Re: [libvirt] [PATCH] virpci:fix Secondary Bus Reset bug

The fact that you are modifying this code implies that you are using it, and that implies that you are still using legacy KVM device assignment (i.e. the pcistub driver) instead of VFIO device assignment. (I say this because the function that calls this function you've patched, virPCIDeviceReset(), has a check very early on that short-circuits the entire function if the device is bound to vfio-pci (and if the device is bound to a host driver, then you shouldn't be resetting it anyway, so the only reason it would be executed is if you're using legacy KVM device assignment or are calling virsh nodedev-reset when you shouldn't). Legacy PCI device assignment was deprecated over 5 years ago, and was removed from the kernel sometime after that. We are seriously considering removing all vestigal support for legacy KVM device assignment from libvirt, since any kernel that is less than 5 years old have VFIO, and most don't even support legacy KVM device assignment at all. So are you actually using legacy KVM device assignment, or did you just find this bug by manual code inspection? If the latter, then you need to seriously look into using VFIO instead. If the latter, then this patch can safely be pushed, but it's not going to actually do anything (and the code it's in will likely no longer exist within the next 6 months). On 8/15/19 5:44 AM, hexin900110(a)163.com wrote: