On Thursday 11 December 2014 13:26:48 Daniel P. Berrange wrote: > On Thu, Dec 11, 2014 at 01:24:17PM +0000, Daniel P. Berrange wrote: > > On Thu, Dec 11, 2014 at 11:32:14AM +0100, Peter Krempa wrote: > > > On 12/10/14 14:30, Dmitry Guryanov wrote: > > > > Use guest disk name to determine disk position on > > > > bus, because Openstack/nova don't set virDomainDeviceInfo. > > > > > > > > Signed-off-by: Dmitry Guryanov <dguryanov@xxxxxxxxxxxxx> > > > > --- > > > > > > > > src/parallels/parallels_sdk.c | 10 +++++++++- > > > > 1 file changed, 9 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/src/parallels/parallels_sdk.c > > > > b/src/parallels/parallels_sdk.c > > > > index e2a1e6c..c85e6d9 100644 > > > > --- a/src/parallels/parallels_sdk.c > > > > +++ b/src/parallels/parallels_sdk.c > > > > @@ -2467,6 +2467,7 @@ static int prlsdkAddDisk(PRL_HANDLE sdkdom, > > > > virDomainDiskDefPtr disk)> > > > > > > int ret = -1; > > > > PRL_VM_DEV_EMULATION_TYPE emutype; > > > > PRL_MASS_STORAGE_INTERFACE_TYPE sdkbus; > > > > > > > > + int idx; > > > > > > > > if (prlsdkCheckDiskUnsupportedParams(disk) < 0) > > > > > > > > return -1; > > > > > > > > @@ -2535,7 +2536,14 @@ static int prlsdkAddDisk(PRL_HANDLE sdkdom, > > > > virDomainDiskDefPtr disk)> > > > > > > pret = PrlVmDev_SetIfaceType(sdkdisk, sdkbus); > > > > prlsdkCheckRetGoto(pret, cleanup); > > > > > > > > - pret = PrlVmDev_SetStackIndex(sdkdisk, > > > > disk->info.addr.drive.target); > > > > + idx = virDiskNameToIndex(disk->dst); > > > > > > Shouldn't this be used only if disk->info.addr.drive.target doesn't > > > contain any info? Does anybody ever set it meaningfully so that it does > > > work? > > > > Strictly speaking the right thing todo would be to have a method called > > before starting the guest, which populates the disk->info.addr based > > on disk->dst, if the user hasn't already provided that info. Then this > > particularly code here would be unchanged. eg something needs to be > > calling virDomainDiskDefAssignAddress > > And actually, the domain XML parser is calling that when no address is > set, so the address should be present. That said, shouldn't this code > be using the 'unit' attribute rather than the 'target' attribute, or > a combo of both of them, depending onthe controller type. > Sorry, missed this message. You're right, I've sent new version of patches. > Regards, > Daniel -- Dmitry Guryanov -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list