Comment 29 for bug 1832138

Testing upstream kernel, I can tell this situation is not a Ubuntu kernel specific issue.
I can reproduce using upstream kernel.

This is where I'm at the moment:
BAD kernel : v4.16.18
GOOD kernel : v4.17-rc1

Leaving us to the following 'vmwgfx' commits to bisect:

----------------
$ git log --oneline v4.16.18..v4.17-rc1 --grep="drm/vmwgfx:"
----------------
320b164abb32 Merge tag 'drm-for-v4.17' of git://people.freedesktop.org/~airlied/linux
2a2553cc45c8 Merge branch 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux into drm-next
43bfefedd028 drm/vmwgfx: Bump version patchlevel and date
37efe80ce85f drm/vmwgfx: use monotonic event timestamps
20fb5a635a0c drm/vmwgfx: Unpin the screen object backup buffer when not used
89dc15b76fd3 drm/vmwgfx: Stricter count of legacy surface device resources
6073a09210e0 drm/vmwgfx: Use kasprintf
4e3e733b45df drm/vmwgfx: Get rid of the device-private suspended member
c3b9b1657344 drm/vmwgfx: Improve on hibernation
bf833fd36f9b drm/vmwgfx: Avoid pinning fbdev framebuffers
dc366364c4ef drm/vmwgfx: Fix multiple command buffer context use
ef86cfee7d74 drm/vmwgfx: Use the cpu blit utility for framebuffer to screen target blits
79273e1b7eb0 drm/vmwgfx: Add a cpu blit utility that can be used for page-backed bos
25db875401c8 drm/vmwgfx: Cursor update fixes
904efd9e3f4c drm/vmwgfx: Send the correct nonblock option for atomic_commit
ac3069e67f56 drm/vmwgfx: Move the stdu vblank event to atomic function
aa64b3f18aeb drm/vmwgfx: Move screen object page flip to atomic function
3cbe87fcf026 drm/vmwgfx: Remove drm_crtc_arm_vblank_event from atomic flush
4e2f9fa7ffb5 drm/vmwgfx: Move surface copy cmd to atomic function
91e9f352cd1b drm/vmwgfx: Avoid iterating over display unit if crtc is available
25a28906ebee drm/vmwgfx: replace drm_*_unreference with drm_*_put
4f4becef17d6 drm/vmwgfx: Use drm_mode_get_hv_timing() to populate plane clip rectangle
----------------

As I said, I'll start the bisection on Monday.
My first feeling about the potential good candidate assumption was wrong but with recent testing, we now know the fix is amongst the commits listed above ^

- Eric