Re: xen 4.17.2 qemu-system-i386 crashes

[xen@xxxxxxxx]

Am 22.08.23 um 01:14 schrieb xen@xxxxxxxx:
> Am 21.08.23 um 22:15 schrieb Chuck Zmudzinski:
> > On 8/18/2023 10:49 PM, Chuck Zmudzinski wrote:
> > > On 8/18/2023 7:42 PM, xen@xxxxxxxx wrote:
> > > > Am 18.08.23 um 02:32 schrieb Chuck Zmudzinski:
> > > > > On 8/17/23 7:22 PM, xen@xxxxxxxx wrote:
> > > > > > Hello,
> > > > > >
> > > > > > I am using Archlinux. Last month I built and installed xen-4.17.2pre
> > > > > > which works perfectly. Today I built and installed the newly released
> > > > > > xen-4.17.2.
> > > > > > Now qemu-system-i386 crashes.
> > > > > >
> > > > > > I attached an excerpt from the log.
> > > > > Hmm. I am on Fedora 38, and they just pushed out the 4.17.2 update 
> > > > > to me
> > > > > yesterday and qemu-system-i386 is working fine with it. This is the 
> > > > > version
> > > > > info on Fedora 38 that is working:
> > > > >
> > > > > Xen version info from sudo xl dmesg on Fedora 38:
> > > > >
> > > > > (XEN) Xen version 4.17.2 (mockbuild@) (gcc (GCC) 13.2.1 20230728 
> > > > > (Red Hat 13.2.1-1)) debug=n Thu Aug 10 22:11:44 UTC 2023
> > > > > Qemu version info on Fedora 38:
> > > > >
> > > > > [user@fedora ~]$ qemu-system-i386 -version
> > > > > QEMU emulator version 7.2.1 (qemu-7.2.1-2.fc38)
> > > > > Copyright (c) 2003-2022 Fabrice Bellard and the QEMU Project 
> > > > > developers
> > > > > [user@fedora ~]$
> > > > >
> > > > > So, I think maybe you are using a different Qemu version that does not
> > > > > work with Xen 4.17.2.
> > > With the new data this suggestion I made yesterday seems wrong now.
> > >
> > > > Xen 4.17.2pre runs on this hardware with both Qemu 8.0.2 and Qemu 8.0.4
> > > > without problems>
> > > > On older hardware (Intel Haswell processor) Xen 4.17.2 runs without
> > > > problems with Qemu 8.0.4. Obviously only my newer hardware is affected.
> > > My working result is also on older hardware (Intel Haswell processor) so
> > > we agree the problem you are seeing is related to newer hardware. It
> > > might help to mention the model of processor in your newer hardware.
> It is an Intel i9 Coffee Lake processor.
>
> > > Unfortunately I do not have newer hardware to test my Qemu 7.2.x with
> > > Xen 4.17.2 so I won't be able to help you out much any more. (Today
> > > Fedora pushed out Qemu 7.2.4 to me and that also works fine with Xen 
> > > 4.17.2
> > > on the Haswell processor).
> > >
> > > With this additional data it seems the changes in Xen (not changes in 
> > > Qemu)
> > > are causing the core dump in Qemu on the newer hardware. If no one here
> > > on xen-users has any ideas you might try posting this bug on the 
> > > xen-devel
> > > mailing list instead. Mention Haswell always works, but something goes
> > > bad with Qemu when upgrading from 4.17.2pre to 4.17.2 on the newer
> > > hardware, and mention the model of processor of the newer hardware and
> > > the excerpt from the journal, and maybe some of the developers can help
> > > you troubleshoot the problem. There presumably are not a lot of changes
> > > between Xen 4.17.2pre and 4.17.2 so hopefully the developers can find 
> > > the
> > > problem and fix it for the next release.
> > To follow up, I just checked the shortlog of the 4.17 branch on 
> > xenbits.xen.org:
> > https://xenbits.xen.org/gitweb/?p=xen.git;a=shortlog;h=refs/heads/stable-4.17
> >
> > and there are I think 80 commits (give or take a few) between the 
> > 4.17.2pre
> > version commit and the 4.17.2 release. So you should be able to bisect 
> > and
> > find the offending commit that is causing qemu-system-i386 to crash on 
> > your
> > newer hardware, but it may take a few builds to narrow it down with about
> > 80 commits to check. If you find the offending commit, I suggest you 
> > report
> > it on the xen-devel mailing list so the developers can diagnose the cause
> > and find a solution.
> I will try with git bisect, but of course this takes some time.
I found the offending commit, it is b9ade05c
https://xenbits.xenproject.org/gitweb/?p=xen.git;a=commit;h=b9ade05cbba977673d5a08bc7a5940c5fd8add0e

Maybe it has something to do with a microcode update from August 2023. 
First I need to check if this update is already installed.