[Intel-gfx] [CI 3/5] drm/i915: Disable semaphores on Sandybridge

[Chris Wilson]

Quoting Jani Nikula (2017-11-21 10:23:44)
> On Tue, 21 Nov 2017, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > Quoting Jani Nikula (2017-11-21 09:57:30)
> >> On Mon, 20 Nov 2017, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> >> > I should have admitted defeat long ago as there has been a rare but
> >> > persistent error on Sandybridge where semaphore signaling did not
> >> > propagate to the waiter, leading to a GPU hang.
> >> 
> >> I'm late to the party, but I think the choice of patch ordering here is
> >> a bit unfortunate. Or is it intentional that you can't backport this to
> >> stable as-is?
> >
> > The unmentioned problem with this patch, is that we were using
> > semaphores to mask the missed-breadcrumb-syndrome on Sandybridge (and
> > that without semaphores we are even more at risk). So I am not too
> > concerned in backporting this particular patch further than we are able
> > to inspect ourselves.
> 
> Should I read between the lines that "the missed-breadcrumb-syndrome"
> has been fixed in dinq?

I wouldn't use fixed; just sufficiently papered over. The runtime of
tests that hunt for such nuisances far exceed the time budget (i.e.
stability testing is more or less limited to the pair of snb I have); so
we have to watch over a few weeks worth of test results to see what the
frequency of occurrence is. Another issue is that since this is timing
related, the extra debugging CI does alters the result...

https://bugs.freedesktop.org/show_bug.cgi?id=100007 is one tracker.
-Chris