Comment 69 for bug 1832138

And the problem is indeed the kernel advertising DRM_CAP_TIMESTAMP_MONOTONIC but instead of using CLOCK_MONOTONIC it is really using CLOCK_REALTIME. This confirms kernel commit 37efe80ce85f is the fix, but it may well have some prerequisite commits we haven't fully worked out...

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?h=v5.2-rc5&id=37efe80ce85f

I'll see if I can hack around it, but the simplicity of the mistake in the vmwgfx kernel code makes me think we should be preferring a kernel fix still.