2.6.23-rc1, KVM-AMD problem

2.6.23-rc1, KVM-AMD problem

[Alistair John Strachan]

Hi,

I'm getting periodic oopses running KVM-33 on 2.6.23-rc1. Here is a digital 
photo of the oops. Alarmingly, a lot of the time it triple faults the machine 
and I don't get a chance to grab it. This time I was lucky, though.

http://devzero.co.uk/~alistair/kvm-2.6.23-rc1.jpg

Unfortunately, some of the oops text scrolled out of the screen. I will 
endeavour to reproduce the bug over serial console, but I can make no 
guarantees.

The CPU is an AMD X2 BE-2350, chipset is AMD 690G.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.

Re: 2.6.23-rc1, KVM-AMD problem

[Avi Kivity]

[-- Attachment #1: Type: text/plain, Size: 822 bytes --]

Alistair John Strachan wrote:
> Hi,
>
> I'm getting periodic oopses running KVM-33 on 2.6.23-rc1. Here is a digital 
> photo of the oops. Alarmingly, a lot of the time it triple faults the machine 
> and I don't get a chance to grab it. This time I was lucky, though.
>
> http://devzero.co.uk/~alistair/kvm-2.6.23-rc1.jpg
>
> Unfortunately, some of the oops text scrolled out of the screen. I will 
> endeavour to reproduce the bug over serial console, but I can make no 
> guarantees.
>
> The CPU is an AMD X2 BE-2350, chipset is AMD 690G.
>
>   

If you are using the modules from 2.6.23-rc1, try upgrading to latest 
-git, which contains a patch that might fix this problem.  If you are 
using the modules from kvm-33, try applying the attached patch.


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


[-- Attachment #2: fix-0f-01.patch --]
[-- Type: text/x-patch, Size: 903 bytes --]

commit bfa6c62f98bd0602025d7b48e267d817082f5d07
Author: Aurelien Jarno <aurelien@aurel32.net>
Date:   Wed Jul 25 10:19:54 2007 +0200

    KVM: disable writeback for 0x0f 0x01 instructions.
    
    0x0f 0x01 instructions (ie lgdt, lidt, smsw, lmsw and invlpg) does
    not use writeback. This patch set no_wb=1 when emulating those
    instructions.
    
    This fixes a regression booting the FreeBSD kernel on AMD.
    
    Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
    Signed-off-by: Avi Kivity <avi@qumranet.com>

diff --git a/drivers/kvm/x86_emulate.c b/drivers/kvm/x86_emulate.c
index 21ce977..cbbb9c5 100644
--- a/drivers/kvm/x86_emulate.c
+++ b/drivers/kvm/x86_emulate.c
@@ -1183,6 +1183,8 @@ pop_instruction:
 twobyte_insn:
 	switch (b) {
 	case 0x01: /* lgdt, lidt, lmsw */
+		/* Disable writeback. */
+		no_wb = 1;
 		switch (modrm_reg) {
 			u16 size;
 			unsigned long address;

Re: 2.6.23-rc1, KVM-AMD problem

[Alistair John Strachan]

On Sunday 29 July 2007 09:16:43 Avi Kivity wrote:
> Alistair John Strachan wrote:
> > Hi,
> >
> > I'm getting periodic oopses running KVM-33 on 2.6.23-rc1. Here is a
> > digital photo of the oops. Alarmingly, a lot of the time it triple faults
> > the machine and I don't get a chance to grab it. This time I was lucky,
> > though.
> >
> > http://devzero.co.uk/~alistair/kvm-2.6.23-rc1.jpg
> >
> > Unfortunately, some of the oops text scrolled out of the screen. I will
> > endeavour to reproduce the bug over serial console, but I can make no
> > guarantees.
> >
> > The CPU is an AMD X2 BE-2350, chipset is AMD 690G.
>
> If you are using the modules from 2.6.23-rc1, try upgrading to latest
> -git, which contains a patch that might fix this problem.  If you are
> using the modules from kvm-33, try applying the attached patch.

Doesn't help, I still get the same crashes. I tried 2.6.22 again and it's rock 
solid by comparison.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.

Re: 2.6.23-rc1, KVM-AMD problem

[Avi Kivity]

Alistair John Strachan wrote:
> On Sunday 29 July 2007 09:16:43 Avi Kivity wrote:
>   
>> Alistair John Strachan wrote:
>>     
>>> Hi,
>>>
>>> I'm getting periodic oopses running KVM-33 on 2.6.23-rc1. Here is a
>>> digital photo of the oops. Alarmingly, a lot of the time it triple faults
>>> the machine and I don't get a chance to grab it. This time I was lucky,
>>> though.
>>>
>>> http://devzero.co.uk/~alistair/kvm-2.6.23-rc1.jpg
>>>
>>> Unfortunately, some of the oops text scrolled out of the screen. I will
>>> endeavour to reproduce the bug over serial console, but I can make no
>>> guarantees.
>>>
>>> The CPU is an AMD X2 BE-2350, chipset is AMD 690G.
>>>       
>> If you are using the modules from 2.6.23-rc1, try upgrading to latest
>> -git, which contains a patch that might fix this problem.  If you are
>> using the modules from kvm-33, try applying the attached patch.
>>     
>
> Doesn't help, I still get the same crashes. I tried 2.6.22 again and it's rock 
> solid by comparison.
>
>   

Do you mean, kvm-33 on top of 2.6.22, or the kvm modules from 2.6.22?  
Please describe your configuration *exactly*.

And what do you mean, "rock solid by comparison"?  Either it's rock 
solid or it isn't.

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

Re: 2.6.23-rc1, KVM-AMD problem

[Alistair John Strachan]

On Sunday 29 July 2007 12:34:28 you wrote:
[snip]
> > Doesn't help, I still get the same crashes. I tried 2.6.22 again and it's
> > rock solid by comparison.
>
> Do you mean, kvm-33 on top of 2.6.22, or the kvm modules from 2.6.22?
> Please describe your configuration *exactly*.

I'm using the kvm-33 *userspace* package (based on Debian's kvm-28 packaging) 
and 2.6.23-rc1's KVM modules. I patched 2.6.23-rc1 with the patch you 
provided in your last email. So I'm not using -git HEAD.

Maybe there's been additional necessary fixes to -git requiring me to update 
to HEAD? That wasn't clear from your last email.

> And what do you mean, "rock solid by comparison"?  Either it's rock
> solid or it isn't.

It's rock solid. It does not crash.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.

Re: 2.6.23-rc1, KVM-AMD problem

[Avi Kivity]

Alistair John Strachan wrote:
> On Sunday 29 July 2007 12:34:28 you wrote:
> [snip]
>   
>>> Doesn't help, I still get the same crashes. I tried 2.6.22 again and it's
>>> rock solid by comparison.
>>>       
>> Do you mean, kvm-33 on top of 2.6.22, or the kvm modules from 2.6.22?
>> Please describe your configuration *exactly*.
>>     
>
> I'm using the kvm-33 *userspace* package (based on Debian's kvm-28 packaging) 
> and 2.6.23-rc1's KVM modules. I patched 2.6.23-rc1 with the patch you 
> provided in your last email. So I'm not using -git HEAD.
>
> Maybe there's been additional necessary fixes to -git requiring me to update 
> to HEAD? That wasn't clear from your last email.
>   

No,  that patch is the only potential fix post -rc1.  There are a few 
other fixes there, but they are intended to avoid guest crashes, not 
host crashes.

What guest are you running?  Maybe I can reproduce it here.

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

Re: 2.6.23-rc1, KVM-AMD problem

[Alistair John Strachan]

On Sunday 29 July 2007 14:47:57 you wrote:
> Alistair John Strachan wrote:
> > On Sunday 29 July 2007 12:34:28 you wrote:
> > [snip]
> >
> >>> Doesn't help, I still get the same crashes. I tried 2.6.22 again and
> >>> it's rock solid by comparison.
> >>
> >> Do you mean, kvm-33 on top of 2.6.22, or the kvm modules from 2.6.22?
> >> Please describe your configuration *exactly*.
> >
> > I'm using the kvm-33 *userspace* package (based on Debian's kvm-28
> > packaging) and 2.6.23-rc1's KVM modules. I patched 2.6.23-rc1 with the
> > patch you provided in your last email. So I'm not using -git HEAD.
> >
> > Maybe there's been additional necessary fixes to -git requiring me to
> > update to HEAD? That wasn't clear from your last email.
>
> No,  that patch is the only potential fix post -rc1.  There are a few
> other fixes there, but they are intended to avoid guest crashes, not
> host crashes.
>
> What guest are you running?  Maybe I can reproduce it here.

Right now, Windows XP. I'm pretty sure Linux (well, Debian Etch) works fine. I 
could only get Windows to install with -no-acpi, but I run it with the 
following (if this is at all useful):

kvm -no-acpi -m 256 -hda $IMAGE -net nic -net tap,script=/etc/kvm/kvm-ifup

Basically, the installer seems to work fine, but Windows seemed to have 
problems after installing post-SP2 updates. Maybe that's why not everybody is 
seeing it yet.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.

Re: 2.6.23-rc1, KVM-AMD problem

[Avi Kivity]

[-- Attachment #1: Type: text/plain, Size: 1695 bytes --]

Alistair John Strachan wrote:
> On Sunday 29 July 2007 14:47:57 you wrote:
>   
>> Alistair John Strachan wrote:
>>     
>>> On Sunday 29 July 2007 12:34:28 you wrote:
>>> [snip]
>>>
>>>       
>>>>> Doesn't help, I still get the same crashes. I tried 2.6.22 again and
>>>>> it's rock solid by comparison.
>>>>>           
>>>> Do you mean, kvm-33 on top of 2.6.22, or the kvm modules from 2.6.22?
>>>> Please describe your configuration *exactly*.
>>>>         
>>> I'm using the kvm-33 *userspace* package (based on Debian's kvm-28
>>> packaging) and 2.6.23-rc1's KVM modules. I patched 2.6.23-rc1 with the
>>> patch you provided in your last email. So I'm not using -git HEAD.
>>>
>>> Maybe there's been additional necessary fixes to -git requiring me to
>>> update to HEAD? That wasn't clear from your last email.
>>>       
>> No,  that patch is the only potential fix post -rc1.  There are a few
>> other fixes there, but they are intended to avoid guest crashes, not
>> host crashes.
>>
>> What guest are you running?  Maybe I can reproduce it here.
>>     
>
> Right now, Windows XP. I'm pretty sure Linux (well, Debian Etch) works fine. I 
> could only get Windows to install with -no-acpi, but I run it with the 
> following (if this is at all useful):
>
> kvm -no-acpi -m 256 -hda $IMAGE -net nic -net tap,script=/etc/kvm/kvm-ifup
>
> Basically, the installer seems to work fine, but Windows seemed to have 
> problems after installing post-SP2 updates. Maybe that's why not everybody is 
> seeing it yet.
>
>   

How about the attached patch? (I haven't yet tried to reproduce, but 
this can cause an AMD-only oops).

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


[-- Attachment #2: kvm-fix-debug-registers-on-amd.patch --]
[-- Type: text/x-patch, Size: 855 bytes --]

commit 80917728e43e248155c019f743655806b582b099
Author: Avi Kivity <avi@qumranet.com>
Date:   Mon Jul 30 15:56:36 2007 +0300

    KVM: x86 emulator: disable writeback for debug register instructions
    
    These are handled internally by the instruction.
    
    Signed-off-by: Avi Kivity <avi@qumranet.com>

diff --git a/drivers/kvm/x86_emulate.c b/drivers/kvm/x86_emulate.c
index 1a90ba0..2136da5 100644
--- a/drivers/kvm/x86_emulate.c
+++ b/drivers/kvm/x86_emulate.c
@@ -1222,11 +1222,13 @@ twobyte_insn:
 		}
 		break;
 	case 0x21: /* mov from dr to reg */
+		no_wb = 1;
 		if (modrm_mod != 3)
 			goto cannot_emulate;
 		rc = emulator_get_dr(ctxt, modrm_reg, &_regs[modrm_rm]);
 		break;
 	case 0x23: /* mov from reg to dr */
+		no_wb = 1;
 		if (modrm_mod != 3)
 			goto cannot_emulate;
 		rc = emulator_set_dr(ctxt, modrm_reg, _regs[modrm_rm]);

Re: 2.6.23-rc1, KVM-AMD problem

[Alistair John Strachan]

On Monday 30 July 2007 14:00:13 Avi Kivity wrote:
> How about the attached patch? (I haven't yet tried to reproduce, but
> this can cause an AMD-only oops).

This seems to have fixed the problem. Thanks Avi.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.

Re: 2.6.23-rc1, KVM-AMD problem

[Avi Kivity]

Alistair John Strachan wrote:
> On Monday 30 July 2007 14:00:13 Avi Kivity wrote:
>   
>> How about the attached patch? (I haven't yet tried to reproduce, but
>> this can cause an AMD-only oops).
>>     
>
> This seems to have fixed the problem. Thanks Avi.
>
>   

Excellent.  I'll submit this fix for 2.6.23.


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

Re: 2.6.23-rc1, KVM-AMD problem

[Prakash Punnoor]

[-- Attachment #1: Type: text/plain, Size: 1682 bytes --]

Am Montag 30 Juli 2007 schrieb Avi Kivity:
> Alistair John Strachan wrote:
[...]
> > Basically, the installer seems to work fine, but Windows seemed to have
> > problems after installing post-SP2 updates. Maybe that's why not
> > everybody is seeing it yet.
>
> How about the attached patch? (I haven't yet tried to reproduce, but
> this can cause an AMD-only oops).

commit 80917728e43e248155c019f743655806b582b099
Author: Avi Kivity <avi@qumranet.com>
Date:   Mon Jul 30 15:56:36 2007 +0300

    KVM: x86 emulator: disable writeback for debug register instructions
    
    These are handled internally by the instruction.
    
    Signed-off-by: Avi Kivity <avi@qumranet.com>

diff --git a/drivers/kvm/x86_emulate.c b/drivers/kvm/x86_emulate.c
index 1a90ba0..2136da5 100644
--- a/drivers/kvm/x86_emulate.c
+++ b/drivers/kvm/x86_emulate.c
@@ -1222,11 +1222,13 @@ twobyte_insn:
                }
                break;
        case 0x21: /* mov from dr to reg */
+               no_wb = 1;
                if (modrm_mod != 3)
                        goto cannot_emulate;
                rc = emulator_get_dr(ctxt, modrm_reg, &_regs[modrm_rm]);
                break;
        case 0x23: /* mov from reg to dr */
+               no_wb = 1;
                if (modrm_mod != 3)
                        goto cannot_emulate;
                rc = emulator_set_dr(ctxt, modrm_reg, _regs[modrm_rm]);


Unfortunately this doesn't seem to have made it into 2.6.23-rc2. I need it as 
well, to make Windows XP boot up w/o hanging or reebooting my host machine.

Cheers,
-- 
(°=                 =°)
//\ Prakash Punnoor /\\
V_/                 \_V

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: 2.6.23-rc1, KVM-AMD problem

[Alistair John Strachan]

On Saturday 04 August 2007 14:17:34 Prakash Punnoor wrote:
> Unfortunately this doesn't seem to have made it into 2.6.23-rc2. I need it
> as well, to make Windows XP boot up w/o hanging or reebooting my host
> machine.

It isn't in 2.6.23-rc2. I guess Avi should re-send to Linus and hopefully 
it'll be in -rc3. Thanks for pointing this out.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.