[Intel-gfx] [PATCH] drm/i915: Don't scream if there's no context for reset stats

Ben Widawsky ben at bwidawsk.net
Mon Mar 10 22:41:36 CET 2014


On Mon, Mar 10, 2014 at 09:30:22PM +0100, Daniel Vetter wrote:
> On Mon, Mar 10, 2014 at 7:30 PM, Ben Widawsky <ben at bwidawsk.net> wrote:
> > On Mon, Mar 10, 2014 at 09:44:22AM +0100, Daniel Vetter wrote:
> >> It can happen ...
> >>
> >> Fix up the check to match pre-gen6 reality where we don't have hw
> >> contexts and hence also don't need to set the reset status on them.
> >>
> >> This blows up when running any gpu reset testcase since for pre-gen6
> >> request->ctx is NULL. With this my ilk here is happy again.
> >>
> >> This regression has been introduced in
> >>
> >> commit 44e2c0705a19e09d7b0f30a591f92e473e5ef89e
> >> Author: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> >> Date:   Thu Jan 30 16:01:15 2014 +0200
> >>
> >>     drm/i915: Use i915_hw_context to set reset stats
> >>
> >> Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> >> Cc: Ben Widawsky <ben at bwidawsk.net>
> >> Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> >
> > Did you try playing around with setting last_context to
> > private_default_context? That is more in line with the original outlined
> > approach of "every platform has a context, be they fake, or real."
> 
> Nope. Currently the tests for that are a bit busted, so I think we
> should do that when we fix things up generally. Mika seems to be
> working on this. For now this keeps my machines happy. There's also
> the problem that our QA seems to have missed this ...
> -Daniel

As long as you've thought about it, lgtm, although I've not investigated
if other areas of the code need it. Without last_context always != NULL,
I wouldn't be surprised if you hit something similar elsewhere.

Reviewed-by: Ben Widawsky <ben at bwidawsk.net>

-- 
Ben Widawsky, Intel Open Source Technology Center



More information about the Intel-gfx mailing list