On Fri, 26 May 2023, Jan Beulich wrote: > On 25.05.2023 21:32, Stefano Stabellini wrote: > > Like I wrote, personally I am happy with whatever gets us to have the PVH > > test in gitlab-ci faster. > > > > However, on the specific problem of PCI devices used by Xen and how to > > deal with them for Dom0 PVH, I think they should be completely hidden. > > Hidden in the sense that they don't appear on the Dom0 PCI bus. If the > > hidden device is a function of a multi-function PCI device, then the > > entire multi-function PCI device should be hidden. > > > > I don't think this case is very important because devices used by Xen > > are timers, IOMMUs, UARTs, > > ... USB debug ports (EHCI, XHCI), ... > > > all devices that typically are not multi-function, > > except for the ones added. Furthermore see video_endboot() for a case > of also hiding the VGA device, which isn't unlikely to have secondary > functions (sound controllers are not uncommon). Hence ... > > > so it is OK to be extra careful and remove the entire > > device from Dom0 in the odd case that the device is both multi-function > > and only partially used by Xen. This is what I would do for Xen on ARM > > too. > > ... at best I would see this as a non-default mode of operation. Of > course we could also play more funny games with vPCI, like surfacing > a "stub" device in place of a hidden one, so the other functions can > still be found.
>From our use-case point of view, non-default is OK. The PCI stub idea is a cool trick that might work but hopefully we won't need it at least initially. But it is something to consider if it turns out there is an important multi-function device with one of the devices hidden.
