saved_regs; ++ if (maybe_ne (saved_prs_size, 0)) + { + /* If we have any vector registers to save above the predicate registers, + the offset of the vector register save slots need to be a multiple +@@ -6932,10 +6934,10 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + else + { +- if (known_le (offset, vector_save_size)) +- offset = vector_save_size; +- else if (known_le (offset, vector_save_size * 2)) +- offset = vector_save_size * 2; ++ if (known_le (saved_prs_size, vector_save_size)) ++ offset = frame.bytes_below_saved_regs + vector_save_size; ++ else if (known_le (saved_prs_size, vector_save_size * 2)) ++ offset = frame.bytes_below_saved_regs + vector_save_size * 2; + else + gcc_unreachable (); + } +@@ -6952,9 +6954,10 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- bool saves_below_hard_fp_p = maybe_ne (offset, 0); +- frame.below_hard_fp_saved_regs_size = offset; +- frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; ++ frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p ++ = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ frame.bytes_below_hard_fp = offset; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -7005,9 +7008,10 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset; ++ frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; ++ poly_int64 varargs_and_saved_regs_size ++ = frame.saved_regs_size + frame.saved_varargs_size; + + poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size +@@ -7430,9 +7434,7 @@ aarch64_save_callee_saves (poly_int64 bytes_below_sp, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -7539,9 +7541,7 @@ aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -7669,14 +7669,12 @@ aarch64_get_separate_components (void) + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection + && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, 0)) ++ && known_eq (offset, frame.bytes_below_saved_regs)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -7823,9 +7821,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -7877,9 +7873,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= frame.below_hard_fp_saved_regs_size; +- else +- offset2 += frame.bytes_below_saved_regs; ++ offset2 -= frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -7995,7 +7989,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p + && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] ++ - frame.bytes_below_saved_regs); + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 8d8916c7227e..1c3c62639cbf 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -809,6 +809,9 @@ extern enum aarch64_processor aarch64_tune; + #ifdef HAVE_POLY_INT_H + struct GTY (()) aarch64_frame + { ++ /* The offset from the bottom of the static frame (the bottom of the ++ outgoing arguments) of each register save slot, or -2 if no save is ++ needed. */ + poly_int64 reg_offset[LAST_SAVED_REGNUM + 1]; + + /* The number of extra stack bytes taken up by register varargs. +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0035-aarch64-Simplify-top-of-frame-allocation.patch b/packages/gcc/10.5.0/0035-aarch64-Simplify-top-of-frame-allocation.patch new file mode 100644 index 00000000..52eac0d8 --- /dev/null +++ b/packages/gcc/10.5.0/0035-aarch64-Simplify-top-of-frame-allocation.patch @@ -0,0 +1,55 @@ +From ff7193c5313d736e20aa66c33abeaf102cc6c152 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 17:44:39 +0100 +Subject: [PATCH 35/42] aarch64: Simplify top of frame allocation + +After previous patches, it no longer really makes sense to allocate +the top of the frame in terms of varargs_and_saved_regs_size and +saved_regs_and_above. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Simplify + the allocation of the top of the frame. +--- + gcc/config/aarch64/aarch64.c | 23 ++++++++--------------- + 1 file changed, 8 insertions(+), 15 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 1ed335f54879..70542c6e13f5 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7010,23 +7010,16 @@ aarch64_layout_frame (void) + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size +- = frame.saved_regs_size + frame.saved_varargs_size; ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ poly_int64 top_of_locals = offset; + +- poly_int64 saved_regs_and_above +- = aligned_upper_bound (varargs_and_saved_regs_size +- + get_frame_size (), +- STACK_BOUNDARY / BITS_PER_UNIT); ++ offset += frame.saved_varargs_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.frame_size = offset; + +- frame.bytes_above_hard_fp +- = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; +- +- /* Both these values are already aligned. */ +- gcc_assert (multiple_p (frame.bytes_below_saved_regs, +- STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; +- +- frame.bytes_above_locals = frame.saved_varargs_size; ++ frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0036-aarch64-Minor-initial-adjustment-tweak.patch b/packages/gcc/10.5.0/0036-aarch64-Minor-initial-adjustment-tweak.patch new file mode 100644 index 00000000..be080c56 --- /dev/null +++ b/packages/gcc/10.5.0/0036-aarch64-Minor-initial-adjustment-tweak.patch @@ -0,0 +1,38 @@ +From 5854b895a92e9fe8b26a051442d44a6fd685512a Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 22 Jun 2023 22:18:45 +0100 +Subject: [PATCH 36/42] aarch64: Minor initial adjustment tweak + +This patch just changes a calculation of initial_adjust +to one that makes it slightly more obvious that the total +adjustment is frame.frame_size. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Tweak + calculation of initial_adjust for frames in which all saves + are SVE saves. +--- + gcc/config/aarch64/aarch64.c | 5 ++--- + 1 file changed, 2 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 70542c6e13f5..08a8667bbc18 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7071,11 +7071,10 @@ aarch64_layout_frame (void) + { + /* Frame in which all saves are SVE saves: + +- sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size ++ sub sp, sp, frame_size - bytes_below_saved_regs + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.bytes_above_hard_fp +- + frame.below_hard_fp_saved_regs_size); ++ frame.initial_adjust = frame.frame_size - frame.bytes_below_saved_regs; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0037-aarch64-Tweak-stack-clash-boundary-condition.patch b/packages/gcc/10.5.0/0037-aarch64-Tweak-stack-clash-boundary-condition.patch new file mode 100644 index 00000000..3f8ea9a4 --- /dev/null +++ b/packages/gcc/10.5.0/0037-aarch64-Tweak-stack-clash-boundary-condition.patch @@ -0,0 +1,125 @@ +From af6bac9775e7d2bc94e4e3600740c3c475e117b5 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 15:11:44 +0100 +Subject: [PATCH 37/42] aarch64: Tweak stack clash boundary condition + +The AArch64 ABI says that, when stack clash protection is used, +there can be a maximum of 1KiB of unprobed space at sp on entry +to a function. Therefore, we need to probe when allocating +>= guard_size - 1KiB of data (>= rather than >). This is what +GCC does. + +If an allocation is exactly guard_size bytes, it is enough to allocate +those bytes and probe once at offset 1024. It isn't possible to use a +single probe at any other offset: higher would conmplicate later code, +by leaving more unprobed space than usual, while lower would risk +leaving an entire page unprobed. For simplicity, the code probes all +allocations at offset 1024. + +Some register saves also act as probes. If we need to allocate +more space below the last such register save probe, we need to +probe the allocation if it is > 1KiB. Again, this allocation is +then sometimes (but not always) probed at offset 1024. This sort of +allocation is currently only used for outgoing arguments, which are +rarely this big. + +However, the code also probed if this final outgoing-arguments +allocation was == 1KiB, rather than just > 1KiB. This isn't +necessary, since the register save then probes at offset 1024 +as required. Continuing to probe allocations of exactly 1KiB +would complicate later patches. + +gcc/ + * config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): + Don't probe final allocations that are exactly 1KiB in size (after + unprobed space above the final allocation has been deducted). + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: New test. +--- + gcc/config/aarch64/aarch64.c | 4 +- + .../aarch64/stack-check-prologue-17.c | 55 +++++++++++++++++++ + 2 files changed, 58 insertions(+), 1 deletion(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 08a8667bbc18..43240af8ae43 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7962,9 +7962,11 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; ++ HOST_WIDE_INT byte_sp_alignment = STACK_BOUNDARY / BITS_PER_UNIT; ++ gcc_assert (multiple_p (poly_size, byte_sp_alignment)); + HOST_WIDE_INT min_probe_threshold + = (final_adjustment_p +- ? guard_used_by_caller ++ ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); + /* When doing the final adjustment for the outgoing arguments, take into + account any unprobed space there is above the current SP. There are +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +new file mode 100644 +index 000000000000..0d8a25d73a24 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -0,0 +1,55 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0038-aarch64-Put-LR-save-probe-in-first-16-bytes.patch b/packages/gcc/10.5.0/0038-aarch64-Put-LR-save-probe-in-first-16-bytes.patch new file mode 100644 index 00000000..4e9409c6 --- /dev/null +++ b/packages/gcc/10.5.0/0038-aarch64-Put-LR-save-probe-in-first-16-bytes.patch @@ -0,0 +1,270 @@ +From 128abc59aedc06b4418ac57d08a484e1fd92dee2 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 15:12:55 +0100 +Subject: [PATCH 38/42] aarch64: Put LR save probe in first 16 bytes + +-fstack-clash-protection uses the save of LR as a probe for the next +allocation. The next allocation could be: + +* another part of the static frame, e.g. when allocating SVE save slots + or outgoing arguments + +* an alloca in the same function + +* an allocation made by a callee function + +However, when -fomit-frame-pointer is used, the LR save slot is placed +above the other GPR save slots. It could therefore be up to 80 bytes +above the base of the GPR save area (which is also the hard fp address). + +aarch64_allocate_and_probe_stack_space took this into account when +deciding how much subsequent space could be allocated without needing +a probe. However, it interacted badly with: + + /* If doing a small final adjustment, we always probe at offset 0. + This is done to avoid issues when LR is not at position 0 or when + the final adjustment is smaller than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +which forces any allocation that is smaller than the guard page size +to be probed at offset 0 rather than the usual offset 1024. It was +therefore possible to construct cases in which we had: + +* a probe using LR at SP + 80 bytes (or some other value >= 16) +* an allocation of the guard page size - 16 bytes +* a probe at SP + 0 + +which allocates guard page size + 64 consecutive unprobed bytes. + +This patch requires the LR probe to be in the first 16 bytes of the +save area when stack clash protection is active. Doing it +unconditionally would cause code-quality regressions, but a later +patch deals with that. + +The new comment doesn't say that the probe register is required +to be LR, since a later patch removes that restriction. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Ensure that + the LR save slot is in the first 16 bytes of the register save area. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + when LR was not in the first 16 bytes. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-18.c: New test. +--- + gcc/config/aarch64/aarch64.c | 65 +++++------ + .../aarch64/stack-check-prologue-18.c | 103 ++++++++++++++++++ + 2 files changed, 130 insertions(+), 38 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 43240af8ae43..3d73c5f352f6 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6958,26 +6958,38 @@ aarch64_layout_frame (void) + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); + frame.bytes_below_hard_fp = offset; ++ ++#define ALLOCATE_GPR_SLOT(REGNO) \ ++ do \ ++ { \ ++ frame.reg_offset[REGNO] = offset; \ ++ if (frame.wb_candidate1 == INVALID_REGNUM) \ ++ frame.wb_candidate1 = (REGNO); \ ++ else if (frame.wb_candidate2 == INVALID_REGNUM) \ ++ frame.wb_candidate2 = (REGNO); \ ++ offset += UNITS_PER_WORD; \ ++ } \ ++ while (0) ++ + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +- frame.reg_offset[R29_REGNUM] = offset; +- frame.wb_candidate1 = R29_REGNUM; +- frame.reg_offset[R30_REGNUM] = offset + UNITS_PER_WORD; +- frame.wb_candidate2 = R30_REGNUM; +- offset += 2 * UNITS_PER_WORD; ++ ALLOCATE_GPR_SLOT (R29_REGNUM); ++ ALLOCATE_GPR_SLOT (R30_REGNUM); + } ++ else if (flag_stack_clash_protection ++ && known_eq (frame.reg_offset[R30_REGNUM], SLOT_REQUIRED)) ++ /* Put the LR save slot first, since it makes a good choice of probe ++ for stack clash purposes. The idea is that the link register usually ++ has to be saved before a call anyway, and so we lose little by ++ stopping it from being individually shrink-wrapped. */ ++ ALLOCATE_GPR_SLOT (R30_REGNUM); + + for (regno = R0_REGNUM; regno <= R30_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) +- { +- frame.reg_offset[regno] = offset; +- if (frame.wb_candidate1 == INVALID_REGNUM) +- frame.wb_candidate1 = regno; +- else if (frame.wb_candidate2 == INVALID_REGNUM) +- frame.wb_candidate2 = regno; +- offset += UNITS_PER_WORD; +- } ++ ALLOCATE_GPR_SLOT (regno); ++ ++#undef ALLOCATE_GPR_SLOT + + poly_int64 max_int_offset = offset; + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -7968,29 +7980,6 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + = (final_adjustment_p + ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); +- /* When doing the final adjustment for the outgoing arguments, take into +- account any unprobed space there is above the current SP. There are +- two cases: +- +- - When saving SVE registers below the hard frame pointer, we force +- the lowest save to take place in the prologue before doing the final +- adjustment (i.e. we don't allow the save to be shrink-wrapped). +- This acts as a probe at SP, so there is no unprobed space. +- +- - When there are no SVE register saves, we use the store of the link +- register as a probe. We can't assume that LR was saved at position 0 +- though, so treat any space below it as unprobed. */ +- if (final_adjustment_p +- && known_eq (frame.below_hard_fp_saved_regs_size, 0)) +- { +- poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] +- - frame.bytes_below_saved_regs); +- if (known_ge (lr_offset, 0)) +- min_probe_threshold -= lr_offset.to_constant (); +- else +- gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); +- } +- + poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ +@@ -8170,8 +8159,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; + /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when LR is not at position 0 or when +- the final adjustment is smaller than the probing offset. */ ++ This is done to avoid issues when the final adjustment is smaller ++ than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +new file mode 100644 +index 000000000000..b646f040b54e +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -0,0 +1,103 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** ... ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** ... ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** ... ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0039-aarch64-Simplify-probe-of-final-frame-allocation.patch b/packages/gcc/10.5.0/0039-aarch64-Simplify-probe-of-final-frame-allocation.patch new file mode 100644 index 00000000..843b0efc --- /dev/null +++ b/packages/gcc/10.5.0/0039-aarch64-Simplify-probe-of-final-frame-allocation.patch @@ -0,0 +1,99 @@ +From c940aadecb4f623a39abe16b6c62b307e2b10638 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 8 Aug 2023 01:57:26 +0100 +Subject: [PATCH 39/42] aarch64: Simplify probe of final frame allocation + +Previous patches ensured that the final frame allocation only needs +a probe when the size is strictly greater than 1KiB. It's therefore +safe to use the normal 1024 probe offset in all cases. + +The main motivation for doing this is to simplify the code and +remove the number of special cases. + +gcc/ + * config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): + Always probe the residual allocation at offset 1024, asserting + that that is in range. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: Expect the probe + to be at offset 1024 rather than offset 0. + * gcc.target/aarch64/stack-check-prologue-18.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 12 ++++-------- + .../gcc.target/aarch64/stack-check-prologue-17.c | 2 +- + .../gcc.target/aarch64/stack-check-prologue-18.c | 4 ++-- + 3 files changed, 7 insertions(+), 11 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 3d73c5f352f6..37c6219b07ab 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8153,16 +8153,12 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + are still safe. */ + if (residual) + { +- HOST_WIDE_INT residual_probe_offset = guard_used_by_caller; ++ gcc_assert (guard_used_by_caller + byte_sp_alignment <= size); ++ + /* If we're doing final adjustments, and we've done any full page + allocations then any residual needs to be probed. */ + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; +- /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when the final adjustment is smaller +- than the probing offset. */ +- else if (final_adjustment_p && rounded_size == 0) +- residual_probe_offset = 0; + + aarch64_sub_sp (temp1, temp2, residual, frame_related_p); + if (residual >= min_probe_threshold) +@@ -8173,8 +8169,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT_PRINT_DEC " bytes, probing will be required." + "\n", residual); + +- emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, +- residual_probe_offset)); ++ emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, ++ guard_used_by_caller)); + emit_insn (gen_blockage ()); + } + } +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +index 0d8a25d73a24..f0ec1389771d 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -33,7 +33,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +index b646f040b54e..71d33ba34e9e 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -10,7 +10,7 @@ void g(); + ** str x30, \[sp\] + ** ... + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -52,7 +52,7 @@ int test1(int z) { + ** str x30, \[sp\] + ** ... + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0040-aarch64-Explicitly-record-probe-registers-in-frame-i.patch b/packages/gcc/10.5.0/0040-aarch64-Explicitly-record-probe-registers-in-frame-i.patch new file mode 100644 index 00000000..70bbfb78 --- /dev/null +++ b/packages/gcc/10.5.0/0040-aarch64-Explicitly-record-probe-registers-in-frame-i.patch @@ -0,0 +1,277 @@ +From 3c73e7bc86a918c2c24b3ad96b2d44716a5e2614 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:50:34 +0100 +Subject: [PATCH 40/42] aarch64: Explicitly record probe registers in frame + info + +The stack frame is currently divided into three areas: + +A: the area above the hard frame pointer +B: the SVE saves below the hard frame pointer +C: the outgoing arguments + +If the stack frame is allocated in one chunk, the allocation needs a +probe if the frame size is >= guard_size - 1KiB. In addition, if the +function is not a leaf function, it must probe an address no more than +1KiB above the outgoing SP. We ensured the second condition by + +(1) using single-chunk allocations for non-leaf functions only if + the link register save slot is within 512 bytes of the bottom + of the frame; and + +(2) using the link register save as a probe (meaning, for instance, + that it can't be individually shrink wrapped) + +If instead the stack is allocated in multiple chunks, then: + +* an allocation involving only the outgoing arguments (C above) requires + a probe if the allocation size is > 1KiB + +* any other allocation requires a probe if the allocation size + is >= guard_size - 1KiB + +* second and subsequent allocations require the previous allocation + to probe at the bottom of the allocated area, regardless of the size + of that previous allocation + +The final point means that, unlike for single allocations, +it can be necessary to have both a non-SVE register probe and +an SVE register probe. For example: + +* allocate A, probe using a non-SVE register save +* allocate B, probe using an SVE register save +* allocate C + +The non-SVE register used in this case was again the link register. +It was previously used even if the link register save slot was some +bytes above the bottom of the non-SVE register saves, but an earlier +patch avoided that by putting the link register save slot first. + +As a belt-and-braces fix, this patch explicitly records which +probe registers we're using and allows the non-SVE probe to be +whichever register comes first (as for SVE). + +The patch also avoids unnecessary probes in sve/pcs/stack_clash_3.c. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::sve_save_and_probe) + (aarch64_frame::hard_fp_save_and_probe): New fields. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize them. + Rather than asserting that a leaf function saves LR, instead assert + that a leaf function saves something. + (aarch64_get_separate_components): Prevent the chosen probe + registers from being individually shrink-wrapped. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + probe registers that aren't at the bottom of the previous allocation. + +gcc/testsuite/ + * gcc.target/aarch64/sve/pcs/stack_clash_3.c: Avoid redundant probes. +--- + gcc/config/aarch64/aarch64.c | 67 +++++++++++++++---- + gcc/config/aarch64/aarch64.h | 8 +++ + .../aarch64/sve/pcs/stack_clash_3.c | 6 +- + 3 files changed, 64 insertions(+), 17 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 37c6219b07ab..2dea9b5617d3 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6897,13 +6897,10 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + +- /* With stack-clash, LR must be saved in non-leaf functions. */ +- gcc_assert (crtl->is_leaf +- || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); +- + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.bytes_below_saved_regs = offset; ++ frame.sve_save_and_probe = INVALID_REGNUM; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -6911,6 +6908,8 @@ aarch64_layout_frame (void) + for (regno = P0_REGNUM; regno <= P15_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += BYTES_PER_SVE_PRED; + } +@@ -6948,6 +6947,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += vector_save_size; + } +@@ -6957,11 +6958,19 @@ aarch64_layout_frame (void) + frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ gcc_assert (!saves_below_hard_fp_p ++ || (frame.sve_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.sve_save_and_probe], ++ frame.bytes_below_saved_regs))); ++ + frame.bytes_below_hard_fp = offset; ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; + + #define ALLOCATE_GPR_SLOT(REGNO) \ + do \ + { \ ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) \ ++ frame.hard_fp_save_and_probe = (REGNO); \ + frame.reg_offset[REGNO] = offset; \ + if (frame.wb_candidate1 == INVALID_REGNUM) \ + frame.wb_candidate1 = (REGNO); \ +@@ -6998,6 +7007,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + /* If there is an alignment gap between integer and fp callee-saves, + allocate the last fp register to it if possible. */ + if (regno == last_fp_reg +@@ -7021,6 +7032,17 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (frame.saved_regs_size, ++ frame.below_hard_fp_saved_regs_size) ++ || (frame.hard_fp_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], ++ frame.bytes_below_hard_fp))); ++ ++ /* With stack-clash, a register must be saved in non-leaf functions. ++ The saving of the bottommost register counts as an implicit probe, ++ which allows us to maintain the invariant described in the comment ++ at expand_prologue. */ ++ gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -7120,6 +7142,25 @@ aarch64_layout_frame (void) + frame.final_adjust = frame.bytes_below_saved_regs; + } + ++ /* The frame is allocated in pieces, with each non-final piece ++ including a register save at offset 0 that acts as a probe for ++ the following piece. In addition, the save of the bottommost register ++ acts as a probe for callees and allocas. Roll back any probes that ++ aren't needed. ++ ++ A probe isn't needed if it is associated with the final allocation ++ (including callees and allocas) that happens before the epilogue is ++ executed. */ ++ if (crtl->is_leaf ++ && !cfun->calls_alloca ++ && known_eq (frame.final_adjust, 0)) ++ { ++ if (maybe_ne (frame.sve_callee_adjust, 0)) ++ frame.sve_save_and_probe = INVALID_REGNUM; ++ else ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; ++ } ++ + /* Make sure the individual adjustments add up to the full frame size. */ + gcc_assert (known_eq (frame.initial_adjust + + frame.callee_adjust +@@ -7669,13 +7710,6 @@ aarch64_get_separate_components (void) + + poly_int64 offset = frame.reg_offset[regno]; + +- /* If the register is saved in the first SVE save slot, we use +- it as a stack probe for -fstack-clash-protection. */ +- if (flag_stack_clash_protection +- && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, frame.bytes_below_saved_regs)) +- continue; +- + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) + offset -= frame.bytes_below_hard_fp; +@@ -7710,6 +7744,13 @@ aarch64_get_separate_components (void) + + bitmap_clear_bit (components, LR_REGNUM); + bitmap_clear_bit (components, SP_REGNUM); ++ if (flag_stack_clash_protection) ++ { ++ if (frame.sve_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.sve_save_and_probe); ++ if (frame.hard_fp_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.hard_fp_save_and_probe); ++ } + + return components; + } +@@ -8246,8 +8287,8 @@ aarch64_epilogue_uses (int regno) + When probing is needed, we emit a probe at the start of the prologue + and every PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE bytes thereafter. + +- We have to track how much space has been allocated and the only stores +- to the stack we track as implicit probes are the FP/LR stores. ++ We can also use register saves as probes. These are stored in ++ sve_save_and_probe and hard_fp_save_and_probe. + + For outgoing arguments we probe if the size is larger than 1KB, such that + the ABI specified buffer is maintained for the next callee. +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 1c3c62639cbf..5279503fd090 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -877,6 +877,14 @@ struct GTY (()) aarch64_frame + This is the register they should use. */ + unsigned spare_pred_reg; + ++ /* An SVE register that is saved below the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned sve_save_and_probe; ++ ++ /* A register that is saved at the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned hard_fp_save_and_probe; ++ + bool laid_out; + }; + +diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +index 3e01ec36c3a4..3530a0d504ba 100644 +--- a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c ++++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +@@ -11,11 +11,10 @@ + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +@@ -39,13 +38,12 @@ test_1 (int n) + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** str p5, \[sp, #1, mul vl\] + ** str p6, \[sp, #2, mul vl\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0041-aarch64-Remove-below_hard_fp_saved_regs_size.patch b/packages/gcc/10.5.0/0041-aarch64-Remove-below_hard_fp_saved_regs_size.patch new file mode 100644 index 00000000..0610853b --- /dev/null +++ b/packages/gcc/10.5.0/0041-aarch64-Remove-below_hard_fp_saved_regs_size.patch @@ -0,0 +1,158 @@ +From 1c659316abe90c98b98494c9f13bd040e25063d8 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 22 Jun 2023 22:15:39 +0100 +Subject: [PATCH 41/42] aarch64: Remove below_hard_fp_saved_regs_size + +After previous patches, it's no longer necessary to store +saved_regs_size and below_hard_fp_saved_regs_size in the frame info. +All measurements instead use the top or bottom of the frame as +reference points. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::saved_regs_size) + (aarch64_frame::below_hard_fp_saved_regs_size): Delete. + * config/aarch64/aarch64.c (aarch64_layout_frame): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 46 +++++++++++++++++------------------- + gcc/config/aarch64/aarch64.h | 7 ------ + 2 files changed, 22 insertions(+), 31 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 2dea9b5617d3..91833473c751 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6955,9 +6955,9 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; +- bool saves_below_hard_fp_p +- = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ poly_int64 below_hard_fp_saved_regs_size ++ = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p = maybe_ne (below_hard_fp_saved_regs_size, 0); + gcc_assert (!saves_below_hard_fp_p + || (frame.sve_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.sve_save_and_probe], +@@ -7031,9 +7031,8 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset - frame.bytes_below_saved_regs; +- gcc_assert (known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size) ++ poly_int64 saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (saved_regs_size, below_hard_fp_saved_regs_size) + || (frame.hard_fp_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], + frame.bytes_below_hard_fp))); +@@ -7042,7 +7041,7 @@ aarch64_layout_frame (void) + The saving of the bottommost register counts as an implicit probe, + which allows us to maintain the invariant described in the comment + at expand_prologue. */ +- gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); ++ gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -7068,7 +7067,7 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; +- if (known_eq (frame.saved_regs_size, 0)) ++ if (known_eq (saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +@@ -7081,7 +7080,7 @@ aarch64_layout_frame (void) + frame.callee_adjust = const_size; + } + else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) +- && frame.saved_regs_size.is_constant (&const_saved_regs_size) ++ && saved_regs_size.is_constant (&const_saved_regs_size) + && const_below_saved_regs + const_saved_regs_size < 512 + /* We could handle this case even with data below the saved + registers, provided that that data left us with valid offsets +@@ -7100,8 +7099,7 @@ aarch64_layout_frame (void) + frame.initial_adjust = frame.frame_size; + } + else if (saves_below_hard_fp_p +- && known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size)) ++ && known_eq (saved_regs_size, below_hard_fp_saved_regs_size)) + { + /* Frame in which all saves are SVE saves: + +@@ -7123,7 +7121,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_above_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else +@@ -7138,7 +7136,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.bytes_above_hard_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + +@@ -8252,17 +8250,17 @@ aarch64_epilogue_uses (int regno) + | local variables | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | \ +- +-------------------------------+ | +- | callee-saved registers | | frame.saved_regs_size +- +-------------------------------+ | +- | LR' | | +- +-------------------------------+ | +- | FP' | | +- +-------------------------------+ |<- hard_frame_pointer_rtx (aligned) +- | SVE vector registers | | \ +- +-------------------------------+ | | below_hard_fp_saved_regs_size +- | SVE predicate registers | / / ++ | padding | ++ +-------------------------------+ ++ | callee-saved registers | ++ +-------------------------------+ ++ | LR' | ++ +-------------------------------+ ++ | FP' | ++ +-------------------------------+ <-- hard_frame_pointer_rtx (aligned) ++ | SVE vector registers | ++ +-------------------------------+ ++ | SVE predicate registers | + +-------------------------------+ + | dynamic allocation | + +-------------------------------+ +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 5279503fd090..b0d13cf08ece 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -820,18 +820,11 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + HOST_WIDE_INT saved_varargs_size; + +- /* The size of the callee-save registers with a slot in REG_OFFSET. */ +- poly_int64 saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the bottom of the register save area. + This value is always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_saved_regs; + +- /* The size of the callee-save registers with a slot in REG_OFFSET that +- are saved below the hard frame pointer. */ +- poly_int64 below_hard_fp_saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the hard frame pointer. This value is + always a multiple of STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/10.5.0/0042-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/10.5.0/0042-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..f7c235e8 --- /dev/null +++ b/packages/gcc/10.5.0/0042-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,299 @@ +From dad1c82acf67a8a1cc68e2acf3ec89a503e361fc Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 15 Jun 2023 19:16:52 +0100 +Subject: [PATCH 42/42] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 47 +++++++-- + .../gcc.target/aarch64/stack-protector-8.c | 95 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 +++++++ + 3 files changed, 169 insertions(+), 6 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 91833473c751..ce701b91a3f2 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -6820,6 +6820,20 @@ aarch64_needs_frame_chain (void) + return aarch64_use_frame_pointer; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -6831,6 +6845,7 @@ aarch64_layout_frame (void) + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); + bool frame_related_fp_reg_p = false; + aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 top_of_locals = -1; + + frame.emit_frame_chain = aarch64_needs_frame_chain (); + +@@ -6897,8 +6912,16 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); ++ + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ if (regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + frame.bytes_below_saved_regs = offset; + frame.sve_save_and_probe = INVALID_REGNUM; + +@@ -7043,15 +7066,18 @@ aarch64_layout_frame (void) + at expand_prologue. */ + gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + +- offset += get_frame_size (); +- offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +- poly_int64 top_of_locals = offset; +- ++ if (!regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + offset += frame.saved_varargs_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = offset; + + frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ gcc_assert (known_ge (top_of_locals, 0)); + frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; +@@ -8247,10 +8273,10 @@ aarch64_epilogue_uses (int regno) + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | ++ | padding (1) | + +-------------------------------+ + | callee-saved registers | + +-------------------------------+ +@@ -8262,6 +8288,10 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | SVE predicate registers | + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -8271,6 +8301,9 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. +@@ -8462,6 +8495,8 @@ aarch64_expand_prologue (void) + gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); ++ if (emit_frame_chain && maybe_ne (final_adjust, 0)) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..e71d820e3654 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,95 @@ ++/* { dg-options " -O -fstack-protector-strong -mstack-protector-guard=sysreg -mstack-protector-guard-reg=tpidr2_el0 -mstack-protector-guard-offset=16" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++__SVBool_t *h(void *); ++ ++/* ++** test1: ++** sub sp, sp, #288 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?288 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** sub sp, sp, #1040 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?1032\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?1040 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} ++ ++#pragma GCC target "+sve" ++ ++/* ++** test3: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** addvl sp, sp, #-18 ++** ... ++** str p4, \[sp\] ++** ... ++** sub sp, sp, #272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl h ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?272 ++** ... ++** ldr p4, \[sp\] ++** ... ++** addvl sp, sp, #18 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++__SVBool_t test3() { ++ int y[0x40]; ++ return *h(y); ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..58f322aa480a +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=neoverse-v1 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0011-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch b/packages/gcc/11.4.0/0011-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch new file mode 100644 index 00000000..759214b0 --- /dev/null +++ b/packages/gcc/11.4.0/0011-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch @@ -0,0 +1,349 @@ +From 52816ab48f97968f3fbfb5656250f3de7c00166d Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:43 +0100 +Subject: [PATCH 11/29] aarch64: Use local frame vars in shrink-wrapping code + +aarch64_layout_frame uses a shorthand for referring to +cfun->machine->frame: + + aarch64_frame &frame = cfun->machine->frame; + +This patch does the same for some other heavy users of the structure. +No functional change intended. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_callee_saves): Use + a local shorthand for cfun->machine->frame. + (aarch64_restore_callee_saves, aarch64_get_separate_components): + (aarch64_process_components): Likewise. + (aarch64_allocate_and_probe_stack_space): Likewise. + (aarch64_expand_prologue, aarch64_expand_epilogue): Likewise. + (aarch64_layout_frame): Use existing shorthand for one more case. +--- + gcc/config/aarch64/aarch64.c | 115 ++++++++++++++++++----------------- + 1 file changed, 60 insertions(+), 55 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 391a93f30184..77c1d1300a5c 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7994,6 +7994,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx_insn *insn; + unsigned regno; + unsigned regno2; +@@ -8008,8 +8009,8 @@ aarch64_save_callee_saves (poly_int64 start_offset, + bool frame_related_p = aarch64_emit_cfi_for_reg_p (regno); + + if (skip_wb +- && (regno == cfun->machine->frame.wb_candidate1 +- || regno == cfun->machine->frame.wb_candidate2)) ++ && (regno == frame.wb_candidate1 ++ || regno == frame.wb_candidate2)) + continue; + + if (cfun->machine->reg_is_wrapped_separately[regno]) +@@ -8017,7 +8018,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -8030,7 +8031,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + { + gcc_assert (known_eq (start_offset, 0)); + poly_int64 fp_offset +- = cfun->machine->frame.below_hard_fp_saved_regs_size; ++ = frame.below_hard_fp_saved_regs_size; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -8052,8 +8053,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -8103,6 +8103,7 @@ static void + aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { ++ aarch64_frame &frame = cfun->machine->frame; + unsigned regno; + unsigned regno2; + poly_int64 offset; +@@ -8119,13 +8120,13 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + rtx reg, mem; + + if (skip_wb +- && (regno == cfun->machine->frame.wb_candidate1 +- || regno == cfun->machine->frame.wb_candidate2)) ++ && (regno == frame.wb_candidate1 ++ || regno == frame.wb_candidate2)) + continue; + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -8136,8 +8137,7 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -8242,6 +8242,7 @@ offset_12bit_unsigned_scaled_p (machine_mode mode, poly_int64 offset) + static sbitmap + aarch64_get_separate_components (void) + { ++ aarch64_frame &frame = cfun->machine->frame; + sbitmap components = sbitmap_alloc (LAST_SAVED_REGNUM + 1); + bitmap_clear (components); + +@@ -8258,18 +8259,18 @@ aarch64_get_separate_components (void) + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + continue; + +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + + /* If the register is saved in the first SVE save slot, we use + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection +- && maybe_ne (cfun->machine->frame.below_hard_fp_saved_regs_size, 0) ++ && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) + && known_eq (offset, 0)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -8288,11 +8289,11 @@ aarch64_get_separate_components (void) + /* If the spare predicate register used by big-endian SVE code + is call-preserved, it must be saved in the main prologue + before any saves that use it. */ +- if (cfun->machine->frame.spare_pred_reg != INVALID_REGNUM) +- bitmap_clear_bit (components, cfun->machine->frame.spare_pred_reg); ++ if (frame.spare_pred_reg != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.spare_pred_reg); + +- unsigned reg1 = cfun->machine->frame.wb_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_candidate2; ++ unsigned reg1 = frame.wb_candidate1; ++ unsigned reg2 = frame.wb_candidate2; + /* If registers have been chosen to be stored/restored with + writeback don't interfere with them to avoid having to output explicit + stack adjustment instructions. */ +@@ -8401,6 +8402,7 @@ aarch64_get_next_set_bit (sbitmap bmp, unsigned int start) + static void + aarch64_process_components (sbitmap components, bool prologue_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx ptr_reg = gen_rtx_REG (Pmode, frame_pointer_needed + ? HARD_FRAME_POINTER_REGNUM + : STACK_POINTER_REGNUM); +@@ -8415,9 +8417,9 @@ aarch64_process_components (sbitmap components, bool prologue_p) + machine_mode mode = aarch64_reg_save_mode (regno); + + rtx reg = gen_rtx_REG (mode, regno); +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -8442,14 +8444,14 @@ aarch64_process_components (sbitmap components, bool prologue_p) + break; + } + +- poly_int64 offset2 = cfun->machine->frame.reg_offset[regno2]; ++ poly_int64 offset2 = frame.reg_offset[regno2]; + /* The next register is not of the same class or its offset is not + mergeable with the current one into a pair. */ + if (aarch64_sve_mode_p (mode) + || !satisfies_constraint_Ump (mem) + || GP_REGNUM_P (regno) != GP_REGNUM_P (regno2) + || (crtl->abi->id () == ARM_PCS_SIMD && FP_REGNUM_P (regno)) +- || maybe_ne ((offset2 - cfun->machine->frame.reg_offset[regno]), ++ || maybe_ne ((offset2 - frame.reg_offset[regno]), + GET_MODE_SIZE (mode))) + { + insn = emit_insn (set); +@@ -8471,7 +8473,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset2 -= frame.below_hard_fp_saved_regs_size; + else + offset2 += crtl->outgoing_args_size; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); +@@ -8566,6 +8568,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + bool frame_related_p, + bool final_adjustment_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; +@@ -8586,25 +8589,25 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + register as a probe. We can't assume that LR was saved at position 0 + though, so treat any space below it as unprobed. */ + if (final_adjustment_p +- && known_eq (cfun->machine->frame.below_hard_fp_saved_regs_size, 0)) ++ && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = cfun->machine->frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else + gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); + } + +- poly_int64 frame_size = cfun->machine->frame.frame_size; ++ poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ + gcc_assert (min_probe_threshold > 0); + + if (flag_stack_clash_protection && !final_adjustment_p) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; + + if (known_eq (frame_size, 0)) + { +@@ -8893,17 +8896,18 @@ aarch64_epilogue_uses (int regno) + void + aarch64_expand_prologue (void) + { +- poly_int64 frame_size = cfun->machine->frame.frame_size; +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 frame_size = frame.frame_size; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_candidate2; +- bool emit_frame_chain = cfun->machine->frame.emit_frame_chain; ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_candidate1; ++ unsigned reg2 = frame.wb_candidate2; ++ bool emit_frame_chain = frame.emit_frame_chain; + rtx_insn *insn; + + if (flag_stack_clash_protection && known_eq (callee_adjust, 0)) +@@ -8969,7 +8973,7 @@ aarch64_expand_prologue (void) + + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - cfun->machine->frame.hard_fp_offset); ++ - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + + /* The offset of the bottom of the save area from the current SP. */ +@@ -9072,15 +9076,16 @@ aarch64_use_return_insn_p (void) + void + aarch64_expand_epilogue (bool for_sibcall) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_candidate2; ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_candidate1; ++ unsigned reg2 = frame.wb_candidate2; + rtx cfi_ops = NULL; + rtx_insn *insn; + /* A stack clash protection prologue may not have left EP0_REGNUM or +@@ -9113,7 +9118,7 @@ aarch64_expand_epilogue (bool for_sibcall) + /* We need to add memory barrier to prevent read from deallocated stack. */ + bool need_barrier_p + = maybe_ne (get_frame_size () +- + cfun->machine->frame.saved_varargs_size, 0); ++ + frame.saved_varargs_size, 0); + + /* Emit a barrier to prevent loads from a deallocated stack. */ + if (maybe_gt (final_adjust, crtl->outgoing_args_size) +@@ -11744,24 +11749,24 @@ aarch64_can_eliminate (const int from ATTRIBUTE_UNUSED, const int to) + poly_int64 + aarch64_initial_elimination_offset (unsigned from, unsigned to) + { ++ aarch64_frame &frame = cfun->machine->frame; ++ + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ return frame.hard_fp_offset - frame.locals_offset; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return frame.frame_size - frame.locals_offset; + } + +- return cfun->machine->frame.frame_size; ++ return frame.frame_size; + } + + +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0012-aarch64-Avoid-a-use-of-callee_offset.patch b/packages/gcc/11.4.0/0012-aarch64-Avoid-a-use-of-callee_offset.patch new file mode 100644 index 00000000..75bb2718 --- /dev/null +++ b/packages/gcc/11.4.0/0012-aarch64-Avoid-a-use-of-callee_offset.patch @@ -0,0 +1,73 @@ +From a2a57f7ec7912e77eb26919545807d90065584ff Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:44 +0100 +Subject: [PATCH 12/29] aarch64: Avoid a use of callee_offset + +When we emit the frame chain, i.e. when we reach Here in this statement +of aarch64_expand_prologue: + + if (emit_frame_chain) + { + // Here + ... + } + +the stack is in one of two states: + +- We've allocated up to the frame chain, but no more. + +- We've allocated the whole frame, and the frame chain is within easy + reach of the new SP. + +The offset of the frame chain from the current SP is available +in aarch64_frame as callee_offset. It is also available as the +chain_offset local variable, where the latter is calculated from other +data. (However, chain_offset is not always equal to callee_offset when +!emit_frame_chain, so chain_offset isn't redundant.) + +In c600df9a4060da3c6121ff4d0b93f179eafd69d1 I switched to using +chain_offset for the initialisation of the hard frame pointer: + + aarch64_add_offset (Pmode, hard_frame_pointer_rtx, +- stack_pointer_rtx, callee_offset, ++ stack_pointer_rtx, chain_offset, + tmp1_rtx, tmp0_rtx, frame_pointer_needed); + +But the later REG_CFA_ADJUST_CFA handling still used callee_offset. + +I think the difference is harmless, but it's more logical for the +CFA note to be in sync, and it's more convenient for later patches +if it uses chain_offset. + +gcc/ + * config/aarch64/aarch64.c (aarch64_expand_prologue): Use + chain_offset rather than callee_offset. +--- + gcc/config/aarch64/aarch64.c | 4 +--- + 1 file changed, 1 insertion(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 77c1d1300a5c..6bc026bd08f9 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8901,7 +8901,6 @@ aarch64_expand_prologue (void) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size + = frame.below_hard_fp_saved_regs_size; +@@ -9010,8 +9009,7 @@ aarch64_expand_prologue (void) + implicit. */ + if (!find_reg_note (insn, REG_CFA_ADJUST_CFA, NULL_RTX)) + { +- rtx src = plus_constant (Pmode, stack_pointer_rtx, +- callee_offset); ++ rtx src = plus_constant (Pmode, stack_pointer_rtx, chain_offset); + add_reg_note (insn, REG_CFA_ADJUST_CFA, + gen_rtx_SET (hard_frame_pointer_rtx, src)); + } +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0013-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/11.4.0/0013-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..16d6cad2 --- /dev/null +++ b/packages/gcc/11.4.0/0013-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,49 @@ +From 5efdcc8ed19d9d9e708a001f5dc695560411496d Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:44 +0100 +Subject: [PATCH 13/29] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +And if we allocate the frame in one go, it should be allocated +as the initial_adjust rather than the final_adjust. This allows the +frame size to grow to guard_size - guard_used_by_caller before a stack +probe is needed. (A frame with no register saves is necessarily a +leaf frame.) + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.c | 8 +++++--- + 1 file changed, 5 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 6bc026bd08f9..05e6ae8c0c96 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7609,9 +7609,11 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; +- if (frame.frame_size.is_constant (&const_size) +- && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ if (known_eq (frame.saved_regs_size, 0)) ++ frame.initial_adjust = frame.frame_size; ++ else if (frame.frame_size.is_constant (&const_size) ++ && const_size < max_push_offset ++ && known_eq (frame.hard_fp_offset, const_size)) + { + /* Simple, small frame with no outgoing arguments: + +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0014-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch b/packages/gcc/11.4.0/0014-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch new file mode 100644 index 00000000..321e4da9 --- /dev/null +++ b/packages/gcc/11.4.0/0014-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch @@ -0,0 +1,233 @@ +From a8385d14318634f2e3a08a75bd2d6e2810f8cec9 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:45 +0100 +Subject: [PATCH 14/29] aarch64: Add bytes_below_saved_regs to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_saved_regs): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_saved_regs instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 71 ++++++++++++++++++------------------ + gcc/config/aarch64/aarch64.h | 5 +++ + 2 files changed, 41 insertions(+), 35 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 05e6ae8c0c96..8fa5a0b25455 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7476,6 +7476,8 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + ++ frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small + offset range. These saves happen below the hard frame pointer. */ +@@ -7580,18 +7582,18 @@ aarch64_layout_frame (void) + + poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; + +- poly_int64 above_outgoing_args ++ poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + + frame.hard_fp_offset +- = above_outgoing_args - frame.below_hard_fp_saved_regs_size; ++ = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +- gcc_assert (multiple_p (crtl->outgoing_args_size, ++ gcc_assert (multiple_p (frame.bytes_below_saved_regs, + STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = above_outgoing_args + crtl->outgoing_args_size; ++ frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + + frame.locals_offset = frame.saved_varargs_size; + +@@ -7607,7 +7609,7 @@ aarch64_layout_frame (void) + else if (frame.wb_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; +@@ -7615,31 +7617,31 @@ aarch64_layout_frame (void) + && const_size < max_push_offset + && known_eq (frame.hard_fp_offset, const_size)) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + frame.callee_adjust = const_size; + } +- else if (crtl->outgoing_args_size.is_constant (&const_outgoing_args_size) ++ else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) + && frame.saved_regs_size.is_constant (&const_saved_regs_size) +- && const_outgoing_args_size + const_saved_regs_size < 512 +- /* We could handle this case even with outgoing args, provided +- that the number of args left us with valid offsets for all +- predicate and vector save slots. It's such a rare case that +- it hardly seems worth the effort though. */ +- && (!saves_below_hard_fp_p || const_outgoing_args_size == 0) ++ && const_below_saved_regs + const_saved_regs_size < 512 ++ /* We could handle this case even with data below the saved ++ registers, provided that that data left us with valid offsets ++ for all predicate and vector save slots. It's such a rare ++ case that it hardly seems worth the effort though. */ ++ && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca + && frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset)) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_saved_regs] ++ stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_outgoing_args_size; ++ frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -7649,30 +7651,29 @@ aarch64_layout_frame (void) + + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = (frame.hard_fp_offset + + frame.below_hard_fp_saved_regs_size); +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments or SVE saves, but with +- a small local area: ++ /* Frame with large area below the saved registers, or with SVE saves, ++ but with a small area above: + + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else + { +- /* Frame with large local area and outgoing arguments or SVE saves, +- using frame pointer: ++ /* General case: + + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] +@@ -7680,10 +7681,10 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.hard_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + + /* Make sure the individual adjustments add up to the full frame size. */ +@@ -8274,7 +8275,7 @@ aarch64_get_separate_components (void) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -8423,7 +8424,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -8477,7 +8478,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset2 -= frame.below_hard_fp_saved_regs_size; + else +- offset2 += crtl->outgoing_args_size; ++ offset2 += frame.bytes_below_saved_regs; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -8551,10 +8552,10 @@ aarch64_stack_clash_protection_alloca_probe_range (void) + registers. If POLY_SIZE is not large enough to require a probe this function + will only adjust the stack. When allocating the stack space + FRAME_RELATED_P is then used to indicate if the allocation is frame related. +- FINAL_ADJUSTMENT_P indicates whether we are allocating the outgoing +- arguments. If we are then we ensure that any allocation larger than the ABI +- defined buffer needs a probe so that the invariant of having a 1KB buffer is +- maintained. ++ FINAL_ADJUSTMENT_P indicates whether we are allocating the area below ++ the saved registers. If we are then we ensure that any allocation ++ larger than the ABI defined buffer needs a probe so that the ++ invariant of having a 1KB buffer is maintained. + + We emit barriers after each stack adjustment to prevent optimizations from + breaking the invariant that we never drop the stack more than a page. This +@@ -8763,7 +8764,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + /* Handle any residuals. Residuals of at least MIN_PROBE_THRESHOLD have to + be probed. This maintains the requirement that each page is probed at + least once. For initial probing we probe only if the allocation is +- more than GUARD_SIZE - buffer, and for the outgoing arguments we probe ++ more than GUARD_SIZE - buffer, and below the saved registers we probe + if the amount is larger than buffer. GUARD_SIZE - buffer + buffer == + GUARD_SIZE. This works that for any allocation that is large enough to + trigger a probe here, we'll have at least one, and if they're not large +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index bb383acfae86..6f0b8c7107e7 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -837,6 +837,11 @@ struct GTY (()) aarch64_frame + /* The size of the callee-save registers with a slot in REG_OFFSET. */ + poly_int64 saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the bottom of the register save area. ++ This value is always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_saved_regs; ++ + /* The size of the callee-save registers with a slot in REG_OFFSET that + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0015-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/11.4.0/0015-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..3c2bd77d --- /dev/null +++ b/packages/gcc/11.4.0/0015-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,84 @@ +From d3f6ceecc8a7f128a9e6cb7d8aecf0de81ed9705 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:45 +0100 +Subject: [PATCH 15/29] aarch64: Add bytes_below_hard_fp to frame info + +Following on from the previous bytes_below_saved_regs patch, this one +records the number of bytes that are below the hard frame pointer. +This eventually replaces below_hard_fp_saved_regs_size. + +If a frame pointer is not needed, the epilogue adds final_adjust +to the stack pointer before restoring registers: + + aarch64_add_sp (tmp1_rtx, tmp0_rtx, final_adjust, true); + +Therefore, if the epilogue needs to restore the stack pointer from +the hard frame pointer, the directly corresponding offset is: + + -bytes_below_hard_fp + final_adjust + +i.e. go from the hard frame pointer to the bottom of the frame, +then add the same amount as if we were using the stack pointer +from the outset. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it. + (aarch64_expand_epilogue): Use it instead of + below_hard_fp_saved_regs_size. +--- + gcc/config/aarch64/aarch64.c | 6 +++--- + gcc/config/aarch64/aarch64.h | 5 +++++ + 2 files changed, 8 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 8fa5a0b25455..e03adf57226f 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7528,6 +7528,7 @@ aarch64_layout_frame (void) + of the callee save area. */ + bool saves_below_hard_fp_p = maybe_ne (offset, 0); + frame.below_hard_fp_saved_regs_size = offset; ++ frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -9083,8 +9084,7 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 final_adjust = frame.final_adjust; + poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_candidate1; + unsigned reg2 = frame.wb_candidate2; + rtx cfi_ops = NULL; +@@ -9140,7 +9140,7 @@ aarch64_expand_epilogue (bool for_sibcall) + is restored on the instruction doing the writeback. */ + aarch64_add_offset (Pmode, stack_pointer_rtx, + hard_frame_pointer_rtx, +- -callee_offset - below_hard_fp_saved_regs_size, ++ -bytes_below_hard_fp + final_adjust, + tmp1_rtx, tmp0_rtx, callee_adjust == 0); + else + /* The case where we need to re-use the register here is very rare, so +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 6f0b8c7107e7..21ac920a3fe1 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -846,6 +846,11 @@ struct GTY (()) aarch64_frame + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0016-aarch64-Tweak-aarch64_save-restore_callee_saves.patch b/packages/gcc/11.4.0/0016-aarch64-Tweak-aarch64_save-restore_callee_saves.patch new file mode 100644 index 00000000..6367f16b --- /dev/null +++ b/packages/gcc/11.4.0/0016-aarch64-Tweak-aarch64_save-restore_callee_saves.patch @@ -0,0 +1,221 @@ +From e8a7ec87fcdbaa5f7c7bd499aebe5cefacbf8687 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:46 +0100 +Subject: [PATCH 16/29] aarch64: Tweak aarch64_save/restore_callee_saves + +aarch64_save_callee_saves and aarch64_restore_callee_saves took +a parameter called start_offset that gives the offset of the +bottom of the saved register area from the current stack pointer. +However, it's more convenient for later patches if we use the +bottom of the entire frame as the reference point, rather than +the bottom of the saved registers. + +Doing that removes the need for the callee_offset field. +Other than that, this is not a win on its own. It only really +makes sense in combination with the follow-on patches. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::callee_offset): Delete. + * config/aarch64/aarch64.c (aarch64_layout_frame): Remove + callee_offset handling. + (aarch64_save_callee_saves): Replace the start_offset parameter + with a bytes_below_sp parameter. + (aarch64_restore_callee_saves): Likewise. + (aarch64_expand_prologue): Update accordingly. + (aarch64_expand_epilogue): Likewise. +--- + gcc/config/aarch64/aarch64.c | 56 ++++++++++++++++++------------------ + gcc/config/aarch64/aarch64.h | 4 --- + 2 files changed, 28 insertions(+), 32 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index e03adf57226f..96e99f6c17ae 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7602,7 +7602,6 @@ aarch64_layout_frame (void) + frame.final_adjust = 0; + frame.callee_adjust = 0; + frame.sve_callee_adjust = 0; +- frame.callee_offset = 0; + + HOST_WIDE_INT max_push_offset = 0; + if (frame.wb_candidate2 != INVALID_REGNUM) +@@ -7642,7 +7641,6 @@ aarch64_layout_frame (void) + stp reg1, reg2, [sp, bytes_below_saved_regs] + stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -7989,12 +7987,13 @@ aarch64_add_cfa_expression (rtx_insn *insn, rtx reg, + } + + /* Emit code to save the callee-saved registers from register number START +- to LIMIT to the stack at the location starting at offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. HARD_FP_VALID_P +- is true if the hard frame pointer has been set up. */ ++ to LIMIT to the stack. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. HARD_FP_VALID_P is true if the hard ++ frame pointer has been set up. */ + + static void +-aarch64_save_callee_saves (poly_int64 start_offset, ++aarch64_save_callee_saves (poly_int64 bytes_below_sp, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { +@@ -8022,7 +8021,9 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -8033,9 +8034,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + else if (GP_REGNUM_P (regno) + && (!offset.is_constant (&const_offset) || const_offset >= 512)) + { +- gcc_assert (known_eq (start_offset, 0)); +- poly_int64 fp_offset +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 fp_offset = frame.bytes_below_hard_fp - bytes_below_sp; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -8099,12 +8098,13 @@ aarch64_save_callee_saves (poly_int64 start_offset, + } + + /* Emit code to restore the callee registers from register number START +- up to and including LIMIT. Restore from the stack offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. Write the +- appropriate REG_CFA_RESTORE notes into CFI_OPS. */ ++ up to and including LIMIT. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. Write the appropriate REG_CFA_RESTORE ++ notes into CFI_OPS. */ + + static void +-aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, ++aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { + aarch64_frame &frame = cfun->machine->frame; +@@ -8130,7 +8130,9 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -8906,8 +8908,6 @@ aarch64_expand_prologue (void) + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; + unsigned reg1 = frame.wb_candidate1; + unsigned reg2 = frame.wb_candidate2; + bool emit_frame_chain = frame.emit_frame_chain; +@@ -8979,8 +8979,8 @@ aarch64_expand_prologue (void) + - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + +- /* The offset of the bottom of the save area from the current SP. */ +- poly_int64 saved_regs_offset = chain_offset - below_hard_fp_saved_regs_size; ++ /* The offset of the current SP from the bottom of the static frame. */ ++ poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { +@@ -8988,7 +8988,7 @@ aarch64_expand_prologue (void) + { + reg1 = R29_REGNUM; + reg2 = R30_REGNUM; +- aarch64_save_callee_saves (saved_regs_offset, reg1, reg2, ++ aarch64_save_callee_saves (bytes_below_sp, reg1, reg2, + false, false); + } + else +@@ -9028,7 +9028,7 @@ aarch64_expand_prologue (void) + emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + +- aarch64_save_callee_saves (saved_regs_offset, R0_REGNUM, R30_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, R0_REGNUM, R30_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + if (maybe_ne (sve_callee_adjust, 0)) +@@ -9038,16 +9038,17 @@ aarch64_expand_prologue (void) + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, + sve_callee_adjust, + !frame_pointer_needed, false); +- saved_regs_offset += sve_callee_adjust; ++ bytes_below_sp -= sve_callee_adjust; + } +- aarch64_save_callee_saves (saved_regs_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, P0_REGNUM, P15_REGNUM, + false, emit_frame_chain); +- aarch64_save_callee_saves (saved_regs_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, V0_REGNUM, V31_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + + /* We may need to probe the final adjustment if it is larger than the guard + that is assumed by the called. */ ++ gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); + } +@@ -9082,7 +9083,6 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_candidate1; +@@ -9150,13 +9150,13 @@ aarch64_expand_epilogue (bool for_sibcall) + + /* Restore the vector registers before the predicate registers, + so that we can use P4 as a temporary for big-endian SVE frames. */ +- aarch64_restore_callee_saves (callee_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, V0_REGNUM, V31_REGNUM, + callee_adjust != 0, &cfi_ops); +- aarch64_restore_callee_saves (callee_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, P0_REGNUM, P15_REGNUM, + false, &cfi_ops); + if (maybe_ne (sve_callee_adjust, 0)) + aarch64_add_sp (NULL_RTX, NULL_RTX, sve_callee_adjust, true); +- aarch64_restore_callee_saves (callee_offset - sve_callee_adjust, ++ aarch64_restore_callee_saves (final_adjust + sve_callee_adjust, + R0_REGNUM, R30_REGNUM, + callee_adjust != 0, &cfi_ops); + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 21ac920a3fe1..57e672177456 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -873,10 +873,6 @@ struct GTY (()) aarch64_frame + It is zero when no push is used. */ + HOST_WIDE_INT callee_adjust; + +- /* The offset from SP to the callee-save registers after initial_adjust. +- It may be non-zero if no push is used (ie. callee_adjust == 0). */ +- poly_int64 callee_offset; +- + /* The size of the stack adjustment before saving or after restoring + SVE registers. */ + poly_int64 sve_callee_adjust; +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0017-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch b/packages/gcc/11.4.0/0017-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch new file mode 100644 index 00000000..8a50afce --- /dev/null +++ b/packages/gcc/11.4.0/0017-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch @@ -0,0 +1,45 @@ +From 7356df0319aefe4c68ef57ec4c6bd18c72188a34 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:46 +0100 +Subject: [PATCH 17/29] aarch64: Only calculate chain_offset if there is a + chain + +After previous patches, it is no longer necessary to calculate +a chain_offset in cases where there is no chain record. + +gcc/ + * config/aarch64/aarch64.c (aarch64_expand_prologue): Move the + calculation of chain_offset into the emit_frame_chain block. +--- + gcc/config/aarch64/aarch64.c | 10 +++++----- + 1 file changed, 5 insertions(+), 5 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 96e99f6c17ae..cf5244b7ec08 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8974,16 +8974,16 @@ aarch64_expand_prologue (void) + if (callee_adjust != 0) + aarch64_push_regs (reg1, reg2, callee_adjust); + +- /* The offset of the frame chain record (if any) from the current SP. */ +- poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); +- gcc_assert (known_ge (chain_offset, 0)); +- + /* The offset of the current SP from the bottom of the static frame. */ + poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { ++ /* The offset of the frame chain record (if any) from the current SP. */ ++ poly_int64 chain_offset = (initial_adjust + callee_adjust ++ - frame.hard_fp_offset); ++ gcc_assert (known_ge (chain_offset, 0)); ++ + if (callee_adjust == 0) + { + reg1 = R29_REGNUM; +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0018-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/11.4.0/0018-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..7c989274 --- /dev/null +++ b/packages/gcc/11.4.0/0018-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,91 @@ +From 82fb69e75c21010f7afc72bb842751164fe8fc72 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:46 +0100 +Subject: [PATCH 18/29] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 6 +++--- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 6 insertions(+), 6 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index cf5244b7ec08..d54f7a89306f 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7596,7 +7596,7 @@ aarch64_layout_frame (void) + STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + +- frame.locals_offset = frame.saved_varargs_size; ++ frame.bytes_above_locals = frame.saved_varargs_size; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +@@ -11758,13 +11758,13 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.locals_offset; ++ return frame.hard_fp_offset - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return frame.frame_size - frame.locals_offset; ++ return frame.frame_size - frame.bytes_above_locals; + } + + return frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 57e672177456..3c5e3dd429dc 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -851,10 +851,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 locals_offset; ++ poly_int64 bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0019-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/11.4.0/0019-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..e4c47306 --- /dev/null +++ b/packages/gcc/11.4.0/0019-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,148 @@ +From fa6600b55b49ee14d8288f13719ceea2a75eea60 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:47 +0100 +Subject: [PATCH 19/29] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.c | 26 +++++++++++++------------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 16 insertions(+), 16 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index d54f7a89306f..23cb084e5a75 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7588,7 +7588,7 @@ aarch64_layout_frame (void) + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.hard_fp_offset ++ frame.bytes_above_hard_fp + = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +@@ -7609,13 +7609,13 @@ aarch64_layout_frame (void) + else if (frame.wb_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ && known_eq (frame.bytes_above_hard_fp, const_size)) + { + /* Simple, small frame with no data below the saved registers. + +@@ -7632,8 +7632,8 @@ aarch64_layout_frame (void) + case that it hardly seems worth the effort though. */ + && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca +- && frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset)) ++ && frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset)) + { + /* Frame with small area below the saved registers: + +@@ -7651,12 +7651,12 @@ aarch64_layout_frame (void) + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.hard_fp_offset ++ frame.initial_adjust = (frame.bytes_above_hard_fp + + frame.below_hard_fp_saved_regs_size); + frame.final_adjust = frame.bytes_below_saved_regs; + } +- else if (frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset) ++ else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset) + { + /* Frame with large area below the saved registers, or with SVE saves, + but with a small area above: +@@ -7666,7 +7666,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.callee_adjust = const_fp_offset; ++ frame.callee_adjust = const_above_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -7681,7 +7681,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = frame.hard_fp_offset; ++ frame.initial_adjust = frame.bytes_above_hard_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -8981,7 +8981,7 @@ aarch64_expand_prologue (void) + { + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); ++ - frame.bytes_above_hard_fp); + gcc_assert (known_ge (chain_offset, 0)); + + if (callee_adjust == 0) +@@ -11755,10 +11755,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return frame.hard_fp_offset; ++ return frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.bytes_above_locals; ++ return frame.bytes_above_hard_fp - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 3c5e3dd429dc..9291cfd3ec8b 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -856,10 +856,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 hard_fp_offset; ++ poly_int64 bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0020-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/11.4.0/0020-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..d9f68fd2 --- /dev/null +++ b/packages/gcc/11.4.0/0020-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,35 @@ +From b8cd5a0229da78c2d1289d54731fbef0126617d5 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:47 +0100 +Subject: [PATCH 20/29] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 9291cfd3ec8b..82883ad5a0d8 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -861,8 +861,8 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- frame (incomming SP) to the stack_pointer. This value is always ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always + a multiple of STACK_BOUNDARY. */ + poly_int64 frame_size; + +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0021-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch b/packages/gcc/11.4.0/0021-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch new file mode 100644 index 00000000..b0a2d328 --- /dev/null +++ b/packages/gcc/11.4.0/0021-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch @@ -0,0 +1,196 @@ +From 999c4a81cffddb850d6ab0f6d3a8de3e704d2f7a Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:48 +0100 +Subject: [PATCH 21/29] aarch64: Measure reg_offset from the bottom of the + frame + +reg_offset was measured from the bottom of the saved register area. +This made perfect sense with the original layout, since the bottom +of the saved register area was also the hard frame pointer address. +It became slightly less obvious with SVE, since we save SVE +registers below the hard frame pointer, but it still made sense. + +However, if we want to allow different frame layouts, it's more +convenient and obvious to measure reg_offset from the bottom of +the frame. After previous patches, it's also a slight simplification +in its own right. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame): Add comment above + reg_offset. + * config/aarch64/aarch64.c (aarch64_layout_frame): Walk offsets + from the bottom of the frame, rather than the bottom of the saved + register area. Measure reg_offset from the bottom of the frame + rather than the bottom of the saved register area. + (aarch64_save_callee_saves): Update accordingly. + (aarch64_restore_callee_saves): Likewise. + (aarch64_get_separate_components): Likewise. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 53 ++++++++++++++++-------------------- + gcc/config/aarch64/aarch64.h | 3 ++ + 2 files changed, 27 insertions(+), 29 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 23cb084e5a75..45ff664cba6b 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7398,7 +7398,6 @@ aarch64_needs_frame_chain (void) + static void + aarch64_layout_frame (void) + { +- poly_int64 offset = 0; + int regno, last_fp_reg = INVALID_REGNUM; + machine_mode vector_save_mode = aarch64_reg_save_mode (V8_REGNUM); + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); +@@ -7476,7 +7475,9 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + +- frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ poly_int64 offset = crtl->outgoing_args_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.bytes_below_saved_regs = offset; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -7488,7 +7489,8 @@ aarch64_layout_frame (void) + offset += BYTES_PER_SVE_PRED; + } + +- if (maybe_ne (offset, 0)) ++ poly_int64 saved_prs_size = offset - frame.bytes_below_saved_regs; ++ if (maybe_ne (saved_prs_size, 0)) + { + /* If we have any vector registers to save above the predicate registers, + the offset of the vector register save slots need to be a multiple +@@ -7506,10 +7508,10 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + else + { +- if (known_le (offset, vector_save_size)) +- offset = vector_save_size; +- else if (known_le (offset, vector_save_size * 2)) +- offset = vector_save_size * 2; ++ if (known_le (saved_prs_size, vector_save_size)) ++ offset = frame.bytes_below_saved_regs + vector_save_size; ++ else if (known_le (saved_prs_size, vector_save_size * 2)) ++ offset = frame.bytes_below_saved_regs + vector_save_size * 2; + else + gcc_unreachable (); + } +@@ -7526,9 +7528,10 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- bool saves_below_hard_fp_p = maybe_ne (offset, 0); +- frame.below_hard_fp_saved_regs_size = offset; +- frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; ++ frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p ++ = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ frame.bytes_below_hard_fp = offset; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -7579,9 +7582,10 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset; ++ frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; ++ poly_int64 varargs_and_saved_regs_size ++ = frame.saved_regs_size + frame.saved_varargs_size; + + poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size +@@ -8021,9 +8025,7 @@ aarch64_save_callee_saves (poly_int64 bytes_below_sp, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -8130,9 +8132,7 @@ aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -8271,14 +8271,12 @@ aarch64_get_separate_components (void) + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection + && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, 0)) ++ && known_eq (offset, frame.bytes_below_saved_regs)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -8425,9 +8423,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -8479,9 +8475,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= frame.below_hard_fp_saved_regs_size; +- else +- offset2 += frame.bytes_below_saved_regs; ++ offset2 -= frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -8597,7 +8591,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p + && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] ++ - frame.bytes_below_saved_regs); + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 82883ad5a0d8..c8ec3d584957 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -826,6 +826,9 @@ extern enum aarch64_processor aarch64_tune; + #ifdef HAVE_POLY_INT_H + struct GTY (()) aarch64_frame + { ++ /* The offset from the bottom of the static frame (the bottom of the ++ outgoing arguments) of each register save slot, or -2 if no save is ++ needed. */ + poly_int64 reg_offset[LAST_SAVED_REGNUM + 1]; + + /* The number of extra stack bytes taken up by register varargs. +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0022-aarch64-Simplify-top-of-frame-allocation.patch b/packages/gcc/11.4.0/0022-aarch64-Simplify-top-of-frame-allocation.patch new file mode 100644 index 00000000..0f50f4d0 --- /dev/null +++ b/packages/gcc/11.4.0/0022-aarch64-Simplify-top-of-frame-allocation.patch @@ -0,0 +1,55 @@ +From 8b664cc8f05c8130e8ca73a59ae2751cdef8a0ea Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:48 +0100 +Subject: [PATCH 22/29] aarch64: Simplify top of frame allocation + +After previous patches, it no longer really makes sense to allocate +the top of the frame in terms of varargs_and_saved_regs_size and +saved_regs_and_above. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Simplify + the allocation of the top of the frame. +--- + gcc/config/aarch64/aarch64.c | 23 ++++++++--------------- + 1 file changed, 8 insertions(+), 15 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 45ff664cba6b..779547d03447 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7584,23 +7584,16 @@ aarch64_layout_frame (void) + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size +- = frame.saved_regs_size + frame.saved_varargs_size; ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ auto top_of_locals = offset; + +- poly_int64 saved_regs_and_above +- = aligned_upper_bound (varargs_and_saved_regs_size +- + get_frame_size (), +- STACK_BOUNDARY / BITS_PER_UNIT); ++ offset += frame.saved_varargs_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.frame_size = offset; + +- frame.bytes_above_hard_fp +- = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; +- +- /* Both these values are already aligned. */ +- gcc_assert (multiple_p (frame.bytes_below_saved_regs, +- STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; +- +- frame.bytes_above_locals = frame.saved_varargs_size; ++ frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0023-aarch64-Minor-initial-adjustment-tweak.patch b/packages/gcc/11.4.0/0023-aarch64-Minor-initial-adjustment-tweak.patch new file mode 100644 index 00000000..89ee42e6 --- /dev/null +++ b/packages/gcc/11.4.0/0023-aarch64-Minor-initial-adjustment-tweak.patch @@ -0,0 +1,38 @@ +From bb4600071acc3a02db4f37ffb95c8495ad76a140 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:49 +0100 +Subject: [PATCH 23/29] aarch64: Minor initial adjustment tweak + +This patch just changes a calculation of initial_adjust +to one that makes it slightly more obvious that the total +adjustment is frame.frame_size. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Tweak + calculation of initial_adjust for frames in which all saves + are SVE saves. +--- + gcc/config/aarch64/aarch64.c | 5 ++--- + 1 file changed, 2 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 779547d03447..0b8992ada747 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7645,11 +7645,10 @@ aarch64_layout_frame (void) + { + /* Frame in which all saves are SVE saves: + +- sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size ++ sub sp, sp, frame_size - bytes_below_saved_regs + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.bytes_above_hard_fp +- + frame.below_hard_fp_saved_regs_size); ++ frame.initial_adjust = frame.frame_size - frame.bytes_below_saved_regs; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0024-aarch64-Tweak-stack-clash-boundary-condition.patch b/packages/gcc/11.4.0/0024-aarch64-Tweak-stack-clash-boundary-condition.patch new file mode 100644 index 00000000..eb77ea72 --- /dev/null +++ b/packages/gcc/11.4.0/0024-aarch64-Tweak-stack-clash-boundary-condition.patch @@ -0,0 +1,125 @@ +From f22329d5efbacf80edf4a2d45ebadd93f283252c Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:49 +0100 +Subject: [PATCH 24/29] aarch64: Tweak stack clash boundary condition + +The AArch64 ABI says that, when stack clash protection is used, +there can be a maximum of 1KiB of unprobed space at sp on entry +to a function. Therefore, we need to probe when allocating +>= guard_size - 1KiB of data (>= rather than >). This is what +GCC does. + +If an allocation is exactly guard_size bytes, it is enough to allocate +those bytes and probe once at offset 1024. It isn't possible to use a +single probe at any other offset: higher would conmplicate later code, +by leaving more unprobed space than usual, while lower would risk +leaving an entire page unprobed. For simplicity, the code probes all +allocations at offset 1024. + +Some register saves also act as probes. If we need to allocate +more space below the last such register save probe, we need to +probe the allocation if it is > 1KiB. Again, this allocation is +then sometimes (but not always) probed at offset 1024. This sort of +allocation is currently only used for outgoing arguments, which are +rarely this big. + +However, the code also probed if this final outgoing-arguments +allocation was == 1KiB, rather than just > 1KiB. This isn't +necessary, since the register save then probes at offset 1024 +as required. Continuing to probe allocations of exactly 1KiB +would complicate later patches. + +gcc/ + * config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): + Don't probe final allocations that are exactly 1KiB in size (after + unprobed space above the final allocation has been deducted). + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: New test. +--- + gcc/config/aarch64/aarch64.c | 4 +- + .../aarch64/stack-check-prologue-17.c | 55 +++++++++++++++++++ + 2 files changed, 58 insertions(+), 1 deletion(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 0b8992ada747..bfd248761951 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8564,9 +8564,11 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; ++ HOST_WIDE_INT byte_sp_alignment = STACK_BOUNDARY / BITS_PER_UNIT; ++ gcc_assert (multiple_p (poly_size, byte_sp_alignment)); + HOST_WIDE_INT min_probe_threshold + = (final_adjustment_p +- ? guard_used_by_caller ++ ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); + /* When doing the final adjustment for the outgoing arguments, take into + account any unprobed space there is above the current SP. There are +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +new file mode 100644 +index 000000000000..0d8a25d73a24 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -0,0 +1,55 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0025-aarch64-Put-LR-save-probe-in-first-16-bytes.patch b/packages/gcc/11.4.0/0025-aarch64-Put-LR-save-probe-in-first-16-bytes.patch new file mode 100644 index 00000000..ca10742e --- /dev/null +++ b/packages/gcc/11.4.0/0025-aarch64-Put-LR-save-probe-in-first-16-bytes.patch @@ -0,0 +1,263 @@ +From 174a9747491e591ef2abb3e20a0332303f11003a Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:49 +0100 +Subject: [PATCH 25/29] aarch64: Put LR save probe in first 16 bytes + +-fstack-clash-protection uses the save of LR as a probe for the next +allocation. The next allocation could be: + +* another part of the static frame, e.g. when allocating SVE save slots + or outgoing arguments + +* an alloca in the same function + +* an allocation made by a callee function + +However, when -fomit-frame-pointer is used, the LR save slot is placed +above the other GPR save slots. It could therefore be up to 80 bytes +above the base of the GPR save area (which is also the hard fp address). + +aarch64_allocate_and_probe_stack_space took this into account when +deciding how much subsequent space could be allocated without needing +a probe. However, it interacted badly with: + + /* If doing a small final adjustment, we always probe at offset 0. + This is done to avoid issues when LR is not at position 0 or when + the final adjustment is smaller than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +which forces any allocation that is smaller than the guard page size +to be probed at offset 0 rather than the usual offset 1024. It was +therefore possible to construct cases in which we had: + +* a probe using LR at SP + 80 bytes (or some other value >= 16) +* an allocation of the guard page size - 16 bytes +* a probe at SP + 0 + +which allocates guard page size + 64 consecutive unprobed bytes. + +This patch requires the LR probe to be in the first 16 bytes of the +save area when stack clash protection is active. Doing it +unconditionally would cause code-quality regressions, but a later +patch deals with that. + +The new comment doesn't say that the probe register is required +to be LR, since a later patch removes that restriction. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Ensure that + the LR save slot is in the first 16 bytes of the register save area. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + when LR was not in the first 16 bytes. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-18.c: New test. +--- + gcc/config/aarch64/aarch64.c | 61 ++++------- + .../aarch64/stack-check-prologue-18.c | 100 ++++++++++++++++++ + 2 files changed, 123 insertions(+), 38 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index bfd248761951..3f2b10de987d 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7532,26 +7532,34 @@ aarch64_layout_frame (void) + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); + frame.bytes_below_hard_fp = offset; ++ ++ auto allocate_gpr_slot = [&](unsigned int regno) ++ { ++ frame.reg_offset[regno] = offset; ++ if (frame.wb_candidate1 == INVALID_REGNUM) ++ frame.wb_candidate1 = regno; ++ else if (frame.wb_candidate2 == INVALID_REGNUM) ++ frame.wb_candidate2 = regno; ++ offset += UNITS_PER_WORD; ++ }; ++ + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +- frame.reg_offset[R29_REGNUM] = offset; +- frame.wb_candidate1 = R29_REGNUM; +- frame.reg_offset[R30_REGNUM] = offset + UNITS_PER_WORD; +- frame.wb_candidate2 = R30_REGNUM; +- offset += 2 * UNITS_PER_WORD; ++ allocate_gpr_slot (R29_REGNUM); ++ allocate_gpr_slot (R30_REGNUM); + } ++ else if (flag_stack_clash_protection ++ && known_eq (frame.reg_offset[R30_REGNUM], SLOT_REQUIRED)) ++ /* Put the LR save slot first, since it makes a good choice of probe ++ for stack clash purposes. The idea is that the link register usually ++ has to be saved before a call anyway, and so we lose little by ++ stopping it from being individually shrink-wrapped. */ ++ allocate_gpr_slot (R30_REGNUM); + + for (regno = R0_REGNUM; regno <= R30_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) +- { +- frame.reg_offset[regno] = offset; +- if (frame.wb_candidate1 == INVALID_REGNUM) +- frame.wb_candidate1 = regno; +- else if (frame.wb_candidate2 == INVALID_REGNUM) +- frame.wb_candidate2 = regno; +- offset += UNITS_PER_WORD; +- } ++ allocate_gpr_slot (regno); + + poly_int64 max_int_offset = offset; + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8570,29 +8578,6 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + = (final_adjustment_p + ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); +- /* When doing the final adjustment for the outgoing arguments, take into +- account any unprobed space there is above the current SP. There are +- two cases: +- +- - When saving SVE registers below the hard frame pointer, we force +- the lowest save to take place in the prologue before doing the final +- adjustment (i.e. we don't allow the save to be shrink-wrapped). +- This acts as a probe at SP, so there is no unprobed space. +- +- - When there are no SVE register saves, we use the store of the link +- register as a probe. We can't assume that LR was saved at position 0 +- though, so treat any space below it as unprobed. */ +- if (final_adjustment_p +- && known_eq (frame.below_hard_fp_saved_regs_size, 0)) +- { +- poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] +- - frame.bytes_below_saved_regs); +- if (known_ge (lr_offset, 0)) +- min_probe_threshold -= lr_offset.to_constant (); +- else +- gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); +- } +- + poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ +@@ -8772,8 +8757,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; + /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when LR is not at position 0 or when +- the final adjustment is smaller than the probing offset. */ ++ This is done to avoid issues when the final adjustment is smaller ++ than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +new file mode 100644 +index 000000000000..82447d20fff5 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -0,0 +1,100 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0026-aarch64-Simplify-probe-of-final-frame-allocation.patch b/packages/gcc/11.4.0/0026-aarch64-Simplify-probe-of-final-frame-allocation.patch new file mode 100644 index 00000000..f3cfd734 --- /dev/null +++ b/packages/gcc/11.4.0/0026-aarch64-Simplify-probe-of-final-frame-allocation.patch @@ -0,0 +1,99 @@ +From e932e11c353be52256dd30d30d924f4e834e3ca3 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:51 +0100 +Subject: [PATCH 26/29] aarch64: Simplify probe of final frame allocation + +Previous patches ensured that the final frame allocation only needs +a probe when the size is strictly greater than 1KiB. It's therefore +safe to use the normal 1024 probe offset in all cases. + +The main motivation for doing this is to simplify the code and +remove the number of special cases. + +gcc/ + * config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): + Always probe the residual allocation at offset 1024, asserting + that that is in range. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: Expect the probe + to be at offset 1024 rather than offset 0. + * gcc.target/aarch64/stack-check-prologue-18.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 12 ++++-------- + .../gcc.target/aarch64/stack-check-prologue-17.c | 2 +- + .../gcc.target/aarch64/stack-check-prologue-18.c | 4 ++-- + 3 files changed, 7 insertions(+), 11 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 3f2b10de987d..4b9cd687525e 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -8751,16 +8751,12 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + are still safe. */ + if (residual) + { +- HOST_WIDE_INT residual_probe_offset = guard_used_by_caller; ++ gcc_assert (guard_used_by_caller + byte_sp_alignment <= size); ++ + /* If we're doing final adjustments, and we've done any full page + allocations then any residual needs to be probed. */ + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; +- /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when the final adjustment is smaller +- than the probing offset. */ +- else if (final_adjustment_p && rounded_size == 0) +- residual_probe_offset = 0; + + aarch64_sub_sp (temp1, temp2, residual, frame_related_p); + if (residual >= min_probe_threshold) +@@ -8771,8 +8767,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT_PRINT_DEC " bytes, probing will be required." + "\n", residual); + +- emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, +- residual_probe_offset)); ++ emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, ++ guard_used_by_caller)); + emit_insn (gen_blockage ()); + } + } +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +index 0d8a25d73a24..f0ec1389771d 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -33,7 +33,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +index 82447d20fff5..6383bec5ebcd 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -9,7 +9,7 @@ void g(); + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -50,7 +50,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0027-aarch64-Explicitly-record-probe-registers-in-frame-i.patch b/packages/gcc/11.4.0/0027-aarch64-Explicitly-record-probe-registers-in-frame-i.patch new file mode 100644 index 00000000..29e673a6 --- /dev/null +++ b/packages/gcc/11.4.0/0027-aarch64-Explicitly-record-probe-registers-in-frame-i.patch @@ -0,0 +1,278 @@ +From 9ed9fd54b2b471745c9489e83496c091a7b64904 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:52 +0100 +Subject: [PATCH 27/29] aarch64: Explicitly record probe registers in frame + info + +The stack frame is currently divided into three areas: + +A: the area above the hard frame pointer +B: the SVE saves below the hard frame pointer +C: the outgoing arguments + +If the stack frame is allocated in one chunk, the allocation needs a +probe if the frame size is >= guard_size - 1KiB. In addition, if the +function is not a leaf function, it must probe an address no more than +1KiB above the outgoing SP. We ensured the second condition by + +(1) using single-chunk allocations for non-leaf functions only if + the link register save slot is within 512 bytes of the bottom + of the frame; and + +(2) using the link register save as a probe (meaning, for instance, + that it can't be individually shrink wrapped) + +If instead the stack is allocated in multiple chunks, then: + +* an allocation involving only the outgoing arguments (C above) requires + a probe if the allocation size is > 1KiB + +* any other allocation requires a probe if the allocation size + is >= guard_size - 1KiB + +* second and subsequent allocations require the previous allocation + to probe at the bottom of the allocated area, regardless of the size + of that previous allocation + +The final point means that, unlike for single allocations, +it can be necessary to have both a non-SVE register probe and +an SVE register probe. For example: + +* allocate A, probe using a non-SVE register save +* allocate B, probe using an SVE register save +* allocate C + +The non-SVE register used in this case was again the link register. +It was previously used even if the link register save slot was some +bytes above the bottom of the non-SVE register saves, but an earlier +patch avoided that by putting the link register save slot first. + +As a belt-and-braces fix, this patch explicitly records which +probe registers we're using and allows the non-SVE probe to be +whichever register comes first (as for SVE). + +The patch also avoids unnecessary probes in sve/pcs/stack_clash_3.c. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::sve_save_and_probe) + (aarch64_frame::hard_fp_save_and_probe): New fields. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize them. + Rather than asserting that a leaf function saves LR, instead assert + that a leaf function saves something. + (aarch64_get_separate_components): Prevent the chosen probe + registers from being individually shrink-wrapped. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + probe registers that aren't at the bottom of the previous allocation. + +gcc/testsuite/ + * gcc.target/aarch64/sve/pcs/stack_clash_3.c: Avoid redundant probes. +--- + gcc/config/aarch64/aarch64.c | 68 +++++++++++++++---- + gcc/config/aarch64/aarch64.h | 8 +++ + .../aarch64/sve/pcs/stack_clash_3.c | 6 +- + 3 files changed, 64 insertions(+), 18 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 4b9cd687525e..ef4b3b671ba2 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7469,15 +7469,11 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + +- /* With stack-clash, LR must be saved in non-leaf functions. The saving of +- LR counts as an implicit probe which allows us to maintain the invariant +- described in the comment at expand_prologue. */ +- gcc_assert (crtl->is_leaf +- || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.bytes_below_saved_regs = offset; ++ frame.sve_save_and_probe = INVALID_REGNUM; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -7485,6 +7481,8 @@ aarch64_layout_frame (void) + for (regno = P0_REGNUM; regno <= P15_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += BYTES_PER_SVE_PRED; + } +@@ -7522,6 +7520,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += vector_save_size; + } +@@ -7531,10 +7531,18 @@ aarch64_layout_frame (void) + frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ gcc_assert (!saves_below_hard_fp_p ++ || (frame.sve_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.sve_save_and_probe], ++ frame.bytes_below_saved_regs))); ++ + frame.bytes_below_hard_fp = offset; ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; + + auto allocate_gpr_slot = [&](unsigned int regno) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + frame.reg_offset[regno] = offset; + if (frame.wb_candidate1 == INVALID_REGNUM) + frame.wb_candidate1 = regno; +@@ -7568,6 +7576,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + /* If there is an alignment gap between integer and fp callee-saves, + allocate the last fp register to it if possible. */ + if (regno == last_fp_reg +@@ -7591,6 +7601,17 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (frame.saved_regs_size, ++ frame.below_hard_fp_saved_regs_size) ++ || (frame.hard_fp_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], ++ frame.bytes_below_hard_fp))); ++ ++ /* With stack-clash, a register must be saved in non-leaf functions. ++ The saving of the bottommost register counts as an implicit probe, ++ which allows us to maintain the invariant described in the comment ++ at expand_prologue. */ ++ gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -7690,6 +7711,25 @@ aarch64_layout_frame (void) + frame.final_adjust = frame.bytes_below_saved_regs; + } + ++ /* The frame is allocated in pieces, with each non-final piece ++ including a register save at offset 0 that acts as a probe for ++ the following piece. In addition, the save of the bottommost register ++ acts as a probe for callees and allocas. Roll back any probes that ++ aren't needed. ++ ++ A probe isn't needed if it is associated with the final allocation ++ (including callees and allocas) that happens before the epilogue is ++ executed. */ ++ if (crtl->is_leaf ++ && !cfun->calls_alloca ++ && known_eq (frame.final_adjust, 0)) ++ { ++ if (maybe_ne (frame.sve_callee_adjust, 0)) ++ frame.sve_save_and_probe = INVALID_REGNUM; ++ else ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; ++ } ++ + /* Make sure the individual adjustments add up to the full frame size. */ + gcc_assert (known_eq (frame.initial_adjust + + frame.callee_adjust +@@ -8267,13 +8307,6 @@ aarch64_get_separate_components (void) + + poly_int64 offset = frame.reg_offset[regno]; + +- /* If the register is saved in the first SVE save slot, we use +- it as a stack probe for -fstack-clash-protection. */ +- if (flag_stack_clash_protection +- && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, frame.bytes_below_saved_regs)) +- continue; +- + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) + offset -= frame.bytes_below_hard_fp; +@@ -8308,6 +8341,13 @@ aarch64_get_separate_components (void) + + bitmap_clear_bit (components, LR_REGNUM); + bitmap_clear_bit (components, SP_REGNUM); ++ if (flag_stack_clash_protection) ++ { ++ if (frame.sve_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.sve_save_and_probe); ++ if (frame.hard_fp_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.hard_fp_save_and_probe); ++ } + + return components; + } +@@ -8844,8 +8884,8 @@ aarch64_epilogue_uses (int regno) + When probing is needed, we emit a probe at the start of the prologue + and every PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE bytes thereafter. + +- We have to track how much space has been allocated and the only stores +- to the stack we track as implicit probes are the FP/LR stores. ++ We can also use register saves as probes. These are stored in ++ sve_save_and_probe and hard_fp_save_and_probe. + + For outgoing arguments we probe if the size is larger than 1KB, such that + the ABI specified buffer is maintained for the next callee. +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index c8ec3d584957..97173e48598b 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -911,6 +911,14 @@ struct GTY (()) aarch64_frame + This is the register they should use. */ + unsigned spare_pred_reg; + ++ /* An SVE register that is saved below the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned sve_save_and_probe; ++ ++ /* A register that is saved at the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned hard_fp_save_and_probe; ++ + bool laid_out; + }; + +diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +index 3e01ec36c3a4..3530a0d504ba 100644 +--- a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c ++++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +@@ -11,11 +11,10 @@ + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +@@ -39,13 +38,12 @@ test_1 (int n) + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** str p5, \[sp, #1, mul vl\] + ** str p6, \[sp, #2, mul vl\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0028-aarch64-Remove-below_hard_fp_saved_regs_size.patch b/packages/gcc/11.4.0/0028-aarch64-Remove-below_hard_fp_saved_regs_size.patch new file mode 100644 index 00000000..6019b6ea --- /dev/null +++ b/packages/gcc/11.4.0/0028-aarch64-Remove-below_hard_fp_saved_regs_size.patch @@ -0,0 +1,157 @@ +From 4bbf7b6cdd02b0d547ddd6a630f2065680bf2f6b Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:52 +0100 +Subject: [PATCH 28/29] aarch64: Remove below_hard_fp_saved_regs_size + +After previous patches, it's no longer necessary to store +saved_regs_size and below_hard_fp_saved_regs_size in the frame info. +All measurements instead use the top or bottom of the frame as +reference points. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::saved_regs_size) + (aarch64_frame::below_hard_fp_saved_regs_size): Delete. + * config/aarch64/aarch64.c (aarch64_layout_frame): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 45 +++++++++++++++++------------------- + gcc/config/aarch64/aarch64.h | 7 ------ + 2 files changed, 21 insertions(+), 31 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index ef4b3b671ba2..385718a475b6 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7528,9 +7528,8 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; +- bool saves_below_hard_fp_p +- = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ auto below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p = maybe_ne (below_hard_fp_saved_regs_size, 0); + gcc_assert (!saves_below_hard_fp_p + || (frame.sve_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.sve_save_and_probe], +@@ -7600,9 +7599,8 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset - frame.bytes_below_saved_regs; +- gcc_assert (known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size) ++ auto saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (saved_regs_size, below_hard_fp_saved_regs_size) + || (frame.hard_fp_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], + frame.bytes_below_hard_fp))); +@@ -7611,7 +7609,7 @@ aarch64_layout_frame (void) + The saving of the bottommost register counts as an implicit probe, + which allows us to maintain the invariant described in the comment + at expand_prologue. */ +- gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); ++ gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -7637,7 +7635,7 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; +- if (known_eq (frame.saved_regs_size, 0)) ++ if (known_eq (saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +@@ -7650,7 +7648,7 @@ aarch64_layout_frame (void) + frame.callee_adjust = const_size; + } + else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) +- && frame.saved_regs_size.is_constant (&const_saved_regs_size) ++ && saved_regs_size.is_constant (&const_saved_regs_size) + && const_below_saved_regs + const_saved_regs_size < 512 + /* We could handle this case even with data below the saved + registers, provided that that data left us with valid offsets +@@ -7669,8 +7667,7 @@ aarch64_layout_frame (void) + frame.initial_adjust = frame.frame_size; + } + else if (saves_below_hard_fp_p +- && known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size)) ++ && known_eq (saved_regs_size, below_hard_fp_saved_regs_size)) + { + /* Frame in which all saves are SVE saves: + +@@ -7692,7 +7689,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_above_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else +@@ -7707,7 +7704,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.bytes_above_hard_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + +@@ -8849,17 +8846,17 @@ aarch64_epilogue_uses (int regno) + | local variables | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | \ +- +-------------------------------+ | +- | callee-saved registers | | frame.saved_regs_size +- +-------------------------------+ | +- | LR' | | +- +-------------------------------+ | +- | FP' | | +- +-------------------------------+ |<- hard_frame_pointer_rtx (aligned) +- | SVE vector registers | | \ +- +-------------------------------+ | | below_hard_fp_saved_regs_size +- | SVE predicate registers | / / ++ | padding | ++ +-------------------------------+ ++ | callee-saved registers | ++ +-------------------------------+ ++ | LR' | ++ +-------------------------------+ ++ | FP' | ++ +-------------------------------+ <-- hard_frame_pointer_rtx (aligned) ++ | SVE vector registers | ++ +-------------------------------+ ++ | SVE predicate registers | + +-------------------------------+ + | dynamic allocation | + +-------------------------------+ +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 97173e48598b..9084b1cfb9d0 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -837,18 +837,11 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + HOST_WIDE_INT saved_varargs_size; + +- /* The size of the callee-save registers with a slot in REG_OFFSET. */ +- poly_int64 saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the bottom of the register save area. + This value is always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_saved_regs; + +- /* The size of the callee-save registers with a slot in REG_OFFSET that +- are saved below the hard frame pointer. */ +- poly_int64 below_hard_fp_saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the hard frame pointer. This value is + always a multiple of STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/11.4.0/0029-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/11.4.0/0029-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..520c1f02 --- /dev/null +++ b/packages/gcc/11.4.0/0029-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,299 @@ +From bea0985749c12fcc264710586addb7838cc61e6d Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:19:52 +0100 +Subject: [PATCH 29/29] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 46 +++++++-- + .../gcc.target/aarch64/stack-protector-8.c | 95 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 +++++++ + 3 files changed, 168 insertions(+), 6 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 385718a475b6..3ccfd3c30fc7 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -7392,6 +7392,20 @@ aarch64_needs_frame_chain (void) + return aarch64_use_frame_pointer; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -7403,6 +7417,7 @@ aarch64_layout_frame (void) + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); + bool frame_related_fp_reg_p = false; + aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 top_of_locals = -1; + + frame.emit_frame_chain = aarch64_needs_frame_chain (); + +@@ -7469,9 +7484,16 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); + + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ if (regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + frame.bytes_below_saved_regs = offset; + frame.sve_save_and_probe = INVALID_REGNUM; + +@@ -7611,15 +7633,18 @@ aarch64_layout_frame (void) + at expand_prologue. */ + gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + +- offset += get_frame_size (); +- offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +- auto top_of_locals = offset; +- ++ if (!regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + offset += frame.saved_varargs_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = offset; + + frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ gcc_assert (known_ge (top_of_locals, 0)); + frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; +@@ -8843,10 +8868,10 @@ aarch64_epilogue_uses (int regno) + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | ++ | padding (1) | + +-------------------------------+ + | callee-saved registers | + +-------------------------------+ +@@ -8858,6 +8883,10 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | SVE predicate registers | + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -8867,6 +8896,9 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. +@@ -9058,6 +9090,8 @@ aarch64_expand_prologue (void) + gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); ++ if (emit_frame_chain && maybe_ne (final_adjust, 0)) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..e71d820e3654 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,95 @@ ++/* { dg-options " -O -fstack-protector-strong -mstack-protector-guard=sysreg -mstack-protector-guard-reg=tpidr2_el0 -mstack-protector-guard-offset=16" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++__SVBool_t *h(void *); ++ ++/* ++** test1: ++** sub sp, sp, #288 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?288 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** sub sp, sp, #1040 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?1032\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?1040 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} ++ ++#pragma GCC target "+sve" ++ ++/* ++** test3: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** addvl sp, sp, #-18 ++** ... ++** str p4, \[sp\] ++** ... ++** sub sp, sp, #272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl h ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?272 ++** ... ++** ldr p4, \[sp\] ++** ... ++** addvl sp, sp, #18 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++__SVBool_t test3() { ++ int y[0x40]; ++ return *h(y); ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..58f322aa480a +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=neoverse-v1 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0010-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch b/packages/gcc/12.3.0/0010-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch new file mode 100644 index 00000000..622eaf6d --- /dev/null +++ b/packages/gcc/12.3.0/0010-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch @@ -0,0 +1,378 @@ +From 62fbb215cc817e9f2c1ca80282a64f4ee30806bc Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:48 +0100 +Subject: [PATCH 10/28] aarch64: Use local frame vars in shrink-wrapping code + +aarch64_layout_frame uses a shorthand for referring to +cfun->machine->frame: + + aarch64_frame &frame = cfun->machine->frame; + +This patch does the same for some other heavy users of the structure. +No functional change intended. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_save_callee_saves): Use + a local shorthand for cfun->machine->frame. + (aarch64_restore_callee_saves, aarch64_get_separate_components): + (aarch64_process_components): Likewise. + (aarch64_allocate_and_probe_stack_space): Likewise. + (aarch64_expand_prologue, aarch64_expand_epilogue): Likewise. + (aarch64_layout_frame): Use existing shorthand for one more case. +--- + gcc/config/aarch64/aarch64.cc | 123 ++++++++++++++++++---------------- + 1 file changed, 64 insertions(+), 59 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 226dc9dffd47..ae42ffdedbeb 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8351,7 +8351,7 @@ aarch64_layout_frame (void) + frame.is_scs_enabled + = (!crtl->calls_eh_return + && sanitize_flags_p (SANITIZE_SHADOW_CALL_STACK) +- && known_ge (cfun->machine->frame.reg_offset[LR_REGNUM], 0)); ++ && known_ge (frame.reg_offset[LR_REGNUM], 0)); + + /* When shadow call stack is enabled, the scs_pop in the epilogue will + restore x30, and we don't need to pop x30 again in the traditional +@@ -8763,6 +8763,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx_insn *insn; + unsigned regno; + unsigned regno2; +@@ -8777,8 +8778,8 @@ aarch64_save_callee_saves (poly_int64 start_offset, + bool frame_related_p = aarch64_emit_cfi_for_reg_p (regno); + + if (skip_wb +- && (regno == cfun->machine->frame.wb_push_candidate1 +- || regno == cfun->machine->frame.wb_push_candidate2)) ++ && (regno == frame.wb_push_candidate1 ++ || regno == frame.wb_push_candidate2)) + continue; + + if (cfun->machine->reg_is_wrapped_separately[regno]) +@@ -8786,7 +8787,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -8799,7 +8800,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + { + gcc_assert (known_eq (start_offset, 0)); + poly_int64 fp_offset +- = cfun->machine->frame.below_hard_fp_saved_regs_size; ++ = frame.below_hard_fp_saved_regs_size; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -8821,8 +8822,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -8872,6 +8872,7 @@ static void + aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { ++ aarch64_frame &frame = cfun->machine->frame; + unsigned regno; + unsigned regno2; + poly_int64 offset; +@@ -8888,13 +8889,13 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + rtx reg, mem; + + if (skip_wb +- && (regno == cfun->machine->frame.wb_pop_candidate1 +- || regno == cfun->machine->frame.wb_pop_candidate2)) ++ && (regno == frame.wb_pop_candidate1 ++ || regno == frame.wb_pop_candidate2)) + continue; + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -8905,8 +8906,7 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -9011,6 +9011,7 @@ offset_12bit_unsigned_scaled_p (machine_mode mode, poly_int64 offset) + static sbitmap + aarch64_get_separate_components (void) + { ++ aarch64_frame &frame = cfun->machine->frame; + sbitmap components = sbitmap_alloc (LAST_SAVED_REGNUM + 1); + bitmap_clear (components); + +@@ -9027,18 +9028,18 @@ aarch64_get_separate_components (void) + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + continue; + +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + + /* If the register is saved in the first SVE save slot, we use + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection +- && maybe_ne (cfun->machine->frame.below_hard_fp_saved_regs_size, 0) ++ && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) + && known_eq (offset, 0)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -9057,11 +9058,11 @@ aarch64_get_separate_components (void) + /* If the spare predicate register used by big-endian SVE code + is call-preserved, it must be saved in the main prologue + before any saves that use it. */ +- if (cfun->machine->frame.spare_pred_reg != INVALID_REGNUM) +- bitmap_clear_bit (components, cfun->machine->frame.spare_pred_reg); ++ if (frame.spare_pred_reg != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.spare_pred_reg); + +- unsigned reg1 = cfun->machine->frame.wb_push_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_push_candidate2; ++ unsigned reg1 = frame.wb_push_candidate1; ++ unsigned reg2 = frame.wb_push_candidate2; + /* If registers have been chosen to be stored/restored with + writeback don't interfere with them to avoid having to output explicit + stack adjustment instructions. */ +@@ -9170,6 +9171,7 @@ aarch64_get_next_set_bit (sbitmap bmp, unsigned int start) + static void + aarch64_process_components (sbitmap components, bool prologue_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx ptr_reg = gen_rtx_REG (Pmode, frame_pointer_needed + ? HARD_FRAME_POINTER_REGNUM + : STACK_POINTER_REGNUM); +@@ -9184,9 +9186,9 @@ aarch64_process_components (sbitmap components, bool prologue_p) + machine_mode mode = aarch64_reg_save_mode (regno); + + rtx reg = gen_rtx_REG (mode, regno); +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -9211,14 +9213,14 @@ aarch64_process_components (sbitmap components, bool prologue_p) + break; + } + +- poly_int64 offset2 = cfun->machine->frame.reg_offset[regno2]; ++ poly_int64 offset2 = frame.reg_offset[regno2]; + /* The next register is not of the same class or its offset is not + mergeable with the current one into a pair. */ + if (aarch64_sve_mode_p (mode) + || !satisfies_constraint_Ump (mem) + || GP_REGNUM_P (regno) != GP_REGNUM_P (regno2) + || (crtl->abi->id () == ARM_PCS_SIMD && FP_REGNUM_P (regno)) +- || maybe_ne ((offset2 - cfun->machine->frame.reg_offset[regno]), ++ || maybe_ne ((offset2 - frame.reg_offset[regno]), + GET_MODE_SIZE (mode))) + { + insn = emit_insn (set); +@@ -9240,7 +9242,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset2 -= frame.below_hard_fp_saved_regs_size; + else + offset2 += crtl->outgoing_args_size; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); +@@ -9335,6 +9337,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + bool frame_related_p, + bool final_adjustment_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; +@@ -9355,25 +9358,25 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + register as a probe. We can't assume that LR was saved at position 0 + though, so treat any space below it as unprobed. */ + if (final_adjustment_p +- && known_eq (cfun->machine->frame.below_hard_fp_saved_regs_size, 0)) ++ && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = cfun->machine->frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else + gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); + } + +- poly_int64 frame_size = cfun->machine->frame.frame_size; ++ poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ + gcc_assert (min_probe_threshold > 0); + + if (flag_stack_clash_protection && !final_adjustment_p) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; + + if (known_eq (frame_size, 0)) + { +@@ -9662,17 +9665,18 @@ aarch64_epilogue_uses (int regno) + void + aarch64_expand_prologue (void) + { +- poly_int64 frame_size = cfun->machine->frame.frame_size; +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 frame_size = frame.frame_size; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_push_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_push_candidate2; +- bool emit_frame_chain = cfun->machine->frame.emit_frame_chain; ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_push_candidate1; ++ unsigned reg2 = frame.wb_push_candidate2; ++ bool emit_frame_chain = frame.emit_frame_chain; + rtx_insn *insn; + + if (flag_stack_clash_protection && known_eq (callee_adjust, 0)) +@@ -9703,7 +9707,7 @@ aarch64_expand_prologue (void) + } + + /* Push return address to shadow call stack. */ +- if (cfun->machine->frame.is_scs_enabled) ++ if (frame.is_scs_enabled) + emit_insn (gen_scs_push ()); + + if (flag_stack_usage_info) +@@ -9742,7 +9746,7 @@ aarch64_expand_prologue (void) + + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - cfun->machine->frame.hard_fp_offset); ++ - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + + /* The offset of the bottom of the save area from the current SP. */ +@@ -9845,16 +9849,17 @@ aarch64_use_return_insn_p (void) + void + aarch64_expand_epilogue (bool for_sibcall) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_pop_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_pop_candidate2; +- unsigned int last_gpr = (cfun->machine->frame.is_scs_enabled ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_pop_candidate1; ++ unsigned reg2 = frame.wb_pop_candidate2; ++ unsigned int last_gpr = (frame.is_scs_enabled + ? R29_REGNUM : R30_REGNUM); + rtx cfi_ops = NULL; + rtx_insn *insn; +@@ -9888,7 +9893,7 @@ aarch64_expand_epilogue (bool for_sibcall) + /* We need to add memory barrier to prevent read from deallocated stack. */ + bool need_barrier_p + = maybe_ne (get_frame_size () +- + cfun->machine->frame.saved_varargs_size, 0); ++ + frame.saved_varargs_size, 0); + + /* Emit a barrier to prevent loads from a deallocated stack. */ + if (maybe_gt (final_adjust, crtl->outgoing_args_size) +@@ -9969,7 +9974,7 @@ aarch64_expand_epilogue (bool for_sibcall) + } + + /* Pop return address from shadow call stack. */ +- if (cfun->machine->frame.is_scs_enabled) ++ if (frame.is_scs_enabled) + { + machine_mode mode = aarch64_reg_save_mode (R30_REGNUM); + rtx reg = gen_rtx_REG (mode, R30_REGNUM); +@@ -12564,24 +12569,24 @@ aarch64_can_eliminate (const int from ATTRIBUTE_UNUSED, const int to) + poly_int64 + aarch64_initial_elimination_offset (unsigned from, unsigned to) + { ++ aarch64_frame &frame = cfun->machine->frame; ++ + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ return frame.hard_fp_offset - frame.locals_offset; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return frame.frame_size - frame.locals_offset; + } + +- return cfun->machine->frame.frame_size; ++ return frame.frame_size; + } + + +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0011-aarch64-Avoid-a-use-of-callee_offset.patch b/packages/gcc/12.3.0/0011-aarch64-Avoid-a-use-of-callee_offset.patch new file mode 100644 index 00000000..dc9fdb33 --- /dev/null +++ b/packages/gcc/12.3.0/0011-aarch64-Avoid-a-use-of-callee_offset.patch @@ -0,0 +1,73 @@ +From 12a8889de169f892d2e927584c00d20b8b7e456f Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:49 +0100 +Subject: [PATCH 11/28] aarch64: Avoid a use of callee_offset + +When we emit the frame chain, i.e. when we reach Here in this statement +of aarch64_expand_prologue: + + if (emit_frame_chain) + { + // Here + ... + } + +the stack is in one of two states: + +- We've allocated up to the frame chain, but no more. + +- We've allocated the whole frame, and the frame chain is within easy + reach of the new SP. + +The offset of the frame chain from the current SP is available +in aarch64_frame as callee_offset. It is also available as the +chain_offset local variable, where the latter is calculated from other +data. (However, chain_offset is not always equal to callee_offset when +!emit_frame_chain, so chain_offset isn't redundant.) + +In c600df9a4060da3c6121ff4d0b93f179eafd69d1 I switched to using +chain_offset for the initialisation of the hard frame pointer: + + aarch64_add_offset (Pmode, hard_frame_pointer_rtx, +- stack_pointer_rtx, callee_offset, ++ stack_pointer_rtx, chain_offset, + tmp1_rtx, tmp0_rtx, frame_pointer_needed); + +But the later REG_CFA_ADJUST_CFA handling still used callee_offset. + +I think the difference is harmless, but it's more logical for the +CFA note to be in sync, and it's more convenient for later patches +if it uses chain_offset. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_expand_prologue): Use + chain_offset rather than callee_offset. +--- + gcc/config/aarch64/aarch64.cc | 4 +--- + 1 file changed, 1 insertion(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index ae42ffdedbeb..79253322fd7c 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9670,7 +9670,6 @@ aarch64_expand_prologue (void) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size + = frame.below_hard_fp_saved_regs_size; +@@ -9783,8 +9782,7 @@ aarch64_expand_prologue (void) + implicit. */ + if (!find_reg_note (insn, REG_CFA_ADJUST_CFA, NULL_RTX)) + { +- rtx src = plus_constant (Pmode, stack_pointer_rtx, +- callee_offset); ++ rtx src = plus_constant (Pmode, stack_pointer_rtx, chain_offset); + add_reg_note (insn, REG_CFA_ADJUST_CFA, + gen_rtx_SET (hard_frame_pointer_rtx, src)); + } +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0012-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/12.3.0/0012-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..1454b85a --- /dev/null +++ b/packages/gcc/12.3.0/0012-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,49 @@ +From 03d5e89e7f3be53fd7142556e8e0a2774c653dca Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:49 +0100 +Subject: [PATCH 12/28] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +And if we allocate the frame in one go, it should be allocated +as the initial_adjust rather than the final_adjust. This allows the +frame size to grow to guard_size - guard_used_by_caller before a stack +probe is needed. (A frame with no register saves is necessarily a +leaf frame.) + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.cc | 8 +++++--- + 1 file changed, 5 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 79253322fd7c..e1f21230c15e 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8378,9 +8378,11 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; +- if (frame.frame_size.is_constant (&const_size) +- && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ if (known_eq (frame.saved_regs_size, 0)) ++ frame.initial_adjust = frame.frame_size; ++ else if (frame.frame_size.is_constant (&const_size) ++ && const_size < max_push_offset ++ && known_eq (frame.hard_fp_offset, const_size)) + { + /* Simple, small frame with no outgoing arguments: + +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0013-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch b/packages/gcc/12.3.0/0013-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch new file mode 100644 index 00000000..cb625c6c --- /dev/null +++ b/packages/gcc/12.3.0/0013-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch @@ -0,0 +1,233 @@ +From 49c2eb7616756c323b7f6b18d8616ec945eb1263 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:49 +0100 +Subject: [PATCH 13/28] aarch64: Add bytes_below_saved_regs to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_saved_regs): New + field. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_saved_regs instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 71 ++++++++++++++++++----------------- + gcc/config/aarch64/aarch64.h | 5 +++ + 2 files changed, 41 insertions(+), 35 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index e1f21230c15e..94e1b6865849 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8217,6 +8217,8 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + ++ frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small + offset range. These saves happen below the hard frame pointer. */ +@@ -8321,18 +8323,18 @@ aarch64_layout_frame (void) + + poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; + +- poly_int64 above_outgoing_args ++ poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + + frame.hard_fp_offset +- = above_outgoing_args - frame.below_hard_fp_saved_regs_size; ++ = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +- gcc_assert (multiple_p (crtl->outgoing_args_size, ++ gcc_assert (multiple_p (frame.bytes_below_saved_regs, + STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = above_outgoing_args + crtl->outgoing_args_size; ++ frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + + frame.locals_offset = frame.saved_varargs_size; + +@@ -8376,7 +8378,7 @@ aarch64_layout_frame (void) + else if (frame.wb_pop_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; +@@ -8384,31 +8386,31 @@ aarch64_layout_frame (void) + && const_size < max_push_offset + && known_eq (frame.hard_fp_offset, const_size)) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + frame.callee_adjust = const_size; + } +- else if (crtl->outgoing_args_size.is_constant (&const_outgoing_args_size) ++ else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) + && frame.saved_regs_size.is_constant (&const_saved_regs_size) +- && const_outgoing_args_size + const_saved_regs_size < 512 +- /* We could handle this case even with outgoing args, provided +- that the number of args left us with valid offsets for all +- predicate and vector save slots. It's such a rare case that +- it hardly seems worth the effort though. */ +- && (!saves_below_hard_fp_p || const_outgoing_args_size == 0) ++ && const_below_saved_regs + const_saved_regs_size < 512 ++ /* We could handle this case even with data below the saved ++ registers, provided that that data left us with valid offsets ++ for all predicate and vector save slots. It's such a rare ++ case that it hardly seems worth the effort though. */ ++ && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca + && frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset)) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_saved_regs] ++ stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_outgoing_args_size; ++ frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -8418,30 +8420,29 @@ aarch64_layout_frame (void) + + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = (frame.hard_fp_offset + + frame.below_hard_fp_saved_regs_size); +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments or SVE saves, but with +- a small local area: ++ /* Frame with large area below the saved registers, or with SVE saves, ++ but with a small area above: + + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else + { +- /* Frame with large local area and outgoing arguments or SVE saves, +- using frame pointer: ++ /* General case: + + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] +@@ -8449,10 +8450,10 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.hard_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + + /* Make sure the individual adjustments add up to the full frame size. */ +@@ -9043,7 +9044,7 @@ aarch64_get_separate_components (void) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -9192,7 +9193,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -9246,7 +9247,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset2 -= frame.below_hard_fp_saved_regs_size; + else +- offset2 += crtl->outgoing_args_size; ++ offset2 += frame.bytes_below_saved_regs; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -9320,10 +9321,10 @@ aarch64_stack_clash_protection_alloca_probe_range (void) + registers. If POLY_SIZE is not large enough to require a probe this function + will only adjust the stack. When allocating the stack space + FRAME_RELATED_P is then used to indicate if the allocation is frame related. +- FINAL_ADJUSTMENT_P indicates whether we are allocating the outgoing +- arguments. If we are then we ensure that any allocation larger than the ABI +- defined buffer needs a probe so that the invariant of having a 1KB buffer is +- maintained. ++ FINAL_ADJUSTMENT_P indicates whether we are allocating the area below ++ the saved registers. If we are then we ensure that any allocation ++ larger than the ABI defined buffer needs a probe so that the ++ invariant of having a 1KB buffer is maintained. + + We emit barriers after each stack adjustment to prevent optimizations from + breaking the invariant that we never drop the stack more than a page. This +@@ -9532,7 +9533,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + /* Handle any residuals. Residuals of at least MIN_PROBE_THRESHOLD have to + be probed. This maintains the requirement that each page is probed at + least once. For initial probing we probe only if the allocation is +- more than GUARD_SIZE - buffer, and for the outgoing arguments we probe ++ more than GUARD_SIZE - buffer, and below the saved registers we probe + if the amount is larger than buffer. GUARD_SIZE - buffer + buffer == + GUARD_SIZE. This works that for any allocation that is large enough to + trigger a probe here, we'll have at least one, and if they're not large +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 6834c3e99226..1e105e12db8d 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -871,6 +871,11 @@ struct GTY (()) aarch64_frame + /* The size of the callee-save registers with a slot in REG_OFFSET. */ + poly_int64 saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the bottom of the register save area. ++ This value is always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_saved_regs; ++ + /* The size of the callee-save registers with a slot in REG_OFFSET that + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0014-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/12.3.0/0014-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..826b4ccf --- /dev/null +++ b/packages/gcc/12.3.0/0014-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,84 @@ +From 34081079ea4de0c98331843f574b5f6f94d7b234 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:50 +0100 +Subject: [PATCH 14/28] aarch64: Add bytes_below_hard_fp to frame info + +Following on from the previous bytes_below_saved_regs patch, this one +records the number of bytes that are below the hard frame pointer. +This eventually replaces below_hard_fp_saved_regs_size. + +If a frame pointer is not needed, the epilogue adds final_adjust +to the stack pointer before restoring registers: + + aarch64_add_sp (tmp1_rtx, tmp0_rtx, final_adjust, true); + +Therefore, if the epilogue needs to restore the stack pointer from +the hard frame pointer, the directly corresponding offset is: + + -bytes_below_hard_fp + final_adjust + +i.e. go from the hard frame pointer to the bottom of the frame, +then add the same amount as if we were using the stack pointer +from the outset. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Initialize it. + (aarch64_expand_epilogue): Use it instead of + below_hard_fp_saved_regs_size. +--- + gcc/config/aarch64/aarch64.cc | 6 +++--- + gcc/config/aarch64/aarch64.h | 5 +++++ + 2 files changed, 8 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 94e1b6865849..c7d84245fbfc 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8269,6 +8269,7 @@ aarch64_layout_frame (void) + of the callee save area. */ + bool saves_below_hard_fp_p = maybe_ne (offset, 0); + frame.below_hard_fp_saved_regs_size = offset; ++ frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -9856,8 +9857,7 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 final_adjust = frame.final_adjust; + poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_pop_candidate1; + unsigned reg2 = frame.wb_pop_candidate2; + unsigned int last_gpr = (frame.is_scs_enabled +@@ -9915,7 +9915,7 @@ aarch64_expand_epilogue (bool for_sibcall) + is restored on the instruction doing the writeback. */ + aarch64_add_offset (Pmode, stack_pointer_rtx, + hard_frame_pointer_rtx, +- -callee_offset - below_hard_fp_saved_regs_size, ++ -bytes_below_hard_fp + final_adjust, + tmp1_rtx, tmp0_rtx, callee_adjust == 0); + else + /* The case where we need to re-use the register here is very rare, so +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 1e105e12db8d..de68ff7202fc 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -880,6 +880,11 @@ struct GTY (()) aarch64_frame + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0015-aarch64-Tweak-aarch64_save-restore_callee_saves.patch b/packages/gcc/12.3.0/0015-aarch64-Tweak-aarch64_save-restore_callee_saves.patch new file mode 100644 index 00000000..c1873245 --- /dev/null +++ b/packages/gcc/12.3.0/0015-aarch64-Tweak-aarch64_save-restore_callee_saves.patch @@ -0,0 +1,225 @@ +From 187861af7c51db9eddc6f954b589c121b210fc74 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:50 +0100 +Subject: [PATCH 15/28] aarch64: Tweak aarch64_save/restore_callee_saves + +aarch64_save_callee_saves and aarch64_restore_callee_saves took +a parameter called start_offset that gives the offset of the +bottom of the saved register area from the current stack pointer. +However, it's more convenient for later patches if we use the +bottom of the entire frame as the reference point, rather than +the bottom of the saved registers. + +Doing that removes the need for the callee_offset field. +Other than that, this is not a win on its own. It only really +makes sense in combination with the follow-on patches. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::callee_offset): Delete. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Remove + callee_offset handling. + (aarch64_save_callee_saves): Replace the start_offset parameter + with a bytes_below_sp parameter. + (aarch64_restore_callee_saves): Likewise. + (aarch64_expand_prologue): Update accordingly. + (aarch64_expand_epilogue): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 56 +++++++++++++++++------------------ + gcc/config/aarch64/aarch64.h | 4 --- + 2 files changed, 28 insertions(+), 32 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index c7d84245fbfc..e79551af41df 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8343,7 +8343,6 @@ aarch64_layout_frame (void) + frame.final_adjust = 0; + frame.callee_adjust = 0; + frame.sve_callee_adjust = 0; +- frame.callee_offset = 0; + + frame.wb_pop_candidate1 = frame.wb_push_candidate1; + frame.wb_pop_candidate2 = frame.wb_push_candidate2; +@@ -8411,7 +8410,6 @@ aarch64_layout_frame (void) + stp reg1, reg2, [sp, bytes_below_saved_regs] + stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -8758,12 +8756,13 @@ aarch64_add_cfa_expression (rtx_insn *insn, rtx reg, + } + + /* Emit code to save the callee-saved registers from register number START +- to LIMIT to the stack at the location starting at offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. HARD_FP_VALID_P +- is true if the hard frame pointer has been set up. */ ++ to LIMIT to the stack. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. HARD_FP_VALID_P is true if the hard ++ frame pointer has been set up. */ + + static void +-aarch64_save_callee_saves (poly_int64 start_offset, ++aarch64_save_callee_saves (poly_int64 bytes_below_sp, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { +@@ -8791,7 +8790,9 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -8802,9 +8803,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + else if (GP_REGNUM_P (regno) + && (!offset.is_constant (&const_offset) || const_offset >= 512)) + { +- gcc_assert (known_eq (start_offset, 0)); +- poly_int64 fp_offset +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 fp_offset = frame.bytes_below_hard_fp - bytes_below_sp; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -8868,12 +8867,13 @@ aarch64_save_callee_saves (poly_int64 start_offset, + } + + /* Emit code to restore the callee registers from register number START +- up to and including LIMIT. Restore from the stack offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. Write the +- appropriate REG_CFA_RESTORE notes into CFI_OPS. */ ++ up to and including LIMIT. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. Write the appropriate REG_CFA_RESTORE ++ notes into CFI_OPS. */ + + static void +-aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, ++aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { + aarch64_frame &frame = cfun->machine->frame; +@@ -8899,7 +8899,9 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -9675,8 +9677,6 @@ aarch64_expand_prologue (void) + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; + unsigned reg1 = frame.wb_push_candidate1; + unsigned reg2 = frame.wb_push_candidate2; + bool emit_frame_chain = frame.emit_frame_chain; +@@ -9752,8 +9752,8 @@ aarch64_expand_prologue (void) + - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + +- /* The offset of the bottom of the save area from the current SP. */ +- poly_int64 saved_regs_offset = chain_offset - below_hard_fp_saved_regs_size; ++ /* The offset of the current SP from the bottom of the static frame. */ ++ poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { +@@ -9761,7 +9761,7 @@ aarch64_expand_prologue (void) + { + reg1 = R29_REGNUM; + reg2 = R30_REGNUM; +- aarch64_save_callee_saves (saved_regs_offset, reg1, reg2, ++ aarch64_save_callee_saves (bytes_below_sp, reg1, reg2, + false, false); + } + else +@@ -9801,7 +9801,7 @@ aarch64_expand_prologue (void) + emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + +- aarch64_save_callee_saves (saved_regs_offset, R0_REGNUM, R30_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, R0_REGNUM, R30_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + if (maybe_ne (sve_callee_adjust, 0)) +@@ -9811,16 +9811,17 @@ aarch64_expand_prologue (void) + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, + sve_callee_adjust, + !frame_pointer_needed, false); +- saved_regs_offset += sve_callee_adjust; ++ bytes_below_sp -= sve_callee_adjust; + } +- aarch64_save_callee_saves (saved_regs_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, P0_REGNUM, P15_REGNUM, + false, emit_frame_chain); +- aarch64_save_callee_saves (saved_regs_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, V0_REGNUM, V31_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + + /* We may need to probe the final adjustment if it is larger than the guard + that is assumed by the called. */ ++ gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); + } +@@ -9855,7 +9856,6 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_pop_candidate1; +@@ -9925,9 +9925,9 @@ aarch64_expand_epilogue (bool for_sibcall) + + /* Restore the vector registers before the predicate registers, + so that we can use P4 as a temporary for big-endian SVE frames. */ +- aarch64_restore_callee_saves (callee_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, V0_REGNUM, V31_REGNUM, + callee_adjust != 0, &cfi_ops); +- aarch64_restore_callee_saves (callee_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, P0_REGNUM, P15_REGNUM, + false, &cfi_ops); + if (maybe_ne (sve_callee_adjust, 0)) + aarch64_add_sp (NULL_RTX, NULL_RTX, sve_callee_adjust, true); +@@ -9935,7 +9935,7 @@ aarch64_expand_epilogue (bool for_sibcall) + /* When shadow call stack is enabled, the scs_pop in the epilogue will + restore x30, we don't need to restore x30 again in the traditional + way. */ +- aarch64_restore_callee_saves (callee_offset - sve_callee_adjust, ++ aarch64_restore_callee_saves (final_adjust + sve_callee_adjust, + R0_REGNUM, last_gpr, + callee_adjust != 0, &cfi_ops); + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index de68ff7202fc..94fca4b94716 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -907,10 +907,6 @@ struct GTY (()) aarch64_frame + It is zero when no push is used. */ + HOST_WIDE_INT callee_adjust; + +- /* The offset from SP to the callee-save registers after initial_adjust. +- It may be non-zero if no push is used (ie. callee_adjust == 0). */ +- poly_int64 callee_offset; +- + /* The size of the stack adjustment before saving or after restoring + SVE registers. */ + poly_int64 sve_callee_adjust; +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0016-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch b/packages/gcc/12.3.0/0016-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch new file mode 100644 index 00000000..edf1ad56 --- /dev/null +++ b/packages/gcc/12.3.0/0016-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch @@ -0,0 +1,45 @@ +From 2b983f9064d808daf909bde1d4a13980934a7e6e Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:51 +0100 +Subject: [PATCH 16/28] aarch64: Only calculate chain_offset if there is a + chain + +After previous patches, it is no longer necessary to calculate +a chain_offset in cases where there is no chain record. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_expand_prologue): Move the + calculation of chain_offset into the emit_frame_chain block. +--- + gcc/config/aarch64/aarch64.cc | 10 +++++----- + 1 file changed, 5 insertions(+), 5 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index e79551af41df..d71a042d6112 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9747,16 +9747,16 @@ aarch64_expand_prologue (void) + if (callee_adjust != 0) + aarch64_push_regs (reg1, reg2, callee_adjust); + +- /* The offset of the frame chain record (if any) from the current SP. */ +- poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); +- gcc_assert (known_ge (chain_offset, 0)); +- + /* The offset of the current SP from the bottom of the static frame. */ + poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { ++ /* The offset of the frame chain record (if any) from the current SP. */ ++ poly_int64 chain_offset = (initial_adjust + callee_adjust ++ - frame.hard_fp_offset); ++ gcc_assert (known_ge (chain_offset, 0)); ++ + if (callee_adjust == 0) + { + reg1 = R29_REGNUM; +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0017-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/12.3.0/0017-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..399f589c --- /dev/null +++ b/packages/gcc/12.3.0/0017-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,91 @@ +From 0a0a824808d1dec51004fb5805c1a0ae2a35433f Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:51 +0100 +Subject: [PATCH 17/28] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.cc (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.cc | 6 +++--- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 6 insertions(+), 6 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index d71a042d6112..d4ec352ba98a 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8337,7 +8337,7 @@ aarch64_layout_frame (void) + STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + +- frame.locals_offset = frame.saved_varargs_size; ++ frame.bytes_above_locals = frame.saved_varargs_size; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +@@ -12578,13 +12578,13 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.locals_offset; ++ return frame.hard_fp_offset - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return frame.frame_size - frame.locals_offset; ++ return frame.frame_size - frame.bytes_above_locals; + } + + return frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 94fca4b94716..bf46e6124aa9 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -885,10 +885,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 locals_offset; ++ poly_int64 bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0018-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/12.3.0/0018-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..37844832 --- /dev/null +++ b/packages/gcc/12.3.0/0018-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,148 @@ +From 3fbf0789202b30a67b12e1fb785c7130f098d665 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:52 +0100 +Subject: [PATCH 18/28] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.cc (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 26 +++++++++++++------------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 16 insertions(+), 16 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index d4ec352ba98a..3c4052740e7a 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8329,7 +8329,7 @@ aarch64_layout_frame (void) + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.hard_fp_offset ++ frame.bytes_above_hard_fp + = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +@@ -8378,13 +8378,13 @@ aarch64_layout_frame (void) + else if (frame.wb_pop_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ && known_eq (frame.bytes_above_hard_fp, const_size)) + { + /* Simple, small frame with no data below the saved registers. + +@@ -8401,8 +8401,8 @@ aarch64_layout_frame (void) + case that it hardly seems worth the effort though. */ + && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca +- && frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset)) ++ && frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset)) + { + /* Frame with small area below the saved registers: + +@@ -8420,12 +8420,12 @@ aarch64_layout_frame (void) + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.hard_fp_offset ++ frame.initial_adjust = (frame.bytes_above_hard_fp + + frame.below_hard_fp_saved_regs_size); + frame.final_adjust = frame.bytes_below_saved_regs; + } +- else if (frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset) ++ else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset) + { + /* Frame with large area below the saved registers, or with SVE saves, + but with a small area above: +@@ -8435,7 +8435,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.callee_adjust = const_fp_offset; ++ frame.callee_adjust = const_above_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -8450,7 +8450,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = frame.hard_fp_offset; ++ frame.initial_adjust = frame.bytes_above_hard_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -9754,7 +9754,7 @@ aarch64_expand_prologue (void) + { + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); ++ - frame.bytes_above_hard_fp); + gcc_assert (known_ge (chain_offset, 0)); + + if (callee_adjust == 0) +@@ -12575,10 +12575,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return frame.hard_fp_offset; ++ return frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.bytes_above_locals; ++ return frame.bytes_above_hard_fp - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index bf46e6124aa9..dd1f403f9393 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -890,10 +890,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 hard_fp_offset; ++ poly_int64 bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0019-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/12.3.0/0019-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..b80eae96 --- /dev/null +++ b/packages/gcc/12.3.0/0019-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,35 @@ +From aac8b31379ac3bbd14fc6427dce23f56e54e8485 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:52 +0100 +Subject: [PATCH 19/28] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index dd1f403f9393..700524ae22bf 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -895,8 +895,8 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- frame (incomming SP) to the stack_pointer. This value is always ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always + a multiple of STACK_BOUNDARY. */ + poly_int64 frame_size; + +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0020-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch b/packages/gcc/12.3.0/0020-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch new file mode 100644 index 00000000..30de1b36 --- /dev/null +++ b/packages/gcc/12.3.0/0020-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch @@ -0,0 +1,196 @@ +From 8d5506a8aeb8dd7e8b209a3663b07688478f76b9 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:53 +0100 +Subject: [PATCH 20/28] aarch64: Measure reg_offset from the bottom of the + frame + +reg_offset was measured from the bottom of the saved register area. +This made perfect sense with the original layout, since the bottom +of the saved register area was also the hard frame pointer address. +It became slightly less obvious with SVE, since we save SVE +registers below the hard frame pointer, but it still made sense. + +However, if we want to allow different frame layouts, it's more +convenient and obvious to measure reg_offset from the bottom of +the frame. After previous patches, it's also a slight simplification +in its own right. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame): Add comment above + reg_offset. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Walk offsets + from the bottom of the frame, rather than the bottom of the saved + register area. Measure reg_offset from the bottom of the frame + rather than the bottom of the saved register area. + (aarch64_save_callee_saves): Update accordingly. + (aarch64_restore_callee_saves): Likewise. + (aarch64_get_separate_components): Likewise. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 53 ++++++++++++++++------------------- + gcc/config/aarch64/aarch64.h | 3 ++ + 2 files changed, 27 insertions(+), 29 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 3c4052740e7a..97dd077844b4 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8139,7 +8139,6 @@ aarch64_needs_frame_chain (void) + static void + aarch64_layout_frame (void) + { +- poly_int64 offset = 0; + int regno, last_fp_reg = INVALID_REGNUM; + machine_mode vector_save_mode = aarch64_reg_save_mode (V8_REGNUM); + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); +@@ -8217,7 +8216,9 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + +- frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ poly_int64 offset = crtl->outgoing_args_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.bytes_below_saved_regs = offset; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -8229,7 +8230,8 @@ aarch64_layout_frame (void) + offset += BYTES_PER_SVE_PRED; + } + +- if (maybe_ne (offset, 0)) ++ poly_int64 saved_prs_size = offset - frame.bytes_below_saved_regs; ++ if (maybe_ne (saved_prs_size, 0)) + { + /* If we have any vector registers to save above the predicate registers, + the offset of the vector register save slots need to be a multiple +@@ -8247,10 +8249,10 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + else + { +- if (known_le (offset, vector_save_size)) +- offset = vector_save_size; +- else if (known_le (offset, vector_save_size * 2)) +- offset = vector_save_size * 2; ++ if (known_le (saved_prs_size, vector_save_size)) ++ offset = frame.bytes_below_saved_regs + vector_save_size; ++ else if (known_le (saved_prs_size, vector_save_size * 2)) ++ offset = frame.bytes_below_saved_regs + vector_save_size * 2; + else + gcc_unreachable (); + } +@@ -8267,9 +8269,10 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- bool saves_below_hard_fp_p = maybe_ne (offset, 0); +- frame.below_hard_fp_saved_regs_size = offset; +- frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; ++ frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p ++ = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ frame.bytes_below_hard_fp = offset; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -8320,9 +8323,10 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset; ++ frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; ++ poly_int64 varargs_and_saved_regs_size ++ = frame.saved_regs_size + frame.saved_varargs_size; + + poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size +@@ -8790,9 +8794,7 @@ aarch64_save_callee_saves (poly_int64 bytes_below_sp, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -8899,9 +8901,7 @@ aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -9040,14 +9040,12 @@ aarch64_get_separate_components (void) + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection + && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, 0)) ++ && known_eq (offset, frame.bytes_below_saved_regs)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -9194,9 +9192,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -9248,9 +9244,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= frame.below_hard_fp_saved_regs_size; +- else +- offset2 += frame.bytes_below_saved_regs; ++ offset2 -= frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -9366,7 +9360,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p + && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] ++ - frame.bytes_below_saved_regs); + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 700524ae22bf..b61358370732 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -860,6 +860,9 @@ extern enum aarch64_processor aarch64_tune; + #ifdef HAVE_POLY_INT_H + struct GTY (()) aarch64_frame + { ++ /* The offset from the bottom of the static frame (the bottom of the ++ outgoing arguments) of each register save slot, or -2 if no save is ++ needed. */ + poly_int64 reg_offset[LAST_SAVED_REGNUM + 1]; + + /* The number of extra stack bytes taken up by register varargs. +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0021-aarch64-Simplify-top-of-frame-allocation.patch b/packages/gcc/12.3.0/0021-aarch64-Simplify-top-of-frame-allocation.patch new file mode 100644 index 00000000..78ca6aa6 --- /dev/null +++ b/packages/gcc/12.3.0/0021-aarch64-Simplify-top-of-frame-allocation.patch @@ -0,0 +1,55 @@ +From b47766614df3b9df878262efb2ad73aaac108363 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:53 +0100 +Subject: [PATCH 21/28] aarch64: Simplify top of frame allocation + +After previous patches, it no longer really makes sense to allocate +the top of the frame in terms of varargs_and_saved_regs_size and +saved_regs_and_above. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Simplify + the allocation of the top of the frame. +--- + gcc/config/aarch64/aarch64.cc | 23 ++++++++--------------- + 1 file changed, 8 insertions(+), 15 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 97dd077844b4..81935852d5b2 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8325,23 +8325,16 @@ aarch64_layout_frame (void) + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size +- = frame.saved_regs_size + frame.saved_varargs_size; ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ auto top_of_locals = offset; + +- poly_int64 saved_regs_and_above +- = aligned_upper_bound (varargs_and_saved_regs_size +- + get_frame_size (), +- STACK_BOUNDARY / BITS_PER_UNIT); ++ offset += frame.saved_varargs_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.frame_size = offset; + +- frame.bytes_above_hard_fp +- = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; +- +- /* Both these values are already aligned. */ +- gcc_assert (multiple_p (frame.bytes_below_saved_regs, +- STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; +- +- frame.bytes_above_locals = frame.saved_varargs_size; ++ frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0022-aarch64-Minor-initial-adjustment-tweak.patch b/packages/gcc/12.3.0/0022-aarch64-Minor-initial-adjustment-tweak.patch new file mode 100644 index 00000000..16d3ad30 --- /dev/null +++ b/packages/gcc/12.3.0/0022-aarch64-Minor-initial-adjustment-tweak.patch @@ -0,0 +1,38 @@ +From 08f71b4bb28fb74d20e8d2927a557e8119ce9f4d Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:54 +0100 +Subject: [PATCH 22/28] aarch64: Minor initial adjustment tweak + +This patch just changes a calculation of initial_adjust +to one that makes it slightly more obvious that the total +adjustment is frame.frame_size. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Tweak + calculation of initial_adjust for frames in which all saves + are SVE saves. +--- + gcc/config/aarch64/aarch64.cc | 5 ++--- + 1 file changed, 2 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 81935852d5b2..4d9fcf3d1623 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8414,11 +8414,10 @@ aarch64_layout_frame (void) + { + /* Frame in which all saves are SVE saves: + +- sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size ++ sub sp, sp, frame_size - bytes_below_saved_regs + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.bytes_above_hard_fp +- + frame.below_hard_fp_saved_regs_size); ++ frame.initial_adjust = frame.frame_size - frame.bytes_below_saved_regs; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0023-aarch64-Tweak-stack-clash-boundary-condition.patch b/packages/gcc/12.3.0/0023-aarch64-Tweak-stack-clash-boundary-condition.patch new file mode 100644 index 00000000..44e82a3d --- /dev/null +++ b/packages/gcc/12.3.0/0023-aarch64-Tweak-stack-clash-boundary-condition.patch @@ -0,0 +1,125 @@ +From f22315d5c19e8310e4dc880fd509678fd291fca8 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:54 +0100 +Subject: [PATCH 23/28] aarch64: Tweak stack clash boundary condition + +The AArch64 ABI says that, when stack clash protection is used, +there can be a maximum of 1KiB of unprobed space at sp on entry +to a function. Therefore, we need to probe when allocating +>= guard_size - 1KiB of data (>= rather than >). This is what +GCC does. + +If an allocation is exactly guard_size bytes, it is enough to allocate +those bytes and probe once at offset 1024. It isn't possible to use a +single probe at any other offset: higher would conmplicate later code, +by leaving more unprobed space than usual, while lower would risk +leaving an entire page unprobed. For simplicity, the code probes all +allocations at offset 1024. + +Some register saves also act as probes. If we need to allocate +more space below the last such register save probe, we need to +probe the allocation if it is > 1KiB. Again, this allocation is +then sometimes (but not always) probed at offset 1024. This sort of +allocation is currently only used for outgoing arguments, which are +rarely this big. + +However, the code also probed if this final outgoing-arguments +allocation was == 1KiB, rather than just > 1KiB. This isn't +necessary, since the register save then probes at offset 1024 +as required. Continuing to probe allocations of exactly 1KiB +would complicate later patches. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_allocate_and_probe_stack_space): + Don't probe final allocations that are exactly 1KiB in size (after + unprobed space above the final allocation has been deducted). + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: New test. +--- + gcc/config/aarch64/aarch64.cc | 4 +- + .../aarch64/stack-check-prologue-17.c | 55 +++++++++++++++++++ + 2 files changed, 58 insertions(+), 1 deletion(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 4d9fcf3d1623..34c1d8614cd9 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9333,9 +9333,11 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; ++ HOST_WIDE_INT byte_sp_alignment = STACK_BOUNDARY / BITS_PER_UNIT; ++ gcc_assert (multiple_p (poly_size, byte_sp_alignment)); + HOST_WIDE_INT min_probe_threshold + = (final_adjustment_p +- ? guard_used_by_caller ++ ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); + /* When doing the final adjustment for the outgoing arguments, take into + account any unprobed space there is above the current SP. There are +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +new file mode 100644 +index 000000000000..0d8a25d73a24 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -0,0 +1,55 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0024-aarch64-Put-LR-save-probe-in-first-16-bytes.patch b/packages/gcc/12.3.0/0024-aarch64-Put-LR-save-probe-in-first-16-bytes.patch new file mode 100644 index 00000000..7ed364f2 --- /dev/null +++ b/packages/gcc/12.3.0/0024-aarch64-Put-LR-save-probe-in-first-16-bytes.patch @@ -0,0 +1,406 @@ +From 15e18831bf98fd25af098b970ebf0c9a6200a34b Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:55 +0100 +Subject: [PATCH 24/28] aarch64: Put LR save probe in first 16 bytes + +-fstack-clash-protection uses the save of LR as a probe for the next +allocation. The next allocation could be: + +* another part of the static frame, e.g. when allocating SVE save slots + or outgoing arguments + +* an alloca in the same function + +* an allocation made by a callee function + +However, when -fomit-frame-pointer is used, the LR save slot is placed +above the other GPR save slots. It could therefore be up to 80 bytes +above the base of the GPR save area (which is also the hard fp address). + +aarch64_allocate_and_probe_stack_space took this into account when +deciding how much subsequent space could be allocated without needing +a probe. However, it interacted badly with: + + /* If doing a small final adjustment, we always probe at offset 0. + This is done to avoid issues when LR is not at position 0 or when + the final adjustment is smaller than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +which forces any allocation that is smaller than the guard page size +to be probed at offset 0 rather than the usual offset 1024. It was +therefore possible to construct cases in which we had: + +* a probe using LR at SP + 80 bytes (or some other value >= 16) +* an allocation of the guard page size - 16 bytes +* a probe at SP + 0 + +which allocates guard page size + 64 consecutive unprobed bytes. + +This patch requires the LR probe to be in the first 16 bytes of the +save area when stack clash protection is active. Doing it +unconditionally would cause code-quality regressions. + +Putting LR before other registers prevents push/pop allocation +when shadow call stacks are enabled, since LR is restored +separately from the other callee-saved registers. + +The new comment doesn't say that the probe register is required +to be LR, since a later patch removes that restriction. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Ensure that + the LR save slot is in the first 16 bytes of the register save area. + Only form STP/LDP push/pop candidates if both registers are valid. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + when LR was not in the first 16 bytes. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-18.c: New test. + * gcc.target/aarch64/stack-check-prologue-19.c: Likewise. + * gcc.target/aarch64/stack-check-prologue-20.c: Likewise. +--- + gcc/config/aarch64/aarch64.cc | 72 ++++++------- + .../aarch64/stack-check-prologue-18.c | 100 ++++++++++++++++++ + .../aarch64/stack-check-prologue-19.c | 100 ++++++++++++++++++ + .../aarch64/stack-check-prologue-20.c | 3 + + 4 files changed, 233 insertions(+), 42 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 34c1d8614cd9..16433fb70f4f 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8273,26 +8273,34 @@ aarch64_layout_frame (void) + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); + frame.bytes_below_hard_fp = offset; ++ ++ auto allocate_gpr_slot = [&](unsigned int regno) ++ { ++ frame.reg_offset[regno] = offset; ++ if (frame.wb_push_candidate1 == INVALID_REGNUM) ++ frame.wb_push_candidate1 = regno; ++ else if (frame.wb_push_candidate2 == INVALID_REGNUM) ++ frame.wb_push_candidate2 = regno; ++ offset += UNITS_PER_WORD; ++ }; ++ + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +- frame.reg_offset[R29_REGNUM] = offset; +- frame.wb_push_candidate1 = R29_REGNUM; +- frame.reg_offset[R30_REGNUM] = offset + UNITS_PER_WORD; +- frame.wb_push_candidate2 = R30_REGNUM; +- offset += 2 * UNITS_PER_WORD; ++ allocate_gpr_slot (R29_REGNUM); ++ allocate_gpr_slot (R30_REGNUM); + } ++ else if (flag_stack_clash_protection ++ && known_eq (frame.reg_offset[R30_REGNUM], SLOT_REQUIRED)) ++ /* Put the LR save slot first, since it makes a good choice of probe ++ for stack clash purposes. The idea is that the link register usually ++ has to be saved before a call anyway, and so we lose little by ++ stopping it from being individually shrink-wrapped. */ ++ allocate_gpr_slot (R30_REGNUM); + + for (regno = R0_REGNUM; regno <= R30_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) +- { +- frame.reg_offset[regno] = offset; +- if (frame.wb_push_candidate1 == INVALID_REGNUM) +- frame.wb_push_candidate1 = regno; +- else if (frame.wb_push_candidate2 == INVALID_REGNUM) +- frame.wb_push_candidate2 = regno; +- offset += UNITS_PER_WORD; +- } ++ allocate_gpr_slot (regno); + + poly_int64 max_int_offset = offset; + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8370,10 +8378,13 @@ aarch64_layout_frame (void) + max_push_offset to 0, because no registers are popped at this time, + so callee_adjust cannot be adjusted. */ + HOST_WIDE_INT max_push_offset = 0; +- if (frame.wb_pop_candidate2 != INVALID_REGNUM) +- max_push_offset = 512; +- else if (frame.wb_pop_candidate1 != INVALID_REGNUM) +- max_push_offset = 256; ++ if (frame.wb_pop_candidate1 != INVALID_REGNUM) ++ { ++ if (frame.wb_pop_candidate2 != INVALID_REGNUM) ++ max_push_offset = 512; ++ else ++ max_push_offset = 256; ++ } + + HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; +@@ -9339,29 +9350,6 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + = (final_adjustment_p + ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); +- /* When doing the final adjustment for the outgoing arguments, take into +- account any unprobed space there is above the current SP. There are +- two cases: +- +- - When saving SVE registers below the hard frame pointer, we force +- the lowest save to take place in the prologue before doing the final +- adjustment (i.e. we don't allow the save to be shrink-wrapped). +- This acts as a probe at SP, so there is no unprobed space. +- +- - When there are no SVE register saves, we use the store of the link +- register as a probe. We can't assume that LR was saved at position 0 +- though, so treat any space below it as unprobed. */ +- if (final_adjustment_p +- && known_eq (frame.below_hard_fp_saved_regs_size, 0)) +- { +- poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] +- - frame.bytes_below_saved_regs); +- if (known_ge (lr_offset, 0)) +- min_probe_threshold -= lr_offset.to_constant (); +- else +- gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); +- } +- + poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ +@@ -9541,8 +9529,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; + /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when LR is not at position 0 or when +- the final adjustment is smaller than the probing offset. */ ++ This is done to avoid issues when the final adjustment is smaller ++ than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +new file mode 100644 +index 000000000000..82447d20fff5 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -0,0 +1,100 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +new file mode 100644 +index 000000000000..73ac3e4e4eb0 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +@@ -0,0 +1,100 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12 -fsanitize=shadow-call-stack -ffixed-x18" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c +new file mode 100644 +index 000000000000..690aae8dfd5b +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c +@@ -0,0 +1,3 @@ ++/* { dg-options "-O2 -fstack-protector-all -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12 -fsanitize=shadow-call-stack -ffixed-x18" } */ ++ ++#include "stack-check-prologue-19.c" +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0025-aarch64-Simplify-probe-of-final-frame-allocation.patch b/packages/gcc/12.3.0/0025-aarch64-Simplify-probe-of-final-frame-allocation.patch new file mode 100644 index 00000000..fcd778e4 --- /dev/null +++ b/packages/gcc/12.3.0/0025-aarch64-Simplify-probe-of-final-frame-allocation.patch @@ -0,0 +1,123 @@ +From c4f0e121faa36342f1d21919e54a05ad841c4f86 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:55 +0100 +Subject: [PATCH 25/28] aarch64: Simplify probe of final frame allocation + +Previous patches ensured that the final frame allocation only needs +a probe when the size is strictly greater than 1KiB. It's therefore +safe to use the normal 1024 probe offset in all cases. + +The main motivation for doing this is to simplify the code and +remove the number of special cases. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_allocate_and_probe_stack_space): + Always probe the residual allocation at offset 1024, asserting + that that is in range. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: Expect the probe + to be at offset 1024 rather than offset 0. + * gcc.target/aarch64/stack-check-prologue-18.c: Likewise. + * gcc.target/aarch64/stack-check-prologue-19.c: Likewise. +--- + gcc/config/aarch64/aarch64.cc | 12 ++++-------- + .../gcc.target/aarch64/stack-check-prologue-17.c | 2 +- + .../gcc.target/aarch64/stack-check-prologue-18.c | 4 ++-- + .../gcc.target/aarch64/stack-check-prologue-19.c | 4 ++-- + 4 files changed, 9 insertions(+), 13 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 16433fb70f4f..8abf3d7a1e2b 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9523,16 +9523,12 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + are still safe. */ + if (residual) + { +- HOST_WIDE_INT residual_probe_offset = guard_used_by_caller; ++ gcc_assert (guard_used_by_caller + byte_sp_alignment <= size); ++ + /* If we're doing final adjustments, and we've done any full page + allocations then any residual needs to be probed. */ + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; +- /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when the final adjustment is smaller +- than the probing offset. */ +- else if (final_adjustment_p && rounded_size == 0) +- residual_probe_offset = 0; + + aarch64_sub_sp (temp1, temp2, residual, frame_related_p); + if (residual >= min_probe_threshold) +@@ -9543,8 +9539,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT_PRINT_DEC " bytes, probing will be required." + "\n", residual); + +- emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, +- residual_probe_offset)); ++ emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, ++ guard_used_by_caller)); + emit_insn (gen_blockage ()); + } + } +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +index 0d8a25d73a24..f0ec1389771d 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -33,7 +33,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +index 82447d20fff5..6383bec5ebcd 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -9,7 +9,7 @@ void g(); + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -50,7 +50,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +index 73ac3e4e4eb0..562039b5e9b8 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +@@ -9,7 +9,7 @@ void g(); + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -50,7 +50,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0026-aarch64-Explicitly-record-probe-registers-in-frame-i.patch b/packages/gcc/12.3.0/0026-aarch64-Explicitly-record-probe-registers-in-frame-i.patch new file mode 100644 index 00000000..4f13c582 --- /dev/null +++ b/packages/gcc/12.3.0/0026-aarch64-Explicitly-record-probe-registers-in-frame-i.patch @@ -0,0 +1,278 @@ +From 6f0ab0a9f46a17b68349ff6035aa776bf65f0575 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:56 +0100 +Subject: [PATCH 26/28] aarch64: Explicitly record probe registers in frame + info + +The stack frame is currently divided into three areas: + +A: the area above the hard frame pointer +B: the SVE saves below the hard frame pointer +C: the outgoing arguments + +If the stack frame is allocated in one chunk, the allocation needs a +probe if the frame size is >= guard_size - 1KiB. In addition, if the +function is not a leaf function, it must probe an address no more than +1KiB above the outgoing SP. We ensured the second condition by + +(1) using single-chunk allocations for non-leaf functions only if + the link register save slot is within 512 bytes of the bottom + of the frame; and + +(2) using the link register save as a probe (meaning, for instance, + that it can't be individually shrink wrapped) + +If instead the stack is allocated in multiple chunks, then: + +* an allocation involving only the outgoing arguments (C above) requires + a probe if the allocation size is > 1KiB + +* any other allocation requires a probe if the allocation size + is >= guard_size - 1KiB + +* second and subsequent allocations require the previous allocation + to probe at the bottom of the allocated area, regardless of the size + of that previous allocation + +The final point means that, unlike for single allocations, +it can be necessary to have both a non-SVE register probe and +an SVE register probe. For example: + +* allocate A, probe using a non-SVE register save +* allocate B, probe using an SVE register save +* allocate C + +The non-SVE register used in this case was again the link register. +It was previously used even if the link register save slot was some +bytes above the bottom of the non-SVE register saves, but an earlier +patch avoided that by putting the link register save slot first. + +As a belt-and-braces fix, this patch explicitly records which +probe registers we're using and allows the non-SVE probe to be +whichever register comes first (as for SVE). + +The patch also avoids unnecessary probes in sve/pcs/stack_clash_3.c. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::sve_save_and_probe) + (aarch64_frame::hard_fp_save_and_probe): New fields. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Initialize them. + Rather than asserting that a leaf function saves LR, instead assert + that a leaf function saves something. + (aarch64_get_separate_components): Prevent the chosen probe + registers from being individually shrink-wrapped. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + probe registers that aren't at the bottom of the previous allocation. + +gcc/testsuite/ + * gcc.target/aarch64/sve/pcs/stack_clash_3.c: Avoid redundant probes. +--- + gcc/config/aarch64/aarch64.cc | 68 +++++++++++++++---- + gcc/config/aarch64/aarch64.h | 8 +++ + .../aarch64/sve/pcs/stack_clash_3.c | 6 +- + 3 files changed, 64 insertions(+), 18 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 8abf3d7a1e2b..a8d907df8843 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8210,15 +8210,11 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + +- /* With stack-clash, LR must be saved in non-leaf functions. The saving of +- LR counts as an implicit probe which allows us to maintain the invariant +- described in the comment at expand_prologue. */ +- gcc_assert (crtl->is_leaf +- || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.bytes_below_saved_regs = offset; ++ frame.sve_save_and_probe = INVALID_REGNUM; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -8226,6 +8222,8 @@ aarch64_layout_frame (void) + for (regno = P0_REGNUM; regno <= P15_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += BYTES_PER_SVE_PRED; + } +@@ -8263,6 +8261,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += vector_save_size; + } +@@ -8272,10 +8272,18 @@ aarch64_layout_frame (void) + frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ gcc_assert (!saves_below_hard_fp_p ++ || (frame.sve_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.sve_save_and_probe], ++ frame.bytes_below_saved_regs))); ++ + frame.bytes_below_hard_fp = offset; ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; + + auto allocate_gpr_slot = [&](unsigned int regno) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + frame.reg_offset[regno] = offset; + if (frame.wb_push_candidate1 == INVALID_REGNUM) + frame.wb_push_candidate1 = regno; +@@ -8309,6 +8317,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + /* If there is an alignment gap between integer and fp callee-saves, + allocate the last fp register to it if possible. */ + if (regno == last_fp_reg +@@ -8332,6 +8342,17 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (frame.saved_regs_size, ++ frame.below_hard_fp_saved_regs_size) ++ || (frame.hard_fp_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], ++ frame.bytes_below_hard_fp))); ++ ++ /* With stack-clash, a register must be saved in non-leaf functions. ++ The saving of the bottommost register counts as an implicit probe, ++ which allows us to maintain the invariant described in the comment ++ at expand_prologue. */ ++ gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8462,6 +8483,25 @@ aarch64_layout_frame (void) + frame.final_adjust = frame.bytes_below_saved_regs; + } + ++ /* The frame is allocated in pieces, with each non-final piece ++ including a register save at offset 0 that acts as a probe for ++ the following piece. In addition, the save of the bottommost register ++ acts as a probe for callees and allocas. Roll back any probes that ++ aren't needed. ++ ++ A probe isn't needed if it is associated with the final allocation ++ (including callees and allocas) that happens before the epilogue is ++ executed. */ ++ if (crtl->is_leaf ++ && !cfun->calls_alloca ++ && known_eq (frame.final_adjust, 0)) ++ { ++ if (maybe_ne (frame.sve_callee_adjust, 0)) ++ frame.sve_save_and_probe = INVALID_REGNUM; ++ else ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; ++ } ++ + /* Make sure the individual adjustments add up to the full frame size. */ + gcc_assert (known_eq (frame.initial_adjust + + frame.callee_adjust +@@ -9039,13 +9079,6 @@ aarch64_get_separate_components (void) + + poly_int64 offset = frame.reg_offset[regno]; + +- /* If the register is saved in the first SVE save slot, we use +- it as a stack probe for -fstack-clash-protection. */ +- if (flag_stack_clash_protection +- && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, frame.bytes_below_saved_regs)) +- continue; +- + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) + offset -= frame.bytes_below_hard_fp; +@@ -9080,6 +9113,13 @@ aarch64_get_separate_components (void) + + bitmap_clear_bit (components, LR_REGNUM); + bitmap_clear_bit (components, SP_REGNUM); ++ if (flag_stack_clash_protection) ++ { ++ if (frame.sve_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.sve_save_and_probe); ++ if (frame.hard_fp_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.hard_fp_save_and_probe); ++ } + + return components; + } +@@ -9616,8 +9656,8 @@ aarch64_epilogue_uses (int regno) + When probing is needed, we emit a probe at the start of the prologue + and every PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE bytes thereafter. + +- We have to track how much space has been allocated and the only stores +- to the stack we track as implicit probes are the FP/LR stores. ++ We can also use register saves as probes. These are stored in ++ sve_save_and_probe and hard_fp_save_and_probe. + + For outgoing arguments we probe if the size is larger than 1KB, such that + the ABI specified buffer is maintained for the next callee. +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index b61358370732..46d4693e2064 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -957,6 +957,14 @@ struct GTY (()) aarch64_frame + This is the register they should use. */ + unsigned spare_pred_reg; + ++ /* An SVE register that is saved below the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned sve_save_and_probe; ++ ++ /* A register that is saved at the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned hard_fp_save_and_probe; ++ + bool laid_out; + + /* True if shadow call stack should be enabled for the current function. */ +diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +index 3e01ec36c3a4..3530a0d504ba 100644 +--- a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c ++++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +@@ -11,11 +11,10 @@ + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +@@ -39,13 +38,12 @@ test_1 (int n) + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** str p5, \[sp, #1, mul vl\] + ** str p6, \[sp, #2, mul vl\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0027-aarch64-Remove-below_hard_fp_saved_regs_size.patch b/packages/gcc/12.3.0/0027-aarch64-Remove-below_hard_fp_saved_regs_size.patch new file mode 100644 index 00000000..bfb7a318 --- /dev/null +++ b/packages/gcc/12.3.0/0027-aarch64-Remove-below_hard_fp_saved_regs_size.patch @@ -0,0 +1,157 @@ +From 8254e1b9cd500e0c278465a3657543477e9d1250 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:56 +0100 +Subject: [PATCH 27/28] aarch64: Remove below_hard_fp_saved_regs_size + +After previous patches, it's no longer necessary to store +saved_regs_size and below_hard_fp_saved_regs_size in the frame info. +All measurements instead use the top or bottom of the frame as +reference points. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::saved_regs_size) + (aarch64_frame::below_hard_fp_saved_regs_size): Delete. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Update accordingly. +--- + gcc/config/aarch64/aarch64.cc | 45 ++++++++++++++++------------------- + gcc/config/aarch64/aarch64.h | 7 ------ + 2 files changed, 21 insertions(+), 31 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index a8d907df8843..ac3d3b336a37 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8269,9 +8269,8 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; +- bool saves_below_hard_fp_p +- = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ auto below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p = maybe_ne (below_hard_fp_saved_regs_size, 0); + gcc_assert (!saves_below_hard_fp_p + || (frame.sve_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.sve_save_and_probe], +@@ -8341,9 +8340,8 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset - frame.bytes_below_saved_regs; +- gcc_assert (known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size) ++ auto saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (saved_regs_size, below_hard_fp_saved_regs_size) + || (frame.hard_fp_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], + frame.bytes_below_hard_fp))); +@@ -8352,7 +8350,7 @@ aarch64_layout_frame (void) + The saving of the bottommost register counts as an implicit probe, + which allows us to maintain the invariant described in the comment + at expand_prologue. */ +- gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); ++ gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8409,7 +8407,7 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; +- if (known_eq (frame.saved_regs_size, 0)) ++ if (known_eq (saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +@@ -8422,7 +8420,7 @@ aarch64_layout_frame (void) + frame.callee_adjust = const_size; + } + else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) +- && frame.saved_regs_size.is_constant (&const_saved_regs_size) ++ && saved_regs_size.is_constant (&const_saved_regs_size) + && const_below_saved_regs + const_saved_regs_size < 512 + /* We could handle this case even with data below the saved + registers, provided that that data left us with valid offsets +@@ -8441,8 +8439,7 @@ aarch64_layout_frame (void) + frame.initial_adjust = frame.frame_size; + } + else if (saves_below_hard_fp_p +- && known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size)) ++ && known_eq (saved_regs_size, below_hard_fp_saved_regs_size)) + { + /* Frame in which all saves are SVE saves: + +@@ -8464,7 +8461,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_above_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else +@@ -8479,7 +8476,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.bytes_above_hard_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + +@@ -9621,17 +9618,17 @@ aarch64_epilogue_uses (int regno) + | local variables | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | \ +- +-------------------------------+ | +- | callee-saved registers | | frame.saved_regs_size +- +-------------------------------+ | +- | LR' | | +- +-------------------------------+ | +- | FP' | | +- +-------------------------------+ |<- hard_frame_pointer_rtx (aligned) +- | SVE vector registers | | \ +- +-------------------------------+ | | below_hard_fp_saved_regs_size +- | SVE predicate registers | / / ++ | padding | ++ +-------------------------------+ ++ | callee-saved registers | ++ +-------------------------------+ ++ | LR' | ++ +-------------------------------+ ++ | FP' | ++ +-------------------------------+ <-- hard_frame_pointer_rtx (aligned) ++ | SVE vector registers | ++ +-------------------------------+ ++ | SVE predicate registers | + +-------------------------------+ + | dynamic allocation | + +-------------------------------+ +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 46d4693e2064..01f7751bc783 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -871,18 +871,11 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + HOST_WIDE_INT saved_varargs_size; + +- /* The size of the callee-save registers with a slot in REG_OFFSET. */ +- poly_int64 saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the bottom of the register save area. + This value is always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_saved_regs; + +- /* The size of the callee-save registers with a slot in REG_OFFSET that +- are saved below the hard frame pointer. */ +- poly_int64 below_hard_fp_saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the hard frame pointer. This value is + always a multiple of STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/12.3.0/0028-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/12.3.0/0028-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..db03001f --- /dev/null +++ b/packages/gcc/12.3.0/0028-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,299 @@ +From 75c37e031408262263442f5b4cdb83d3777b6422 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:08:57 +0100 +Subject: [PATCH 28/28] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.cc | 46 +++++++-- + .../gcc.target/aarch64/stack-protector-8.c | 95 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 +++++++ + 3 files changed, 168 insertions(+), 6 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index ac3d3b336a37..96c3f48fdc49 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8133,6 +8133,20 @@ aarch64_needs_frame_chain (void) + return aarch64_use_frame_pointer; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -8144,6 +8158,7 @@ aarch64_layout_frame (void) + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); + bool frame_related_fp_reg_p = false; + aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 top_of_locals = -1; + + frame.emit_frame_chain = aarch64_needs_frame_chain (); + +@@ -8210,9 +8225,16 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); + + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ if (regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + frame.bytes_below_saved_regs = offset; + frame.sve_save_and_probe = INVALID_REGNUM; + +@@ -8352,15 +8374,18 @@ aarch64_layout_frame (void) + at expand_prologue. */ + gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + +- offset += get_frame_size (); +- offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +- auto top_of_locals = offset; +- ++ if (!regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + offset += frame.saved_varargs_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = offset; + + frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ gcc_assert (known_ge (top_of_locals, 0)); + frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; +@@ -9615,10 +9640,10 @@ aarch64_epilogue_uses (int regno) + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | ++ | padding (1) | + +-------------------------------+ + | callee-saved registers | + +-------------------------------+ +@@ -9630,6 +9655,10 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | SVE predicate registers | + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -9639,6 +9668,9 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. +@@ -9834,6 +9866,8 @@ aarch64_expand_prologue (void) + gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); ++ if (emit_frame_chain && maybe_ne (final_adjust, 0)) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..e71d820e3654 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,95 @@ ++/* { dg-options " -O -fstack-protector-strong -mstack-protector-guard=sysreg -mstack-protector-guard-reg=tpidr2_el0 -mstack-protector-guard-offset=16" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++__SVBool_t *h(void *); ++ ++/* ++** test1: ++** sub sp, sp, #288 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?288 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** sub sp, sp, #1040 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?1032\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?1040 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} ++ ++#pragma GCC target "+sve" ++ ++/* ++** test3: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** addvl sp, sp, #-18 ++** ... ++** str p4, \[sp\] ++** ... ++** sub sp, sp, #272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl h ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?272 ++** ... ++** ldr p4, \[sp\] ++** ... ++** addvl sp, sp, #18 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++__SVBool_t test3() { ++ int y[0x40]; ++ return *h(y); ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..58f322aa480a +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=neoverse-v1 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0014-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch b/packages/gcc/13.2.0/0014-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch new file mode 100644 index 00000000..d3502cd2 --- /dev/null +++ b/packages/gcc/13.2.0/0014-aarch64-Use-local-frame-vars-in-shrink-wrapping-code.patch @@ -0,0 +1,378 @@ +From 71a2aa2127283f450c623d3604dbcabe0e14a8d4 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:12 +0100 +Subject: [PATCH 14/32] aarch64: Use local frame vars in shrink-wrapping code + +aarch64_layout_frame uses a shorthand for referring to +cfun->machine->frame: + + aarch64_frame &frame = cfun->machine->frame; + +This patch does the same for some other heavy users of the structure. +No functional change intended. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_save_callee_saves): Use + a local shorthand for cfun->machine->frame. + (aarch64_restore_callee_saves, aarch64_get_separate_components): + (aarch64_process_components): Likewise. + (aarch64_allocate_and_probe_stack_space): Likewise. + (aarch64_expand_prologue, aarch64_expand_epilogue): Likewise. + (aarch64_layout_frame): Use existing shorthand for one more case. +--- + gcc/config/aarch64/aarch64.cc | 123 ++++++++++++++++++---------------- + 1 file changed, 64 insertions(+), 59 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 822a2b49a46e..5d473d161d97 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8612,7 +8612,7 @@ aarch64_layout_frame (void) + frame.is_scs_enabled + = (!crtl->calls_eh_return + && sanitize_flags_p (SANITIZE_SHADOW_CALL_STACK) +- && known_ge (cfun->machine->frame.reg_offset[LR_REGNUM], 0)); ++ && known_ge (frame.reg_offset[LR_REGNUM], 0)); + + /* When shadow call stack is enabled, the scs_pop in the epilogue will + restore x30, and we don't need to pop x30 again in the traditional +@@ -9078,6 +9078,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx_insn *insn; + unsigned regno; + unsigned regno2; +@@ -9092,8 +9093,8 @@ aarch64_save_callee_saves (poly_int64 start_offset, + bool frame_related_p = aarch64_emit_cfi_for_reg_p (regno); + + if (skip_wb +- && (regno == cfun->machine->frame.wb_push_candidate1 +- || regno == cfun->machine->frame.wb_push_candidate2)) ++ && (regno == frame.wb_push_candidate1 ++ || regno == frame.wb_push_candidate2)) + continue; + + if (cfun->machine->reg_is_wrapped_separately[regno]) +@@ -9101,7 +9102,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -9114,7 +9115,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + { + gcc_assert (known_eq (start_offset, 0)); + poly_int64 fp_offset +- = cfun->machine->frame.below_hard_fp_saved_regs_size; ++ = frame.below_hard_fp_saved_regs_size; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -9136,8 +9137,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -9187,6 +9187,7 @@ static void + aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { ++ aarch64_frame &frame = cfun->machine->frame; + unsigned regno; + unsigned regno2; + poly_int64 offset; +@@ -9203,13 +9204,13 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + rtx reg, mem; + + if (skip_wb +- && (regno == cfun->machine->frame.wb_pop_candidate1 +- || regno == cfun->machine->frame.wb_pop_candidate2)) ++ && (regno == frame.wb_pop_candidate1 ++ || regno == frame.wb_pop_candidate2)) + continue; + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + cfun->machine->frame.reg_offset[regno]; ++ offset = start_offset + frame.reg_offset[regno]; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -9220,8 +9221,7 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + && (regno2 = aarch64_next_callee_save (regno + 1, limit)) <= limit + && !cfun->machine->reg_is_wrapped_separately[regno2] + && known_eq (GET_MODE_SIZE (mode), +- cfun->machine->frame.reg_offset[regno2] +- - cfun->machine->frame.reg_offset[regno])) ++ frame.reg_offset[regno2] - frame.reg_offset[regno])) + { + rtx reg2 = gen_rtx_REG (mode, regno2); + rtx mem2; +@@ -9326,6 +9326,7 @@ offset_12bit_unsigned_scaled_p (machine_mode mode, poly_int64 offset) + static sbitmap + aarch64_get_separate_components (void) + { ++ aarch64_frame &frame = cfun->machine->frame; + sbitmap components = sbitmap_alloc (LAST_SAVED_REGNUM + 1); + bitmap_clear (components); + +@@ -9342,18 +9343,18 @@ aarch64_get_separate_components (void) + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + continue; + +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + + /* If the register is saved in the first SVE save slot, we use + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection +- && maybe_ne (cfun->machine->frame.below_hard_fp_saved_regs_size, 0) ++ && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) + && known_eq (offset, 0)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -9372,11 +9373,11 @@ aarch64_get_separate_components (void) + /* If the spare predicate register used by big-endian SVE code + is call-preserved, it must be saved in the main prologue + before any saves that use it. */ +- if (cfun->machine->frame.spare_pred_reg != INVALID_REGNUM) +- bitmap_clear_bit (components, cfun->machine->frame.spare_pred_reg); ++ if (frame.spare_pred_reg != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.spare_pred_reg); + +- unsigned reg1 = cfun->machine->frame.wb_push_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_push_candidate2; ++ unsigned reg1 = frame.wb_push_candidate1; ++ unsigned reg2 = frame.wb_push_candidate2; + /* If registers have been chosen to be stored/restored with + writeback don't interfere with them to avoid having to output explicit + stack adjustment instructions. */ +@@ -9485,6 +9486,7 @@ aarch64_get_next_set_bit (sbitmap bmp, unsigned int start) + static void + aarch64_process_components (sbitmap components, bool prologue_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + rtx ptr_reg = gen_rtx_REG (Pmode, frame_pointer_needed + ? HARD_FRAME_POINTER_REGNUM + : STACK_POINTER_REGNUM); +@@ -9499,9 +9501,9 @@ aarch64_process_components (sbitmap components, bool prologue_p) + machine_mode mode = aarch64_reg_save_mode (regno); + + rtx reg = gen_rtx_REG (mode, regno); +- poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset -= frame.below_hard_fp_saved_regs_size; + else + offset += crtl->outgoing_args_size; + +@@ -9526,14 +9528,14 @@ aarch64_process_components (sbitmap components, bool prologue_p) + break; + } + +- poly_int64 offset2 = cfun->machine->frame.reg_offset[regno2]; ++ poly_int64 offset2 = frame.reg_offset[regno2]; + /* The next register is not of the same class or its offset is not + mergeable with the current one into a pair. */ + if (aarch64_sve_mode_p (mode) + || !satisfies_constraint_Ump (mem) + || GP_REGNUM_P (regno) != GP_REGNUM_P (regno2) + || (crtl->abi->id () == ARM_PCS_SIMD && FP_REGNUM_P (regno)) +- || maybe_ne ((offset2 - cfun->machine->frame.reg_offset[regno]), ++ || maybe_ne ((offset2 - frame.reg_offset[regno]), + GET_MODE_SIZE (mode))) + { + insn = emit_insn (set); +@@ -9555,7 +9557,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= cfun->machine->frame.below_hard_fp_saved_regs_size; ++ offset2 -= frame.below_hard_fp_saved_regs_size; + else + offset2 += crtl->outgoing_args_size; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); +@@ -9650,6 +9652,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + bool frame_related_p, + bool final_adjustment_p) + { ++ aarch64_frame &frame = cfun->machine->frame; + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; +@@ -9670,25 +9673,25 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + register as a probe. We can't assume that LR was saved at position 0 + though, so treat any space below it as unprobed. */ + if (final_adjustment_p +- && known_eq (cfun->machine->frame.below_hard_fp_saved_regs_size, 0)) ++ && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = cfun->machine->frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else + gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); + } + +- poly_int64 frame_size = cfun->machine->frame.frame_size; ++ poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ + gcc_assert (min_probe_threshold > 0); + + if (flag_stack_clash_protection && !final_adjustment_p) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; + + if (known_eq (frame_size, 0)) + { +@@ -9977,17 +9980,18 @@ aarch64_epilogue_uses (int regno) + void + aarch64_expand_prologue (void) + { +- poly_int64 frame_size = cfun->machine->frame.frame_size; +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 frame_size = frame.frame_size; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_push_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_push_candidate2; +- bool emit_frame_chain = cfun->machine->frame.emit_frame_chain; ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_push_candidate1; ++ unsigned reg2 = frame.wb_push_candidate2; ++ bool emit_frame_chain = frame.emit_frame_chain; + rtx_insn *insn; + + if (flag_stack_clash_protection && known_eq (callee_adjust, 0)) +@@ -10018,7 +10022,7 @@ aarch64_expand_prologue (void) + } + + /* Push return address to shadow call stack. */ +- if (cfun->machine->frame.is_scs_enabled) ++ if (frame.is_scs_enabled) + emit_insn (gen_scs_push ()); + + if (flag_stack_usage_info) +@@ -10057,7 +10061,7 @@ aarch64_expand_prologue (void) + + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - cfun->machine->frame.hard_fp_offset); ++ - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + + /* The offset of the bottom of the save area from the current SP. */ +@@ -10160,16 +10164,17 @@ aarch64_use_return_insn_p (void) + void + aarch64_expand_epilogue (bool for_sibcall) + { +- poly_int64 initial_adjust = cfun->machine->frame.initial_adjust; +- HOST_WIDE_INT callee_adjust = cfun->machine->frame.callee_adjust; +- poly_int64 final_adjust = cfun->machine->frame.final_adjust; +- poly_int64 callee_offset = cfun->machine->frame.callee_offset; +- poly_int64 sve_callee_adjust = cfun->machine->frame.sve_callee_adjust; ++ aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 initial_adjust = frame.initial_adjust; ++ HOST_WIDE_INT callee_adjust = frame.callee_adjust; ++ poly_int64 final_adjust = frame.final_adjust; ++ poly_int64 callee_offset = frame.callee_offset; ++ poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size +- = cfun->machine->frame.below_hard_fp_saved_regs_size; +- unsigned reg1 = cfun->machine->frame.wb_pop_candidate1; +- unsigned reg2 = cfun->machine->frame.wb_pop_candidate2; +- unsigned int last_gpr = (cfun->machine->frame.is_scs_enabled ++ = frame.below_hard_fp_saved_regs_size; ++ unsigned reg1 = frame.wb_pop_candidate1; ++ unsigned reg2 = frame.wb_pop_candidate2; ++ unsigned int last_gpr = (frame.is_scs_enabled + ? R29_REGNUM : R30_REGNUM); + rtx cfi_ops = NULL; + rtx_insn *insn; +@@ -10203,7 +10208,7 @@ aarch64_expand_epilogue (bool for_sibcall) + /* We need to add memory barrier to prevent read from deallocated stack. */ + bool need_barrier_p + = maybe_ne (get_frame_size () +- + cfun->machine->frame.saved_varargs_size, 0); ++ + frame.saved_varargs_size, 0); + + /* Emit a barrier to prevent loads from a deallocated stack. */ + if (maybe_gt (final_adjust, crtl->outgoing_args_size) +@@ -10284,7 +10289,7 @@ aarch64_expand_epilogue (bool for_sibcall) + } + + /* Pop return address from shadow call stack. */ +- if (cfun->machine->frame.is_scs_enabled) ++ if (frame.is_scs_enabled) + { + machine_mode mode = aarch64_reg_save_mode (R30_REGNUM); + rtx reg = gen_rtx_REG (mode, R30_REGNUM); +@@ -12740,24 +12745,24 @@ aarch64_can_eliminate (const int from ATTRIBUTE_UNUSED, const int to) + poly_int64 + aarch64_initial_elimination_offset (unsigned from, unsigned to) + { ++ aarch64_frame &frame = cfun->machine->frame; ++ + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ return frame.hard_fp_offset - frame.locals_offset; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return frame.frame_size - frame.locals_offset; + } + +- return cfun->machine->frame.frame_size; ++ return frame.frame_size; + } + + +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0015-aarch64-Avoid-a-use-of-callee_offset.patch b/packages/gcc/13.2.0/0015-aarch64-Avoid-a-use-of-callee_offset.patch new file mode 100644 index 00000000..8a30f5a6 --- /dev/null +++ b/packages/gcc/13.2.0/0015-aarch64-Avoid-a-use-of-callee_offset.patch @@ -0,0 +1,73 @@ +From 89a9fa287706c5011f61926eaf65e7b996b963a3 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:12 +0100 +Subject: [PATCH 15/32] aarch64: Avoid a use of callee_offset + +When we emit the frame chain, i.e. when we reach Here in this statement +of aarch64_expand_prologue: + + if (emit_frame_chain) + { + // Here + ... + } + +the stack is in one of two states: + +- We've allocated up to the frame chain, but no more. + +- We've allocated the whole frame, and the frame chain is within easy + reach of the new SP. + +The offset of the frame chain from the current SP is available +in aarch64_frame as callee_offset. It is also available as the +chain_offset local variable, where the latter is calculated from other +data. (However, chain_offset is not always equal to callee_offset when +!emit_frame_chain, so chain_offset isn't redundant.) + +In c600df9a4060da3c6121ff4d0b93f179eafd69d1 I switched to using +chain_offset for the initialisation of the hard frame pointer: + + aarch64_add_offset (Pmode, hard_frame_pointer_rtx, +- stack_pointer_rtx, callee_offset, ++ stack_pointer_rtx, chain_offset, + tmp1_rtx, tmp0_rtx, frame_pointer_needed); + +But the later REG_CFA_ADJUST_CFA handling still used callee_offset. + +I think the difference is harmless, but it's more logical for the +CFA note to be in sync, and it's more convenient for later patches +if it uses chain_offset. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_expand_prologue): Use + chain_offset rather than callee_offset. +--- + gcc/config/aarch64/aarch64.cc | 4 +--- + 1 file changed, 1 insertion(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 5d473d161d97..4f233c95140e 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9985,7 +9985,6 @@ aarch64_expand_prologue (void) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 below_hard_fp_saved_regs_size + = frame.below_hard_fp_saved_regs_size; +@@ -10098,8 +10097,7 @@ aarch64_expand_prologue (void) + implicit. */ + if (!find_reg_note (insn, REG_CFA_ADJUST_CFA, NULL_RTX)) + { +- rtx src = plus_constant (Pmode, stack_pointer_rtx, +- callee_offset); ++ rtx src = plus_constant (Pmode, stack_pointer_rtx, chain_offset); + add_reg_note (insn, REG_CFA_ADJUST_CFA, + gen_rtx_SET (hard_frame_pointer_rtx, src)); + } +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0016-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/13.2.0/0016-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..c19ea80c --- /dev/null +++ b/packages/gcc/13.2.0/0016-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,49 @@ +From b36a2a78040722dab6124366c5d6baf8eaf80aef Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:13 +0100 +Subject: [PATCH 16/32] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +And if we allocate the frame in one go, it should be allocated +as the initial_adjust rather than the final_adjust. This allows the +frame size to grow to guard_size - guard_used_by_caller before a stack +probe is needed. (A frame with no register saves is necessarily a +leaf frame.) + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.cc | 8 +++++--- + 1 file changed, 5 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 4f233c95140e..37643041ffb1 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8639,9 +8639,11 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; +- if (frame.frame_size.is_constant (&const_size) +- && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ if (known_eq (frame.saved_regs_size, 0)) ++ frame.initial_adjust = frame.frame_size; ++ else if (frame.frame_size.is_constant (&const_size) ++ && const_size < max_push_offset ++ && known_eq (frame.hard_fp_offset, const_size)) + { + /* Simple, small frame with no outgoing arguments: + +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0017-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch b/packages/gcc/13.2.0/0017-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch new file mode 100644 index 00000000..761f7c53 --- /dev/null +++ b/packages/gcc/13.2.0/0017-aarch64-Add-bytes_below_saved_regs-to-frame-info.patch @@ -0,0 +1,233 @@ +From ada2ab0093596be707f23a3466ac82cff59fcffe Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:13 +0100 +Subject: [PATCH 17/32] aarch64: Add bytes_below_saved_regs to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_saved_regs): New + field. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_saved_regs instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 71 ++++++++++++++++++----------------- + gcc/config/aarch64/aarch64.h | 5 +++ + 2 files changed, 41 insertions(+), 35 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 37643041ffb1..dacc2b0e4dd2 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8478,6 +8478,8 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + ++ frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small + offset range. These saves happen below the hard frame pointer. */ +@@ -8582,18 +8584,18 @@ aarch64_layout_frame (void) + + poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; + +- poly_int64 above_outgoing_args ++ poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + + frame.hard_fp_offset +- = above_outgoing_args - frame.below_hard_fp_saved_regs_size; ++ = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +- gcc_assert (multiple_p (crtl->outgoing_args_size, ++ gcc_assert (multiple_p (frame.bytes_below_saved_regs, + STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = above_outgoing_args + crtl->outgoing_args_size; ++ frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + + frame.locals_offset = frame.saved_varargs_size; + +@@ -8637,7 +8639,7 @@ aarch64_layout_frame (void) + else if (frame.wb_pop_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_outgoing_args_size, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; +@@ -8645,31 +8647,31 @@ aarch64_layout_frame (void) + && const_size < max_push_offset + && known_eq (frame.hard_fp_offset, const_size)) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + frame.callee_adjust = const_size; + } +- else if (crtl->outgoing_args_size.is_constant (&const_outgoing_args_size) ++ else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) + && frame.saved_regs_size.is_constant (&const_saved_regs_size) +- && const_outgoing_args_size + const_saved_regs_size < 512 +- /* We could handle this case even with outgoing args, provided +- that the number of args left us with valid offsets for all +- predicate and vector save slots. It's such a rare case that +- it hardly seems worth the effort though. */ +- && (!saves_below_hard_fp_p || const_outgoing_args_size == 0) ++ && const_below_saved_regs + const_saved_regs_size < 512 ++ /* We could handle this case even with data below the saved ++ registers, provided that that data left us with valid offsets ++ for all predicate and vector save slots. It's such a rare ++ case that it hardly seems worth the effort though. */ ++ && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca + && frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset)) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_saved_regs] ++ stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_outgoing_args_size; ++ frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -8679,30 +8681,29 @@ aarch64_layout_frame (void) + + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = (frame.hard_fp_offset + + frame.below_hard_fp_saved_regs_size); +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments or SVE saves, but with +- a small local area: ++ /* Frame with large area below the saved registers, or with SVE saves, ++ but with a small area above: + + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + else + { +- /* Frame with large local area and outgoing arguments or SVE saves, +- using frame pointer: ++ /* General case: + + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] +@@ -8710,10 +8711,10 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, 16] + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.hard_fp_offset; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; +- frame.final_adjust = crtl->outgoing_args_size; ++ frame.final_adjust = frame.bytes_below_saved_regs; + } + + /* Make sure the individual adjustments add up to the full frame size. */ +@@ -9358,7 +9359,7 @@ aarch64_get_separate_components (void) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -9507,7 +9508,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset -= frame.below_hard_fp_saved_regs_size; + else +- offset += crtl->outgoing_args_size; ++ offset += frame.bytes_below_saved_regs; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -9561,7 +9562,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + if (frame_pointer_needed) + offset2 -= frame.below_hard_fp_saved_regs_size; + else +- offset2 += crtl->outgoing_args_size; ++ offset2 += frame.bytes_below_saved_regs; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -9635,10 +9636,10 @@ aarch64_stack_clash_protection_alloca_probe_range (void) + registers. If POLY_SIZE is not large enough to require a probe this function + will only adjust the stack. When allocating the stack space + FRAME_RELATED_P is then used to indicate if the allocation is frame related. +- FINAL_ADJUSTMENT_P indicates whether we are allocating the outgoing +- arguments. If we are then we ensure that any allocation larger than the ABI +- defined buffer needs a probe so that the invariant of having a 1KB buffer is +- maintained. ++ FINAL_ADJUSTMENT_P indicates whether we are allocating the area below ++ the saved registers. If we are then we ensure that any allocation ++ larger than the ABI defined buffer needs a probe so that the ++ invariant of having a 1KB buffer is maintained. + + We emit barriers after each stack adjustment to prevent optimizations from + breaking the invariant that we never drop the stack more than a page. This +@@ -9847,7 +9848,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + /* Handle any residuals. Residuals of at least MIN_PROBE_THRESHOLD have to + be probed. This maintains the requirement that each page is probed at + least once. For initial probing we probe only if the allocation is +- more than GUARD_SIZE - buffer, and for the outgoing arguments we probe ++ more than GUARD_SIZE - buffer, and below the saved registers we probe + if the amount is larger than buffer. GUARD_SIZE - buffer + buffer == + GUARD_SIZE. This works that for any allocation that is large enough to + trigger a probe here, we'll have at least one, and if they're not large +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 73b09e205083..0b6faa3ddf17 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -777,6 +777,11 @@ struct GTY (()) aarch64_frame + /* The size of the callee-save registers with a slot in REG_OFFSET. */ + poly_int64 saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the bottom of the register save area. ++ This value is always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_saved_regs; ++ + /* The size of the callee-save registers with a slot in REG_OFFSET that + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0018-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/13.2.0/0018-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..bc9379b4 --- /dev/null +++ b/packages/gcc/13.2.0/0018-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,84 @@ +From 82f6b3e1b596ef0f4e3ac3bb9c6e88fb4458f402 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:14 +0100 +Subject: [PATCH 18/32] aarch64: Add bytes_below_hard_fp to frame info + +Following on from the previous bytes_below_saved_regs patch, this one +records the number of bytes that are below the hard frame pointer. +This eventually replaces below_hard_fp_saved_regs_size. + +If a frame pointer is not needed, the epilogue adds final_adjust +to the stack pointer before restoring registers: + + aarch64_add_sp (tmp1_rtx, tmp0_rtx, final_adjust, true); + +Therefore, if the epilogue needs to restore the stack pointer from +the hard frame pointer, the directly corresponding offset is: + + -bytes_below_hard_fp + final_adjust + +i.e. go from the hard frame pointer to the bottom of the frame, +then add the same amount as if we were using the stack pointer +from the outset. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Initialize it. + (aarch64_expand_epilogue): Use it instead of + below_hard_fp_saved_regs_size. +--- + gcc/config/aarch64/aarch64.cc | 6 +++--- + gcc/config/aarch64/aarch64.h | 5 +++++ + 2 files changed, 8 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index dacc2b0e4dd2..a3f7aabcc594 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8530,6 +8530,7 @@ aarch64_layout_frame (void) + of the callee save area. */ + bool saves_below_hard_fp_p = maybe_ne (offset, 0); + frame.below_hard_fp_saved_regs_size = offset; ++ frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -10171,8 +10172,7 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 final_adjust = frame.final_adjust; + poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_pop_candidate1; + unsigned reg2 = frame.wb_pop_candidate2; + unsigned int last_gpr = (frame.is_scs_enabled +@@ -10230,7 +10230,7 @@ aarch64_expand_epilogue (bool for_sibcall) + is restored on the instruction doing the writeback. */ + aarch64_add_offset (Pmode, stack_pointer_rtx, + hard_frame_pointer_rtx, +- -callee_offset - below_hard_fp_saved_regs_size, ++ -bytes_below_hard_fp + final_adjust, + tmp1_rtx, tmp0_rtx, callee_adjust == 0); + else + /* The case where we need to re-use the register here is very rare, so +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 0b6faa3ddf17..4263d29d29d7 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -786,6 +786,11 @@ struct GTY (()) aarch64_frame + are saved below the hard frame pointer. */ + poly_int64 below_hard_fp_saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0019-aarch64-Tweak-aarch64_save-restore_callee_saves.patch b/packages/gcc/13.2.0/0019-aarch64-Tweak-aarch64_save-restore_callee_saves.patch new file mode 100644 index 00000000..ebe3fe89 --- /dev/null +++ b/packages/gcc/13.2.0/0019-aarch64-Tweak-aarch64_save-restore_callee_saves.patch @@ -0,0 +1,225 @@ +From 86fa43e9fe4a8bf954f2919f07cbe3646d1d1df3 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:14 +0100 +Subject: [PATCH 19/32] aarch64: Tweak aarch64_save/restore_callee_saves + +aarch64_save_callee_saves and aarch64_restore_callee_saves took +a parameter called start_offset that gives the offset of the +bottom of the saved register area from the current stack pointer. +However, it's more convenient for later patches if we use the +bottom of the entire frame as the reference point, rather than +the bottom of the saved registers. + +Doing that removes the need for the callee_offset field. +Other than that, this is not a win on its own. It only really +makes sense in combination with the follow-on patches. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::callee_offset): Delete. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Remove + callee_offset handling. + (aarch64_save_callee_saves): Replace the start_offset parameter + with a bytes_below_sp parameter. + (aarch64_restore_callee_saves): Likewise. + (aarch64_expand_prologue): Update accordingly. + (aarch64_expand_epilogue): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 56 +++++++++++++++++------------------ + gcc/config/aarch64/aarch64.h | 4 --- + 2 files changed, 28 insertions(+), 32 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index a3f7aabcc594..46ae5cf76735 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8604,7 +8604,6 @@ aarch64_layout_frame (void) + frame.final_adjust = 0; + frame.callee_adjust = 0; + frame.sve_callee_adjust = 0; +- frame.callee_offset = 0; + + frame.wb_pop_candidate1 = frame.wb_push_candidate1; + frame.wb_pop_candidate2 = frame.wb_push_candidate2; +@@ -8672,7 +8671,6 @@ aarch64_layout_frame (void) + stp reg1, reg2, [sp, bytes_below_saved_regs] + stp reg3, reg4, [sp, bytes_below_saved_regs + 16] */ + frame.initial_adjust = frame.frame_size; +- frame.callee_offset = const_below_saved_regs; + } + else if (saves_below_hard_fp_p + && known_eq (frame.saved_regs_size, +@@ -9073,12 +9071,13 @@ aarch64_add_cfa_expression (rtx_insn *insn, rtx reg, + } + + /* Emit code to save the callee-saved registers from register number START +- to LIMIT to the stack at the location starting at offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. HARD_FP_VALID_P +- is true if the hard frame pointer has been set up. */ ++ to LIMIT to the stack. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. HARD_FP_VALID_P is true if the hard ++ frame pointer has been set up. */ + + static void +-aarch64_save_callee_saves (poly_int64 start_offset, ++aarch64_save_callee_saves (poly_int64 bytes_below_sp, + unsigned start, unsigned limit, bool skip_wb, + bool hard_fp_valid_p) + { +@@ -9106,7 +9105,9 @@ aarch64_save_callee_saves (poly_int64 start_offset, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -9117,9 +9118,7 @@ aarch64_save_callee_saves (poly_int64 start_offset, + else if (GP_REGNUM_P (regno) + && (!offset.is_constant (&const_offset) || const_offset >= 512)) + { +- gcc_assert (known_eq (start_offset, 0)); +- poly_int64 fp_offset +- = frame.below_hard_fp_saved_regs_size; ++ poly_int64 fp_offset = frame.bytes_below_hard_fp - bytes_below_sp; + if (hard_fp_valid_p) + base_rtx = hard_frame_pointer_rtx; + else +@@ -9183,12 +9182,13 @@ aarch64_save_callee_saves (poly_int64 start_offset, + } + + /* Emit code to restore the callee registers from register number START +- up to and including LIMIT. Restore from the stack offset START_OFFSET, +- skipping any write-back candidates if SKIP_WB is true. Write the +- appropriate REG_CFA_RESTORE notes into CFI_OPS. */ ++ up to and including LIMIT. The stack pointer is currently BYTES_BELOW_SP ++ bytes above the bottom of the static frame. Skip any write-back ++ candidates if SKIP_WB is true. Write the appropriate REG_CFA_RESTORE ++ notes into CFI_OPS. */ + + static void +-aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, ++aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + unsigned limit, bool skip_wb, rtx *cfi_ops) + { + aarch64_frame &frame = cfun->machine->frame; +@@ -9214,7 +9214,9 @@ aarch64_restore_callee_saves (poly_int64 start_offset, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = start_offset + frame.reg_offset[regno]; ++ offset = (frame.reg_offset[regno] ++ + frame.bytes_below_saved_regs ++ - bytes_below_sp); + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -9990,8 +9992,6 @@ aarch64_expand_prologue (void) + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; +- poly_int64 below_hard_fp_saved_regs_size +- = frame.below_hard_fp_saved_regs_size; + unsigned reg1 = frame.wb_push_candidate1; + unsigned reg2 = frame.wb_push_candidate2; + bool emit_frame_chain = frame.emit_frame_chain; +@@ -10067,8 +10067,8 @@ aarch64_expand_prologue (void) + - frame.hard_fp_offset); + gcc_assert (known_ge (chain_offset, 0)); + +- /* The offset of the bottom of the save area from the current SP. */ +- poly_int64 saved_regs_offset = chain_offset - below_hard_fp_saved_regs_size; ++ /* The offset of the current SP from the bottom of the static frame. */ ++ poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { +@@ -10076,7 +10076,7 @@ aarch64_expand_prologue (void) + { + reg1 = R29_REGNUM; + reg2 = R30_REGNUM; +- aarch64_save_callee_saves (saved_regs_offset, reg1, reg2, ++ aarch64_save_callee_saves (bytes_below_sp, reg1, reg2, + false, false); + } + else +@@ -10116,7 +10116,7 @@ aarch64_expand_prologue (void) + emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + +- aarch64_save_callee_saves (saved_regs_offset, R0_REGNUM, R30_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, R0_REGNUM, R30_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + if (maybe_ne (sve_callee_adjust, 0)) +@@ -10126,16 +10126,17 @@ aarch64_expand_prologue (void) + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, + sve_callee_adjust, + !frame_pointer_needed, false); +- saved_regs_offset += sve_callee_adjust; ++ bytes_below_sp -= sve_callee_adjust; + } +- aarch64_save_callee_saves (saved_regs_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, P0_REGNUM, P15_REGNUM, + false, emit_frame_chain); +- aarch64_save_callee_saves (saved_regs_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_save_callee_saves (bytes_below_sp, V0_REGNUM, V31_REGNUM, + callee_adjust != 0 || emit_frame_chain, + emit_frame_chain); + + /* We may need to probe the final adjustment if it is larger than the guard + that is assumed by the called. */ ++ gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); + } +@@ -10170,7 +10171,6 @@ aarch64_expand_epilogue (bool for_sibcall) + poly_int64 initial_adjust = frame.initial_adjust; + HOST_WIDE_INT callee_adjust = frame.callee_adjust; + poly_int64 final_adjust = frame.final_adjust; +- poly_int64 callee_offset = frame.callee_offset; + poly_int64 sve_callee_adjust = frame.sve_callee_adjust; + poly_int64 bytes_below_hard_fp = frame.bytes_below_hard_fp; + unsigned reg1 = frame.wb_pop_candidate1; +@@ -10240,9 +10240,9 @@ aarch64_expand_epilogue (bool for_sibcall) + + /* Restore the vector registers before the predicate registers, + so that we can use P4 as a temporary for big-endian SVE frames. */ +- aarch64_restore_callee_saves (callee_offset, V0_REGNUM, V31_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, V0_REGNUM, V31_REGNUM, + callee_adjust != 0, &cfi_ops); +- aarch64_restore_callee_saves (callee_offset, P0_REGNUM, P15_REGNUM, ++ aarch64_restore_callee_saves (final_adjust, P0_REGNUM, P15_REGNUM, + false, &cfi_ops); + if (maybe_ne (sve_callee_adjust, 0)) + aarch64_add_sp (NULL_RTX, NULL_RTX, sve_callee_adjust, true); +@@ -10250,7 +10250,7 @@ aarch64_expand_epilogue (bool for_sibcall) + /* When shadow call stack is enabled, the scs_pop in the epilogue will + restore x30, we don't need to restore x30 again in the traditional + way. */ +- aarch64_restore_callee_saves (callee_offset - sve_callee_adjust, ++ aarch64_restore_callee_saves (final_adjust + sve_callee_adjust, + R0_REGNUM, last_gpr, + callee_adjust != 0, &cfi_ops); + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 4263d29d29d7..fd820b1be4eb 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -813,10 +813,6 @@ struct GTY (()) aarch64_frame + It is zero when no push is used. */ + HOST_WIDE_INT callee_adjust; + +- /* The offset from SP to the callee-save registers after initial_adjust. +- It may be non-zero if no push is used (ie. callee_adjust == 0). */ +- poly_int64 callee_offset; +- + /* The size of the stack adjustment before saving or after restoring + SVE registers. */ + poly_int64 sve_callee_adjust; +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0020-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch b/packages/gcc/13.2.0/0020-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch new file mode 100644 index 00000000..0c62ada0 --- /dev/null +++ b/packages/gcc/13.2.0/0020-aarch64-Only-calculate-chain_offset-if-there-is-a-ch.patch @@ -0,0 +1,45 @@ +From 8ae9181426f2700c2e5a2909487fa630e6fa406b Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:15 +0100 +Subject: [PATCH 20/32] aarch64: Only calculate chain_offset if there is a + chain + +After previous patches, it is no longer necessary to calculate +a chain_offset in cases where there is no chain record. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_expand_prologue): Move the + calculation of chain_offset into the emit_frame_chain block. +--- + gcc/config/aarch64/aarch64.cc | 10 +++++----- + 1 file changed, 5 insertions(+), 5 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 46ae5cf76735..0e9b9717c085 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -10062,16 +10062,16 @@ aarch64_expand_prologue (void) + if (callee_adjust != 0) + aarch64_push_regs (reg1, reg2, callee_adjust); + +- /* The offset of the frame chain record (if any) from the current SP. */ +- poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); +- gcc_assert (known_ge (chain_offset, 0)); +- + /* The offset of the current SP from the bottom of the static frame. */ + poly_int64 bytes_below_sp = frame_size - initial_adjust - callee_adjust; + + if (emit_frame_chain) + { ++ /* The offset of the frame chain record (if any) from the current SP. */ ++ poly_int64 chain_offset = (initial_adjust + callee_adjust ++ - frame.hard_fp_offset); ++ gcc_assert (known_ge (chain_offset, 0)); ++ + if (callee_adjust == 0) + { + reg1 = R29_REGNUM; +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0021-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/13.2.0/0021-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..f00913e8 --- /dev/null +++ b/packages/gcc/13.2.0/0021-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,91 @@ +From 375794feb614cee1f41b710b9cc1b6f25da6c1cb Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:15 +0100 +Subject: [PATCH 21/32] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.cc (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.cc | 6 +++--- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 6 insertions(+), 6 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 0e9b9717c085..0a22f91520e5 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8598,7 +8598,7 @@ aarch64_layout_frame (void) + STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; + +- frame.locals_offset = frame.saved_varargs_size; ++ frame.bytes_above_locals = frame.saved_varargs_size; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +@@ -12754,13 +12754,13 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + return frame.hard_fp_offset; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.locals_offset; ++ return frame.hard_fp_offset - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return frame.frame_size - frame.locals_offset; ++ return frame.frame_size - frame.bytes_above_locals; + } + + return frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index fd820b1be4eb..7ae12d13e2b4 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -791,10 +791,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 locals_offset; ++ poly_int64 bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0022-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/13.2.0/0022-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..b8fac757 --- /dev/null +++ b/packages/gcc/13.2.0/0022-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,148 @@ +From 1a9ea1c45c75615ffbfabe652b3598a1d7be2168 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:16 +0100 +Subject: [PATCH 22/32] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.cc (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 26 +++++++++++++------------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 16 insertions(+), 16 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 0a22f91520e5..95499ae49ba2 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8590,7 +8590,7 @@ aarch64_layout_frame (void) + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.hard_fp_offset ++ frame.bytes_above_hard_fp + = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; + + /* Both these values are already aligned. */ +@@ -8639,13 +8639,13 @@ aarch64_layout_frame (void) + else if (frame.wb_pop_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- HOST_WIDE_INT const_size, const_below_saved_regs, const_fp_offset; ++ HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; + if (known_eq (frame.saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +- && known_eq (frame.hard_fp_offset, const_size)) ++ && known_eq (frame.bytes_above_hard_fp, const_size)) + { + /* Simple, small frame with no data below the saved registers. + +@@ -8662,8 +8662,8 @@ aarch64_layout_frame (void) + case that it hardly seems worth the effort though. */ + && (!saves_below_hard_fp_p || const_below_saved_regs == 0) + && !(cfun->calls_alloca +- && frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset)) ++ && frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset)) + { + /* Frame with small area below the saved registers: + +@@ -8681,12 +8681,12 @@ aarch64_layout_frame (void) + sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.hard_fp_offset ++ frame.initial_adjust = (frame.bytes_above_hard_fp + + frame.below_hard_fp_saved_regs_size); + frame.final_adjust = frame.bytes_below_saved_regs; + } +- else if (frame.hard_fp_offset.is_constant (&const_fp_offset) +- && const_fp_offset < max_push_offset) ++ else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) ++ && const_above_fp < max_push_offset) + { + /* Frame with large area below the saved registers, or with SVE saves, + but with a small area above: +@@ -8696,7 +8696,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.callee_adjust = const_fp_offset; ++ frame.callee_adjust = const_above_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -8711,7 +8711,7 @@ aarch64_layout_frame (void) + [sub sp, sp, below_hard_fp_saved_regs_size] + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = frame.hard_fp_offset; ++ frame.initial_adjust = frame.bytes_above_hard_fp; + frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } +@@ -10069,7 +10069,7 @@ aarch64_expand_prologue (void) + { + /* The offset of the frame chain record (if any) from the current SP. */ + poly_int64 chain_offset = (initial_adjust + callee_adjust +- - frame.hard_fp_offset); ++ - frame.bytes_above_hard_fp); + gcc_assert (known_ge (chain_offset, 0)); + + if (callee_adjust == 0) +@@ -12751,10 +12751,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return frame.hard_fp_offset; ++ return frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return frame.hard_fp_offset - frame.bytes_above_locals; ++ return frame.bytes_above_hard_fp - frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 7ae12d13e2b4..3808f49e9ca5 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -796,10 +796,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 hard_fp_offset; ++ poly_int64 bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0023-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/13.2.0/0023-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..c6f66660 --- /dev/null +++ b/packages/gcc/13.2.0/0023-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,35 @@ +From d202ce1ecf60a36a3e1009917dd76109248ce9be Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:16 +0100 +Subject: [PATCH 23/32] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 3808f49e9ca5..108a5731b0d7 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -801,8 +801,8 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- frame (incomming SP) to the stack_pointer. This value is always ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always + a multiple of STACK_BOUNDARY. */ + poly_int64 frame_size; + +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0024-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch b/packages/gcc/13.2.0/0024-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch new file mode 100644 index 00000000..709a632b --- /dev/null +++ b/packages/gcc/13.2.0/0024-aarch64-Measure-reg_offset-from-the-bottom-of-the-fr.patch @@ -0,0 +1,196 @@ +From f2b585375205b0a1802d79c682ba33766ecd1f0f Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:17 +0100 +Subject: [PATCH 24/32] aarch64: Measure reg_offset from the bottom of the + frame + +reg_offset was measured from the bottom of the saved register area. +This made perfect sense with the original layout, since the bottom +of the saved register area was also the hard frame pointer address. +It became slightly less obvious with SVE, since we save SVE +registers below the hard frame pointer, but it still made sense. + +However, if we want to allow different frame layouts, it's more +convenient and obvious to measure reg_offset from the bottom of +the frame. After previous patches, it's also a slight simplification +in its own right. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame): Add comment above + reg_offset. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Walk offsets + from the bottom of the frame, rather than the bottom of the saved + register area. Measure reg_offset from the bottom of the frame + rather than the bottom of the saved register area. + (aarch64_save_callee_saves): Update accordingly. + (aarch64_restore_callee_saves): Likewise. + (aarch64_get_separate_components): Likewise. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.cc | 53 ++++++++++++++++------------------- + gcc/config/aarch64/aarch64.h | 3 ++ + 2 files changed, 27 insertions(+), 29 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 95499ae49ba2..af99807ef8ab 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8400,7 +8400,6 @@ aarch64_needs_frame_chain (void) + static void + aarch64_layout_frame (void) + { +- poly_int64 offset = 0; + int regno, last_fp_reg = INVALID_REGNUM; + machine_mode vector_save_mode = aarch64_reg_save_mode (V8_REGNUM); + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); +@@ -8478,7 +8477,9 @@ aarch64_layout_frame (void) + gcc_assert (crtl->is_leaf + || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + +- frame.bytes_below_saved_regs = crtl->outgoing_args_size; ++ poly_int64 offset = crtl->outgoing_args_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.bytes_below_saved_regs = offset; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -8490,7 +8491,8 @@ aarch64_layout_frame (void) + offset += BYTES_PER_SVE_PRED; + } + +- if (maybe_ne (offset, 0)) ++ poly_int64 saved_prs_size = offset - frame.bytes_below_saved_regs; ++ if (maybe_ne (saved_prs_size, 0)) + { + /* If we have any vector registers to save above the predicate registers, + the offset of the vector register save slots need to be a multiple +@@ -8508,10 +8510,10 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + else + { +- if (known_le (offset, vector_save_size)) +- offset = vector_save_size; +- else if (known_le (offset, vector_save_size * 2)) +- offset = vector_save_size * 2; ++ if (known_le (saved_prs_size, vector_save_size)) ++ offset = frame.bytes_below_saved_regs + vector_save_size; ++ else if (known_le (saved_prs_size, vector_save_size * 2)) ++ offset = frame.bytes_below_saved_regs + vector_save_size * 2; + else + gcc_unreachable (); + } +@@ -8528,9 +8530,10 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- bool saves_below_hard_fp_p = maybe_ne (offset, 0); +- frame.below_hard_fp_saved_regs_size = offset; +- frame.bytes_below_hard_fp = offset + frame.bytes_below_saved_regs; ++ frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p ++ = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ frame.bytes_below_hard_fp = offset; + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -8581,9 +8584,10 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset; ++ frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size = offset + frame.saved_varargs_size; ++ poly_int64 varargs_and_saved_regs_size ++ = frame.saved_regs_size + frame.saved_varargs_size; + + poly_int64 saved_regs_and_above + = aligned_upper_bound (varargs_and_saved_regs_size +@@ -9105,9 +9109,7 @@ aarch64_save_callee_saves (poly_int64 bytes_below_sp, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + poly_int64 sp_offset = offset; + +@@ -9214,9 +9216,7 @@ aarch64_restore_callee_saves (poly_int64 bytes_below_sp, unsigned start, + + machine_mode mode = aarch64_reg_save_mode (regno); + reg = gen_rtx_REG (mode, regno); +- offset = (frame.reg_offset[regno] +- + frame.bytes_below_saved_regs +- - bytes_below_sp); ++ offset = frame.reg_offset[regno] - bytes_below_sp; + rtx base_rtx = stack_pointer_rtx; + if (mode == VNx2DImode && BYTES_BIG_ENDIAN) + aarch64_adjust_sve_callee_save_base (mode, base_rtx, anchor_reg, +@@ -9355,14 +9355,12 @@ aarch64_get_separate_components (void) + it as a stack probe for -fstack-clash-protection. */ + if (flag_stack_clash_protection + && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, 0)) ++ && known_eq (offset, frame.bytes_below_saved_regs)) + continue; + + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ +@@ -9509,9 +9507,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + poly_int64 offset = frame.reg_offset[regno]; + if (frame_pointer_needed) +- offset -= frame.below_hard_fp_saved_regs_size; +- else +- offset += frame.bytes_below_saved_regs; ++ offset -= frame.bytes_below_hard_fp; + + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); +@@ -9563,9 +9559,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (frame_pointer_needed) +- offset2 -= frame.below_hard_fp_saved_regs_size; +- else +- offset2 += frame.bytes_below_saved_regs; ++ offset2 -= frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -9681,7 +9675,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p + && known_eq (frame.below_hard_fp_saved_regs_size, 0)) + { +- poly_int64 lr_offset = frame.reg_offset[LR_REGNUM]; ++ poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] ++ - frame.bytes_below_saved_regs); + if (known_ge (lr_offset, 0)) + min_probe_threshold -= lr_offset.to_constant (); + else +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 108a5731b0d7..c8becb098c88 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -766,6 +766,9 @@ extern enum aarch64_processor aarch64_tune; + #ifdef HAVE_POLY_INT_H + struct GTY (()) aarch64_frame + { ++ /* The offset from the bottom of the static frame (the bottom of the ++ outgoing arguments) of each register save slot, or -2 if no save is ++ needed. */ + poly_int64 reg_offset[LAST_SAVED_REGNUM + 1]; + + /* The number of extra stack bytes taken up by register varargs. +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0025-aarch64-Simplify-top-of-frame-allocation.patch b/packages/gcc/13.2.0/0025-aarch64-Simplify-top-of-frame-allocation.patch new file mode 100644 index 00000000..509050a0 --- /dev/null +++ b/packages/gcc/13.2.0/0025-aarch64-Simplify-top-of-frame-allocation.patch @@ -0,0 +1,55 @@ +From 79faabda181d0d9fd29a3cf5726ba65bdee945b5 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:17 +0100 +Subject: [PATCH 25/32] aarch64: Simplify top of frame allocation + +After previous patches, it no longer really makes sense to allocate +the top of the frame in terms of varargs_and_saved_regs_size and +saved_regs_and_above. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Simplify + the allocation of the top of the frame. +--- + gcc/config/aarch64/aarch64.cc | 23 ++++++++--------------- + 1 file changed, 8 insertions(+), 15 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index af99807ef8ab..31b00094c2a7 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8586,23 +8586,16 @@ aarch64_layout_frame (void) + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; + +- poly_int64 varargs_and_saved_regs_size +- = frame.saved_regs_size + frame.saved_varargs_size; ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ auto top_of_locals = offset; + +- poly_int64 saved_regs_and_above +- = aligned_upper_bound (varargs_and_saved_regs_size +- + get_frame_size (), +- STACK_BOUNDARY / BITS_PER_UNIT); ++ offset += frame.saved_varargs_size; ++ gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ frame.frame_size = offset; + +- frame.bytes_above_hard_fp +- = saved_regs_and_above - frame.below_hard_fp_saved_regs_size; +- +- /* Both these values are already aligned. */ +- gcc_assert (multiple_p (frame.bytes_below_saved_regs, +- STACK_BOUNDARY / BITS_PER_UNIT)); +- frame.frame_size = saved_regs_and_above + frame.bytes_below_saved_regs; +- +- frame.bytes_above_locals = frame.saved_varargs_size; ++ frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; + frame.final_adjust = 0; +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0026-aarch64-Minor-initial-adjustment-tweak.patch b/packages/gcc/13.2.0/0026-aarch64-Minor-initial-adjustment-tweak.patch new file mode 100644 index 00000000..4de884f3 --- /dev/null +++ b/packages/gcc/13.2.0/0026-aarch64-Minor-initial-adjustment-tweak.patch @@ -0,0 +1,38 @@ +From 4e62049e403b141e6f916176160dac8cbd65fe47 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:18 +0100 +Subject: [PATCH 26/32] aarch64: Minor initial adjustment tweak + +This patch just changes a calculation of initial_adjust +to one that makes it slightly more obvious that the total +adjustment is frame.frame_size. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Tweak + calculation of initial_adjust for frames in which all saves + are SVE saves. +--- + gcc/config/aarch64/aarch64.cc | 5 ++--- + 1 file changed, 2 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 31b00094c2a7..1aa79da0673c 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8675,11 +8675,10 @@ aarch64_layout_frame (void) + { + /* Frame in which all saves are SVE saves: + +- sub sp, sp, hard_fp_offset + below_hard_fp_saved_regs_size ++ sub sp, sp, frame_size - bytes_below_saved_regs + save SVE registers relative to SP + sub sp, sp, bytes_below_saved_regs */ +- frame.initial_adjust = (frame.bytes_above_hard_fp +- + frame.below_hard_fp_saved_regs_size); ++ frame.initial_adjust = frame.frame_size - frame.bytes_below_saved_regs; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else if (frame.bytes_above_hard_fp.is_constant (&const_above_fp) +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0027-aarch64-Tweak-stack-clash-boundary-condition.patch b/packages/gcc/13.2.0/0027-aarch64-Tweak-stack-clash-boundary-condition.patch new file mode 100644 index 00000000..cc376eb0 --- /dev/null +++ b/packages/gcc/13.2.0/0027-aarch64-Tweak-stack-clash-boundary-condition.patch @@ -0,0 +1,125 @@ +From aaa1a0a5912d9e5d571e5f1c6f09ceac99544ab5 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:18 +0100 +Subject: [PATCH 27/32] aarch64: Tweak stack clash boundary condition + +The AArch64 ABI says that, when stack clash protection is used, +there can be a maximum of 1KiB of unprobed space at sp on entry +to a function. Therefore, we need to probe when allocating +>= guard_size - 1KiB of data (>= rather than >). This is what +GCC does. + +If an allocation is exactly guard_size bytes, it is enough to allocate +those bytes and probe once at offset 1024. It isn't possible to use a +single probe at any other offset: higher would conmplicate later code, +by leaving more unprobed space than usual, while lower would risk +leaving an entire page unprobed. For simplicity, the code probes all +allocations at offset 1024. + +Some register saves also act as probes. If we need to allocate +more space below the last such register save probe, we need to +probe the allocation if it is > 1KiB. Again, this allocation is +then sometimes (but not always) probed at offset 1024. This sort of +allocation is currently only used for outgoing arguments, which are +rarely this big. + +However, the code also probed if this final outgoing-arguments +allocation was == 1KiB, rather than just > 1KiB. This isn't +necessary, since the register save then probes at offset 1024 +as required. Continuing to probe allocations of exactly 1KiB +would complicate later patches. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_allocate_and_probe_stack_space): + Don't probe final allocations that are exactly 1KiB in size (after + unprobed space above the final allocation has been deducted). + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: New test. +--- + gcc/config/aarch64/aarch64.cc | 4 +- + .../aarch64/stack-check-prologue-17.c | 55 +++++++++++++++++++ + 2 files changed, 58 insertions(+), 1 deletion(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 1aa79da0673c..5cad847977a4 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9648,9 +9648,11 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT guard_size + = 1 << param_stack_clash_protection_guard_size; + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; ++ HOST_WIDE_INT byte_sp_alignment = STACK_BOUNDARY / BITS_PER_UNIT; ++ gcc_assert (multiple_p (poly_size, byte_sp_alignment)); + HOST_WIDE_INT min_probe_threshold + = (final_adjustment_p +- ? guard_used_by_caller ++ ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); + /* When doing the final adjustment for the outgoing arguments, take into + account any unprobed space there is above the current SP. There are +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +new file mode 100644 +index 000000000000..0d8a25d73a24 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -0,0 +1,55 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0028-aarch64-Put-LR-save-probe-in-first-16-bytes.patch b/packages/gcc/13.2.0/0028-aarch64-Put-LR-save-probe-in-first-16-bytes.patch new file mode 100644 index 00000000..4ba2df93 --- /dev/null +++ b/packages/gcc/13.2.0/0028-aarch64-Put-LR-save-probe-in-first-16-bytes.patch @@ -0,0 +1,406 @@ +From 8433953434a7b58c0923140d39eb3c5988c1d097 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:19 +0100 +Subject: [PATCH 28/32] aarch64: Put LR save probe in first 16 bytes + +-fstack-clash-protection uses the save of LR as a probe for the next +allocation. The next allocation could be: + +* another part of the static frame, e.g. when allocating SVE save slots + or outgoing arguments + +* an alloca in the same function + +* an allocation made by a callee function + +However, when -fomit-frame-pointer is used, the LR save slot is placed +above the other GPR save slots. It could therefore be up to 80 bytes +above the base of the GPR save area (which is also the hard fp address). + +aarch64_allocate_and_probe_stack_space took this into account when +deciding how much subsequent space could be allocated without needing +a probe. However, it interacted badly with: + + /* If doing a small final adjustment, we always probe at offset 0. + This is done to avoid issues when LR is not at position 0 or when + the final adjustment is smaller than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +which forces any allocation that is smaller than the guard page size +to be probed at offset 0 rather than the usual offset 1024. It was +therefore possible to construct cases in which we had: + +* a probe using LR at SP + 80 bytes (or some other value >= 16) +* an allocation of the guard page size - 16 bytes +* a probe at SP + 0 + +which allocates guard page size + 64 consecutive unprobed bytes. + +This patch requires the LR probe to be in the first 16 bytes of the +save area when stack clash protection is active. Doing it +unconditionally would cause code-quality regressions. + +Putting LR before other registers prevents push/pop allocation +when shadow call stacks are enabled, since LR is restored +separately from the other callee-saved registers. + +The new comment doesn't say that the probe register is required +to be LR, since a later patch removes that restriction. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_layout_frame): Ensure that + the LR save slot is in the first 16 bytes of the register save area. + Only form STP/LDP push/pop candidates if both registers are valid. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + when LR was not in the first 16 bytes. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-18.c: New test. + * gcc.target/aarch64/stack-check-prologue-19.c: Likewise. + * gcc.target/aarch64/stack-check-prologue-20.c: Likewise. +--- + gcc/config/aarch64/aarch64.cc | 72 ++++++------- + .../aarch64/stack-check-prologue-18.c | 100 ++++++++++++++++++ + .../aarch64/stack-check-prologue-19.c | 100 ++++++++++++++++++ + .../aarch64/stack-check-prologue-20.c | 3 + + 4 files changed, 233 insertions(+), 42 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 5cad847977a4..a765f92329d3 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8534,26 +8534,34 @@ aarch64_layout_frame (void) + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); + frame.bytes_below_hard_fp = offset; ++ ++ auto allocate_gpr_slot = [&](unsigned int regno) ++ { ++ frame.reg_offset[regno] = offset; ++ if (frame.wb_push_candidate1 == INVALID_REGNUM) ++ frame.wb_push_candidate1 = regno; ++ else if (frame.wb_push_candidate2 == INVALID_REGNUM) ++ frame.wb_push_candidate2 = regno; ++ offset += UNITS_PER_WORD; ++ }; ++ + if (frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +- frame.reg_offset[R29_REGNUM] = offset; +- frame.wb_push_candidate1 = R29_REGNUM; +- frame.reg_offset[R30_REGNUM] = offset + UNITS_PER_WORD; +- frame.wb_push_candidate2 = R30_REGNUM; +- offset += 2 * UNITS_PER_WORD; ++ allocate_gpr_slot (R29_REGNUM); ++ allocate_gpr_slot (R30_REGNUM); + } ++ else if (flag_stack_clash_protection ++ && known_eq (frame.reg_offset[R30_REGNUM], SLOT_REQUIRED)) ++ /* Put the LR save slot first, since it makes a good choice of probe ++ for stack clash purposes. The idea is that the link register usually ++ has to be saved before a call anyway, and so we lose little by ++ stopping it from being individually shrink-wrapped. */ ++ allocate_gpr_slot (R30_REGNUM); + + for (regno = R0_REGNUM; regno <= R30_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) +- { +- frame.reg_offset[regno] = offset; +- if (frame.wb_push_candidate1 == INVALID_REGNUM) +- frame.wb_push_candidate1 = regno; +- else if (frame.wb_push_candidate2 == INVALID_REGNUM) +- frame.wb_push_candidate2 = regno; +- offset += UNITS_PER_WORD; +- } ++ allocate_gpr_slot (regno); + + poly_int64 max_int_offset = offset; + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8631,10 +8639,13 @@ aarch64_layout_frame (void) + max_push_offset to 0, because no registers are popped at this time, + so callee_adjust cannot be adjusted. */ + HOST_WIDE_INT max_push_offset = 0; +- if (frame.wb_pop_candidate2 != INVALID_REGNUM) +- max_push_offset = 512; +- else if (frame.wb_pop_candidate1 != INVALID_REGNUM) +- max_push_offset = 256; ++ if (frame.wb_pop_candidate1 != INVALID_REGNUM) ++ { ++ if (frame.wb_pop_candidate2 != INVALID_REGNUM) ++ max_push_offset = 512; ++ else ++ max_push_offset = 256; ++ } + + HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; +@@ -9654,29 +9665,6 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + = (final_adjustment_p + ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller); +- /* When doing the final adjustment for the outgoing arguments, take into +- account any unprobed space there is above the current SP. There are +- two cases: +- +- - When saving SVE registers below the hard frame pointer, we force +- the lowest save to take place in the prologue before doing the final +- adjustment (i.e. we don't allow the save to be shrink-wrapped). +- This acts as a probe at SP, so there is no unprobed space. +- +- - When there are no SVE register saves, we use the store of the link +- register as a probe. We can't assume that LR was saved at position 0 +- though, so treat any space below it as unprobed. */ +- if (final_adjustment_p +- && known_eq (frame.below_hard_fp_saved_regs_size, 0)) +- { +- poly_int64 lr_offset = (frame.reg_offset[LR_REGNUM] +- - frame.bytes_below_saved_regs); +- if (known_ge (lr_offset, 0)) +- min_probe_threshold -= lr_offset.to_constant (); +- else +- gcc_assert (!flag_stack_clash_protection || known_eq (poly_size, 0)); +- } +- + poly_int64 frame_size = frame.frame_size; + + /* We should always have a positive probe threshold. */ +@@ -9856,8 +9844,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; + /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when LR is not at position 0 or when +- the final adjustment is smaller than the probing offset. */ ++ This is done to avoid issues when the final adjustment is smaller ++ than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +new file mode 100644 +index 000000000000..82447d20fff5 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -0,0 +1,100 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +new file mode 100644 +index 000000000000..73ac3e4e4eb0 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +@@ -0,0 +1,100 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12 -fsanitize=shadow-call-stack -ffixed-x18" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c +new file mode 100644 +index 000000000000..690aae8dfd5b +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-20.c +@@ -0,0 +1,3 @@ ++/* { dg-options "-O2 -fstack-protector-all -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12 -fsanitize=shadow-call-stack -ffixed-x18" } */ ++ ++#include "stack-check-prologue-19.c" +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0029-aarch64-Simplify-probe-of-final-frame-allocation.patch b/packages/gcc/13.2.0/0029-aarch64-Simplify-probe-of-final-frame-allocation.patch new file mode 100644 index 00000000..5e031466 --- /dev/null +++ b/packages/gcc/13.2.0/0029-aarch64-Simplify-probe-of-final-frame-allocation.patch @@ -0,0 +1,123 @@ +From eea1759073e09dd1aefbc9a881601ab1eebfdd18 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:19 +0100 +Subject: [PATCH 29/32] aarch64: Simplify probe of final frame allocation + +Previous patches ensured that the final frame allocation only needs +a probe when the size is strictly greater than 1KiB. It's therefore +safe to use the normal 1024 probe offset in all cases. + +The main motivation for doing this is to simplify the code and +remove the number of special cases. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_allocate_and_probe_stack_space): + Always probe the residual allocation at offset 1024, asserting + that that is in range. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: Expect the probe + to be at offset 1024 rather than offset 0. + * gcc.target/aarch64/stack-check-prologue-18.c: Likewise. + * gcc.target/aarch64/stack-check-prologue-19.c: Likewise. +--- + gcc/config/aarch64/aarch64.cc | 12 ++++-------- + .../gcc.target/aarch64/stack-check-prologue-17.c | 2 +- + .../gcc.target/aarch64/stack-check-prologue-18.c | 4 ++-- + .../gcc.target/aarch64/stack-check-prologue-19.c | 4 ++-- + 4 files changed, 9 insertions(+), 13 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index a765f92329d3..37809a306f79 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -9838,16 +9838,12 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + are still safe. */ + if (residual) + { +- HOST_WIDE_INT residual_probe_offset = guard_used_by_caller; ++ gcc_assert (guard_used_by_caller + byte_sp_alignment <= size); ++ + /* If we're doing final adjustments, and we've done any full page + allocations then any residual needs to be probed. */ + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; +- /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when the final adjustment is smaller +- than the probing offset. */ +- else if (final_adjustment_p && rounded_size == 0) +- residual_probe_offset = 0; + + aarch64_sub_sp (temp1, temp2, residual, frame_related_p); + if (residual >= min_probe_threshold) +@@ -9858,8 +9854,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT_PRINT_DEC " bytes, probing will be required." + "\n", residual); + +- emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, +- residual_probe_offset)); ++ emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, ++ guard_used_by_caller)); + emit_insn (gen_blockage ()); + } + } +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +index 0d8a25d73a24..f0ec1389771d 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -33,7 +33,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +index 82447d20fff5..6383bec5ebcd 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -9,7 +9,7 @@ void g(); + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -50,7 +50,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +index 73ac3e4e4eb0..562039b5e9b8 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-19.c +@@ -9,7 +9,7 @@ void g(); + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -50,7 +50,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0030-aarch64-Explicitly-record-probe-registers-in-frame-i.patch b/packages/gcc/13.2.0/0030-aarch64-Explicitly-record-probe-registers-in-frame-i.patch new file mode 100644 index 00000000..8e3ae4b7 --- /dev/null +++ b/packages/gcc/13.2.0/0030-aarch64-Explicitly-record-probe-registers-in-frame-i.patch @@ -0,0 +1,278 @@ +From 96d85187c3b9c9a7efc2fd698c3d452e80d8aa47 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:20 +0100 +Subject: [PATCH 30/32] aarch64: Explicitly record probe registers in frame + info + +The stack frame is currently divided into three areas: + +A: the area above the hard frame pointer +B: the SVE saves below the hard frame pointer +C: the outgoing arguments + +If the stack frame is allocated in one chunk, the allocation needs a +probe if the frame size is >= guard_size - 1KiB. In addition, if the +function is not a leaf function, it must probe an address no more than +1KiB above the outgoing SP. We ensured the second condition by + +(1) using single-chunk allocations for non-leaf functions only if + the link register save slot is within 512 bytes of the bottom + of the frame; and + +(2) using the link register save as a probe (meaning, for instance, + that it can't be individually shrink wrapped) + +If instead the stack is allocated in multiple chunks, then: + +* an allocation involving only the outgoing arguments (C above) requires + a probe if the allocation size is > 1KiB + +* any other allocation requires a probe if the allocation size + is >= guard_size - 1KiB + +* second and subsequent allocations require the previous allocation + to probe at the bottom of the allocated area, regardless of the size + of that previous allocation + +The final point means that, unlike for single allocations, +it can be necessary to have both a non-SVE register probe and +an SVE register probe. For example: + +* allocate A, probe using a non-SVE register save +* allocate B, probe using an SVE register save +* allocate C + +The non-SVE register used in this case was again the link register. +It was previously used even if the link register save slot was some +bytes above the bottom of the non-SVE register saves, but an earlier +patch avoided that by putting the link register save slot first. + +As a belt-and-braces fix, this patch explicitly records which +probe registers we're using and allows the non-SVE probe to be +whichever register comes first (as for SVE). + +The patch also avoids unnecessary probes in sve/pcs/stack_clash_3.c. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::sve_save_and_probe) + (aarch64_frame::hard_fp_save_and_probe): New fields. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Initialize them. + Rather than asserting that a leaf function saves LR, instead assert + that a leaf function saves something. + (aarch64_get_separate_components): Prevent the chosen probe + registers from being individually shrink-wrapped. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + probe registers that aren't at the bottom of the previous allocation. + +gcc/testsuite/ + * gcc.target/aarch64/sve/pcs/stack_clash_3.c: Avoid redundant probes. +--- + gcc/config/aarch64/aarch64.cc | 68 +++++++++++++++---- + gcc/config/aarch64/aarch64.h | 8 +++ + .../aarch64/sve/pcs/stack_clash_3.c | 6 +- + 3 files changed, 64 insertions(+), 18 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 37809a306f79..6c59c39a639d 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8471,15 +8471,11 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + +- /* With stack-clash, LR must be saved in non-leaf functions. The saving of +- LR counts as an implicit probe which allows us to maintain the invariant +- described in the comment at expand_prologue. */ +- gcc_assert (crtl->is_leaf +- || maybe_ne (frame.reg_offset[R30_REGNUM], SLOT_NOT_REQUIRED)); + + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.bytes_below_saved_regs = offset; ++ frame.sve_save_and_probe = INVALID_REGNUM; + + /* Now assign stack slots for the registers. Start with the predicate + registers, since predicate LDR and STR have a relatively small +@@ -8487,6 +8483,8 @@ aarch64_layout_frame (void) + for (regno = P0_REGNUM; regno <= P15_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += BYTES_PER_SVE_PRED; + } +@@ -8524,6 +8522,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.sve_save_and_probe == INVALID_REGNUM) ++ frame.sve_save_and_probe = regno; + frame.reg_offset[regno] = offset; + offset += vector_save_size; + } +@@ -8533,10 +8533,18 @@ aarch64_layout_frame (void) + frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; + bool saves_below_hard_fp_p + = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ gcc_assert (!saves_below_hard_fp_p ++ || (frame.sve_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.sve_save_and_probe], ++ frame.bytes_below_saved_regs))); ++ + frame.bytes_below_hard_fp = offset; ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; + + auto allocate_gpr_slot = [&](unsigned int regno) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + frame.reg_offset[regno] = offset; + if (frame.wb_push_candidate1 == INVALID_REGNUM) + frame.wb_push_candidate1 = regno; +@@ -8570,6 +8578,8 @@ aarch64_layout_frame (void) + for (regno = V0_REGNUM; regno <= V31_REGNUM; regno++) + if (known_eq (frame.reg_offset[regno], SLOT_REQUIRED)) + { ++ if (frame.hard_fp_save_and_probe == INVALID_REGNUM) ++ frame.hard_fp_save_and_probe = regno; + /* If there is an alignment gap between integer and fp callee-saves, + allocate the last fp register to it if possible. */ + if (regno == last_fp_reg +@@ -8593,6 +8603,17 @@ aarch64_layout_frame (void) + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + + frame.saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (frame.saved_regs_size, ++ frame.below_hard_fp_saved_regs_size) ++ || (frame.hard_fp_save_and_probe != INVALID_REGNUM ++ && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], ++ frame.bytes_below_hard_fp))); ++ ++ /* With stack-clash, a register must be saved in non-leaf functions. ++ The saving of the bottommost register counts as an implicit probe, ++ which allows us to maintain the invariant described in the comment ++ at expand_prologue. */ ++ gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8723,6 +8744,25 @@ aarch64_layout_frame (void) + frame.final_adjust = frame.bytes_below_saved_regs; + } + ++ /* The frame is allocated in pieces, with each non-final piece ++ including a register save at offset 0 that acts as a probe for ++ the following piece. In addition, the save of the bottommost register ++ acts as a probe for callees and allocas. Roll back any probes that ++ aren't needed. ++ ++ A probe isn't needed if it is associated with the final allocation ++ (including callees and allocas) that happens before the epilogue is ++ executed. */ ++ if (crtl->is_leaf ++ && !cfun->calls_alloca ++ && known_eq (frame.final_adjust, 0)) ++ { ++ if (maybe_ne (frame.sve_callee_adjust, 0)) ++ frame.sve_save_and_probe = INVALID_REGNUM; ++ else ++ frame.hard_fp_save_and_probe = INVALID_REGNUM; ++ } ++ + /* Make sure the individual adjustments add up to the full frame size. */ + gcc_assert (known_eq (frame.initial_adjust + + frame.callee_adjust +@@ -9354,13 +9394,6 @@ aarch64_get_separate_components (void) + + poly_int64 offset = frame.reg_offset[regno]; + +- /* If the register is saved in the first SVE save slot, we use +- it as a stack probe for -fstack-clash-protection. */ +- if (flag_stack_clash_protection +- && maybe_ne (frame.below_hard_fp_saved_regs_size, 0) +- && known_eq (offset, frame.bytes_below_saved_regs)) +- continue; +- + /* Get the offset relative to the register we'll use. */ + if (frame_pointer_needed) + offset -= frame.bytes_below_hard_fp; +@@ -9395,6 +9428,13 @@ aarch64_get_separate_components (void) + + bitmap_clear_bit (components, LR_REGNUM); + bitmap_clear_bit (components, SP_REGNUM); ++ if (flag_stack_clash_protection) ++ { ++ if (frame.sve_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.sve_save_and_probe); ++ if (frame.hard_fp_save_and_probe != INVALID_REGNUM) ++ bitmap_clear_bit (components, frame.hard_fp_save_and_probe); ++ } + + return components; + } +@@ -9931,8 +9971,8 @@ aarch64_epilogue_uses (int regno) + When probing is needed, we emit a probe at the start of the prologue + and every PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE bytes thereafter. + +- We have to track how much space has been allocated and the only stores +- to the stack we track as implicit probes are the FP/LR stores. ++ We can also use register saves as probes. These are stored in ++ sve_save_and_probe and hard_fp_save_and_probe. + + For outgoing arguments we probe if the size is larger than 1KB, such that + the ABI specified buffer is maintained for the next callee. +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index c8becb098c88..fbfb73545bae 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -863,6 +863,14 @@ struct GTY (()) aarch64_frame + This is the register they should use. */ + unsigned spare_pred_reg; + ++ /* An SVE register that is saved below the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned sve_save_and_probe; ++ ++ /* A register that is saved at the hard frame pointer and that acts ++ as a probe for later allocations, or INVALID_REGNUM if none. */ ++ unsigned hard_fp_save_and_probe; ++ + bool laid_out; + + /* True if shadow call stack should be enabled for the current function. */ +diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +index 3e01ec36c3a4..3530a0d504ba 100644 +--- a/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c ++++ b/gcc/testsuite/gcc.target/aarch64/sve/pcs/stack_clash_3.c +@@ -11,11 +11,10 @@ + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +@@ -39,13 +38,12 @@ test_1 (int n) + ** mov x11, sp + ** ... + ** sub sp, sp, x13 +-** str p4, \[sp\] + ** cbz w0, [^\n]* ++** str p4, \[sp\] + ** str p5, \[sp, #1, mul vl\] + ** str p6, \[sp, #2, mul vl\] + ** ... + ** ptrue p0\.b, all +-** ldr p4, \[sp\] + ** addvl sp, sp, #1 + ** ldr x24, \[sp\], 32 + ** ret +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0031-aarch64-Remove-below_hard_fp_saved_regs_size.patch b/packages/gcc/13.2.0/0031-aarch64-Remove-below_hard_fp_saved_regs_size.patch new file mode 100644 index 00000000..c9bc9153 --- /dev/null +++ b/packages/gcc/13.2.0/0031-aarch64-Remove-below_hard_fp_saved_regs_size.patch @@ -0,0 +1,157 @@ +From 56df065080950bb30dda9c260f71be54269bdda5 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:20 +0100 +Subject: [PATCH 31/32] aarch64: Remove below_hard_fp_saved_regs_size + +After previous patches, it's no longer necessary to store +saved_regs_size and below_hard_fp_saved_regs_size in the frame info. +All measurements instead use the top or bottom of the frame as +reference points. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::saved_regs_size) + (aarch64_frame::below_hard_fp_saved_regs_size): Delete. + * config/aarch64/aarch64.cc (aarch64_layout_frame): Update accordingly. +--- + gcc/config/aarch64/aarch64.cc | 45 ++++++++++++++++------------------- + gcc/config/aarch64/aarch64.h | 7 ------ + 2 files changed, 21 insertions(+), 31 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index 6c59c39a639d..b95e805a8cc2 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8530,9 +8530,8 @@ aarch64_layout_frame (void) + + /* OFFSET is now the offset of the hard frame pointer from the bottom + of the callee save area. */ +- frame.below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; +- bool saves_below_hard_fp_p +- = maybe_ne (frame.below_hard_fp_saved_regs_size, 0); ++ auto below_hard_fp_saved_regs_size = offset - frame.bytes_below_saved_regs; ++ bool saves_below_hard_fp_p = maybe_ne (below_hard_fp_saved_regs_size, 0); + gcc_assert (!saves_below_hard_fp_p + || (frame.sve_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.sve_save_and_probe], +@@ -8602,9 +8601,8 @@ aarch64_layout_frame (void) + + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); + +- frame.saved_regs_size = offset - frame.bytes_below_saved_regs; +- gcc_assert (known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size) ++ auto saved_regs_size = offset - frame.bytes_below_saved_regs; ++ gcc_assert (known_eq (saved_regs_size, below_hard_fp_saved_regs_size) + || (frame.hard_fp_save_and_probe != INVALID_REGNUM + && known_eq (frame.reg_offset[frame.hard_fp_save_and_probe], + frame.bytes_below_hard_fp))); +@@ -8613,7 +8611,7 @@ aarch64_layout_frame (void) + The saving of the bottommost register counts as an implicit probe, + which allows us to maintain the invariant described in the comment + at expand_prologue. */ +- gcc_assert (crtl->is_leaf || maybe_ne (frame.saved_regs_size, 0)); ++ gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + + offset += get_frame_size (); + offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -8670,7 +8668,7 @@ aarch64_layout_frame (void) + + HOST_WIDE_INT const_size, const_below_saved_regs, const_above_fp; + HOST_WIDE_INT const_saved_regs_size; +- if (known_eq (frame.saved_regs_size, 0)) ++ if (known_eq (saved_regs_size, 0)) + frame.initial_adjust = frame.frame_size; + else if (frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +@@ -8683,7 +8681,7 @@ aarch64_layout_frame (void) + frame.callee_adjust = const_size; + } + else if (frame.bytes_below_saved_regs.is_constant (&const_below_saved_regs) +- && frame.saved_regs_size.is_constant (&const_saved_regs_size) ++ && saved_regs_size.is_constant (&const_saved_regs_size) + && const_below_saved_regs + const_saved_regs_size < 512 + /* We could handle this case even with data below the saved + registers, provided that that data left us with valid offsets +@@ -8702,8 +8700,7 @@ aarch64_layout_frame (void) + frame.initial_adjust = frame.frame_size; + } + else if (saves_below_hard_fp_p +- && known_eq (frame.saved_regs_size, +- frame.below_hard_fp_saved_regs_size)) ++ && known_eq (saved_regs_size, below_hard_fp_saved_regs_size)) + { + /* Frame in which all saves are SVE saves: + +@@ -8725,7 +8722,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.callee_adjust = const_above_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + else +@@ -8740,7 +8737,7 @@ aarch64_layout_frame (void) + [save SVE registers relative to SP] + sub sp, sp, bytes_below_saved_regs */ + frame.initial_adjust = frame.bytes_above_hard_fp; +- frame.sve_callee_adjust = frame.below_hard_fp_saved_regs_size; ++ frame.sve_callee_adjust = below_hard_fp_saved_regs_size; + frame.final_adjust = frame.bytes_below_saved_regs; + } + +@@ -9936,17 +9933,17 @@ aarch64_epilogue_uses (int regno) + | local variables | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | \ +- +-------------------------------+ | +- | callee-saved registers | | frame.saved_regs_size +- +-------------------------------+ | +- | LR' | | +- +-------------------------------+ | +- | FP' | | +- +-------------------------------+ |<- hard_frame_pointer_rtx (aligned) +- | SVE vector registers | | \ +- +-------------------------------+ | | below_hard_fp_saved_regs_size +- | SVE predicate registers | / / ++ | padding | ++ +-------------------------------+ ++ | callee-saved registers | ++ +-------------------------------+ ++ | LR' | ++ +-------------------------------+ ++ | FP' | ++ +-------------------------------+ <-- hard_frame_pointer_rtx (aligned) ++ | SVE vector registers | ++ +-------------------------------+ ++ | SVE predicate registers | + +-------------------------------+ + | dynamic allocation | + +-------------------------------+ +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index fbfb73545bae..cfeaf4657abb 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -777,18 +777,11 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + HOST_WIDE_INT saved_varargs_size; + +- /* The size of the callee-save registers with a slot in REG_OFFSET. */ +- poly_int64 saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the bottom of the register save area. + This value is always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_saved_regs; + +- /* The size of the callee-save registers with a slot in REG_OFFSET that +- are saved below the hard frame pointer. */ +- poly_int64 below_hard_fp_saved_regs_size; +- + /* The number of bytes between the bottom of the static frame (the bottom + of the outgoing arguments) and the hard frame pointer. This value is + always a multiple of STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/13.2.0/0032-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/13.2.0/0032-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..774109d4 --- /dev/null +++ b/packages/gcc/13.2.0/0032-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,299 @@ +From b96e66fd4ef3e36983969fb8cdd1956f551a074b Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 12 Sep 2023 16:07:21 +0100 +Subject: [PATCH 32/32] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.cc (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.cc | 46 +++++++-- + .../gcc.target/aarch64/stack-protector-8.c | 95 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 +++++++ + 3 files changed, 168 insertions(+), 6 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc +index b95e805a8cc2..389c0e293536 100644 +--- a/gcc/config/aarch64/aarch64.cc ++++ b/gcc/config/aarch64/aarch64.cc +@@ -8394,6 +8394,20 @@ aarch64_needs_frame_chain (void) + return aarch64_use_frame_pointer; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -8405,6 +8419,7 @@ aarch64_layout_frame (void) + poly_int64 vector_save_size = GET_MODE_SIZE (vector_save_mode); + bool frame_related_fp_reg_p = false; + aarch64_frame &frame = cfun->machine->frame; ++ poly_int64 top_of_locals = -1; + + frame.emit_frame_chain = aarch64_needs_frame_chain (); + +@@ -8471,9 +8486,16 @@ aarch64_layout_frame (void) + && !crtl->abi->clobbers_full_reg_p (regno)) + frame.reg_offset[regno] = SLOT_REQUIRED; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); + + poly_int64 offset = crtl->outgoing_args_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); ++ if (regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + frame.bytes_below_saved_regs = offset; + frame.sve_save_and_probe = INVALID_REGNUM; + +@@ -8613,15 +8635,18 @@ aarch64_layout_frame (void) + at expand_prologue. */ + gcc_assert (crtl->is_leaf || maybe_ne (saved_regs_size, 0)); + +- offset += get_frame_size (); +- offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); +- auto top_of_locals = offset; +- ++ if (!regs_at_top_p) ++ { ++ offset += get_frame_size (); ++ offset = aligned_upper_bound (offset, STACK_BOUNDARY / BITS_PER_UNIT); ++ top_of_locals = offset; ++ } + offset += frame.saved_varargs_size; + gcc_assert (multiple_p (offset, STACK_BOUNDARY / BITS_PER_UNIT)); + frame.frame_size = offset; + + frame.bytes_above_hard_fp = frame.frame_size - frame.bytes_below_hard_fp; ++ gcc_assert (known_ge (top_of_locals, 0)); + frame.bytes_above_locals = frame.frame_size - top_of_locals; + + frame.initial_adjust = 0; +@@ -9930,10 +9955,10 @@ aarch64_epilogue_uses (int regno) + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | ++ | padding (1) | + +-------------------------------+ + | callee-saved registers | + +-------------------------------+ +@@ -9945,6 +9970,10 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | SVE predicate registers | + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -9954,6 +9983,9 @@ aarch64_epilogue_uses (int regno) + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. +@@ -10149,6 +10181,8 @@ aarch64_expand_prologue (void) + gcc_assert (known_eq (bytes_below_sp, final_adjust)); + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); ++ if (emit_frame_chain && maybe_ne (final_adjust, 0)) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..e71d820e3654 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,95 @@ ++/* { dg-options " -O -fstack-protector-strong -mstack-protector-guard=sysreg -mstack-protector-guard-reg=tpidr2_el0 -mstack-protector-guard-offset=16" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++__SVBool_t *h(void *); ++ ++/* ++** test1: ++** sub sp, sp, #288 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?288 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** sub sp, sp, #1040 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?1032\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?1040 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} ++ ++#pragma GCC target "+sve" ++ ++/* ++** test3: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** addvl sp, sp, #-18 ++** ... ++** str p4, \[sp\] ++** ... ++** sub sp, sp, #272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, #?0 ++** add x0, sp, #?8 ++** bl h ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?272 ++** ... ++** ldr p4, \[sp\] ++** ... ++** addvl sp, sp, #18 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++__SVBool_t test3() { ++ int y[0x40]; ++ return *h(y); ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..58f322aa480a +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=neoverse-v1 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0024-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/7.5.0/0024-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..b2c9005e --- /dev/null +++ b/packages/gcc/7.5.0/0024-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,41 @@ +From 04553281f42e5c4340d554a93d1a73006365d8fb Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 17:00:51 +0100 +Subject: [PATCH 24/30] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.c | 6 ++++-- + 1 file changed, 4 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 4e94be3b0b4d..bb4f6020e5df 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -2901,8 +2901,10 @@ aarch64_layout_frame (void) + else if (cfun->machine->frame.wb_candidate1 != INVALID_REGNUM) + max_push_offset = 256; + +- if (cfun->machine->frame.frame_size < max_push_offset +- && crtl->outgoing_args_size == 0) ++ if (cfun->machine->frame.saved_regs_size == 0) ++ cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; ++ else if (cfun->machine->frame.frame_size < max_push_offset ++ && crtl->outgoing_args_size == 0) + { + /* Simple, small frame with no outgoing arguments: + stp reg1, reg2, [sp, -frame_size]! +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0025-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/7.5.0/0025-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..aeb08bba --- /dev/null +++ b/packages/gcc/7.5.0/0025-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,166 @@ +From e97c93de7ceadad75774f897424a50ece3215129 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 16:55:12 +0100 +Subject: [PATCH 25/30] aarch64: Add bytes_below_hard_fp to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_hard_fp instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 40 +++++++++++++++++++++--------------- + gcc/config/aarch64/aarch64.h | 6 +++++- + 2 files changed, 28 insertions(+), 18 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index bb4f6020e5df..12df8bb783c7 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -2826,6 +2826,8 @@ aarch64_layout_frame (void) + last_fp_reg = regno; + } + ++ cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; ++ + if (frame_pointer_needed) + { + /* FP and LR are placed in the linkage record. */ +@@ -2885,7 +2887,7 @@ aarch64_layout_frame (void) + + cfun->machine->frame.frame_size + = ROUND_UP (cfun->machine->frame.hard_fp_offset +- + crtl->outgoing_args_size, ++ + cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT); + + cfun->machine->frame.locals_offset = cfun->machine->frame.saved_varargs_size; +@@ -2904,32 +2906,33 @@ aarch64_layout_frame (void) + if (cfun->machine->frame.saved_regs_size == 0) + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + else if (cfun->machine->frame.frame_size < max_push_offset +- && crtl->outgoing_args_size == 0) ++ && cfun->machine->frame.bytes_below_hard_fp == 0) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + cfun->machine->frame.callee_adjust = cfun->machine->frame.frame_size; + } +- else if ((crtl->outgoing_args_size ++ else if ((cfun->machine->frame.bytes_below_hard_fp + + cfun->machine->frame.saved_regs_size < 512) + && !(cfun->calls_alloca + && cfun->machine->frame.hard_fp_offset < max_push_offset)) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_hard_fp] ++ stp reg3, reg4, [sp, bytes_below_hard_fp + 16] */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + cfun->machine->frame.callee_offset + = cfun->machine->frame.frame_size - cfun->machine->frame.hard_fp_offset; + } + else if (cfun->machine->frame.hard_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments but a small local area: ++ /* Frame with large area below the saved registers, but with a ++ small area above: + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.callee_adjust = cfun->machine->frame.hard_fp_offset; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.callee_adjust; +@@ -2945,17 +2948,19 @@ aarch64_layout_frame (void) + cfun->machine->frame.callee_adjust = varargs_and_saved_regs_size; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.callee_adjust; ++ cfun->machine->frame.bytes_below_hard_fp ++ = cfun->machine->frame.final_adjust; + cfun->machine->frame.hard_fp_offset = cfun->machine->frame.callee_adjust; + cfun->machine->frame.locals_offset = cfun->machine->frame.hard_fp_offset; + } + else + { +- /* Frame with large local area and outgoing arguments using frame pointer: ++ /* General case: + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] + add x29, sp, 0 + stp reg3, reg4, [sp, 16] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.hard_fp_offset; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.initial_adjust; +@@ -3313,9 +3318,11 @@ aarch64_get_separate_components (void) + if (aarch64_register_saved_on_entry (regno)) + { + HOST_WIDE_INT offset = cfun->machine->frame.reg_offset[regno]; ++ ++ /* Get the offset relative to the register we'll use. */ + if (!frame_pointer_needed) +- offset += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset += cfun->machine->frame.bytes_below_hard_fp; ++ + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ + if (offset_12bit_unsigned_scaled_p (DImode, offset)) +@@ -3418,8 +3425,8 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + HOST_WIDE_INT offset = cfun->machine->frame.reg_offset[regno]; + if (!frame_pointer_needed) +- offset += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset += cfun->machine->frame.bytes_below_hard_fp; ++ + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); + +@@ -3460,8 +3467,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (!frame_pointer_needed) +- offset2 += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset2 += cfun->machine->frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index ddf833ebfe85..6dc6ef2b989b 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -558,9 +558,13 @@ struct GTY (()) aarch64_frame + HOST_WIDE_INT saved_varargs_size; + + /* The size of the saved callee-save int/FP registers. */ +- + HOST_WIDE_INT saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ HOST_WIDE_INT bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0026-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/7.5.0/0026-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..ba5f2634 --- /dev/null +++ b/packages/gcc/7.5.0/0026-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,104 @@ +From e628f34ef3a7db3a75aa146b3e75099a9c431d20 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:25:40 +0100 +Subject: [PATCH 26/30] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 12 +++++++----- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 10 insertions(+), 8 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 12df8bb783c7..27dd3b7a62b3 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -2890,7 +2890,8 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT); + +- cfun->machine->frame.locals_offset = cfun->machine->frame.saved_varargs_size; ++ cfun->machine->frame.bytes_above_locals ++ = cfun->machine->frame.saved_varargs_size; + + cfun->machine->frame.initial_adjust = 0; + cfun->machine->frame.final_adjust = 0; +@@ -2951,7 +2952,8 @@ aarch64_layout_frame (void) + cfun->machine->frame.bytes_below_hard_fp + = cfun->machine->frame.final_adjust; + cfun->machine->frame.hard_fp_offset = cfun->machine->frame.callee_adjust; +- cfun->machine->frame.locals_offset = cfun->machine->frame.hard_fp_offset; ++ cfun->machine->frame.bytes_above_locals ++ = cfun->machine->frame.hard_fp_offset; + } + else + { +@@ -5653,14 +5655,14 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + + if (from == FRAME_POINTER_REGNUM) + return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ - cfun->machine->frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return cfun->machine->frame.frame_size ++ - cfun->machine->frame.bytes_above_locals; + } + + return cfun->machine->frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 6dc6ef2b989b..79af237ada67 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -565,10 +565,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + HOST_WIDE_INT bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- HOST_WIDE_INT locals_offset; ++ HOST_WIDE_INT bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0027-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/7.5.0/0027-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..6d4dcccd --- /dev/null +++ b/packages/gcc/7.5.0/0027-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,141 @@ +From da0132a4cd2eb26ecaff20578bf2753b20014223 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:28:11 +0100 +Subject: [PATCH 27/30] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.c | 29 ++++++++++++++++------------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 19 insertions(+), 16 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 27dd3b7a62b3..e3c36abb1e35 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -2881,12 +2881,12 @@ aarch64_layout_frame (void) + HOST_WIDE_INT varargs_and_saved_regs_size + = offset + cfun->machine->frame.saved_varargs_size; + +- cfun->machine->frame.hard_fp_offset ++ cfun->machine->frame.bytes_above_hard_fp + = ROUND_UP (varargs_and_saved_regs_size + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); + + cfun->machine->frame.frame_size +- = ROUND_UP (cfun->machine->frame.hard_fp_offset ++ = ROUND_UP (cfun->machine->frame.bytes_above_hard_fp + + cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT); + +@@ -2917,7 +2917,7 @@ aarch64_layout_frame (void) + else if ((cfun->machine->frame.bytes_below_hard_fp + + cfun->machine->frame.saved_regs_size < 512) + && !(cfun->calls_alloca +- && cfun->machine->frame.hard_fp_offset < max_push_offset)) ++ && cfun->machine->frame.bytes_above_hard_fp < max_push_offset)) + { + /* Frame with small area below the saved registers: + sub sp, sp, frame_size +@@ -2925,16 +2925,17 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, bytes_below_hard_fp + 16] */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + cfun->machine->frame.callee_offset +- = cfun->machine->frame.frame_size - cfun->machine->frame.hard_fp_offset; ++ = cfun->machine->frame.frame_size - cfun->machine->frame.bytes_above_hard_fp; + } +- else if (cfun->machine->frame.hard_fp_offset < max_push_offset) ++ else if (cfun->machine->frame.bytes_above_hard_fp < max_push_offset) + { + /* Frame with large area below the saved registers, but with a + small area above: +- stp reg1, reg2, [sp, -hard_fp_offset]! ++ stp reg1, reg2, [sp, -bytes_above_hard_fp]! + stp reg3, reg4, [sp, 16] + sub sp, sp, bytes_below_hard_fp */ +- cfun->machine->frame.callee_adjust = cfun->machine->frame.hard_fp_offset; ++ cfun->machine->frame.callee_adjust ++ = cfun->machine->frame.bytes_above_hard_fp; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.callee_adjust; + } +@@ -2951,19 +2952,21 @@ aarch64_layout_frame (void) + = cfun->machine->frame.frame_size - cfun->machine->frame.callee_adjust; + cfun->machine->frame.bytes_below_hard_fp + = cfun->machine->frame.final_adjust; +- cfun->machine->frame.hard_fp_offset = cfun->machine->frame.callee_adjust; ++ cfun->machine->frame.bytes_above_hard_fp ++ = cfun->machine->frame.callee_adjust; + cfun->machine->frame.bytes_above_locals +- = cfun->machine->frame.hard_fp_offset; ++ = cfun->machine->frame.bytes_above_hard_fp; + } + else + { + /* General case: +- sub sp, sp, hard_fp_offset ++ sub sp, sp, bytes_above_hard_fp + stp x29, x30, [sp, 0] + add x29, sp, 0 + stp reg3, reg4, [sp, 16] + sub sp, sp, bytes_below_hard_fp */ +- cfun->machine->frame.initial_adjust = cfun->machine->frame.hard_fp_offset; ++ cfun->machine->frame.initial_adjust ++ = cfun->machine->frame.bytes_above_hard_fp; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.initial_adjust; + } +@@ -5651,10 +5654,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return cfun->machine->frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset ++ return cfun->machine->frame.bytes_above_hard_fp + - cfun->machine->frame.bytes_above_locals; + } + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 79af237ada67..35ba9681e03d 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -570,10 +570,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + HOST_WIDE_INT bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- HOST_WIDE_INT hard_fp_offset; ++ HOST_WIDE_INT bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + * frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0028-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/7.5.0/0028-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..14d418fd --- /dev/null +++ b/packages/gcc/7.5.0/0028-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,37 @@ +From 19776fcf06856ceff521f99a1ceddaf113cf0d09 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 22 Jun 2023 22:26:30 +0100 +Subject: [PATCH 28/30] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 6 +++--- + 1 file changed, 3 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 35ba9681e03d..19b7082187a5 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -575,9 +575,9 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + HOST_WIDE_INT bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- * frame (incomming SP) to the stack_pointer. This value is always +- * a multiple of STACK_BOUNDARY. */ ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always ++ a multiple of STACK_BOUNDARY. */ + HOST_WIDE_INT frame_size; + + /* The size of the initial stack adjustment before saving callee-saves. */ +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0029-Backport-check-function-bodies-support.patch b/packages/gcc/7.5.0/0029-Backport-check-function-bodies-support.patch new file mode 100644 index 00000000..82d48cc2 --- /dev/null +++ b/packages/gcc/7.5.0/0029-Backport-check-function-bodies-support.patch @@ -0,0 +1,211 @@ +From 69c18301a2dea6dd686d4351ba3c095490707d16 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 15 Aug 2023 19:05:30 +0100 +Subject: [PATCH 29/30] Backport check-function-bodies support + +--- + gcc/testsuite/lib/scanasm.exp | 191 ++++++++++++++++++++++++++++++++++ + 1 file changed, 191 insertions(+) + +diff --git a/gcc/testsuite/lib/scanasm.exp b/gcc/testsuite/lib/scanasm.exp +index bab23e8e1656..1103f86c6023 100644 +--- a/gcc/testsuite/lib/scanasm.exp ++++ b/gcc/testsuite/lib/scanasm.exp +@@ -514,3 +514,194 @@ proc scan-lto-assembler { args } { + verbose "output_file: $output_file" + dg-scan "scan-assembler" 1 $testcase $output_file $args + } ++ ++# Read assembly file FILENAME and store a mapping from function names ++# to function bodies in array RESULT. FILENAME has already been uploaded ++# locally where necessary and is known to exist. ++ ++proc parse_function_bodies { filename result } { ++ upvar $result up_result ++ ++ # Regexp for the start of a function definition (name in \1). ++ set label {^([a-zA-Z_]\S+):$} ++ ++ # Regexp for the end of a function definition. ++ set terminator {^\s*\.size} ++ ++ # Regexp for lines that aren't interesting. ++ set fluff {^\s*(?:\.|//|@|$)} ++ ++ set fd [open $filename r] ++ set in_function 0 ++ while { [gets $fd line] >= 0 } { ++ if { [regexp $label $line dummy function_name] } { ++ set in_function 1 ++ set function_body "" ++ } elseif { $in_function } { ++ if { [regexp $terminator $line] } { ++ set up_result($function_name) $function_body ++ set in_function 0 ++ } elseif { ![regexp $fluff $line] } { ++ append function_body $line "\n" ++ } ++ } ++ } ++ close $fd ++} ++ ++# FUNCTIONS is an array that maps function names to function bodies. ++# Return true if it contains a definition of function NAME and if ++# that definition matches BODY_REGEXP. ++ ++proc check_function_body { functions name body_regexp } { ++ upvar $functions up_functions ++ ++ if { ![info exists up_functions($name)] } { ++ return 0 ++ } ++ set fn_res [regexp "^$body_regexp\$" $up_functions($name)] ++ if { !$fn_res } { ++ verbose -log "body: $body_regexp" ++ verbose -log "against: $up_functions($name)" ++ } ++ return $fn_res ++} ++ ++# Check the implementations of functions against expected output. Used as: ++# ++# { dg-do { check-function-bodies PREFIX TERMINATOR[ OPTION[ SELECTOR]] } } ++# ++# See sourcebuild.texi for details. ++ ++proc check-function-bodies { args } { ++ if { [llength $args] < 2 } { ++ error "too few arguments to check-function-bodies" ++ } ++ if { [llength $args] > 4 } { ++ error "too many arguments to check-function-bodies" ++ } ++ ++ if { [llength $args] >= 3 } { ++ set required_flags [lindex $args 2] ++ ++ upvar 2 dg-extra-tool-flags extra_tool_flags ++ set flags $extra_tool_flags ++ ++ global torture_current_flags ++ if { [info exists torture_current_flags] } { ++ append flags " " $torture_current_flags ++ } ++ foreach required_flag $required_flags { ++ switch -- $required_flag { ++ target - ++ xfail { ++ error "misplaced $required_flag in check-function-bodies" ++ } ++ } ++ } ++ foreach required_flag $required_flags { ++ if { ![regexp " $required_flag " $flags] } { ++ return ++ } ++ } ++ } ++ ++ set xfail_all 0 ++ if { [llength $args] >= 4 } { ++ switch [dg-process-target [lindex $args 3]] { ++ "S" { } ++ "N" { return } ++ "F" { set xfail_all 1 } ++ "P" { } ++ } ++ } ++ ++ set testcase [testname-for-summary] ++ # The name might include a list of options; extract the file name. ++ set filename [lindex $testcase 0] ++ ++ global srcdir ++ set input_filename "$srcdir/$filename" ++ set output_filename "[file rootname [file tail $filename]].s" ++ ++ set prefix [lindex $args 0] ++ set prefix_len [string length $prefix] ++ set terminator [lindex $args 1] ++ if { [string equal $terminator ""] } { ++ set terminator "*/" ++ } ++ set terminator_len [string length $terminator] ++ ++ set have_bodies 0 ++ if { [is_remote host] } { ++ remote_upload host "$filename" ++ } ++ if { [file exists $output_filename] } { ++ parse_function_bodies $output_filename functions ++ set have_bodies 1 ++ } else { ++ verbose -log "$testcase: output file does not exist" ++ } ++ ++ set count 0 ++ set function_regexp "" ++ set label {^(\S+):$} ++ ++ set lineno 1 ++ set fd [open $input_filename r] ++ set in_function 0 ++ while { [gets $fd line] >= 0 } { ++ if { [string equal -length $prefix_len $line $prefix] } { ++ set line [string trim [string range $line $prefix_len end]] ++ if { !$in_function } { ++ if { [regexp "^(.*?\\S)\\s+{(.*)}\$" $line dummy \ ++ line selector] } { ++ set selector [dg-process-target $selector] ++ } else { ++ set selector "P" ++ } ++ if { ![regexp $label $line dummy function_name] } { ++ close $fd ++ error "check-function-bodies: line $lineno does not have a function label" ++ } ++ set in_function 1 ++ set function_regexp "" ++ } elseif { [string equal $line "("] } { ++ append function_regexp "(?:" ++ } elseif { [string equal $line "|"] } { ++ append function_regexp "|" ++ } elseif { [string equal $line ")"] } { ++ append function_regexp ")" ++ } elseif { [string equal $line "..."] } { ++ append function_regexp ".*" ++ } else { ++ append function_regexp "\t" $line "\n" ++ } ++ } elseif { [string equal -length $terminator_len $line $terminator] } { ++ if { ![string equal $selector "N"] } { ++ if { $xfail_all || [string equal $selector "F"] } { ++ setup_xfail "*-*-*" ++ } ++ set testname "$testcase check-function-bodies $function_name" ++ if { !$have_bodies } { ++ unresolved $testname ++ } elseif { [check_function_body functions $function_name \ ++ $function_regexp] } { ++ pass $testname ++ } else { ++ fail $testname ++ } ++ } ++ set in_function 0 ++ incr count ++ } ++ incr lineno ++ } ++ close $fd ++ if { $in_function } { ++ error "check-function-bodies: missing \"$terminator\"" ++ } ++ if { $count == 0 } { ++ error "check-function-bodies: no matches found" ++ } ++} +-- +2.42.0 + diff --git a/packages/gcc/7.5.0/0030-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/7.5.0/0030-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..039125ab --- /dev/null +++ b/packages/gcc/7.5.0/0030-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,282 @@ +From c2c772172f0478315ab94386ebf9963efd36eb75 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 15 Jun 2023 19:16:52 +0100 +Subject: [PATCH 30/30] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 45 +++++++++- + .../gcc.target/aarch64/stack-protector-8.c | 86 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 +++++++ + 3 files changed, 161 insertions(+), 3 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index e3c36abb1e35..8708b4fe4ea2 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -2780,6 +2780,20 @@ aarch64_frame_pointer_required (void) + return false; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -2828,6 +2842,16 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); ++ ++ if (regs_at_top_p) ++ { ++ cfun->machine->frame.bytes_below_hard_fp += get_frame_size (); ++ cfun->machine->frame.bytes_below_hard_fp ++ = ROUND_UP (cfun->machine->frame.bytes_below_hard_fp, ++ STACK_BOUNDARY / BITS_PER_UNIT); ++ } ++ + if (frame_pointer_needed) + { + /* FP and LR are placed in the linkage record. */ +@@ -2881,8 +2905,11 @@ aarch64_layout_frame (void) + HOST_WIDE_INT varargs_and_saved_regs_size + = offset + cfun->machine->frame.saved_varargs_size; + ++ cfun->machine->frame.bytes_above_hard_fp = varargs_and_saved_regs_size; ++ if (!regs_at_top_p) ++ cfun->machine->frame.bytes_above_hard_fp += get_frame_size (); + cfun->machine->frame.bytes_above_hard_fp +- = ROUND_UP (varargs_and_saved_regs_size + get_frame_size (), ++ = ROUND_UP (cfun->machine->frame.bytes_above_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT); + + cfun->machine->frame.frame_size +@@ -2892,6 +2919,9 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_above_locals + = cfun->machine->frame.saved_varargs_size; ++ if (regs_at_top_p) ++ cfun->machine->frame.bytes_above_locals ++ += cfun->machine->frame.saved_regs_size; + + cfun->machine->frame.initial_adjust = 0; + cfun->machine->frame.final_adjust = 0; +@@ -3537,10 +3567,10 @@ aarch64_set_handled_components (sbitmap components) + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding0 | \ ++ | padding (1) | \ + +-------------------------------+ | + | callee-saved registers | | frame.saved_regs_size + +-------------------------------+ | +@@ -3548,6 +3578,10 @@ aarch64_set_handled_components (sbitmap components) + +-------------------------------+ | + | FP' | / <- hard_frame_pointer_rtx (aligned) + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -3557,6 +3591,9 @@ aarch64_set_handled_components (sbitmap components) + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. */ +@@ -3626,6 +3663,8 @@ aarch64_expand_prologue (void) + aarch64_save_callee_saves (DFmode, callee_offset, V0_REGNUM, V31_REGNUM, + callee_adjust != 0 || frame_pointer_needed); + aarch64_sub_sp (IP1_REGNUM, final_adjust, !frame_pointer_needed); ++ if (frame_pointer_needed && final_adjust != 0) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..2575abbec739 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,86 @@ ++/* { dg-options " -O -fstack-protector-strong" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++ ++/* ++** test1: ++** sub sp, sp, #304 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** str (x[0-9]+), \[sp, #?288\] ++** ... ++** ldr (x[0-9]+), \[\1\] ++** str \2, \[(?:sp, #?264|x29, #?-8)\] ++** mov \2, *0 ++** ( ++** add x0, sp, #?8 ++** | ++** sub x0, x29, #?264 ++** ) ++** bl g ++** ... ++** ldr x[0-9]+, \[\1\] ++** ... ++** ( ++** bne .* ++** | ++** cbnz .* ++** ) ++** ... ++** ldr \1, \[sp, #?288\] ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?304 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-32\]! ++** ( ++** mov x29, sp ++** | ++** add x29, sp, #?0 ++** ) ++** str (x[0-9]+), \[sp, #?16\] ++** sub sp, sp, #1040 ++** ... ++** ldr (x[0-9]+), \[\1\] ++** str \2, \[(?:sp, #?1032|x29, #?-8)\] ++** mov \2, *0 ++** ( ++** add x0, sp, #?8 ++** | ++** sub x0, x29, #?1032 ++** ) ++** bl g ++** ... ++** ldr x[0-9]+, \[\1\] ++** ... ++** ( ++** bne .* ++** | ++** cbnz .* ++** ) ++** ... ++** ( ++** add sp, sp, #?1040 ++** | ++** add sp, x29, #?0 ++** ) ++** ldr \1, \[sp, #?16\] ++** ldp x29, x30, \[sp\], #?32 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..1d2673438061 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=cortex-a73 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0026-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/8.5.0/0026-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..7e32ce8e --- /dev/null +++ b/packages/gcc/8.5.0/0026-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,43 @@ +From f9e1c86b3397faa8dee4bbaeaf8b56492e3a63df Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 17:00:51 +0100 +Subject: [PATCH 26/32] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.c | 8 +++++--- + 1 file changed, 5 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 4ad73220337f..ed9e2a28f914 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4215,9 +4215,11 @@ aarch64_layout_frame (void) + max_push_offset = 256; + + HOST_WIDE_INT const_size, const_fp_offset; +- if (cfun->machine->frame.frame_size.is_constant (&const_size) +- && const_size < max_push_offset +- && known_eq (crtl->outgoing_args_size, 0)) ++ if (cfun->machine->frame.saved_regs_size == 0) ++ cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; ++ else if (cfun->machine->frame.frame_size.is_constant (&const_size) ++ && const_size < max_push_offset ++ && known_eq (crtl->outgoing_args_size, 0)) + { + /* Simple, small frame with no outgoing arguments: + stp reg1, reg2, [sp, -frame_size]! +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0027-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/8.5.0/0027-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..f9ee88d6 --- /dev/null +++ b/packages/gcc/8.5.0/0027-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,165 @@ +From b4780c8d6cbd3fc8f2997dd7dcd360b3039c6690 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 16:55:12 +0100 +Subject: [PATCH 27/32] aarch64: Add bytes_below_hard_fp to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_hard_fp instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 40 ++++++++++++++++++++---------------- + gcc/config/aarch64/aarch64.h | 6 +++++- + 2 files changed, 27 insertions(+), 19 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index ed9e2a28f914..996c898888cd 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4136,6 +4136,8 @@ aarch64_layout_frame (void) + last_fp_reg = regno; + } + ++ cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; ++ + if (cfun->machine->frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -4195,11 +4197,11 @@ aarch64_layout_frame (void) + STACK_BOUNDARY / BITS_PER_UNIT); + + /* Both these values are already aligned. */ +- gcc_assert (multiple_p (crtl->outgoing_args_size, ++ gcc_assert (multiple_p (cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT)); + cfun->machine->frame.frame_size + = (cfun->machine->frame.hard_fp_offset +- + crtl->outgoing_args_size); ++ + cfun->machine->frame.bytes_below_hard_fp); + + cfun->machine->frame.locals_offset = cfun->machine->frame.saved_varargs_size; + +@@ -4219,23 +4221,23 @@ aarch64_layout_frame (void) + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + else if (cfun->machine->frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +- && known_eq (crtl->outgoing_args_size, 0)) ++ && known_eq (cfun->machine->frame.bytes_below_hard_fp, 0)) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + cfun->machine->frame.callee_adjust = const_size; + } +- else if (known_lt (crtl->outgoing_args_size ++ else if (known_lt (cfun->machine->frame.bytes_below_hard_fp + + cfun->machine->frame.saved_regs_size, 512) + && !(cfun->calls_alloca + && known_lt (cfun->machine->frame.hard_fp_offset, + max_push_offset))) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_hard_fp] ++ stp reg3, reg4, [sp, bytes_below_hard_fp + 16] */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + cfun->machine->frame.callee_offset + = cfun->machine->frame.frame_size - cfun->machine->frame.hard_fp_offset; +@@ -4243,22 +4245,23 @@ aarch64_layout_frame (void) + else if (cfun->machine->frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments but a small local area: ++ /* Frame with large area below the saved registers, but with a ++ small area above: + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.callee_adjust = const_fp_offset; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.callee_adjust; + } + else + { +- /* Frame with large local area and outgoing arguments using frame pointer: ++ /* General case: + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] + add x29, sp, 0 + stp reg3, reg4, [sp, 16] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.hard_fp_offset; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.initial_adjust; +@@ -4666,9 +4669,11 @@ aarch64_get_separate_components (void) + if (aarch64_register_saved_on_entry (regno)) + { + poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ ++ /* Get the offset relative to the register we'll use. */ + if (!frame_pointer_needed) +- offset += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset += cfun->machine->frame.bytes_below_hard_fp; ++ + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ + if (offset_12bit_unsigned_scaled_p (DImode, offset)) +@@ -4786,8 +4791,8 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + poly_int64 offset = cfun->machine->frame.reg_offset[regno]; + if (!frame_pointer_needed) +- offset += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset += cfun->machine->frame.bytes_below_hard_fp; ++ + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); + +@@ -4828,8 +4833,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (!frame_pointer_needed) +- offset2 += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset2 += cfun->machine->frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index f4d27fa64580..cac7a57ed475 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -647,9 +647,13 @@ struct GTY (()) aarch64_frame + HOST_WIDE_INT saved_varargs_size; + + /* The size of the saved callee-save int/FP registers. */ +- + HOST_WIDE_INT saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0028-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/8.5.0/0028-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..c96071e2 --- /dev/null +++ b/packages/gcc/8.5.0/0028-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,94 @@ +From 746f8815cf7efc16c6973420e74b2560b82a644c Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:25:40 +0100 +Subject: [PATCH 28/32] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 9 +++++---- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 8 insertions(+), 7 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 996c898888cd..32a9857cb111 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4203,7 +4203,8 @@ aarch64_layout_frame (void) + = (cfun->machine->frame.hard_fp_offset + + cfun->machine->frame.bytes_below_hard_fp); + +- cfun->machine->frame.locals_offset = cfun->machine->frame.saved_varargs_size; ++ cfun->machine->frame.bytes_above_locals ++ = cfun->machine->frame.saved_varargs_size; + + cfun->machine->frame.initial_adjust = 0; + cfun->machine->frame.final_adjust = 0; +@@ -7628,14 +7629,14 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + + if (from == FRAME_POINTER_REGNUM) + return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ - cfun->machine->frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return cfun->machine->frame.frame_size ++ - cfun->machine->frame.bytes_above_locals; + } + + return cfun->machine->frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index cac7a57ed475..eb69e38bd537 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -654,10 +654,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 locals_offset; ++ poly_int64 bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0029-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/8.5.0/0029-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..385b7594 --- /dev/null +++ b/packages/gcc/8.5.0/0029-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,130 @@ +From 8711ad4ae2c4c9fa693bab65a6a2fe0b191f7678 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:28:11 +0100 +Subject: [PATCH 29/32] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.c | 21 +++++++++++---------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 14 insertions(+), 13 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 32a9857cb111..861e928ebdc2 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4191,7 +4191,7 @@ aarch64_layout_frame (void) + HOST_WIDE_INT varargs_and_saved_regs_size + = offset + cfun->machine->frame.saved_varargs_size; + +- cfun->machine->frame.hard_fp_offset ++ cfun->machine->frame.bytes_above_hard_fp + = aligned_upper_bound (varargs_and_saved_regs_size + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); +@@ -4200,7 +4200,7 @@ aarch64_layout_frame (void) + gcc_assert (multiple_p (cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT)); + cfun->machine->frame.frame_size +- = (cfun->machine->frame.hard_fp_offset ++ = (cfun->machine->frame.bytes_above_hard_fp + + cfun->machine->frame.bytes_below_hard_fp); + + cfun->machine->frame.bytes_above_locals +@@ -4232,7 +4232,7 @@ aarch64_layout_frame (void) + else if (known_lt (cfun->machine->frame.bytes_below_hard_fp + + cfun->machine->frame.saved_regs_size, 512) + && !(cfun->calls_alloca +- && known_lt (cfun->machine->frame.hard_fp_offset, ++ && known_lt (cfun->machine->frame.bytes_above_hard_fp, + max_push_offset))) + { + /* Frame with small area below the saved registers: +@@ -4241,14 +4241,14 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, bytes_below_hard_fp + 16] */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + cfun->machine->frame.callee_offset +- = cfun->machine->frame.frame_size - cfun->machine->frame.hard_fp_offset; ++ = cfun->machine->frame.frame_size - cfun->machine->frame.bytes_above_hard_fp; + } +- else if (cfun->machine->frame.hard_fp_offset.is_constant (&const_fp_offset) ++ else if (cfun->machine->frame.bytes_above_hard_fp.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { + /* Frame with large area below the saved registers, but with a + small area above: +- stp reg1, reg2, [sp, -hard_fp_offset]! ++ stp reg1, reg2, [sp, -bytes_above_hard_fp]! + stp reg3, reg4, [sp, 16] + sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.callee_adjust = const_fp_offset; +@@ -4258,12 +4258,13 @@ aarch64_layout_frame (void) + else + { + /* General case: +- sub sp, sp, hard_fp_offset ++ sub sp, sp, bytes_above_hard_fp + stp x29, x30, [sp, 0] + add x29, sp, 0 + stp reg3, reg4, [sp, 16] + sub sp, sp, bytes_below_hard_fp */ +- cfun->machine->frame.initial_adjust = cfun->machine->frame.hard_fp_offset; ++ cfun->machine->frame.initial_adjust ++ = cfun->machine->frame.bytes_above_hard_fp; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.initial_adjust; + } +@@ -7625,10 +7626,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return cfun->machine->frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset ++ return cfun->machine->frame.bytes_above_hard_fp + - cfun->machine->frame.bytes_above_locals; + } + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index eb69e38bd537..55ef6ae08a54 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -659,10 +659,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 hard_fp_offset; ++ poly_int64 bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0030-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/8.5.0/0030-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..129a7179 --- /dev/null +++ b/packages/gcc/8.5.0/0030-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,35 @@ +From d1cfa6a4f7c148000b9cd591479accf001fc739b Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 22 Jun 2023 22:26:30 +0100 +Subject: [PATCH 30/32] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 55ef6ae08a54..fbcfb9891b73 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -664,8 +664,8 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- frame (incomming SP) to the stack_pointer. This value is always ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always + a multiple of STACK_BOUNDARY. */ + poly_int64 frame_size; + +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0031-Backport-check-function-bodies-support.patch b/packages/gcc/8.5.0/0031-Backport-check-function-bodies-support.patch new file mode 100644 index 00000000..ba96f85e --- /dev/null +++ b/packages/gcc/8.5.0/0031-Backport-check-function-bodies-support.patch @@ -0,0 +1,211 @@ +From 30455227307a5b078d8d857f337d931191c2d9f0 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 15 Aug 2023 19:05:30 +0100 +Subject: [PATCH 31/32] Backport check-function-bodies support + +--- + gcc/testsuite/lib/scanasm.exp | 191 ++++++++++++++++++++++++++++++++++ + 1 file changed, 191 insertions(+) + +diff --git a/gcc/testsuite/lib/scanasm.exp b/gcc/testsuite/lib/scanasm.exp +index e653b1e7597e..191d38686cd3 100644 +--- a/gcc/testsuite/lib/scanasm.exp ++++ b/gcc/testsuite/lib/scanasm.exp +@@ -542,3 +542,194 @@ proc scan-lto-assembler { args } { + verbose "output_file: $output_file" + dg-scan "scan-assembler" 1 $testcase $output_file $args + } ++ ++# Read assembly file FILENAME and store a mapping from function names ++# to function bodies in array RESULT. FILENAME has already been uploaded ++# locally where necessary and is known to exist. ++ ++proc parse_function_bodies { filename result } { ++ upvar $result up_result ++ ++ # Regexp for the start of a function definition (name in \1). ++ set label {^([a-zA-Z_]\S+):$} ++ ++ # Regexp for the end of a function definition. ++ set terminator {^\s*\.size} ++ ++ # Regexp for lines that aren't interesting. ++ set fluff {^\s*(?:\.|//|@|$)} ++ ++ set fd [open $filename r] ++ set in_function 0 ++ while { [gets $fd line] >= 0 } { ++ if { [regexp $label $line dummy function_name] } { ++ set in_function 1 ++ set function_body "" ++ } elseif { $in_function } { ++ if { [regexp $terminator $line] } { ++ set up_result($function_name) $function_body ++ set in_function 0 ++ } elseif { ![regexp $fluff $line] } { ++ append function_body $line "\n" ++ } ++ } ++ } ++ close $fd ++} ++ ++# FUNCTIONS is an array that maps function names to function bodies. ++# Return true if it contains a definition of function NAME and if ++# that definition matches BODY_REGEXP. ++ ++proc check_function_body { functions name body_regexp } { ++ upvar $functions up_functions ++ ++ if { ![info exists up_functions($name)] } { ++ return 0 ++ } ++ set fn_res [regexp "^$body_regexp\$" $up_functions($name)] ++ if { !$fn_res } { ++ verbose -log "body: $body_regexp" ++ verbose -log "against: $up_functions($name)" ++ } ++ return $fn_res ++} ++ ++# Check the implementations of functions against expected output. Used as: ++# ++# { dg-do { check-function-bodies PREFIX TERMINATOR[ OPTION[ SELECTOR]] } } ++# ++# See sourcebuild.texi for details. ++ ++proc check-function-bodies { args } { ++ if { [llength $args] < 2 } { ++ error "too few arguments to check-function-bodies" ++ } ++ if { [llength $args] > 4 } { ++ error "too many arguments to check-function-bodies" ++ } ++ ++ if { [llength $args] >= 3 } { ++ set required_flags [lindex $args 2] ++ ++ upvar 2 dg-extra-tool-flags extra_tool_flags ++ set flags $extra_tool_flags ++ ++ global torture_current_flags ++ if { [info exists torture_current_flags] } { ++ append flags " " $torture_current_flags ++ } ++ foreach required_flag $required_flags { ++ switch -- $required_flag { ++ target - ++ xfail { ++ error "misplaced $required_flag in check-function-bodies" ++ } ++ } ++ } ++ foreach required_flag $required_flags { ++ if { ![regexp " $required_flag " $flags] } { ++ return ++ } ++ } ++ } ++ ++ set xfail_all 0 ++ if { [llength $args] >= 4 } { ++ switch [dg-process-target [lindex $args 3]] { ++ "S" { } ++ "N" { return } ++ "F" { set xfail_all 1 } ++ "P" { } ++ } ++ } ++ ++ set testcase [testname-for-summary] ++ # The name might include a list of options; extract the file name. ++ set filename [lindex $testcase 0] ++ ++ global srcdir ++ set input_filename "$srcdir/$filename" ++ set output_filename "[file rootname [file tail $filename]].s" ++ ++ set prefix [lindex $args 0] ++ set prefix_len [string length $prefix] ++ set terminator [lindex $args 1] ++ if { [string equal $terminator ""] } { ++ set terminator "*/" ++ } ++ set terminator_len [string length $terminator] ++ ++ set have_bodies 0 ++ if { [is_remote host] } { ++ remote_upload host "$filename" ++ } ++ if { [file exists $output_filename] } { ++ parse_function_bodies $output_filename functions ++ set have_bodies 1 ++ } else { ++ verbose -log "$testcase: output file does not exist" ++ } ++ ++ set count 0 ++ set function_regexp "" ++ set label {^(\S+):$} ++ ++ set lineno 1 ++ set fd [open $input_filename r] ++ set in_function 0 ++ while { [gets $fd line] >= 0 } { ++ if { [string equal -length $prefix_len $line $prefix] } { ++ set line [string trim [string range $line $prefix_len end]] ++ if { !$in_function } { ++ if { [regexp "^(.*?\\S)\\s+{(.*)}\$" $line dummy \ ++ line selector] } { ++ set selector [dg-process-target $selector] ++ } else { ++ set selector "P" ++ } ++ if { ![regexp $label $line dummy function_name] } { ++ close $fd ++ error "check-function-bodies: line $lineno does not have a function label" ++ } ++ set in_function 1 ++ set function_regexp "" ++ } elseif { [string equal $line "("] } { ++ append function_regexp "(?:" ++ } elseif { [string equal $line "|"] } { ++ append function_regexp "|" ++ } elseif { [string equal $line ")"] } { ++ append function_regexp ")" ++ } elseif { [string equal $line "..."] } { ++ append function_regexp ".*" ++ } else { ++ append function_regexp "\t" $line "\n" ++ } ++ } elseif { [string equal -length $terminator_len $line $terminator] } { ++ if { ![string equal $selector "N"] } { ++ if { $xfail_all || [string equal $selector "F"] } { ++ setup_xfail "*-*-*" ++ } ++ set testname "$testcase check-function-bodies $function_name" ++ if { !$have_bodies } { ++ unresolved $testname ++ } elseif { [check_function_body functions $function_name \ ++ $function_regexp] } { ++ pass $testname ++ } else { ++ fail $testname ++ } ++ } ++ set in_function 0 ++ incr count ++ } ++ incr lineno ++ } ++ close $fd ++ if { $in_function } { ++ error "check-function-bodies: missing \"$terminator\"" ++ } ++ if { $count == 0 } { ++ error "check-function-bodies: no matches found" ++ } ++} +-- +2.42.0 + diff --git a/packages/gcc/8.5.0/0032-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/8.5.0/0032-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..b074c42d --- /dev/null +++ b/packages/gcc/8.5.0/0032-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,259 @@ +From ee9111b567ddcd2f1b16da4f38b69b1b979df9ef Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 15 Jun 2023 19:16:52 +0100 +Subject: [PATCH 32/32] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 46 ++++++++++++-- + .../gcc.target/aarch64/stack-protector-8.c | 62 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 ++++++++++ + 3 files changed, 137 insertions(+), 4 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 861e928ebdc2..5c0e3fd484da 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4078,6 +4078,20 @@ aarch64_output_probe_stack_range (rtx reg1, rtx reg2) + return ""; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -4138,6 +4152,16 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); ++ ++ if (regs_at_top_p) ++ { ++ cfun->machine->frame.bytes_below_hard_fp += get_frame_size (); ++ cfun->machine->frame.bytes_below_hard_fp ++ = aligned_upper_bound (cfun->machine->frame.bytes_below_hard_fp, ++ STACK_BOUNDARY / BITS_PER_UNIT); ++ } ++ + if (cfun->machine->frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -4191,9 +4215,11 @@ aarch64_layout_frame (void) + HOST_WIDE_INT varargs_and_saved_regs_size + = offset + cfun->machine->frame.saved_varargs_size; + ++ cfun->machine->frame.bytes_above_hard_fp = varargs_and_saved_regs_size; ++ if (!regs_at_top_p) ++ cfun->machine->frame.bytes_above_hard_fp += get_frame_size (); + cfun->machine->frame.bytes_above_hard_fp +- = aligned_upper_bound (varargs_and_saved_regs_size +- + get_frame_size (), ++ = aligned_upper_bound (cfun->machine->frame.bytes_above_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT); + + /* Both these values are already aligned. */ +@@ -4205,6 +4231,9 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_above_locals + = cfun->machine->frame.saved_varargs_size; ++ if (regs_at_top_p) ++ cfun->machine->frame.bytes_above_locals ++ += cfun->machine->frame.saved_regs_size; + + cfun->machine->frame.initial_adjust = 0; + cfun->machine->frame.final_adjust = 0; +@@ -4912,10 +4941,10 @@ aarch64_add_cfa_expression (rtx_insn *insn, unsigned int reg, + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding0 | \ ++ | padding (1) | \ + +-------------------------------+ | + | callee-saved registers | | frame.saved_regs_size + +-------------------------------+ | +@@ -4923,6 +4952,10 @@ aarch64_add_cfa_expression (rtx_insn *insn, unsigned int reg, + +-------------------------------+ | + | FP' | / <- hard_frame_pointer_rtx (aligned) + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -4932,6 +4965,9 @@ aarch64_add_cfa_expression (rtx_insn *insn, unsigned int reg, + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. */ +@@ -5042,6 +5078,8 @@ aarch64_expand_prologue (void) + aarch64_save_callee_saves (DFmode, callee_offset, V0_REGNUM, V31_REGNUM, + callee_adjust != 0 || emit_frame_chain); + aarch64_sub_sp (ip1_rtx, ip0_rtx, final_adjust, !frame_pointer_needed); ++ if (emit_frame_chain && maybe_ne (final_adjust, 0)) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..b808735762fd +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,62 @@ ++/* { dg-options " -O -fstack-protector-strong" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++ ++/* ++** test1: ++** sub sp, sp, #304 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** str (x[0-9]+), \[sp, #?288\] ++** ... ++** ldr (x[0-9]+), \[\1\] ++** str \2, \[sp, #?264\] ++** mov \2, *0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** ldr x[0-9]+, \[\1\] ++** ... ++** bne .* ++** ... ++** ldr \1, \[sp, #?288\] ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?304 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-32\]! ++** mov x29, sp ++** str (x[0-9]+), \[sp, #?16\] ++** sub sp, sp, #1040 ++** ... ++** ldr (x[0-9]+), \[\1\] ++** str \2, \[sp, #?1032\] ++** mov \2, *0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** ldr x[0-9]+, \[\1\] ++** ... ++** bne .* ++** ... ++** add sp, sp, #?1040 ++** ldr \1, \[sp, #?16\] ++** ldp x29, x30, \[sp\], #?32 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..58f322aa480a +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=neoverse-v1 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0021-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch b/packages/gcc/9.5.0/0021-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch new file mode 100644 index 00000000..c1180f52 --- /dev/null +++ b/packages/gcc/9.5.0/0021-aarch64-Explicitly-handle-frames-with-no-saved-regis.patch @@ -0,0 +1,49 @@ +From 78ebdb7b12d5e258b9811bab715734454268fd0c Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 17:00:51 +0100 +Subject: [PATCH 21/30] aarch64: Explicitly handle frames with no saved + registers + +If a frame has no saved registers, it can be allocated in one go. +There is no need to treat the areas below and above the saved +registers as separate. + +And if we allocate the frame in one go, it should be allocated +as the initial_adjust rather than the final_adjust. This allows the +frame size to grow to guard_size - guard_used_by_caller before a stack +probe is needed. (A frame with no register saves is necessarily a +leaf frame.) + +This is a no-op as thing stand, since a leaf function will have +no outgoing arguments, and so all the frame will be above where +the saved registers normally go. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Explicitly + allocate the frame in one go if there are no saved registers. +--- + gcc/config/aarch64/aarch64.c | 8 +++++--- + 1 file changed, 5 insertions(+), 3 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index a35dceab9fc7..e9dad682738f 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4771,9 +4771,11 @@ aarch64_layout_frame (void) + max_push_offset = 256; + + HOST_WIDE_INT const_size, const_fp_offset; +- if (cfun->machine->frame.frame_size.is_constant (&const_size) +- && const_size < max_push_offset +- && known_eq (crtl->outgoing_args_size, 0)) ++ if (cfun->machine->frame.saved_regs_size == 0) ++ cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; ++ else if (cfun->machine->frame.frame_size.is_constant (&const_size) ++ && const_size < max_push_offset ++ && known_eq (crtl->outgoing_args_size, 0)) + { + /* Simple, small frame with no outgoing arguments: + stp reg1, reg2, [sp, -frame_size]! +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0022-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch b/packages/gcc/9.5.0/0022-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch new file mode 100644 index 00000000..039c352f --- /dev/null +++ b/packages/gcc/9.5.0/0022-aarch64-Add-bytes_below_hard_fp-to-frame-info.patch @@ -0,0 +1,189 @@ +From 347487fffa0266d43bf18f1f91878410881f596e Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Fri, 16 Jun 2023 16:55:12 +0100 +Subject: [PATCH 22/30] aarch64: Add bytes_below_hard_fp to frame info + +The frame layout code currently hard-codes the assumption that +the number of bytes below the saved registers is equal to the +size of the outgoing arguments. This patch abstracts that +value into a new field of aarch64_frame. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::bytes_below_hard_fp): New + field. + * config/aarch64/aarch64.c (aarch64_layout_frame): Initialize it, + and use it instead of crtl->outgoing_args_size. + (aarch64_get_separate_components): Use bytes_below_hard_fp instead + of outgoing_args_size. + (aarch64_process_components): Likewise. +--- + gcc/config/aarch64/aarch64.c | 50 +++++++++++++++++++----------------- + gcc/config/aarch64/aarch64.h | 6 ++++- + 2 files changed, 32 insertions(+), 24 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index e9dad682738f..25cf10cc4b94 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4684,6 +4684,8 @@ aarch64_layout_frame (void) + last_fp_reg = regno; + } + ++ cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; ++ + if (cfun->machine->frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +@@ -4751,11 +4753,11 @@ aarch64_layout_frame (void) + STACK_BOUNDARY / BITS_PER_UNIT); + + /* Both these values are already aligned. */ +- gcc_assert (multiple_p (crtl->outgoing_args_size, ++ gcc_assert (multiple_p (cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT)); + cfun->machine->frame.frame_size + = (cfun->machine->frame.hard_fp_offset +- + crtl->outgoing_args_size); ++ + cfun->machine->frame.bytes_below_hard_fp); + + cfun->machine->frame.locals_offset = cfun->machine->frame.saved_varargs_size; + +@@ -4775,23 +4777,23 @@ aarch64_layout_frame (void) + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + else if (cfun->machine->frame.frame_size.is_constant (&const_size) + && const_size < max_push_offset +- && known_eq (crtl->outgoing_args_size, 0)) ++ && known_eq (cfun->machine->frame.bytes_below_hard_fp, 0)) + { +- /* Simple, small frame with no outgoing arguments: ++ /* Simple, small frame with no data below the saved registers. + stp reg1, reg2, [sp, -frame_size]! + stp reg3, reg4, [sp, 16] */ + cfun->machine->frame.callee_adjust = const_size; + } +- else if (known_lt (crtl->outgoing_args_size ++ else if (known_lt (cfun->machine->frame.bytes_below_hard_fp + + cfun->machine->frame.saved_regs_size, 512) + && !(cfun->calls_alloca + && known_lt (cfun->machine->frame.hard_fp_offset, + max_push_offset))) + { +- /* Frame with small outgoing arguments: ++ /* Frame with small area below the saved registers: + sub sp, sp, frame_size +- stp reg1, reg2, [sp, outgoing_args_size] +- stp reg3, reg4, [sp, outgoing_args_size + 16] */ ++ stp reg1, reg2, [sp, bytes_below_hard_fp] ++ stp reg3, reg4, [sp, bytes_below_hard_fp + 16] */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + cfun->machine->frame.callee_offset + = cfun->machine->frame.frame_size - cfun->machine->frame.hard_fp_offset; +@@ -4799,22 +4801,23 @@ aarch64_layout_frame (void) + else if (cfun->machine->frame.hard_fp_offset.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { +- /* Frame with large outgoing arguments but a small local area: ++ /* Frame with large area below the saved registers, but with a ++ small area above: + stp reg1, reg2, [sp, -hard_fp_offset]! + stp reg3, reg4, [sp, 16] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.callee_adjust = const_fp_offset; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.callee_adjust; + } + else + { +- /* Frame with large local area and outgoing arguments using frame pointer: ++ /* General case: + sub sp, sp, hard_fp_offset + stp x29, x30, [sp, 0] + add x29, sp, 0 + stp reg3, reg4, [sp, 16] +- sub sp, sp, outgoing_args_size */ ++ sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.hard_fp_offset; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.initial_adjust; +@@ -5243,9 +5246,11 @@ aarch64_get_separate_components (void) + if (aarch64_register_saved_on_entry (regno)) + { + poly_int64 offset = cfun->machine->frame.reg_offset[regno]; ++ ++ /* Get the offset relative to the register we'll use. */ + if (!frame_pointer_needed) +- offset += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset += cfun->machine->frame.bytes_below_hard_fp; ++ + /* Check that we can access the stack slot of the register with one + direct load with no adjustments needed. */ + if (offset_12bit_unsigned_scaled_p (DImode, offset)) +@@ -5367,8 +5372,8 @@ aarch64_process_components (sbitmap components, bool prologue_p) + rtx reg = gen_rtx_REG (mode, regno); + poly_int64 offset = cfun->machine->frame.reg_offset[regno]; + if (!frame_pointer_needed) +- offset += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset += cfun->machine->frame.bytes_below_hard_fp; ++ + rtx addr = plus_constant (Pmode, ptr_reg, offset); + rtx mem = gen_frame_mem (mode, addr); + +@@ -5410,8 +5415,7 @@ aarch64_process_components (sbitmap components, bool prologue_p) + /* REGNO2 can be saved/restored in a pair with REGNO. */ + rtx reg2 = gen_rtx_REG (mode, regno2); + if (!frame_pointer_needed) +- offset2 += cfun->machine->frame.frame_size +- - cfun->machine->frame.hard_fp_offset; ++ offset2 += cfun->machine->frame.bytes_below_hard_fp; + rtx addr2 = plus_constant (Pmode, ptr_reg, offset2); + rtx mem2 = gen_frame_mem (mode, addr2); + rtx set2 = prologue_p ? gen_rtx_SET (mem2, reg2) +@@ -5478,10 +5482,10 @@ aarch64_stack_clash_protection_alloca_probe_range (void) + registers. If POLY_SIZE is not large enough to require a probe this function + will only adjust the stack. When allocating the stack space + FRAME_RELATED_P is then used to indicate if the allocation is frame related. +- FINAL_ADJUSTMENT_P indicates whether we are allocating the outgoing +- arguments. If we are then we ensure that any allocation larger than the ABI +- defined buffer needs a probe so that the invariant of having a 1KB buffer is +- maintained. ++ FINAL_ADJUSTMENT_P indicates whether we are allocating the area below ++ the saved registers. If we are then we ensure that any allocation ++ larger than the ABI defined buffer needs a probe so that the ++ invariant of having a 1KB buffer is maintained. + + We emit barriers after each stack adjustment to prevent optimizations from + breaking the invariant that we never drop the stack more than a page. This +@@ -5671,7 +5675,7 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + /* Handle any residuals. Residuals of at least MIN_PROBE_THRESHOLD have to + be probed. This maintains the requirement that each page is probed at + least once. For initial probing we probe only if the allocation is +- more than GUARD_SIZE - buffer, and for the outgoing arguments we probe ++ more than GUARD_SIZE - buffer, and below the saved registers we probe + if the amount is larger than buffer. GUARD_SIZE - buffer + buffer == + GUARD_SIZE. This works that for any allocation that is large enough to + trigger a probe here, we'll have at least one, and if they're not large +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index af0bc3f18817..95831637ba73 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -712,9 +712,13 @@ struct GTY (()) aarch64_frame + HOST_WIDE_INT saved_varargs_size; + + /* The size of the saved callee-save int/FP registers. */ +- + HOST_WIDE_INT saved_regs_size; + ++ /* The number of bytes between the bottom of the static frame (the bottom ++ of the outgoing arguments) and the hard frame pointer. This value is ++ always a multiple of STACK_BOUNDARY. */ ++ poly_int64 bytes_below_hard_fp; ++ + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0023-aarch64-Rename-locals_offset-to-bytes_above_locals.patch b/packages/gcc/9.5.0/0023-aarch64-Rename-locals_offset-to-bytes_above_locals.patch new file mode 100644 index 00000000..2a727bb3 --- /dev/null +++ b/packages/gcc/9.5.0/0023-aarch64-Rename-locals_offset-to-bytes_above_locals.patch @@ -0,0 +1,94 @@ +From 4604c4cd0a6c4c26d6594ec9a0383b4d9197d9df Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:25:40 +0100 +Subject: [PATCH 23/30] aarch64: Rename locals_offset to bytes_above_locals +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +locals_offset was described as: + + /* Offset from the base of the frame (incomming SP) to the + top of the locals area. This value is always a multiple of + STACK_BOUNDARY. */ + +This is implicitly an “upside down” view of the frame: the incoming +SP is at offset 0, and anything N bytes below the incoming SP is at +offset N (rather than -N). + +However, reg_offset instead uses a “right way up” view; that is, +it views offsets in address terms. Something above X is at a +positive offset from X and something below X is at a negative +offset from X. + +Also, even on FRAME_GROWS_DOWNWARD targets like AArch64, +target-independent code views offsets in address terms too: +locals are allocated at negative offsets to virtual_stack_vars. + +It seems confusing to have *_offset fields of the same structure +using different polarities like this. This patch tries to avoid +that by renaming locals_offset to bytes_above_locals. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::locals_offset): Rename to... + (aarch64_frame::bytes_above_locals): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_initial_elimination_offset): Update accordingly. +--- + gcc/config/aarch64/aarch64.c | 9 +++++---- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 8 insertions(+), 7 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 25cf10cc4b94..dcaf491af428 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4759,7 +4759,8 @@ aarch64_layout_frame (void) + = (cfun->machine->frame.hard_fp_offset + + cfun->machine->frame.bytes_below_hard_fp); + +- cfun->machine->frame.locals_offset = cfun->machine->frame.saved_varargs_size; ++ cfun->machine->frame.bytes_above_locals ++ = cfun->machine->frame.saved_varargs_size; + + cfun->machine->frame.initial_adjust = 0; + cfun->machine->frame.final_adjust = 0; +@@ -8566,14 +8567,14 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + + if (from == FRAME_POINTER_REGNUM) + return cfun->machine->frame.hard_fp_offset +- - cfun->machine->frame.locals_offset; ++ - cfun->machine->frame.bytes_above_locals; + } + + if (to == STACK_POINTER_REGNUM) + { + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.frame_size +- - cfun->machine->frame.locals_offset; ++ return cfun->machine->frame.frame_size ++ - cfun->machine->frame.bytes_above_locals; + } + + return cfun->machine->frame.frame_size; +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index 95831637ba73..a079a88b4f46 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -719,10 +719,10 @@ struct GTY (()) aarch64_frame + always a multiple of STACK_BOUNDARY. */ + poly_int64 bytes_below_hard_fp; + +- /* Offset from the base of the frame (incomming SP) to the +- top of the locals area. This value is always a multiple of ++ /* The number of bytes between the top of the locals area and the top ++ of the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 locals_offset; ++ poly_int64 bytes_above_locals; + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0024-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch b/packages/gcc/9.5.0/0024-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch new file mode 100644 index 00000000..6bd05091 --- /dev/null +++ b/packages/gcc/9.5.0/0024-aarch64-Rename-hard_fp_offset-to-bytes_above_hard_fp.patch @@ -0,0 +1,130 @@ +From 16016465ff28a75f5e0540cbaeb4eb102fdc3230 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 11:28:11 +0100 +Subject: [PATCH 24/30] aarch64: Rename hard_fp_offset to bytes_above_hard_fp +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +Similarly to the previous locals_offset patch, hard_fp_offset +was described as: + + /* Offset from the base of the frame (incomming SP) to the + hard_frame_pointer. This value is always a multiple of + STACK_BOUNDARY. */ + poly_int64 hard_fp_offset; + +which again took an “upside-down” view: higher offsets meant lower +addresses. This patch renames the field to bytes_above_hard_fp instead. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::hard_fp_offset): Rename + to... + (aarch64_frame::bytes_above_hard_fp): ...this. + * config/aarch64/aarch64.c (aarch64_layout_frame) + (aarch64_expand_prologue): Update accordingly. + (aarch64_initial_elimination_offset): Likewise. +--- + gcc/config/aarch64/aarch64.c | 21 +++++++++++---------- + gcc/config/aarch64/aarch64.h | 6 +++--- + 2 files changed, 14 insertions(+), 13 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index dcaf491af428..2681e0c2bb90 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4747,7 +4747,7 @@ aarch64_layout_frame (void) + HOST_WIDE_INT varargs_and_saved_regs_size + = offset + cfun->machine->frame.saved_varargs_size; + +- cfun->machine->frame.hard_fp_offset ++ cfun->machine->frame.bytes_above_hard_fp + = aligned_upper_bound (varargs_and_saved_regs_size + + get_frame_size (), + STACK_BOUNDARY / BITS_PER_UNIT); +@@ -4756,7 +4756,7 @@ aarch64_layout_frame (void) + gcc_assert (multiple_p (cfun->machine->frame.bytes_below_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT)); + cfun->machine->frame.frame_size +- = (cfun->machine->frame.hard_fp_offset ++ = (cfun->machine->frame.bytes_above_hard_fp + + cfun->machine->frame.bytes_below_hard_fp); + + cfun->machine->frame.bytes_above_locals +@@ -4788,7 +4788,7 @@ aarch64_layout_frame (void) + else if (known_lt (cfun->machine->frame.bytes_below_hard_fp + + cfun->machine->frame.saved_regs_size, 512) + && !(cfun->calls_alloca +- && known_lt (cfun->machine->frame.hard_fp_offset, ++ && known_lt (cfun->machine->frame.bytes_above_hard_fp, + max_push_offset))) + { + /* Frame with small area below the saved registers: +@@ -4797,14 +4797,14 @@ aarch64_layout_frame (void) + stp reg3, reg4, [sp, bytes_below_hard_fp + 16] */ + cfun->machine->frame.initial_adjust = cfun->machine->frame.frame_size; + cfun->machine->frame.callee_offset +- = cfun->machine->frame.frame_size - cfun->machine->frame.hard_fp_offset; ++ = cfun->machine->frame.frame_size - cfun->machine->frame.bytes_above_hard_fp; + } +- else if (cfun->machine->frame.hard_fp_offset.is_constant (&const_fp_offset) ++ else if (cfun->machine->frame.bytes_above_hard_fp.is_constant (&const_fp_offset) + && const_fp_offset < max_push_offset) + { + /* Frame with large area below the saved registers, but with a + small area above: +- stp reg1, reg2, [sp, -hard_fp_offset]! ++ stp reg1, reg2, [sp, -bytes_above_hard_fp]! + stp reg3, reg4, [sp, 16] + sub sp, sp, bytes_below_hard_fp */ + cfun->machine->frame.callee_adjust = const_fp_offset; +@@ -4814,12 +4814,13 @@ aarch64_layout_frame (void) + else + { + /* General case: +- sub sp, sp, hard_fp_offset ++ sub sp, sp, bytes_above_hard_fp + stp x29, x30, [sp, 0] + add x29, sp, 0 + stp reg3, reg4, [sp, 16] + sub sp, sp, bytes_below_hard_fp */ +- cfun->machine->frame.initial_adjust = cfun->machine->frame.hard_fp_offset; ++ cfun->machine->frame.initial_adjust ++ = cfun->machine->frame.bytes_above_hard_fp; + cfun->machine->frame.final_adjust + = cfun->machine->frame.frame_size - cfun->machine->frame.initial_adjust; + } +@@ -8563,10 +8564,10 @@ aarch64_initial_elimination_offset (unsigned from, unsigned to) + if (to == HARD_FRAME_POINTER_REGNUM) + { + if (from == ARG_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset; ++ return cfun->machine->frame.bytes_above_hard_fp; + + if (from == FRAME_POINTER_REGNUM) +- return cfun->machine->frame.hard_fp_offset ++ return cfun->machine->frame.bytes_above_hard_fp + - cfun->machine->frame.bytes_above_locals; + } + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index a079a88b4f46..eab6da84a02e 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -724,10 +724,10 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_locals; + +- /* Offset from the base of the frame (incomming SP) to the +- hard_frame_pointer. This value is always a multiple of ++ /* The number of bytes between the hard_frame_pointer and the top of ++ the frame (the incomming SP). This value is always a multiple of + STACK_BOUNDARY. */ +- poly_int64 hard_fp_offset; ++ poly_int64 bytes_above_hard_fp; + + /* The size of the frame. This value is the offset from base of the + frame (incomming SP) to the stack_pointer. This value is always +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0025-aarch64-Tweak-frame_size-comment.patch b/packages/gcc/9.5.0/0025-aarch64-Tweak-frame_size-comment.patch new file mode 100644 index 00000000..43596741 --- /dev/null +++ b/packages/gcc/9.5.0/0025-aarch64-Tweak-frame_size-comment.patch @@ -0,0 +1,35 @@ +From eb2271eb6bb68ec3c9aa9ae4746ea1ee5f18874a Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 22 Jun 2023 22:26:30 +0100 +Subject: [PATCH 25/30] aarch64: Tweak frame_size comment +MIME-Version: 1.0 +Content-Type: text/plain; charset=UTF-8 +Content-Transfer-Encoding: 8bit + +This patch fixes another case in which a value was described with +an “upside-down” view. + +gcc/ + * config/aarch64/aarch64.h (aarch64_frame::frame_size): Tweak comment. +--- + gcc/config/aarch64/aarch64.h | 4 ++-- + 1 file changed, 2 insertions(+), 2 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.h b/gcc/config/aarch64/aarch64.h +index eab6da84a02e..7c4b65ec55be 100644 +--- a/gcc/config/aarch64/aarch64.h ++++ b/gcc/config/aarch64/aarch64.h +@@ -729,8 +729,8 @@ struct GTY (()) aarch64_frame + STACK_BOUNDARY. */ + poly_int64 bytes_above_hard_fp; + +- /* The size of the frame. This value is the offset from base of the +- frame (incomming SP) to the stack_pointer. This value is always ++ /* The size of the frame, i.e. the number of bytes between the bottom ++ of the outgoing arguments and the incoming SP. This value is always + a multiple of STACK_BOUNDARY. */ + poly_int64 frame_size; + +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0026-Backport-check-function-bodies-support.patch b/packages/gcc/9.5.0/0026-Backport-check-function-bodies-support.patch new file mode 100644 index 00000000..b0148d68 --- /dev/null +++ b/packages/gcc/9.5.0/0026-Backport-check-function-bodies-support.patch @@ -0,0 +1,211 @@ +From cfed3b87e9351edff1568ade4ef666edc9887639 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 15 Aug 2023 19:05:30 +0100 +Subject: [PATCH 26/30] Backport check-function-bodies support + +--- + gcc/testsuite/lib/scanasm.exp | 191 ++++++++++++++++++++++++++++++++++ + 1 file changed, 191 insertions(+) + +diff --git a/gcc/testsuite/lib/scanasm.exp b/gcc/testsuite/lib/scanasm.exp +index 35ccbc86fc03..c9af27bf47aa 100644 +--- a/gcc/testsuite/lib/scanasm.exp ++++ b/gcc/testsuite/lib/scanasm.exp +@@ -546,3 +546,194 @@ proc scan-lto-assembler { args } { + verbose "output_file: $output_file" + dg-scan "scan-lto-assembler" 1 $testcase $output_file $args + } ++ ++# Read assembly file FILENAME and store a mapping from function names ++# to function bodies in array RESULT. FILENAME has already been uploaded ++# locally where necessary and is known to exist. ++ ++proc parse_function_bodies { filename result } { ++ upvar $result up_result ++ ++ # Regexp for the start of a function definition (name in \1). ++ set label {^([a-zA-Z_]\S+):$} ++ ++ # Regexp for the end of a function definition. ++ set terminator {^\s*\.size} ++ ++ # Regexp for lines that aren't interesting. ++ set fluff {^\s*(?:\.|//|@|$)} ++ ++ set fd [open $filename r] ++ set in_function 0 ++ while { [gets $fd line] >= 0 } { ++ if { [regexp $label $line dummy function_name] } { ++ set in_function 1 ++ set function_body "" ++ } elseif { $in_function } { ++ if { [regexp $terminator $line] } { ++ set up_result($function_name) $function_body ++ set in_function 0 ++ } elseif { ![regexp $fluff $line] } { ++ append function_body $line "\n" ++ } ++ } ++ } ++ close $fd ++} ++ ++# FUNCTIONS is an array that maps function names to function bodies. ++# Return true if it contains a definition of function NAME and if ++# that definition matches BODY_REGEXP. ++ ++proc check_function_body { functions name body_regexp } { ++ upvar $functions up_functions ++ ++ if { ![info exists up_functions($name)] } { ++ return 0 ++ } ++ set fn_res [regexp "^$body_regexp\$" $up_functions($name)] ++ if { !$fn_res } { ++ verbose -log "body: $body_regexp" ++ verbose -log "against: $up_functions($name)" ++ } ++ return $fn_res ++} ++ ++# Check the implementations of functions against expected output. Used as: ++# ++# { dg-do { check-function-bodies PREFIX TERMINATOR[ OPTION[ SELECTOR]] } } ++# ++# See sourcebuild.texi for details. ++ ++proc check-function-bodies { args } { ++ if { [llength $args] < 2 } { ++ error "too few arguments to check-function-bodies" ++ } ++ if { [llength $args] > 4 } { ++ error "too many arguments to check-function-bodies" ++ } ++ ++ if { [llength $args] >= 3 } { ++ set required_flags [lindex $args 2] ++ ++ upvar 2 dg-extra-tool-flags extra_tool_flags ++ set flags $extra_tool_flags ++ ++ global torture_current_flags ++ if { [info exists torture_current_flags] } { ++ append flags " " $torture_current_flags ++ } ++ foreach required_flag $required_flags { ++ switch -- $required_flag { ++ target - ++ xfail { ++ error "misplaced $required_flag in check-function-bodies" ++ } ++ } ++ } ++ foreach required_flag $required_flags { ++ if { ![regexp " $required_flag " $flags] } { ++ return ++ } ++ } ++ } ++ ++ set xfail_all 0 ++ if { [llength $args] >= 4 } { ++ switch [dg-process-target [lindex $args 3]] { ++ "S" { } ++ "N" { return } ++ "F" { set xfail_all 1 } ++ "P" { } ++ } ++ } ++ ++ set testcase [testname-for-summary] ++ # The name might include a list of options; extract the file name. ++ set filename [lindex $testcase 0] ++ ++ global srcdir ++ set input_filename "$srcdir/$filename" ++ set output_filename "[file rootname [file tail $filename]].s" ++ ++ set prefix [lindex $args 0] ++ set prefix_len [string length $prefix] ++ set terminator [lindex $args 1] ++ if { [string equal $terminator ""] } { ++ set terminator "*/" ++ } ++ set terminator_len [string length $terminator] ++ ++ set have_bodies 0 ++ if { [is_remote host] } { ++ remote_upload host "$filename" ++ } ++ if { [file exists $output_filename] } { ++ parse_function_bodies $output_filename functions ++ set have_bodies 1 ++ } else { ++ verbose -log "$testcase: output file does not exist" ++ } ++ ++ set count 0 ++ set function_regexp "" ++ set label {^(\S+):$} ++ ++ set lineno 1 ++ set fd [open $input_filename r] ++ set in_function 0 ++ while { [gets $fd line] >= 0 } { ++ if { [string equal -length $prefix_len $line $prefix] } { ++ set line [string trim [string range $line $prefix_len end]] ++ if { !$in_function } { ++ if { [regexp "^(.*?\\S)\\s+{(.*)}\$" $line dummy \ ++ line selector] } { ++ set selector [dg-process-target $selector] ++ } else { ++ set selector "P" ++ } ++ if { ![regexp $label $line dummy function_name] } { ++ close $fd ++ error "check-function-bodies: line $lineno does not have a function label" ++ } ++ set in_function 1 ++ set function_regexp "" ++ } elseif { [string equal $line "("] } { ++ append function_regexp "(?:" ++ } elseif { [string equal $line "|"] } { ++ append function_regexp "|" ++ } elseif { [string equal $line ")"] } { ++ append function_regexp ")" ++ } elseif { [string equal $line "..."] } { ++ append function_regexp ".*" ++ } else { ++ append function_regexp "\t" $line "\n" ++ } ++ } elseif { [string equal -length $terminator_len $line $terminator] } { ++ if { ![string equal $selector "N"] } { ++ if { $xfail_all || [string equal $selector "F"] } { ++ setup_xfail "*-*-*" ++ } ++ set testname "$testcase check-function-bodies $function_name" ++ if { !$have_bodies } { ++ unresolved $testname ++ } elseif { [check_function_body functions $function_name \ ++ $function_regexp] } { ++ pass $testname ++ } else { ++ fail $testname ++ } ++ } ++ set in_function 0 ++ incr count ++ } ++ incr lineno ++ } ++ close $fd ++ if { $in_function } { ++ error "check-function-bodies: missing \"$terminator\"" ++ } ++ if { $count == 0 } { ++ error "check-function-bodies: no matches found" ++ } ++} +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0027-aarch64-Tweak-stack-clash-boundary-condition.patch b/packages/gcc/9.5.0/0027-aarch64-Tweak-stack-clash-boundary-condition.patch new file mode 100644 index 00000000..06ce00e6 --- /dev/null +++ b/packages/gcc/9.5.0/0027-aarch64-Tweak-stack-clash-boundary-condition.patch @@ -0,0 +1,132 @@ +From 4dd8925d95d3d6d89779b494b5f4cfadcf9fa96e Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 15:11:44 +0100 +Subject: [PATCH 27/30] aarch64: Tweak stack clash boundary condition + +The AArch64 ABI says that, when stack clash protection is used, +there can be a maximum of 1KiB of unprobed space at sp on entry +to a function. Therefore, we need to probe when allocating +>= guard_size - 1KiB of data (>= rather than >). This is what +GCC does. + +If an allocation is exactly guard_size bytes, it is enough to allocate +those bytes and probe once at offset 1024. It isn't possible to use a +single probe at any other offset: higher would conmplicate later code, +by leaving more unprobed space than usual, while lower would risk +leaving an entire page unprobed. For simplicity, the code probes all +allocations at offset 1024. + +Some register saves also act as probes. If we need to allocate +more space below the last such register save probe, we need to +probe the allocation if it is > 1KiB. Again, this allocation is +then sometimes (but not always) probed at offset 1024. This sort of +allocation is currently only used for outgoing arguments, which are +rarely this big. + +However, the code also probed if this final outgoing-arguments +allocation was == 1KiB, rather than just > 1KiB. This isn't +necessary, since the register save then probes at offset 1024 +as required. Continuing to probe allocations of exactly 1KiB +would complicate later patches. + +gcc/ + * config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): + Don't probe final allocations that are exactly 1KiB in size (after + unprobed space above the final allocation has been deducted). + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: New test. +--- + gcc/config/aarch64/aarch64.c | 6 +- + .../aarch64/stack-check-prologue-17.c | 55 +++++++++++++++++++ + 2 files changed, 60 insertions(+), 1 deletion(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 2681e0c2bb90..4c9e11cd7cff 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -5506,6 +5506,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT guard_size + = 1 << PARAM_VALUE (PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE); + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; ++ HOST_WIDE_INT byte_sp_alignment = STACK_BOUNDARY / BITS_PER_UNIT; ++ gcc_assert (multiple_p (poly_size, byte_sp_alignment)); + /* When doing the final adjustment for the outgoing argument size we can't + assume that LR was saved at position 0. So subtract it's offset from the + ABI safe buffer so that we don't accidentally allow an adjustment that +@@ -5513,7 +5515,9 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + probing. */ + HOST_WIDE_INT min_probe_threshold + = final_adjustment_p +- ? guard_used_by_caller - cfun->machine->frame.reg_offset[LR_REGNUM] ++ ? (guard_used_by_caller ++ + byte_sp_alignment ++ - cfun->machine->frame.reg_offset[LR_REGNUM]) + : guard_size - guard_used_by_caller; + + poly_int64 frame_size = cfun->machine->frame.frame_size; +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +new file mode 100644 +index 000000000000..0d8a25d73a24 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -0,0 +1,55 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0028-aarch64-Put-LR-save-probe-in-first-16-bytes.patch b/packages/gcc/9.5.0/0028-aarch64-Put-LR-save-probe-in-first-16-bytes.patch new file mode 100644 index 00000000..ae60cb55 --- /dev/null +++ b/packages/gcc/9.5.0/0028-aarch64-Put-LR-save-probe-in-first-16-bytes.patch @@ -0,0 +1,254 @@ +From 12517baf6c88447e3bda3a459ac4c29d61f84e6c Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 27 Jun 2023 15:12:55 +0100 +Subject: [PATCH 28/30] aarch64: Put LR save probe in first 16 bytes + +-fstack-clash-protection uses the save of LR as a probe for the next +allocation. The next allocation could be: + +* another part of the static frame, e.g. when allocating SVE save slots + or outgoing arguments + +* an alloca in the same function + +* an allocation made by a callee function + +However, when -fomit-frame-pointer is used, the LR save slot is placed +above the other GPR save slots. It could therefore be up to 80 bytes +above the base of the GPR save area (which is also the hard fp address). + +aarch64_allocate_and_probe_stack_space took this into account when +deciding how much subsequent space could be allocated without needing +a probe. However, it interacted badly with: + + /* If doing a small final adjustment, we always probe at offset 0. + This is done to avoid issues when LR is not at position 0 or when + the final adjustment is smaller than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +which forces any allocation that is smaller than the guard page size +to be probed at offset 0 rather than the usual offset 1024. It was +therefore possible to construct cases in which we had: + +* a probe using LR at SP + 80 bytes (or some other value >= 16) +* an allocation of the guard page size - 16 bytes +* a probe at SP + 0 + +which allocates guard page size + 64 consecutive unprobed bytes. + +This patch requires the LR probe to be in the first 16 bytes of the +save area when stack clash protection is active. Doing it +unconditionally would cause code-quality regressions. + +gcc/ + * config/aarch64/aarch64.c (aarch64_layout_frame): Ensure that + the LR save slot is in the first 16 bytes of the register save area. + (aarch64_allocate_and_probe_stack_space): Remove workaround for + when LR was not in the first 16 bytes. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-18.c: New test. +--- + gcc/config/aarch64/aarch64.c | 50 +++++---- + .../aarch64/stack-check-prologue-18.c | 100 ++++++++++++++++++ + 2 files changed, 127 insertions(+), 23 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 4c9e11cd7cff..1e8467fdd03f 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4686,15 +4686,31 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; + ++#define ALLOCATE_GPR_SLOT(REGNO) \ ++ do \ ++ { \ ++ cfun->machine->frame.reg_offset[REGNO] = offset; \ ++ if (cfun->machine->frame.wb_candidate1 == INVALID_REGNUM) \ ++ cfun->machine->frame.wb_candidate1 = (REGNO); \ ++ else if (cfun->machine->frame.wb_candidate2 == INVALID_REGNUM) \ ++ cfun->machine->frame.wb_candidate2 = (REGNO); \ ++ offset += UNITS_PER_WORD; \ ++ } \ ++ while (0) ++ + if (cfun->machine->frame.emit_frame_chain) + { + /* FP and LR are placed in the linkage record. */ +- cfun->machine->frame.reg_offset[R29_REGNUM] = 0; +- cfun->machine->frame.wb_candidate1 = R29_REGNUM; +- cfun->machine->frame.reg_offset[R30_REGNUM] = UNITS_PER_WORD; +- cfun->machine->frame.wb_candidate2 = R30_REGNUM; +- offset = 2 * UNITS_PER_WORD; ++ ALLOCATE_GPR_SLOT (R29_REGNUM); ++ ALLOCATE_GPR_SLOT (R30_REGNUM); + } ++ else if (flag_stack_clash_protection ++ && cfun->machine->frame.reg_offset[R30_REGNUM] == SLOT_REQUIRED) ++ /* Put the LR save slot first, since it makes a good choice of probe ++ for stack clash purposes. The idea is that the link register usually ++ has to be saved before a call anyway, and so we lose little by ++ stopping it from being individually shrink-wrapped. */ ++ ALLOCATE_GPR_SLOT (R30_REGNUM); + + /* With stack-clash, LR must be saved in non-leaf functions. */ + gcc_assert (crtl->is_leaf +@@ -4704,14 +4720,9 @@ aarch64_layout_frame (void) + /* Now assign stack slots for them. */ + for (regno = R0_REGNUM; regno <= R30_REGNUM; regno++) + if (cfun->machine->frame.reg_offset[regno] == SLOT_REQUIRED) +- { +- cfun->machine->frame.reg_offset[regno] = offset; +- if (cfun->machine->frame.wb_candidate1 == INVALID_REGNUM) +- cfun->machine->frame.wb_candidate1 = regno; +- else if (cfun->machine->frame.wb_candidate2 == INVALID_REGNUM) +- cfun->machine->frame.wb_candidate2 = regno; +- offset += UNITS_PER_WORD; +- } ++ ALLOCATE_GPR_SLOT (regno); ++ ++#undef ALLOCATE_GPR_SLOT + + HOST_WIDE_INT max_int_offset = offset; + offset = ROUND_UP (offset, STACK_BOUNDARY / BITS_PER_UNIT); +@@ -5508,16 +5519,9 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT guard_used_by_caller = STACK_CLASH_CALLER_GUARD; + HOST_WIDE_INT byte_sp_alignment = STACK_BOUNDARY / BITS_PER_UNIT; + gcc_assert (multiple_p (poly_size, byte_sp_alignment)); +- /* When doing the final adjustment for the outgoing argument size we can't +- assume that LR was saved at position 0. So subtract it's offset from the +- ABI safe buffer so that we don't accidentally allow an adjustment that +- would result in an allocation larger than the ABI buffer without +- probing. */ + HOST_WIDE_INT min_probe_threshold + = final_adjustment_p +- ? (guard_used_by_caller +- + byte_sp_alignment +- - cfun->machine->frame.reg_offset[LR_REGNUM]) ++ ? guard_used_by_caller + byte_sp_alignment + : guard_size - guard_used_by_caller; + + poly_int64 frame_size = cfun->machine->frame.frame_size; +@@ -5697,8 +5701,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; + /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when LR is not at position 0 or when +- the final adjustment is smaller than the probing offset. */ ++ This is done to avoid issues when the final adjustment is smaller ++ than the probing offset. */ + else if (final_adjustment_p && rounded_size == 0) + residual_probe_offset = 0; + +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +new file mode 100644 +index 000000000000..82447d20fff5 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -0,0 +1,100 @@ ++/* { dg-options "-O2 -fstack-clash-protection -fomit-frame-pointer --param stack-clash-protection-guard-size=12" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void f(int, ...); ++void g(); ++ ++/* ++** test1: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #4064 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++** str x26, \[sp, #?4128\] ++** ... ++*/ ++int test1(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test2: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1040 ++** str xzr, \[sp\] ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test2(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x); ++ } ++ g(); ++ return 1; ++} ++ ++/* ++** test3: ++** ... ++** str x30, \[sp\] ++** sub sp, sp, #1024 ++** cbnz w0, .* ++** bl g ++** ... ++*/ ++int test3(int z) { ++ __uint128_t x = 0; ++ int y[0x400]; ++ if (z) ++ { ++ asm volatile ("" ::: ++ "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26"); ++ f(0, 0, 0, 0, 0, 0, 0, &y, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, ++ x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x); ++ } ++ g(); ++ return 1; ++} +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0029-aarch64-Simplify-probe-of-final-frame-allocation.patch b/packages/gcc/9.5.0/0029-aarch64-Simplify-probe-of-final-frame-allocation.patch new file mode 100644 index 00000000..321a0022 --- /dev/null +++ b/packages/gcc/9.5.0/0029-aarch64-Simplify-probe-of-final-frame-allocation.patch @@ -0,0 +1,111 @@ +From f2684e63652bb251d22c79e40081c646df1f36b6 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Tue, 8 Aug 2023 01:57:26 +0100 +Subject: [PATCH 29/30] aarch64: Simplify probe of final frame allocation + +Previous patches ensured that the final frame allocation only needs +a probe when the size is strictly greater than 1KiB. It's therefore +safe to use the normal 1024 probe offset in all cases. + +The main motivation for doing this is to simplify the code and +remove the number of special cases. + +gcc/ + * config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): + Always probe the residual allocation at offset 1024, asserting + that that is in range. + +gcc/testsuite/ + * gcc.target/aarch64/stack-check-prologue-17.c: Expect the probe + to be at offset 1024 rather than offset 0. + * gcc.target/aarch64/stack-check-prologue-18.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 12 ++++-------- + .../gcc.target/aarch64/stack-check-prologue-17.c | 2 +- + .../gcc.target/aarch64/stack-check-prologue-18.c | 7 +++++-- + 3 files changed, 10 insertions(+), 11 deletions(-) + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 1e8467fdd03f..705f719a2eaa 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -5695,16 +5695,12 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + are still safe. */ + if (residual) + { +- HOST_WIDE_INT residual_probe_offset = guard_used_by_caller; ++ gcc_assert (guard_used_by_caller + byte_sp_alignment <= size); ++ + /* If we're doing final adjustments, and we've done any full page + allocations then any residual needs to be probed. */ + if (final_adjustment_p && rounded_size != 0) + min_probe_threshold = 0; +- /* If doing a small final adjustment, we always probe at offset 0. +- This is done to avoid issues when the final adjustment is smaller +- than the probing offset. */ +- else if (final_adjustment_p && rounded_size == 0) +- residual_probe_offset = 0; + + aarch64_sub_sp (temp1, temp2, residual, frame_related_p); + if (residual >= min_probe_threshold) +@@ -5715,8 +5711,8 @@ aarch64_allocate_and_probe_stack_space (rtx temp1, rtx temp2, + HOST_WIDE_INT_PRINT_DEC " bytes, probing will be required." + "\n", residual); + +- emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, +- residual_probe_offset)); ++ emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx, ++ guard_used_by_caller)); + emit_insn (gen_blockage ()); + } + } +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +index 0d8a25d73a24..f0ec1389771d 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-17.c +@@ -33,7 +33,7 @@ int test1(int z) { + ** ... + ** str x30, \[sp\] + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +index 82447d20fff5..71d33ba34e9e 100644 +--- a/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c ++++ b/gcc/testsuite/gcc.target/aarch64/stack-check-prologue-18.c +@@ -8,8 +8,9 @@ void g(); + ** test1: + ** ... + ** str x30, \[sp\] ++** ... + ** sub sp, sp, #4064 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -49,8 +50,9 @@ int test1(int z) { + ** test2: + ** ... + ** str x30, \[sp\] ++** ... + ** sub sp, sp, #1040 +-** str xzr, \[sp\] ++** str xzr, \[sp, #?1024\] + ** cbnz w0, .* + ** bl g + ** ... +@@ -77,6 +79,7 @@ int test2(int z) { + ** test3: + ** ... + ** str x30, \[sp\] ++** ... + ** sub sp, sp, #1024 + ** cbnz w0, .* + ** bl g +-- +2.42.0 + diff --git a/packages/gcc/9.5.0/0030-aarch64-Make-stack-smash-canary-protect-saved-regist.patch b/packages/gcc/9.5.0/0030-aarch64-Make-stack-smash-canary-protect-saved-regist.patch new file mode 100644 index 00000000..d2441669 --- /dev/null +++ b/packages/gcc/9.5.0/0030-aarch64-Make-stack-smash-canary-protect-saved-regist.patch @@ -0,0 +1,255 @@ +From bf3eeaa0182a92987570d9c787bd45079eebf528 Mon Sep 17 00:00:00 2001 +From: Richard Sandiford +Date: Thu, 15 Jun 2023 19:16:52 +0100 +Subject: [PATCH 30/30] aarch64: Make stack smash canary protect saved + registers + +AArch64 normally puts the saved registers near the bottom of the frame, +immediately above any dynamic allocations. But this means that a +stack-smash attack on those dynamic allocations could overwrite the +saved registers without needing to reach as far as the stack smash +canary. + +The same thing could also happen for variable-sized arguments that are +passed by value, since those are allocated before a call and popped on +return. + +This patch avoids that by putting the locals (and thus the canary) below +the saved registers when stack smash protection is active. + +The patch fixes CVE-2023-4039. + +gcc/ + * config/aarch64/aarch64.c (aarch64_save_regs_above_locals_p): + New function. + (aarch64_layout_frame): Use it to decide whether locals should + go above or below the saved registers. + (aarch64_expand_prologue): Update stack layout comment. + Emit a stack tie after the final adjustment. + +gcc/testsuite/ + * gcc.target/aarch64/stack-protector-8.c: New test. + * gcc.target/aarch64/stack-protector-9.c: Likewise. +--- + gcc/config/aarch64/aarch64.c | 46 +++++++++++++-- + .../gcc.target/aarch64/stack-protector-8.c | 58 +++++++++++++++++++ + .../gcc.target/aarch64/stack-protector-9.c | 33 +++++++++++ + 3 files changed, 133 insertions(+), 4 deletions(-) + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-8.c + create mode 100644 gcc/testsuite/gcc.target/aarch64/stack-protector-9.c + +diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c +index 705f719a2eaa..3d094214fac2 100644 +--- a/gcc/config/aarch64/aarch64.c ++++ b/gcc/config/aarch64/aarch64.c +@@ -4622,6 +4622,20 @@ aarch64_needs_frame_chain (void) + return aarch64_use_frame_pointer; + } + ++/* Return true if the current function should save registers above ++ the locals area, rather than below it. */ ++ ++static bool ++aarch64_save_regs_above_locals_p () ++{ ++ /* When using stack smash protection, make sure that the canary slot ++ comes between the locals and the saved registers. Otherwise, ++ it would be possible for a carefully sized smash attack to change ++ the saved registers (particularly LR and FP) without reaching the ++ canary. */ ++ return crtl->stack_protect_guard; ++} ++ + /* Mark the registers that need to be saved by the callee and calculate + the size of the callee-saved registers area and frame record (both FP + and LR may be omitted). */ +@@ -4686,6 +4700,16 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_below_hard_fp = crtl->outgoing_args_size; + ++ bool regs_at_top_p = aarch64_save_regs_above_locals_p (); ++ ++ if (regs_at_top_p) ++ { ++ cfun->machine->frame.bytes_below_hard_fp += get_frame_size (); ++ cfun->machine->frame.bytes_below_hard_fp ++ = aligned_upper_bound (cfun->machine->frame.bytes_below_hard_fp, ++ STACK_BOUNDARY / BITS_PER_UNIT); ++ } ++ + #define ALLOCATE_GPR_SLOT(REGNO) \ + do \ + { \ +@@ -4758,9 +4782,11 @@ aarch64_layout_frame (void) + HOST_WIDE_INT varargs_and_saved_regs_size + = offset + cfun->machine->frame.saved_varargs_size; + ++ cfun->machine->frame.bytes_above_hard_fp = varargs_and_saved_regs_size; ++ if (!regs_at_top_p) ++ cfun->machine->frame.bytes_above_hard_fp += get_frame_size (); + cfun->machine->frame.bytes_above_hard_fp +- = aligned_upper_bound (varargs_and_saved_regs_size +- + get_frame_size (), ++ = aligned_upper_bound (cfun->machine->frame.bytes_above_hard_fp, + STACK_BOUNDARY / BITS_PER_UNIT); + + /* Both these values are already aligned. */ +@@ -4772,6 +4798,9 @@ aarch64_layout_frame (void) + + cfun->machine->frame.bytes_above_locals + = cfun->machine->frame.saved_varargs_size; ++ if (regs_at_top_p) ++ cfun->machine->frame.bytes_above_locals ++ += cfun->machine->frame.saved_regs_size; + + cfun->machine->frame.initial_adjust = 0; + cfun->machine->frame.final_adjust = 0; +@@ -5764,10 +5793,10 @@ aarch64_add_cfa_expression (rtx_insn *insn, unsigned int reg, + | for register varargs | + | | + +-------------------------------+ +- | local variables | <-- frame_pointer_rtx ++ | local variables (1) | <-- frame_pointer_rtx + | | + +-------------------------------+ +- | padding | \ ++ | padding (1) | \ + +-------------------------------+ | + | callee-saved registers | | frame.saved_regs_size + +-------------------------------+ | +@@ -5775,6 +5804,10 @@ aarch64_add_cfa_expression (rtx_insn *insn, unsigned int reg, + +-------------------------------+ | + | FP' | / <- hard_frame_pointer_rtx (aligned) + +-------------------------------+ ++ | local variables (2) | ++ +-------------------------------+ ++ | padding (2) | ++ +-------------------------------+ + | dynamic allocation | + +-------------------------------+ + | padding | +@@ -5784,6 +5817,9 @@ aarch64_add_cfa_expression (rtx_insn *insn, unsigned int reg, + +-------------------------------+ + | | <-- stack_pointer_rtx (aligned) + ++ The regions marked (1) and (2) are mutually exclusive. (2) is used ++ when aarch64_save_regs_above_locals_p is true. ++ + Dynamic stack allocations via alloca() decrease stack_pointer_rtx + but leave frame_pointer_rtx and hard_frame_pointer_rtx + unchanged. +@@ -5937,6 +5973,8 @@ aarch64_expand_prologue (void) + that is assumed by the called. */ + aarch64_allocate_and_probe_stack_space (tmp1_rtx, tmp0_rtx, final_adjust, + !frame_pointer_needed, true); ++ if (emit_frame_chain && maybe_ne (final_adjust, 0)) ++ emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); + } + + /* Return TRUE if we can use a simple_return insn. +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +new file mode 100644 +index 000000000000..c5e7deef6c10 +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-8.c +@@ -0,0 +1,58 @@ ++/* { dg-options " -O -fstack-protector-strong -mstack-protector-guard=sysreg -mstack-protector-guard-reg=tpidr2_el0 -mstack-protector-guard-offset=16" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++void g(void *); ++ ++/* ++** test1: ++** sub sp, sp, #288 ++** stp x29, x30, \[sp, #?272\] ++** add x29, sp, #?272 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?264\] ++** mov \2, *0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** ldp x29, x30, \[sp, #?272\] ++** add sp, sp, #?288 ++** ret ++** bl __stack_chk_fail ++*/ ++int test1() { ++ int y[0x40]; ++ g(y); ++ return 1; ++} ++ ++/* ++** test2: ++** stp x29, x30, \[sp, #?-16\]! ++** mov x29, sp ++** sub sp, sp, #1040 ++** mrs (x[0-9]+), tpidr2_el0 ++** ldr (x[0-9]+), \[\1, #?16\] ++** str \2, \[sp, #?1032\] ++** mov \2, *0 ++** add x0, sp, #?8 ++** bl g ++** ... ++** mrs .* ++** ... ++** bne .* ++** ... ++** add sp, sp, #?1040 ++** ldp x29, x30, \[sp\], #?16 ++** ret ++** bl __stack_chk_fail ++*/ ++int test2() { ++ int y[0x100]; ++ g(y); ++ return 1; ++} +diff --git a/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +new file mode 100644 +index 000000000000..58f322aa480a +--- /dev/null ++++ b/gcc/testsuite/gcc.target/aarch64/stack-protector-9.c +@@ -0,0 +1,33 @@ ++/* { dg-options "-O2 -mcpu=neoverse-v1 -fstack-protector-all" } */ ++/* { dg-final { check-function-bodies "**" "" } } */ ++ ++/* ++** main: ++** ... ++** stp x29, x30, \[sp, #?-[0-9]+\]! ++** ... ++** sub sp, sp, #[0-9]+ ++** ... ++** str x[0-9]+, \[x29, #?-8\] ++** ... ++*/ ++int f(const char *); ++void g(void *); ++int main(int argc, char* argv[]) ++{ ++ int a; ++ int b; ++ char c[2+f(argv[1])]; ++ int d[0x100]; ++ char y; ++ ++ y=42; a=4; b=10; ++ c[0] = 'h'; c[1] = '\0'; ++ ++ c[f(argv[2])] = '\0'; ++ ++ __builtin_printf("%d %d\n%s\n", a, b, c); ++ g(d); ++ ++ return 0; ++} +-- +2.42.0 +