Hi,
I have been reading about vhost-user and specifically vhost-user-gpu, which is used alongside VirGL/Venus on QEMU to provide "pieces"/"slices" of a GPU from the host to Guests/VMs.
The problem with this is that using `dpdk`, this is running in user-space instead of kernel space, which is a problem for security.
My idea is:
1. Pass through host GPU to guest VM with VFIO.
2. Install dpdk, vhost utilities and VirGL/Venus on Guest.
3. Use RPC (QEMUGuestAgent) to let other guests use "slices" of the GPU (passed through to the original guest).
I have a theoretical (but incomplete) understanding of the first two steps, but the third step is where I'm completely stumped: I have not being able to find any tutorials on how do something like this over RPC.
If anyone has done this, please comment! I would greatly appreciate any leads on this. Thanks!
I have been reading about vhost-user and specifically vhost-user-gpu, which is used alongside VirGL/Venus on QEMU to provide "pieces"/"slices" of a GPU from the host to Guests/VMs.
The problem with this is that using `dpdk`, this is running in user-space instead of kernel space, which is a problem for security.
My idea is:
1. Pass through host GPU to guest VM with VFIO.
2. Install dpdk, vhost utilities and VirGL/Venus on Guest.
3. Use RPC (QEMUGuestAgent) to let other guests use "slices" of the GPU (passed through to the original guest).
I have a theoretical (but incomplete) understanding of the first two steps, but the third step is where I'm completely stumped: I have not being able to find any tutorials on how do something like this over RPC.
If anyone has done this, please comment! I would greatly appreciate any leads on this. Thanks!