Re: [libvirt] [PATCH v2 1/2] qemu: Check for existing hostdev address for cold attach device

On 06/13/2018 07:16 AM, John Ferlan wrote: The difference with this is that any existing domain with this error would have failed to start qemu in previous versions, so it surely would have been fixed. And assignment of addresses to PCI devices that are "coldplugged" (ie those that are added individually with AttachDevice, but that are only added to the persistent config, *not* to the running guest, ie "CONFIG") is handled by a manually added call to virDomainDefPostParse() that is called directly from qemuDomainAttachDeviceConfig:         qemuDomainAttachDeviceConfig           virDomainDefPostParse             qemuDomainDefAssignAddresses (aka assignAddressCallback)               qemuDomainAssignAddresses Is there a reason why SCSI addresses aren't assigned by the same function. If we did that then at least everyone would have the same set of problems :-P) On the other hand, for hotplugged PCI devices ("LIVE"), both assigning addresses to devices with no supplied address, and checking for duplicate addresses is handled by qemuDomainEnsurePCIAddress (which calls virDomainPCIAddressEnsureAddr. No, I did not name either of them) - that is called directly from qemuDomainAttach${Blah}Device in qemu_hotplug.c. So in both cases, nothing associated with *PCI* device addresses is done during the *Device* post parse callback, although in the case of CONFIG, addresses are assigned/checked by a "POST-postparse" manual call to the *Domain* PostParse callback. Sigh. I was hoping to arrive at some nugget of good information that would point to the best solution. Instead I've just pointed out how convoluted everything is, and given myself more ideas of other possible bugs. Maybe if I look at it again in the morning with a fresh brain...