Re: [libvirt] [PATCH] qemu: auto-create pci controller alias if missing from domain status

https://bugzilla.redhat.com/show_bug.cgi?id=1047429 The problem occurs when a host with a running guest is upgraded from pre-1.0.2 libvirt, and then a later attempt is made to hotplug a new PCI device (emulated or passthrough) to that guest without having power cycled / save-restored / migrated the guest. The hotplug fails with this error: internal error: Device alias was not set for PCI controller with index 0 required for device at address 0000:00:xx.x

I'm not 100% convinced that it's necessary to handle *this* case either (upgrading from pre-1.0.2), but thought I should send the patch to the list anyway to see what others think. My opinion is that it's creating permanent clutter in the code in order to handle a situation that will come up just once in the entire lifetime of a host (and even then only if libvirt is upgraded without somehow restarting a guest, and *then* a device is hotplugged to that guest), so it may be better if we just left the workaround ("restart/save-restore/migrate the domain if you need to hotplug a device") as a footnote somewhere instead.

+ /* this could be a case where the domain had been + * started with an earlier version of libvirt that + * had no device entry for the PCI bus + * (pre-1.0.2), and in that case no alias would be + * set in the status XML, so we create one using + * the standard "pci.N" pattern. + */ + if (virAsprintf(&tempAlias, "pci.%u", info->addr.pci.bus) < 0) + goto cleanup; + contAlias = tempAlias;

} break; } @@ -3118,6 +3123,7 @@ qemuBuildDeviceAddressStr(virBufferPtr buf, ret = 0; cleanup: VIR_FREE(devStr); + VIR_FREE(tempAlias);