From fc26e29e257c8d737b78e4581f7ffd9be338a70c Mon Sep 17 00:00:00 2001
From: Dave Stevenson <dave.stevenson@raspberrypi.com>
Date: Tue, 27 Apr 2021 14:24:21 +0200
Subject: [PATCH] drm/vc4: Add support for gamma on BCM2711

BCM2711 changes from a 256 entry lookup table to a 16 point
piecewise linear function as the pipeline bitdepth has increased
to make a LUT unwieldy.

Implement a simple conversion from a 256 entry LUT that userspace
is likely to expect to 16 evenly spread points in the PWL. This
could be improved with curve fitting at a later date.

Co-developed-by: Juerg Haefliger <juergh@canonical.com>
Signed-off-by: Juerg Haefliger <juergh@canonical.com>
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
---
 drivers/gpu/drm/vc4/vc4_crtc.c | 35 +++++++++++---
 drivers/gpu/drm/vc4/vc4_drv.h  | 28 +++++++++--
 drivers/gpu/drm/vc4/vc4_hvs.c  | 87 ++++++++++++++++++++++++++++++++--
 drivers/gpu/drm/vc4/vc4_regs.h | 22 +++++++++
 4 files changed, 159 insertions(+), 13 deletions(-)

--- a/drivers/gpu/drm/vc4/vc4_crtc.c
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -1148,19 +1148,42 @@ int vc4_crtc_init(struct drm_device *drm
 
 	if (!vc4->hvs->hvs5) {
 		drm_mode_crtc_set_gamma_size(crtc, ARRAY_SIZE(vc4_crtc->lut_r));
+	} else {
+		/* This is a lie for hvs5 which uses a 16 point PWL, but it
+		 * allows for something smarter than just 16 linearly spaced
+		 * segments. Conversion is done in vc5_hvs_update_gamma_lut.
+		 */
+		drm_mode_crtc_set_gamma_size(crtc, 256);
+	}
 
-		drm_crtc_enable_color_mgmt(crtc, 0, false, crtc->gamma_size);
+	drm_crtc_enable_color_mgmt(crtc, 0, false, crtc->gamma_size);
 
+	if (!vc4->hvs->hvs5) {
 		/* We support CTM, but only for one CRTC at a time. It's therefore
 		 * implemented as private driver state in vc4_kms, not here.
 		 */
 		drm_crtc_enable_color_mgmt(crtc, 0, true, crtc->gamma_size);
-	}
 
-	for (i = 0; i < crtc->gamma_size; i++) {
-		vc4_crtc->lut_r[i] = i;
-		vc4_crtc->lut_g[i] = i;
-		vc4_crtc->lut_b[i] = i;
+		/* Initialize the VC4 gamma LUTs */
+		for (i = 0; i < crtc->gamma_size; i++) {
+			vc4_crtc->lut_r[i] = i;
+			vc4_crtc->lut_g[i] = i;
+			vc4_crtc->lut_b[i] = i;
+		}
+	} else {
+		/* Initialize the VC5 gamma PWL entries. Assume 12-bit pipeline,
+		 * evenly spread over full range.
+		 */
+		for (i = 0; i < SCALER5_DSPGAMMA_NUM_POINTS; i++) {
+			vc4_crtc->pwl_r[i] =
+				VC5_HVS_SET_GAMMA_ENTRY(i << 8, i << 12, 1 << 8);
+			vc4_crtc->pwl_g[i] =
+				VC5_HVS_SET_GAMMA_ENTRY(i << 8, i << 12, 1 << 8);
+			vc4_crtc->pwl_b[i] =
+				VC5_HVS_SET_GAMMA_ENTRY(i << 8, i << 12, 1 << 8);
+			vc4_crtc->pwl_a[i] =
+				VC5_HVS_SET_GAMMA_ENTRY(i << 8, i << 12, 1 << 8);
+		}
 	}
 
 	return 0;
--- a/drivers/gpu/drm/vc4/vc4_drv.h
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -19,6 +19,7 @@
 #include <drm/drm_modeset_lock.h>
 
 #include "uapi/drm/vc4_drm.h"
+#include "vc4_regs.h"
 
 struct drm_device;
 struct drm_gem_object;
@@ -481,6 +482,17 @@ struct vc4_pv_data {
 
 };
 
+struct vc5_gamma_entry {
+	u32 x_c_terms;
+	u32 grad_term;
+};
+
+#define VC5_HVS_SET_GAMMA_ENTRY(x, c, g) (struct vc5_gamma_entry){	\
+	.x_c_terms = VC4_SET_FIELD((x), SCALER5_DSPGAMMA_OFF_X) | 	\
+		     VC4_SET_FIELD((c), SCALER5_DSPGAMMA_OFF_C),	\
+	.grad_term = (g)						\
+}
+
 struct vc4_crtc {
 	struct drm_crtc base;
 	struct platform_device *pdev;
@@ -490,9 +502,19 @@ struct vc4_crtc {
 	/* Timestamp at start of vblank irq - unaffected by lock delays. */
 	ktime_t t_vblank;
 
-	u8 lut_r[256];
-	u8 lut_g[256];
-	u8 lut_b[256];
+	union {
+		struct {  /* VC4 gamma LUT */
+			u8 lut_r[256];
+			u8 lut_g[256];
+			u8 lut_b[256];
+		};
+		struct {  /* VC5 gamma PWL entries */
+			struct vc5_gamma_entry pwl_r[SCALER5_DSPGAMMA_NUM_POINTS];
+			struct vc5_gamma_entry pwl_g[SCALER5_DSPGAMMA_NUM_POINTS];
+			struct vc5_gamma_entry pwl_b[SCALER5_DSPGAMMA_NUM_POINTS];
+			struct vc5_gamma_entry pwl_a[SCALER5_DSPGAMMA_NUM_POINTS];
+		};
+	};
 
 	struct drm_pending_vblank_event *event;
 
--- a/drivers/gpu/drm/vc4/vc4_hvs.c
+++ b/drivers/gpu/drm/vc4/vc4_hvs.c
@@ -259,6 +259,80 @@ u8 vc4_hvs_get_fifo_frame_count(struct d
 	return field;
 }
 
+static void vc5_hvs_write_gamma_entry(struct vc4_dev *vc4,
+				      u32 offset,
+				      struct vc5_gamma_entry *gamma)
+{
+	HVS_WRITE(offset, gamma->x_c_terms);
+	HVS_WRITE(offset + 4, gamma->grad_term);
+}
+
+static void vc5_hvs_lut_load(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
+	u32 i;
+	u32 offset = SCALER5_DSPGAMMA_START +
+		vc4_state->assigned_channel * SCALER5_DSPGAMMA_CHAN_OFFSET;
+
+	for (i = 0; i < SCALER5_DSPGAMMA_NUM_POINTS; i++, offset += 8)
+		vc5_hvs_write_gamma_entry(vc4, offset, &vc4_crtc->pwl_r[i]);
+	for (i = 0; i < SCALER5_DSPGAMMA_NUM_POINTS; i++, offset += 8)
+		vc5_hvs_write_gamma_entry(vc4, offset, &vc4_crtc->pwl_g[i]);
+	for (i = 0; i < SCALER5_DSPGAMMA_NUM_POINTS; i++, offset += 8)
+		vc5_hvs_write_gamma_entry(vc4, offset, &vc4_crtc->pwl_b[i]);
+
+	if (vc4_state->assigned_channel == 2) {
+		/* Alpha only valid on channel 2 */
+		for (i = 0; i < SCALER5_DSPGAMMA_NUM_POINTS; i++, offset += 8)
+			vc5_hvs_write_gamma_entry(vc4, offset, &vc4_crtc->pwl_a[i]);
+	}
+}
+
+static void vc5_hvs_update_gamma_lut(struct drm_crtc *crtc)
+{
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct drm_color_lut *lut = crtc->state->gamma_lut->data;
+	unsigned int step, i;
+	u32 start, end;
+
+#define VC5_HVS_UPDATE_GAMMA_ENTRY_FROM_LUT(pwl, chan)			\
+	start = drm_color_lut_extract(lut[i * step].chan, 12);		\
+	end = drm_color_lut_extract(lut[(i + 1) * step - 1].chan, 12);	\
+									\
+	/* Negative gradients not permitted by the hardware, so		\
+	 * flatten such points out.					\
+	 */								\
+	if (end < start)						\
+		end = start;						\
+									\
+	/* Assume 12bit pipeline.					\
+	 * X evenly spread over full range (12 bit).			\
+	 * C as U12.4 format.						\
+	 * Gradient as U4.8 format.					\
+	*/								\
+	vc4_crtc->pwl[i] =						\
+		VC5_HVS_SET_GAMMA_ENTRY(i << 8, start << 4,		\
+				((end - start) << 4) / (step - 1))
+
+	/* HVS5 has a 16 point piecewise linear function for each colour
+	 * channel (including alpha on channel 2) on each display channel.
+	 *
+	 * Currently take a crude subsample of the gamma LUT, but this could
+	 * be improved to implement curve fitting.
+	 */
+	step = crtc->gamma_size / SCALER5_DSPGAMMA_NUM_POINTS;
+	for (i = 0; i < SCALER5_DSPGAMMA_NUM_POINTS; i++) {
+		VC5_HVS_UPDATE_GAMMA_ENTRY_FROM_LUT(pwl_r, red);
+		VC5_HVS_UPDATE_GAMMA_ENTRY_FROM_LUT(pwl_g, green);
+		VC5_HVS_UPDATE_GAMMA_ENTRY_FROM_LUT(pwl_b, blue);
+	}
+
+	vc5_hvs_lut_load(crtc);
+}
+
 int vc4_hvs_get_fifo_from_output(struct drm_device *dev, unsigned int output)
 {
 	struct vc4_dev *vc4 = to_vc4_dev(dev);
@@ -352,14 +426,16 @@ static int vc4_hvs_init_channel(struct v
 	dispbkgndx &= ~SCALER_DISPBKGND_INTERLACE;
 
 	HVS_WRITE(SCALER_DISPBKGNDX(chan), dispbkgndx |
-		  SCALER_DISPBKGND_AUTOHS |
-		  ((!vc4->hvs->hvs5) ? SCALER_DISPBKGND_GAMMA : 0) |
+		  SCALER_DISPBKGND_AUTOHS | SCALER_DISPBKGND_GAMMA |
 		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
 
 	/* Reload the LUT, since the SRAMs would have been disabled if
 	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
 	 */
-	vc4_hvs_lut_load(crtc);
+	if (!vc4->hvs->hvs5)
+		vc4_hvs_lut_load(crtc);
+	else
+		vc5_hvs_lut_load(crtc);
 
 	return 0;
 }
@@ -557,7 +633,10 @@ void vc4_hvs_atomic_flush(struct drm_crt
 		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(vc4_state->assigned_channel));
 
 		if (crtc->state->gamma_lut) {
-			vc4_hvs_update_gamma_lut(crtc);
+			if (!vc4->hvs->hvs5)
+				vc4_hvs_update_gamma_lut(crtc);
+			else
+				vc5_hvs_update_gamma_lut(crtc);
 			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
 		} else {
 			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
--- a/drivers/gpu/drm/vc4/vc4_regs.h
+++ b/drivers/gpu/drm/vc4/vc4_regs.h
@@ -505,6 +505,28 @@
 #define SCALER_DLIST_START                      0x00002000
 #define SCALER_DLIST_SIZE                       0x00004000
 
+/* Gamma PWL for each channel. 16 points for each of 4 colour channels (alpha
+ * only on channel 2). 8 bytes per entry, offsets first, then gradient:
+ *   Y = GRAD * X + C
+ *
+ * Values for X and C are left justified, and vary depending on the width of
+ * the HVS channel:
+ *    8-bit pipeline: X uses [31:24], C is U8.8 format, and GRAD is U4.8.
+ *   12-bit pipeline: X uses [31:20], C is U12.4 format, and GRAD is U4.8.
+ *
+ * The 3 HVS channels start at 0x400 offsets (ie chan 1 starts at 0x2400, and
+ * chan 2 at 0x2800).
+ */
+#define SCALER5_DSPGAMMA_NUM_POINTS		16
+#define SCALER5_DSPGAMMA_START			0x00002000
+#define SCALER5_DSPGAMMA_CHAN_OFFSET		0x400
+# define SCALER5_DSPGAMMA_OFF_X_MASK		VC4_MASK(31, 20)
+# define SCALER5_DSPGAMMA_OFF_X_SHIFT		20
+# define SCALER5_DSPGAMMA_OFF_C_MASK		VC4_MASK(15, 0)
+# define SCALER5_DSPGAMMA_OFF_C_SHIFT		0
+# define SCALER5_DSPGAMMA_GRAD_MASK		VC4_MASK(11, 0)
+# define SCALER5_DSPGAMMA_GRAD_SHIFT		0
+
 #define SCALER5_DLIST_START			0x00004000
 
 # define VC4_HDMI_SW_RESET_FORMAT_DETECT	BIT(1)