Op 20-12-16 om 19:29 schreef Daniel Vetter: > On Mon, Dec 19, 2016 at 10:47:25AM +0100, Maarten Lankhorst wrote: >> Op 19-12-16 om 10:22 schreef Daniel Vetter: >>> On Mon, Dec 19, 2016 at 09:58:28AM +0100, Daniel Vetter wrote: >>>> This gets rid of the last users of for_each_intel_connector(), remove >>>> that too. >>>> >>>> The one exception is the loop in verify_encoder_state - that needs to >>>> switch over to for_each_connector_in_state. >>>> >>>> v2: Maarten corrected me that in the state verifier we indeed must >>>> only walk connectors in the atomic commit, not all of them! >>> Ok, CI just told me this was a stupid idea. Since we don't walk all >>> connectors, but still walk all encoders there's not plenty of state >>> mismatches (e.g. if you have 2 crtc with different connectors enabled and >>> you're doing a modeset on just one). >>> >>> Not exactly sure how to best fix this, since replacing the encoder >>> walking with a connnector walking and then dereferencing >>> connector->state->best_encoder to get at only the encoders relevant for us >>> defeats the point of the cross check. >> Assuming we trust the atomic state, we could fix this by iterating either state and ORing (old,new)_crtc_state->encoder_mask. >> This way we'll have all encoders that we care about. > Yeah, I think we have to rely on the correctness of the atomic states. For > merging I think it's better to get v1 in to correct the connector_list > enumeration, then fix up the encoder checking in a separate patch on top. > Does that sound good? Yes sounds good. ~Maarten _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
