diff --git a/repos/base-sel4/patches/intel_vmx_disable_vpid.patch b/repos/base-sel4/patches/intel_vmx_disable_vpid.patch new file mode 100644 index 0000000000..e98051f653 --- /dev/null +++ b/repos/base-sel4/patches/intel_vmx_disable_vpid.patch @@ -0,0 +1,21 @@ +--- src/kernel/sel4/src/arch/x86/object/vcpu.c ++++ src/kernel/sel4/src/arch/x86/object/vcpu.c +@@ -285,13 +288,13 @@ init_vtx_fixed_values(bool_t useTrueMsrs) + cr4_low = x86_rdmsr_low(IA32_VMX_CR4_FIXED1_MSR); + + /* Check for VPID support */ +- if (!(secondary_control_low & BIT(5))) { ++// if (!(secondary_control_low & BIT(5))) { + vmx_feature_vpid = 0; + printf("vt-x: VPIDs are not supported. Expect performance degredation\n"); +- } else { +- vmx_feature_vpid = 1; +- secondary_control_mask |= BIT(5); +- } ++// } else { ++// vmx_feature_vpid = 1; ++// secondary_control_mask |= BIT(5); ++// } + + /* Check for load perf global control */ + if (!(exit_control_low & BIT(12))) { diff --git a/repos/base-sel4/patches/intel_vmx_full_state.patch b/repos/base-sel4/patches/intel_vmx_full_state.patch new file mode 100644 index 0000000000..1d78c44c99 --- /dev/null +++ b/repos/base-sel4/patches/intel_vmx_full_state.patch @@ -0,0 +1,87 @@ +--- src/kernel/sel4/src/arch/x86/object/vcpu.c ++++ src/kernel/sel4/src/arch/x86/object/vcpu.c +@@ -88,6 +88,9 @@ static bool_t vmx_feature_ack_on_exit; + static vcpu_t *x86KSVPIDTable[VPID_LAST + 1]; + static vpid_t x86KSNextVPID = VPID_FIRST; + ++static void ++setMRs_vmexit(uint32_t reason, word_t qualification, tcb_t *tcb); ++ + static inline bool_t + vmxon(paddr_t vmxon_region) + { +@@ -967,10 +987,8 @@ vcpu_update_state_sysvmenter(vcpu_t *vcpu) + void + vcpu_sysvmenter_reply_to_user(tcb_t *tcb) + { +- word_t *buffer; + vcpu_t *vcpu; + +- buffer = lookupIPCBuffer(true, tcb); + vcpu = tcb->tcbArch.tcbVCPU; + + assert(vcpu); +@@ -979,11 +997,9 @@ vcpu_sysvmenter_reply_to_user(tcb_t *tcb) + switchVCPU(vcpu); + } + +- setMR(tcb, buffer, SEL4_VMENTER_CALL_EIP_MR, vmread(VMX_GUEST_RIP)); +- setMR(tcb, buffer, SEL4_VMENTER_CALL_CONTROL_PPC_MR, vmread(VMX_CONTROL_PRIMARY_PROCESSOR_CONTROLS)); ++ setMRs_vmexit(/* unused */ -1, /* unused */ -1, tcb); + +- setMR(tcb, buffer, SEL4_VMENTER_CALL_CONTROL_ENTRY_MR, vmread(VMX_CONTROL_ENTRY_INTERRUPTION_INFO)); +- setRegister(tcb, msgInfoRegister, 0); ++ setRegister(tcb, msgInfoRegister, 0 /* notification that this is no VM exit */); + } + + exception_t +@@ -1113,27 +1141,27 @@ vtx_init(void) + } + + static void +-setMRs_vmexit(uint32_t reason, word_t qualification) ++setMRs_vmexit(uint32_t reason, word_t qualification, tcb_t *tcb) + { + word_t *buffer; + int i; + +- buffer = lookupIPCBuffer(true, NODE_STATE(ksCurThread)); ++ buffer = lookupIPCBuffer(true, tcb); + +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_CALL_EIP_MR, vmread(VMX_GUEST_RIP)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_CALL_CONTROL_PPC_MR, vmread(VMX_CONTROL_PRIMARY_PROCESSOR_CONTROLS)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_CALL_CONTROL_ENTRY_MR, vmread(VMX_CONTROL_ENTRY_INTERRUPTION_INFO)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_REASON_MR, reason); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_QUALIFICATION_MR, qualification); ++ setMR(tcb, buffer, SEL4_VMENTER_CALL_EIP_MR, vmread(VMX_GUEST_RIP)); ++ setMR(tcb, buffer, SEL4_VMENTER_CALL_CONTROL_PPC_MR, vmread(VMX_CONTROL_PRIMARY_PROCESSOR_CONTROLS)); ++ setMR(tcb, buffer, SEL4_VMENTER_CALL_CONTROL_ENTRY_MR, vmread(VMX_CONTROL_ENTRY_INTERRUPTION_INFO)); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_REASON_MR, reason); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_QUALIFICATION_MR, qualification); + +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_INSTRUCTION_LEN_MR, vmread(VMX_DATA_EXIT_INSTRUCTION_LENGTH)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_GUEST_PHYSICAL_MR, vmread(VMX_DATA_GUEST_PHYSICAL)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_RFLAGS_MR, vmread(VMX_GUEST_RFLAGS)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_GUEST_INT_MR, vmread(VMX_GUEST_INTERRUPTABILITY)); +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_CR3_MR, vmread(VMX_GUEST_CR3)); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_INSTRUCTION_LEN_MR, vmread(VMX_DATA_EXIT_INSTRUCTION_LENGTH)); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_GUEST_PHYSICAL_MR, vmread(VMX_DATA_GUEST_PHYSICAL)); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_RFLAGS_MR, vmread(VMX_GUEST_RFLAGS)); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_GUEST_INT_MR, vmread(VMX_GUEST_INTERRUPTABILITY)); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_CR3_MR, vmread(VMX_GUEST_CR3)); + + for (i = 0; i < n_vcpu_gp_register; i++) { +- setMR(NODE_STATE(ksCurThread), buffer, SEL4_VMENTER_FAULT_EAX + i, NODE_STATE(ksCurThread)->tcbArch.tcbVCPU->gp_registers[i]); ++ setMR(tcb, buffer, SEL4_VMENTER_FAULT_EAX + i, tcb->tcbArch.tcbVCPU->gp_registers[i]); + } + } + +@@ -1143,7 +1171,7 @@ handleVmxFault(uint32_t reason, word_t qualification) + /* Indicate that we are returning the from VMEnter with a fault */ + setRegister(NODE_STATE(ksCurThread), msgInfoRegister, SEL4_VMENTER_RESULT_FAULT); + +- setMRs_vmexit(reason, qualification); ++ setMRs_vmexit(reason, qualification, NODE_STATE(ksCurThread)); + + /* Set the thread back to running */ + setThreadState(NODE_STATE(ksCurThread), ThreadState_Running); diff --git a/repos/base-sel4/ports/sel4.hash b/repos/base-sel4/ports/sel4.hash index 94fbbf1f97..70350f0055 100644 --- a/repos/base-sel4/ports/sel4.hash +++ b/repos/base-sel4/ports/sel4.hash @@ -1 +1 @@ -34b8f0e01692d1d2ba2f02c98bafe321fc09de22 +7935487f91a31c0cd8aaf09278f6312af56bb935 diff --git a/repos/base-sel4/src/core/include/vm_space.h b/repos/base-sel4/src/core/include/vm_space.h index 66ea903d1a..c8dc959553 100644 --- a/repos/base-sel4/src/core/include/vm_space.h +++ b/repos/base-sel4/src/core/include/vm_space.h @@ -134,6 +134,8 @@ class Genode::Vm_space */ using Selector_allocator = Bit_allocator<1UL << NUM_VM_SEL_LOG2>; + class Alloc_page_table_failed : Exception { }; + private: Selector_allocator _sel_alloc { }; @@ -247,8 +249,6 @@ class Genode::Vm_space long _invalidate_page(Genode::Cap_sel const &, seL4_Word const, seL4_Word const); - class Alloc_page_table_failed : Exception { }; - /** * Allocate and install page structures for the protection domain. * diff --git a/repos/base-sel4/src/core/spec/x86/vm_session_component.cc b/repos/base-sel4/src/core/spec/x86/vm_session_component.cc index 5d74cad990..b674e827ee 100644 --- a/repos/base-sel4/src/core/spec/x86/vm_session_component.cc +++ b/repos/base-sel4/src/core/spec/x86/vm_session_component.cc @@ -251,22 +251,24 @@ void Vm_session_component::_attach_vm_memory(Dataspace_component &dsc, while (page.valid()) { enum { NO_FLUSH = false, FLUSH = true }; try { - _vm_space.alloc_guest_page_tables(page.hotspot, 1 << page.log2_order); - _vm_space.map_guest(page.addr, page.hotspot, - (1 << page.log2_order) / 4096, - dsc.cacheability(), - dsc.writable() && attribute.writeable, - attribute.executable, NO_FLUSH); - } catch (Page_table_registry::Mapping_cache_full full) { - if (full.reason == Page_table_registry::Mapping_cache_full::MEMORY) { - if (_ram_quota_guard().limit().value > 4 * 1024 * 1024) - /* we get in trouble in core if we use too much memory */ - throw Vm_space::Selector_allocator::Out_of_indices(); - throw Out_of_ram(); + try { + _vm_space.alloc_guest_page_tables(page.hotspot, 1 << page.log2_order); + _vm_space.map_guest(page.addr, page.hotspot, + (1 << page.log2_order) / 4096, + dsc.cacheability(), + dsc.writable() && attribute.writeable, + attribute.executable, NO_FLUSH); + } catch (Page_table_registry::Mapping_cache_full full) { + if (full.reason == Page_table_registry::Mapping_cache_full::MEMORY) { + if (_ram_quota_guard().limit().value > 4 * 1024 * 1024) + /* we get in trouble in core if we use too much memory */ + throw Vm_space::Selector_allocator::Out_of_indices(); + throw Out_of_ram(); + } + if (full.reason == Page_table_registry::Mapping_cache_full::CAPS) + throw Out_of_caps(); + return; } - if (full.reason == Page_table_registry::Mapping_cache_full::CAPS) - throw Out_of_caps(); - return; } catch (Vm_space::Selector_allocator::Out_of_indices) { Genode::warning("run out of indices - flush all - cap=", _cap_quota_guard().used(), "/", @@ -277,12 +279,6 @@ void Vm_session_component::_attach_vm_memory(Dataspace_component &dsc, _ram_quota_guard().limit(), " guest=", Genode::Hex(0UL - _map.avail())); - _vm_space.map_guest(page.addr, page.hotspot, - (1 << page.log2_order) / 4096, - dsc.cacheability(), - dsc.writable() && attribute.writeable, - attribute.executable, FLUSH); - /* drop all attachment to limit ram usage of this session */ while (true) { addr_t out_addr = 0; @@ -292,9 +288,14 @@ void Vm_session_component::_attach_vm_memory(Dataspace_component &dsc, detach(out_addr); } - } catch (...) { - // Alloc_page_table_failed - Genode::error("alloc_guest_page_table exception"); + + _vm_space.map_guest(page.addr, page.hotspot, + (1 << page.log2_order) / 4096, + dsc.cacheability(), + dsc.writable() && attribute.writeable, + attribute.executable, FLUSH); + } catch (Vm_space::Alloc_page_table_failed) { + Genode::error("alloc page table failed"); return; } diff --git a/repos/base-sel4/src/lib/base/x86/vm_session.cc b/repos/base-sel4/src/lib/base/x86/vm_session.cc index 993a12c1e0..062000f57e 100644 --- a/repos/base-sel4/src/lib/base/x86/vm_session.cc +++ b/repos/base-sel4/src/lib/base/x86/vm_session.cc @@ -87,6 +87,8 @@ struct Vcpu : Genode::Thread addr_t const cr4_mask = CR4_VMX; addr_t const cr4_set = CR4_VMX; + seL4_VCPUContext _recent_gpr { }; + void entry() override { /* trigger that thread is up */ @@ -113,7 +115,7 @@ struct Vcpu : Genode::Thread /* initial startup VM exit to get valid VM state */ state.exit_reason = VMEXIT_STARTUP; - _read_sel4_state_async(service, state); + _read_sel4_state(service, state); Genode::Signal_transmitter(_signal).submit(); @@ -155,12 +157,9 @@ struct Vcpu : Genode::Thread state = Vm_state {}; - state.ip.value(seL4_GetMR(SEL4_VMENTER_CALL_EIP_MR)); - state.ctrl_primary.value(seL4_GetMR(SEL4_VMENTER_CALL_CONTROL_PPC_MR)); - state.inj_info.value(seL4_GetMR(SEL4_VMENTER_CALL_CONTROL_ENTRY_MR)); - state.exit_reason = VMEXIT_RECALL; - _read_sel4_state_async(service, state); + + _read_sel4_state(service, state); /* notify VM handler */ Genode::Signal_transmitter(_signal).submit(); @@ -189,36 +188,14 @@ struct Vcpu : Genode::Thread state = Vm_state {}; - if (res == SEL4_VMENTER_RESULT_FAULT) { - state.ip.value(seL4_GetMR(SEL4_VMENTER_CALL_EIP_MR)); - state.ctrl_primary.value(seL4_GetMR(SEL4_VMENTER_CALL_CONTROL_PPC_MR)); + if (res != SEL4_VMENTER_RESULT_FAULT) + state.exit_reason = VMEXIT_RECALL; + else state.exit_reason = seL4_GetMR(SEL4_VMENTER_FAULT_REASON_MR); - state.ip_len.value(seL4_GetMR(SEL4_VMENTER_FAULT_INSTRUCTION_LEN_MR)); - state.qual_primary.value(seL4_GetMR(SEL4_VMENTER_FAULT_QUALIFICATION_MR)); - state.qual_secondary.value(seL4_GetMR(SEL4_VMENTER_FAULT_GUEST_PHYSICAL_MR)); - - state.flags.value(seL4_GetMR(SEL4_VMENTER_FAULT_RFLAGS_MR)); - state.intr_state.value(seL4_GetMR(SEL4_VMENTER_FAULT_GUEST_INT_MR)); - state.cr3.value(seL4_GetMR(SEL4_VMENTER_FAULT_CR3_MR)); - - state.ax.value(seL4_GetMR(SEL4_VMENTER_FAULT_EAX)); - state.bx.value(seL4_GetMR(SEL4_VMENTER_FAULT_EBX)); - state.cx.value(seL4_GetMR(SEL4_VMENTER_FAULT_ECX)); - state.dx.value(seL4_GetMR(SEL4_VMENTER_FAULT_EDX)); - state.si.value(seL4_GetMR(SEL4_VMENTER_FAULT_ESI)); - state.di.value(seL4_GetMR(SEL4_VMENTER_FAULT_EDI)); - state.bp.value(seL4_GetMR(SEL4_VMENTER_FAULT_EBP)); - - _read_sel4_state(service, state); - } else { - state.ip.value(seL4_GetMR(SEL4_VMENTER_CALL_EIP_MR)); - state.ctrl_primary.value(seL4_GetMR(SEL4_VMENTER_CALL_CONTROL_PPC_MR)); - /* what about the other GPR stuff ? XXX */ - - state.exit_reason = VMEXIT_RECALL; - _read_sel4_state_async(service, state); + _read_sel4_state(service, state); + if (res != SEL4_VMENTER_RESULT_FAULT) { Lock::Guard guard(_remote_lock); if (_remote == PAUSE) { _remote = NONE; @@ -342,23 +319,6 @@ struct Vcpu : Genode::Thread " ", res.written); } - void _write_gpr(seL4_X86_VCPU const service, Vm_state &state) - { - seL4_VCPUContext regs; - regs.eax = state.ax.value(); - regs.ebx = state.bx.value(); - regs.ecx = state.cx.value(); - regs.edx = state.dx.value(); - regs.esi = state.si.value(); - regs.edi = state.di.value(); - regs.ebp = state.bp.value(); - - seL4_Error res = seL4_X86_VCPU_WriteRegisters(service, ®s); - if (res != seL4_NoError) - Genode::error("setting general purpose register failed ", - (int)res); - } - /* * Convert to Intel format comprising 32 bits. */ @@ -373,12 +333,24 @@ struct Vcpu : Genode::Thread void _write_sel4_state(seL4_X86_VCPU const service, Vm_state &state) { + if (state.ax.valid()) _recent_gpr.eax = state.ax.value(); + if (state.bx.valid()) _recent_gpr.ebx = state.bx.value(); + if (state.cx.valid()) _recent_gpr.ecx = state.cx.value(); + if (state.dx.valid()) _recent_gpr.edx = state.dx.value(); + if (state.si.valid()) _recent_gpr.esi = state.si.value(); + if (state.di.valid()) _recent_gpr.edi = state.di.value(); + if (state.bp.valid()) _recent_gpr.ebp = state.bp.value(); + if (state.ax.valid() || state.cx.valid() || state.dx.valid() || state.bx.valid() || state.bp.valid() || state.di.valid() || - state.si.valid()) { - /* XXX read first all values and write back only the changed ones ... */ - _write_gpr(service, state); + state.si.valid()) + { + seL4_Error res = seL4_X86_VCPU_WriteRegisters(service, + &_recent_gpr); + if (res != seL4_NoError) + Genode::error("setting general purpose registers failed ", + (int)res); } if (state.r8.valid() || state.r9.valid() || @@ -597,36 +569,35 @@ struct Vcpu : Genode::Thread uint32_t _read_vmcs_32(seL4_X86_VCPU const service, enum Vmcs const field) { return _read_vmcsX(service, field); } - void _read_sel4_state_async(seL4_X86_VCPU const service, Vm_state &state) - { -#if 0 - state.ax.value(state.ax.value()); /* XXX ? */ - state.cx.value(state.cx.value()); - state.dx.value(state.dx.value()); - state.bx.value(state.bx.value()); - - state.di.value(state.di.value()); /* XXX ? */ - state.si.value(state.si.value()); - state.bp.value(state.bp.value()); -#endif - - state.flags.value(_read_vmcs(service, Vmcs::RFLAGS)); - - state.ip.value(_read_vmcs(service, Vmcs::RIP)); - state.ip_len.value(_read_vmcs(service, Vmcs::EXIT_INST_LEN)); - - state.cr3.value(_read_vmcs(service, Vmcs::CR3)); - - state.qual_primary.value(state.qual_primary.value()); /* XXX ? */ - state.qual_secondary.value(state.qual_secondary.value()); /* XXX ? */ - - state.ctrl_primary.value(_read_vmcs(service, Vmcs::CTRL_0)); - - _read_sel4_state(service, state); - } - void _read_sel4_state(seL4_X86_VCPU const service, Vm_state &state) { + state.ip.value(seL4_GetMR(SEL4_VMENTER_CALL_EIP_MR)); + state.ctrl_primary.value(seL4_GetMR(SEL4_VMENTER_CALL_CONTROL_PPC_MR)); + + state.ip_len.value(seL4_GetMR(SEL4_VMENTER_FAULT_INSTRUCTION_LEN_MR)); + state.qual_primary.value(seL4_GetMR(SEL4_VMENTER_FAULT_QUALIFICATION_MR)); + state.qual_secondary.value(seL4_GetMR(SEL4_VMENTER_FAULT_GUEST_PHYSICAL_MR)); + + state.flags.value(seL4_GetMR(SEL4_VMENTER_FAULT_RFLAGS_MR)); + state.intr_state.value(seL4_GetMR(SEL4_VMENTER_FAULT_GUEST_INT_MR)); + state.cr3.value(seL4_GetMR(SEL4_VMENTER_FAULT_CR3_MR)); + + state.ax.value(seL4_GetMR(SEL4_VMENTER_FAULT_EAX)); + state.bx.value(seL4_GetMR(SEL4_VMENTER_FAULT_EBX)); + state.cx.value(seL4_GetMR(SEL4_VMENTER_FAULT_ECX)); + state.dx.value(seL4_GetMR(SEL4_VMENTER_FAULT_EDX)); + state.si.value(seL4_GetMR(SEL4_VMENTER_FAULT_ESI)); + state.di.value(seL4_GetMR(SEL4_VMENTER_FAULT_EDI)); + state.bp.value(seL4_GetMR(SEL4_VMENTER_FAULT_EBP)); + + _recent_gpr.eax = state.ax.value(); + _recent_gpr.ebx = state.bx.value(); + _recent_gpr.ecx = state.cx.value(); + _recent_gpr.edx = state.dx.value(); + _recent_gpr.esi = state.si.value(); + _recent_gpr.edi = state.di.value(); + _recent_gpr.ebp = state.bp.value(); + state.sp.value(_read_vmcs(service, Vmcs::RSP)); state.dr7.value(_read_vmcs(service, Vmcs::DR7)); diff --git a/repos/ports/run/seoul.inc b/repos/ports/run/seoul.inc index 1e08a6e593..c69eebad1b 100644 --- a/repos/ports/run/seoul.inc +++ b/repos/ports/run/seoul.inc @@ -21,6 +21,7 @@ set vmm_vcpu_same_cpu "no" if {[have_spec sel4]} { set map_small "yes" + set vmm_vcpu_same_cpu "yes" # seL4 has no AMD SVM support if {[have_include "power_on/qemu"]} { @@ -274,7 +275,7 @@ append_if $use_framebuffer config { if {!$use_fancy_stuff} { append config { - + } append config " "