Here the respin of my series of vblank bug fixes for stable Linux 4.4/4.5 with all review comments by Daniel and Ville taken into account and reviewd-by's and tested-by's added.
I've removed the stable patch "drm: Prevent vblank counter jumps with timestamp based update method." for the moment. I'll send a respin of that patch later for drm-next / Linux 4.6, taking Daniels and Villes comments into account.
Instead a stable patch 6/6 for nouveau now solves the problem of large vblank counter jumps for that driver on suspend/resume. As nouveau is currently the only user of the timestamping based vblank counter path, this should fix that regression properly for Linux 4.4/4.5. Tested with a couple of suspend/resume cycles composited/non-composited, with/without vblank clients on a laptop with nv-50.
thanks, -mario