Re: [PATCH 07/14] drm/nouveau: use drm_crtc_send_vblank_event()

[Gustavo Padovan <gustavo.padovan@collabora.co.uk>]

Hi Mario,

2016-04-25 Mario Kleiner <mario.kleiner.de@gmail.com>:

> On 04/14/2016 07:48 PM, Gustavo Padovan wrote:
> >From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> >
> >Replace the legacy drm_send_vblank_event() with the new helper function.
> >
> >Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> >---
> >  drivers/gpu/drm/nouveau/nouveau_display.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> >
> >diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
> >index 7ce7fa5..973c2d9 100644
> >--- a/drivers/gpu/drm/nouveau/nouveau_display.c
> >+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
> >@@ -841,10 +841,12 @@ nouveau_finish_page_flip(struct nouveau_channel *chan,
> >
> >  	s = list_first_entry(&fctx->flip, struct nouveau_page_flip_state, head);
> 
> Hi Gustavo
> 
> >  	if (s->event) {
> >+		struct drm_crtc *crtc = drm_crtc_find(dev, s->crtc);
> >+
> 
> I don't think this would work. s->crtc is a nouveau internal display pipe
> index, not a drm mode object id, so i don't think drm_crtc_find() will do
> what you need. Also it takes a mutex, which might_sleep() and i think
> nouveau_finish_page_flip gets called from irq context and holds a
> spin_lock_irqsave, so would end badly.

You are right. I didn't pay attention that the id wasn't a object id.

> 
> You'd probably have to extend struct nouveau_page_flip_state to carry around
> a reference to the required drm_crtc, set up in nouveau_crtc_page_flip().

Yeah, that may be a good solution.

	Gustavo