
On 09/03/2013 08:06 AM, Gerd Hoffmann wrote:
Hi,
How does this work if we start off a guest with 3 disks attached to the usb-bot SCSI controller. Then hot-unplug the 2nd disk. You can't hotplug individual luns anyway. How does hotplug/unplug work in the context of usb-bot ?
AFAIK we need to be able to run
device_add usb_bot drive_add file... device_add scsi-hd
And the reverse, to unplug it, if we're to have feature parity with usb-storage. Hot-unplug is easy. You can remove the usb-bot device which will also remove all child devices.
Hot-plug doesn't work at the moment, and I don't see any obvious way to fix that properly :-(
We need some way to hotplug a *group* of devices (usb-bot + all children) as usb-bot itself is hotpluggable but the individual scsi devices connected to it are not.
There is another case where hot-plug/unplug of a group of devices would be useful - if you want to hotplug/unplug a pci passthrough device using vfio *and* that device is a part of an iommu group that contains other devices *and* you're okay with having those other devices assigned to the guest as well *AND* you want to use "managed='yes'" (where libvirt automatically unbinds the devices from the host driver and binds them to vfio-pci before passing them off to the guest), then you need to be able to plug and unplug all the devices in that iommu group in a single operation, otherwise the plug fails (due to attempting to assign one device in a group when the other devices haven't been bound to vfio-pci or pci-stub). As it stands now you can only hot-plug/unplug with vfio if a device is in an iommu group all by itself, or if you don't use managed='yes', and instead deal with binding the devices to vfio-pci manually. If none of that makes sense, consider yourself lucky :-)
I could allow hotplug on usb-bot as workaround, then you can do
stop device_add usb_bot device_add scsi-{hd,cd,whatever} cont
but that would be more a gross hack than a solution ...
cheers, Gerd
-- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list