[REGRESSION 5.0.x] Windows XP broken on KVM

[REGRESSION 5.0.x] Windows XP broken on KVM

[Takashi Iwai]

Hi,

we've got a regression report on the recent 5.0.x kernel, starting
from 5.0.6, where Windows XP can't boot on KVM any longer.

The culprit seems to be the patch
   KVM: x86: update %rip after emulating IO
with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
Reverting this alone fixed the problem.

The report is found at openSUSE bugzilla:
  https://bugzilla.suse.com/show_bug.cgi?id=1132694

Is there already a followup fix?  If not, we need to revert it from
stable, at least.


thanks,

Takashi

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Greg Kroah-Hartman]

On Thu, Apr 18, 2019 at 09:38:52AM +0200, Takashi Iwai wrote:
> Hi,
> 
> we've got a regression report on the recent 5.0.x kernel, starting
> from 5.0.6, where Windows XP can't boot on KVM any longer.
> 
> The culprit seems to be the patch
>    KVM: x86: update %rip after emulating IO
> with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> Reverting this alone fixed the problem.
> 
> The report is found at openSUSE bugzilla:
>   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> 
> Is there already a followup fix?  If not, we need to revert it from
> stable, at least.

Is this also a problem in 5.1-rc5?

Those KVM tests being integrated by Linaro into the stable testing
process can't come fast enough, I was told yesterday that they would be
in there "soon"...

thanks,

greg k-h

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Takashi Iwai]

On Thu, 18 Apr 2019 09:53:28 +0200,
Greg Kroah-Hartman wrote:
> 
> On Thu, Apr 18, 2019 at 09:38:52AM +0200, Takashi Iwai wrote:
> > Hi,
> > 
> > we've got a regression report on the recent 5.0.x kernel, starting
> > from 5.0.6, where Windows XP can't boot on KVM any longer.
> > 
> > The culprit seems to be the patch
> >    KVM: x86: update %rip after emulating IO
> > with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> > Reverting this alone fixed the problem.
> > 
> > The report is found at openSUSE bugzilla:
> >   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> > 
> > Is there already a followup fix?  If not, we need to revert it from
> > stable, at least.
> 
> Is this also a problem in 5.1-rc5?

Only 5.0.x has been tested, so far.
I'll ask the reporter to try 5.1-rc if Paolo can't reproduce the
problem in his side.


thanks,

Takashi

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Sean Christopherson]

On Thu, Apr 18, 2019 at 10:05:34AM +0200, Takashi Iwai wrote:
> On Thu, 18 Apr 2019 09:53:28 +0200,
> Greg Kroah-Hartman wrote:
> > 
> > On Thu, Apr 18, 2019 at 09:38:52AM +0200, Takashi Iwai wrote:
> > > Hi,
> > > 
> > > we've got a regression report on the recent 5.0.x kernel, starting
> > > from 5.0.6, where Windows XP can't boot on KVM any longer.
> > > 
> > > The culprit seems to be the patch
> > >    KVM: x86: update %rip after emulating IO
> > > with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> > > Reverting this alone fixed the problem.
> > > 
> > > The report is found at openSUSE bugzilla:
> > >   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> > > 
> > > Is there already a followup fix?  If not, we need to revert it from
> > > stable, at least.
> > 
> > Is this also a problem in 5.1-rc5?
> 
> Only 5.0.x has been tested, so far.
> I'll ask the reporter to try 5.1-rc if Paolo can't reproduce the
> problem in his side.

I'd be surprised if 5.1-rc* changes anything relative to 5.0.x, the
commit in question is fairly isolated and specific.

Can we get the Qemu version and command line that is being run?

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Takashi Iwai]

On Thu, 18 Apr 2019 10:05:34 +0200,
Takashi Iwai wrote:
> 
> On Thu, 18 Apr 2019 09:53:28 +0200,
> Greg Kroah-Hartman wrote:
> > 
> > On Thu, Apr 18, 2019 at 09:38:52AM +0200, Takashi Iwai wrote:
> > > Hi,
> > > 
> > > we've got a regression report on the recent 5.0.x kernel, starting
> > > from 5.0.6, where Windows XP can't boot on KVM any longer.
> > > 
> > > The culprit seems to be the patch
> > >    KVM: x86: update %rip after emulating IO
> > > with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> > > Reverting this alone fixed the problem.
> > > 
> > > The report is found at openSUSE bugzilla:
> > >   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> > > 
> > > Is there already a followup fix?  If not, we need to revert it from
> > > stable, at least.
> > 
> > Is this also a problem in 5.1-rc5?
> 
> Only 5.0.x has been tested, so far.
> I'll ask the reporter to try 5.1-rc if Paolo can't reproduce the
> problem in his side.

The reporter confirmed that the problem is present with the latest
5.1-rc kernel, too.

FWIW, it's a Ryzen system.  Check the bugzilla URL above for details.


thanks,

Takashi

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Paolo Bonzini]

On 18/04/19 10:05, Takashi Iwai wrote:
> On Thu, 18 Apr 2019 09:53:28 +0200,
> Greg Kroah-Hartman wrote:
>>
>> On Thu, Apr 18, 2019 at 09:38:52AM +0200, Takashi Iwai wrote:
>>> Hi,
>>>
>>> we've got a regression report on the recent 5.0.x kernel, starting
>>> from 5.0.6, where Windows XP can't boot on KVM any longer.
>>>
>>> The culprit seems to be the patch
>>>    KVM: x86: update %rip after emulating IO
>>> with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
>>> Reverting this alone fixed the problem.
>>>
>>> The report is found at openSUSE bugzilla:
>>>   https://bugzilla.suse.com/show_bug.cgi?id=1132694
>>>
>>> Is there already a followup fix?  If not, we need to revert it from
>>> stable, at least.
>>
>> Is this also a problem in 5.1-rc5?
> 
> Only 5.0.x has been tested, so far.
> I'll ask the reporter to try 5.1-rc if Paolo can't reproduce the
> problem in his side.

Back from vacation.  I'll try AMD, that's the most likely culprit.

Paolo

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Harald Arnesen]

Greg Kroah-Hartman [18.04.2019 09:53]:

> On Thu, Apr 18, 2019 at 09:38:52AM +0200, Takashi Iwai wrote:
>> Hi,
>> 
>> we've got a regression report on the recent 5.0.x kernel, starting
>> from 5.0.6, where Windows XP can't boot on KVM any longer.
>> 
>> The culprit seems to be the patch
>>    KVM: x86: update %rip after emulating IO
>> with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
>> Reverting this alone fixed the problem.
>> 
>> The report is found at openSUSE bugzilla:
>>   https://bugzilla.suse.com/show_bug.cgi?id=1132694
>> 
>> Is there already a followup fix?  If not, we need to revert it from
>> stable, at least.
> 
> Is this also a problem in 5.1-rc5?

My previously installed Windows XP boots and runs fine in 5.1-rc5.
-- 
Hilsen Harald

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Paolo Bonzini]

On 18/04/19 09:38, Takashi Iwai wrote:
> Hi,
> 
> we've got a regression report on the recent 5.0.x kernel, starting
> from 5.0.6, where Windows XP can't boot on KVM any longer.
> 
> The culprit seems to be the patch
>    KVM: x86: update %rip after emulating IO
> with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> Reverting this alone fixed the problem.
> 
> The report is found at openSUSE bugzilla:
>   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> 
> Is there already a followup fix?  If not, we need to revert it from
> stable, at least.

No, it's the first time I hear this and I actually test Windows XP
before every pull request I send to Linus...  I'll download 5.0.x and
test it there.

Paolo

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Greg Kroah-Hartman]

On Thu, Apr 18, 2019 at 09:56:02AM +0200, Paolo Bonzini wrote:
> On 18/04/19 09:38, Takashi Iwai wrote:
> > Hi,
> > 
> > we've got a regression report on the recent 5.0.x kernel, starting
> > from 5.0.6, where Windows XP can't boot on KVM any longer.
> > 
> > The culprit seems to be the patch
> >    KVM: x86: update %rip after emulating IO
> > with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> > Reverting this alone fixed the problem.
> > 
> > The report is found at openSUSE bugzilla:
> >   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> > 
> > Is there already a followup fix?  If not, we need to revert it from
> > stable, at least.
> 
> No, it's the first time I hear this and I actually test Windows XP
> before every pull request I send to Linus...  I'll download 5.0.x and
> test it there.

Any further ideas about this?

thanks,

greg k-h

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Sean Christopherson]

On Tue, Apr 23, 2019 at 10:23:17PM +0200, Greg Kroah-Hartman wrote:
> On Thu, Apr 18, 2019 at 09:56:02AM +0200, Paolo Bonzini wrote:
> > On 18/04/19 09:38, Takashi Iwai wrote:
> > > Hi,
> > > 
> > > we've got a regression report on the recent 5.0.x kernel, starting
> > > from 5.0.6, where Windows XP can't boot on KVM any longer.
> > > 
> > > The culprit seems to be the patch
> > >    KVM: x86: update %rip after emulating IO
> > > with the upstream commit 45def77ebf79e2e8942b89ed79294d97ce914fa0.
> > > Reverting this alone fixed the problem.
> > > 
> > > The report is found at openSUSE bugzilla:
> > >   https://bugzilla.suse.com/show_bug.cgi?id=1132694
> > > 
> > > Is there already a followup fix?  If not, we need to revert it from
> > > stable, at least.
> > 
> > No, it's the first time I hear this and I actually test Windows XP
> > before every pull request I send to Linus...  I'll download 5.0.x and
> > test it there.
> 
> Any further ideas about this?

I followed up on the bugzilla to request more information.  My best guess
at this point is that the issue is related to an older version of Qemu or
a specific emulated device, but without additional details we're stuck.

Re: [REGRESSION 5.0.x] Windows XP broken on KVM

[Harald Arnesen]

Takashi Iwai [18.04.2019 09:38]:

> Hi,
> 
> we've got a regression report on the recent 5.0.x kernel, starting
> from 5.0.6, where Windows XP can't boot on KVM any longer.

Not for all configurations. I just checked with 5.0.8, and Windows XP
boots just fine on KVM.
-- 
Hilsen Harald