[Intel-gfx] [PATCH 1/2] drm/i915: Remove use of the autoreported ringbuffer HEAD position

Daniel Vetter daniel at ffwll.ch
Wed Feb 8 19:02:47 CET 2012

On Wed, Feb 8, 2012 at 18:36, Keith Packard <keithp at keithp.com> wrote:
> <#part sign=pgpmime>
> On Wed, 8 Feb 2012 15:54:25 +0100, Daniel Vetter <daniel at ffwll.ch> wrote:
>> I'm on the fence whether we should include this in -fixes. On one hand it
>> fixes a severe issue on snb, but introduces a perf regression without the
>> second patch. Otoh we've never shipped snb without it broken like this.
>> But if it turns out that this is broken on ilk and earlier, too, I think
>> we definitely need these two patches in -fixes.
> The first one is a simple bug fix -- not attempting to use the status
> page value. That seems reasonable for -fixes.

The issue is that the first one introduces a pretty decent perf
regression, iirc Chris mentions something much large than 10x slowdown
on certain cairo traces on sna. So we can only merge the first one
together with the second one.

> The second one has a bunch of global changes which seems less suitable
> for -fixes, especially as it now requires that the driver reliably track
> ring positions, which seems a bit perilous to me.

Yeah, that's kinda the issue which I've failed to phrase clearly in my mail ;-)

Without this issue being confirmed on older machines I'm leaning
towards merging both patches to -next - after all snb never really
regressed because the offending commit changed behaviour from "dies as
soon as the ring wraps" to "sometimes corrupts the ring".
Daniel Vetter
daniel.vetter at ffwll.ch - +41 (0) 79 364 57 48 - http://blog.ffwll.ch

More information about the Intel-gfx mailing list