ce/gf100: fix incorrect CE0 address calculation on some GPUs

ce/gf100: fix incorrect CE0 address calculation on some GPUs

commit 93f43ed81abec8c805e1b77eb1d20dbc51a24dc4 upstream.

The code which constructs the modules for each engine present on the GPU
passes -1 for 'instance' on non-instanced engines, which affects how the
name for a sub-device is generated.  This is then stored as 'instance 0'
in nvkm_subdev.inst, so code can potentially be shared with earlier GPUs
that only had a single instance of an engine.

However, GF100's CE constructor uses this value to calculate the address
of its falcon before it's translated, resulting in CE0 getting the wrong
address.

This slightly modifies the approach, always passing a valid instance for
engines that *can* have multiple copies, and having the code for earlier
GPUs explicitly ask for non-instanced name generation.

Bug: https://gitlab.freedesktop.org/drm/nouveau/-/issues/91

Fixes: 50551b15c760 ("drm/nouveau/ce: switch to instanced constructor")
Cc: <stable@vger.kernel.org> # v5.12+
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Tested-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211103011057.15344-1-skeggsb@gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>