[vfio-users] UEFI Win7 VM randomly does not start

[Alexander Petrenz]

Hi Laszlo,

currently I'm using kernel package vfio-lts 4.1.12. Before it was version
4.1.10 and the one before that was something like 4.1.6 or so.
The ovmf package according to yaourt is ovmf-git 698.0bfd7c-1 and the
installed qemu package is qemu-git 2.5.r41574.g40fe17b-1.
Anyway what I want to say is, that with this packages I seem to not having
the issue.

Regards
Alexander

On Tue, Oct 20, 2015 at 12:37 AM, Laszlo Ersek <lersek at redhat.com> wrote:

> On 10/17/15 15:15, Alex Williamson wrote:
> > On Sat, 2015-10-17 at 10:50 +0200, Alexander Petrenz wrote:
> >> Hi all,
> >>
> >> I have the problem that my UEFI VM for a Windows 7 guest does not always
> >> starts properly. When it doesn´t work I have to try it again and
> finally it
> >> will work after something between 3 to 6 attempts. The only error
> message
> >> is:
> >> qemu-system-x86_64: Guest moved used index from 13 to 0
> >
> > This is a virtio issue that ring indexes are out of sync.  If you can
> > come up with a reproducible scenario on latest kernel and qemu, it
> > should be reported to the qemu-devel mailing list.  I've seen it
> > occasionally, but not consistently enough to debug.  Thanks,
>
> Very interesting; just the other day I fixed a theoretical issue in OVMF
> that could explain this -- although I had never experienced these
> symptoms caused *directly* by OVMF.
>
> The reproducer script <http://pastebin.com/yWubnz9Y> seems to use
> virtio-blk with dataplane; that could be a contributing factor.
>
> (1) I have seen the same error once with SeaBIOS:
>
>   http://code.coreboot.org/p/seabios/source/commit/5f2d17d35b23
>
> (2) I can tell from the qemu command line here that under OVMF, the
> virtio-net NIC is bound by the iPXE oprom that is bundled with QEMU.
> This is revelant because -- in addition to SeaBIOS (see above) -- I've
> also seen the same from iPXE. It is fixed now in iPXE upstream, but I'm
> unsure about the binaries bundled with QEMU.
>
>   http://thread.gmane.org/gmane.network.ipxe.devel/3918
>
>   http://git.ipxe.org/ipxe.git/commitdiff/755d2b8
>
> (3) In any case, Gerd's most recent OVMF RPMs (built at git commit
> 368002a) contain the upstream fix for the problem, *if* (and that's a
> large if!) it was indeed caused by OVMF itself:
>
>   http://thread.gmane.org/gmane.comp.bios.edk2.devel/3156
>
>   https://github.com/tianocore/edk2/commit/fbc80813eeb0
>   https://github.com/tianocore/edk2/commit/fcb636ea4244
>
> Alexander, if you can more-or-less consistently reproduce the issue,
> would you mind retesting with the most recent build from Gerd? (Please
> CC me on any replies.)
>
> Thanks
> Laszlo
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20151115/6a1bfdce/attachment.htm>