drm/nouveau/fifo/gk104-: require explicit runlist selection for channel allocation
authorBen Skeggs <bskeggs@redhat.com>
Tue, 8 May 2018 10:39:46 +0000 (20:39 +1000)
committerBen Skeggs <bskeggs@redhat.com>
Fri, 18 May 2018 05:01:21 +0000 (15:01 +1000)
commita7cf01809bf23b95413d8047bd91cdc3cedd1ca1
tree92a72d7f2fe690c30a6b423350d09a671eabfca3
parentcc36205085bb6e3a4eed1edbe413fd2235cadb27
drm/nouveau/fifo/gk104-: require explicit runlist selection for channel allocation

We didn't used to be aware that runlist/engine IDs weren't the same thing,
or that there was such variability in configuration between GPUs.

By exposing this information to a client, and giving it explicit control
of which runlist it's allocating a channel on, we're able to make better
choices.

The immediate effect of this is that on GPUs where CE0 is the "GRCE", we
will now be allocating a copy engine running asynchronously to GR for BO
migrations - as intended.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/include/nvif/cla06f.h
drivers/gpu/drm/nouveau/nouveau_abi16.c
drivers/gpu/drm/nouveau/nouveau_chan.c
drivers/gpu/drm/nouveau/nouveau_drm.c
drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c