Re: [Qemu-devel] [PATCH v8 5/7] libqos: Added test case for configuration changes in virtio-blk test

["Marc Marí" <marc.mari.barcelo@gmail.com>]

El Mon, 1 Sep 2014 18:27:34 +0200
Marc Marí <marc.mari.barcelo@gmail.com> escribió:
> El Mon, 1 Sep 2014 18:09:09 +0200
> Greg Kurz <gkurz@linux.vnet.ibm.com> escribió:
> > On Mon,  1 Sep 2014 12:07:58 +0200
> > Marc Marí <marc.mari.barcelo@gmail.com> wrote:
> > > Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > Signed-off-by: Marc Marí <marc.mari.barcelo@gmail.com>
> > > ---
> > 
> > Hi Marc,
> > 
> > I gave it a try for various host/target combinations involving ppc64
> > and x86_64. Here is what I get:
> > 
> >         |   x86_64   |    ppc64    | TARGET
> >         +------------+-------------+--------
> > x86_64  |     OK     | assert  (1) |
> > --------+------------+-------------+
> > ppc64   | assert (2) | assert  (1) |
> > --------+------------+-------------+
> > ppc64le |     OK     | assert  (1) |
> > --------+------------+-------------+
> >   HOST  |
> > 
> > where OK means:
> > 
> > /virtio/blk/pci/basic: OK
> > /virtio/blk/pci/indirect: OK
> > /virtio/blk/pci/config: OK
> > /virtio/blk/pci/msix: OK
> > /virtio/blk/pci/idx: OK
> > 
> > and assert (1) is:
> > 
> > tests/virtio-blk-test.c:87:virtio_blk_init: assertion failed:
> > (dev != NULL)
> > 
> > and assert (2) is:
> > 
> > tests/virtio-blk-test.c:171:pci_basic: assertion failed (status ==
> > 0): (2 == 0)
> > 
> > I will investigate further but the first column in the array ^^
> > seems to indicate that there's some endianness bug.
> > 
> > Cheers.
> > 
> > --
> > Greg
> 
> Hi
> 
> I will also investigate. Just add that with assert (2), status
> is 2, which means status is VIRTIO_BLK_S_UNSUPP, which means "request
> unsupported by host". So, for some reason, ppc64 host does not support
> a simple write request.
> 
> Thanks for your tests
> Marc

Hi

The problem seems to be that it is not implemented. virtio-blk-test is
now using pci-pc, that has the addresses 0xcf8 and 0xcfc hardcoded.
These addresses are the key-value for PCI space, to search for devices,
and I suppose it is specific to i386 and x86_64.

I also searched information to see how easy it is to add another
architecture and I could not find much information on ppc64, and even
less when mixing ppc64 and PCI.

Thanks
Marc