drm/nve0: allow specification of channel engine type in abi16 call

Previously, if either vram/gart handles were specified as ~0, the ioctl
call would fail.  In order to hack engine selection into the ioctl for
kepler, we now define (fb_ctxdma_handle == ~0) to mean "engine mask is
in tt_ctxdma_handle".

This approach also allows new userspace to detect lack of support for
non-PGRAPH channels on older kernels.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

1 files changed, 1 insertions, 1 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index 049c6b23e1d7..1363578bf945 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -346,7 +346,7 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart)
         /* allocate software object class (used for fences on <= nv05, and
          * to signal flip completion), bind it to a subchannel.
          */
-        if (chan != chan->drm->cechan) {
+        if ((device->card_type < NV_E0) || gart /* nve0: want_nvsw */) {
                 ret = nouveau_object_new(nv_object(client), chan->handle,
                                          NvSw, nouveau_abi16_swclass(chan->drm),
                                          NULL, 0, &object);