Re: [libvirt] [PATCH v6 09/13] conf: Allocate/release 'uid' and 'fid' in PCI address

On Tue, 2018-10-16 at 11:28 +0800, Yi Min Zhao wrote: > 在 2018/10/11 下午10:50, Andrea Bolognani 写道: >> On Fri, 2018-09-28 at 16:46 +0800, Yi Min Zhao wrote: >>> # conf/device_conf.h >>> +virDeviceInfoPCIAddressExtensionIsPresent; >>> +virDeviceInfoPCIAddressExtensionIsWanted; >>> virDeviceInfoPCIAddressIsPresent; >>> virDeviceInfoPCIAddressIsWanted; >>> virDomainDeviceAddressIsValid; >>> @@ -119,6 +121,9 @@ virDomainCCWAddressSetFree; >>> virDomainPCIAddressBusIsFullyReserved; >>> virDomainPCIAddressBusSetModel; >>> virDomainPCIAddressEnsureAddr; >>> +virDomainPCIAddressExtensionReleaseAddr; >>> +virDomainPCIAddressExtensionReserveAddr; >>> +virDomainPCIAddressExtensionReserveNextAddr; >> I'm not quite quire we need to export these functions. >> >> With the recent changes, we've gotten to the point that we're not >> even passing a virZPCIDeviceAddress to them, but rather they have >> the very same signature as the regular virPCIDeviceAddress... >> >> So it should be possible to just make them static and only call >> them from the virDomainPCIAddressReserveAddr() and friends, right? >> Which is where I was hoping we could eventually get. Or did I >> miss something? > I think this would make things complex. If either PCI address or > zPCI address exists, we have to do more checks for calling > virDomainPCIAddressReserveAddr(). And there are amounts of > code calling ***IsWanted() to call ***ReserveNext***(). I think > keeping them separately is better. Again, I might be missing something because I haven't actually tried implementing any of this, but at least from the theoretical point of view I don't see how keeping them separate would make things simpler: if anything, it seems to me like it would make them more complicated for the calling code because now you have to worry about the PCI address extensions *in addition* to the PCI address itself.