Linux Virtualization - Feb 2013 - [Qemu-devel] Headsup: windows virtio networking does not work on current git

On Thu, Feb 07, 2013 at 09:53:39PM +1100, Vadim Rozenfeld
wrote:
> On Thu, 2013-02-07 at 12:18 +0200, Michael S. Tsirkin wrote:
> > On Thu, Feb 07, 2013 at 08:24:10PM +1100, Vadim Rozenfeld wrote:
> > > On Thu, 2013-02-07 at 11:33 +1030, Rusty Russell wrote:
> > > > Vadim Rozenfeld <vrozenfe at redhat.com> writes:
> > > > > On Tue, 2013-02-05 at 13:58 +0200, Michael S. Tsirkin
wrote:
> > > > >> On Tue, Feb 05, 2013 at 03:45:38PM +0400, Michael
Tokarev wrote:
> > > > >>  Is it really
> > > > >> > that bad that the config space size changed? 
Why it has this effect?
> > > > > Because in this case it's hard to distinguish
between resource's
> > > > > corruption and HW update.
> > > > 
> > > > But it's also true that if we'd incremented revid
you'd have the same
> > > > failure in this case, right?
> > > 
> > > It depends. If we have explicitly specified revision id in inf
file and
> > > this id doesn't mach the new revision id, Windows will not
try to load
> > > the "incompatible" driver, and finish up with
"device driver not found"
> > > dialog.
> > > 
> > > Best regards,
> > > Vadim.
> > >   
> > > > 
> > > > Cheers,
> > > > Rusty.
> > 
> > Well that's all in theory, in practice it does not look like
revision ID
> > is specified in the NetKVM inf so this won't work?
> > 
> > From what I see this inf specifies:
> > 
> > NetKVM/wlh/netkvm.inf:%kvmnet6.DeviceDesc%    = kvmnet6.ndi,
PCI\VEN_1AF4&DEV_1000&SUBSYS_00011AF4
> > NetKVM/wxp/netkvm.inf:%kvmnet5.DeviceDesc%    = kvmnet5.ndi,
PCI\VEN_1AF4&DEV_1000&SUBSYS_00011AF4
> > NetKVM/wxp/netkvm2k.inf:%kvmnet5.DeviceDesc%    = kvmnet5.ndi,
PCI\VEN_1AF4&DEV_1000&SUBSYS_00011AF4
> > 
> > So we can tweak any of vendor device and subsystem id.
> > 
> Unfortunately, it won't. Only balloon has revision id, specified as a
> part of device HW descriptor. But it's only because virtio doesn't
use
> revision ids. Otherwise it differential will be there. 
So your driver will load and attempt to work on rev=2 devices?
If yes it's a bug.  virtio spec specifies revision id as an ABI version.
Linux driver does:

        if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
                printk(KERN_ERR "virtio_pci: expected ABI version %d, got
%d\n",
                       VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
                return -ENODEV;
        }

> > 
> > Changing subsystem vendor ID actually will be completely
> > transparent to linux which for some reason looks at the
> > subsystem device ID (why? no idea) but not the subsystem vendor ID.
> > Of course this requires a valid vendor ID, getting this
> > costs $3000 I think.
> > We could tweak device ID too but that might break some other guests
> > which don't copy the crazy 'replace device id with subsystem
device id'
> > logic from Linux.
> > 
> 
Apropos, would you guys like to start to copy your patches to
virtualization at lists.linux-foundation.org?
If you do, you might get some review and feedback, allowing
us to catch such forward compatibility issues earlier.

Of course it's your project so entirely up to you.


-- 
MST

On Thu, 2013-02-07 at 13:33 +0200, Michael S. Tsirkin
wrote:
> On Thu, Feb 07, 2013 at 09:53:39PM +1100, Vadim Rozenfeld wrote:
> > On Thu, 2013-02-07 at 12:18 +0200, Michael S. Tsirkin wrote:
> > > On Thu, Feb 07, 2013 at 08:24:10PM +1100, Vadim Rozenfeld wrote:
> > > > On Thu, 2013-02-07 at 11:33 +1030, Rusty Russell wrote:
> > > > > Vadim Rozenfeld <vrozenfe at redhat.com> writes:
> > > > > > On Tue, 2013-02-05 at 13:58 +0200, Michael S.
Tsirkin wrote:
> > > > > >> On Tue, Feb 05, 2013 at 03:45:38PM +0400,
Michael Tokarev wrote:
> > > > > >>  Is it really
> > > > > >> > that bad that the config space size
changed?  Why it has this effect?
> > > > > > Because in this case it's hard to distinguish
between resource's
> > > > > > corruption and HW update.
> > > > > 
> > > > > But it's also true that if we'd incremented
revid you'd have the same
> > > > > failure in this case, right?
> > > > 
> > > > It depends. If we have explicitly specified revision id in
inf file and
> > > > this id doesn't mach the new revision id, Windows will
not try to load
> > > > the "incompatible" driver, and finish up with
"device driver not found"
> > > > dialog.
> > > > 
> > > > Best regards,
> > > > Vadim.
> > > >   
> > > > > 
> > > > > Cheers,
> > > > > Rusty.
> > > 
> > > Well that's all in theory, in practice it does not look like
revision ID
> > > is specified in the NetKVM inf so this won't work?
> > > 
> > > From what I see this inf specifies:
> > > 
> > > NetKVM/wlh/netkvm.inf:%kvmnet6.DeviceDesc%    = kvmnet6.ndi,
PCI\VEN_1AF4&DEV_1000&SUBSYS_00011AF4
> > > NetKVM/wxp/netkvm.inf:%kvmnet5.DeviceDesc%    = kvmnet5.ndi,
PCI\VEN_1AF4&DEV_1000&SUBSYS_00011AF4
> > > NetKVM/wxp/netkvm2k.inf:%kvmnet5.DeviceDesc%    = kvmnet5.ndi,
PCI\VEN_1AF4&DEV_1000&SUBSYS_00011AF4
> > > 
> > > So we can tweak any of vendor device and subsystem id.
> > > 
> > Unfortunately, it won't. Only balloon has revision id, specified
as a
> > part of device HW descriptor. But it's only because virtio
doesn't use
> > revision ids. Otherwise it differential will be there. 
> 
> So your driver will load and attempt to work on rev=2 devices?
All virtio-win drivers (net, serial, block, and scsi), except for
balloon will.
 
> If yes it's a bug.  virtio spec specifies revision id as an ABI
version.
> Linux driver does:
> 
>         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
>                 printk(KERN_ERR "virtio_pci: expected ABI version %d,
got %d\n",
>                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
>                 return -ENODEV;
>         }
> 
> 
> > > 
> > > Changing subsystem vendor ID actually will be completely
> > > transparent to linux which for some reason looks at the
> > > subsystem device ID (why? no idea) but not the subsystem vendor
ID.
> > > Of course this requires a valid vendor ID, getting this
> > > costs $3000 I think.
> > > We could tweak device ID too but that might break some other
guests
> > > which don't copy the crazy 'replace device id with
subsystem device id'
> > > logic from Linux.
> > > 
> > 
> 
> Apropos, would you guys like to start to copy your patches to
> virtualization at lists.linux-foundation.org?
> If you do, you might get some review and feedback, allowing
> us to catch such forward compatibility issues earlier.
> 
> Of course it's your project so entirely up to you.
> 
>