Re: [Qemu-devel] [PATCH 3/5] qdev/compat: virtio-blk-pci 0.10 compatibility.

[Michael S. Tsirkin]

On Thu, Jul 09, 2009 at 09:42:47PM +0200, Filip Navara wrote:
> On Thu, Jul 9, 2009 at 8:20 PM, Michael S. Tsirkin<address@hidden> wrote:
> > On Thu, Jul 09, 2009 at 05:09:41PM +0200, Gerd Hoffmann wrote:
> >> On 07/09/09 16:52, Michael S. Tsirkin wrote:
> >>>>       .compat_props = (CompatProperty[]) {
> >>>> +        {
> >>>> +            .driver   = "virtio-blk-pci",
> >>>> +            .property = "class",
> >>>> +            .value    = "0x0180", /* PCI_CLASS_STORAGE_OTHER */
> >>>
> >>> it seems annoying that we can't use the symbolic name. Ideas how to fix 
> >>> this?
> >>
> >> We could add a special property type instead of using hex32.  Then we
> >> can have a string <-> int mapping and use something like
> >>
> >>              .value = "storage-other",
> >>
> >> Not sure it is worth the trouble though.
> >>
> >>>> -
> >>>> -    uint16_t vendor;
> >>>> -    uint16_t device;
> >>>> -    uint16_t subvendor;
> >>>> -    uint16_t class_code;
> >>>> -    uint8_t pif;
> >>>
> >>> Are the other fields unused? If yes can be a separate patch ...
> >>
> >> Yes, they are all unused to date.  This patch puts class_code into use.
> >>
> >>>> +    if (proxy->class_code != PCI_CLASS_STORAGE_SCSI&&
> >>>> +        proxy->class_code != PCI_CLASS_STORAGE_OTHER)
> >>>> +        proxy->class_code = PCI_CLASS_STORAGE_SCSI;
> >>>> +
> >>>
> >>> what does this do?
> >>
> >> Make sure proxy->class_code has one of the two allowed values.
> >>
> >>>>       vdev = virtio_blk_init(&pci_dev->qdev);
> >>>>       virtio_init_pci(proxy, vdev,
> >>>>                       PCI_VENDOR_ID_REDHAT_QUMRANET,
> >>>>                       PCI_DEVICE_ID_VIRTIO_BLOCK,
> >>>> -                    PCI_CLASS_STORAGE_OTHER,
> >>>> -                    0x00);
> >>>> +                    proxy->class_code, 0x00);
> >>>>   }
> >>>>
> >>>>   static void virtio_console_init_pci(PCIDevice *pci_dev)
> >>>
> >>> does this mean that virtio block was broken by some previous
> >>> patch? It's not a good way to split changes: bisecting won't work.
> >>
> >> Huh?  virtio block wasn't broken.  What makes you think it was?
> >>
> >> cheers,
> >>   Gerd
> >
> > This is adding a parameter to a function. Before this patch it was missing
> > this parameter so was broken?
> >
> It is not! Two parameters removed, two added.
> 

Looks like my eyes need to be checked. Sorry about the noise.

-- 
MST