Jetson nano headless boot and GPU passthrough to KVM/qemu


Is it possible to boot Jetson nano in headless mode and then run a KVM/qemu vm with GPU assigned to VM, and running tensorflow model inside VM?

Not planning to run any graphics in vm too.

If I could just unbind the driver from host, and assign the dtb nodes in vm device tree? Is it possible?

It is easy to do a vfio pcu passthrough, not sure what dependencies are there.

Sorry, we don’t have such case so don’t know if it can work.

Maybe other users can share their experience.

i have tried to boot headless and unbending the driver it works.
Need to find out in VM what dependency of other devices is required so that they can be assigned to VM itself.
Wayne I believe you are from nvidia, can you provide some pointers. Is there a gpu reference manual?


I am not sure what kind of info you need here. There is no gpu document.

There is one nvgpu driver in our kernel, which is open source.