[Intel-gfx] [PATCH 2/2] drm/i915: Only check pipe state for fast modeset when it's possible.

[Maarten Lankhorst]

Op 14-09-15 om 19:10 schreef Daniel Stone:
> Hi,
>
> On 14 September 2015 at 10:30, Maarten Lankhorst
> <maarten.lankhorst at linux.intel.com> wrote:
>> @@ -13013,14 +13013,15 @@ static int intel_atomic_check(struct drm_device *dev,
>>                 if (ret)
>>                         return ret;
>>
>> -               if (intel_pipe_config_compare(state->dev,
>> -                                       to_intel_crtc_state(crtc->state),
>> -                                       pipe_config, true)) {
>> +               if (!crtc_state->connectors_changed &&
>> +                   !crtc_state->active_changed && look
>> +                   crtc_state->active &&
>> +                   intel_pipe_config_compare(state->dev,
>> +                                             to_intel_crtc_state(crtc->state),
>> +                                             pipe_config, true)) {
>>                         crtc_state->mode_changed = false;
>> -                       to_intel_crtc_state(crtc_state)->update_pipe = true;
>> -               }
>> -
>> -               if (needs_modeset(crtc_state)) {
>> +                       pipe_config->update_pipe = true;
>> +               } else {
>>                         any_ms = true;
> The change from only setting any_ms if needs_modeset() is true, to
> always if we can't do a fastset, seems correct but maybe a bit subtle.
It's exactly the same thing, just made a bit more explicit.

before: any_ms = needs_modeset() with mode_changed = !update_pipe.
After: any_ms = !update_pipe.

> Was that intended? At the moment it does look like it'll widen the net
> a little bit, but I _suspect_ that's a good thing. Pending igt:
> Acked-by: Daniel Stone <daniels at collabora.com>
>
> Cheers,
> Daniel