[Intel-gfx] [PATCH 2/4] drm/i915: prevent tiling changes on framebuffer backing storage

[Ville Syrjälä]

On Thu, Oct 10, 2013 at 12:09:48AM +0200, Daniel Vetter wrote:
> On Thu, Oct 10, 2013 at 12:02:17AM +0200, Daniel Vetter wrote:
> > On Wed, Oct 09, 2013 at 10:29:39PM +0100, Chris Wilson wrote:
> > > On Wed, Oct 09, 2013 at 09:23:52PM +0200, Daniel Vetter wrote:
> > > > Assuming that all framebuffer related metadata is invariant simplifies
> > > > our userspace input data checking. And current userspace always first
> > > > updates the tiling of an object before creating a framebuffer with it.
> > > 
> > > Userspace already changes the tiling layout whilst keeping the fb id.
> > 
> > How exactly does that work? You can't really change the fb pitch without
> > creating a new one ... That leaves switching from tiled to untiled and
> > back I think.
> 
> Meh, didn't read sna carefully enough. On a second read we only seem to
> have the code added
> 
> commit 0dd20381364aabede2e1306945abe21d57c1d7b4
> Author: Chris Wilson <chris at chris-wilson.co.uk>
> Date:   Sun Sep 29 11:19:46 2013 +0100
> 
>     sna: Resize an existing framebuffer if possible
> 
> That seems to just be for the cache, should be able to cope with failures
> and we can fix it by moving the rmfb up before the set_tiling. It's also
> only 2 weeks old.
> 
> So if there's nothing else I've missed I strongly vote to break the
> pre-release over saner intefaces - allowing tiling to change kinda wreaks
> a bit havoc with out in-kernel checks ...

It would be a bit simpler if we can trust that things don't change
after we've created the fb.

-- 
Ville Syrjälä
Intel OTC