On 04/06/14 21:15, Keith Packard wrote:
Right, it's reasonable for the kernel to not care whether the buffer is idle or not when flipping. However, the application *does* care, and so we need a way to not flip until rendering is complete. Whether that's additional mechanism in the kernel, or additional code up in user space is the question.
In the case of PVR driver, we end up sending the flip down to the kernel whether or not the buffer is idle. We then have a mechanism for waiting for all rendering to complete before executing the flip via mmio. This is mostly why the whole "always async flip if supported" thing doesn't work for us and the reason for my patch to always sync flips unless instructed not to by the client: http://lists.x.org/archives/xorg-devel/2014-May/042604.html _______________________________________________ [email protected]: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
