If the processor does not have an "Always Running APIC Timer" (aka ARAT),
we should not give guests direct access to MWAIT. The LAPIC timer would
stop ticking in deep C-states, so any host deadlines would not wakeup the
host kernel.
The host kernel intel_idle driver handles this by switching to broadcast
mode when ARAT is not available and MWAIT is issued with a deep C-state
that would stop the LAPIC timer. When MWAIT is passed through, we can not
tell when MWAIT is issued.
So just disable this capability when LAPIC ARAT is not available. I am not
even sure if there are any CPUs with VMX support but no LAPIC ARAT or not.
Cc: Paolo Bonzini <[email protected]>
Cc: Radim Krčmář <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: H. Peter Anvin <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Reported-by: Wanpeng Li <[email protected]>
Signed-off-by: KarimAllah Ahmed <[email protected]>
---
arch/x86/kvm/x86.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index b2ff74b..0334b25 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2819,7 +2819,8 @@ static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
static inline bool kvm_can_mwait_in_guest(void)
{
return boot_cpu_has(X86_FEATURE_MWAIT) &&
- !boot_cpu_has_bug(X86_BUG_MONITOR);
+ !boot_cpu_has_bug(X86_BUG_MONITOR) &&
+ boot_cpu_has(X86_FEATURE_ARAT);
}
int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
--
2.7.4
On 11/04/2018 11:16, KarimAllah Ahmed wrote:
> If the processor does not have an "Always Running APIC Timer" (aka ARAT),
> we should not give guests direct access to MWAIT. The LAPIC timer would
> stop ticking in deep C-states, so any host deadlines would not wakeup the
> host kernel.
>
> The host kernel intel_idle driver handles this by switching to broadcast
> mode when ARAT is not available and MWAIT is issued with a deep C-state
> that would stop the LAPIC timer. When MWAIT is passed through, we can not
> tell when MWAIT is issued.
>
> So just disable this capability when LAPIC ARAT is not available. I am not
> even sure if there are any CPUs with VMX support but no LAPIC ARAT or not.
The alternative would be to remove ARAT from __do_cpuid_ent when the
host lacks it and userspace has requested direct access to MWAIT.
However, this is simpler.
Queued, thanks.
Paolo
On Wed, 2018-04-11 at 11:16 +0200, KarimAllah Ahmed wrote:
> If the processor does not have an "Always Running APIC Timer" (aka ARAT),
> we should not give guests direct access to MWAIT. The LAPIC timer would
> stop ticking in deep C-states, so any host deadlines would not wakeup the
> host kernel.
>
> The host kernel intel_idle driver handles this by switching to broadcast
> mode when ARAT is not available and MWAIT is issued with a deep C-state
> that would stop the LAPIC timer. When MWAIT is passed through, we can not
> tell when MWAIT is issued.
>
> So just disable this capability when LAPIC ARAT is not available. I am not
> even sure if there are any CPUs with VMX support but no LAPIC ARAT or not.
ARAT was added on WSM, so NHM, the Core 2 family and a few PSC SKUs
support VMX+MWAIT but not ARAT.
> Cc: Paolo Bonzini <[email protected]>
> Cc: Radim Krčmář <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: H. Peter Anvin <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> Reported-by: Wanpeng Li <[email protected]>
> Signed-off-by: KarimAllah Ahmed <[email protected]>
> ---
> arch/x86/kvm/x86.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index b2ff74b..0334b25 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2819,7 +2819,8 @@ static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
> static inline bool kvm_can_mwait_in_guest(void)
> {
> return boot_cpu_has(X86_FEATURE_MWAIT) &&
> - !boot_cpu_has_bug(X86_BUG_MONITOR);
> + !boot_cpu_has_bug(X86_BUG_MONITOR) &&
> + boot_cpu_has(X86_FEATURE_ARAT);
> }
>
> int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)