[Nouveau] [PATCH] drm/ttm: Fix race condition in ttm_bo_delayed_delete

[Luca Barbieri]

> Doing it without software methods means you need to have a semaphore
> that "exists" in the cpu domain and therefore cannot be used again
> until the cpu is sure it's done. So that will probably require a
> rotating queue of several semaphores and a fallback to cpu waiting.

Why would it need a semaphore in CPU domain?
Couldn't it work this way:
1. Allocate a BO in the kernel
2. Emit on one channel a "notify" request on that BO
3. Emit on the other channel a "wait" request on that BO
4. Emit a fence on the wait channel
5. Use the existing delayed-destroy mechanism to get rid of the BO
once the wait channel fence expires

Then the GPU would presumably switch away from the "waiting" context
and not reexecute it until the notify happens, or something similar.

The problem is, of course, whether the GPU supports the "wait" request.