On Wed, 2016-07-20 at 14:17 +0100, Daniel P. Berrange wrote: > On Tue, Jul 19, 2016 at 07:04:36PM +0200, Andrea Bolognani wrote: > > > > On Tue, 2016-07-19 at 16:17 +0200, Andrea Bolognani wrote: > > > > > > Commit 8156493d8db9 changed libvirt so that '-device pci-ohci' > > > would be used instead of '-usb' on ppc64 when no specific USB > > > controller model had been specified in the guest configuration. > > > > > > While the device that ends up being presented by the guest is > > > exactly the same, '-usb' causes it to be assigned to PCI address > > > 00:00.0 while '-device pci-ohci', being subject to the regular > > > PCI address assignment logic, will be at a different address. > > > > > > This PCI address mismatch breaks migration of existing guests > > > to new libvirt versions. > > > > > > Luckily, when QEMU has switched its default '-usb' controller > > > from pci-ohci to nec-usb-xhci (QEMU commit 57040d451315), it > > > has done so without affecting older machine types, which means > > > we can keep using '-usb' without risking guest ABI breakage. > > > > > > Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1357468 > > > --- > > > src/qemu/qemu_command.c | 8 +++++--- > > > tests/qemuxml2argvdata/qemuxml2argv-ppc64-usb-controller.args | 2 +- > > > tests/qemuxml2argvdata/qemuxml2argv-pseries-usb-default.args | 2 +- > > > 3 files changed, 7 insertions(+), 5 deletions(-) > > > > Ján pointed out that applying this would un-break migration > > for ppc64 guests created before 8156493d8db9, but at the > > same time break migration for those created since, eg. in > > the last ~6 months. > > > > I'm kinda out of ideas here, so if anyone has a brilliant > > plan to make migration work for both old and new guests, > > please do share :) > > It seems that we could solve this by just changing the logic in > qemuDomainAssignDevicePCISlots() so that when it is auto-assigning > addresses, it always uses 00:00.0 for the USB controller when guest > arch is ppc64. > > Existing guests deployed from a libvirt version using -device won't > be affected, because we'll have recorded a PCI address for that > in the XML and continue to use that. ie the auto-allocation logic > won't run for them. > > Existing guests deployed from a libvirt version using -usb should > then get the fixed PCI address of 00:00.0 when they upgrade to the > fix libvirt (assuming they didn't get run with a broken libvirt > in between). Mh, that could actually work! Thanks :) I'll try to cook up a patch tomorrow. > Regardless though, I'm very much against any change that tales us > back to using -usb. This is long since obsolete syntax we should > aim to never use - bringing ppc inline with other arches using > -device is very much desirable. Agreed - that was the reasoning behind the original, sadly flawed, attempt to get rid of it. -- Andrea Bolognani / Red Hat / Virtualization -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list