All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH v2 1/3] gpu/drm: ingenic: Add trick to support 16bpp on 24-bit panels
@ 2020-02-26  4:30 ` Paul Cercueil
  0 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-02-26  4:30 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: od, dri-devel, linux-kernel, Paul Cercueil

If the panel interface is 24-bit but our primary plane is 16bpp,
configure as if the panel was 18-bit. This tricks permits the display
of 16bpp data on a 24-bit panel by wiring each color component to the
MSBs of the 24-bit interface.

v2: Check bytes-per-pixel count instead of fourcc format

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 drivers/gpu/drm/ingenic/ingenic-drm.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
index 6d47ef7b148c..5493a80d7d2f 100644
--- a/drivers/gpu/drm/ingenic/ingenic-drm.c
+++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
@@ -400,6 +400,8 @@ static void ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
 	struct drm_display_mode *mode = &crtc_state->adjusted_mode;
 	struct drm_connector *conn = conn_state->connector;
 	struct drm_display_info *info = &conn->display_info;
+	struct drm_plane_state *plane_state = crtc_state->crtc->primary->state;
+	const struct drm_format_info *finfo = NULL;
 	unsigned int cfg;
 
 	priv->panel_is_sharp = info->bus_flags & DRM_BUS_FLAG_SHARP_SIGNALS;
@@ -435,7 +437,21 @@ static void ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
 				cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
 				break;
 			case MEDIA_BUS_FMT_RGB888_1X24:
-				cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
+				if (plane_state && plane_state->fb)
+					finfo = plane_state->fb->format;
+
+				/*
+				 * If the panel interface is 24-bit but our
+				 * primary plane is 16bpp, configure as if the
+				 * panel was 18-bit. This tricks permits the
+				 * display of 16bpp data on a 24-bit panel by
+				 * wiring each color component to the MSBs of
+				 * the 24-bit interface.
+				 */
+				if (finfo && finfo->cpp[0] < 3)
+					cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
+				else
+					cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
 				break;
 			case MEDIA_BUS_FMT_RGB888_3X8:
 				cfg |= JZ_LCD_CFG_MODE_8BIT_SERIAL;
-- 
2.25.0


^ permalink raw reply related	[flat|nested] 12+ messages in thread

* [PATCH v2 1/3] gpu/drm: ingenic: Add trick to support 16bpp on 24-bit panels
@ 2020-02-26  4:30 ` Paul Cercueil
  0 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-02-26  4:30 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: Paul Cercueil, od, linux-kernel, dri-devel

If the panel interface is 24-bit but our primary plane is 16bpp,
configure as if the panel was 18-bit. This tricks permits the display
of 16bpp data on a 24-bit panel by wiring each color component to the
MSBs of the 24-bit interface.

v2: Check bytes-per-pixel count instead of fourcc format

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 drivers/gpu/drm/ingenic/ingenic-drm.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
index 6d47ef7b148c..5493a80d7d2f 100644
--- a/drivers/gpu/drm/ingenic/ingenic-drm.c
+++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
@@ -400,6 +400,8 @@ static void ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
 	struct drm_display_mode *mode = &crtc_state->adjusted_mode;
 	struct drm_connector *conn = conn_state->connector;
 	struct drm_display_info *info = &conn->display_info;
+	struct drm_plane_state *plane_state = crtc_state->crtc->primary->state;
+	const struct drm_format_info *finfo = NULL;
 	unsigned int cfg;
 
 	priv->panel_is_sharp = info->bus_flags & DRM_BUS_FLAG_SHARP_SIGNALS;
@@ -435,7 +437,21 @@ static void ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
 				cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
 				break;
 			case MEDIA_BUS_FMT_RGB888_1X24:
-				cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
+				if (plane_state && plane_state->fb)
+					finfo = plane_state->fb->format;
+
+				/*
+				 * If the panel interface is 24-bit but our
+				 * primary plane is 16bpp, configure as if the
+				 * panel was 18-bit. This tricks permits the
+				 * display of 16bpp data on a 24-bit panel by
+				 * wiring each color component to the MSBs of
+				 * the 24-bit interface.
+				 */
+				if (finfo && finfo->cpp[0] < 3)
+					cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
+				else
+					cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
 				break;
 			case MEDIA_BUS_FMT_RGB888_3X8:
 				cfg |= JZ_LCD_CFG_MODE_8BIT_SERIAL;
-- 
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply related	[flat|nested] 12+ messages in thread

* [PATCH v2 2/3] gpu/drm: ingenic: Switch emulated fbdev to 16bpp
  2020-02-26  4:30 ` Paul Cercueil
@ 2020-02-26  4:30   ` Paul Cercueil
  -1 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-02-26  4:30 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: od, dri-devel, linux-kernel, Paul Cercueil

The fbdev emulation is only ever used on Ingenic SoCs to run old SDL1
based games at 16bpp (rgb565). Recent applications generally talk to
DRM directly, and can request their favourite pixel format; so we can
make everybody happy by switching the emulated fbdev to 16bpp.

v2: No change

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 drivers/gpu/drm/ingenic/ingenic-drm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
index 5493a80d7d2f..3f8cc98d41fe 100644
--- a/drivers/gpu/drm/ingenic/ingenic-drm.c
+++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
@@ -807,7 +807,7 @@ static int ingenic_drm_probe(struct platform_device *pdev)
 		goto err_devclk_disable;
 	}
 
-	ret = drm_fbdev_generic_setup(drm, 32);
+	ret = drm_fbdev_generic_setup(drm, 16);
 	if (ret)
 		dev_warn(dev, "Unable to start fbdev emulation: %i", ret);
 
-- 
2.25.0


^ permalink raw reply related	[flat|nested] 12+ messages in thread

* [PATCH v2 2/3] gpu/drm: ingenic: Switch emulated fbdev to 16bpp
@ 2020-02-26  4:30   ` Paul Cercueil
  0 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-02-26  4:30 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: Paul Cercueil, od, linux-kernel, dri-devel

The fbdev emulation is only ever used on Ingenic SoCs to run old SDL1
based games at 16bpp (rgb565). Recent applications generally talk to
DRM directly, and can request their favourite pixel format; so we can
make everybody happy by switching the emulated fbdev to 16bpp.

v2: No change

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 drivers/gpu/drm/ingenic/ingenic-drm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
index 5493a80d7d2f..3f8cc98d41fe 100644
--- a/drivers/gpu/drm/ingenic/ingenic-drm.c
+++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
@@ -807,7 +807,7 @@ static int ingenic_drm_probe(struct platform_device *pdev)
 		goto err_devclk_disable;
 	}
 
-	ret = drm_fbdev_generic_setup(drm, 32);
+	ret = drm_fbdev_generic_setup(drm, 16);
 	if (ret)
 		dev_warn(dev, "Unable to start fbdev emulation: %i", ret);
 
-- 
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply related	[flat|nested] 12+ messages in thread

* [PATCH v2 3/3] gpu/drm: ingenic: Add option to mmap GEM buffers cached
  2020-02-26  4:30 ` Paul Cercueil
@ 2020-02-26  4:30   ` Paul Cercueil
  -1 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-02-26  4:30 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: od, dri-devel, linux-kernel, Paul Cercueil

Ingenic SoCs are most notably used in cheap chinese handheld gaming
consoles. There, the games and applications generally render in software
directly in the emulated framebuffer using SDL1.

Since the emulated framebuffer is mapped as write-combine by default,
these applications start to run really slow as soon as alpha-blending is
used.

Add a 'cached_gem_buffers' option to the ingenic-drm driver to mmap the
GEM buffers as fully cached to address this issue.

v2: Use standard noncoherent DMA APIs

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 drivers/gpu/drm/ingenic/ingenic-drm.c | 35 +++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
index 3f8cc98d41fe..e51ac8d62d27 100644
--- a/drivers/gpu/drm/ingenic/ingenic-drm.c
+++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
@@ -6,6 +6,8 @@
 
 #include <linux/clk.h>
 #include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
+#include <linux/io.h>
 #include <linux/module.h>
 #include <linux/of_device.h>
 #include <linux/platform_device.h>
@@ -30,6 +32,11 @@
 #include <drm/drm_probe_helper.h>
 #include <drm/drm_vblank.h>
 
+static bool ingenic_drm_cached_gem_buf;
+module_param_named(cached_gem_buffers, ingenic_drm_cached_gem_buf, bool, 0400);
+MODULE_PARM_DESC(cached_gem_buffers,
+		 "Enable fully cached GEM buffers [default=false]");
+
 #define JZ_REG_LCD_CFG				0x00
 #define JZ_REG_LCD_VSYNC			0x04
 #define JZ_REG_LCD_HSYNC			0x08
@@ -379,15 +386,23 @@ static void ingenic_drm_plane_atomic_update(struct drm_plane *plane,
 	struct drm_plane_state *state = plane->state;
 	unsigned int width, height, cpp;
 	dma_addr_t addr;
+	uint32_t len;
 
 	if (state && state->fb) {
 		addr = drm_fb_cma_get_gem_addr(state->fb, state, 0);
+
 		width = state->src_w >> 16;
 		height = state->src_h >> 16;
 		cpp = state->fb->format->cpp[plane->index];
+		len = width * height * cpp;
+
+		if (ingenic_drm_cached_gem_buf) {
+			dma_cache_sync(priv->dev, phys_to_virt(addr),
+				       len, DMA_TO_DEVICE);
+		}
 
 		priv->dma_hwdesc->addr = addr;
-		priv->dma_hwdesc->cmd = width * height * cpp / 4;
+		priv->dma_hwdesc->cmd = len / 4;
 		priv->dma_hwdesc->cmd |= JZ_LCD_CMD_EOF_IRQ;
 	}
 }
@@ -532,6 +547,22 @@ static void ingenic_drm_disable_vblank(struct drm_crtc *crtc)
 
 DEFINE_DRM_GEM_CMA_FOPS(ingenic_drm_fops);
 
+static int ingenic_drm_gem_mmap(struct drm_gem_object *obj,
+				struct vm_area_struct *vma)
+{
+	struct drm_gem_cma_object *cma_obj = to_drm_gem_cma_obj(obj);
+	struct ingenic_drm *priv = drm_device_get_priv(obj->dev);
+	unsigned long attrs = DMA_ATTR_NON_CONSISTENT;
+
+	if (!ingenic_drm_cached_gem_buf)
+		return drm_gem_cma_prime_mmap(obj, vma);
+
+	vma->vm_page_prot = dma_pgprot(priv->dev, vma->vm_page_prot, attrs);
+
+	return dma_mmap_attrs(priv->dev, vma, cma_obj->vaddr, cma_obj->paddr,
+			      vma->vm_end - vma->vm_start, attrs);
+}
+
 static struct drm_driver ingenic_drm_driver_data = {
 	.driver_features	= DRIVER_MODESET | DRIVER_GEM | DRIVER_ATOMIC,
 	.name			= "ingenic-drm",
@@ -553,7 +584,7 @@ static struct drm_driver ingenic_drm_driver_data = {
 	.gem_prime_import_sg_table = drm_gem_cma_prime_import_sg_table,
 	.gem_prime_vmap		= drm_gem_cma_prime_vmap,
 	.gem_prime_vunmap	= drm_gem_cma_prime_vunmap,
-	.gem_prime_mmap		= drm_gem_cma_prime_mmap,
+	.gem_prime_mmap		= ingenic_drm_gem_mmap,
 
 	.irq_handler		= ingenic_drm_irq_handler,
 	.release		= ingenic_drm_release,
-- 
2.25.0


^ permalink raw reply related	[flat|nested] 12+ messages in thread

* [PATCH v2 3/3] gpu/drm: ingenic: Add option to mmap GEM buffers cached
@ 2020-02-26  4:30   ` Paul Cercueil
  0 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-02-26  4:30 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: Paul Cercueil, od, linux-kernel, dri-devel

Ingenic SoCs are most notably used in cheap chinese handheld gaming
consoles. There, the games and applications generally render in software
directly in the emulated framebuffer using SDL1.

Since the emulated framebuffer is mapped as write-combine by default,
these applications start to run really slow as soon as alpha-blending is
used.

Add a 'cached_gem_buffers' option to the ingenic-drm driver to mmap the
GEM buffers as fully cached to address this issue.

v2: Use standard noncoherent DMA APIs

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 drivers/gpu/drm/ingenic/ingenic-drm.c | 35 +++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
index 3f8cc98d41fe..e51ac8d62d27 100644
--- a/drivers/gpu/drm/ingenic/ingenic-drm.c
+++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
@@ -6,6 +6,8 @@
 
 #include <linux/clk.h>
 #include <linux/dma-mapping.h>
+#include <linux/dma-noncoherent.h>
+#include <linux/io.h>
 #include <linux/module.h>
 #include <linux/of_device.h>
 #include <linux/platform_device.h>
@@ -30,6 +32,11 @@
 #include <drm/drm_probe_helper.h>
 #include <drm/drm_vblank.h>
 
+static bool ingenic_drm_cached_gem_buf;
+module_param_named(cached_gem_buffers, ingenic_drm_cached_gem_buf, bool, 0400);
+MODULE_PARM_DESC(cached_gem_buffers,
+		 "Enable fully cached GEM buffers [default=false]");
+
 #define JZ_REG_LCD_CFG				0x00
 #define JZ_REG_LCD_VSYNC			0x04
 #define JZ_REG_LCD_HSYNC			0x08
@@ -379,15 +386,23 @@ static void ingenic_drm_plane_atomic_update(struct drm_plane *plane,
 	struct drm_plane_state *state = plane->state;
 	unsigned int width, height, cpp;
 	dma_addr_t addr;
+	uint32_t len;
 
 	if (state && state->fb) {
 		addr = drm_fb_cma_get_gem_addr(state->fb, state, 0);
+
 		width = state->src_w >> 16;
 		height = state->src_h >> 16;
 		cpp = state->fb->format->cpp[plane->index];
+		len = width * height * cpp;
+
+		if (ingenic_drm_cached_gem_buf) {
+			dma_cache_sync(priv->dev, phys_to_virt(addr),
+				       len, DMA_TO_DEVICE);
+		}
 
 		priv->dma_hwdesc->addr = addr;
-		priv->dma_hwdesc->cmd = width * height * cpp / 4;
+		priv->dma_hwdesc->cmd = len / 4;
 		priv->dma_hwdesc->cmd |= JZ_LCD_CMD_EOF_IRQ;
 	}
 }
@@ -532,6 +547,22 @@ static void ingenic_drm_disable_vblank(struct drm_crtc *crtc)
 
 DEFINE_DRM_GEM_CMA_FOPS(ingenic_drm_fops);
 
+static int ingenic_drm_gem_mmap(struct drm_gem_object *obj,
+				struct vm_area_struct *vma)
+{
+	struct drm_gem_cma_object *cma_obj = to_drm_gem_cma_obj(obj);
+	struct ingenic_drm *priv = drm_device_get_priv(obj->dev);
+	unsigned long attrs = DMA_ATTR_NON_CONSISTENT;
+
+	if (!ingenic_drm_cached_gem_buf)
+		return drm_gem_cma_prime_mmap(obj, vma);
+
+	vma->vm_page_prot = dma_pgprot(priv->dev, vma->vm_page_prot, attrs);
+
+	return dma_mmap_attrs(priv->dev, vma, cma_obj->vaddr, cma_obj->paddr,
+			      vma->vm_end - vma->vm_start, attrs);
+}
+
 static struct drm_driver ingenic_drm_driver_data = {
 	.driver_features	= DRIVER_MODESET | DRIVER_GEM | DRIVER_ATOMIC,
 	.name			= "ingenic-drm",
@@ -553,7 +584,7 @@ static struct drm_driver ingenic_drm_driver_data = {
 	.gem_prime_import_sg_table = drm_gem_cma_prime_import_sg_table,
 	.gem_prime_vmap		= drm_gem_cma_prime_vmap,
 	.gem_prime_vunmap	= drm_gem_cma_prime_vunmap,
-	.gem_prime_mmap		= drm_gem_cma_prime_mmap,
+	.gem_prime_mmap		= ingenic_drm_gem_mmap,
 
 	.irq_handler		= ingenic_drm_irq_handler,
 	.release		= ingenic_drm_release,
-- 
2.25.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply related	[flat|nested] 12+ messages in thread

* Re: [PATCH v2 1/3] gpu/drm: ingenic: Add trick to support 16bpp on 24-bit panels
  2020-02-26  4:30 ` Paul Cercueil
@ 2020-04-08  9:51   ` Paul Cercueil
  -1 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-04-08  9:51 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: od, dri-devel, linux-kernel

Hi,

Any feedback for this patchset?

Thanks,
-Paul


Le mer. 26 févr. 2020 à 1:30, Paul Cercueil <paul@crapouillou.net> a 
écrit :
> If the panel interface is 24-bit but our primary plane is 16bpp,
> configure as if the panel was 18-bit. This tricks permits the display
> of 16bpp data on a 24-bit panel by wiring each color component to the
> MSBs of the 24-bit interface.
> 
> v2: Check bytes-per-pixel count instead of fourcc format
> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
>  drivers/gpu/drm/ingenic/ingenic-drm.c | 18 +++++++++++++++++-
>  1 file changed, 17 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c 
> b/drivers/gpu/drm/ingenic/ingenic-drm.c
> index 6d47ef7b148c..5493a80d7d2f 100644
> --- a/drivers/gpu/drm/ingenic/ingenic-drm.c
> +++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
> @@ -400,6 +400,8 @@ static void 
> ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
>  	struct drm_display_mode *mode = &crtc_state->adjusted_mode;
>  	struct drm_connector *conn = conn_state->connector;
>  	struct drm_display_info *info = &conn->display_info;
> +	struct drm_plane_state *plane_state = 
> crtc_state->crtc->primary->state;
> +	const struct drm_format_info *finfo = NULL;
>  	unsigned int cfg;
> 
>  	priv->panel_is_sharp = info->bus_flags & DRM_BUS_FLAG_SHARP_SIGNALS;
> @@ -435,7 +437,21 @@ static void 
> ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
>  				cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
>  				break;
>  			case MEDIA_BUS_FMT_RGB888_1X24:
> -				cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
> +				if (plane_state && plane_state->fb)
> +					finfo = plane_state->fb->format;
> +
> +				/*
> +				 * If the panel interface is 24-bit but our
> +				 * primary plane is 16bpp, configure as if the
> +				 * panel was 18-bit. This tricks permits the
> +				 * display of 16bpp data on a 24-bit panel by
> +				 * wiring each color component to the MSBs of
> +				 * the 24-bit interface.
> +				 */
> +				if (finfo && finfo->cpp[0] < 3)
> +					cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
> +				else
> +					cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
>  				break;
>  			case MEDIA_BUS_FMT_RGB888_3X8:
>  				cfg |= JZ_LCD_CFG_MODE_8BIT_SERIAL;
> --
> 2.25.0
> 



^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [PATCH v2 1/3] gpu/drm: ingenic: Add trick to support 16bpp on 24-bit panels
@ 2020-04-08  9:51   ` Paul Cercueil
  0 siblings, 0 replies; 12+ messages in thread
From: Paul Cercueil @ 2020-04-08  9:51 UTC (permalink / raw)
  To: Daniel Vetter, David Airlie; +Cc: od, linux-kernel, dri-devel

Hi,

Any feedback for this patchset?

Thanks,
-Paul


Le mer. 26 févr. 2020 à 1:30, Paul Cercueil <paul@crapouillou.net> a 
écrit :
> If the panel interface is 24-bit but our primary plane is 16bpp,
> configure as if the panel was 18-bit. This tricks permits the display
> of 16bpp data on a 24-bit panel by wiring each color component to the
> MSBs of the 24-bit interface.
> 
> v2: Check bytes-per-pixel count instead of fourcc format
> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
>  drivers/gpu/drm/ingenic/ingenic-drm.c | 18 +++++++++++++++++-
>  1 file changed, 17 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c 
> b/drivers/gpu/drm/ingenic/ingenic-drm.c
> index 6d47ef7b148c..5493a80d7d2f 100644
> --- a/drivers/gpu/drm/ingenic/ingenic-drm.c
> +++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
> @@ -400,6 +400,8 @@ static void 
> ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
>  	struct drm_display_mode *mode = &crtc_state->adjusted_mode;
>  	struct drm_connector *conn = conn_state->connector;
>  	struct drm_display_info *info = &conn->display_info;
> +	struct drm_plane_state *plane_state = 
> crtc_state->crtc->primary->state;
> +	const struct drm_format_info *finfo = NULL;
>  	unsigned int cfg;
> 
>  	priv->panel_is_sharp = info->bus_flags & DRM_BUS_FLAG_SHARP_SIGNALS;
> @@ -435,7 +437,21 @@ static void 
> ingenic_drm_encoder_atomic_mode_set(struct drm_encoder *encoder,
>  				cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
>  				break;
>  			case MEDIA_BUS_FMT_RGB888_1X24:
> -				cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
> +				if (plane_state && plane_state->fb)
> +					finfo = plane_state->fb->format;
> +
> +				/*
> +				 * If the panel interface is 24-bit but our
> +				 * primary plane is 16bpp, configure as if the
> +				 * panel was 18-bit. This tricks permits the
> +				 * display of 16bpp data on a 24-bit panel by
> +				 * wiring each color component to the MSBs of
> +				 * the 24-bit interface.
> +				 */
> +				if (finfo && finfo->cpp[0] < 3)
> +					cfg |= JZ_LCD_CFG_MODE_GENERIC_18BIT;
> +				else
> +					cfg |= JZ_LCD_CFG_MODE_GENERIC_24BIT;
>  				break;
>  			case MEDIA_BUS_FMT_RGB888_3X8:
>  				cfg |= JZ_LCD_CFG_MODE_8BIT_SERIAL;
> --
> 2.25.0
> 


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [PATCH v2 2/3] gpu/drm: ingenic: Switch emulated fbdev to 16bpp
  2020-02-26  4:30   ` Paul Cercueil
@ 2020-04-08 12:04     ` Daniel Vetter
  -1 siblings, 0 replies; 12+ messages in thread
From: Daniel Vetter @ 2020-04-08 12:04 UTC (permalink / raw)
  To: Paul Cercueil; +Cc: Daniel Vetter, David Airlie, od, dri-devel, linux-kernel

On Wed, Feb 26, 2020 at 01:30:40AM -0300, Paul Cercueil wrote:
> The fbdev emulation is only ever used on Ingenic SoCs to run old SDL1
> based games at 16bpp (rgb565). Recent applications generally talk to
> DRM directly, and can request their favourite pixel format; so we can
> make everybody happy by switching the emulated fbdev to 16bpp.
> 
> v2: No change
> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
>  drivers/gpu/drm/ingenic/ingenic-drm.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
> index 5493a80d7d2f..3f8cc98d41fe 100644
> --- a/drivers/gpu/drm/ingenic/ingenic-drm.c
> +++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
> @@ -807,7 +807,7 @@ static int ingenic_drm_probe(struct platform_device *pdev)
>  		goto err_devclk_disable;
>  	}
>  
> -	ret = drm_fbdev_generic_setup(drm, 32);
> +	ret = drm_fbdev_generic_setup(drm, 16);

This will conflict with Thomas' work to remove the error code, so might
need a rebase. Now for the patch itself your supposed to set
drm_device->mode_config.preferred_bpp and leave this here as 0. That way
userspace can also know what's the best option.

Maybe in the future we could auto-compute the preferred_bpp from just
taking the first format of the first primary plane, but atm not there.

With the above change:

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>  	if (ret)
>  		dev_warn(dev, "Unable to start fbdev emulation: %i", ret);
>  
> -- 
> 2.25.0
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [PATCH v2 2/3] gpu/drm: ingenic: Switch emulated fbdev to 16bpp
@ 2020-04-08 12:04     ` Daniel Vetter
  0 siblings, 0 replies; 12+ messages in thread
From: Daniel Vetter @ 2020-04-08 12:04 UTC (permalink / raw)
  To: Paul Cercueil; +Cc: David Airlie, od, dri-devel, linux-kernel

On Wed, Feb 26, 2020 at 01:30:40AM -0300, Paul Cercueil wrote:
> The fbdev emulation is only ever used on Ingenic SoCs to run old SDL1
> based games at 16bpp (rgb565). Recent applications generally talk to
> DRM directly, and can request their favourite pixel format; so we can
> make everybody happy by switching the emulated fbdev to 16bpp.
> 
> v2: No change
> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
>  drivers/gpu/drm/ingenic/ingenic-drm.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
> index 5493a80d7d2f..3f8cc98d41fe 100644
> --- a/drivers/gpu/drm/ingenic/ingenic-drm.c
> +++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
> @@ -807,7 +807,7 @@ static int ingenic_drm_probe(struct platform_device *pdev)
>  		goto err_devclk_disable;
>  	}
>  
> -	ret = drm_fbdev_generic_setup(drm, 32);
> +	ret = drm_fbdev_generic_setup(drm, 16);

This will conflict with Thomas' work to remove the error code, so might
need a rebase. Now for the patch itself your supposed to set
drm_device->mode_config.preferred_bpp and leave this here as 0. That way
userspace can also know what's the best option.

Maybe in the future we could auto-compute the preferred_bpp from just
taking the first format of the first primary plane, but atm not there.

With the above change:

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>  	if (ret)
>  		dev_warn(dev, "Unable to start fbdev emulation: %i", ret);
>  
> -- 
> 2.25.0
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [PATCH v2 3/3] gpu/drm: ingenic: Add option to mmap GEM buffers cached
  2020-02-26  4:30   ` Paul Cercueil
@ 2020-04-08 12:06     ` Daniel Vetter
  -1 siblings, 0 replies; 12+ messages in thread
From: Daniel Vetter @ 2020-04-08 12:06 UTC (permalink / raw)
  To: Paul Cercueil; +Cc: Daniel Vetter, David Airlie, od, dri-devel, linux-kernel

On Wed, Feb 26, 2020 at 01:30:41AM -0300, Paul Cercueil wrote:
> Ingenic SoCs are most notably used in cheap chinese handheld gaming
> consoles. There, the games and applications generally render in software
> directly in the emulated framebuffer using SDL1.
> 
> Since the emulated framebuffer is mapped as write-combine by default,
> these applications start to run really slow as soon as alpha-blending is
> used.
> 
> Add a 'cached_gem_buffers' option to the ingenic-drm driver to mmap the
> GEM buffers as fully cached to address this issue.
> 
> v2: Use standard noncoherent DMA APIs
> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>

Dumb buffers don't have any cache flushing controls, so without a ->dirty
callback I'm not sure how this exactly works. I think you need a pile more
here.

Also there's a prefere_shadow bit that you're supposed to set in this
case, and which userspace is supposed to obey.

Also kinda surprised that fbdev userspace is this bad, since generally all
of fbdev is wc. Traditionally at least.
-Daniel

> ---
>  drivers/gpu/drm/ingenic/ingenic-drm.c | 35 +++++++++++++++++++++++++--
>  1 file changed, 33 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
> index 3f8cc98d41fe..e51ac8d62d27 100644
> --- a/drivers/gpu/drm/ingenic/ingenic-drm.c
> +++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
> @@ -6,6 +6,8 @@
>  
>  #include <linux/clk.h>
>  #include <linux/dma-mapping.h>
> +#include <linux/dma-noncoherent.h>
> +#include <linux/io.h>
>  #include <linux/module.h>
>  #include <linux/of_device.h>
>  #include <linux/platform_device.h>
> @@ -30,6 +32,11 @@
>  #include <drm/drm_probe_helper.h>
>  #include <drm/drm_vblank.h>
>  
> +static bool ingenic_drm_cached_gem_buf;
> +module_param_named(cached_gem_buffers, ingenic_drm_cached_gem_buf, bool, 0400);
> +MODULE_PARM_DESC(cached_gem_buffers,
> +		 "Enable fully cached GEM buffers [default=false]");
> +
>  #define JZ_REG_LCD_CFG				0x00
>  #define JZ_REG_LCD_VSYNC			0x04
>  #define JZ_REG_LCD_HSYNC			0x08
> @@ -379,15 +386,23 @@ static void ingenic_drm_plane_atomic_update(struct drm_plane *plane,
>  	struct drm_plane_state *state = plane->state;
>  	unsigned int width, height, cpp;
>  	dma_addr_t addr;
> +	uint32_t len;
>  
>  	if (state && state->fb) {
>  		addr = drm_fb_cma_get_gem_addr(state->fb, state, 0);
> +
>  		width = state->src_w >> 16;
>  		height = state->src_h >> 16;
>  		cpp = state->fb->format->cpp[plane->index];
> +		len = width * height * cpp;
> +
> +		if (ingenic_drm_cached_gem_buf) {
> +			dma_cache_sync(priv->dev, phys_to_virt(addr),
> +				       len, DMA_TO_DEVICE);
> +		}
>  
>  		priv->dma_hwdesc->addr = addr;
> -		priv->dma_hwdesc->cmd = width * height * cpp / 4;
> +		priv->dma_hwdesc->cmd = len / 4;
>  		priv->dma_hwdesc->cmd |= JZ_LCD_CMD_EOF_IRQ;
>  	}
>  }
> @@ -532,6 +547,22 @@ static void ingenic_drm_disable_vblank(struct drm_crtc *crtc)
>  
>  DEFINE_DRM_GEM_CMA_FOPS(ingenic_drm_fops);
>  
> +static int ingenic_drm_gem_mmap(struct drm_gem_object *obj,
> +				struct vm_area_struct *vma)
> +{
> +	struct drm_gem_cma_object *cma_obj = to_drm_gem_cma_obj(obj);
> +	struct ingenic_drm *priv = drm_device_get_priv(obj->dev);
> +	unsigned long attrs = DMA_ATTR_NON_CONSISTENT;
> +
> +	if (!ingenic_drm_cached_gem_buf)
> +		return drm_gem_cma_prime_mmap(obj, vma);
> +
> +	vma->vm_page_prot = dma_pgprot(priv->dev, vma->vm_page_prot, attrs);
> +
> +	return dma_mmap_attrs(priv->dev, vma, cma_obj->vaddr, cma_obj->paddr,
> +			      vma->vm_end - vma->vm_start, attrs);
> +}
> +
>  static struct drm_driver ingenic_drm_driver_data = {
>  	.driver_features	= DRIVER_MODESET | DRIVER_GEM | DRIVER_ATOMIC,
>  	.name			= "ingenic-drm",
> @@ -553,7 +584,7 @@ static struct drm_driver ingenic_drm_driver_data = {
>  	.gem_prime_import_sg_table = drm_gem_cma_prime_import_sg_table,
>  	.gem_prime_vmap		= drm_gem_cma_prime_vmap,
>  	.gem_prime_vunmap	= drm_gem_cma_prime_vunmap,
> -	.gem_prime_mmap		= drm_gem_cma_prime_mmap,
> +	.gem_prime_mmap		= ingenic_drm_gem_mmap,
>  
>  	.irq_handler		= ingenic_drm_irq_handler,
>  	.release		= ingenic_drm_release,
> -- 
> 2.25.0
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [PATCH v2 3/3] gpu/drm: ingenic: Add option to mmap GEM buffers cached
@ 2020-04-08 12:06     ` Daniel Vetter
  0 siblings, 0 replies; 12+ messages in thread
From: Daniel Vetter @ 2020-04-08 12:06 UTC (permalink / raw)
  To: Paul Cercueil; +Cc: David Airlie, od, dri-devel, linux-kernel

On Wed, Feb 26, 2020 at 01:30:41AM -0300, Paul Cercueil wrote:
> Ingenic SoCs are most notably used in cheap chinese handheld gaming
> consoles. There, the games and applications generally render in software
> directly in the emulated framebuffer using SDL1.
> 
> Since the emulated framebuffer is mapped as write-combine by default,
> these applications start to run really slow as soon as alpha-blending is
> used.
> 
> Add a 'cached_gem_buffers' option to the ingenic-drm driver to mmap the
> GEM buffers as fully cached to address this issue.
> 
> v2: Use standard noncoherent DMA APIs
> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>

Dumb buffers don't have any cache flushing controls, so without a ->dirty
callback I'm not sure how this exactly works. I think you need a pile more
here.

Also there's a prefere_shadow bit that you're supposed to set in this
case, and which userspace is supposed to obey.

Also kinda surprised that fbdev userspace is this bad, since generally all
of fbdev is wc. Traditionally at least.
-Daniel

> ---
>  drivers/gpu/drm/ingenic/ingenic-drm.c | 35 +++++++++++++++++++++++++--
>  1 file changed, 33 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ingenic/ingenic-drm.c b/drivers/gpu/drm/ingenic/ingenic-drm.c
> index 3f8cc98d41fe..e51ac8d62d27 100644
> --- a/drivers/gpu/drm/ingenic/ingenic-drm.c
> +++ b/drivers/gpu/drm/ingenic/ingenic-drm.c
> @@ -6,6 +6,8 @@
>  
>  #include <linux/clk.h>
>  #include <linux/dma-mapping.h>
> +#include <linux/dma-noncoherent.h>
> +#include <linux/io.h>
>  #include <linux/module.h>
>  #include <linux/of_device.h>
>  #include <linux/platform_device.h>
> @@ -30,6 +32,11 @@
>  #include <drm/drm_probe_helper.h>
>  #include <drm/drm_vblank.h>
>  
> +static bool ingenic_drm_cached_gem_buf;
> +module_param_named(cached_gem_buffers, ingenic_drm_cached_gem_buf, bool, 0400);
> +MODULE_PARM_DESC(cached_gem_buffers,
> +		 "Enable fully cached GEM buffers [default=false]");
> +
>  #define JZ_REG_LCD_CFG				0x00
>  #define JZ_REG_LCD_VSYNC			0x04
>  #define JZ_REG_LCD_HSYNC			0x08
> @@ -379,15 +386,23 @@ static void ingenic_drm_plane_atomic_update(struct drm_plane *plane,
>  	struct drm_plane_state *state = plane->state;
>  	unsigned int width, height, cpp;
>  	dma_addr_t addr;
> +	uint32_t len;
>  
>  	if (state && state->fb) {
>  		addr = drm_fb_cma_get_gem_addr(state->fb, state, 0);
> +
>  		width = state->src_w >> 16;
>  		height = state->src_h >> 16;
>  		cpp = state->fb->format->cpp[plane->index];
> +		len = width * height * cpp;
> +
> +		if (ingenic_drm_cached_gem_buf) {
> +			dma_cache_sync(priv->dev, phys_to_virt(addr),
> +				       len, DMA_TO_DEVICE);
> +		}
>  
>  		priv->dma_hwdesc->addr = addr;
> -		priv->dma_hwdesc->cmd = width * height * cpp / 4;
> +		priv->dma_hwdesc->cmd = len / 4;
>  		priv->dma_hwdesc->cmd |= JZ_LCD_CMD_EOF_IRQ;
>  	}
>  }
> @@ -532,6 +547,22 @@ static void ingenic_drm_disable_vblank(struct drm_crtc *crtc)
>  
>  DEFINE_DRM_GEM_CMA_FOPS(ingenic_drm_fops);
>  
> +static int ingenic_drm_gem_mmap(struct drm_gem_object *obj,
> +				struct vm_area_struct *vma)
> +{
> +	struct drm_gem_cma_object *cma_obj = to_drm_gem_cma_obj(obj);
> +	struct ingenic_drm *priv = drm_device_get_priv(obj->dev);
> +	unsigned long attrs = DMA_ATTR_NON_CONSISTENT;
> +
> +	if (!ingenic_drm_cached_gem_buf)
> +		return drm_gem_cma_prime_mmap(obj, vma);
> +
> +	vma->vm_page_prot = dma_pgprot(priv->dev, vma->vm_page_prot, attrs);
> +
> +	return dma_mmap_attrs(priv->dev, vma, cma_obj->vaddr, cma_obj->paddr,
> +			      vma->vm_end - vma->vm_start, attrs);
> +}
> +
>  static struct drm_driver ingenic_drm_driver_data = {
>  	.driver_features	= DRIVER_MODESET | DRIVER_GEM | DRIVER_ATOMIC,
>  	.name			= "ingenic-drm",
> @@ -553,7 +584,7 @@ static struct drm_driver ingenic_drm_driver_data = {
>  	.gem_prime_import_sg_table = drm_gem_cma_prime_import_sg_table,
>  	.gem_prime_vmap		= drm_gem_cma_prime_vmap,
>  	.gem_prime_vunmap	= drm_gem_cma_prime_vunmap,
> -	.gem_prime_mmap		= drm_gem_cma_prime_mmap,
> +	.gem_prime_mmap		= ingenic_drm_gem_mmap,
>  
>  	.irq_handler		= ingenic_drm_irq_handler,
>  	.release		= ingenic_drm_release,
> -- 
> 2.25.0
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

^ permalink raw reply	[flat|nested] 12+ messages in thread

end of thread, other threads:[~2020-04-09  7:34 UTC | newest]

Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-02-26  4:30 [PATCH v2 1/3] gpu/drm: ingenic: Add trick to support 16bpp on 24-bit panels Paul Cercueil
2020-02-26  4:30 ` Paul Cercueil
2020-02-26  4:30 ` [PATCH v2 2/3] gpu/drm: ingenic: Switch emulated fbdev to 16bpp Paul Cercueil
2020-02-26  4:30   ` Paul Cercueil
2020-04-08 12:04   ` Daniel Vetter
2020-04-08 12:04     ` Daniel Vetter
2020-02-26  4:30 ` [PATCH v2 3/3] gpu/drm: ingenic: Add option to mmap GEM buffers cached Paul Cercueil
2020-02-26  4:30   ` Paul Cercueil
2020-04-08 12:06   ` Daniel Vetter
2020-04-08 12:06     ` Daniel Vetter
2020-04-08  9:51 ` [PATCH v2 1/3] gpu/drm: ingenic: Add trick to support 16bpp on 24-bit panels Paul Cercueil
2020-04-08  9:51   ` Paul Cercueil

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.