Re: [libvirt] [PATCH v2 09/11] qemu: don't mess with non mandatory hostdevs on reattaching

On Mon, Sep 16, 2019 at 08:58:10 +0000, Nikolay Shirokovskiy wrote: > > > On 16.09.2019 11:30, Peter Krempa wrote: >> On Mon, Sep 09, 2019 at 14:33:12 +0300, Nikolay Shirokovskiy wrote: >>> First I don't want to add code to handle dummy device that is used when >>> host usb device is not present at the moment of starting/migrating etc. >>> Second supporting non mandatory policies would require to handle races >>> when host usb device is plugged to host and libvirtd starts adding >>> device but if in the meanwhile host usb device it unplugged back then >>> current code will use dummy device which is not desired in this case. >> >> I don't think that handling startupPolicy for cases other than VM >> startup makes semantic sense. Could you elaborate what's the goal? I >> didn't quite get it from the commit message. >> > > We have different states for device that was missing on start > and device that was unplugged. In the former case we have stub > device in qemu (bus=0, device=0) while in the latter case we don't > have any correnspondent device in qemu. So I don't want add extra > logic to handle the first case on re-plug. AFAIK a device whose startup policy allows it to be missing is removed fully from the (running) definition on VM startup if it was not present at that time. If it is not like that it's probably a bug because having a device in the definition which is not recognized by the VM should not happen.