Bug 98725

Summary:	[BYT] piglit tests crash due to infrastructure refactor
Product:	piglit	Reporter:	Mark Janes <mark.a.janes>
Component:	infrastructure	Assignee:	Dylan Baker <baker.dylan.c>
Status:	RESOLVED FIXED	QA Contact:	Piglit Mailing List <piglit>
Severity:	normal
Priority:	medium
Version:	unspecified
Hardware:	Other
OS:	All
Whiteboard:
i915 platform:		i915 features:

Description Mark Janes 2016-11-14 17:56:12 UTC

The following tests began to crash on BYT:

piglit.spec.!opengl 1_1.streaming-texture-leak
piglit.spec.!opengl 1_1.max-texture-size
piglit.spec.!opengl 1_2.tex3d-maxsize

bisected to:

908e28423afceee508a33357797703ebbdeb2cc3
tests: Copy profiles to allow them to be run in parallel

3d887afe4762c78f1b280789339c4e73c71bc2dd
framework/programs/run: Only allow --test-list if one profile


This series should not have regressed tests.

Comment 1 Dylan Baker 2016-11-14 20:28:25 UTC

I'm unable to reproduce any changes between the last patch before my series landed and master. The commands and the environments are exactly the same.  Further, the second patch 3e887 doesn't affect the way piglit runs at all (there is in fact a bug in that patch, but it would cause piglit to crash if you hit it). The first (908e28) might have a bug that testing isn't showing and that doesn't show up on my system, but it's concerning to me that this issue doesn't manifest on any other platform (especially other little core platforms like BSW).

I am absolutely certain that the bisection to 3e887 is bogus. Which tests bisect to which commits?

Comment 2 Mark Janes 2016-11-14 20:38:55 UTC

The bisections are probably bogus.  It looks like the tests crash intermittently.

piglit.spec.!opengl 1_1.max-texture-size = piglit 908e28423afceee508a33357797703ebbdeb2cc3
piglit.spec.!opengl 1_1.streaming-texture-leak = piglit 3d887afe4762c78f1b280789339c4e73c71bc2dd
piglit.spec.!opengl 1_2.tex3d-maxsize = piglit 908e28423afceee508a33357797703ebbdeb2cc3

I'll run a few builds to see if I can better isolate where they started failing.

Comment 3 Dylan Baker 2016-11-15 18:20:58 UTC

Okay, after a bit of testing I've determined that the 908e284 patch does increase the memory consumption of piglit at rest. I'm not exactly sure why, but for some reason there are active references to the parent profile. Which might explain why the tests sporadically crash on BYT and only on BYT, since those systems (IIRC) have only 1 Gb of RAM.

I'll do some digging today to see if I can fix it. Otherwise I have some patches I'll send out this week that would make the problem go away, even if they don't actually fix it.

Comment 4 Dylan Baker 2016-11-15 22:44:46 UTC

fixed:
1e5dd0e..1619c37  master -> master

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.