On 09/05/18 10:05, Chris Wilson wrote:
Could there be any more pointer chasing? </chandler>
Thinking about more about how to make this part cleaner, could we not
store the engine->noa_batch and then this all becomes
vma = engine->noa_batch;
if (vma)
return;
Locking! Missed it in the first pass, but we are unlocked here... That
also ties into lifetimes.
I'm thinking more like ctx->noa_batch with that being pinned along with
the context. We need something along those lines to track the
locking/lifetime.
-Chris
In another series tracking pid of submitted requests, I used "empty"
requests to change the state of the engine.
I could use that approach to message-pass what needs to be used in the
engine submission.
-
Lionel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
