[Intel-gfx] [PATCH] drm/i915/hsw: Disable L3 caching of atomic memory operations.

[Ben Widawsky]

On Thu, Oct 03, 2013 at 12:20:43AM +0200, Daniel Vetter wrote:
> On Thu, Oct 3, 2013 at 12:03 AM, Francisco Jerez <currojerez at riseup.net> wrote:
> > +       case I915_PARAM_HAS_ATOMICS:
> > +               value = 1;
> > +               break;
> 
> Generally when we do kernel fixes for gpu hangs like that we don't add
> parameters (would drown in them otherwise) but simply queue it up to
> -fixes and slap a cc: stable on it. Gpu hang fixes are critical enough
> imo for that treatment, even when it's for brand new userspace code.
> 
> Any specific reason why we shouldn't follow this approach here? I'd
> make the patch simpler and we could dump a bit of userspace code, too.
> -Daniel
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> +41 (0) 79 365 57 48 - http://blog.ffwll.ch

They haven't yet enabled this feature in mesa, so it's not exactly
fixing a hang. It is preventing one from ever occurring. Mesa versions
built against an older libdrm will not use atomics.

-- 
Ben Widawsky, Intel Open Source Technology Center