[Qemu-devel] Re: [PATCH] qemu: make virtio-blk PCI compliant by default

[Rusty Russell]

On Tue, 8 Sep 2009 05:28:31 pm Michael S. Tsirkin wrote:
> On Tue, Sep 08, 2009 at 03:40:08AM -0400, john cooper wrote:
> > Michael S. Tsirkin wrote:
> > > commit bf011293faaa7f87e4de83185931e7411b794128 made virtio-blk-pci not
> > > PCI-compliant, since it makes region 0 (which is an i/o region)
> > > size > 256, and, since PCI 2.1, i/o regions are limited to 256 bytes size.
> > > 
> > > When the ATA serial number feature is off, which is the default,
> > > make the device spec compliant again, by making region 0 smaller.
> > 
> > I'd hazard this is the cause of the breakage others
> > encountered -- even when the driver was initialized
> > but unused.  For some odd reason I hadn't seen nor
> > been able to reproduce the failure.
> > 
> > The mock-up of an entire ATA IDENTIFY page is really
> > overkill for what we're trying to accomplish here,
> > namely passing a 20 byte S/N from qemu to the guest.
> > However emulating and passing an IDENTIFY page allows
> > guest apps to interpret the information via an
> > existing interface, with the guest driver doing nothing
> > more than transferring the data as opaque.  During
> > review, other defined fields of the IDENTIFY page were
> > speculated to be potentially useful thus the entire
> > 512 byte page was passed wholesale.  But it is clearly
> > more trouble than benefit at this point.  I'll rework
> > the patch or use an alternate mechanism.
> 
> BTW, will the change you have in mind affect the guest driver as well?
> If yes, maybe revert 1d589bb16b825b3a7b4edd34d997f1f1f953033d in linux
> before 2.6.31 is out? Otherwise we'll be stuck supporting the bad
> interface from upstream 2.6.31 in upstream qemu ...

Too late, but no big deal.  We just add another feature bit for the
new method.  We can have a command virtqueue for queries such as this.

(Interestingly, this same issue broke lguest, with the 8-bit limit on
configuration space sizes).

Thanks,
Rusty.