[Intel-gfx] [PATCH 12/14] drm/i915: Keep vblanks enabled during the entire pipe update

[Chris Wilson]

On Fri, Mar 17, 2017 at 11:18:06PM +0200, ville.syrjala at linux.intel.com wrote:
> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> 
> We currently hold a vblank referenced while trying to evade the
> vblank, but we drop it as soon as we've done that. After all the
> planes have been committed we are quite likely to grab a new vblank
> reference for delivering the flip event. This causes the vblank
> interrupt to do a enable->enable->disable ping-pong during many
> commits. If we instead hang on to the original vblank reference
> across the entire commit we can eliminate that ping-pong.

Does it make sense to take an an extra vblank_get() only if we have an
event to deliver?

> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/intel_sprite.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_sprite.c b/drivers/gpu/drm/i915/intel_sprite.c
> index 26c8bdcd7e72..337e884252e5 100644
> --- a/drivers/gpu/drm/i915/intel_sprite.c
> +++ b/drivers/gpu/drm/i915/intel_sprite.c
> @@ -137,8 +137,6 @@ void intel_pipe_update_start(struct intel_crtc *crtc)
>  
>  	finish_wait(wq, &wait);
>  
	if (crtc->base.state->event)
		drm_crtc_vblank_get(&crtc->base);

	drm_crtc_vblank_put(&crtc->base);

>  	crtc->debug.scanline_start = scanline;
>  	crtc->debug.start_vbl_time = ktime_get();
>  	crtc->debug.start_vbl_count = intel_crtc_get_vblank_counter(crtc);

> @@ -185,6 +183,15 @@ void intel_pipe_update_end(struct intel_crtc *crtc, struct intel_flip_work *work
>  		crtc->base.state->event = NULL;
>  
> +	/*
> +	 * The reference was taken in intel_pipe_update_start(). It could
> +	 * have been dropped as soon as the vblank was evaded, but we hold
> +	 * on to it until this time to avoid the extra vblank interrupt
> +	 * enable->disable->enable ping-pong whenever we have to deliver
> +	 * an event.
> +	 */
> +	drm_crtc_vblank_put(&crtc->base);

	}

>  	local_irq_enable();

-- 
Chris Wilson, Intel Open Source Technology Centre