Re: [Qemu-devel] [PATCH 4/5] ivshmem: use symbolic constant for PCI ID, add to pci-ids.txt

[Michael S. Tsirkin]

On Wed, Dec 12, 2012 at 03:51:16PM +0100, Paolo Bonzini wrote:
> Il 12/12/2012 15:48, Michael S. Tsirkin ha scritto:
> >>>>> > >> > diff --git a/hw/pci.h b/hw/pci.h
> >>>>> > >> > index 0719521..3704d5f 100644
> >>>>> > >> > --- a/hw/pci.h
> >>>>> > >> > +++ b/hw/pci.h
> >>>>> > >> > @@ -78,6 +78,7 @@
> >>>>> > >> >  #define PCI_DEVICE_ID_VIRTIO_SCSI        0x1004
> >>>>> > >> >  #define PCI_DEVICE_ID_VIRTIO_RNG         0x1005
> >>>>> > >> >  #define PCI_DEVICE_ID_VIRTIO_9P          0x1009
> >>>>> > >> > +#define PCI_DEVICE_ID_QUMRANET_IVSHMEM   0x1110
> >>>>> > >> >  
> >>>>> > >> >  #define FMT_PCIBUS                      PRIx64
> >>> > > Why _QUMRANET I wonder?
> >>> > > 
> >> > 
> >> > Because it's under 0x1af4 (PCI_VENDOR_ID_REDHAT_QUMRANET), not 0x1b36
> >> > (PCI_VENDOR_ID_REDHAT).
> > 
> > Yes but so are all virtio devices are they not?
> 
> All device IDs start with PCI_DEVICE_ID_<vendor>.  virtio devices are
> the exception for some historical reason I don't know.
> PCI_DEVICE_ID_VIRTIO_IVSHMEM is wrong, and PCI_DEVICE_ID_IVSHMEM is
> unspecific.  PCI_DEVICE_ID_QEMU_IVSHMEM is wrong because the QEMU vendor
> id is 0x1234.
> 
> So either it stays Qumranet, or the 0x1234 vendor id is renamed to
> PCI_VENDOR_ID_BOCHS and 0x1af4 becomes PCI_VENDOR_ID_REDHAT_QEMU.
> Either is fine for me, but I wanted to minimize the churn.
> 
> Paolo

Thinking about it some more, I'd rather move this stuff
out of pci.h

For virtio, there's actually linux/virtio_ids.h
which defines VIRTIO_ID_NET etc. We should add that,
and have virtio just use it.

Doesn't ivshmem have a linux driver?
When it lands upstream we'll be able to add it to pci_ids.h
meanwhile keeping a number in device .c seems fine as well.

I will do the virtio change I think - meanwhile
could you limit this patch to just the .txt change please?

-- 
MST