Bug 110131

(In reply to Eero Tamminen from comment #0)
> No idea whether these are related to compute hangs bug 108820, or Heaven
> hang bug 103556.

There were Heaven hangs during weekend, but no tessellation test hangs.  If I don't happen to notice more of these by end of the month, I'll close this as WORKSFORME.

(In reply to Eero Tamminen from comment #0)
> Created attachment 143679 [details]
> SKL GT4e SynMark OglTerrainFlyTess hang error state
> 
> Setup:
> * Ubuntu 18.04
> * v5.0+ drm-tip kernel & git version of Xserver
> * Mesa git version
> 
> In last few days I've seen couple of random GPU hangs in tessellation
> related tests:
> 
> - on SKL GT4e, recoverable one once in SynMark2 v7 OglTerrainFlyTess, and
> once in GfxBench v5 GL Aztec Ruins normal (does also lot of other things
> besides tessellation)
> - one system hang in GfxBench tessellation test on KBL GT2 day before
> 
> It's possible that first item is related to starting to use Weston/XWayland
> instead of normal X:
> ----------------------------------------------------
> [ 8231.866172] i915 0000:00:02.0: GPU HANG: ecode 9:1:0xfffffffe, in  [0],
> hang on rcs0
> [ 8231.866174] [drm] GPU hangs can indicate a bug anywhere in the entire gfx
> stack, including userspace.
> [ 8231.866174] [drm] Please file a _new_ bug report on bugs.freedesktop.org
> against DRI -> DRM/Intel
> [ 8231.866175] [drm] drm/i915 developers can then reassign to the right
> component if it's not a kernel issue.
> [ 8231.866175] [drm] The gpu crash dump is required to analyze gpu hangs, so
> please always attach it.
> [ 8231.866175] [drm] GPU crash dump saved to /sys/class/drm/card0/error
> [ 8231.867183] i915 0000:00:02.0: Resetting rcs0 for hang on rcs0
> [ 8239.858844] i915 0000:00:02.0: Resetting rcs0 for hang on rcs0
> [ 8243.313841] Asynchronous wait on fence i915:weston[643]/1:5eb9e timed out
> (hint:intel_atomic_commit_ready+0x0/0x54 [i915])
> [ 8247.858844] i915 0000:00:02.0: Resetting rcs0 for hang on rcs0
> ----------------------------------------------------
> 
> See attachement for error state.
> 
> (Another possibility could be that those started (to be more visible?) after
> the "intel/nir: Vectorize all IO" fix to bug 107510, as that improved
> tessellation tests.)
> 
> Note: I don't actively read dmesg output, so I may have missed most of the
> recoverable GPU hangs unless they've been serious enough to hang the system,
> fail the test, or at least slow it down enough to significantly impact
> performance.  I'll add some better tracking for that.
> 
> No idea whether these are related to compute hangs bug 108820, or Heaven
> hang bug 103556.

This error state has consistent HS/TE/DS stage programming so it would seem to be a different issue from the unigine bug.

(In reply to Lionel Landwerlin from comment #2)
> This error state has consistent HS/TE/DS stage programming so it would seem
> to be a different issue from the unigine bug.

Thanks!

I have now tracking for GPU resets, and I haven't seen any tessellation test hangs since I filed this bug (only bug 108820 & bug 103556 hangs).  If they don't appear by next week, I'll close this as WORKSFORME.

(In reply to Eero Tamminen from comment #3)
> I have now tracking for GPU resets, and I haven't seen any tessellation test
> hangs since I filed this bug (only bug 108820 & bug 103556 hangs).  If they
> don't appear by next week, I'll close this as WORKSFORME.

-> WORKSFORME.

I'm seeing (recoverable) GEN9+ GPU hangs only in Manhattan 3.1, CarChase and AztecRuins (and all of those could be compute issues, bug 108820), not in the tests listed in this bug.