> But I find a interesting thing, if I don't > start the vm after GPU is bind to the vfio, GPU can still be rebind to > the i915 driver. Do you know why? Thanks.
As Alex’ blog [1] pointed out, vfio-pci is basically pci-stub plus creating a vfio device node. pci-stub just prevents other drivers from binding to the device. The vfio device node on the other hand makes the device address space available for Guest OSes, but it does not touch the device itself. So its address space (think of it as GPU ram + internal buffer and registers) is still left in a pristine state until you bind it to a VM. For that reason it makes perfectly sense, that you can rebind it to i915, if you did not start any VM with VGA passthrough. Kind regards, Manuel [1]: https://vfio.blogspot.de/2015/05/vfio-gpu-how-to-series-part-3-host.html _______________________________________________ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users