Re: [libvirt] [PATCH 09/26] conf: Move index number checking to drivers

On 06/12/2017 10:08 PM, Laine Stump wrote: > On 06/02/2017 12:07 PM, Andrea Bolognani wrote: >> pSeries guests will soon be allowed to have multiple >> PHBs (pci-root controllers), which of course means that >> all but one of them will have a non-zero index; hence, >> we'll need to relax the current check. >> >> However, right now the check is performed in the conf >> module, which is generic rather than tied to the QEMU >> driver, and where we don't have information such as the >> guest machine type available. >> >> To make this change of behavior possible down the line, >> we need to move the check from the XML parser to the >> driver. Unfortunately, this means duplicating code :(\ > > > Maybe instead we should just eliminate this check (since the pSeries > case shows that it's an invalid assumption). And since the index is > really just something used internally by libvirt, we really don't even > need to verify that one of the pci controllers has index=0. All we > *really* care about is that there is at least one "root" PCI bus, and > that no device includes a bus# in its PCI address that isn't represented > by the index in one of the PCI controllers. > > (But for the purposes of this patch, I think we can just remove the > validation in conf and not worry about adding it elsewhere).\\ After looking at the next patch, I may have changed my mind. If we remove the validation here, then we would still have to add in validation when the commandline is generated. But I suppose it's better to give an error at domain definition time rather than domain runtime, so this makes sense. I don't think all of those drivers actually use PCI controllers though, do they? (Look for which ones actually call the functions to assign PCI addresses).