[Intel-gfx] [PATCH 1/2] drm/i915: Don't drop pinned fences

[Ville Syrjälä]

On Mon, Jan 27, 2014 at 10:41:57PM +0100, Daniel Vetter wrote:
> On Mon, Jan 27, 2014 at 10:26 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > On Mon, Jan 27, 2014 at 09:49:47PM +0100, Daniel Vetter wrote:
> >> Userspace can currently provoke this when e.g. trying to use a pinned
> >> scanout as a cursor or overlay target. Later on that might lead to
> >> some fun fence pin count mayhem.
> >>
> >> Spurred by Ville's report that something goes wrong here and
> >> originally I've thought that this might slip through the pwrite gtt
> >> fastpath. But that one checks of obj tiling, so should be ok.
> >>
> >> But one thing that _does_ blow up is the vma unbinding with more than
> >> one address space. The next patch will fix this.
> >
> > It's buggy behaviour, so if (WARN_ON(fence->pin_count) return -EBUSY;
> >
> > I did worry that we would then have some unexpected failure, a new EBUSY
> > returning to userspace, but I think along the paths that can trigger it,
> > we have already returned EBUSY in the (dim and distant) past.
> 
> Indeed we seem to have the required tiling checks everywhere. At first
> I've had the WARN but then though userspace could sneak in through the
> cursor/overlay code, but it can't. I'll update the patch.

I've tested these (+ Chris's WARN_ON) and the issue looks to be fixed.

Tested-by: Ville Syrjälä <ville.syrjala at linux.intel.com>

-- 
Ville Syrjälä
Intel OTC