[Qemu-devel] Re: [PATCH 0/2] fix virtio_blk serial pci config breakage

[Anthony Liguori <anthony@codemonkey.ws>]

john cooper wrote:
> This is a correction to the previous version
> which violated the PCI config space 256 byte
> limitation.
>
> The crux of the feature is simply passing a qemu
> received virtio serial number to/through the guest
> driver and making it available to guest userspace
> via a preexisting interface.
>
> To accomplish this, ATA IDENTIFY data as implemented
> by the HDIO_GET_IDENTITY ioctl is returned to the
> caller for the virtio_blk drive of interest.  Content
> of this identify structure is created by qemu and
> passed wholesale to guest userspace without
> interpretation by the guest driver.
>
> The change this patch implements is passing of the
> identify data through a mapping established by
> PCI BAR #5 rather than the PCI config area, the
> latter of which resulted in the above breakage.
>   

This is a massive layering violation.  The virtio-blk ABI cannot make 
demands of the transport.

The better solution would be to move the entire virtio-pci config space 
to a separate BAR that's an MMIO region.  Then there is no practical 
limit on the size of the config area.

Regards,

Anthony Liguori