[Intel-gfx] [PATCH 11/11] drm/i915: Enable PSR by default.

[Daniel Vetter]

On Sun, Jul 13, 2014 at 8:30 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> On Sat, Jul 12, 2014 at 10:48:30PM +0200, Daniel Vetter wrote:
>> On Sat, Jul 12, 2014 at 8:55 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
>> > On Sat, Jul 12, 2014 at 08:15:42PM +0200, Daniel Vetter wrote:
>> >> On Sat, Jul 12, 2014 at 7:51 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
>> >> > On Sat, Jul 12, 2014 at 07:14:33PM +0200, Daniel Vetter wrote:
>> >> >> On Sat, Jul 12, 2014 at 2:02 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
>> >> >> > But PSR is definitely an output property... Please-dont-touch-me is
>> >> >> > indeed a plane property.
>> >> >>
>> >> >> Well I guess we also want this for fbc, and fbc isn't an output
>> >> >> property. I guess in the end it doesn't really matter that much as
>> >> >> long as it's there, but traditional userspace exposes all output
>> >> >> properties to userspace (on X), hence why I think we should hide it a
>> >> >> bit ;-)
>> >> >
>> >> > Hence why I want PSR as an output property. I want to see the status
>> >> > exposed in xrandr.
>> >>
>> >> Hm, why that? I've thought we only want this to give hints to the
>> >> compositor/ddx?
>> >
>> > An interesting factoid to present to the user. Also makes it easy for me
>> > to check everything works in X.
>>
>> Hm not sure we want users to know this stuff. Next they ask to adjust
>> it and then we have the same mess as with kernel options ;-) But I
>> don't care strongly enough really either way. In-kernel I still think
>> we want to keep this on the crtc so that fbc and psr work the same
>> way. The connector property would then just chase the connected crtc
>> to read out the property.
>
> It can't be crtc as it is an connector properrty... And it is immutable.
> I think it will be one of those useful things that people would like to
> check infrequently to make sure everything is functioning as intended
> (like powertop). I think there may be a few properties like this we can
> expose. And if people ask why PSR isn't active, then we should do a
> better job at making sure it stays enabled.

Oh, you mean just a psr property? I've thought in general terms of
discouraging frontbuffer rendering and included fbc. A simple
has_self_refresh_display (don't want to exclude dsi) is obviously a
connector thing. But I thought you want to use that as a "should I
flip or can I frontbuffer render" hint, and I guessed that fbc should
be part of that. At least once we switched it to the sw tracking.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch