[Qemu-devel] solaris x86 in qemu?

[Qemu-devel] solaris x86 in qemu?

[Michael Tokarev]

Hello.

In order to verify some build issues on solaris, I tried to install
sol10 x86 in a kvm vm.  But unfortunately it does not work: after the
grub prompt and choosing "Solaris 10 x86" boot entry, the kernel
gets loaded (there's a row of dots displayed during that), next,
the following message gets displayed:

 SunOS Release 5.10 Version Generic_147148-26 64-bit
 Copyright (c) 1983, 2013 Oracle and/or its affiliates.  All rights reserved.

and the guest stays there for a long time, spinning up 100% of its CPU,
and nothing more happens.

The same happens when run with or without kvm (ie, tcg and kvm behaves
the same way).

When run in kvm, kvm_stats shows just a few exits (about 600/sec) and
nothing more than that.

Any idea what to do with that?

I think that supporting solaris as _guest_ OS is an important goal
for qemu/kvm (as opposed to _host_).

Thanks,

/mjt

Re: [Qemu-devel] solaris x86 in qemu? [bisected]

[Michael Tokarev]

13.06.2013 11:51, Michael Tokarev wrote:
> Hello.
> 
> In order to verify some build issues on solaris, I tried to install
> sol10 x86 in a kvm vm.  But unfortunately it does not work: after the
> grub prompt and choosing "Solaris 10 x86" boot entry, the kernel
> gets loaded (there's a row of dots displayed during that), next,
> the following message gets displayed:
> 
>  SunOS Release 5.10 Version Generic_147148-26 64-bit
>  Copyright (c) 1983, 2013 Oracle and/or its affiliates.  All rights reserved.
> 
> and the guest stays there for a long time, spinning up 100% of its CPU,
> and nothing more happens.
> 
> The same happens when run with or without kvm (ie, tcg and kvm behaves
> the same way).
> 
> When run in kvm, kvm_stats shows just a few exits (about 600/sec) and
> nothing more than that.
> 
> I think that supporting solaris as _guest_ OS is an important goal
> for qemu/kvm (as opposed to _host_).

I tried to bisect this.  It turns out that solaris x86 does not boot in
qemu/kvm for quite long time already, namely, starting from this commit:


 commit 6b034aa138716a515c88f7894940d5d0aff2f3ed
 Author: Gerd Hoffmann <kraxel@redhat.com>
 Date:   Tue Apr 17 10:51:41 2012 +0200

    seabios: update to 1.7.0

    Update roms/seabios and pc-bios/bios.bin to the 1.7.0 release.
    Most noticable new feature is virtio-scsi support.

    Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>


So I went on and tried to bisect seabios (previous version in qemu
was 1.6.3.2, and it worked).

So seabios bisection with qemu-1.1 points to this commit:

 commit 9d3d7cb4b163d3fbcba64a01c4fa42eb6bc53128
 Author: Kevin O'Connor <kevin@koconnor.net>
 Date:   Wed Sep 21 21:19:51 2011 -0400

    Move code from PCI hotplug DSDT macros to methods.

    Simplify the hotplug code by moving the bulk of the logic out of the
    macros and into static method definitions.  This also reduces the ACPI
    DSDT code size.

    Signed-off-by: Kevin O'Connor <kevin@koconnor.net>


Now, I don't really understand what's going on there... ;)

And since this is DSDT, using seabios 1.6 with recent qemu does
not solve the problem, since DSDT is now external/separate.

Thanks,

/mjt

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Kevin O'Connor]

On Fri, Jun 14, 2013 at 12:32:22AM +0400, Michael Tokarev wrote:
> 13.06.2013 11:51, Michael Tokarev wrote:
> > Hello.
> > 
> > In order to verify some build issues on solaris, I tried to install
> > sol10 x86 in a kvm vm.  But unfortunately it does not work: after the
> > grub prompt and choosing "Solaris 10 x86" boot entry, the kernel
> > gets loaded (there's a row of dots displayed during that), next,
> > the following message gets displayed:
> > 
> >  SunOS Release 5.10 Version Generic_147148-26 64-bit
> >  Copyright (c) 1983, 2013 Oracle and/or its affiliates.  All rights reserved.
> > 
> > and the guest stays there for a long time, spinning up 100% of its CPU,
> > and nothing more happens.
> > 
> > The same happens when run with or without kvm (ie, tcg and kvm behaves
> > the same way).
> > 
> > When run in kvm, kvm_stats shows just a few exits (about 600/sec) and
> > nothing more than that.
> > 
> > I think that supporting solaris as _guest_ OS is an important goal
> > for qemu/kvm (as opposed to _host_).
> 
> I tried to bisect this.  It turns out that solaris x86 does not boot in
> qemu/kvm for quite long time already, namely, starting from this commit:
> 
> 
>  commit 6b034aa138716a515c88f7894940d5d0aff2f3ed
>  Author: Gerd Hoffmann <kraxel@redhat.com>
>  Date:   Tue Apr 17 10:51:41 2012 +0200
> 
>     seabios: update to 1.7.0
> 
>     Update roms/seabios and pc-bios/bios.bin to the 1.7.0 release.
>     Most noticable new feature is virtio-scsi support.
> 
>     Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> 
> 
> So I went on and tried to bisect seabios (previous version in qemu
> was 1.6.3.2, and it worked).
> 
> So seabios bisection with qemu-1.1 points to this commit:
> 
>  commit 9d3d7cb4b163d3fbcba64a01c4fa42eb6bc53128
>  Author: Kevin O'Connor <kevin@koconnor.net>
>  Date:   Wed Sep 21 21:19:51 2011 -0400
> 
>     Move code from PCI hotplug DSDT macros to methods.
> 
>     Simplify the hotplug code by moving the bulk of the logic out of the
>     macros and into static method definitions.  This also reduces the ACPI
>     DSDT code size.
> 
>     Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
> 

Can you try with the latest seabios release (v1.7.2.2)?  There have
been other changes to the dsdt that may have corrected your issue.

-Kevin

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Michael Tokarev]

14.06.2013 03:47, Kevin O'Connor wrote:

>>     Move code from PCI hotplug DSDT macros to methods.
>>
>>     Simplify the hotplug code by moving the bulk of the logic out of the
>>     macros and into static method definitions.  This also reduces the ACPI
>>     DSDT code size.
>>
>>     Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
> 
> Can you try with the latest seabios release (v1.7.2.2)?  There have
> been other changes to the dsdt that may have corrected your issue.

It's not exactly trivial to switch bioses after splitting out the
tables.  But I tried the same thing in current qemu git master,
which includes 1.7.2.2, and where solaris guest behaves exactly the
same -- it stalls with 100% CPU usage somewhere at the beginning of
kernel booting.

Hmm.  Speaking of the splitting.  Does the current bios include the
the tables which were split into separate files?  I mean, what about
compatibility between new bios and old qemu and vise versa, where
"old" = "before split" and "new" = "after split" ?

Thanks!

/mjt

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Gerd Hoffmann]

  Hi,

> Hmm.  Speaking of the splitting.  Does the current bios include the
> the tables which were split into separate files?

Yes, they are in out/ too after building seabios.  Use "qemu -L
/path/to/seabios/out" to make sure qemu picks up matching bios.bin and dsdt.

cheers,
  Gerd

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Michael Tokarev]

14.06.2013 16:36, Gerd Hoffmann wrote:
>   Hi,
> 
>> Hmm.  Speaking of the splitting.  Does the current bios include the
>> the tables which were split into separate files?
> 
> Yes, they are in out/ too after building seabios.  Use "qemu -L
> /path/to/seabios/out" to make sure qemu picks up matching bios.bin and dsdt.

Yes I know they're built in seabios/out, and I know one may
use -L option here.  So that, for example, -bios option of
qemu makes less sense now than it was before.

But my question was about something different.

Can I use current seabios with old qemu which does not provide
the separate ACPI tables?  For example, does current bios contain
these tables too, so they're both separate and embedded?

And the reverse, can I use old bios with new qemu which do provide
the separate tables?

Which tables will be used in each case?

Thanks,

/mjt

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Kevin O'Connor]

On Fri, Jun 14, 2013 at 06:14:00PM +0400, Michael Tokarev wrote:
> 14.06.2013 16:36, Gerd Hoffmann wrote:
> >   Hi,
> > 
> >> Hmm.  Speaking of the splitting.  Does the current bios include the
> >> the tables which were split into separate files?
> > 
> > Yes, they are in out/ too after building seabios.  Use "qemu -L
> > /path/to/seabios/out" to make sure qemu picks up matching bios.bin and dsdt.
> 
> Yes I know they're built in seabios/out, and I know one may
> use -L option here.  So that, for example, -bios option of
> qemu makes less sense now than it was before.
> 
> But my question was about something different.
> 
> Can I use current seabios with old qemu which does not provide
> the separate ACPI tables?  For example, does current bios contain
> these tables too, so they're both separate and embedded?

Yes.

> And the reverse, can I use old bios with new qemu which do provide
> the separate tables?

Yes.  The only limitation is that one can not use the q35 machine type
without also having a new version of seabios.

> Which tables will be used in each case?

SeaBIOS has the "piix4" acpi tables compiled into it, and it will use
those if it doesn't receive a dsdt from qemu.

-Kevin

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Gerd Hoffmann]

  Hi,

> Can I use current seabios with old qemu which does not provide
> the separate ACPI tables?  For example, does current bios contain
> these tables too, so they're both separate and embedded?

Yes.  The i440fx acpi tables are compiled in and will be used as
fallback in case qemu doesn't provide acpi tables.

> And the reverse, can I use old bios with new qemu which do provide
> the separate tables?

Yes.  Old seabios will simply ignore the qemu-provides tables and use
the compiled-in ones.  That is limited to the 'pc' machine type though,
q35 will not work this way.  q35 needs other tables and this was the
reason to let qemu provide the tables in the first place.

cheers,
  Gerd

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Michael Tokarev]

14.06.2013 03:47, Kevin O'Connor wrote:
> On Fri, Jun 14, 2013 at 12:32:22AM +0400, Michael Tokarev wrote:
>> 13.06.2013 11:51, Michael Tokarev wrote:
>>> Hello.
>>>
>>> In order to verify some build issues on solaris, I tried to install
>>> sol10 x86 in a kvm vm.  But unfortunately it does not work: after the
>>> grub prompt and choosing "Solaris 10 x86" boot entry, the kernel
>>> gets loaded (there's a row of dots displayed during that), next,
>>> the following message gets displayed:
>>>
>>>  SunOS Release 5.10 Version Generic_147148-26 64-bit
>>>  Copyright (c) 1983, 2013 Oracle and/or its affiliates.  All rights reserved.
>>>
>>> and the guest stays there for a long time, spinning up 100% of its CPU,
>>> and nothing more happens.
>>>
>>> The same happens when run with or without kvm (ie, tcg and kvm behaves
>>> the same way).
>>>
>>> When run in kvm, kvm_stats shows just a few exits (about 600/sec) and
>>> nothing more than that.
>>>
>>> I think that supporting solaris as _guest_ OS is an important goal
>>> for qemu/kvm (as opposed to _host_).
>>
>> I tried to bisect this.  It turns out that solaris x86 does not boot in
>> qemu/kvm for quite long time already, namely, starting from this commit:
>>
>>
>>  commit 6b034aa138716a515c88f7894940d5d0aff2f3ed
>>  Author: Gerd Hoffmann <kraxel@redhat.com>
>>  Date:   Tue Apr 17 10:51:41 2012 +0200
>>
>>     seabios: update to 1.7.0
>>
>>     Update roms/seabios and pc-bios/bios.bin to the 1.7.0 release.
>>     Most noticable new feature is virtio-scsi support.
>>
>>     Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
>>
>>
>> So I went on and tried to bisect seabios (previous version in qemu
>> was 1.6.3.2, and it worked).
>>
>> So seabios bisection with qemu-1.1 points to this commit:
>>
>>  commit 9d3d7cb4b163d3fbcba64a01c4fa42eb6bc53128
>>  Author: Kevin O'Connor <kevin@koconnor.net>
>>  Date:   Wed Sep 21 21:19:51 2011 -0400
>>
>>     Move code from PCI hotplug DSDT macros to methods.
>>
>>     Simplify the hotplug code by moving the bulk of the logic out of the
>>     macros and into static method definitions.  This also reduces the ACPI
>>     DSDT code size.
>>
>>     Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
>>
> 
> Can you try with the latest seabios release (v1.7.2.2)?  There have
> been other changes to the dsdt that may have corrected your issue.

So, with the knowlege that current 1.7.2.2 does not change this, can we
guess what's going on there?  I don't really know solaris kernel internals
(is the source available to start with?), but maybe this particular commit
may give some hint?

Thanks!

/mjt

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Kevin O'Connor]

On Sun, Jun 16, 2013 at 02:07:39PM +0400, Michael Tokarev wrote:
> 14.06.2013 03:47, Kevin O'Connor wrote:
> > On Fri, Jun 14, 2013 at 12:32:22AM +0400, Michael Tokarev wrote:
> >> 13.06.2013 11:51, Michael Tokarev wrote:
> >>> In order to verify some build issues on solaris, I tried to install
> >>> sol10 x86 in a kvm vm.  But unfortunately it does not work: after the
> >>> grub prompt and choosing "Solaris 10 x86" boot entry, the kernel
> >>> gets loaded (there's a row of dots displayed during that), next,
> >>> the following message gets displayed:
> >>>
> >>>  SunOS Release 5.10 Version Generic_147148-26 64-bit
> >>>  Copyright (c) 1983, 2013 Oracle and/or its affiliates.  All rights reserved.
> >>>
> >>> and the guest stays there for a long time, spinning up 100% of its CPU,
> >>> and nothing more happens.
> > 
> > Can you try with the latest seabios release (v1.7.2.2)?  There have
> > been other changes to the dsdt that may have corrected your issue.
> 
> So, with the knowlege that current 1.7.2.2 does not change this, can we
> guess what's going on there?  I don't really know solaris kernel internals
> (is the source available to start with?), but maybe this particular commit
> may give some hint?

Unfortunately not.  There was nothing special about the DSDT changes
in that commit.  Likely the AML interpretter in Solaris is quirky and
one of the clauses in that commit (or some arrangement of clauses) is
confusing Solaris.  This is one of the unfortunate parts of ACPI -
despite the lengthy specification each implementation of ACPI is
different and it's painful to come up with AML code that runs on
multiple platforms.

You could try disabling acpi entirely.

You could also try iterating through the parts of that commit and
commenting out various DSDT entries until you find which one is
aggravating Solaris.

-Kevin

Re: [Qemu-devel] [SeaBIOS] solaris x86 in qemu? [bisected]

[Laszlo Ersek]

On 06/16/13 17:30, Kevin O'Connor wrote:

> Unfortunately not.  There was nothing special about the DSDT changes
> in that commit.  Likely the AML interpretter in Solaris is quirky and
> one of the clauses in that commit (or some arrangement of clauses) is
> confusing Solaris.  This is one of the unfortunate parts of ACPI -
> despite the lengthy specification each implementation of ACPI is
> different and it's painful to come up with AML code that runs on
> multiple platforms.
> 
> You could try disabling acpi entirely.
> 
> You could also try iterating through the parts of that commit and
> commenting out various DSDT entries until you find which one is
> aggravating Solaris.

Also, "Solaris ACPICA" is googleable; two examples:

https://blogs.oracle.com/danasblog/entry/configuring_solaris_acpi_at_boot
https://blogs.oracle.com/danasblog/entry/solaris_acpi_ca_debug_output

Laszlo

Re: [Qemu-devel] solaris x86 in qemu?

[Brad Smith]

On 13/06/13 3:51 AM, Michael Tokarev wrote:
> I think that supporting solaris as _guest_ OS is an important goal
> for qemu/kvm (as opposed to _host_).

IMO both are important goals. I don't think it would be that big of a 
deal to find one or two people to provide Solaris x64 10 and/or 11.1 
buildslaves.

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] solaris x86 in qemu?

[Michael Tokarev]

14.06.2013 01:44, Brad Smith wrote:
> On 13/06/13 3:51 AM, Michael Tokarev wrote:
>> I think that supporting solaris as _guest_ OS is an important goal
>> for qemu/kvm (as opposed to _host_).
> 
> IMO both are important goals. I don't think it would be that big of a deal to find one or two people to provide Solaris x64 10 and/or 11.1 buildslaves.

Well, after discovering how long solaris guest is broken (since
1.0.50), I tend to disagree that it is important... ;)

And this very fact that it's broken makes it more difficult to provide
a build bot.  I wanted to provide one, but I don't have spare hardware
for that, so I intended to set up a virtual machine, but it does not
work.. ;)

Thanks,

/mjt