On 03/23/2014 08:04 AM, Andreas Färber wrote: > Am 22.03.2014 21:54, schrieb Peter Maydell: >> On 22 March 2014 20:18, Andreas Färber <afaer...@suse.de> wrote: >>> Because AFAIU >>> migration is possible without VMSD, just not with VMSD that sets >>> .unmigratable = 1. >> >> Well, the migration won't fail with an error, but on the destination >> end you'll end up with a device in its reset state but a guest which >> may think the device is in some other state. If the device was >> quiescent and doesn't need complex setup it might cope, but >> more likely is that that guest driver will fall over in a heap next >> time you try to use it... > > Well, there is no OHCI state being added, only PCI state. So I'd be > curious to know what in there is the problem because a general review of > PCI devices might be due then - and ideally before we do the release.
Without the patch, if to migrate a machine while it is in SLOF between BARs assigned and USB driver started working, the BAR memory regions were not enabled and BAR accesses were ending up in unassigned_mem_ops. -- Alexey