[PATCH 1/1] arm64: Prevent offline CPU handling IRQs.

[Mark Rutland]

On Wed, Jul 13, 2022 at 09:28:43PM +0800, alvinzhan1234 at gmail.com wrote:
> From: Alvin Zhan <alvinzhan1234 at gmail.com>
> 
> Migrate all of the IRQs on this CPU after setting it offline may still cause
> the IRQs to be dispatched to this CPU.

I was under the impression that __cpu_disable() was called with interrupts
masked, since it's called via stop_machine_cpuslocked(), and hence interrupts
cannot be taken here (ignoring pseudo-NMIs for the moment).

Have you ever seen that actually happen?

Mark.

> To prevent offline CPU handling IRQs, we should migrate the IRQs away from
> this CPU before it is powered off.
> 
> Signed-off-by: Alvin Zhan <alvinzhan1234 at gmail.com>
> ---
>  arch/arm64/kernel/smp.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> index 62ed361a4..c4c3c5ed5 100644
> --- a/arch/arm64/kernel/smp.c
> +++ b/arch/arm64/kernel/smp.c
> @@ -302,6 +302,11 @@ int __cpu_disable(void)
>  	remove_cpu_topology(cpu);
>  	numa_remove_cpu(cpu);
>  
> +	/*
> +	 * Migrate IRQs away from this CPU first.
> +	 */
> +	irq_migrate_all_off_this_cpu();
> +
>  	/*
>  	 * Take this CPU offline.  Once we clear this, we can't return,
>  	 * and we must not schedule until we're ready to give up the cpu.
> @@ -309,11 +314,6 @@ int __cpu_disable(void)
>  	set_cpu_online(cpu, false);
>  	ipi_teardown(cpu);
>  
> -	/*
> -	 * OK - migrate IRQs away from this CPU
> -	 */
> -	irq_migrate_all_off_this_cpu();
> -
>  	return 0;
>  }
>  
> -- 
> 2.34.1
>