On Fri, Jan 30, 2015 at 03:57:53PM +0000, Daniel Stone wrote: > Hi, > > On 30 January 2015 at 14:30, Gustavo Padovan <gustavo@xxxxxxxxxxx> wrote: > > 2015-01-30 Joonyoung Shim <jy0922.shim@xxxxxxxxxxx>: > >> We will lose unfinished prior events by this change. That's why we use > >> linked list. > > > > I think you are right, but I was using exynos_crtc->event to do exactly the > > same as exynos_crtc->pending_flip. So we were losing a event in > > exynos_drm_crtc_dpms() before too. I change this patch to have a page_flip > > list on the crtc. > > The usual approach in other drivers is to return -EBUSY when there is > already an async pageflip pending. This definitely makes sense to me, > as I don't see the point of submitting pageflips faster than the > hardware can actually render, and pretending to the application that > they were actually shown. Yes, right now drm doesn't really support anything like a pageflip queue. Same for atomic really. Even the async pageflip mode works like it, it just ends up flipping faster. Long-term we want a flip queue where subsequent flips can be folded together on the next vblank. That makes benchmark-mode games happy, without resulting in tearing like async flips and still resulting in the lowest possible latency (since the kernel we just commit the flips for which all the buffers are ready and not stall). -Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
