Bug 45816 - [IVB]SPECviewperf 11 cause GPU hung
Summary: [IVB]SPECviewperf 11 cause GPU hung
Status: CLOSED WONTFIX
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: git
Hardware: x86 (IA32) Linux (All)
: medium critical
Assignee: Kenneth Graunke
QA Contact: Intel 3D Bugs Mailing List
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2012-02-09 00:49 UTC by Ouping Zhang
Modified: 2014-12-20 00:36 UTC (History)
3 users (show)

See Also:
i915 platform:
i915 features:


Attachments
dmesg (59.11 KB, text/plain)
2012-02-09 00:49 UTC, Ouping Zhang
Details
apitrace info (28.00 KB, text/plain)
2012-03-02 00:30 UTC, Ouping Zhang
Details

Description Ouping Zhang 2012-02-09 00:49:28 UTC
Created attachment 56795 [details]
dmesg

System Environment:
--------------------------
Libdrm:         (master)2.4.31
Mesa:           (master)7c857a6b159debf76d4661f494fd2c97d205b5b1
Xserver:                (master)xorg-server-1.11.99.902-3-g85cecd981191f9c3dab0fb13310d91eff643d423
Kernel: (drm-intel-testing)396ee23db6b76354e487fd2cfdacbd989442f81d

Running SPECviewperf 11 on x-ivb13(32bit) cause GPU hung, and the issue didn't happen on SNB. It seems that when runing the case maya-03 cause GPU hung. I have attached the dmesg info.
Comment 1 Gordon Jin 2012-02-09 23:50:39 UTC
note: this is the first we run SPECviewperf 11.
Comment 2 Gordon Jin 2012-02-26 19:12:07 UTC
Ken, can you reproduce this? This is the only IVB hang we observe now.
Comment 3 Kenneth Graunke 2012-02-26 21:34:07 UTC
I'll try and take a look this week.
Comment 4 Kenneth Graunke 2012-02-28 16:38:40 UTC
I just ran the whole SPECviewperf 11 suite on my IVB machine without any GPU hangs.  Mesa is ab79d2be2e12, kernel is 5ffca28.

How hard would it be to capture an apitrace that reproduces the hang?
Comment 5 Ouping Zhang 2012-03-02 00:30:23 UTC
Created attachment 57897 [details]
apitrace info

apitrace
Comment 6 Kenneth Graunke 2012-03-02 11:47:27 UTC
I need the actual .trace file, not the ASCII dump/log.
Comment 7 Ouping Zhang 2012-03-08 01:12:27 UTC
when only running maya-03, it caused GPU hung, but when running apitrace trace maya-03, there was any hung.
when only running proe-05, there was any huang, but when running apitrace trace proe-05, it caused GPU hung.
   
(In reply to comment #6)
> I need the actual .trace file, not the ASCII dump/log.
Comment 8 Ouping Zhang 2012-03-08 01:25:25 UTC
9.3G Mar  8 08:55 viewperf.trace
The apitrace file is so big, I can't attach it.
(In reply to comment #6)
> I need the actual .trace file, not the ASCII dump/log.
Comment 9 Kenneth Graunke 2012-03-08 08:35:56 UTC
Yeah, that's not too surprising.  Could you upload it somewhere (e.g. tinderbox.sh.intel.com) and mail me a link?
Comment 10 Ouping Zhang 2012-03-09 00:05:36 UTC
you can download the apitrace file"viewperf.trace.tar.gz" from http://tinderbox.sh.intel.com/games/
(In reply to comment #9)
> Yeah, that's not too surprising.  Could you upload it somewhere (e.g.
> tinderbox.sh.intel.com) and mail me a link?
Comment 11 Kenneth Graunke 2012-03-26 16:10:47 UTC
I just replayed your trace...it worked fine.  No GPU hangs.
Comment 12 Ouping Zhang 2012-03-31 19:05:44 UTC
SPECviewperf11 have 8 testcases: catia-03 ensight-04 lightwave-01 maya-03 proe-05 snx-01 sw-02 tcvis-02.
when running SPECviewperf11, maya-03 caused GPU hang, but when running apitrace trace SPECviewperf11, maya-03 didn't cause GPU hang. but it seems when runinng proe-05, there was something error with the following info:
apitrace: redirecting dlopen("libGL.so.1", 0x102)
Writing PNG file '../../viewperf/viewperf11.0/results/proe-05/grabs/test1full.png'...done.
the process got stuck in proe-05, and the apitrace file of proe-05 is 93G.....untill the test machine is crash.
why is the apitrace file of proe-05 so big?

(In reply to comment #11)
> I just replayed your trace...it worked fine.  No GPU hangs.
Comment 13 Ben Widawsky 2014-12-19 19:26:47 UTC
Can you please try to reproduce this with latest software and attach the error state if it still GPU hangs?
Comment 14 Gordon Jin 2014-12-20 00:36:16 UTC
It's too old and we've removed it from testing, as it looks like no one cared about this and it takes big effort to run. I suggest closing this.


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.