diff mbox series

[RFC] drm/amd/display: fix bandwidth validation failure on DCN 2.1

Message ID 20231229163821.144599-1-mwen@igalia.com (mailing list archive)
State New, archived
Headers show
Series [RFC] drm/amd/display: fix bandwidth validation failure on DCN 2.1 | expand

Commit Message

Melissa Wen Dec. 29, 2023, 4:25 p.m. UTC
IGT `amdgpu/amd_color/crtc-lut-accuracy` fails right at the beginning of
the test execution, during atomic check, because DC rejects the
bandwidth state for a fb sizing 64x64. The test was previously working
with the deprecated dc_commit_state(). Now using
dc_validate_with_context() approach, the atomic check needs to perform a
full state validation. Therefore, set fast_validation to false in the
dc_validate_global_state call for atomic check.

Fixes: b8272241ff9d ("drm/amd/display: Drop dc_commit_state in favor of dc_commit_streams")
Signed-off-by: Melissa Wen <mwen@igalia.com>
---

Hi,

It's a long story. I was inspecting this bug report:
- https://gitlab.freedesktop.org/drm/amd/-/issues/2016
and noticed the IGT test `igt@amdgpu/amd_color@crtc-lut-accuracy`
mentioned there wasn't even being executed on a laptop with DCN 2.1
(HP HP ENVY x360 Convertible 13-ay1xxx/8929). The test fails right at
the beginning due to an atomic check rejection, as below:

Starting subtest: crtc-lut-accuracy
(amd_color:14772) igt_kms-CRITICAL: Test assertion failure function igt_display_commit_atomic, file ../lib/igt_kms.c:4530:
(amd_color:14772) igt_kms-CRITICAL: Failed assertion: ret == 0
(amd_color:14772) igt_kms-CRITICAL: Last errno: 22, Invalid argument
(amd_color:14772) igt_kms-CRITICAL: error: -22 != 0
Stack trace:
  #0 ../lib/igt_core.c:1989 __igt_fail_assert()
  #1 [igt_display_commit_atomic+0x44]
  #2 ../tests/amdgpu/amd_color.c:159 __igt_unique____real_main395()
  #3 ../tests/amdgpu/amd_color.c:395 main()
  #4 ../sysdeps/nptl/libc_start_call_main.h:74 __libc_start_call_main()
  #5 ../csu/libc-start.c:128 __libc_start_main@@GLIBC_2.34()
  #6 [_start+0x21]
Subtest crtc-lut-accuracy failed.

Checking dmesg, we can see that a bandwidth validation failure causes
the atomic check rejection:

[  711.147663] amdgpu 0000:04:00.0: [drm] Mode Validation Warning: Unknown Status failed validation.
[  711.147667] [drm:amdgpu_dm_atomic_check [amdgpu]] DC global validation failure: Bandwidth validation failure (BW and Watermark) (13)
[  711.147772] [drm:amdgpu_irq_dispatch [amdgpu]] Unregistered interrupt src_id: 243 of client_id:8
[  711.148033] [drm:amdgpu_dm_atomic_check [amdgpu]] Atomic check failed with err: -22

I also noticed that the atomic check doesn't fail if I change the fb
width and height used in the test from 64 to 66, and I can get the test
execution back (and with success). However, I recall that all test cases
of IGT `amd_color` were working in the past, so I bisected and found the
first bad commit:

b8272241ff9d drm/amd/display: Drop dc_commit_state in favor of dc_commit_streams

Bringing the `dc_commit_state` machinery back also prevents the
bandwidth validation failure, but the commit above says
dc_commit_streams validation is more complete than dc_commit_state, so I
discarded this approach.

After some debugging and code inspection, I found out that avoiding fast
validation on dc_validate_global_state during atomic check solves the
issue, but I'm not sure if this change may affect performance. I
compared exec time of some IGT tests and didn't see any differences, but
I recognize it doesn't provide enough evidence.

What do you think about this change? Should I examine other things? Do
you see any potential issue that I should investigate? Could you
recommend a better approach to assess any side-effect of not enabling
fast validation in the atomic check?

Please, let me know your thoughts.

Happy New Year!

Melissa

 drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Hamza Mahfooz Jan. 3, 2024, 2:44 p.m. UTC | #1
On 12/29/23 11:25, Melissa Wen wrote:
> IGT `amdgpu/amd_color/crtc-lut-accuracy` fails right at the beginning of
> the test execution, during atomic check, because DC rejects the
> bandwidth state for a fb sizing 64x64. The test was previously working
> with the deprecated dc_commit_state(). Now using
> dc_validate_with_context() approach, the atomic check needs to perform a
> full state validation. Therefore, set fast_validation to false in the
> dc_validate_global_state call for atomic check.
> 
> Fixes: b8272241ff9d ("drm/amd/display: Drop dc_commit_state in favor of dc_commit_streams")
> Signed-off-by: Melissa Wen <mwen@igalia.com>
> ---
> 
> Hi,
> 
> It's a long story. I was inspecting this bug report:
> - https://gitlab.freedesktop.org/drm/amd/-/issues/2016
> and noticed the IGT test `igt@amdgpu/amd_color@crtc-lut-accuracy`
> mentioned there wasn't even being executed on a laptop with DCN 2.1
> (HP HP ENVY x360 Convertible 13-ay1xxx/8929). The test fails right at
> the beginning due to an atomic check rejection, as below:
> 
> Starting subtest: crtc-lut-accuracy
> (amd_color:14772) igt_kms-CRITICAL: Test assertion failure function igt_display_commit_atomic, file ../lib/igt_kms.c:4530:
> (amd_color:14772) igt_kms-CRITICAL: Failed assertion: ret == 0
> (amd_color:14772) igt_kms-CRITICAL: Last errno: 22, Invalid argument
> (amd_color:14772) igt_kms-CRITICAL: error: -22 != 0
> Stack trace:
>    #0 ../lib/igt_core.c:1989 __igt_fail_assert()
>    #1 [igt_display_commit_atomic+0x44]
>    #2 ../tests/amdgpu/amd_color.c:159 __igt_unique____real_main395()
>    #3 ../tests/amdgpu/amd_color.c:395 main()
>    #4 ../sysdeps/nptl/libc_start_call_main.h:74 __libc_start_call_main()
>    #5 ../csu/libc-start.c:128 __libc_start_main@@GLIBC_2.34()
>    #6 [_start+0x21]
> Subtest crtc-lut-accuracy failed.
> 
> Checking dmesg, we can see that a bandwidth validation failure causes
> the atomic check rejection:
> 
> [  711.147663] amdgpu 0000:04:00.0: [drm] Mode Validation Warning: Unknown Status failed validation.
> [  711.147667] [drm:amdgpu_dm_atomic_check [amdgpu]] DC global validation failure: Bandwidth validation failure (BW and Watermark) (13)
> [  711.147772] [drm:amdgpu_irq_dispatch [amdgpu]] Unregistered interrupt src_id: 243 of client_id:8
> [  711.148033] [drm:amdgpu_dm_atomic_check [amdgpu]] Atomic check failed with err: -22
> 
> I also noticed that the atomic check doesn't fail if I change the fb
> width and height used in the test from 64 to 66, and I can get the test
> execution back (and with success). However, I recall that all test cases
> of IGT `amd_color` were working in the past, so I bisected and found the
> first bad commit:
> 
> b8272241ff9d drm/amd/display: Drop dc_commit_state in favor of dc_commit_streams
> 
> Bringing the `dc_commit_state` machinery back also prevents the
> bandwidth validation failure, but the commit above says
> dc_commit_streams validation is more complete than dc_commit_state, so I
> discarded this approach.
> 
> After some debugging and code inspection, I found out that avoiding fast
> validation on dc_validate_global_state during atomic check solves the
> issue, but I'm not sure if this change may affect performance. I
> compared exec time of some IGT tests and didn't see any differences, but
> I recognize it doesn't provide enough evidence.
> 
> What do you think about this change? Should I examine other things? Do
> you see any potential issue that I should investigate? Could you
> recommend a better approach to assess any side-effect of not enabling
> fast validation in the atomic check?
> 
> Please, let me know your thoughts.

We shouldn't be doing fast updates when lock_and_validation_needed is
true, so this seems to be correct.

Which is to say, applied, thanks!

Cc: stable@vger.kernel.org

> 
> Happy New Year!
> 
> Melissa
> 
>   drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> index 2845c884398e..4f51a7ad7a3c 100644
> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> @@ -10745,7 +10745,7 @@ static int amdgpu_dm_atomic_check(struct drm_device *dev,
>   			DRM_DEBUG_DRIVER("drm_dp_mst_atomic_check() failed\n");
>   			goto fail;
>   		}
> -		status = dc_validate_global_state(dc, dm_state->context, true);
> +		status = dc_validate_global_state(dc, dm_state->context, false);
>   		if (status != DC_OK) {
>   			DRM_DEBUG_DRIVER("DC global validation failure: %s (%d)",
>   				       dc_status_to_str(status), status);
Greg Kroah-Hartman Jan. 3, 2024, 3:41 p.m. UTC | #2
On Wed, Jan 03, 2024 at 09:44:18AM -0500, Hamza Mahfooz wrote:
> On 12/29/23 11:25, Melissa Wen wrote:
> > IGT `amdgpu/amd_color/crtc-lut-accuracy` fails right at the beginning of
> > the test execution, during atomic check, because DC rejects the
> > bandwidth state for a fb sizing 64x64. The test was previously working
> > with the deprecated dc_commit_state(). Now using
> > dc_validate_with_context() approach, the atomic check needs to perform a
> > full state validation. Therefore, set fast_validation to false in the
> > dc_validate_global_state call for atomic check.
> > 
> > Fixes: b8272241ff9d ("drm/amd/display: Drop dc_commit_state in favor of dc_commit_streams")
> > Signed-off-by: Melissa Wen <mwen@igalia.com>
> > ---
> > 
> > Hi,
> > 
> > It's a long story. I was inspecting this bug report:
> > - https://gitlab.freedesktop.org/drm/amd/-/issues/2016
> > and noticed the IGT test `igt@amdgpu/amd_color@crtc-lut-accuracy`
> > mentioned there wasn't even being executed on a laptop with DCN 2.1
> > (HP HP ENVY x360 Convertible 13-ay1xxx/8929). The test fails right at
> > the beginning due to an atomic check rejection, as below:
> > 
> > Starting subtest: crtc-lut-accuracy
> > (amd_color:14772) igt_kms-CRITICAL: Test assertion failure function igt_display_commit_atomic, file ../lib/igt_kms.c:4530:
> > (amd_color:14772) igt_kms-CRITICAL: Failed assertion: ret == 0
> > (amd_color:14772) igt_kms-CRITICAL: Last errno: 22, Invalid argument
> > (amd_color:14772) igt_kms-CRITICAL: error: -22 != 0
> > Stack trace:
> >    #0 ../lib/igt_core.c:1989 __igt_fail_assert()
> >    #1 [igt_display_commit_atomic+0x44]
> >    #2 ../tests/amdgpu/amd_color.c:159 __igt_unique____real_main395()
> >    #3 ../tests/amdgpu/amd_color.c:395 main()
> >    #4 ../sysdeps/nptl/libc_start_call_main.h:74 __libc_start_call_main()
> >    #5 ../csu/libc-start.c:128 __libc_start_main@@GLIBC_2.34()
> >    #6 [_start+0x21]
> > Subtest crtc-lut-accuracy failed.
> > 
> > Checking dmesg, we can see that a bandwidth validation failure causes
> > the atomic check rejection:
> > 
> > [  711.147663] amdgpu 0000:04:00.0: [drm] Mode Validation Warning: Unknown Status failed validation.
> > [  711.147667] [drm:amdgpu_dm_atomic_check [amdgpu]] DC global validation failure: Bandwidth validation failure (BW and Watermark) (13)
> > [  711.147772] [drm:amdgpu_irq_dispatch [amdgpu]] Unregistered interrupt src_id: 243 of client_id:8
> > [  711.148033] [drm:amdgpu_dm_atomic_check [amdgpu]] Atomic check failed with err: -22
> > 
> > I also noticed that the atomic check doesn't fail if I change the fb
> > width and height used in the test from 64 to 66, and I can get the test
> > execution back (and with success). However, I recall that all test cases
> > of IGT `amd_color` were working in the past, so I bisected and found the
> > first bad commit:
> > 
> > b8272241ff9d drm/amd/display: Drop dc_commit_state in favor of dc_commit_streams
> > 
> > Bringing the `dc_commit_state` machinery back also prevents the
> > bandwidth validation failure, but the commit above says
> > dc_commit_streams validation is more complete than dc_commit_state, so I
> > discarded this approach.
> > 
> > After some debugging and code inspection, I found out that avoiding fast
> > validation on dc_validate_global_state during atomic check solves the
> > issue, but I'm not sure if this change may affect performance. I
> > compared exec time of some IGT tests and didn't see any differences, but
> > I recognize it doesn't provide enough evidence.
> > 
> > What do you think about this change? Should I examine other things? Do
> > you see any potential issue that I should investigate? Could you
> > recommend a better approach to assess any side-effect of not enabling
> > fast validation in the atomic check?
> > 
> > Please, let me know your thoughts.
> 
> We shouldn't be doing fast updates when lock_and_validation_needed is
> true, so this seems to be correct.
> 
> Which is to say, applied, thanks!
> 
> Cc: stable@vger.kernel.org

<formletter>

This is not the correct way to submit patches for inclusion in the
stable kernel tree.  Please read:
    https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 2845c884398e..4f51a7ad7a3c 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -10745,7 +10745,7 @@  static int amdgpu_dm_atomic_check(struct drm_device *dev,
 			DRM_DEBUG_DRIVER("drm_dp_mst_atomic_check() failed\n");
 			goto fail;
 		}
-		status = dc_validate_global_state(dc, dm_state->context, true);
+		status = dc_validate_global_state(dc, dm_state->context, false);
 		if (status != DC_OK) {
 			DRM_DEBUG_DRIVER("DC global validation failure: %s (%d)",
 				       dc_status_to_str(status), status);