[Intel-gfx] [PATCH v2 15/17] drm/i915: Calculate haswell plane workaround, v2.

[Daniel Stone]

Hi,

On 18 May 2015 at 16:47, Daniel Vetter <daniel at ffwll.ch> wrote:
> On Wed, May 13, 2015 at 10:23:45PM +0200, Maarten Lankhorst wrote:
>> This needs to be a global check because at the time of crtc checking
>> not all modesets have to be calculated yet. Use intel_crtc->atomic
>> because there's no reason to keep it in state.
>
> Hm, intel_crtc->atomic needs to be moved into intel_crtc_state eventually,
> so I don't really understand your reasoning here. Yes it's not really a
> static state, but we carry all the other state transition hints like
> ->need_modeset or ->active_changed around in state objects too. Imo it's
> the right place for this stuff, the only tricky part is that we must clear
> them all in the relevant duplicate_state hooks.
>
> Anyway nothing that blocks this one, just something to keep in mind for
> the road ahead.

What makes me slightly nervous is the number of places where we
duplicate the state, meaning that we could easily bleed those if we're
not extremely careful about how we do it. I argued for this one to go
into intel_crtc->atomic as it was much more suited than the static
state, but if we're implementing a transient state section of
intel_crtc_state (ideally under its own sub-struct, so it can easily
be zeroed), then I guess that wfm.

Cheers,
Daniel