[Intel-gfx] [PATCH] drm/i915: Make i915_gem_context_mark_guilty() safe for unlocked updates

[Chris Wilson]

Quoting Mika Kuoppala (2017-07-10 15:14:19)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > Since we make call i915_gem_context_mark_guilty() concurrently when
> > resetting different engines in parallel, we need to make sure that our
> > updates are safe for the unlocked access.
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Michel Thierry <michel.thierry at intel.com>
> > Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> 
> And we dont ever clear the context bannable flag so
> this should be fine.
> 
> One improvement would be that only the triggering
> ban would spew the message. But now thinking about it,
> the dual message would reveal the concurrent ban and
> is more informative.

It only tells you that we encountered a race condition in handling a
reset on two engines. A double ban message is going to be more confusing
than anything.

Did I send the patch to add the user message for the per-engine reset? I
was having withdrawal symptoms from not seeing it before mesa died.

Otoh, we clearly do propagate the error to userspace, but on the other
hand I'm used to seeing it. It's akin to the kernel reporting when it
sends a terminal signal to a process. I've been trying to find that to
see if they discuss the same issue and in particular why its hidden
behind kconfig and sysctl.
-Chris