[PATCH v3 03/12] drm/i915/display: Use async flip when available for initial plane config

[Ville Syrjälä]

On Thu, Oct 03, 2024 at 10:50:24PM +0200, Maarten Lankhorst wrote:
> Hello,
> 
> Den 2024-10-03 kl. 18:14, skrev Ville Syrjälä:
> > On Thu, Oct 03, 2024 at 05:44:12PM +0200, Maarten Lankhorst wrote:
> >> I'm planning to reorder readout in the Xe sequence in such a way that
> >> interrupts will not be available, so just use an async flip.
> >>
> >> Since the new FB points to the same pages, it will not tear. It also
> >> has the benefit of perhaps being slightly faster.
> >>
> >> Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
> >> ---
> >>   drivers/gpu/drm/i915/display/skl_universal_plane.c | 13 +++++++++++--
> >>   1 file changed, 11 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/i915/display/skl_universal_plane.c b/drivers/gpu/drm/i915/display/skl_universal_plane.c
> >> index fdb141cfa4274..73a3624e34098 100644
> >> --- a/drivers/gpu/drm/i915/display/skl_universal_plane.c
> >> +++ b/drivers/gpu/drm/i915/display/skl_universal_plane.c
> >> @@ -2800,7 +2800,7 @@ bool skl_fixup_initial_plane_config(struct intel_crtc *crtc,
> >>   		to_intel_plane_state(plane->base.state);
> >>   	enum plane_id plane_id = plane->id;
> >>   	enum pipe pipe = crtc->pipe;
> >> -	u32 base;
> >> +	u32 base, plane_ctl;
> >>   
> >>   	if (!plane_state->uapi.visible)
> >>   		return false;
> >> @@ -2814,7 +2814,16 @@ bool skl_fixup_initial_plane_config(struct intel_crtc *crtc,
> >>   	if (plane_config->base == base)
> >>   		return false;
> >>   
> >> +	/* Perform an async flip to the new surface. */
> >> +	plane_ctl = intel_read(i915, PLANE_CTL(pipe, plane_id));
> >> +	plane_ctl |= PLANE_CTL_ASYNC_FLIP;
> >> +
> > 
> > No async flips!
> 
> Can you please explain your reasoning?

Async flips are special. They have all kinds of random hardware
limitations.

> I think async flip would fit 
> perfectly here. We cannot perform a wait_for_vblank as we will not have 
> interrupts enabled yet.

The type of flip is irrelevant when you just poll until it's done.

> Additionally an async flip would cause a faster 
> driver loading. Since the FB is exactly the same except set to a 
> different address, no tearing will occur.

Until we violate some hardware requirement and you get a fault.

-- 
Ville Syrjälä
Intel