The new atomic modesetting/pageflip code for nv50+ for Linux 4.10+ no longer uses pageflip irq's to signal flip completion. Instead it polls for flip completion from within a kthread/work queue. This creates a race between the vblank irq handler updating the vblank count and timestamp for the
On 20.10.2016 00:46, Ben Skeggs wrote: [...] > I'd like to see a mmiotrace of the NVIDIA binary driver on a system > where this WAR breaks things. I applied it to all the GPUs that NVIDIA > told me required it. > > Ben. > Still broken, more than four months, tested with mainline 4.9-rc6.