Re: [PATCH] drm/i915: small semaphore fix

[Ben Widawsky <ben@bwidawsk.net>]

On Fri, Sep 02, 2011 at 09:56:31AM +0100, Chris Wilson wrote:
> On Thu, 1 Sep 2011 19:51:11 -0700, Ben Widawsky <ben@bwidawsk.net> wrote:
> > On Thu, 01 Sep 2011 11:03:07 -0700
> > Eric Anholt <eric@anholt.net> wrote:
> > 
> > > On Wed, 31 Aug 2011 12:47:22 -0700, Ben Widawsky <ben@bwidawsk.net>
> > > wrote:
> > > > Assertion + unsigned helps catch potential issues.
> > > > 
> > > > From the docs it is hard to tell if the global GTT flag is actually
> > > > needed, but it shouldn't hurt.
> > > 
> > > We're updating a register, not the GTT, so I don't see why the flag
> > > would be relevant.
> > 
> > The patch is as (ir)relevant as without it which implies PPGTT. The
> > reason it was there is in another branch I was planning to implement
> > memory based semaphores.
> 
> Memory based semaphores required working ppgtt, afaict.
> -Chris

Hmm, nothing in the docs suggested to me that it wouldn't work. I know for sure
it won't work on the simulator. So long as the semaphore commands are being
issued from a secure batch, I don't see why ppgtt *should* be a requirement. I
think the whole point of the gGTT flag is to allow inter-contexct
synchronization without needing to have identical ppgtt mappings.

I was planning to give it a shot today, but based on the simulator I suspect it
won't work.

Ben