[Intel-gfx] [PATCH 1/6] drm/i915/selftests: Serialise nop reset with retirement

[Chris Wilson]

Quoting Mika Kuoppala (2019-06-26 14:11:49)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > In order for the reset count to be accurate across our selftest, we need
> > to prevent the background retire worker from modifying our expected
> > state.
> >
> 
> Ok, to summarize the irc discussion we had: The above holds true
> for igt_reset_engine_nop only. As there is no race in
> global reset path...currently.
> 
> But there is intent towards symmetry on both paths
> so it makes sense to keep the tests aligned.
> 
> The commit msg could be enhanced on this regard.
> 
> Also while looking through this, we do increase
> the reset_count rather early before the failpaths.
> even with the resets disabled it gets incremented.
> So now it is more of a attempted reset count.

We don't expect it to fail, and if it does we wedge and report -EIO not
just a boring -EINVAL (or in theory we do -- that's generally the
approach we take elsewhere, treating the GPU going south as a more
severe failure than the test itself failing to fulfil its contract).
-Chris