kvm-devel

[kvm-devel] 32-bit binaries failing in 64 bit guests after using vmport

[Soren H. <so...@ub...>]

Esteemed kvm developers!

I've been trying to debug this bug

    https://bugs.launchpad.net/ubuntu/+source/kvm/+bug/219165

It originally revealed itself by failing to run grub (which is a 32 bit
binary) when installing Ubuntu from our live cd. It turned out to be a
more general problem of 32 bit binaries failing to run. The server
install worked like a charm. I eventually discovered that loading the
vmmouse driver triggered it and narrowed it down to the call to
kvm_load_registers in vmport_ioport_read. 

We're releasing on Thursday, and I needed a quick fix, so I reverted the
calls to kvm_{save,load}_registers in vmport_ioport_read to the old code
that simply saved the eax, ebx, ecx, edx, esi, and edi registers, but
I'm supposing kvm_{load,save}_registers really should work here.

I dug a bit further into the code and tried disabling various pieces of
the kvm_load_registers until it finally worked again. The problem seems
to only arise when the lstar msr is loaded. I've looked at the code, but
seeing as three days ago I didn't know there was such a thing as an
lstar msr, I'm finding myself getting stuck. :)

Any pointers in the right direction would be lovely.

-- 
Soren Hansen               | 
Virtualisation specialist  | Ubuntu Server Team
Canonical Ltd.             | http://www.ubuntu.com/

Re: [kvm-devel] 32-bit binaries failing in 64 bit guests after using vmport

[Anthony L. <an...@co...>]

Soren Hansen wrote:
> Esteemed kvm developers!
>
> I've been trying to debug this bug
>
>     https://bugs.launchpad.net/ubuntu/+source/kvm/+bug/219165
>
> It originally revealed itself by failing to run grub (which is a 32 bit
> binary) when installing Ubuntu from our live cd. It turned out to be a
> more general problem of 32 bit binaries failing to run. The server
> install worked like a charm. I eventually discovered that loading the
> vmmouse driver triggered it and narrowed it down to the call to
> kvm_load_registers in vmport_ioport_read. 
>
> We're releasing on Thursday, and I needed a quick fix, so I reverted the
> calls to kvm_{save,load}_registers in vmport_ioport_read to the old code
> that simply saved the eax, ebx, ecx, edx, esi, and edi registers, but
> I'm supposing kvm_{load,save}_registers really should work here.
>   

Ah, you may have missed the fix that updated the KVM load/save functions 
to deal with the in-kernel APIC.

It turns out, vmmouse was horribly broken with SMP guests too.  See

commit 9949bd84ac4dfdfc60b2974557819637b8719911
Author: Anthony Liguori <ali...@us...>
Date:   Thu Apr 3 18:37:16 2008 -0500

commit 5208ce19dca268f84a2b9441c2fbb6129161e44c
Author: Marcelo Tosatti <mto...@re...>
Date:   Thu Apr 3 20:24:37 2008 -0300

commit 85a67aa2a1b942ddccfcbd625d280869367edc95
Author: Marcelo Tosatti <mto...@re...>
Date:   Fri Apr 11 13:24:41 2008 -0300

Regards,

Anthony LIguori

> I dug a bit further into the code and tried disabling various pieces of
> the kvm_load_registers until it finally worked again. The problem seems
> to only arise when the lstar msr is loaded. I've looked at the code, but
> seeing as three days ago I didn't know there was such a thing as an
> lstar msr, I'm finding myself getting stuck. :)
>
> Any pointers in the right direction would be lovely.
>
>   
> ------------------------------------------------------------------------
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
> Don't miss this year's exciting event. There's still time to save $100. 
> Use priority code J8TL2D2. 
> http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
> ------------------------------------------------------------------------
>
> _______________________________________________
> kvm-devel mailing list
> kvm...@li...
> https://lists.sourceforge.net/lists/listinfo/kvm-devel
>   

Re: [kvm-devel] 32-bit binaries failing in 64 bit guests after using vmport

[Soren H. <so...@ub...>]

On Mon, Apr 21, 2008 at 08:51:17AM -0500, Anthony Liguori wrote:
>> We're releasing on Thursday, and I needed a quick fix, so I reverted
>> the calls to kvm_{save,load}_registers in vmport_ioport_read to the
>> old code that simply saved the eax, ebx, ecx, edx, esi, and edi
>> registers, but I'm supposing kvm_{load,save}_registers really should
>> work here.

/me sighs very deeply

Ok, first chance I get, I'm signing up for "Patch management 101". :( I
got some tests mixed around, so it failing is actually dependent on
whether EIP (not LSTAR as I originally thought) is restored or not.

I have a patch that fixes it, but I need to work a few things out first
before I submit it.

> Ah, you may have missed the fix that updated the KVM load/save
> functions  to deal with the in-kernel APIC.

Indeed.

> It turns out, vmmouse was horribly broken with SMP guests too.  See
>
> commit 9949bd84ac4dfdfc60b2974557819637b8719911
> Author: Anthony Liguori <ali...@us...>
> Date:   Thu Apr 3 18:37:16 2008 -0500
>
> commit 5208ce19dca268f84a2b9441c2fbb6129161e44c
> Author: Marcelo Tosatti <mto...@re...>
> Date:   Thu Apr 3 20:24:37 2008 -0300

I did my tests using kvm-65 userland and kernel, so these two should
already be included, afaics.  

> commit 85a67aa2a1b942ddccfcbd625d280869367edc95
> Author: Marcelo Tosatti <mto...@re...>
> Date:   Fri Apr 11 13:24:41 2008 -0300

This did not change anything for me.

-- 
Soren Hansen               | 
Virtualisation specialist  | Ubuntu Server Team
Canonical Ltd.             | http://www.ubuntu.com/