Skip to contentSkip to conversation

Add framebuffer max width/height fields to drm_mode_config

PATCH New Jeykumar Sankaran 4 participants 8 messages Sep 28, 2019 – Oct 14, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Add framebuffer max width/height fields to drm_mode_config

Below two discussion threads will provide the context behind this patch.

https://www.spinics.net/lists/dri-devel/msg229070.html
https://lore.kernel.org/linux-arm-msm/[email protected]/T/

Seperating out the core framework patch from vendor implementation.

Jeykumar Sankaran (1):
drm: add fb max width/height fields to drm_mode_config

 drivers/gpu/drm/drm_framebuffer.c | 17 +++++++++++++----
 include/drm/drm_mode_config.h     |  3 +++
 2 files changed, 16 insertions(+), 4 deletions(-)
Signature · 7 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 1/1 drm: add fb max width/height fields to drm_mode_config

The mode_config max width/height values determine the maximum resolution the pixel reader can handle. But the same values are used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

Vendors setting these fields should fix their mode_set paths too by filtering and validating the modes against the appropriate max fields in their mode_valid() implementations.

---
 drivers/gpu/drm/drm_framebuffer.c | 15 +++++++++++----
 include/drm/drm_mode_config.h     |  3 +++
 2 files changed, 14 insertions(+), 4 deletions(-)
drivers/gpu/drm/drm_framebuffer.c +11 -4
index 5756431..2083168 100644
@@ -300,14 +300,21 @@ struct drm_framebuffer *
 		return ERR_PTR(-EINVAL);
 	}
 
-	if ((config->min_width > r->width) || (r->width > config->max_width)) {
+	if ((config->min_width > r->width) ||
+	    (!config->max_fb_width && r->width > config->max_width) ||
+	    (config->max_fb_width && r->width > config->max_fb_width)) {
 		DRM_DEBUG_KMS("bad framebuffer width %d, should be >= %d && <= %d\n",
-			  r->width, config->min_width, config->max_width);
+			r->width, config->min_width, config->max_fb_width ?
+			config->max_fb_width : config->max_width);
 		return ERR_PTR(-EINVAL);
 	}
-	if ((config->min_height > r->height) || (r->height > config->max_height)) {
+
+	if ((config->min_height > r->height) ||
+	    (!config->max_fb_height && r->height > config->max_height) ||
+	    (config->max_fb_height && r->height > config->max_fb_height)) {
 		DRM_DEBUG_KMS("bad framebuffer height %d, should be >= %d && <= %d\n",
-			  r->height, config->min_height, config->max_height);
+			r->height, config->min_height, config->max_fb_width ?
+			config->max_fb_height : config->max_height);
 		return ERR_PTR(-EINVAL);
 	}
 
include/drm/drm_mode_config.h +3 -0
index 3bcbe30..c6394ed 100644
@@ -339,6 +339,8 @@ struct drm_mode_config_funcs {
  * @min_height: minimum fb pixel height on this device
  * @max_width: maximum fb pixel width on this device
  * @max_height: maximum fb pixel height on this device
+ * @max_fb_width: maximum fb buffer width if differs from max_width
+ * @max_fb_height: maximum fb buffer height if differs from  max_height
  * @funcs: core driver provided mode setting functions
  * @fb_base: base address of the framebuffer
  * @poll_enabled: track polling support for this device
@@ -523,6 +525,7 @@ struct drm_mode_config {
 
 	int min_width, min_height;
 	int max_width, max_height;
+	int max_fb_width, max_fb_height;
 	const struct drm_mode_config_funcs *funcs;
 	resource_size_t fb_base;
 
Signature · 7 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#2 lore raw

Re: [PATCH] drm: add fb max width/height fields to drm_mode_config

Quoted 2 lines · Jeykumar Sankaran

The mode_config max width/height values determine the maximum resolution the pixel reader can handle.

Not according to the docs I "fixed" a while ago.

Quoted 9 lines

But the same values are
used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

If you're going to change the meaning of the old values you need to fix the drivers too.

Personally I don't see too much point in this since you most likely want to validate all the other timings as well, and so likely need some kind of mode_valid implementation anyway. Hence to validate modes there's not much benefit of having global min/max values.

Quoted 71 lines

Vendors setting these fields should fix their mode_set paths too by filtering and validating the modes against the appropriate max fields in their mode_valid() implementations.

---
 drivers/gpu/drm/drm_framebuffer.c | 15 +++++++++++----
 include/drm/drm_mode_config.h     |  3 +++
 2 files changed, 14 insertions(+), 4 deletions(-)
drivers/gpu/drm/drm_framebuffer.c +11 -4
index 5756431..2083168 100644
@@ -300,14 +300,21 @@ struct drm_framebuffer *
 		return ERR_PTR(-EINVAL);
 	}
 
-	if ((config->min_width > r->width) || (r->width > config->max_width)) {
+	if ((config->min_width > r->width) ||
+	    (!config->max_fb_width && r->width > config->max_width) ||
+	    (config->max_fb_width && r->width > config->max_fb_width)) {
 		DRM_DEBUG_KMS("bad framebuffer width %d, should be >= %d && <= %d\n",
-			  r->width, config->min_width, config->max_width);
+			r->width, config->min_width, config->max_fb_width ?
+			config->max_fb_width : config->max_width);
 		return ERR_PTR(-EINVAL);
 	}
-	if ((config->min_height > r->height) || (r->height > config->max_height)) {
+
+	if ((config->min_height > r->height) ||
+	    (!config->max_fb_height && r->height > config->max_height) ||
+	    (config->max_fb_height && r->height > config->max_fb_height)) {
 		DRM_DEBUG_KMS("bad framebuffer height %d, should be >= %d && <= %d\n",
-			  r->height, config->min_height, config->max_height);
+			r->height, config->min_height, config->max_fb_width ?
+			config->max_fb_height : config->max_height);
 		return ERR_PTR(-EINVAL);
 	}
 
include/drm/drm_mode_config.h +3 -0
index 3bcbe30..c6394ed 100644
@@ -339,6 +339,8 @@ struct drm_mode_config_funcs {
  * @min_height: minimum fb pixel height on this device
  * @max_width: maximum fb pixel width on this device
  * @max_height: maximum fb pixel height on this device
+ * @max_fb_width: maximum fb buffer width if differs from max_width
+ * @max_fb_height: maximum fb buffer height if differs from  max_height
  * @funcs: core driver provided mode setting functions
  * @fb_base: base address of the framebuffer
  * @poll_enabled: track polling support for this device
@@ -523,6 +525,7 @@ struct drm_mode_config {
 
 	int min_width, min_height;
 	int max_width, max_height;
+	int max_fb_width, max_fb_height;
 	const struct drm_mode_config_funcs *funcs;
 	resource_size_t fb_base;
 
Signature · 7 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 2 lines
Ville Syrjälä
Intel
#3 lore raw

Re: [PATCH] drm: add fb max width/height fields to drm_mode_config

Quoted 16 lines · Ville Syrjälä
Quoted 2 lines · Jeykumar Sankaran

The mode_config max width/height values determine the maximum resolution the pixel reader can handle.

Not according to the docs I "fixed" a while ago.

Quoted 9 lines

But the same values are
used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

Quoted 8 lines

If you're going to change the meaning of the old values you need to fix the drivers too.

Personally I don't see too much point in this since you most likely want to validate all the other timings as well, and so likely need some kind of mode_valid implementation anyway. Hence to validate modes there's not much benefit of having global min/max values.

https://patchwork.kernel.org/patch/10467155/

I believe you are referring to this patch.

I am primarily interested in the scaling scenario mentioned here. MSM and a few other hardware have scaling block that are used both ways:

1) Where FB limits are larger than the display limits. Scalar blocks are used to downscale the framebuffers and render within display limits.

In this scenario, with your patch, are you suggesting the drivers maintain the display limits locally and use those values in fill_modes() / mode_valid() to filter out invalid modes explicitly instead of mode_config.max_width/height?

2) Where FB limits are smaller than display limits. Enforced for performance reasons on low tier hardware. It reduces the fetch bandwidth and uses post blending scalar block to scale up the pixel stream to match the display resolution.

Any suggestions on how this topology can be handled with a single set of max/min values?

Thanks and Regards,
Jeykumar S.

Quoted 87 lines
Quoted 5 lines

Vendors setting these fields should fix their mode_set paths too by filtering and validating the modes against the appropriate max fields in their mode_valid() implementations.

Review

Quoted 8 lines
---
 drivers/gpu/drm/drm_framebuffer.c | 15 +++++++++++----
 include/drm/drm_mode_config.h     |  3 +++
 2 files changed, 14 insertions(+), 4 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/drm_framebuffer.c

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer * · line 303

Quoted 8 lines

index 5756431..2083168 100644

drivers/gpu/drm/drm_framebuffer.c +0 -1
@@ -300,14 +300,21 @@ struct drm_framebuffer *
 		return ERR_PTR(-EINVAL);
 	}
 
-	if ((config->min_width > r->width) || (r->width >

config->max_width)) {

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer *

Quoted 4 lines
drivers/gpu/drm/drm_framebuffer.c +3 -0
+	if ((config->min_width > r->width) ||
+	    (!config->max_fb_width && r->width > config->max_width) ||
+	    (config->max_fb_width && r->width > config->max_fb_width)) {
 		DRM_DEBUG_KMS("bad framebuffer width %d, should be >= %d

&& <= %d\n",

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer * · line 307

Quoted 2 lines
drivers/gpu/drm/drm_framebuffer.c +1 -1
-			  r->width, config->min_width, config->max_width);
+			r->width, config->min_width, config->max_fb_width

?

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer *

Quoted 4 lines
drivers/gpu/drm/drm_framebuffer.c +1 -1
+			config->max_fb_width : config->max_width);
 		return ERR_PTR(-EINVAL);
 	}
-	if ((config->min_height > r->height) || (r->height >

config->max_height)) {

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer *

Quoted 4 lines
drivers/gpu/drm/drm_framebuffer.c +4 -0
+
+	if ((config->min_height > r->height) ||
+	    (!config->max_fb_height && r->height > config->max_height) ||
+	    (config->max_fb_height && r->height > config->max_fb_height))

{

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer * · line 315

Quoted 1 line
drivers/gpu/drm/drm_framebuffer.c +0 -0
 		DRM_DEBUG_KMS("bad framebuffer height %d, should be >= %d

&& <= %d\n",

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer * · line 310

Quoted 1 line
drivers/gpu/drm/drm_framebuffer.c +0 -1
-			  r->height, config->min_height,

config->max_height);

drivers/gpu/drm/drm_framebuffer.c struct drm_framebuffer * · line 316

Quoted 1 line
drivers/gpu/drm/drm_framebuffer.c +1 -0
+			r->height, config->min_height,

config->max_fb_width ?

Review

Quoted 5 lines
drivers/gpu/drm/drm_framebuffer.c +1 -0
+			config->max_fb_height : config->max_height);
 		return ERR_PTR(-EINVAL);
 	}
 
(quoted diff) +0 -0

b/include/drm/drm_mode_config.h

Quoted 22 lines

index 3bcbe30..c6394ed 100644

include/drm/drm_mode_config.h +2 -0
@@ -339,6 +339,8 @@ struct drm_mode_config_funcs {
  * @min_height: minimum fb pixel height on this device
  * @max_width: maximum fb pixel width on this device
  * @max_height: maximum fb pixel height on this device
+ * @max_fb_width: maximum fb buffer width if differs from max_width
+ * @max_fb_height: maximum fb buffer height if differs from  

max_height
* @funcs: core driver provided mode setting functions
* @fb_base: base address of the framebuffer
* @poll_enabled: track polling support for this device

include/drm/drm_mode_config.h +1 -1
@@ -523,6 +525,7 @@ struct drm_mode_config {
 
 	int min_width, min_height;
 	int max_width, max_height;
+	int max_fb_width, max_fb_height;
 	const struct drm_mode_config_funcs *funcs;
 	resource_size_t fb_base;
 
--
Quoted 1 line

The Qualcomm Innovation Center, Inc. is a member of the Code Aurora

Forum,

Quoted 6 lines

a Linux Foundation Collaborative Project

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

--
Ville Syrjälä
Intel

Signature · 1 line
Jeykumar S
#4 lore raw

Re: [PATCH] drm: add fb max width/height fields to drm_mode_config

Quoted 42 lines · Jeykumar Sankaran
Quoted 16 lines · Ville Syrjälä
Quoted 2 lines · Jeykumar Sankaran

The mode_config max width/height values determine the maximum resolution the pixel reader can handle.

Not according to the docs I "fixed" a while ago.

Quoted 9 lines

But the same values are
used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

Quoted 8 lines

If you're going to change the meaning of the old values you need to fix the drivers too.

Personally I don't see too much point in this since you most likely want to validate all the other timings as well, and so likely need some kind of mode_valid implementation anyway. Hence to validate modes there's not much benefit of having global min/max values.

https://patchwork.kernel.org/patch/10467155/

I believe you are referring to this patch.

I am primarily interested in the scaling scenario mentioned here. MSM and a few other hardware have scaling block that are used both ways:

1) Where FB limits are larger than the display limits. Scalar blocks are used to
downscale the framebuffers and render within display limits.

In this scenario, with your patch, are you suggesting the drivers maintain the
display limits locally and use those values in fill_modes() / mode_valid() to filter
out invalid modes explicitly instead of mode_config.max_width/height?

Quoted 5 lines

2) Where FB limits are smaller than display limits. Enforced for performance reasons on low tier hardware.
It reduces the fetch bandwidth and uses post blending scalar block to scale up the pixel stream
to match the display resolution.

As Daniel mentioned in that discussion your typical userspace assumes that it can use a single unscaled framebuffer with any advertised mode. Hence I believe limiting the mode list based on the max framebuffer size is pretty much required unless you want to break existing userspace.

In i915 I went a bit further than that recently and now we filter the mode list based on the maximum plane size [1] (which can be less than the max fb size and less than the maximum crtc dimensions). And again that's because userspace assumes that it can just use a single unscaled fullscreen plane to cover the entire crtc.

These assumption are also carved into the legacy setcrtc uapi where you can't even specify multiple framebuffers. In theory a driver could internally use multiple planes to overcome some of the limitations, but in i915 at least we don't.

[1] https://cgit.freedesktop.org/drm/drm-intel/commit/?id=2d20411e25a3bf3d2914a2219f47ed48dc57aed5

Quoted 4 lines

Any suggestions on how this topology can be handled with a single set of max/min values?

I think a safe way to relax these rules would be to either:
a) Add a client cap by which userspace can inform the kernel it understands there are more complicated limits at play and thus can't assume that everything will just work
b) Maybe we could just tie that in with the atomic cap since atomic clients are pretty much required to do the TEST_ONLY dance anyway, so one might hope they have a working fallback strategy. Though I suspect eg. the modesetting ddx wouldn't like this. But we no longer allow atomic with X anyway so that partcular argument may not hold much weight anymore.

Signature · 2 lines
Ville Syrjälä
Intel
#5 lore raw

Re: [PATCH] drm: add fb max width/height fields to drm_mode_config

Quoted 75 lines · Ville Syrjälä
Quoted 42 lines · Jeykumar Sankaran
Quoted 16 lines · Ville Syrjälä
Quoted 2 lines · Jeykumar Sankaran

The mode_config max width/height values determine the maximum resolution the pixel reader can handle.

Not according to the docs I "fixed" a while ago.

Quoted 9 lines

But the same values are
used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

Quoted 8 lines

If you're going to change the meaning of the old values you need to fix the drivers too.

Personally I don't see too much point in this since you most likely want to validate all the other timings as well, and so likely need some kind of mode_valid implementation anyway. Hence to validate modes there's not much benefit of having global min/max values.

https://patchwork.kernel.org/patch/10467155/

I believe you are referring to this patch.

I am primarily interested in the scaling scenario mentioned here. MSM and a few other hardware have scaling block that are used both ways:

1) Where FB limits are larger than the display limits. Scalar blocks are used to
downscale the framebuffers and render within display limits.

In this scenario, with your patch, are you suggesting the drivers maintain the
display limits locally and use those values in fill_modes() / mode_valid() to filter
out invalid modes explicitly instead of mode_config.max_width/height?

Quoted 5 lines

2) Where FB limits are smaller than display limits. Enforced for performance reasons on low tier hardware.
It reduces the fetch bandwidth and uses post blending scalar block to scale up the pixel stream
to match the display resolution.

As Daniel mentioned in that discussion your typical userspace assumes that it can use a single unscaled framebuffer with any advertised mode. Hence I believe limiting the mode list based on the max framebuffer size is pretty much required unless you want to break existing userspace.

In i915 I went a bit further than that recently and now we filter the mode list based on the maximum plane size [1] (which can be less than the max fb size and less than the maximum crtc dimensions). And again that's because userspace assumes that it can just use a single unscaled fullscreen plane to cover the entire crtc.

These assumption are also carved into the legacy setcrtc uapi where you can't even specify multiple framebuffers. In theory a driver could internally use multiple planes to overcome some of the limitations, but in i915 at least we don't.

[1] https://cgit.freedesktop.org/drm/drm-intel/commit/?id=2d20411e25a3bf3d2914a2219f47ed48dc57aed5

Quoted 4 lines

Any suggestions on how this topology can be handled with a single set of max/min values?

Quoted 10 lines

I think a safe way to relax these rules would be to either:
a) Add a client cap by which userspace can inform the kernel it understands there are more complicated limits at play and thus can't assume that everything will just work
b) Maybe we could just tie that in with the atomic cap since atomic clients are pretty much required to do the TEST_ONLY dance anyway, so one might hope they have a working fallback strategy. Though I suspect eg. the modesetting ddx wouldn't like this. But we no longer allow atomic with X anyway so that partcular argument may not hold much weight anymore.

What was the conclusion of the hack to not expose atomic to modesetting ddx, due to the brokenness of it's atomic use? I guess that could also make the modesetting case go away..

BR,
-R
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#6 lore raw

Re: [PATCH] drm: add fb max width/height fields to drm_mode_config

Quoted 88 lines · Rob Clark
Quoted 75 lines · Ville Syrjälä
Quoted 42 lines · Jeykumar Sankaran
Quoted 16 lines · Ville Syrjälä
Quoted 2 lines · Jeykumar Sankaran

The mode_config max width/height values determine the maximum resolution the pixel reader can handle.

Not according to the docs I "fixed" a while ago.

Quoted 9 lines

But the same values are
used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

Quoted 8 lines

If you're going to change the meaning of the old values you need to fix the drivers too.

Personally I don't see too much point in this since you most likely want to validate all the other timings as well, and so likely need some kind of mode_valid implementation anyway. Hence to validate modes there's not much benefit of having global min/max values.

https://patchwork.kernel.org/patch/10467155/

I believe you are referring to this patch.

I am primarily interested in the scaling scenario mentioned here. MSM and a few other hardware have scaling block that are used both ways:

1) Where FB limits are larger than the display limits. Scalar blocks are used to
downscale the framebuffers and render within display limits.

In this scenario, with your patch, are you suggesting the drivers maintain the
display limits locally and use those values in fill_modes() / mode_valid() to filter
out invalid modes explicitly instead of mode_config.max_width/height?

Quoted 5 lines

2) Where FB limits are smaller than display limits. Enforced for performance reasons on low tier hardware.
It reduces the fetch bandwidth and uses post blending scalar block to scale up the pixel stream
to match the display resolution.

As Daniel mentioned in that discussion your typical userspace assumes that it can use a single unscaled framebuffer with any advertised mode. Hence I believe limiting the mode list based on the max framebuffer size is pretty much required unless you want to break existing userspace.

In i915 I went a bit further than that recently and now we filter the mode list based on the maximum plane size [1] (which can be less than the max fb size and less than the maximum crtc dimensions). And again that's because userspace assumes that it can just use a single unscaled fullscreen plane to cover the entire crtc.

These assumption are also carved into the legacy setcrtc uapi where you can't even specify multiple framebuffers. In theory a driver could internally use multiple planes to overcome some of the limitations, but in i915 at least we don't.

[1] https://cgit.freedesktop.org/drm/drm-intel/commit/?id=2d20411e25a3bf3d2914a2219f47ed48dc57aed5

Quoted 4 lines

Any suggestions on how this topology can be handled with a single set of max/min values?

Quoted 10 lines

I think a safe way to relax these rules would be to either:
a) Add a client cap by which userspace can inform the kernel it understands there are more complicated limits at play and thus can't assume that everything will just work
b) Maybe we could just tie that in with the atomic cap since atomic clients are pretty much required to do the TEST_ONLY dance anyway, so one might hope they have a working fallback strategy. Though I suspect eg. the modesetting ddx wouldn't like this. But we no longer allow atomic with X anyway so that partcular argument may not hold much weight anymore.

Quoted 3 lines

What was the conclusion of the hack to not expose atomic to modesetting ddx, due to the brokenness of it's atomic use? I guess that could also make the modesetting case go away..

I thought it went in? Maybe I'm mistaken.

Signature · 2 lines
Ville Syrjälä
Intel
#7 lore raw

Re: [PATCH] drm: add fb max width/height fields to drm_mode_config

Quoted 74 lines · Ville Syrjälä
Quoted 73 lines · Rob Clark
Quoted 71 lines · Ville Syrjälä
Quoted 42 lines · Jeykumar Sankaran
Quoted 16 lines · Ville Syrjälä
Quoted 2 lines · Jeykumar Sankaran

The mode_config max width/height values determine the maximum resolution the pixel reader can handle.

Not according to the docs I "fixed" a while ago.

Quoted 9 lines

But the same values are
used to restrict the size of the framebuffer creation. Hardware's with scaling blocks can operate on framebuffers larger/smaller than that of the pixel reader resolutions by scaling them down/up before rendering.

This changes adds a separate framebuffer max width/height fields in drm_mode_config to allow vendors to set if they are different than that of the default max resolution values.

Quoted 8 lines

If you're going to change the meaning of the old values you need to fix the drivers too.

Personally I don't see too much point in this since you most likely want to validate all the other timings as well, and so likely need some kind of mode_valid implementation anyway. Hence to validate modes there's not much benefit of having global min/max values.

https://patchwork.kernel.org/patch/10467155/

I believe you are referring to this patch.

I am primarily interested in the scaling scenario mentioned here. MSM and a few other hardware have scaling block that are used both ways:

1) Where FB limits are larger than the display limits. Scalar blocks are used to
downscale the framebuffers and render within display limits.

In this scenario, with your patch, are you suggesting the drivers maintain the
display limits locally and use those values in fill_modes() / mode_valid() to filter
out invalid modes explicitly instead of mode_config.max_width/height?

Quoted 5 lines

2) Where FB limits are smaller than display limits. Enforced for performance reasons on low tier hardware.
It reduces the fetch bandwidth and uses post blending scalar block to scale up the pixel stream
to match the display resolution.

As Daniel mentioned in that discussion your typical userspace assumes that it can use a single unscaled framebuffer with any advertised mode. Hence I believe limiting the mode list based on the max framebuffer size is pretty much required unless you want to break existing userspace.

In i915 I went a bit further than that recently and now we filter the mode list based on the maximum plane size [1] (which can be less than the max fb size and less than the maximum crtc dimensions). And again that's because userspace assumes that it can just use a single unscaled fullscreen plane to cover the entire crtc.

These assumption are also carved into the legacy setcrtc uapi where you can't even specify multiple framebuffers. In theory a driver could internally use multiple planes to overcome some of the limitations, but in i915 at least we don't.

[1] https://cgit.freedesktop.org/drm/drm-intel/commit/?id=2d20411e25a3bf3d2914a2219f47ed48dc57aed5

Quoted 1 line
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 3 lines

Any suggestions on how this topology can be handled with a single set of max/min values?

I think a safe way to relax these rules would be to either:
a) Add a client cap by which userspace can inform the kernel it understands there are more complicated limits at play and thus can't assume that everything will just work

+1 on this approach. We already have that for 3d modes, another client cap for "modes bigger than max fb" sounds like a good idea.

For "max plane size" I'm leaning towards drivers should virtualize at least the primary plane if that's needed to scan out the biggest resolution. Since there's way too much userspace which will simply not work otherwise (iirc that's what a bunch of dual-pipe dsi drivers did).

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines

b) Maybe we could just tie that in with the atomic cap since atomic clients are pretty much required to do the TEST_ONLY dance anyway, so one might hope they have a working fallback strategy. Though I suspect eg. the modesetting ddx wouldn't like this. But we no longer allow atomic with X anyway so that partcular argument may not hold much weight anymore.

Quoted 5 lines
Quoted 3 lines

What was the conclusion of the hack to not expose atomic to modesetting ddx, due to the brokenness of it's atomic use? I guess that could also make the modesetting case go away..

I thought it went in? Maybe I'm mistaken.

I did:

commit 26b1d3b527e7bf3e24b814d617866ac5199ce68d

Author: Daniel Vetter <[email protected]>
Date:   Thu Sep 5 20:53:18 2019 +0200

drm/atomic: Take the atomic toys away from X

Cheers, Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel