Re: [PATCH] kvm: preemption must be disabled when calling smp_call_function_many

[Manali Shukla <manali.shukla@amd.com>]

On 6/7/2025 4:09 PM, Salah Triki wrote:
> {Disable, Enable} preemption {before, after} calling
> smp_call_function_many().
> 
> Signed-off-by: Salah Triki <salah.triki@gmail.com>
> ---
>  virt/kvm/kvm_main.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index eec82775c5bf..ab9593943846 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -209,7 +209,10 @@ static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
>  	if (cpumask_empty(cpus))
>  		return false;
>  
> +	preempt_disable();
>  	smp_call_function_many(cpus, ack_kick, NULL, wait);
> +	preempt_enable();
> +
>  	return true;
>  }
>  

Hi Salah,

I noticed that kvm_kick_many_cpus() is only invoked from two locations (from 
kvm_make_vcpus_request_mask() and kvm_make_all_cpus_request(), and in both cases, get_cpu() is
called prior to kvm_kick_many_cpus(), with a corresponding put_cpu() afterward. Since get_cpu()
disables preemption and put_cpu() re-enables it, preemption is already disabled during the
execution of kvm_kick_many_cpus().

Given that, could you clarify the motivation behind explicitly adding preempt_disable() and
preempt_enable() within kvm_kick_many_cpus()?

-Manali