[Intel-gfx] [PATCH] drm/i915/skl: Enabling PSR on Skylake

[Daniel Vetter]

On Wed, Jan 21, 2015 at 02:25:23PM +0530, sonika wrote:
> 
> On Wednesday 21 January 2015 02:07 PM, Daniel Vetter wrote:
> >On Tue, Jan 20, 2015 at 02:01:25PM -0800, Rodrigo Vivi wrote:
> >>Hi Sonika, for the login screen my guess is that blinking cursor
> >>waiting for password blocks psr entry.
> >>
> >>For test cases I started the test case enhancement but my psr work is
> >>paused this month for bug maintainance. Any help is welcome.
> >>I was planing also to take a look on new Paulo's fbc tests cases to
> >>see if we could use some pieces on PSR validation. might be worth to
> >>take a look.
> >>
> >>But anyway, for HW tracking I would only consider when I use a KDE for
> >>a while without missing any screen updates and seeing perf counter
> >>increasing. But if you tell this is what happening I trust you and we
> >>can add support for that since it is disabled by default and validate
> >>that carefully also on SKL when trying to enable it by default
> >>everywhere.
> >I'm definitely leaning the other way: Before we take out the sw tracking
> >for some platforms I want the full psr validation to be done (since
> >otherwise no one will do it again and for skl+1 we have the same hilarious
> >saga again). And if we figure out that the hw tracking has indeed become
> >better and is now useful, then we can do the same thing like Paulo's fbc
> >patches and ignore some sw invalidate events selectively. But only when
> >the tests are solid, we're sure we have full coverage and that the hw
> >really works for that use-case.
> >-Daniel
> Ok, I like the idea of selectively ignoring sw invalidates. And then PSR2's
> HW tracking will still hold good
> in those cases. I think we can have a module parameter (hw_tracking) instead
> of checking for platform
> everywhere.

I don't think a hw_tracking variable is all that useful, since I expect
every platform to work slightly different. If you look at the psr code we
already have some platform checks for specific issues (e.g. sprite flips
don't invalidate on hsw but do on bdw). Imo much better to just have
platform base checks which filter out specific sw-invalidate sources.

Please also look at Paulo's fbc series which adds more information to the
invalidate callbacks so that we can filter more precisely.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch