amd64: does it ever work? ;)

amd64: does it ever work? ;)

[Michael Tokarev]

Hello!  A first-time poster (and user) is here... but not a newbie ;)

Some time ago I tried KVM for the first time, on an AMD X2-64 system
(BE-2400 processor, also tried with other similar processors, like
"plain" X2-64 4888+ etc), first with 2.6.23 vanilla kernel compiled
for 64 bits, SMP.  After it didn't work, I also tried KVM modules
from SVN (compiled against the same kernel sources), and when 2.6.24
was out, I tried that too.  Neither of which works.

Userspace tools is from Debian unstable -- 58+dfsg-1.

The effect is as follows.

With 2.6.23 and vanilla kvm modules, while from X, it opens the
"virtual console" (an X window), but that console is entirely
empty (black), and nothing more happens.  I can Ctrl+C the original
process, but nothing more.

With 2.6.23+modules-59, and with vanilla 2.6.24, it again opens
the window, and the HOST machine immediately reboots (goes into
BIOS Post screen).

When not loading kvm modules, it works in all 3 cases.

Where to look at to debug it further?

Thanks!

/mjt

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Amit Shah]

Hello Michael,

On Friday 01 February 2008 13:30:03 Michael Tokarev wrote:
> Hello!  A first-time poster (and user) is here... but not a newbie ;)
>
> Some time ago I tried KVM for the first time, on an AMD X2-64 system
> (BE-2400 processor, also tried with other similar processors, like
> "plain" X2-64 4888+ etc), first with 2.6.23 vanilla kernel compiled
> for 64 bits, SMP.  After it didn't work, I also tried KVM modules
> from SVN (compiled against the same kernel sources), and when 2.6.24
> was out, I tried that too.  Neither of which works.

KVM doesn't have an SVN tree; do you mean git?

> Userspace tools is from Debian unstable -- 58+dfsg-1.
>
> The effect is as follows.
>
> With 2.6.23 and vanilla kvm modules, while from X, it opens the
> "virtual console" (an X window), but that console is entirely
> empty (black), and nothing more happens.  I can Ctrl+C the original
> process, but nothing more.
>
> With 2.6.23+modules-59, and with vanilla 2.6.24, it again opens
> the window, and the HOST machine immediately reboots (goes into
> BIOS Post screen).
>
> When not loading kvm modules, it works in all 3 cases.
>
> Where to look at to debug it further?

Can you check if you have any BIOS updates? This seems like BIOS support for 
SVM might be half-baked.

>
> Thanks!
>
> /mjt

Amit

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Michael Tokarev]

Amit Shah wrote:

> On Friday 01 February 2008 13:30:03 Michael Tokarev wrote:

>> Some time ago I tried KVM for the first time, on an AMD X2-64 system
>> (BE-2400 processor, also tried with other similar processors, like
>> "plain" X2-64 4888+ etc), first with 2.6.23 vanilla kernel compiled
>> for 64 bits, SMP.  After it didn't work, I also tried KVM modules
>> from SVN (compiled against the same kernel sources), and when 2.6.24
>> was out, I tried that too.  Neither of which works.
> 
> KVM doesn't have an SVN tree; do you mean git?

Err...  It was totally screwed up.  I tried kvm-59 release from
sourceforge, not git (and definitely not svn :)

Right now I'm compiling kvm-60.

>> Userspace tools is from Debian unstable -- 58+dfsg-1.
>>
>> The effect is as follows.
>>
>> With 2.6.23 and vanilla kvm modules, while from X, it opens the
>> "virtual console" (an X window), but that console is entirely
>> empty (black), and nothing more happens.  I can Ctrl+C the original
>> process, but nothing more.
>>
>> With 2.6.23+modules-59, and with vanilla 2.6.24, it again opens
>> the window, and the HOST machine immediately reboots (goes into
>> BIOS Post screen).
>>
>> When not loading kvm modules, it works in all 3 cases.
>>
>> Where to look at to debug it further?
> 
> Can you check if you have any BIOS updates? This seems like BIOS support for 
> SVM might be half-baked.

Well..  I tried it on 2 different motherboards with 2 different
CPUs (in all 4 permutations):

Biostar TA690G-AM2 - AMD690G chipset, latest BIOS (dated 09-2007)
ASUS M2NPV-VM - Geforce6150/Nforce430(?) (MCP51), latest BIOS (dated 11-2007)

The effect on both is exactly the same.

So I don't think it's a BIOS problem after all.  Maybe some kernel
option(s) (like tickless (dynticks) for example - but 2.6.23 x86-64
didn't have that option).  But with so many options I'm afraid I can't
tell which to try to turn off... ;)

Thanks!

/mjt

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Michael Tokarev]

Michael Tokarev wrote:
> Amit Shah wrote:
[]
>>> With 2.6.23 and vanilla kvm modules, while from X, it opens the
>>> "virtual console" (an X window), but that console is entirely
>>> empty (black), and nothing more happens.  I can Ctrl+C the original
>>> process, but nothing more.
>>>
>>> With 2.6.23+modules-59, and with vanilla 2.6.24, it again opens
>>> the window, and the HOST machine immediately reboots (goes into
>>> BIOS Post screen).
>>>
>>> When not loading kvm modules, it works in all 3 cases.
>>>
>>> Where to look at to debug it further?
>> Can you check if you have any BIOS updates? This seems like BIOS support for 
>> SVM might be half-baked.
> 
> Well..  I tried it on 2 different motherboards with 2 different
> CPUs (in all 4 permutations):
> 
> Biostar TA690G-AM2 - AMD690G chipset, latest BIOS (dated 09-2007)
> ASUS M2NPV-VM - Geforce6150/Nforce430(?) (MCP51), latest BIOS (dated 11-2007)

Ok.  I compiled kvm-60, loaded the modules and tried something simple
as usual.

Unlike previously, the host didn't reboot, and kvm window stayed
blank.  But now there's at least something in dmesg:

 loaded kvm module (kvm-60)
 emulation failed (pagetable) rip fff0 00 00 00 00

Is it something which can help?

Thanks!

/mjt


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Michael Tokarev]

Michael Tokarev wrote:
> Amit Shah wrote:
> 
>> On Friday 01 February 2008 13:30:03 Michael Tokarev wrote:
[]
>>> With 2.6.23 and vanilla kvm modules, while from X, it opens the
>>> "virtual console" (an X window), but that console is entirely
>>> empty (black), and nothing more happens.  I can Ctrl+C the original
>>> process, but nothing more.
>>>
>>> With 2.6.23+modules-59, and with vanilla 2.6.24, it again opens
>>> the window, and the HOST machine immediately reboots (goes into
>>> BIOS Post screen).
>>>
>>> When not loading kvm modules, it works in all 3 cases.
>>>
>>> Where to look at to debug it further?
>> Can you check if you have any BIOS updates? This seems like BIOS support for 
>> SVM might be half-baked.
> 
> Well..  I tried it on 2 different motherboards with 2 different
> CPUs (in all 4 permutations):
> 
> Biostar TA690G-AM2 - AMD690G chipset, latest BIOS (dated 09-2007)
> ASUS M2NPV-VM - Geforce6150/Nforce430(?) (MCP51), latest BIOS (dated 11-2007)

Ok, I finally got it working.

The problem was 32/64 bits issue.  I'm so used to 64bits kernel and 32bits
userland - not a single thought occured to me that this might be problematic
of some sort.  I was always trying 32bits kvm usermode tools with 64bits
kernels.  Now I booted a 32bits kernel, and it finally worked.  Oh well.

/mjt

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Avi Kivity]

Michael Tokarev wrote:
> Ok, I finally got it working.
>
> The problem was 32/64 bits issue.  I'm so used to 64bits kernel and 32bits
> userland - not a single thought occured to me that this might be problematic
> of some sort.  I was always trying 32bits kvm usermode tools with 64bits
> kernels.  Now I booted a 32bits kernel, and it finally worked.  Oh well.
>
>   

Actually this is intended to work.  I'll look into it.

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Amit Shah]

On Sunday 10 February 2008 17:01:46 Avi Kivity wrote:
> Michael Tokarev wrote:
> > Ok, I finally got it working.
> >
> > The problem was 32/64 bits issue.  I'm so used to 64bits kernel and
> > 32bits userland - not a single thought occured to me that this might be
> > problematic of some sort.  I was always trying 32bits kvm usermode tools
> > with 64bits kernels.  Now I booted a 32bits kernel, and it finally
> > worked.  Oh well.
>
> Actually this is intended to work.  I'll look into it.

I think he means a 64-bit kernel in the guest with 32-bit userspace in 
guest -- that won't work.

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Avi Kivity]

Amit Shah wrote:
> On Sunday 10 February 2008 17:01:46 Avi Kivity wrote:
>   
>> Michael Tokarev wrote:
>>     
>>> Ok, I finally got it working.
>>>
>>> The problem was 32/64 bits issue.  I'm so used to 64bits kernel and
>>> 32bits userland - not a single thought occured to me that this might be
>>> problematic of some sort.  I was always trying 32bits kvm usermode tools
>>> with 64bits kernels.  Now I booted a 32bits kernel, and it finally
>>> worked.  Oh well.
>>>       
>> Actually this is intended to work.  I'll look into it.
>>     
>
> I think he means a 64-bit kernel in the guest with 32-bit userspace in 
> guest -- that won't work.
>   

Why not?

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Michael Tokarev]

Avi Kivity wrote:
> Amit Shah wrote:
>> On Sunday 10 February 2008 17:01:46 Avi Kivity wrote:
>>   
>>> Michael Tokarev wrote:
>>>     
>>>> Ok, I finally got it working.
>>>>
>>>> The problem was 32/64 bits issue.  I'm so used to 64bits kernel and
>>>> 32bits userland - not a single thought occured to me that this might be
>>>> problematic of some sort.  I was always trying 32bits kvm usermode tools
>>>> with 64bits kernels.  Now I booted a 32bits kernel, and it finally
>>>> worked.  Oh well.
>>>>       
>>> Actually this is intended to work.  I'll look into it.
>>>     
>> I think he means a 64-bit kernel in the guest with 32-bit userspace in 
>> guest -- that won't work.

I first tried 64bit HOST kernel and 32bit host userspace.  It didn't matter
what to use as guest - it doesn't even try to boot.

Speaking of 64bit host, 64bit guest kernel and 32bit guest userland - it
just works.

/mjt

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Avi Kivity]

Michael Tokarev wrote:
> I first tried 64bit HOST kernel and 32bit host userspace.  It didn't matter
> what to use as guest - it doesn't even try to boot.
>
>   

Okay.  Fixed that bit for kvm-61.

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/

Re: amd64: does it ever work? ;)

[Amit Shah]

On Monday 11 February 2008 12:58:07 Avi Kivity wrote:
> Amit Shah wrote:
> > On Sunday 10 February 2008 17:01:46 Avi Kivity wrote:
> >> Michael Tokarev wrote:
> >>> Ok, I finally got it working.
> >>>
> >>> The problem was 32/64 bits issue.  I'm so used to 64bits kernel and
> >>> 32bits userland - not a single thought occured to me that this might be
> >>> problematic of some sort.  I was always trying 32bits kvm usermode
> >>> tools with 64bits kernels.  Now I booted a 32bits kernel, and it
> >>> finally worked.  Oh well.
> >>
> >> Actually this is intended to work.  I'll look into it.
> >
> > I think he means a 64-bit kernel in the guest with 32-bit userspace in
> > guest -- that won't work.
>
> Why not?

Hmm, I thought having 64-bit userland with 32-bit compat libraries was the 
only way of making that work.

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/