Re: Re: BSOD "A clock interrupt was not recevied onasecondary processor within the allocated time interval"

[Frank van der Linden <Frank.Vanderlinden@Sun.COM>]

Steven Smith wrote:
>> Alternative Viridian interface support was checked in. When enabled, it
>> ought to be sufficient to disable these bugchecks. Œviridian=1¹ needs to be
>> specified in the domain config file.
> Hmm...  In order for the Viridian stuff to actually solve this
> problem, you need to set the relaxed-timers bit.  It doesn't look like
> the xen-unstable implementation does so.  Something like this might
> help:
> 
> diff -r f6b92526e916 xen/arch/x86/hvm/viridian.c
> --- a/xen/arch/x86/hvm/viridian.c	Tue Jan 06 09:14:39 2009 +0000
> +++ b/xen/arch/x86/hvm/viridian.c	Tue Jan 06 10:32:26 2009 +0000
> @@ -37,6 +37,7 @@
>  
>  /* Viridian CPUID 4000004, Implementation Recommendations. */
>  #define CPUID4A_MSR_BASED_APIC  (1 << 3)
> +#define CPUID4A_RELAX_TIMER_INT_HANDLING (1 << 5)
>  
>  int cpuid_viridian_leaves(unsigned int leaf, unsigned int *eax,
>                            unsigned int *ebx, unsigned int *ecx,
> @@ -84,7 +85,7 @@
>          if ( (d->arch.hvm_domain.viridian.guest_os_id.raw == 0) ||
>               (d->arch.hvm_domain.viridian.guest_os_id.fields.os < 4) )
>              break;
> -        *eax = CPUID4A_MSR_BASED_APIC;
> +        *eax = CPUID4A_MSR_BASED_APIC | CPUID4A_RELAX_TIMER_INT_HANDLING;
>          *ebx = 2047; /* long spin count */
>          break;
>      }
> 
> But note that I haven't tested this *at all*.

That appears to work. I have a test setup where I can introduce an 
arbitrary delay in qemu-dm. It's easy to trigger bug check 0x101 in 
Windows 2008 that way. Using viridian=1, with the relaxed timer bit set, 
I have been unable to crash Windows 2008.

Looks like a good thing to commit.

- Frank