[libvirt] [RFC PATCH] qemu: Add suport for pci-assign.configfd option

[Alex Williamson]

On Wed, 2010-05-19 at 22:34 -0700, Chris Wright wrote:
> * Alex Williamson (alex.williamson at redhat.com) wrote:
> > @@ -1378,6 +1420,9 @@ int qemudExtractVersionInfo(const char *qemu,
> >                            &version, &is_kvm, &kvm_version) == -1)
> >          goto cleanup2;
> >  
> > +    if (flags & QEMUD_CMD_FLAG_DEVICE)
> > +        qemudParsePCIDeviceStrs(qemu, &flags);
> > +
> 
> looks like this is called early enough for both cmdline and monitor
> interface

Yes, it should be.

> > +    /* XXX don't hard code segment */
> 
> You don't need to, just...
> 
> > +    if (virAsprintf(&path, "/sys/bus/pci/devices/0000:%02x:%02x.%01x/config",
>                                                    /
> s/0000/%04x/ --------------------------------------
> 
> > +                    dev->source.subsys.u.pci.bus, dev->source.subsys.u.pci.slot,
> 
> and add dev->source.subsys.u.pci.domain 

Oh good, I meant to look, but obviously forgot.  Fixed.

> > +                    dev->source.subsys.u.pci.function) < 0) {
> > +        virReportOOMError();
> > +        return -1;
> > +    }
> > +
> > +    configfd = open(path, O_RDWR, 0);
> 
> Should probably report an open failure to aid debugging.

Done.

> > --- a/src/qemu/qemu_driver.c
> > +++ b/src/qemu/qemu_driver.c
> > @@ -7567,7 +7567,7 @@ static int qemudDomainAttachHostPciDevice(struct qemud_driver *driver,
> >          if (qemuDomainPCIAddressEnsureAddr(priv->pciaddrs, &hostdev->info) < 0)
> >              goto error;
> >  
> > -        if (!(devstr = qemuBuildPCIHostdevDevStr(hostdev)))
> > +        if (!(devstr = qemuBuildPCIHostdevDevStr(hostdev, -1)))
> 
> Looks like this is dynamically assigning host PCI device to running guest,
> we should still be able to pass fd here.  That will complicate the qemu
> side to be able to accept the fd.

We could certainly open the file and pass an fd, but I don't know how to
let qemu make use of it after it's been spawned, so I reverted to
existing behavior here.  Thanks,

Alex