Balaji Rao wrote:
> Hello,
>
> I was trying to enable the use of nmi watchdog within a linux guest running 
> in kvm. I have done it 
> by allowing direct access to perfmon msrs using the MSR_BITMAP field in vmcs 
> region.
>
> Most of the times the NMI Watchdog Test in the guest fails, but with a finite 
> number of NMI's 
> received  by the guest. But randomly it does work! Whenever it fails, i get 
> this vmwrite error :
>
> vmwrite error: reg 4016 value 80000202 (err 164061)
>
> I have a few questions.
>
> 1. How are NMI's supposed to be delivered to the guest ? I did this by adding 
> a new op to
> kvm_x86_ops.
>   

That's fine. ops are there to be extended.

> 2. How am I supposed to handle perfmon MSRs ? Direct access may pose problems 
> during migration. But 
> am not sure how costly emulation by abstraction would be..
> I have not yet considered saving the MSRS upon vmexits to allow multiple VMs 
> use the MSRs. I think i 
> can do them easily when i get this working.
>   

I don't think there's a sane way to emulate the non-architectural 
perfmon counters. It may be possible to do so for the architectural 
ones, but I'm not sure. So pass-through (with saving and restoring) is 
the only option.

> Here's the code. Please tell me what dumb mistake I am doing.
>
>
> diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c 
> b/arch/x86/kernel/cpu/perfctr-watchdog.c
> index c02541e..276048a 100644
> --- a/arch/x86/kernel/cpu/perfctr-watchdog.c
> +++ b/arch/x86/kernel/cpu/perfctr-watchdog.c
> @@ -342,7 +342,7 @@ static const struct wd_ops k7_wd_ops = {
>  #define P6_EVNTSEL_INT               (1 << 20)
>  #define P6_EVNTSEL_OS                (1 << 17)
>  #define P6_EVNTSEL_USR               (1 << 16)
> -#define P6_EVENT_CPU_CLOCKS_NOT_HALTED       0x79
> +#define P6_EVENT_CPU_CLOCKS_NOT_HALTED       0x3C
>  #define P6_NMI_EVENT         P6_EVENT_CPU_CLOCKS_NOT_HALTED
>   

What's this?


>  
>  static int setup_p6_watchdog(unsigned nmi_hz)
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index 2cbee94..73e9361 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -25,6 +25,8 @@
>  #include <linux/hrtimer.h>
>  #include <linux/io.h>
>  #include <linux/module.h>
> +#include <linux/kdebug.h>
> +#include <linux/notifier.h>
>  #include <asm/processor.h>
>  #include <asm/msr.h>
>  #include <asm/page.h>
> @@ -740,9 +742,12 @@ static void apic_mmio_write(struct kvm_io_device *this,
>               apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
>               break;
>  
> +     case APIC_LVTPC:
> +             /* Enable PC NMI*/
> +             if (val == APIC_DM_NMI)
> +                     apic_write(APIC_LVTPC,val);
>   

You're writing to the precious host apic here.

- need to disallow if the host is using it
- need to prevent illegal values
- need to use some sort of perfmon api rather than directly banging on 
the apic

> @@ -790,6 +795,18 @@ static int apic_mmio_range(struct kvm_io_device *this, 
> gpa_t addr)
>       return ret;
>  }
>  
> +static int nmi_notify(struct notifier_block *self,unsigned long val, void 
> *data) {
> +
> +        struct kvm *kvm;
> +        kvm  = list_entry(vm_list.next, struct kvm, vm_list);
> +     kvm_x86_ops->inject_nmi(kvm->vcpus[0]);
> +        return NOTIFY_STOP;
> +}
>   

You're not guaranteed to be in vcpu context here, that's what's causing 
the vmwrite errors.

Enabling on guest entry and disabling on guest exit is critical, both 
for accuracy and correctness.




-- 
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/
_______________________________________________
kvm-devel mailing list
kvm-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/kvm-devel

Reply via email to