fpudna: fpcurthread == curthread XXXX times

[Sergi]

On 01/06/11 11:36, Sergi wrote:
> On 01/06/11 10:21, Kostik Belousov wrote:
>> On Wed, Jun 01, 2011 at 09:44:23AM +0200, Sergi wrote:
>>> Hello,
>>>
>>> I'm working with full virtual FreeBSD 8.2-RELEASE-p1 domU under debian
>>> squeeze and xen-hypervisor-4.0-amd64.
>>>
>>> If I cfg this hvm with cpu>  4 :
>>>
>>>   vcpus    = 5
>>>
>>> these messages block the server :
>>>
>>>   fpudna: fpcurthread == curthread XXXX times
>>>
>>> The machine is pingable but I'm unable to ssh to it.
>>>
>>> On single user it works fine, fsck an so on ok, but when switching to
>>> multiuser these fpudna messages start flooding.
>>>
>>> I've googled but haven't found anything; something from 2005 about 
>>> fpudna :
>>>
>>>   
>>> http://lists.freebsd.org/pipermail/freebsd-amd64/2005-April/004413.html
>>>
>>> and this link, but I don't have the options he mentions enabled on the
>>> kernel :
>>>
>>>   http://forums.freebsd.org/showthread.php?t=17979
>>>
>>> Has anyone stepped on this behaviour before?, is there any workaround?
>>> The machine really seems to detect cpu's available and responds to 
>>> keyboard
>>> on VNC, but it's impossible to see whats written down because of the
>>> messages flooding the screen.
>> You did not specified the architecture of the domu. From the message, 
>> I can
>> guess that your guest is running amd64 kernel. There are slight 
>> differences
>> in the handling of the FPU in i386 and amd64 that may matter there.
>>
>> The message you reported means that the FreeBSD kernel assumes that FPU
>> is currently loaded with the context of the current thread, but the
>> CR0.TS bit is set, meaning that FPU context is set for switch.
>>
>> AFAIR, HVM means that you run bare-metal kernel, right ? Most likely,
>> it is some issue with Xen itself. I am curious whether the following
>> will cause any usermode-visible regression for you:
>>
>> diff --git a/sys/amd64/amd64/fpu.c b/sys/amd64/amd64/fpu.c
>> index 08e5e57..a5ee853 100644
>> --- a/sys/amd64/amd64/fpu.c
>> +++ b/sys/amd64/amd64/fpu.c
>> @@ -394,14 +394,8 @@ fpudna(void)
>>       struct pcb *pcb;
>>
>>       critical_enter();
>> -    if (PCPU_GET(fpcurthread) == curthread) {
>> -        printf("fpudna: fpcurthread == curthread %d times\n",
>> -            ++err_count);
>> -        stop_emulating();
>> -        critical_exit();
>> -        return;
>> -    }
>> -    if (PCPU_GET(fpcurthread) != NULL) {
>> +    if (PCPU_GET(fpcurthread) != NULL&&
>> +        PCPU_GET(fpcurthread) != curthread) {
>>           printf("fpudna: fpcurthread = %p (%d), curthread = %p (%d)\n",
>>                  PCPU_GET(fpcurthread),
>>                  PCPU_GET(fpcurthread)->td_proc->p_pid,
> Hello,
>
> yes, sorry, amd64, and yes, hvm hardware virtual machine, not 
> paravirtual.
>
> So, you mean patching fpu.c and recompiling the kernel, right?, I'm 
> new to modifiying src files.
>
> Thanks for your help,
> Sergi
>
>
> _______________________________________________
> freebsd-xen at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-xen
> To unsubscribe, send any mail to "freebsd-xen-unsubscribe at freebsd.org"
>
Hello,

well, I patched fpu.c, recompiled the kernel, and booted ok with 4 vcpu.
Then I tried to boot with 5 vcpus and got :

kernel trap 22 with interrupts disabled
...
kernel trap 22 with interrupts disabled
Fatal double fault
rip = 0xffffffff8067865a
rsp = 0xffffff8000000000
rbp = 0xffffff8000000040
cpuid = 4; apic id = 08
panic: double fault
cpuid = 4

4 vcpus is the maximum number of vcpus I can use.

How do you think I can debug this in order to provide more information?

Thanks,
regards,
Sergi