Use NVD_GPU to control which DRM node to open in direct backend mode #137
Conversation
|
Hmm, I swear I wrote a comment on this yesterday, Github seems to have eaten it. I think I'd prefer a With that said, I'd ideally like some way to detect the DRM device from the X11 handle that's passed in (we should have a DRM device if using Wayland), but the only ways I currently know of doing that are using the DRI3 Open method, which wasn't supported the last time I looked, or some wacky NV-GLX calls. I guess we could just scan for the first NVIDIA DRM device, and leave more complicated setups to the NVD_DRM_DEVICE variable. |
|
How about using NVD_GPU to mean the nth nvidia device, and we iterate over the render nodes and look for nvidia devices specifically. Then we can keep using the same variable with it meaning the same thing. |
philipl
force-pushed
the
direct-drm-node
branch
from
421fec4 to
ea63f1a
Compare
|
Updated diff to use |
When running in direct backend mode, we need to open the DRM node ourselves, and the first node may not be the right one. We will iterate through the DRM nodes looking for NVIDIA GPUs and will use the first one found or the one identified by NVD_GPU if set.
philipl
force-pushed
the
direct-drm-node
branch
from
ea63f1a to
ea10d0a
Compare
|
Updated to use |
|
Ok, that looks good. Thanks for that! |
When running in direct backend mode, we need to open the DRM node ourselves, and the first node may not be the right one.
This change uses NVD_GPU to identify which one to use. It might be desirable to use a different env var because it will count non-nvidia GPUs so it could be confusing. That's easy to change.