[Intel-gfx] [PATCH 4/4] drm/i915: creating Haswell rc6 function

[Daniel Vetter]

On Tue, Mar 26, 2013 at 8:37 PM, Rodrigo Vivi <rodrigo.vivi at gmail.com> wrote:
> I forgot to say that I tested patch by patch today on my HSW ULT and just
> noticed real improvement with the patch you already queued.
>
> Where improvements are: at least 0.2W and also less oscillation and more %
> on RC and package C states.

Cool!

> Although I'd like to see the rest of the series applied to have the code as
> clean, organized and matching documentation as possible.

Yeah, sounds like a good plan (at least for hsw).

> On Tue, Mar 26, 2013 at 4:32 PM, Rodrigo Vivi <rodrigo.vivi at gmail.com>
> wrote:
>>
>> I'm in favor of this revert. Although I don't have any argument in values,
>> I always guessed that many of rc6 bugs we have on snb came from the gap
>> between the threashold values used and documented for snb.

The reason I'm still wary is that this helped (according to multiple
people) to more reliably enter rc6. It might be that we've simply
mixed up other rc6 issues and hangs here (or that it's just as flaky
as it seems), but for snb/ivb we need to be careful.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch