[Intel-gfx] [RFC] dpms handling on atomic drivers

[Jesse Barnes]

On Thu, 6 Nov 2014 19:35:51 -0500
Rob Clark <robdclark at gmail.com> wrote:

> On Thu, Nov 6, 2014 at 6:29 PM, Daniel Vetter <daniel at ffwll.ch> wrote:
> >
> > That aside I guess I need to elaborate on what makes dpms special in
> > i915, and why there's a real difference between crtc->enable == true
> > && ->active == false and crtc->enable == false in i915. For complex
> > configs we do resource checking (shared dplls) and that's done in
> > the modeset. For a pipe which has been disabled just with dpms we
> > then guarantee that we'll keep these resources reserve and so will
> > always be able to enable the pipe again. If you disable the pipe
> > completely (i.e. set crtc->enable to false) we'll release these
> > resources. E.g. in the i915 share dpll code we have both an active
> > refcount (does the ppl need to be running) and a reference mask
> > (which crtc is referencing this pll, even when the crtc is disabled
> > with dpms).
> 
> ahh, ok, "reserved but not enabled" makes a lot more sense.. that was
> the distinction that I was missing.  That probably deserves to be in
> headerdoc somewhere..

A rename would be nice too; it's very misleading.  Though with a move
to a boolean DPMS internal state, it should be possible to drop it and
just re-alloc all the resources on DPMS on (iow treat both DPMS off and
on as mode sets).  But that's not something that should block these
changes by any means.

Jesse