Bug 103182

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3312/fi-bwr-2160/igt@kms_busy@basic-flip-b.html


(kms_busy:2060) CRITICAL: Test assertion failure function flip_to_fb, file kms_busy.c:132:
(kms_busy:2060) CRITICAL: Failed assertion: nanosleep(&tv, NULL) == -1
(kms_busy:2060) CRITICAL: flip to fb[1] blocked waiting for busy fbSubtest basic-flip-B failed.

Last seen CI_DRM_3605: 2018-01-08 / 222 runs ago

Re-opened for CNL:

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3921/fi-cnl-drrs/igt@kms_busy@basic-flip-a.html

(kms_busy:3606) CRITICAL: Test assertion failure function flip_to_fb, file ../tests/kms_busy.c:132:
(kms_busy:3606) CRITICAL: Failed assertion: nanosleep(&tv, NULL) == -1
(kms_busy:3606) CRITICAL: flip to fb[1] blocked waiting for busy fbSubtest basic-flip-A failed.

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3972/fi-blb-e6850/igt@prime_vgem@basic-fence-flip.html

(prime_vgem:3594) CRITICAL: Test assertion failure function flip_to_vgem, file ../tests/prime_vgem.c:681:
(prime_vgem:3594) CRITICAL: Failed assertion: nanosleep(&tv, NULL) == -1
(prime_vgem:3594) CRITICAL: flip to busy back blocked
Subtest basic-fence-flip failed.

Also seen on BWR: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_8824/fi-bwr-2160/igt@prime_vgem@basic-fence-flip.html

I moved the platform to ALL, because both BLB and CNL saw the problem, which pretty much means all Intel platforms are affected.

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_5144/fi-gdg-551/igt@prime_vgem@basic-fence-flip.html

Starting subtest: basic-fence-flip
(prime_vgem:1355) CRITICAL: Test assertion failure function flip_to_vgem, file ../tests/prime_vgem.c:722:
(prime_vgem:1355) CRITICAL: Failed assertion: nanosleep(&tv, NULL) == -1
(prime_vgem:1355) CRITICAL: flip to busy back blocked
Subtest basic-fence-flip failed.

<5> [280.491987] Asynchronous wait on fence vgem:unbound:1 timed out (hint:intel_atomic_commit_ready+0x0/0x50 [i915])

Test does test use prime buffers, but the assert is caused by drmModePageFlip being blocked or something. I suggest maybe display folks investigate that first why is that the case in the first place?

Assigning to Jani for further dispatching.

JP, any updates here?

(In reply to Lakshmi from comment #9)
> JP, any updates here?

I'm currently looking into this, I just started with it.

I tried reproducing on HSW which is one of the listed platforms but it doesn't reproduce for me. GTG seems to be most likely candidate for reproducing this, I need to look around if I find such old box anywhere to try on.

The impact of this bug is that flipping would be slow which can't be noticed on normal usage.

Setting the priority to Medium.

This trims about 1s off the runtime -- from looking at the gdg results, it was borderline even on success. Restructuring the wait should avoid false positives.


commit 20f8f52498b9c382076cbc85079df44079e5500f (HEAD, upstream/master)
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Sat Mar 30 23:46:36 2019 +0000

    kms_busy: Use igt_waitchildren_timeout()
    
    Replace the convoluted raising of SIGALRM from the child with an
    interruptible sleep in the parent with the equivalent and far more
    natural igt_waitchildren_timeout().
    
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103182
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
    Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

commit f539e21e934019f0196fee646f351b4e30a8c341 (HEAD, upstream/master)
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Apr 1 08:55:20 2019 +0100

    prime_vgem: Replace nanosleep with igt_waitchildren_timeout
    
    We want to use a child in order to detect an uninterruptable sleep (a
    potential bug we might hit), but we can use igt_waitchildren_timeout()
    to replace our risky self-signaling + nanosleep.
    
    v2: Remove the now redundant signal() setup.
    
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103182
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
    Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

(In reply to Chris Wilson from comment #14)
> commit f539e21e934019f0196fee646f351b4e30a8c341 (HEAD, upstream/master)
> Author: Chris Wilson <chris@chris-wilson.co.uk>
> Date:   Mon Apr 1 08:55:20 2019 +0100
> 
>     prime_vgem: Replace nanosleep with igt_waitchildren_timeout
>     
>     We want to use a child in order to detect an uninterruptable sleep (a
>     potential bug we might hit), but we can use igt_waitchildren_timeout()
>     to replace our risky self-signaling + nanosleep.
>     
>     v2: Remove the now redundant signal() setup.
>     
>     Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103182
>     Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
>     Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Looks good, thanks!

The CI Bug Log issue associated to this bug has been archived.

New failures matching the above filters will not be associated to this bug anymore.