On 10/25/16 21:24, Laszlo Ersek wrote: > On 10/25/16 20:42, Alex Williamson wrote:
>> FWIW, I think the reason >> this hasn't been done to date is that PCI bus addresses (except for >> root bus devices) are not stable. Depending on the system, the address >> of a given device may change, not only based on the slot where the >> device is installed, but whether other devices in other slots are >> populated. > > I agree. > > However, while the addresses are not stable in the face of hardware > changes, I think the addresses don't change haphazardly (that is, > without hardware changes). > > So, if you plug in another card, your current pci-stub.except=... > parameter might become invalid; but that's not very different from the > case when you plug in the second instance of a preexistent card right > now -- then the pci-stub.ids=... filter won't match uniquely anymore, > and assignment vs. host-side use might not work as intented. Sorry about the self-followup; I just wanted to add -- although it might not carry a lot of weight for the host kernel -- that the libvirt domain XML hard-codes the host-side PCI BDF anyway, for assigned devices: http://libvirt.org/formatdomain.html#elementsHostDev """ <devices> <hostdev mode='subsystem' type='pci' managed='yes'> <source> <address domain='0x0000' bus='0x06' slot='0x02' function='0x0'/> </source> ... </hostdev> </devices> ... source The source element describes the device as seen from the host using the following mechanism to describe: ... pci PCI devices can only be described by their address. """ Thanks Laszlo
