Skip to contentSkip to conversation

Support IGT in display driver

PATCH Superseded Shawn Sung 5 participants 30 messages Jun 6, 2024 – Jun 21, 2024

Subsystems (3)

v8 of 9 Superseded Latest is v9 →

lore
Share & tools
lore

Shawn Sung

author
#0 lore raw

v8 cover Support IGT in display driver

From: Hsiao Chien Sung <[email protected]>

Based-on: commit 1613e604df0c ("Linux 6.10-rc1")

This series adds support for running IGT (Intel GPU Tool) tests with MediaTek display driver. The following changes will be applied:

1. Add a new API for creating GCE thread loop to retrieve CRCs from the hardware component
2. Support hardware CRC calculation in both VDOSYS0 and VDOSYS1
3. Support alpha blending in both VDOSYS0 and VDOSYS1

Changes in v8:
- Start/Stop CRC CMDQ thread on when needed
- Squash and rearrange the commits
- Add more information to the commit message and comments

Changes in v7:
- Separate the patch into smaller ones

Changes in v6:
- Use drm_vblank_work to deffer the CRC work into bottom halves
- Separate the patches for "Premultiplied" and "None" alpha blending

Changes in v5:
- Add more descriptions to the codes
- Add DRM mode configs to the driver data
- Squash and rearrange the commits

Changes in v4:
- Separate the patch into smaller ones
- Change the title of some patches
- Revert the changes that are not related to the series

Changes in v3:
- Modify the dt-binding document of Mediatek OVL
- Set DRM mode configs accroding to the hardware capabilities
- Replace cmdq_pkt_jump_absolute() with cmdq_pkt_jump()

Changes in v2:
- Simplify CMDQ by adding commands that are currently used only
- Integrate CRC related codes into new APIs for Mixer and OVL to reuse
- Add CPU version CRC retrieval when CMDQ is disabled

Hsiao Chien Sung (16):
soc: mediatek: Disable 9-bit alpha in ETHDR drm/mediatek: Add OVL compatible name for MT8195 drm/mediatek: Add missing plane settings when async update drm/mediatek: Add DRM_MODE_ROTATE_0 to rotation property drm/mediatek: Set DRM mode configs accordingly drm/mediatek: Turn off the layers with zero width or height drm/mediatek: Support more 10bit formats in OVL drm/mediatek: Support RGBA8888 and RGBX8888 in OVL on MT8195 drm/mediatek: Support "None" blending in OVL drm/mediatek: Support "None" blending in Mixer drm/mediatek: Support "Pre-multiplied" blending in OVL drm/mediatek: Support "Pre-multiplied" blending in Mixer drm/mediatek: Support alpha blending in display driver drm/mediatek: Support CRC in display driver drm/mediatek: Support CRC in OVL
drm/mediatek: Support CRC in OVL adaptor

 drivers/gpu/drm/mediatek/mtk_crtc.c           | 280 ++++++++++++++++
 drivers/gpu/drm/mediatek/mtk_crtc.h           |  38 +++
 drivers/gpu/drm/mediatek/mtk_ddp_comp.c       |  10 +
 drivers/gpu/drm/mediatek/mtk_ddp_comp.h       |  11 +-
 drivers/gpu/drm/mediatek/mtk_disp_drv.h       |  10 +
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c       | 316 ++++++++++++++++--
 .../gpu/drm/mediatek/mtk_disp_ovl_adaptor.c   |  37 +-
 drivers/gpu/drm/mediatek/mtk_drm_drv.c        |  24 ++
 drivers/gpu/drm/mediatek/mtk_drm_drv.h        |   4 +
 drivers/gpu/drm/mediatek/mtk_ethdr.c          | 106 +++++-
 drivers/gpu/drm/mediatek/mtk_ethdr.h          |   7 +
 drivers/gpu/drm/mediatek/mtk_plane.c          |  15 +-
 drivers/soc/mediatek/mtk-mmsys.c              |   1 +
 13 files changed, 816 insertions(+), 43 deletions(-)

--
2.18.0

author
#1 lore raw

PATCH v8 1/16 soc: mediatek: Disable 9-bit alpha in ETHDR

From: Hsiao Chien Sung <[email protected]>

When 9-bit alpha is enabled, its value will be converted from 0-255 to 0-256 (255 = not defined). This is designed for special HDR related calculation, which should be disabled by default, otherwise, alpha blending will not work correctly.

---
 drivers/soc/mediatek/mtk-mmsys.c | 1 +
 1 file changed, 1 insertion(+)
drivers/soc/mediatek/mtk-mmsys.c +1 -0
index f370f4ec4b88..938240714e54 100644
@@ -236,6 +236,7 @@ void mtk_mmsys_mixer_in_config(struct device *dev, int idx, bool alpha_sel, u16
 
 	mtk_mmsys_update_bits(mmsys, MT8195_VDO1_MIXER_IN1_ALPHA + (idx - 1) * 4, ~0,
 			      alpha << 16 | alpha, cmdq_pkt);
+	mtk_mmsys_update_bits(mmsys, MT8195_VDO1_HDR_TOP_CFG, BIT(15 + idx), 0, cmdq_pkt);
 	mtk_mmsys_update_bits(mmsys, MT8195_VDO1_HDR_TOP_CFG, BIT(19 + idx),
 			      alpha_sel << (19 + idx), cmdq_pkt);
 	mtk_mmsys_update_bits(mmsys, MT8195_VDO1_MIXER_IN1_PAD + (idx - 1) * 4,
Signature · 1 line
2.18.0
author
#2 lore raw

PATCH v8 2/16 drm/mediatek: Add OVL compatible name for MT8195

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Add OVL compatible name for MT8195.
Without this commit, DRM won't work after modifying the device tree.

---
 drivers/gpu/drm/mediatek/mtk_drm_drv.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/mediatek/mtk_drm_drv.c +2 -0
index b5f605751b0a..8e047043202b 100644
@@ -743,6 +743,8 @@ static const struct of_device_id mtk_ddp_comp_dt_ids[] = {
 	  .data = (void *)MTK_DISP_OVL },
 	{ .compatible = "mediatek,mt8192-disp-ovl",
 	  .data = (void *)MTK_DISP_OVL },
+	{ .compatible = "mediatek,mt8195-disp-ovl",
+	  .data = (void *)MTK_DISP_OVL },
 	{ .compatible = "mediatek,mt8183-disp-ovl-2l",
 	  .data = (void *)MTK_DISP_OVL_2L },
 	{ .compatible = "mediatek,mt8192-disp-ovl-2l",
Signature · 1 line
2.18.0
author
#3 lore raw

PATCH v8 3/16 drm/mediatek: Add missing plane settings when async update

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Fix an issue that plane coordinate was not saved when calling async update.

Fixes 920fffcc8912 ("drm/mediatek: update cursors by using async atomic update")
---
 drivers/gpu/drm/mediatek/mtk_plane.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/mediatek/mtk_plane.c +2 -0
index 4625deb21d40..a74b26d35985 100644
@@ -227,6 +227,8 @@ static void mtk_plane_atomic_async_update(struct drm_plane *plane,
 	plane->state->src_y = new_state->src_y;
 	plane->state->src_h = new_state->src_h;
 	plane->state->src_w = new_state->src_w;
+	plane->state->dst.x1 = new_state->dst.x1;
+	plane->state->dst.y1 = new_state->dst.y1;
 
 	mtk_plane_update_new_state(new_state, new_plane_state);
 	swap(plane->state->fb, new_state->fb);
Signature · 1 line
2.18.0
author
#4 lore raw

PATCH v8 4/16 drm/mediatek: Add DRM_MODE_ROTATE_0 to rotation property

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Always add DRM_MODE_ROTATE_0 to rotation property to meet IGT's (Intel GPU Tools) requirement.

---
 drivers/gpu/drm/mediatek/mtk_ddp_comp.h |  6 +++++-
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 17 +++++------------
 drivers/gpu/drm/mediatek/mtk_plane.c    |  2 +-
 3 files changed, 11 insertions(+), 14 deletions(-)
drivers/gpu/drm/mediatek/mtk_ddp_comp.h +5 -1
index 26236691ce4c..f7fe2e08dc8e 100644
@@ -192,7 +192,11 @@ unsigned int mtk_ddp_comp_supported_rotations(struct mtk_ddp_comp *comp)
 	if (comp->funcs && comp->funcs->supported_rotations)
 		return comp->funcs->supported_rotations(comp->dev);
 
-	return 0;
+	/*
+	 * In order to pass IGT tests, DRM_MODE_ROTATE_0 is required when
+	 * rotation is not supported.
+	 */
+	return DRM_MODE_ROTATE_0;
 }
 
 static inline unsigned int mtk_ddp_comp_layer_nr(struct mtk_ddp_comp *comp)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +5 -12
index b552a02d7eae..862ab683ed1b 100644
@@ -296,27 +296,20 @@ int mtk_ovl_layer_check(struct device *dev, unsigned int idx,
 			struct mtk_plane_state *mtk_state)
 {
 	struct drm_plane_state *state = &mtk_state->base;
-	unsigned int rotation = 0;
 
-	rotation = drm_rotation_simplify(state->rotation,
-					 DRM_MODE_ROTATE_0 |
-					 DRM_MODE_REFLECT_X |
-					 DRM_MODE_REFLECT_Y);
-	rotation &= ~DRM_MODE_ROTATE_0;
-
-	/* We can only do reflection, not rotation */
-	if ((rotation & DRM_MODE_ROTATE_MASK) != 0)
+	/* check if any unsupported rotation is set */
+	if (state->rotation & ~mtk_ovl_supported_rotations(dev))
 		return -EINVAL;
 
 	/*
 	 * TODO: Rotating/reflecting YUV buffers is not supported at this time.
 	 *	 Only RGB[AX] variants are supported.
+	 *	 Since DRM_MODE_ROTATE_0 means "no rotation", we should not
+	 *	 reject layers with this property.
 	 */
-	if (state->fb->format->is_yuv && rotation != 0)
+	if (state->fb->format->is_yuv && (state->rotation & ~DRM_MODE_ROTATE_0))
 		return -EINVAL;
 
-	state->rotation = rotation;
-
 	return 0;
 }
 
drivers/gpu/drm/mediatek/mtk_plane.c +1 -1
index a74b26d35985..1723d4333f37 100644
@@ -338,7 +338,7 @@ int mtk_plane_init(struct drm_device *dev, struct drm_plane *plane,
 		return err;
 	}
 
-	if (supported_rotations & ~DRM_MODE_ROTATE_0) {
+	if (supported_rotations) {
 		err = drm_plane_create_rotation_property(plane,
 							 DRM_MODE_ROTATE_0,
 							 supported_rotations);
Signature · 1 line
2.18.0
author
#5 lore raw

PATCH v8 5/16 drm/mediatek: Set DRM mode configs accordingly

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Set DRM mode configs limitation according to the hardware capabilities and pass the IGT checks as below:

- The test "graphics.IgtKms.kms_plane" requires a frame buffer with width of 4512 pixels (> 4096).
- The test "graphics.IgtKms.kms_cursor_crc" checks if the cursor size is defined, and run the test with cursor size from 1x1 to 512x512.

Please notice that the test conditions may change as IGT is updated.

---
 drivers/gpu/drm/mediatek/mtk_drm_drv.c | 22 ++++++++++++++++++++++
 drivers/gpu/drm/mediatek/mtk_drm_drv.h |  4 ++++
 2 files changed, 26 insertions(+)
drivers/gpu/drm/mediatek/mtk_drm_drv.c +22 -0
index 8e047043202b..c9cad3a82737 100644
@@ -294,6 +294,9 @@ static const struct mtk_mmsys_driver_data mt8188_vdosys0_driver_data = {
 	.conn_routes = mt8188_mtk_ddp_main_routes,
 	.num_conn_routes = ARRAY_SIZE(mt8188_mtk_ddp_main_routes),
 	.mmsys_dev_num = 2,
+	.max_width = 8191,
+	.min_width = 1,
+	.min_height = 1,
 };
 
 static const struct mtk_mmsys_driver_data mt8192_mmsys_driver_data = {
@@ -308,6 +311,9 @@ static const struct mtk_mmsys_driver_data mt8195_vdosys0_driver_data = {
 	.main_path = mt8195_mtk_ddp_main,
 	.main_len = ARRAY_SIZE(mt8195_mtk_ddp_main),
 	.mmsys_dev_num = 2,
+	.max_width = 8191,
+	.min_width = 1,
+	.min_height = 1,
 };
 
 static const struct mtk_mmsys_driver_data mt8195_vdosys1_driver_data = {
@@ -315,6 +321,9 @@ static const struct mtk_mmsys_driver_data mt8195_vdosys1_driver_data = {
 	.ext_len = ARRAY_SIZE(mt8195_mtk_ddp_ext),
 	.mmsys_id = 1,
 	.mmsys_dev_num = 2,
+	.max_width = 8191,
+	.min_width = 2, /* 2-pixel align when ethdr is bypassed */
+	.min_height = 1,
 };
 
 static const struct of_device_id mtk_drm_of_ids[] = {
@@ -493,6 +502,15 @@ static int mtk_drm_kms_init(struct drm_device *drm)
 		for (j = 0; j < private->data->mmsys_dev_num; j++) {
 			priv_n = private->all_drm_private[j];
 
+			if (priv_n->data->max_width)
+				drm->mode_config.max_width = priv_n->data->max_width;
+
+			if (priv_n->data->min_width)
+				drm->mode_config.min_width = priv_n->data->min_width;
+
+			if (priv_n->data->min_height)
+				drm->mode_config.min_height = priv_n->data->min_height;
+
 			if (i == CRTC_MAIN && priv_n->data->main_len) {
 				ret = mtk_crtc_create(drm, priv_n->data->main_path,
 						      priv_n->data->main_len, j,
@@ -520,6 +538,10 @@ static int mtk_drm_kms_init(struct drm_device *drm)
 		}
 	}
 
+	/* IGT will check if the cursor size is configured */
+	drm->mode_config.cursor_width = drm->mode_config.max_width;
+	drm->mode_config.cursor_height = drm->mode_config.max_height;
+
 	/* Use OVL device for all DMA memory allocations */
 	crtc = drm_crtc_from_index(drm, 0);
 	if (crtc)
drivers/gpu/drm/mediatek/mtk_drm_drv.h +4 -0
index 78d698ede1bf..6cfa790e8df5 100644
@@ -46,6 +46,10 @@ struct mtk_mmsys_driver_data {
 	bool shadow_register;
 	unsigned int mmsys_id;
 	unsigned int mmsys_dev_num;
+
+	int max_width;
+	int min_width;
+	int min_height;
 };
 
 struct mtk_drm_private {
Signature · 1 line
2.18.0
maintainer Reviewed-by on v8 5/16
#6 lore raw

Re: [PATCH v8 05/16] drm/mediatek: Set DRM mode configs accordingly

Il 06/06/24 11:26, Shawn Sung ha scritto:

Quoted 19 lines

From: Hsiao Chien Sung <[email protected]>

Set DRM mode configs limitation according to the hardware capabilities and pass the IGT checks as below:

- The test "graphics.IgtKms.kms_plane" requires a frame buffer with width of 4512 pixels (> 4096).
- The test "graphics.IgtKms.kms_cursor_crc" checks if the cursor size is defined, and run the test with cursor size from 1x1 to 512x512.

Please notice that the test conditions may change as IGT is updated.

---
  drivers/gpu/drm/mediatek/mtk_drm_drv.c | 22 ++++++++++++++++++++++
  drivers/gpu/drm/mediatek/mtk_drm_drv.h |  4 ++++
  2 files changed, 26 insertions(+)

..snip..

drivers/gpu/drm/mediatek/mtk_drm_drv.h struct mtk_mmsys_driver_data { · line 52

Quoted 12 lines
drivers/gpu/drm/mediatek/mtk_drm_drv.h +4 -0
index 78d698ede1bf..6cfa790e8df5 100644
@@ -46,6 +46,10 @@ struct mtk_mmsys_driver_data {
  	bool shadow_register;
  	unsigned int mmsys_id;
  	unsigned int mmsys_dev_num;
+
+	int max_width;
+	int min_width;
+	int min_height;

Only one nit here.

Since {min,max}_{width,height}:
- Will never be negative; and
- Will never need more than 16 bits; and
- Are local to this driver and anyway copied to drm later

Can you please change them to unsigned 16-bits (u16)?

After which,

Cheers,
Angelo

Quoted 3 lines
drivers/gpu/drm/mediatek/mtk_drm_drv.h +0 -0
  };
 
  struct mtk_drm_private {
author
#7 lore raw

PATCH v8 6/16 drm/mediatek: Turn off the layers with zero width or height

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

We found that IGT (Intel GPU Tool) will try to commit layers with zero width or height and lead to undefined behaviors in hardware. Disable the layers in such a situation.

Fixes 777b7bc86a0a ("UPSTREAM: drm/mediatek: Add ovl_adaptor support for MT8195")
Fixes fa97fe71f6f9 ("UPSTREAM: drm/mediatek: Add ETHDR support for MT8195")
---
 drivers/gpu/drm/mediatek/mtk_disp_ovl_adaptor.c | 2 +-
 drivers/gpu/drm/mediatek/mtk_ethdr.c            | 7 ++++++-
 2 files changed, 7 insertions(+), 2 deletions(-)
drivers/gpu/drm/mediatek/mtk_disp_ovl_adaptor.c +1 -1
index 02dd7dcdfedb..2b62d6475918 100644
@@ -158,7 +158,7 @@ void mtk_ovl_adaptor_layer_config(struct device *dev, unsigned int idx,
 	merge = ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_MERGE0 + idx];
 	ethdr = ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_ETHDR0];
 
-	if (!pending->enable) {
+	if (!pending->enable || !pending->width || !pending->height) {
 		mtk_merge_stop_cmdq(merge, cmdq_pkt);
 		mtk_mdp_rdma_stop(rdma_l, cmdq_pkt);
 		mtk_mdp_rdma_stop(rdma_r, cmdq_pkt);
drivers/gpu/drm/mediatek/mtk_ethdr.c +6 -1
index 156c6ff547e8..4ffd0a064861 100644
@@ -160,7 +160,12 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	if (idx >= 4)
 		return;
 
-	if (!pending->enable) {
+	if (!pending->enable || !pending->width || !pending->height) {
+		/*
+		 * instead of disabling layer with MIX_SRC_CON directly
+		 * set the size to 0 to avoid screen shift due to mixer
+		 * mode switch (hardware behavior)
+		 */
 		mtk_ddp_write(cmdq_pkt, 0, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_SIZE(idx));
 		return;
 	}
Signature · 1 line
2.18.0
author
#8 lore raw

PATCH v8 7/16 drm/mediatek: Support more 10bit formats in OVL

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support more 10bit formats in OVL.

---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 32 ++++++++++++++++++++++---
 1 file changed, 29 insertions(+), 3 deletions(-)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +29 -3
index 862ab683ed1b..d970cdce06bc 100644
@@ -71,6 +71,22 @@
 #define	OVL_CON_VIRT_FLIP	BIT(9)
 #define	OVL_CON_HORZ_FLIP	BIT(10)
 
+static inline bool is_10bit_rgb(u32 fmt)
+{
+	switch (fmt) {
+	case DRM_FORMAT_XRGB2101010:
+	case DRM_FORMAT_ARGB2101010:
+	case DRM_FORMAT_RGBX1010102:
+	case DRM_FORMAT_RGBA1010102:
+	case DRM_FORMAT_XBGR2101010:
+	case DRM_FORMAT_ABGR2101010:
+	case DRM_FORMAT_BGRX1010102:
+	case DRM_FORMAT_BGRA1010102:
+		return true;
+	}
+	return false;
+}
+
 static const u32 mt8173_formats[] = {
 	DRM_FORMAT_XRGB8888,
 	DRM_FORMAT_ARGB8888,
@@ -88,12 +104,18 @@ static const u32 mt8173_formats[] = {
 static const u32 mt8195_formats[] = {
 	DRM_FORMAT_XRGB8888,
 	DRM_FORMAT_ARGB8888,
+	DRM_FORMAT_XRGB2101010,
 	DRM_FORMAT_ARGB2101010,
 	DRM_FORMAT_BGRX8888,
 	DRM_FORMAT_BGRA8888,
+	DRM_FORMAT_BGRX1010102,
 	DRM_FORMAT_BGRA1010102,
 	DRM_FORMAT_ABGR8888,
 	DRM_FORMAT_XBGR8888,
+	DRM_FORMAT_XBGR2101010,
+	DRM_FORMAT_ABGR2101010,
+	DRM_FORMAT_RGBX1010102,
+	DRM_FORMAT_RGBA1010102,
 	DRM_FORMAT_RGB888,
 	DRM_FORMAT_BGR888,
 	DRM_FORMAT_RGB565,
@@ -253,9 +275,7 @@ static void mtk_ovl_set_bit_depth(struct device *dev, int idx, u32 format,
 	reg = readl(ovl->regs + DISP_REG_OVL_CLRFMT_EXT);
 	reg &= ~OVL_CON_CLRFMT_BIT_DEPTH_MASK(idx);
 
-	if (format == DRM_FORMAT_RGBA1010102 ||
-	    format == DRM_FORMAT_BGRA1010102 ||
-	    format == DRM_FORMAT_ARGB2101010)
+	if (is_10bit_rgb(format))
 		bit_depth = OVL_CON_CLRFMT_10_BIT;
 
 	reg |= OVL_CON_CLRFMT_BIT_DEPTH(bit_depth, idx);
@@ -368,17 +388,23 @@ static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt)
 		return OVL_CON_CLRFMT_RGB888(ovl) | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_RGBX8888:
 	case DRM_FORMAT_RGBA8888:
+	case DRM_FORMAT_RGBX1010102:
+	case DRM_FORMAT_RGBA1010102:
 		return OVL_CON_CLRFMT_ARGB8888;
 	case DRM_FORMAT_BGRX8888:
 	case DRM_FORMAT_BGRA8888:
+	case DRM_FORMAT_BGRX1010102:
 	case DRM_FORMAT_BGRA1010102:
 		return OVL_CON_CLRFMT_ARGB8888 | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_XRGB8888:
 	case DRM_FORMAT_ARGB8888:
+	case DRM_FORMAT_XRGB2101010:
 	case DRM_FORMAT_ARGB2101010:
 		return OVL_CON_CLRFMT_RGBA8888;
 	case DRM_FORMAT_XBGR8888:
 	case DRM_FORMAT_ABGR8888:
+	case DRM_FORMAT_XBGR2101010:
+	case DRM_FORMAT_ABGR2101010:
 		return OVL_CON_CLRFMT_RGBA8888 | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_UYVY:
 		return OVL_CON_CLRFMT_UYVY | OVL_CON_MTX_YUV_TO_RGB;
Signature · 1 line
2.18.0
author
#10 lore raw

PATCH v8 8/16 drm/mediatek: Support RGBA8888 and RGBX8888 in OVL on MT8195

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support RGBA8888 and RGBX8888 formats in OVL on MT8195.

---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +2 -0
index d970cdce06bc..738244a6164e 100644
@@ -114,6 +114,8 @@ static const u32 mt8195_formats[] = {
 	DRM_FORMAT_XBGR8888,
 	DRM_FORMAT_XBGR2101010,
 	DRM_FORMAT_ABGR2101010,
+	DRM_FORMAT_RGBX8888,
+	DRM_FORMAT_RGBA8888,
 	DRM_FORMAT_RGBX1010102,
 	DRM_FORMAT_RGBA1010102,
 	DRM_FORMAT_RGB888,
Signature · 1 line
2.18.0
Reviewed-by on v8 8/16
#12 lore raw

Re: [PATCH v8 08/16] drm/mediatek: Support RGBA8888 and RGBX8888 in OVL on MT8195

In reply to Shawn Sung

Hi, Shawn:

Quoted 3 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support RGBA8888 and RGBX8888 formats in OVL on MT8195.

Quoted 19 lines
---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +2 -0
index d970cdce06bc..738244a6164e 100644
@@ -114,6 +114,8 @@ static const u32 mt8195_formats[] = {
 	DRM_FORMAT_XBGR8888,
 	DRM_FORMAT_XBGR2101010,
 	DRM_FORMAT_ABGR2101010,
+	DRM_FORMAT_RGBX8888,
+	DRM_FORMAT_RGBA8888,
 	DRM_FORMAT_RGBX1010102,
 	DRM_FORMAT_RGBA1010102,
 	DRM_FORMAT_RGB888,
author
#13 lore raw

PATCH v8 9/16 drm/mediatek: Support "None" blending in OVL

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "None" alpha blending mode on MediaTek's chips.

---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +14 -2
index 738244a6164e..54a6f11aa867 100644
@@ -38,6 +38,7 @@
 #define DISP_REG_OVL_PITCH_MSB(n)		(0x0040 + 0x20 * (n))
 #define OVL_PITCH_MSB_2ND_SUBBUF			BIT(16)
 #define DISP_REG_OVL_PITCH(n)			(0x0044 + 0x20 * (n))
+#define OVL_CONST_BLEND					BIT(28)
 #define DISP_REG_OVL_RDMA_CTRL(n)		(0x00c0 + 0x20 * (n))
 #define DISP_REG_OVL_RDMA_GMC(n)		(0x00c8 + 0x20 * (n))
 #define DISP_REG_OVL_ADDR_MT2701		0x0040
@@ -428,6 +429,8 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 	unsigned int fmt = pending->format;
 	unsigned int offset = (pending->y << 16) | pending->x;
 	unsigned int src_size = (pending->height << 16) | pending->width;
+	unsigned int blend_mode = state->base.pixel_blend_mode;
+	unsigned int ignore_pixel_alpha = 0;
 	unsigned int con;
 	bool is_afbc = pending->modifier != DRM_FORMAT_MOD_LINEAR;
 	union overlay_pitch {
@@ -449,6 +452,15 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 	if (state->base.fb && state->base.fb->format->has_alpha)
 		con |= OVL_CON_AEN | OVL_CON_ALPHA;
 
+	/* CONST_BLD must be enabled for XRGB formats although the alpha channel
+	 * can be ignored, or OVL will still read the value from memory.
+	 * For RGB888 related formats, whether CONST_BLD is enabled or not won't
+	 * affect the result. Therefore we use !has_alpha as the condition.
+	 */
+	if (blend_mode == DRM_MODE_BLEND_PIXEL_NONE ||
+	    (state->base.fb && !state->base.fb->format->has_alpha))
+		ignore_pixel_alpha = OVL_CONST_BLEND;
+
 	if (pending->rotation & DRM_MODE_REFLECT_Y) {
 		con |= OVL_CON_VIRT_FLIP;
 		addr += (pending->height - 1) * pending->pitch;
@@ -464,8 +476,8 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 
 	mtk_ddp_write_relaxed(cmdq_pkt, con, &ovl->cmdq_reg, ovl->regs,
 			      DISP_REG_OVL_CON(idx));
-	mtk_ddp_write_relaxed(cmdq_pkt, overlay_pitch.split_pitch.lsb, &ovl->cmdq_reg, ovl->regs,
-			      DISP_REG_OVL_PITCH(idx));
+	mtk_ddp_write_relaxed(cmdq_pkt, overlay_pitch.split_pitch.lsb | ignore_pixel_alpha,
+			      &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_PITCH(idx));
 	mtk_ddp_write_relaxed(cmdq_pkt, src_size, &ovl->cmdq_reg, ovl->regs,
 			      DISP_REG_OVL_SRC_SIZE(idx));
 	mtk_ddp_write_relaxed(cmdq_pkt, offset, &ovl->cmdq_reg, ovl->regs,

--
2.18.0

#14 lore raw

Re: [PATCH v8 09/16] drm/mediatek: Support "None" blending in OVL

Hi, Shawn:

Quoted 31 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "None" alpha blending mode on MediaTek's chips.

---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
index 738244a6164e..54a6f11aa867 100644
@@ -38,6 +38,7 @@
 #define DISP_REG_OVL_PITCH_MSB(n)		(0x0040 + 0x20 * (n))
 #define OVL_PITCH_MSB_2ND_SUBBUF			BIT(16)
 #define DISP_REG_OVL_PITCH(n)			(0x0044 + 0x20 * (n))
+#define OVL_CONST_BLEND					BIT(28)
 #define DISP_REG_OVL_RDMA_CTRL(n)		(0x00c0 + 0x20 * (n))
 #define DISP_REG_OVL_RDMA_GMC(n)		(0x00c8 + 0x20 * (n))
 #define DISP_REG_OVL_ADDR_MT2701		0x0040
@@ -428,6 +429,8 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 	unsigned int fmt = pending->format;
 	unsigned int offset = (pending->y << 16) | pending->x;
 	unsigned int src_size = (pending->height << 16) | pending->width;
+	unsigned int blend_mode = state->base.pixel_blend_mode;
+	unsigned int ignore_pixel_alpha = 0;
 	unsigned int con;
 	bool is_afbc = pending->modifier != DRM_FORMAT_MOD_LINEAR;
 	union overlay_pitch {

drivers/gpu/drm/mediatek/mtk_disp_ovl.c void mtk_ovl_layer_config(struct device *dev, unsigned int idx, · line 459

Quoted 8 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +5 -0
@@ -449,6 +452,15 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 	if (state->base.fb && state->base.fb->format->has_alpha)
 		con |= OVL_CON_AEN | OVL_CON_ALPHA;
 
+	/* CONST_BLD must be enabled for XRGB formats although the alpha channel
+	 * can be ignored, or OVL will still read the value from memory.
+	 * For RGB888 related formats, whether CONST_BLD is enabled or not won't
+	 * affect the result. Therefore we use !has_alpha as the condition.
+	 */

This patch looks good to me, but OVL driver has already declare some X format. So I would like to separate X format related part to a fix patch.

Regards,
CK

Quoted 21 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +6 -3
+	if (blend_mode == DRM_MODE_BLEND_PIXEL_NONE ||
+	    (state->base.fb && !state->base.fb->format->has_alpha))
+		ignore_pixel_alpha = OVL_CONST_BLEND;
+
 	if (pending->rotation & DRM_MODE_REFLECT_Y) {
 		con |= OVL_CON_VIRT_FLIP;
 		addr += (pending->height - 1) * pending->pitch;
@@ -464,8 +476,8 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 
 	mtk_ddp_write_relaxed(cmdq_pkt, con, &ovl->cmdq_reg, ovl->regs,
 			      DISP_REG_OVL_CON(idx));
-	mtk_ddp_write_relaxed(cmdq_pkt, overlay_pitch.split_pitch.lsb, &ovl->cmdq_reg, ovl->regs,
-			      DISP_REG_OVL_PITCH(idx));
+	mtk_ddp_write_relaxed(cmdq_pkt, overlay_pitch.split_pitch.lsb | ignore_pixel_alpha,
+			      &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_PITCH(idx));
 	mtk_ddp_write_relaxed(cmdq_pkt, src_size, &ovl->cmdq_reg, ovl->regs,
 			      DISP_REG_OVL_SRC_SIZE(idx));
 	mtk_ddp_write_relaxed(cmdq_pkt, offset, &ovl->cmdq_reg, ovl->regs,
--

2.18.0

author
#15 lore raw

PATCH v8 10/16 drm/mediatek: Support "None" blending in Mixer

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "None" alpha blending mode on MediaTek's chips.

---
 drivers/gpu/drm/mediatek/mtk_ethdr.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
drivers/gpu/drm/mediatek/mtk_ethdr.c +12 -2
index 4ffd0a064861..bcced62e455d 100644
@@ -3,6 +3,7 @@
  * Copyright (c) 2021 MediaTek Inc.
  */
 
+#include <drm/drm_blend.h>
 #include <drm/drm_fourcc.h>
 #include <drm/drm_framebuffer.h>
 #include <linux/clk.h>
@@ -154,6 +155,7 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	unsigned int offset = (pending->x & 1) << 31 | pending->y << 16 | pending->x;
 	unsigned int align_width = ALIGN_DOWN(pending->width, 2);
 	unsigned int alpha_con = 0;
+	bool replace_src_a = false;
 
 	dev_dbg(dev, "%s+ idx:%d", __func__, idx);
 
@@ -173,8 +175,16 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	if (state->base.fb && state->base.fb->format->has_alpha)
 		alpha_con = MIXER_ALPHA_AEN | MIXER_ALPHA;
 
-	mtk_mmsys_mixer_in_config(priv->mmsys_dev, idx + 1, alpha_con ? false : true,
-				  DEFAULT_9BIT_ALPHA,
+	if (state->base.pixel_blend_mode == DRM_MODE_BLEND_PIXEL_NONE ||
+	    (state->base.fb && !state->base.fb->format->has_alpha)) {
+		/*
+		 * Mixer doesn't support CONST_BLD mode,
+		 * use a trick to make the output equivalent
+		 */
+		replace_src_a = true;
+	}
+
+	mtk_mmsys_mixer_in_config(priv->mmsys_dev, idx + 1, replace_src_a, MIXER_ALPHA,
 				  pending->x & 1 ? MIXER_INX_MODE_EVEN_EXTEND :
 				  MIXER_INX_MODE_BYPASS, align_width / 2 - 1, cmdq_pkt);
 

--
2.18.0

#16 lore raw

Re: [PATCH v8 10/16] drm/mediatek: Support "None" blending in Mixer

Hi, Shawn:

Quoted 35 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "None" alpha blending mode on MediaTek's chips.

---
 drivers/gpu/drm/mediatek/mtk_ethdr.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
drivers/gpu/drm/mediatek/mtk_ethdr.c +2 -2
index 4ffd0a064861..bcced62e455d 100644
@@ -3,6 +3,7 @@
  * Copyright (c) 2021 MediaTek Inc.
  */
 
+#include <drm/drm_blend.h>
 #include <drm/drm_fourcc.h>
 #include <drm/drm_framebuffer.h>
 #include <linux/clk.h>
@@ -154,6 +155,7 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	unsigned int offset = (pending->x & 1) << 31 | pending->y << 16 | pending->x;
 	unsigned int align_width = ALIGN_DOWN(pending->width, 2);
 	unsigned int alpha_con = 0;
+	bool replace_src_a = false;
 
 	dev_dbg(dev, "%s+ idx:%d", __func__, idx);
 
@@ -173,8 +175,16 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	if (state->base.fb && state->base.fb->format->has_alpha)
 		alpha_con = MIXER_ALPHA_AEN | MIXER_ALPHA;
 
-	mtk_mmsys_mixer_in_config(priv->mmsys_dev, idx + 1, alpha_con ? false : true,
-				  DEFAULT_9BIT_ALPHA,

drivers/gpu/drm/mediatek/mtk_ethdr.c void mtk_ethdr_layer_config(struct device *dev, unsigned int idx, · line 187

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_ethdr.c +10 -0
+	if (state->base.pixel_blend_mode == DRM_MODE_BLEND_PIXEL_NONE ||
+	    (state->base.fb && !state->base.fb->format->has_alpha)) {
+		/*
+		 * Mixer doesn't support CONST_BLD mode,
+		 * use a trick to make the output equivalent
+		 */
+		replace_src_a = true;
+	}
+
+	mtk_mmsys_mixer_in_config(priv->mmsys_dev, idx + 1, replace_src_a, MIXER_ALPHA,

It seems originally replace source alpha with 9 bits constant alpha, but this patch change to 8 bits constant alpha. Changing from 9 bits to 8 bits is not related to this patch, so separate it to another patch.

Regards,
CK

Quoted 7 lines
drivers/gpu/drm/mediatek/mtk_ethdr.c +0 -1
 				  pending->x & 1 ? MIXER_INX_MODE_EVEN_EXTEND :
 				  MIXER_INX_MODE_BYPASS, align_width / 2 - 1, cmdq_pkt);
 
--

2.18.0

author
#17 lore raw

PATCH v8 11/16 drm/mediatek: Support "Pre-multiplied" blending in OVL

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "Pre-multiplied" alpha blending mode on in OVL. Before this patch, only the "coverage" mode is supported.

---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 40 +++++++++++++++++++++----
 1 file changed, 34 insertions(+), 6 deletions(-)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +34 -6
index 54a6f11aa867..de1633988921 100644
@@ -52,13 +52,16 @@
 #define GMC_THRESHOLD_HIGH	((1 << GMC_THRESHOLD_BITS) / 4)
 #define GMC_THRESHOLD_LOW	((1 << GMC_THRESHOLD_BITS) / 8)
 
+#define OVL_CON_CLRFMT_MAN	BIT(23)
 #define OVL_CON_BYTE_SWAP	BIT(24)
-#define OVL_CON_MTX_YUV_TO_RGB	(6 << 16)
+#define OVL_CON_RGB_SWAP	BIT(25)
 #define OVL_CON_CLRFMT_RGB	(1 << 12)
 #define OVL_CON_CLRFMT_RGBA8888	(2 << 12)
 #define OVL_CON_CLRFMT_ARGB8888	(3 << 12)
 #define OVL_CON_CLRFMT_UYVY	(4 << 12)
 #define OVL_CON_CLRFMT_YUYV	(5 << 12)
+#define OVL_CON_MTX_YUV_TO_RGB	(6 << 16)
+#define OVL_CON_CLRFMT_PARGB8888	(OVL_CON_CLRFMT_ARGB8888 | OVL_CON_CLRFMT_MAN)
 #define OVL_CON_CLRFMT_RGB565(ovl)	((ovl)->data->fmt_rgb565_is_0 ? \
 					0 : OVL_CON_CLRFMT_RGB)
 #define OVL_CON_CLRFMT_RGB888(ovl)	((ovl)->data->fmt_rgb565_is_0 ? \
@@ -72,6 +75,8 @@
 #define	OVL_CON_VIRT_FLIP	BIT(9)
 #define	OVL_CON_HORZ_FLIP	BIT(10)
 
+#define OVL_COLOR_ALPHA		GENMASK(31, 24)
+
 static inline bool is_10bit_rgb(u32 fmt)
 {
 	switch (fmt) {
@@ -296,7 +301,13 @@ void mtk_ovl_config(struct device *dev, unsigned int w,
 	if (w != 0 && h != 0)
 		mtk_ddp_write_relaxed(cmdq_pkt, h << 16 | w, &ovl->cmdq_reg, ovl->regs,
 				      DISP_REG_OVL_ROI_SIZE);
-	mtk_ddp_write_relaxed(cmdq_pkt, 0x0, &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_ROI_BGCLR);
+
+	/*
+	 * The background color must be opaque black (ARGB),
+	 * otherwise the alpha blending will have no effect
+	 */
+	mtk_ddp_write_relaxed(cmdq_pkt, OVL_COLOR_ALPHA, &ovl->cmdq_reg,
+			      ovl->regs, DISP_REG_OVL_ROI_BGCLR);
 
 	mtk_ddp_write(cmdq_pkt, 0x1, &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_RST);
 	mtk_ddp_write(cmdq_pkt, 0x0, &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_RST);
@@ -372,7 +383,8 @@ void mtk_ovl_layer_off(struct device *dev, unsigned int idx,
 		      DISP_REG_OVL_RDMA_CTRL(idx));
 }
 
-static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt)
+static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt,
+				    unsigned int blend_mode)
 {
 	/* The return value in switch "MEM_MODE_INPUT_FORMAT_XXX"
 	 * is defined in mediatek HW data sheet.
@@ -391,21 +403,35 @@ static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt)
 		return OVL_CON_CLRFMT_RGB888(ovl) | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_RGBX8888:
 	case DRM_FORMAT_RGBA8888:
+		return blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_ARGB8888 :
+		       OVL_CON_CLRFMT_PARGB8888;
 	case DRM_FORMAT_RGBX1010102:
 	case DRM_FORMAT_RGBA1010102:
 		return OVL_CON_CLRFMT_ARGB8888;
 	case DRM_FORMAT_BGRX8888:
 	case DRM_FORMAT_BGRA8888:
+		return OVL_CON_BYTE_SWAP |
+		       (blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_ARGB8888 :
+		       OVL_CON_CLRFMT_PARGB8888);
 	case DRM_FORMAT_BGRX1010102:
 	case DRM_FORMAT_BGRA1010102:
 		return OVL_CON_CLRFMT_ARGB8888 | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_XRGB8888:
 	case DRM_FORMAT_ARGB8888:
+		return blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_RGBA8888 :
+		       OVL_CON_CLRFMT_PARGB8888;
 	case DRM_FORMAT_XRGB2101010:
 	case DRM_FORMAT_ARGB2101010:
 		return OVL_CON_CLRFMT_RGBA8888;
 	case DRM_FORMAT_XBGR8888:
 	case DRM_FORMAT_ABGR8888:
+		return OVL_CON_RGB_SWAP |
+		       (blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_RGBA8888 :
+		       OVL_CON_CLRFMT_PARGB8888);
 	case DRM_FORMAT_XBGR2101010:
 	case DRM_FORMAT_ABGR2101010:
 		return OVL_CON_CLRFMT_RGBA8888 | OVL_CON_BYTE_SWAP;
@@ -448,9 +474,11 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 		return;
 	}
 
-	con = ovl_fmt_convert(ovl, fmt);
-	if (state->base.fb && state->base.fb->format->has_alpha)
-		con |= OVL_CON_AEN | OVL_CON_ALPHA;
+	con = ovl_fmt_convert(ovl, fmt, blend_mode);
+	if (state->base.fb) {
+		con |= OVL_CON_AEN;
+		con |= state->base.alpha & OVL_CON_ALPHA;
+	}
 
 	/* CONST_BLD must be enabled for XRGB formats although the alpha channel
 	 * can be ignored, or OVL will still read the value from memory.
Signature · 1 line
2.18.0
#18 lore raw

Re: [PATCH v8 11/16] drm/mediatek: Support "Pre-multiplied" blending in OVL

Hi, Shawn:

Quoted 43 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "Pre-multiplied" alpha blending mode on in OVL. Before this patch, only the "coverage" mode is supported.

---
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 40 +++++++++++++++++++++----
 1 file changed, 34 insertions(+), 6 deletions(-)
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +6 -1
index 54a6f11aa867..de1633988921 100644
@@ -52,13 +52,16 @@
 #define GMC_THRESHOLD_HIGH	((1 << GMC_THRESHOLD_BITS) / 4)
 #define GMC_THRESHOLD_LOW	((1 << GMC_THRESHOLD_BITS) / 8)
 
+#define OVL_CON_CLRFMT_MAN	BIT(23)
 #define OVL_CON_BYTE_SWAP	BIT(24)
-#define OVL_CON_MTX_YUV_TO_RGB	(6 << 16)
+#define OVL_CON_RGB_SWAP	BIT(25)
 #define OVL_CON_CLRFMT_RGB	(1 << 12)
 #define OVL_CON_CLRFMT_RGBA8888	(2 << 12)
 #define OVL_CON_CLRFMT_ARGB8888	(3 << 12)
 #define OVL_CON_CLRFMT_UYVY	(4 << 12)
 #define OVL_CON_CLRFMT_YUYV	(5 << 12)
+#define OVL_CON_MTX_YUV_TO_RGB	(6 << 16)
+#define OVL_CON_CLRFMT_PARGB8888	(OVL_CON_CLRFMT_ARGB8888 | OVL_CON_CLRFMT_MAN)
 #define OVL_CON_CLRFMT_RGB565(ovl)	((ovl)->data->fmt_rgb565_is_0 ? \
 					0 : OVL_CON_CLRFMT_RGB)
 #define OVL_CON_CLRFMT_RGB888(ovl)	((ovl)->data->fmt_rgb565_is_0 ? \
@@ -72,6 +75,8 @@
 #define	OVL_CON_VIRT_FLIP	BIT(9)
 #define	OVL_CON_HORZ_FLIP	BIT(10)
 
+#define OVL_COLOR_ALPHA		GENMASK(31, 24)
+
 static inline bool is_10bit_rgb(u32 fmt)
 {
 	switch (fmt) {
@@ -296,7 +301,13 @@ void mtk_ovl_config(struct device *dev, unsigned int w,
 	if (w != 0 && h != 0)

drivers/gpu/drm/mediatek/mtk_disp_ovl.c void mtk_ovl_config(struct device *dev, unsigned int w, · line 310

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +7 -1
 		mtk_ddp_write_relaxed(cmdq_pkt, h << 16 | w, &ovl->cmdq_reg, ovl->regs,
 				      DISP_REG_OVL_ROI_SIZE);
-	mtk_ddp_write_relaxed(cmdq_pkt, 0x0, &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_ROI_BGCLR);
+
+	/*
+	 * The background color must be opaque black (ARGB),
+	 * otherwise the alpha blending will have no effect
+	 */
+	mtk_ddp_write_relaxed(cmdq_pkt, OVL_COLOR_ALPHA, &ovl->cmdq_reg,
+			      ovl->regs, DISP_REG_OVL_ROI_BGCLR);

The coverage mode formula is:
dst.RGB = src.RGB * src.A + dst.RGB * (1 - src.A)

The pre-multiplied mode formula is:
dst.RGB = src.RGB + dst.RGB * (1 - src.A)

Both formula has no destination alpha (I think background color is last destination), why coverage mode work fine but pre-multiplied mode has something wrong?

Quoted 27 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +7 -1
 	mtk_ddp_write(cmdq_pkt, 0x1, &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_RST);
 	mtk_ddp_write(cmdq_pkt, 0x0, &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_RST);
@@ -372,7 +383,8 @@ void mtk_ovl_layer_off(struct device *dev, unsigned int idx,
 		      DISP_REG_OVL_RDMA_CTRL(idx));
 }
 
-static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt)
+static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt,
+				    unsigned int blend_mode)
 {
 	/* The return value in switch "MEM_MODE_INPUT_FORMAT_XXX"
 	 * is defined in mediatek HW data sheet.
@@ -391,21 +403,35 @@ static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt)
 		return OVL_CON_CLRFMT_RGB888(ovl) | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_RGBX8888:
 	case DRM_FORMAT_RGBA8888:
+		return blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_ARGB8888 :
+		       OVL_CON_CLRFMT_PARGB8888;
 	case DRM_FORMAT_RGBX1010102:
 	case DRM_FORMAT_RGBA1010102:
 		return OVL_CON_CLRFMT_ARGB8888;
 	case DRM_FORMAT_BGRX8888:
 	case DRM_FORMAT_BGRA8888:
+		return OVL_CON_BYTE_SWAP |
+		       (blend_mode == DRM_MODE_BLEND_COVERAGE ?

drivers/gpu/drm/mediatek/mtk_disp_ovl.c static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt) · line 425

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +5 -0
+		       OVL_CON_CLRFMT_ARGB8888 :
+		       OVL_CON_CLRFMT_PARGB8888);
 	case DRM_FORMAT_BGRX1010102:
 	case DRM_FORMAT_BGRA1010102:
 		return OVL_CON_CLRFMT_ARGB8888 | OVL_CON_BYTE_SWAP;
 	case DRM_FORMAT_XRGB8888:
 	case DRM_FORMAT_ARGB8888:
+		return blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_RGBA8888 :
+		       OVL_CON_CLRFMT_PARGB8888;

In pre-multiplied mode, DRM_FORMAT_RGBA8888 and DRM_FORMAT_ARGB8888 would return the same value OVL_CON_CLRFMT_PARGB8888, so how does hardware distinguish these two format?

drivers/gpu/drm/mediatek/mtk_disp_ovl.c static unsigned int ovl_fmt_convert(struct mtk_disp_ovl *ovl, unsigned int fmt) · line 434

Quoted 9 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
 	case DRM_FORMAT_XRGB2101010:
 	case DRM_FORMAT_ARGB2101010:
 		return OVL_CON_CLRFMT_RGBA8888;
 	case DRM_FORMAT_XBGR8888:
 	case DRM_FORMAT_ABGR8888:
+		return OVL_CON_RGB_SWAP |
+		       (blend_mode == DRM_MODE_BLEND_COVERAGE ?
+		       OVL_CON_CLRFMT_RGBA8888 :
+		       OVL_CON_CLRFMT_PARGB8888);

Originally, DRM_MODE_BLEND_COVERAGE and DRM_FORMAT_ABGR8888 would return

OVL_CON_CLRFMT_RGBA8888 | OVL_CON_BYTE_SWAP

but this patch return

OVL_CON_CLRFMT_RGBA8888 | OVL_CON_RGB_SWAP

This is not related to pre-multiplied, so separate this to another patch.

Quoted 4 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +0 -0
 	case DRM_FORMAT_XBGR2101010:
 	case DRM_FORMAT_ABGR2101010:
 		return OVL_CON_CLRFMT_RGBA8888 | OVL_CON_BYTE_SWAP;

drivers/gpu/drm/mediatek/mtk_disp_ovl.c void mtk_ovl_layer_config(struct device *dev, unsigned int idx, · line 480

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -3
@@ -448,9 +474,11 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 		return;
 	}
 
-	con = ovl_fmt_convert(ovl, fmt);
-	if (state->base.fb && state->base.fb->format->has_alpha)
-		con |= OVL_CON_AEN | OVL_CON_ALPHA;
+	con = ovl_fmt_convert(ovl, fmt, blend_mode);
+	if (state->base.fb) {
+		con |= OVL_CON_AEN;
+		con |= state->base.alpha & OVL_CON_ALPHA;

This modification is to support constant alpha, not related to pre-multiplied alpha. So separate this to another patch.

Regards,
CK

Quoted 4 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +1 -0
+	}
 
 	/* CONST_BLD must be enabled for XRGB formats although the alpha channel
 	 * can be ignored, or OVL will still read the value from memory.
author
#19 lore raw

PATCH v8 12/16 drm/mediatek: Support "Pre-multiplied" blending in Mixer

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "Pre-multiplied" alpha blending mode in Mixer. Before this patch, only the coverage mode is supported.

---
 drivers/gpu/drm/mediatek/mtk_ethdr.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/gpu/drm/mediatek/mtk_ethdr.c +9 -4
index bcced62e455d..d01f65819816 100644
@@ -6,6 +6,7 @@
 #include <drm/drm_blend.h>
 #include <drm/drm_fourcc.h>
 #include <drm/drm_framebuffer.h>
+#include <drm/drm_blend.h>
 #include <linux/clk.h>
 #include <linux/component.h>
 #include <linux/of.h>
@@ -36,6 +37,7 @@
 #define MIX_SRC_L0_EN				BIT(0)
 #define MIX_L_SRC_CON(n)		(0x28 + 0x18 * (n))
 #define NON_PREMULTI_SOURCE			(2 << 12)
+#define PREMULTI_SOURCE				(3 << 12)
 #define MIX_L_SRC_SIZE(n)		(0x30 + 0x18 * (n))
 #define MIX_L_SRC_OFFSET(n)		(0x34 + 0x18 * (n))
 #define MIX_FUNC_DCM0			0x120
@@ -172,8 +174,12 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 		return;
 	}
 
-	if (state->base.fb && state->base.fb->format->has_alpha)
-		alpha_con = MIXER_ALPHA_AEN | MIXER_ALPHA;
+	alpha_con |= MIXER_ALPHA_AEN | (state->base.alpha & MIXER_ALPHA);
+
+	if (state->base.pixel_blend_mode != DRM_MODE_BLEND_COVERAGE)
+		alpha_con |= PREMULTI_SOURCE;
+	else
+		alpha_con |= NON_PREMULTI_SOURCE;
 
 	if (state->base.pixel_blend_mode == DRM_MODE_BLEND_PIXEL_NONE ||
 	    (state->base.fb && !state->base.fb->format->has_alpha)) {
@@ -191,8 +197,7 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	mtk_ddp_write(cmdq_pkt, pending->height << 16 | align_width, &mixer->cmdq_base,
 		      mixer->regs, MIX_L_SRC_SIZE(idx));
 	mtk_ddp_write(cmdq_pkt, offset, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_OFFSET(idx));
-	mtk_ddp_write_mask(cmdq_pkt, alpha_con, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_CON(idx),
-			   0x1ff);
+	mtk_ddp_write(cmdq_pkt, alpha_con, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_CON(idx));
 	mtk_ddp_write_mask(cmdq_pkt, BIT(idx), &mixer->cmdq_base, mixer->regs, MIX_SRC_CON,
 			   BIT(idx));
 }
Signature · 1 line
2.18.0
#20 lore raw

Re: [PATCH v8 12/16] drm/mediatek: Support "Pre-multiplied" blending in Mixer

Hi, Shawn:

Quoted 31 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "Pre-multiplied" alpha blending mode in Mixer. Before this patch, only the coverage mode is supported.

---
 drivers/gpu/drm/mediatek/mtk_ethdr.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/gpu/drm/mediatek/mtk_ethdr.c +2 -0
index bcced62e455d..d01f65819816 100644
@@ -6,6 +6,7 @@
 #include <drm/drm_blend.h>
 #include <drm/drm_fourcc.h>
 #include <drm/drm_framebuffer.h>
+#include <drm/drm_blend.h>
 #include <linux/clk.h>
 #include <linux/component.h>
 #include <linux/of.h>
@@ -36,6 +37,7 @@
 #define MIX_SRC_L0_EN				BIT(0)
 #define MIX_L_SRC_CON(n)		(0x28 + 0x18 * (n))
 #define NON_PREMULTI_SOURCE			(2 << 12)
+#define PREMULTI_SOURCE				(3 << 12)
 #define MIX_L_SRC_SIZE(n)		(0x30 + 0x18 * (n))
 #define MIX_L_SRC_OFFSET(n)		(0x34 + 0x18 * (n))
 #define MIX_FUNC_DCM0			0x120

drivers/gpu/drm/mediatek/mtk_ethdr.c void mtk_ethdr_layer_config(struct device *dev, unsigned int idx, · line 177

Quoted 6 lines
drivers/gpu/drm/mediatek/mtk_ethdr.c +1 -2
@@ -172,8 +174,12 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 		return;
 	}
 
-	if (state->base.fb && state->base.fb->format->has_alpha)
-		alpha_con = MIXER_ALPHA_AEN | MIXER_ALPHA;
+	alpha_con |= MIXER_ALPHA_AEN | (state->base.alpha & MIXER_ALPHA);

This is constant alpha, not related to pre-multiplied alpha, so separate it to another patch.

Regards,
CK

Quoted 18 lines
drivers/gpu/drm/mediatek/mtk_ethdr.c +6 -2
+
+	if (state->base.pixel_blend_mode != DRM_MODE_BLEND_COVERAGE)
+		alpha_con |= PREMULTI_SOURCE;
+	else
+		alpha_con |= NON_PREMULTI_SOURCE;
 
 	if (state->base.pixel_blend_mode == DRM_MODE_BLEND_PIXEL_NONE ||
 	    (state->base.fb && !state->base.fb->format->has_alpha)) {
@@ -191,8 +197,7 @@ void mtk_ethdr_layer_config(struct device *dev, unsigned int idx,
 	mtk_ddp_write(cmdq_pkt, pending->height << 16 | align_width, &mixer->cmdq_base,
 		      mixer->regs, MIX_L_SRC_SIZE(idx));
 	mtk_ddp_write(cmdq_pkt, offset, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_OFFSET(idx));
-	mtk_ddp_write_mask(cmdq_pkt, alpha_con, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_CON(idx),
-			   0x1ff);
+	mtk_ddp_write(cmdq_pkt, alpha_con, &mixer->cmdq_base, mixer->regs, MIX_L_SRC_CON(idx));
 	mtk_ddp_write_mask(cmdq_pkt, BIT(idx), &mixer->cmdq_base, mixer->regs, MIX_SRC_CON,
 			   BIT(idx));
 }
author
#21 lore raw

PATCH v8 13/16 drm/mediatek: Support alpha blending in display driver

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Support "Pre-multiplied" and "None" blend mode on MediaTek's chips by adding correct blend mode property when the planes init. Before this patch, only the "Coverage" mode (default) is supported.

For more information, there are three pixel blend modes in DRM driver:
"None", "Pre-multiplied", and "Coverage".

To understand the difference between these modes, let's take a look at the following two approaches to do alpha blending:

1. Straight:
dst.RGB = src.RGB * src.A + dst.RGB * (1 - src.A)
This is straightforward and easy to understand, when the source layer is compositing with the destination layer, it's alpha will affect the result. This is also known as "post-multiplied", or "Coverage" mode.

2. Pre-multiplied:
dst.RGB = src.RGB + dst.RGB * (1 - src.A)
Since the source RGB have already multiplied its alpha, only destination RGB need to multiply it. This is the "Pre-multiplied" mode in DRM.

For the "None" blend mode in DRM, it means the pixel alpha is ignored when compositing the layers, only the constant alpha for the composited layer will take effects.

---
 drivers/gpu/drm/mediatek/mtk_plane.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
drivers/gpu/drm/mediatek/mtk_plane.c +11 -0
index 1723d4333f37..5bf757a3ef20 100644
@@ -346,6 +346,17 @@ int mtk_plane_init(struct drm_device *dev, struct drm_plane *plane,
 			DRM_INFO("Create rotation property failed\n");
 	}
 
+	err = drm_plane_create_alpha_property(plane);
+	if (err)
+		DRM_ERROR("failed to create property: alpha\n");
+
+	err = drm_plane_create_blend_mode_property(plane,
+						   BIT(DRM_MODE_BLEND_PREMULTI) |
+						   BIT(DRM_MODE_BLEND_COVERAGE) |
+						   BIT(DRM_MODE_BLEND_PIXEL_NONE));
+	if (err)
+		DRM_ERROR("failed to create property: blend_mode\n");
+
 	drm_plane_helper_add(plane, &mtk_plane_helper_funcs);
 
 	return 0;
Signature · 1 line
2.18.0
author
#22 lore raw

PATCH v8 14/16 drm/mediatek: Support CRC in display driver

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Register CRC related function pointers to support
CRC retrieval.

---
 drivers/gpu/drm/mediatek/mtk_crtc.c     | 280 ++++++++++++++++++++++++
 drivers/gpu/drm/mediatek/mtk_crtc.h     |  38 ++++
 drivers/gpu/drm/mediatek/mtk_ddp_comp.h |   5 +
 3 files changed, 323 insertions(+)
drivers/gpu/drm/mediatek/mtk_crtc.c +280 -0
index 6f34f573e127..be7cf61b9f9b 100644
@@ -18,6 +18,7 @@
 #include <drm/drm_atomic_helper.h>
 #include <drm/drm_probe_helper.h>
 #include <drm/drm_vblank.h>
+#include <drm/drm_vblank_work.h>
 
 #include "mtk_crtc.h"
 #include "mtk_ddp_comp.h"
@@ -69,6 +70,9 @@ struct mtk_crtc {
 	/* lock for display hardware access */
 	struct mutex			hw_lock;
 	bool				config_updating;
+
+	struct mtk_ddp_comp		*crc_provider;
+	struct drm_vblank_work		crc_work;
 };
 
 struct mtk_crtc_state {
@@ -703,6 +707,88 @@ static void mtk_crtc_update_output(struct drm_crtc *crtc,
 	}
 }
 
+static void mtk_crtc_crc_work(struct kthread_work *base)
+{
+	struct drm_vblank_work *work = to_drm_vblank_work(base);
+	struct mtk_crtc *mtk_crtc =
+		container_of(work, typeof(*mtk_crtc), crc_work);
+	struct mtk_ddp_comp *comp = mtk_crtc->crc_provider;
+
+	if (!comp) {
+		DRM_WARN("%s(crtc-%d): no crc provider\n",
+			 __func__, drm_crtc_index(&mtk_crtc->base));
+		return;
+	}
+
+	if (mtk_crtc->base.crc.opened) {
+		u64 vblank = drm_crtc_vblank_count(&mtk_crtc->base);
+
+		comp->funcs->crc_read(comp->dev);
+
+		/* could take more than 50ms to finish */
+		drm_crtc_add_crc_entry(&mtk_crtc->base, true, vblank,
+				       comp->funcs->crc_entry(comp->dev));
+
+		drm_vblank_work_schedule(&mtk_crtc->crc_work, vblank + 1, true);
+	} else {
+		comp->funcs->crc_stop(comp->dev);
+	}
+}
+
+static int mtk_crtc_set_crc_source(struct drm_crtc *crtc, const char *src)
+{
+	struct mtk_crtc *mtk_crtc = to_mtk_crtc(crtc);
+	struct mtk_ddp_comp *comp = mtk_crtc->crc_provider;
+
+	if (!comp) {
+		DRM_ERROR("%s(crtc-%d): no crc provider\n",
+			  __func__, drm_crtc_index(crtc));
+		return -ENOENT;
+	}
+
+	if (!src)
+		return -EINVAL;
+
+	if (strcmp(src, "auto") != 0) {
+		DRM_ERROR("%s(crtc-%d): unknown source '%s'\n",
+			  __func__, drm_crtc_index(crtc), src);
+		return -EINVAL;
+	}
+
+	comp->funcs->crc_start(comp->dev);
+
+	/*
+	 * skip the first crc because the first frame (vblank + 1) is configured
+	 * by mtk_crtc_ddp_hw_init() when atomic enable
+	 */
+	drm_vblank_work_schedule(&mtk_crtc->crc_work,
+				 drm_crtc_vblank_count(crtc) + 2, false);
+	return 0;
+}
+
+static int mtk_crtc_verify_crc_source(struct drm_crtc *crtc, const char *src,
+				      size_t *cnt)
+{
+	struct mtk_crtc *mtk_crtc = to_mtk_crtc(crtc);
+	struct mtk_ddp_comp *comp = mtk_crtc->crc_provider;
+
+	if (!comp) {
+		DRM_ERROR("%s(crtc-%d): no crc provider\n",
+			  __func__, drm_crtc_index(crtc));
+		return -ENOENT;
+	}
+
+	if (src && strcmp(src, "auto") != 0) {
+		DRM_ERROR("%s(crtc-%d): unknown source '%s'\n",
+			  __func__, drm_crtc_index(crtc), src);
+		return -EINVAL;
+	}
+
+	*cnt = comp->funcs->crc_cnt(comp->dev);
+
+	return 0;
+}
+
 int mtk_crtc_plane_check(struct drm_crtc *crtc, struct drm_plane *plane,
 			 struct mtk_plane_state *state)
 {
@@ -751,6 +837,8 @@ static void mtk_crtc_atomic_enable(struct drm_crtc *crtc,
 
 	drm_crtc_vblank_on(crtc);
 	mtk_crtc->enabled = true;
+
+	drm_vblank_work_init(&mtk_crtc->crc_work, crtc, mtk_crtc_crc_work);
 }
 
 static void mtk_crtc_atomic_disable(struct drm_crtc *crtc,
@@ -840,6 +928,8 @@ static const struct drm_crtc_funcs mtk_crtc_funcs = {
 	.atomic_destroy_state	= mtk_crtc_destroy_state,
 	.enable_vblank		= mtk_crtc_enable_vblank,
 	.disable_vblank		= mtk_crtc_disable_vblank,
+	.set_crc_source		= mtk_crtc_set_crc_source,
+	.verify_crc_source	= mtk_crtc_verify_crc_source,
 };
 
 static const struct drm_crtc_helper_funcs mtk_crtc_helper_funcs = {
@@ -1033,6 +1123,13 @@ int mtk_crtc_create(struct drm_device *drm_dev, const unsigned int *path,
 
 			if (comp->funcs->ctm_set)
 				has_ctm = true;
+
+			if (comp->funcs->crc_cnt   &&
+			    comp->funcs->crc_entry &&
+			    comp->funcs->crc_read  &&
+			    comp->funcs->crc_start &&
+			    comp->funcs->crc_stop)
+				mtk_crtc->crc_provider = comp;
 		}
 
 		mtk_ddp_comp_register_vblank_cb(comp, mtk_crtc_ddp_irq,
@@ -1136,3 +1233,186 @@ int mtk_crtc_create(struct drm_device *drm_dev, const unsigned int *path,
 
 	return 0;
 }
+
+void mtk_crtc_init_crc(struct mtk_crtc_crc *crc, const u32 *crc_offset_table,
+		       size_t crc_count, u32 reset_offset, u32 reset_mask)
+{
+	crc->ofs = crc_offset_table;
+	crc->cnt = crc_count;
+	crc->rst_ofs = reset_offset;
+	crc->rst_msk = reset_mask;
+	crc->va = kcalloc(crc->cnt, sizeof(*crc->va), GFP_KERNEL);
+	if (!crc->va) {
+		DRM_ERROR("failed to allocate memory for crc\n");
+		crc->cnt = 0;
+	}
+}
+
+void mtk_crtc_read_crc(struct mtk_crtc_crc *crc, void __iomem *reg)
+{
+	if (!crc->cnt || !crc->ofs || !crc->va)
+		return;
+
+#if IS_REACHABLE(CONFIG_MTK_CMDQ)
+	/* sync to see the most up-to-date copy of the DMA buffer */
+	dma_sync_single_for_cpu(crc->cmdq_client.chan->mbox->dev,
+				crc->pa, crc->cnt * sizeof(*crc->va),
+				DMA_FROM_DEVICE);
+#endif
+}
+
+void mtk_crtc_destroy_crc(struct mtk_crtc_crc *crc)
+{
+	if (!crc->cnt)
+		return;
+
+#if IS_REACHABLE(CONFIG_MTK_CMDQ)
+	if (crc->pa) {
+		dma_unmap_single(crc->cmdq_client.chan->mbox->dev,
+				 crc->pa, crc->cnt * sizeof(*crc->va),
+				 DMA_TO_DEVICE);
+		crc->pa = 0;
+	}
+	if (crc->cmdq_client.chan) {
+		mtk_drm_cmdq_pkt_destroy(&crc->cmdq_handle);
+		mbox_free_channel(crc->cmdq_client.chan);
+		crc->cmdq_client.chan = NULL;
+	}
+#endif
+	kfree(crc->va);
+	crc->va = NULL;
+	crc->cnt = 0;
+}
+
+#if IS_REACHABLE(CONFIG_MTK_CMDQ)
+/**
+ * mtk_crtc_create_crc_cmdq - Create a CMDQ thread for syncing the CRCs
+ * @dev: Kernel device node of the CRC provider
+ * @crc: Pointer of the CRC to init
+ *
+ * This function will create a looping thread on GCE (Global Command Engine) to
+ * keep the CRC up to date by monitoring the assigned event (usually the frame
+ * done event) of the CRC provider, and read the CRCs from the registers to a
+ * shared memory for the workqueue to read. To start/stop the looping thread,
+ * please call `mtk_crtc_start_crc_cmdq()` and `mtk_crtc_stop_crc_cmdq()`
+ * defined blow.
+ *
+ * The reason why we don't update the CRCs with CPU is that the front porch of
+ * 4K60 timing in CEA-861 is less than 60us, and register read/write speed is
+ * relatively unreliable comparing to GCE due to the bus design.
+ *
+ * We must create a new thread instead of using the original one for plane
+ * update is because:
+ * 1. We cannot add another wait-for-event command at the end of cmdq packet, or
+ *    the cmdq callback will delay for too long
+ * 2. Will get the CRC of the previous frame if using the existed wait-for-event
+ *    command which is at the beginning of the packet
+ */
+void mtk_crtc_create_crc_cmdq(struct device *dev, struct mtk_crtc_crc *crc)
+{
+	int i;
+
+	if (!crc->cnt) {
+		dev_warn(dev, "%s: not support\n", __func__);
+		goto cleanup;
+	}
+
+	if (!crc->ofs) {
+		dev_warn(dev, "%s: not defined\n", __func__);
+		goto cleanup;
+	}
+
+	crc->cmdq_client.client.dev = dev;
+	crc->cmdq_client.client.tx_block = false;
+	crc->cmdq_client.client.knows_txdone = true;
+	crc->cmdq_client.client.rx_callback = NULL;
+	crc->cmdq_client.chan = mbox_request_channel(&crc->cmdq_client.client, 0);
+	if (IS_ERR(crc->cmdq_client.chan)) {
+		dev_warn(dev, "%s: failed to create mailbox client\n", __func__);
+		crc->cmdq_client.chan = NULL;
+		goto cleanup;
+	}
+
+	if (mtk_drm_cmdq_pkt_create(&crc->cmdq_client, &crc->cmdq_handle, PAGE_SIZE)) {
+		dev_warn(dev, "%s: failed to create cmdq packet\n", __func__);
+		goto cleanup;
+	}
+
+	if (!crc->va) {
+		dev_warn(dev, "%s: no memory\n", __func__);
+		goto cleanup;
+	}
+
+	/* map the entry to get a dma address for cmdq to store the crc */
+	crc->pa = dma_map_single(crc->cmdq_client.chan->mbox->dev,
+				 crc->va, crc->cnt * sizeof(*crc->va),
+				 DMA_FROM_DEVICE);
+
+	if (dma_mapping_error(crc->cmdq_client.chan->mbox->dev, crc->pa)) {
+		dev_err(dev, "%s: failed to map dma\n", __func__);
+		goto cleanup;
+	}
+
+	if (crc->cmdq_event)
+		cmdq_pkt_wfe(&crc->cmdq_handle, crc->cmdq_event, true);
+
+	for (i = 0; i < crc->cnt; i++) {
+		/* put crc to spr1 register */
+		cmdq_pkt_read_s(&crc->cmdq_handle, crc->cmdq_reg->subsys,
+				crc->cmdq_reg->offset + crc->ofs[i],
+				CMDQ_THR_SPR_IDX1);
+
+		/* copy spr1 register to physical address of the crc */
+		cmdq_pkt_assign(&crc->cmdq_handle, CMDQ_THR_SPR_IDX0,
+				CMDQ_ADDR_HIGH(crc->pa + i * sizeof(*crc->va)));
+		cmdq_pkt_write_s(&crc->cmdq_handle, CMDQ_THR_SPR_IDX0,
+				 CMDQ_ADDR_LOW(crc->pa + i * sizeof(*crc->va)),
+				 CMDQ_THR_SPR_IDX1);
+	}
+	/* reset crc */
+	mtk_ddp_write_mask(&crc->cmdq_handle, ~0, crc->cmdq_reg, 0,
+			   crc->rst_ofs, crc->rst_msk);
+
+	/* clear reset bit */
+	mtk_ddp_write_mask(&crc->cmdq_handle, 0, crc->cmdq_reg, 0,
+			   crc->rst_ofs, crc->rst_msk);
+
+	/* jump to head of the cmdq packet */
+	cmdq_pkt_jump_abs(&crc->cmdq_handle, crc->cmdq_handle.pa_base,
+			  cmdq_get_shift_pa(crc->cmdq_client.chan));
+
+	return;
+cleanup:
+	mtk_crtc_destroy_crc(crc);
+}
+
+/**
+ * mtk_crtc_start_crc_cmdq - Start the GCE looping thread for CRC update
+ * @crc: Pointer of the CRC information
+ */
+void mtk_crtc_start_crc_cmdq(struct mtk_crtc_crc *crc)
+{
+	if (!crc->cmdq_client.chan)
+		return;
+
+	dma_sync_single_for_device(crc->cmdq_client.chan->mbox->dev,
+				   crc->cmdq_handle.pa_base,
+				   crc->cmdq_handle.cmd_buf_size,
+				   DMA_TO_DEVICE);
+	mbox_send_message(crc->cmdq_client.chan, &crc->cmdq_handle);
+	mbox_client_txdone(crc->cmdq_client.chan, 0);
+}
+
+/**
+ * mtk_crtc_stop_crc_cmdq - Stop the GCE looping thread for CRC update
+ * @crc: Pointer of the CRC information
+ */
+void mtk_crtc_stop_crc_cmdq(struct mtk_crtc_crc *crc)
+{
+	if (!crc->cmdq_client.chan)
+		return;
+
+	/* remove all the commands from the cmdq packet */
+	mbox_flush(crc->cmdq_client.chan, 2000);
+}
+#endif
drivers/gpu/drm/mediatek/mtk_crtc.h +38 -0
index 388e900b6f4d..a79c4611754e 100644
@@ -14,6 +14,34 @@
 #define MTK_MAX_BPC	10
 #define MTK_MIN_BPC	3
 
+/**
+ * struct mtk_crtc_crc - crc related information
+ * @ofs: register offset of crc
+ * @rst_ofs: register offset of crc reset
+ * @rst_msk: register mask of crc reset
+ * @cnt: count of crc
+ * @va: pointer to the start of crc array
+ * @pa: physical address of the crc for gce to access
+ * @cmdq_event: the event to trigger the cmdq
+ * @cmdq_reg: address of the register that cmdq is going to access
+ * @cmdq_client: handler to control cmdq (mbox channel, thread ...etc.)
+ * @cmdq_handle: cmdq packet to store the commands
+ */
+struct mtk_crtc_crc {
+	const u32 *ofs;
+	u32 rst_ofs;
+	u32 rst_msk;
+	size_t cnt;
+	u32 *va;
+#if IS_REACHABLE(CONFIG_MTK_CMDQ)
+	dma_addr_t pa;
+	u32 cmdq_event;
+	struct cmdq_client_reg *cmdq_reg;
+	struct cmdq_client cmdq_client;
+	struct cmdq_pkt cmdq_handle;
+#endif
+};
+
 void mtk_crtc_commit(struct drm_crtc *crtc);
 int mtk_crtc_create(struct drm_device *drm_dev, const unsigned int *path,
 		    unsigned int path_len, int priv_data_index,
@@ -25,4 +53,14 @@ void mtk_crtc_async_update(struct drm_crtc *crtc, struct drm_plane *plane,
 			   struct drm_atomic_state *plane_state);
 struct device *mtk_crtc_dma_dev_get(struct drm_crtc *crtc);
 
+void mtk_crtc_init_crc(struct mtk_crtc_crc *crc, const u32 *crc_offset_table,
+		       size_t crc_count, u32 reset_offset, u32 reset_mask);
+void mtk_crtc_read_crc(struct mtk_crtc_crc *crc, void __iomem *reg);
+void mtk_crtc_destroy_crc(struct mtk_crtc_crc *crc);
+#if IS_REACHABLE(CONFIG_MTK_CMDQ)
+void mtk_crtc_create_crc_cmdq(struct device *dev, struct mtk_crtc_crc *crc);
+void mtk_crtc_start_crc_cmdq(struct mtk_crtc_crc *crc);
+void mtk_crtc_stop_crc_cmdq(struct mtk_crtc_crc *crc);
+#endif
+
 #endif /* MTK_CRTC_H */
drivers/gpu/drm/mediatek/mtk_ddp_comp.h +5 -0
index f7fe2e08dc8e..b220a672d182 100644
@@ -88,6 +88,11 @@ struct mtk_ddp_comp_funcs {
 	void (*remove)(struct device *dev, struct mtk_mutex *mutex);
 	unsigned int (*encoder_index)(struct device *dev);
 	enum drm_mode_status (*mode_valid)(struct device *dev, const struct drm_display_mode *mode);
+	size_t (*crc_cnt)(struct device *dev);
+	u32 *(*crc_entry)(struct device *dev);
+	void (*crc_read)(struct device *dev);
+	void (*crc_start)(struct device *dev);
+	void (*crc_stop)(struct device *dev);
 };
 
 struct mtk_ddp_comp {

--
2.18.0

#23 lore raw

Re: [PATCH v8 14/16] drm/mediatek: Support CRC in display driver

Hi, Shawn:

Quoted 7 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Register CRC related function pointers to support
CRC retrieval.

---

[snip]

Quoted 12 lines
drivers/gpu/drm/mediatek/mtk_crtc.c +11 -0
+static void mtk_crtc_crc_work(struct kthread_work *base)
+{
+	struct drm_vblank_work *work = to_drm_vblank_work(base);
+	struct mtk_crtc *mtk_crtc =
+		container_of(work, typeof(*mtk_crtc), crc_work);
+	struct mtk_ddp_comp *comp = mtk_crtc->crc_provider;
+
+	if (!comp) {
+		DRM_WARN("%s(crtc-%d): no crc provider\n",
+			 __func__, drm_crtc_index(&mtk_crtc->base));
+		return;

drivers/gpu/drm/mediatek/mtk_crtc.c static void mtk_crtc_update_output(struct drm_crtc *crtc, · line 730

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_crtc.c +10 -0
+	}
+
+	if (mtk_crtc->base.crc.opened) {
+		u64 vblank = drm_crtc_vblank_count(&mtk_crtc->base);
+
+		comp->funcs->crc_read(comp->dev);
+
+		/* could take more than 50ms to finish */
+		drm_crtc_add_crc_entry(&mtk_crtc->base, true, vblank,
+				       comp->funcs->crc_entry(comp->dev));

It seems that you could regenerate cmdq packet for crc here. So crtc atomic flush and crc could use the same mailbox channel.

Regards,
CK

Quoted 7 lines
drivers/gpu/drm/mediatek/mtk_crtc.c +7 -0
+
+		drm_vblank_work_schedule(&mtk_crtc->crc_work, vblank + 1, true);
+	} else {
+		comp->funcs->crc_stop(comp->dev);
+	}
+}
+
#24 lore raw

Re: [PATCH v8 14/16] drm/mediatek: Support CRC in display driver

Hi CK,

Quoted 35 lines · CK Hu (胡俊光)

Hi, Shawn:

Quoted 7 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

Register CRC related function pointers to support
CRC retrieval.

---

[snip]

Quoted 22 lines
drivers/gpu/drm/mediatek/mtk_crtc.c +21 -0
+static void mtk_crtc_crc_work(struct kthread_work *base)
+{
+	struct drm_vblank_work *work = to_drm_vblank_work(base);
+	struct mtk_crtc *mtk_crtc =
+		container_of(work, typeof(*mtk_crtc), crc_work);
+	struct mtk_ddp_comp *comp = mtk_crtc->crc_provider;
+
+	if (!comp) {
+		DRM_WARN("%s(crtc-%d): no crc provider\n",
+			 __func__, drm_crtc_index(&mtk_crtc->base));
+		return;
+	}
+
+	if (mtk_crtc->base.crc.opened) {
+		u64 vblank = drm_crtc_vblank_count(&mtk_crtc->base);
+
+		comp->funcs->crc_read(comp->dev);
+
+		/* could take more than 50ms to finish */
+		drm_crtc_add_crc_entry(&mtk_crtc->base, true, vblank,
+				       comp->funcs->crc_entry(comp-
Quoted 8 lines
Quoted 1 line
Quoted 1 line

dev));

It seems that you could regenerate cmdq packet for crc here. So crtc atomic flush and crc could use the same mailbox channel.

Regards,
CK

It seems that it's possible to call mtk_ddp_update_config() here and add a new pending_read_crc flag to insert the read crc related instructions into the same cmdq_pkt and use the same mailbox channel as atomic flush.

I'll try this, but I'll need some time to modify this and test it to make sure no other issues arise.

Regards,
Jason-Jh Lin

Quoted 8 lines
Quoted 8 lines
(quoted diff) +2 -0
+
+		drm_vblank_work_schedule(&mtk_crtc->crc_work, vblank +

1, true);

drivers/gpu/drm/mediatek/mtk_crtc.c +5 -0
+	} else {
+		comp->funcs->crc_stop(comp->dev);
+	}
+}
+
#25 lore raw

Re: [PATCH v8 14/16] drm/mediatek: Support CRC in display driver

In reply to Shawn Sung
kernel test robot noticed the following build warnings:

Hi Shawn,

kernel test robot noticed the following build warnings:

[auto build test WARNING on v6.10-rc2]
[also build test WARNING on linus/master next-20240613] [cannot apply to chrome-os/chromeos-6.1]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Shawn-Sung/soc-mediatek-Disable-9-bit-alpha-in-ETHDR/20240606-172904
base:   v6.10-rc2
patch link:    https://lore.kernel.org/r/20240606092635.27981-15-shawn.sung%40mediatek.com
patch subject: [PATCH v8 14/16] drm/mediatek: Support CRC in display driver
config: arm64-randconfig-r113-20240615 (https://download.01.org/0day-ci/archive/20240616/[email protected]/config)
compiler: aarch64-linux-gcc (GCC) 13.2.0
reproduce: (https://download.01.org/0day-ci/archive/20240616/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

sparse warnings: (new ones prefixed by >>)

>> drivers/gpu/drm/mediatek/mtk_crtc.c:1373:66: sparse: sparse: Using plain integer as NULL pointer

drivers/gpu/drm/mediatek/mtk_crtc.c:1377:65: sparse: sparse: Using plain integer as NULL pointer

vim +1373 drivers/gpu/drm/mediatek/mtk_crtc.c

1286
1287 #if IS_REACHABLE(CONFIG_MTK_CMDQ)
1288 /**
1289 * mtk_crtc_create_crc_cmdq - Create a CMDQ thread for syncing the CRCs
1290 * @dev: Kernel device node of the CRC provider
1291 * @crc: Pointer of the CRC to init
1292 *
1293 * This function will create a looping thread on GCE (Global Command Engine) to
1294 * keep the CRC up to date by monitoring the assigned event (usually the frame
1295 * done event) of the CRC provider, and read the CRCs from the registers to a
1296 * shared memory for the workqueue to read. To start/stop the looping thread,
1297 * please call mtk_crtc_start_crc_cmdq() and mtk_crtc_stop_crc_cmdq()
1298 * defined blow.
1299 *
1300 * The reason why we don't update the CRCs with CPU is that the front porch of
1301 * 4K60 timing in CEA-861 is less than 60us, and register read/write speed is
1302 * relatively unreliable comparing to GCE due to the bus design.
1303 *
1304 * We must create a new thread instead of using the original one for plane
1305 * update is because:
1306 * 1. We cannot add another wait-for-event command at the end of cmdq packet, or
1307 * the cmdq callback will delay for too long
1308 * 2. Will get the CRC of the previous frame if using the existed wait-for-event
1309 * command which is at the beginning of the packet
1310 */
1311 void mtk_crtc_create_crc_cmdq(struct device *dev, struct mtk_crtc_crc *crc)
1312 {
1313 int i;
1314
1315 if (!crc->cnt) {
1316 dev_warn(dev, "%s: not support\n", __func__);
1317 goto cleanup;
1318 }
1319
1320 if (!crc->ofs) {
1321 dev_warn(dev, "%s: not defined\n", __func__);
1322 goto cleanup;
1323 }
1324
1325 crc->cmdq_client.client.dev = dev;
1326 crc->cmdq_client.client.tx_block = false;
1327 crc->cmdq_client.client.knows_txdone = true;
1328 crc->cmdq_client.client.rx_callback = NULL;
1329 crc->cmdq_client.chan = mbox_request_channel(&crc->cmdq_client.client, 0);
1330 if (IS_ERR(crc->cmdq_client.chan)) {
1331 dev_warn(dev, "%s: failed to create mailbox client\n", __func__);
1332 crc->cmdq_client.chan = NULL;
1333 goto cleanup;
1334 }
1335
1336 if (mtk_drm_cmdq_pkt_create(&crc->cmdq_client, &crc->cmdq_handle, PAGE_SIZE)) {
1337 dev_warn(dev, "%s: failed to create cmdq packet\n", __func__);
1338 goto cleanup;
1339 }
1340
1341 if (!crc->va) {
1342 dev_warn(dev, "%s: no memory\n", __func__);
1343 goto cleanup;
1344 }
1345
1346 /* map the entry to get a dma address for cmdq to store the crc */
1347 crc->pa = dma_map_single(crc->cmdq_client.chan->mbox->dev,
1348 crc->va, crc->cnt * sizeof(*crc->va),
1349 DMA_FROM_DEVICE);
1350
1351 if (dma_mapping_error(crc->cmdq_client.chan->mbox->dev, crc->pa)) {
1352 dev_err(dev, "%s: failed to map dma\n", __func__);
1353 goto cleanup;
1354 }
1355
1356 if (crc->cmdq_event)
1357 cmdq_pkt_wfe(&crc->cmdq_handle, crc->cmdq_event, true);
1358
1359 for (i = 0; i < crc->cnt; i++) {
1360 /* put crc to spr1 register */
1361 cmdq_pkt_read_s(&crc->cmdq_handle, crc->cmdq_reg->subsys,
1362 crc->cmdq_reg->offset + crc->ofs[i],
1363 CMDQ_THR_SPR_IDX1);
1364
1365 /* copy spr1 register to physical address of the crc */
1366 cmdq_pkt_assign(&crc->cmdq_handle, CMDQ_THR_SPR_IDX0,
1367 CMDQ_ADDR_HIGH(crc->pa + i * sizeof(*crc->va)));
1368 cmdq_pkt_write_s(&crc->cmdq_handle, CMDQ_THR_SPR_IDX0,
1369 CMDQ_ADDR_LOW(crc->pa + i * sizeof(*crc->va)),
1370 CMDQ_THR_SPR_IDX1);
1371 }
1372 /* reset crc */

> 1373		mtk_ddp_write_mask(&crc->cmdq_handle, ~0, crc->cmdq_reg, 0,

1374 crc->rst_ofs, crc->rst_msk);
1375
1376 /* clear reset bit */
1377 mtk_ddp_write_mask(&crc->cmdq_handle, 0, crc->cmdq_reg, 0,
1378 crc->rst_ofs, crc->rst_msk);
1379
1380 /* jump to head of the cmdq packet */
1381 cmdq_pkt_jump_abs(&crc->cmdq_handle, crc->cmdq_handle.pa_base,
1382 cmdq_get_shift_pa(crc->cmdq_client.chan));
1383
1384 return;
1385 cleanup:
1386 mtk_crtc_destroy_crc(crc);
1387 }
1388

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
author
#26 lore raw

PATCH v8 15/16 drm/mediatek: Support CRC in OVL

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

We choose OVL as the CRC generator from other hardware components that are also capable of calculating CRCs, since its frame done event triggers vblanks, it can be used as a signal to know when is safe to retrieve CRC of the frame.

Please note that position of the hardware component that is chosen as CRC generator in the display path is significant. For example, while OVL is the first module in VDOSYS0, its CRC won't be affected by the modules after it, which means effects applied by PQ, Gamma, Dither or any other components after OVL won't be calculated in CRC generation.

---
 drivers/gpu/drm/mediatek/mtk_ddp_comp.c |   5 +
 drivers/gpu/drm/mediatek/mtk_disp_drv.h |   5 +
 drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 209 ++++++++++++++++++++++--
 3 files changed, 209 insertions(+), 10 deletions(-)
drivers/gpu/drm/mediatek/mtk_ddp_comp.c +5 -0
index 17b036411292..07c3b5c5e14e 100644
@@ -351,6 +351,11 @@ static const struct mtk_ddp_comp_funcs ddp_ovl = {
 	.clk_enable = mtk_ovl_clk_enable,
 	.clk_disable = mtk_ovl_clk_disable,
 	.config = mtk_ovl_config,
+	.crc_cnt = mtk_ovl_crc_cnt,
+	.crc_entry = mtk_ovl_crc_entry,
+	.crc_read = mtk_ovl_crc_read,
+	.crc_start = mtk_ovl_crc_start,
+	.crc_stop = mtk_ovl_crc_stop,
 	.start = mtk_ovl_start,
 	.stop = mtk_ovl_stop,
 	.register_vblank_cb = mtk_ovl_register_vblank_cb,
drivers/gpu/drm/mediatek/mtk_disp_drv.h +5 -0
index 082ac18fe04a..a03d7a10847a 100644
@@ -105,6 +105,11 @@ void mtk_ovl_enable_vblank(struct device *dev);
 void mtk_ovl_disable_vblank(struct device *dev);
 const u32 *mtk_ovl_get_formats(struct device *dev);
 size_t mtk_ovl_get_num_formats(struct device *dev);
+size_t mtk_ovl_crc_cnt(struct device *dev);
+u32 *mtk_ovl_crc_entry(struct device *dev);
+void mtk_ovl_crc_read(struct device *dev);
+void mtk_ovl_crc_start(struct device *dev);
+void mtk_ovl_crc_stop(struct device *dev);
 
 void mtk_ovl_adaptor_add_comp(struct device *dev, struct mtk_mutex *mutex);
 void mtk_ovl_adaptor_remove_comp(struct device *dev, struct mtk_mutex *mutex);
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +199 -10

304 lines not on this page · raw message

Signature · 1 line
2.18.0
#27 lore raw

Re: [PATCH v8 15/16] drm/mediatek: Support CRC in OVL

Hi, Shawn:

Quoted 18 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

We choose OVL as the CRC generator from other hardware components that are also capable of calculating CRCs, since its frame done event triggers vblanks, it can be used as a signal to know when is safe to retrieve CRC of the frame.

Please note that position of the hardware component that is chosen as CRC generator in the display path is significant. For example, while OVL is the first module in VDOSYS0, its CRC won't be affected by the modules after it, which means effects applied by PQ, Gamma, Dither or any other components after OVL won't be calculated in CRC generation.

---

[snip]

Quoted 17 lines

void mtk_ovl_stop(struct device *dev)

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +11 -0
@@ -489,6 +579,83 @@ void mtk_ovl_layer_config(struct device *dev, unsigned int idx,
 	    (state->base.fb && !state->base.fb->format->has_alpha))
 		ignore_pixel_alpha = OVL_CONST_BLEND;
 
+	/*
+	 * OVL only supports 8 bits data in CRC calculation, transform 10-bit
+	 * RGB to 8-bit RGB by leveraging the ability of the Y2R (YUV-to-RGB)
+	 * hardware to multiply coefficients, although there is nothing to do
+	 * with the YUV format.
+	 */
+	if (ovl->data->supports_clrfmt_ext) {
+		u32 y2r_coef = 0, y2r_offset = 0, r2r_coef = 0, csc_en = 0;
+
+		if (is_10bit_rgb(fmt)) {
+			con |= OVL_CON_MTX_AUTO_DIS | OVL_CON_MTX_EN | OVL_CON_MTX_PROGRAMMABLE;

drivers/gpu/drm/mediatek/mtk_disp_ovl.c void mtk_ovl_layer_config(struct device *dev, unsigned int idx, · line 602

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +10 -0
+
+			/*
+			 * Y2R coefficient setting
+			 * bit 13 is 2^1, bit 12 is 2^0, bit 11 is 2^-1,
+			 * bit 10 is 2^-2 = 0.25
+			 */
+			y2r_coef = BIT(10);
+
+			/* -1 in 10bit */
+			y2r_offset = GENMASK(10, 0) - 1;

I don't know why do this? If an input value is 0x100, then

0x100 right shit 2 bit become 0x40.
0x40 - 1 = 0x3f.
0x3f left shift 2 bit become 0xfc.

So input 0x100 and output 0xfc. Why?

Quoted 8 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +8 -0
+
+			/*
+			 * R2R coefficient setting
+			 * bit 19 is 2^1, bit 18 is 2^0, bit 17 is 2^-1,
+			 * bit 20 is 2^2 = 4
+			 */
+			r2r_coef = BIT(20);
+

drivers/gpu/drm/mediatek/mtk_disp_ovl.c void mtk_ovl_layer_config(struct device *dev, unsigned int idx, · line 620

Quoted 10 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +10 -0
+			/* CSC_EN is for R2R */
+			csc_en = OVL_CLRFMT_EXT1_CSC_EN(idx);
+
+			/*
+			 * 1. YUV input data - 1 and shift right for 2 bits to remove it
+			 * [R']   [0.25    0    0]   [Y in - 1]
+			 * [G'] = [   0 0.25    0] * [U in - 1]
+			 * [B']   [   0    0 0.25]   [V in - 1]
+			 *
+			 * 2. shift left for 2 bit letting the last 2 bits become 0

You truncate the last two bit, so some quality lost. I think the quality is main function and CRC is just for debug. So it's better that in normal case we keep quality and only for debug to lost the quality.

I have another question. You just truncate the last two bit but it is still 10 bit value, so CRC could calculate this 10 bit value? I don't know why you say CRC just for 8 bit?

Regards,
CK

Quoted 42 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +38 -0
+			 * [R out]   [ 4  0  0]   [R']
+			 * [G out] = [ 0  4  0] * [G']
+			 * [B out]   [ 0  0  4]   [B']
+			 */
+		}
+
+		mtk_ddp_write_mask(cmdq_pkt, y2r_coef,
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_Y2R_PARA_R0(idx),
+				   OVL_Y2R_PARA_C_CF_RMY);
+		mtk_ddp_write_mask(cmdq_pkt, (y2r_coef << 16),
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_Y2R_PARA_G0(idx),
+				   OVL_Y2R_PARA_C_CF_GMU);
+		mtk_ddp_write_mask(cmdq_pkt, y2r_coef,
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_Y2R_PARA_B1(idx),
+				   OVL_Y2R_PARA_C_CF_BMV);
+
+		mtk_ddp_write_mask(cmdq_pkt, y2r_offset,
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_Y2R_PARA_YUV_A_0(idx),
+				   OVL_Y2R_PARA_C_CF_YA);
+		mtk_ddp_write_mask(cmdq_pkt, (y2r_offset << 16),
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_Y2R_PARA_YUV_A_0(idx),
+				   OVL_Y2R_PARA_C_CF_UA);
+		mtk_ddp_write_mask(cmdq_pkt, y2r_offset,
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_Y2R_PARA_YUV_A_1(idx),
+				   OVL_Y2R_PARA_C_CF_VA);
+
+		mtk_ddp_write_relaxed(cmdq_pkt, r2r_coef,
+				      &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_R2R_R0(idx));
+		mtk_ddp_write_relaxed(cmdq_pkt, r2r_coef,
+				      &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_R2R_G1(idx));
+		mtk_ddp_write_relaxed(cmdq_pkt, r2r_coef,
+				      &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_R2R_B2(idx));
+
+		mtk_ddp_write_mask(cmdq_pkt, csc_en,
+				   &ovl->cmdq_reg, ovl->regs, DISP_REG_OVL_CLRFMT_EXT1,
+				   OVL_CLRFMT_EXT1_CSC_EN(idx));
+	}
+
 	if (pending->rotation & DRM_MODE_REFLECT_Y) {
 		con |= OVL_CON_VIRT_FLIP;
 		addr += (pending->height - 1) * pending->pitch;
#28 lore raw

Re: [PATCH v8 15/16] drm/mediatek: Support CRC in OVL

Hi CK,

Quoted 61 lines · CK Hu (胡俊光)

Hi, Shawn:

Quoted 18 lines · Shawn Sung

From: Hsiao Chien Sung <[email protected]>

We choose OVL as the CRC generator from other hardware components that are also capable of calculating CRCs, since its frame done event triggers vblanks, it can be used as a signal to know when is safe to retrieve CRC of the frame.

Please note that position of the hardware component that is chosen as CRC generator in the display path is significant. For example, while OVL is the first module in VDOSYS0, its CRC won't be affected by the modules after it, which means effects applied by PQ, Gamma, Dither or any other components after OVL won't be calculated in CRC generation.

---

[snip]

Quoted 28 lines

void mtk_ovl_stop(struct device *dev)

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +2 -0
@@ -489,6 +579,83 @@ void mtk_ovl_layer_config(struct device *dev,
 unsigned int idx,
 	    (state->base.fb && !state->base.fb->format->has_alpha))
 		ignore_pixel_alpha = OVL_CONST_BLEND;
 
+	/*
+	 * OVL only supports 8 bits data in CRC calculation, transform

10-bit

(quoted diff) +1 -0
+	 * RGB to 8-bit RGB by leveraging the ability of the Y2R (YUV-

to-RGB)

(quoted diff) +1 -0
+	 * hardware to multiply coefficients, although there is nothing

to do

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
+	 * with the YUV format.
+	 */
+	if (ovl->data->supports_clrfmt_ext) {
+		u32 y2r_coef = 0, y2r_offset = 0, r2r_coef = 0, csc_en

= 0;

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+
+		if (is_10bit_rgb(fmt)) {
+			con |= OVL_CON_MTX_AUTO_DIS | OVL_CON_MTX_EN |

OVL_CON_MTX_PROGRAMMABLE;

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
+
+			/*
+			 * Y2R coefficient setting
+			 * bit 13 is 2^1, bit 12 is 2^0, bit 11 is 2^-

1,

drivers/gpu/drm/mediatek/mtk_disp_ovl.c void mtk_ovl_layer_config(struct device *dev, unsigned int idx, · line 602

Quoted 6 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +6 -0
+			 * bit 10 is 2^-2 = 0.25
+			 */
+			y2r_coef = BIT(10);
+
+			/* -1 in 10bit */
+			y2r_offset = GENMASK(10, 0) - 1;

I don't know why do this? If an input value is 0x100, then

Quoted 6 lines

0x100 right shit 2 bit become 0x40.
0x40 - 1 = 0x3f.
0x3f left shift 2 bit become 0xfc.

So input 0x100 and output 0xfc. Why?

This should be -2, not -1.
It is to rid of the last 2 bit.

OVL HW does not use >> 2 and then << 2, but uses *0.25 and then rounds to a positive integer and then *4, so it is necessary to remove the last 2 bits by -2 before doing the operation.

If the formula is: round(input data *0.25, 0) *4 The input data is 3, and it will become 4 after calculation.

If the formula is: round((input data - 2) *0.25, 0) *4 The input data is 3, and it will become 0 after calculation.

Quoted 19 lines
Quoted 19 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
+
+			/*
+			 * R2R coefficient setting
+			 * bit 19 is 2^1, bit 18 is 2^0, bit 17 is 2^-

1,

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +9 -0
+			 * bit 20 is 2^2 = 4
+			 */
+			r2r_coef = BIT(20);
+
+			/* CSC_EN is for R2R */
+			csc_en = OVL_CLRFMT_EXT1_CSC_EN(idx);
+
+			/*
+			 * 1. YUV input data - 1 and shift right for 2

bits to remove it

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
+			 * [R']   [0.25    0    0]   [Y in - 1]
+			 * [G'] = [   0 0.25    0] * [U in - 1]
+			 * [B']   [   0    0 0.25]   [V in - 1]
+			 *
Quoted 8 lines
Quoted 2 lines
(quoted diff) +1 -0
+			 * 2. shift left for 2 bit letting the last 2

bits become 0

You truncate the last two bit, so some quality lost. I think the quality is main function and CRC is just for debug. So it's better that
in normal case we keep quality and only for debug to lost the quality.

I agree with this, I'll add crc.opend flag to bypass these settings.

Quoted 5 lines

I have another question. You just truncate the last two bit but it is still 10 bit value, so CRC could calculate this 10 bit value? I don't know why you say CRC just for 8 bit?

OVL is 10-bit per RGB channel, so the 8-bit value will be padding with 2-bits 0 in the end, so the CRC truncating the last 2-bit in 10-bit value would be the same as original case.

Quoted 55 lines

Regards,
CK

Quoted 52 lines
drivers/gpu/drm/mediatek/mtk_disp_ovl.c +8 -0
+			 * [R out]   [ 4  0  0]   [R']
+			 * [G out] = [ 0  4  0] * [G']
+			 * [B out]   [ 0  0  4]   [B']
+			 */
+		}
+
+		mtk_ddp_write_mask(cmdq_pkt, y2r_coef,
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_Y2R_PARA_R0(idx),

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+				   OVL_Y2R_PARA_C_CF_RMY);
+		mtk_ddp_write_mask(cmdq_pkt, (y2r_coef << 16),
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_Y2R_PARA_G0(idx),

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+				   OVL_Y2R_PARA_C_CF_GMU);
+		mtk_ddp_write_mask(cmdq_pkt, y2r_coef,
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_Y2R_PARA_B1(idx),

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
+				   OVL_Y2R_PARA_C_CF_BMV);
+
+		mtk_ddp_write_mask(cmdq_pkt, y2r_offset,
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_Y2R_PARA_YUV_A_0(idx),

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+				   OVL_Y2R_PARA_C_CF_YA);
+		mtk_ddp_write_mask(cmdq_pkt, (y2r_offset << 16),
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_Y2R_PARA_YUV_A_0(idx),

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+				   OVL_Y2R_PARA_C_CF_UA);
+		mtk_ddp_write_mask(cmdq_pkt, y2r_offset,
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_Y2R_PARA_YUV_A_1(idx),

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +4 -0
+				   OVL_Y2R_PARA_C_CF_VA);
+
+		mtk_ddp_write_relaxed(cmdq_pkt, r2r_coef,
+				      &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_R2R_R0(idx));

(quoted diff) +2 -0
+		mtk_ddp_write_relaxed(cmdq_pkt, r2r_coef,
+				      &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_R2R_G1(idx));

(quoted diff) +2 -0
+		mtk_ddp_write_relaxed(cmdq_pkt, r2r_coef,
+				      &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_R2R_B2(idx));

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+
+		mtk_ddp_write_mask(cmdq_pkt, csc_en,
+				   &ovl->cmdq_reg, ovl->regs,

DISP_REG_OVL_CLRFMT_EXT1,

drivers/gpu/drm/mediatek/mtk_disp_ovl.c +3 -0
+				   OVL_CLRFMT_EXT1_CSC_EN(idx));
+	}
+
 	if (pending->rotation & DRM_MODE_REFLECT_Y) {
 		con |= OVL_CON_VIRT_FLIP;
 		addr += (pending->height - 1) * pending->pitch;
author
#29 lore raw

PATCH v8 16/16 drm/mediatek: Support CRC in OVL adaptor

In reply to Shawn Sung

From: Hsiao Chien Sung <[email protected]>

We choose Mixer as CRC generator in OVL adaptor since its frame done event will trigger vblanks, we can know when is safe to retrieve CRC of the frame.

In OVL adaptor, there's no image procession after Mixer, unlike the OVL in VDOSYS0, Mixer's CRC will include all the effects that are applied to the frame.

---
 drivers/gpu/drm/mediatek/mtk_ddp_comp.c       |  5 ++
 drivers/gpu/drm/mediatek/mtk_disp_drv.h       |  5 ++
 .../gpu/drm/mediatek/mtk_disp_ovl_adaptor.c   | 35 +++++++++
 drivers/gpu/drm/mediatek/mtk_ethdr.c          | 72 +++++++++++++++++++
 drivers/gpu/drm/mediatek/mtk_ethdr.h          |  7 ++
 5 files changed, 124 insertions(+)
drivers/gpu/drm/mediatek/mtk_ddp_comp.c +5 -0
index 07c3b5c5e14e..473460ab3a7a 100644
@@ -408,6 +408,11 @@ static const struct mtk_ddp_comp_funcs ddp_ovl_adaptor = {
 	.clk_enable = mtk_ovl_adaptor_clk_enable,
 	.clk_disable = mtk_ovl_adaptor_clk_disable,
 	.config = mtk_ovl_adaptor_config,
+	.crc_cnt = mtk_ovl_adaptor_crc_cnt,
+	.crc_entry = mtk_ovl_adaptor_crc_entry,
+	.crc_read = mtk_ovl_adaptor_crc_read,
+	.crc_start = mtk_ovl_adaptor_crc_start,
+	.crc_stop = mtk_ovl_adaptor_crc_stop,
 	.start = mtk_ovl_adaptor_start,
 	.stop = mtk_ovl_adaptor_stop,
 	.layer_nr = mtk_ovl_adaptor_layer_nr,
drivers/gpu/drm/mediatek/mtk_disp_drv.h +5 -0
index a03d7a10847a..0ef32bc3b996 100644
@@ -140,6 +140,11 @@ const u32 *mtk_ovl_adaptor_get_formats(struct device *dev);
 size_t mtk_ovl_adaptor_get_num_formats(struct device *dev);
 enum drm_mode_status mtk_ovl_adaptor_mode_valid(struct device *dev,
 						const struct drm_display_mode *mode);
+size_t mtk_ovl_adaptor_crc_cnt(struct device *dev);
+u32 *mtk_ovl_adaptor_crc_entry(struct device *dev);
+void mtk_ovl_adaptor_crc_read(struct device *dev);
+void mtk_ovl_adaptor_crc_start(struct device *dev);
+void mtk_ovl_adaptor_crc_stop(struct device *dev);
 
 void mtk_rdma_bypass_shadow(struct device *dev);
 int mtk_rdma_clk_enable(struct device *dev);
drivers/gpu/drm/mediatek/mtk_disp_ovl_adaptor.c +35 -0
index 2b62d6475918..02a617ac8265 100644
@@ -207,6 +207,41 @@ void mtk_ovl_adaptor_layer_config(struct device *dev, unsigned int idx,
 	mtk_ethdr_layer_config(ethdr, idx, state, cmdq_pkt);
 }
 
+size_t mtk_ovl_adaptor_crc_cnt(struct device *dev)
+{
+	struct mtk_disp_ovl_adaptor *ovl_adaptor = dev_get_drvdata(dev);
+
+	return mtk_ethdr_crc_cnt(ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_ETHDR0]);
+}
+
+u32 *mtk_ovl_adaptor_crc_entry(struct device *dev)
+{
+	struct mtk_disp_ovl_adaptor *ovl_adaptor = dev_get_drvdata(dev);
+
+	return mtk_ethdr_crc_entry(ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_ETHDR0]);
+}
+
+void mtk_ovl_adaptor_crc_read(struct device *dev)
+{
+	struct mtk_disp_ovl_adaptor *ovl_adaptor = dev_get_drvdata(dev);
+
+	mtk_ethdr_crc_read(ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_ETHDR0]);
+}
+
+void mtk_ovl_adaptor_crc_start(struct device *dev)
+{
+	struct mtk_disp_ovl_adaptor *ovl_adaptor = dev_get_drvdata(dev);
+
+	mtk_ethdr_crc_start(ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_ETHDR0]);
+}
+
+void mtk_ovl_adaptor_crc_stop(struct device *dev)
+{
+	struct mtk_disp_ovl_adaptor *ovl_adaptor = dev_get_drvdata(dev);
+
+	mtk_ethdr_crc_stop(ovl_adaptor->ovl_adaptor_comp[OVL_ADAPTOR_ETHDR0]);
+}
+
 void mtk_ovl_adaptor_config(struct device *dev, unsigned int w,
 			    unsigned int h, unsigned int vrefresh,
 			    unsigned int bpc, struct cmdq_pkt *cmdq_pkt)
drivers/gpu/drm/mediatek/mtk_ethdr.c +72 -0

120 lines not on this page · raw message

drivers/gpu/drm/mediatek/mtk_ethdr.h +7 -0
index 81af9edea3f7..6c479c460ac0 100644
@@ -22,4 +22,11 @@ void mtk_ethdr_register_vblank_cb(struct device *dev,
 void mtk_ethdr_unregister_vblank_cb(struct device *dev);
 void mtk_ethdr_enable_vblank(struct device *dev);
 void mtk_ethdr_disable_vblank(struct device *dev);
+
+size_t mtk_ethdr_crc_cnt(struct device *dev);
+u32 *mtk_ethdr_crc_entry(struct device *dev);
+void mtk_ethdr_crc_read(struct device *dev);
+void mtk_ethdr_crc_start(struct device *dev);
+void mtk_ethdr_crc_stop(struct device *dev);
+
 #endif
Signature · 1 line
2.18.0