[Intel-gfx] [PATCH] lib/core: Check for kernel error messages and FAIL if any are found

[Chris Wilson]

On Wed, Sep 17, 2014 at 06:13:56PM +0200, Daniel Vetter wrote:
> On Wed, Sep 17, 2014 at 6:01 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > On Wed, Sep 17, 2014 at 05:54:52PM +0200, Daniel Vetter wrote:
> >> On Wed, Sep 17, 2014 at 12:34:46PM +0100, Chris Wilson wrote:
> >> > At the end of a subtest, check for any WARNs or ERRORs (or worse!)
> >> > emitted since the start of our test and FAIL the subtest if any are
> >> > found. This will prevent silent failures due to oops from going amiss or
> >> > being falsely reported as TIMEOUTs.
> >> >
> >> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> >>
> >> We already have this in piglit, including filtering for non-i915 issues
> >> (which especially on s/r tests happen a lot). So this just duplicates
> >> that.
> >
> > What piglit? I don't see QA reports involving pigligt and they seem to
> > mistake kernel OOPSes for benign TIMEOUTs quite frequently.
> 
> Can you please reply with the relevant bugzillas? Since about 2 months
> QA is supposed to be using the piglit runner for their framework, so
> any difference in test results compared to what piglit would report is
> fail.

All reproduction recipes still use the bare test runner, e.g.
https://bugs.freedesktop.org/show_bug.cgi?id=83969 from today. Which is
also a good example for the test missing the kernel warning that
triggered the actual fail. I don't see any problem with having the bare
test runner being able to detect an oops during a subtest. I am pretty
sure there have been timeouts within the last month or so that have been
mutex deadlocks due to a driver oops - but that would require a bit of
digging to confirm.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre