[PATCH] x86/kvm: Don't use PVspinlock when mwait is advertised

[PATCH] x86/kvm: Don't use PVspinlock when mwait is advertised

[lirongqing]

From: Li RongQing <lirongqing@baidu.com>

MWAIT is advertised in host is not overcommitted scenario, however,
pvspinlock should be enabled in host overcommitted scenario. Let's
add the MWAIT checking when enabling pvspinlock

Signed-off-by: Li RongQing <lirongqing@baidu.com>
---
 arch/x86/kernel/kvm.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
index 1cceac5..dfa1451 100644
--- a/arch/x86/kernel/kvm.c
+++ b/arch/x86/kernel/kvm.c
@@ -1087,6 +1087,11 @@ void __init kvm_spinlock_init(void)
 		goto out;
 	}
 
+	if (boot_cpu_has(X86_FEATURE_MWAIT)) {
+		pr_info("PV spinlocks disabled with mwait\n");
+		goto out;
+	}
+
 	if (num_possible_cpus() == 1) {
 		pr_info("PV spinlocks disabled, single CPU\n");
 		goto out;
-- 
2.9.4

Re: [PATCH] x86/kvm: Don't use PVspinlock when mwait is advertised

[Sean Christopherson]

On Sat, Feb 11, 2023, lirongqing@baidu.com wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> MWAIT is advertised in host is not overcommitted scenario, however,
> pvspinlock should be enabled in host overcommitted scenario. Let's
> add the MWAIT checking when enabling pvspinlock
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
>  arch/x86/kernel/kvm.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
> index 1cceac5..dfa1451 100644
> --- a/arch/x86/kernel/kvm.c
> +++ b/arch/x86/kernel/kvm.c
> @@ -1087,6 +1087,11 @@ void __init kvm_spinlock_init(void)
>  		goto out;
>  	}
>  
> +	if (boot_cpu_has(X86_FEATURE_MWAIT)) {

IMO this is a bad idea.  Odds are good that there are less-than-perfect VM configs
that advertise MONITOR+MWAIT but don't disable interception.  Ugh, but we've already
added this logic for other features guaraded by KVM_HINTS_REALTIME in commit
40cd58dbf121 ("x86/kvm: Don't use PV TLB/yield when mwait is advertised")

Why doesn't the host simply advertise KVM_HINTS_REALTIME?