Bug 70598

Summary: [PNV/945]igt/gen3_mixed_blits randomly causes [drm:i915_report_and_clear_eir] *ERROR* EIR stuck: 0x00000010, masking
Product: DRI Reporter: lu hua <huax.lu>
Component: DRM/IntelAssignee: Intel GFX Bugs mailing list <intel-gfx-bugs>
Status: CLOSED WONTFIX QA Contact: Intel GFX Bugs mailing list <intel-gfx-bugs>
Severity: normal    
Priority: medium CC: intel-gfx-bugs
Version: unspecified   
Hardware: All   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
dmesg
none
error informaiton none

Description lu hua 2013-10-18 06:41:15 UTC
Created attachment 87813 [details]
dmesg

System Environment:
--------------------------
Platform:  Pineview/945
Kernel:   (drm-intel-fixes)cd66407810869709e0d6764409137d48c6812cfc

Bug detailed description:
-----------------------------
Run ./gen3_mixed_blits, [drm:i915_report_and_clear_eir] *ERROR* EIR stuck: 0x00000010, masking randomly appears in dmesg.It happens 1 in 5 runs.
It happens on pineview and 945 with -fixes kernel.

output:
Using 384 1MiB buffers
Verifying initialisation...done
Cyclic blits, forward...verifying...done
Cyclic blits, backward...verifying...done
Random blits...verifying...done

Reproduce steps:
----------------------------
1../gen3_mixed_blits
Comment 1 Chris Wilson 2013-10-18 12:20:18 UTC
Can you also attach the corresponding error-state?
Comment 2 Chris Wilson 2013-10-18 12:22:30 UTC
Display A: invalid tiling.

Hopefully, the error-state will shed more light on the issue.
Comment 3 lu hua 2013-10-21 06:59:22 UTC
Created attachment 87909 [details]
error informaiton
Comment 4 Chris Wilson 2013-10-21 08:31:11 UTC
The display state seems valid - the block that faulted is pointing to an uncached buffer, the scanout is not being written to by the render engine and there is not even a nearby Y-fence.

I think this is a WONTFIX - but lets see if we can understand the trigger a bit better first. Does this appear to a recent regression? Would it be possible to bisect?
Comment 5 lu hua 2013-10-22 07:42:27 UTC
(In reply to comment #4)
> The display state seems valid - the block that faulted is pointing to an
> uncached buffer, the scanout is not being written to by the render engine
> and there is not even a nearby Y-fence.
> 
> I think this is a WONTFIX - but lets see if we can understand the trigger a
> bit better first. Does this appear to a recent regression? Would it be
> possible to bisect?

It is not a recent regression.
Test on earlier commit, It also has this issue.
Test commit:8abbbaf6adb46157b6bd416f7616b555cc6a332f.
commit 8abbbaf6adb46157b6bd416f7616b555cc6a332f
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Mar 27 21:42:00 2013 +0100

    drm: don't unlock in the addfb error paths

    We don't grab the modeset locks any more since
Comment 6 Chris Wilson 2013-10-22 08:02:42 UTC
Unless someone has a bright idea, I am afraid that we need to consign this to the annals of history.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.