Re: [PATCH 00/15] perf: KVM: Fix, optimize, and clean up callbacks

[Peter Zijlstra]

On Fri, Aug 27, 2021 at 04:01:45PM +0800, Like Xu wrote:
> On 27/8/2021 3:44 pm, Peter Zijlstra wrote:

> > You just have to make sure all static_call() invocations that started
> > before unreg are finished before continuing with the unload.
> > synchronize_rcu() can help with that.
> 
> Do you mean something like that:
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 64e310ff4f3a..e7d310af7509 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -8465,6 +8465,7 @@ void kvm_arch_exit(void)
>  #endif
>  	kvm_lapic_exit();
>  	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
> +	synchronize_rcu();
> 
>  	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
>  		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index e466fc8176e1..63ae56c5d133 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -6508,6 +6508,7 @@ EXPORT_SYMBOL_GPL(perf_register_guest_info_callbacks);
>  int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *cbs)
>  {
>  	perf_guest_cbs = NULL;
> +	arch_perf_update_guest_cbs();

I'm thinking the synchronize_rcu() should go here, and access to
perf_guest_cbs should be wrapped to yell when called with preemption
enabled.

But yes..

>  	return 0;
>  }
>  EXPORT_SYMBOL_GPL(perf_unregister_guest_info_callbacks);
> 
> > 
> > This is module unload 101. Nothing specific to static_call().
> >