DG2: '[drm] *ERROR* Fault errors on pipe A: 0x00000800' errors when moving the cursor

How to reproduce:

• start graphical interface
• move the mouse around
• [drm] *ERROR* Fault errors on pipe A: 0x00000800 errors printed in dmesg

This is a regression as a few weeks ago this was not reproducible. Whole dmesg: dmesg.txt

added BUG label

Error = GEN9_PIPE_CURSOR_FAULT, is cursor programmed correctly?

Seems to be i915 bugs too, do you get the bug with same intensity if you try with i915?

Not able to reproduce with i915 with the same drm-xe-next kernel and same DG2.

Possibly related i915 bug, drm/intel#2017

@zehortigoza You mentioned that it is a regression. Can you tell us the version for which you were not seeing the issue?

I don't have a working commit hash, just said that it worked fine in the past.

kms_cursor_legacy shows the errors too. Should make it easier to track down. Even single-move hits it.

My guess is that it worked before the GGTT invalidation commits were reversed.

I am trying to bisect the issue. Is it Possible to share the hash for these patches?

No need to bisect, but if you want a hash of what should be a working commit: commit 28247b0d (gitlab/initial-plane-fixups) Author: Matthew Brost matthew.brost@intel.com Date: Wed Feb 1 11:26:11 2023 -0800

drm/xe: Pad GGTT mapping with an extra page pointing to scratch

The revert appears to be lost somewhere in a squash.

https://patchwork.freedesktop.org/series/119009/ should probably fix it.

Also needs a null check in xe/display/xe_fb_pin.c intel_plane_unpin_fb to work.

The whole patch would be like this?

diff --git a/drivers/gpu/drm/i915/display/intel_cursor.c b/drivers/gpu/drm/i915/display/intel_cursor.c
index 8dd9dac3389c5..a139b4e645ba1 100644
--- a/drivers/gpu/drm/i915/display/intel_cursor.c
+++ b/drivers/gpu/drm/i915/display/intel_cursor.c
@@ -688,9 +688,8 @@ intel_legacy_cursor_update(struct drm_plane *_plane,
        if (ret)
                goto out_free;

-       ret = intel_plane_pin_fb(new_plane_state);
-       if (ret)
-               goto out_free;
+       /* magic trick! */
+       swap(new_plane_state->ggtt_vma, old_plane_state->ggtt_vma);

        intel_frontbuffer_flush(to_intel_frontbuffer(new_plane_state->hw.fb),
                                ORIGIN_CURSOR_UPDATE);
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
index 83c9245dd3624..f17ce7f8758c1 100644
--- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -291,6 +291,8 @@ int intel_plane_pin_fb(struct intel_plane_state *plane_state)

 void intel_plane_unpin_fb(struct intel_plane_state *old_plane_state)
 {
-       __xe_unpin_fb_vma(old_plane_state->ggtt_vma);
-       old_plane_state->ggtt_vma = NULL;
+       if (old_plane_state->ggtt_vma) {
+               __xe_unpin_fb_vma(old_plane_state->ggtt_vma);
+               old_plane_state->ggtt_vma = NULL;
+       }
 }

This fixed the dmesg warnings, can't verify if it is visually fixed as I don't have a local DG2. @pzanoni can you please verify if it fixes the visual artifacts?

@mlankhorst I was using two tests to reproduce the error - igt@kms_cursor_legacy@cursor-vs-flip-atomic-transitions and igt@kms_cursor_crc@cursor-dpms. With the patch above the legacy path seems to be fixed and I don't see the error anymore. But the issue is still seen with igt@kms_cursor_crc@cursor-dpms. Anything we can do for the atomic path?

Latest version is https://patchwork.freedesktop.org/series/119676/ still waiting review. But it don't do anything in intel_plane_unpin_fb() and has one additional patch. Will give a try tomorrow on RIL.

Tested, it is missing the intel_plane_unpin_fb() change block. With that it still fixes the dmesg warnings. Can't verify visually as I only have remote access to DG2.

mentioned in issue #296 (closed)

mentioned in issue #293 (closed)

It seems the series is given issues with device busy errors. @mlankhorst, @zehortigoza Can you please check once. @suresh_kurmi FYI

I don't get any busy errors while running Xorg + Mesa tests with it.

Hi @zehortigoza If you refer the CI results, some of the tests like below give error: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_120018v1/shard-dg2-11/igt@kms_async_flips@test-cursor@pipe-b-dp-4.html

More details in results here: https://patchwork.freedesktop.org/series/120018/

This series is missing this: https://github.com/zehortigoza/linux/commit/5da8fee9df1d2302c98a4048b83a1ec69e708443

But without this it would crash, not get busy errors...

The tests that I run don't stress async flips, could be that.

Yeah seems like that, need a closer look to narrow down the issue here.

@mlankhorst @uma_mnnit any updates here? this bug has a huge impact in user experience in DG2.

Still reproduced with latest drm-xe-next.

mentioned in issue #325 (closed)

The CI Bug Log issue associated to this bug has been updated by adelaryb.

New filters associated

• DG2: igt@kms_cursor_.* - dmesg-warn/incomplete - ERROR Fault errors on pipe .* (No new failures associated)

mentioned in merge request mesa/mesa!23045 (merged)

mentioned in issue #509 (closed)

A CI Bug Log filter associated to this bug has been updated by tillipix.

Description: DG2: igAll t@kmes_curtsor_.* - dmesg-warn/dmesg-fail/incomplete - *ERROR* Fault errors on pipe .*

Equivalent query: runconfig_tag IS IN ["xe"] AND machine_tag IS IN ["DG2"] AND ((testsuite_name = "IGT" AND status_name IS IN ["incomplete", "skip", "dmesg-warn", "dmesg-fail"])) AND dmesg ~= '\*ERROR\* Fault errors on pipe .*'

No new failures caught with the new filter

The CI Bug Log issue associated to this bug has been updated by tillipix.

Removed filters

• DG2: Few tests - dmesg-warn/incomplete/skip - ERROR Fault errors on pipe .* (added on 1 week, 1 day ago)

A CI Bug Log filter associated to this bug has been updated by tillipix.

Description: ADL_P DG2: All tests - dmesg-warn/dmesg-fail/incomplete - *ERROR* Fault errors on pipe .*: 0x00000080

Equivalent query: runconfig_tag IS IN ["xe"] AND machine_tag IS IN ["DG2", "ADL-P"] AND ((testsuite_name = "IGT" AND status_name IS IN ["incomplete", "dmesg-warn", "dmesg-fail"])) AND dmesg ~= '\*ERROR\* Fault errors on pipe .*: 0x00000080'

No new failures caught with the new filter

A CI Bug Log filter associated to this bug has been updated by tillipix.

Description: ADL_P DG2: All tests - dmesg-warn/dmesg-fail/incomplete - *ERROR* Fault errors on pipe .*: 0x000000800

Equivalent query: runconfig_tag IS IN ["xe"] AND machine_tag IS IN ["DG2", "ADL-P"] AND ((testsuite_name = "IGT" AND status_name IS IN ["incomplete", "dmesg-warn", "dmesg-fail"])) AND dmesg ~= '\*ERROR\* Fault errors on pipe .*: 0x000000800'

No new failures caught with the new filter