[Intel-gfx] [PATCH] drm: Pass crtc to .best_encoder()

[Ville Syrjälä]

On Tue, Jun 26, 2018 at 05:23:25PM +0200, Daniel Vetter wrote:
> On Fri, Jun 15, 2018 at 10:52:21PM +0300, Ville Syrjala wrote:
> > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > 
> > To pick the correct MST encoder i915 wants to know which crtc is going
> > to be feeding us. To that end let's pass the crtc to the .best_encoder()
> > hook. The atomic variant already knows the crtc via the connector state,
> > but the non-atomic hooks is still being used by the fb_helper even on
> > atomic drivers.
> 
> Can't we instead fix the fb helpers to use atomic_best_encoder?

We'd need a connector state for that. Not sure we want to construct a
fake one on demand or something.

> Or just
> drop that code, userspace doesn't really know any better either and just
> needs to figure this out without the help of ->best_encoder, so it should
> be possible.

I have to admit I don't even remmber how userspace does this. OK, yeah
so each connector has a list of encoders and each encoder has the
possible_crtcs bitmask. So we should be able to just iterate through
all the encoders for the connector. Not sure there couldn't be some
some false positives if best_encoder() doesn't quite agree with the
answer for some reason.

But yeah, this does seem like a nicer option because we can then nuke
the extra .best_encoder() hook for i915 mst.

Now the question becomes whether I get to review the
connecor->encoder_ids[] assignments in every driver. I suppose they
should be mostly OK if userspace manages to work.

-- 
Ville Syrjälä
Intel