Re: PCI/VGA passthrough on >Xen-4.2 - (XEN) AMD-Vi: IO_PAGE_FAULT:

[Hans Mueller <mcbeagle@gmx.de>]

On Friday, 9. November 2012 16:33:47 Ian Campbell wrote:
> On Fri, 2012-11-09 at 16:25 +0000, Hans Mueller wrote:
> > On Thursday, 8. November 2012 20:02:51 feral wrote:
> > > Upgraded from Xen-4.1 to 4.2/4.3/4.3-unstable.  Tested against all
> > > three.
> > > 
> > > VGA passthrough previously working with 4.1 is no longer working with
> > > >4.2.
> > > 
> > > Setup - Working:
> > > Mint13, kernel-3.2.0-32, Xen-4.1 (from Ubuntu-Precise repos),
> > > libvirt-bin,
> > > xend
> > > Asus Radeon HD 7870 DirectCU.
> > > Works perfect.  Benches on par with bare metal.
> > > 
> > > Setup - Broken:
> > > Mint13, kernel-3.2.0-32, Xen-4.2,Xen-4.3,Xen-4.3-unstable (from GIT), xl
> > > toolstack.
> > > Asus Radeon HD 7870 DirectCU.
> > > Some times the guest boots correctly but display is never enabled.
> > > 
> > >  Sometimes BSOD, doesn't seem to be any rhyme or reason?
> > > 
> > > No errors in logs.   Found the following in 'xl dmesg':
> > > (XEN) AMD-Vi: IO_PAGE_FAULT: domain = 2, device id = 0x600, fault
> > > address =
> > > 0x1ac800000, flags = 0
> > 
> > I had a similar problem since hg revision 25818:50adc933faaf which changes
> > the qemu tag.
> > It looks like the qemu commit
> > 3e66da7266c84638c0e22a09c9d2b07529802576
> > (qemu-xen-trad: fix msi_translate with PV event delivery)
> > causes this behavior.
> > 
> > I solved the problem for me by using 'pci_msitranslate=0' in the domU
> > config.
> I don't follow this stuff closely but I could have sworn we had disabled
> msitranslate by default.
> 
> In any case I don't think that changeset was intended to cause anything
> like this -- Stefano?
> 
> > In addition I had to reboot (perhaps poweroff?) the dom0 once the
> > IO_PAGE_FAULT occured.

One more info for you - afair i noticed that it was not only the IO_PAGE_FAULT 
which occured but also the qemu process in dom0 crashed w/o any messages in 
the logs, perhaps feral can confirm this?