[Bug 112198] New: GPU hang when decoding MJPEGs via VA-API

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=112198

            Bug ID: 112198
           Summary: GPU hang when decoding MJPEGs via VA-API
           Product: DRI
           Version: unspecified
          Hardware: x86-64 (AMD64)
                OS: Linux (All)
            Status: NEW
          Severity: normal
          Priority: not set
         Component: DRM/Intel
          Assignee: intel-gfx-bugs at lists.freedesktop.org
          Reporter: sgunderson at bigfoot.com
        QA Contact: intel-gfx-bugs at lists.freedesktop.org
                CC: intel-gfx-bugs at lists.freedesktop.org

Created attachment 145878
  --> https://bugs.freedesktop.org/attachment.cgi?id=145878&action=edit
Dump of /sys/class/drm/card0/error

Hi,

I've got a desktop and a laptop (both Haswell) that use VA-API to decode JPEGs
as part of a larger video pipeline. This used to work fine around March-April
or so, but recently (November), I've started seeing GPU hangs on both. The
kernel says:

[515715.657023] DMAR: DRHD: handling fault status reg 3
[515715.657030] DMAR: [DMA Write] Request device [00:02.0] fault addr f72e5000
[fault reason 05] PTE Write access is not set
[515789.233234] DMAR: DRHD: handling fault status reg 3
[515789.233240] DMAR: [DMA Write] Request device [00:02.0] fault addr ed5df000
[fault reason 05] PTE Write access is not set
[515809.358568] i915 0000:00:02.0: Resetting chip for hang on rcs0
[515817.358404] i915 0000:00:02.0: Resetting chip for hang on rcs0
[515825.358428] i915 0000:00:02.0: Resetting chip for hang on rcs0

I rebooted with intel_iommu=igfx_off, and ran the program again. After ~10
minutes of running, it hung again, with:

[  792.028358] i915 0000:00:02.0: GPU HANG: ecode 7:1:0xfffffffe, in futatabi
[2319], hang on rcs0
[  792.028361] [drm] GPU hangs can indicate a bug anywhere in the entire gfx
stack, including userspace.
[  792.028361] [drm] Please file a _new_ bug report on bugs.freedesktop.org
against DRI -> DRM/Intel
[  792.028362] [drm] drm/i915 developers can then reassign to the right
component if it's not a kernel issue.
[  792.028363] [drm] The gpu crash dump is required to analyze gpu hangs, so
please always attach it.
[  792.028364] [drm] GPU crash dump saved to /sys/class/drm/card0/error
[  792.028654] i915 0000:00:02.0: Resetting chip for hang on rcs0
[  799.996625] i915 0000:00:02.0: Resetting chip for hang on rcs0
[  807.996604] i915 0000:00:02.0: Resetting chip for hang on rcs0

If it's interesting, you can find the decoder source at
https://git.sesse.net/?p=nageru;a=blob;f=futatabi/vaapi_jpeg_decoder.cpp;h=d18a8735c11a23853ea6109b340c031dfee2a19c;hb=HEAD
.
I'll be filing a copy of /sys/class/drm/card0/error as an attachment.

-- 
You are receiving this mail because:
You are the assignee for the bug.
You are on the CC list for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20191102/36230899/attachment.html>