Re: [PATCH v6] arm64: sdei: abort running SDEI handlers during crash

[James Morse]

Hi Scott,

On 27/06/2023 01:29, D Scott Phillips wrote:
> Interrupts are blocked in SDEI context, per the SDEI spec: "The client
> interrupts cannot preempt the event handler." If we crashed in the SDEI
> handler-running context (as with ACPI's AGDI) then we need to clean up the
> SDEI state before proceeding to the crash kernel so that the crash kernel
> can have working interrupts.
> 
> Track the active SDEI handler per-cpu so that we can COMPLETE_AND_RESUME
> the handler, discarding the interrupted context.

I still argue this is a firmware bug. That preempt text was supposed to mean "PSTATE.DAIF
get set", the whole "GIC abstraction" thing got shoehorned in much later.

But I agree we need to work around it in linux.


> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
> index f9040bd61081..285fe7ad490d 100644
> --- a/drivers/firmware/arm_sdei.c
> +++ b/drivers/firmware/arm_sdei.c
> @@ -1095,3 +1095,22 @@ int sdei_event_handler(struct pt_regs *regs,
>  	return err;
>  }
>  NOKPROBE_SYMBOL(sdei_event_handler);
> +
> +void sdei_handler_abort(void)
> +{
> +	/*
> +	 * If the crash happened in an SDEI event handler then we need to
> +	 * finish the handler with the firmware so that we can have working
> +	 * interrupts in the crash kernel.
> +	 */
> +	if (__this_cpu_read(sdei_active_critical_event)) {
> +	        pr_warn("still in SDEI critical event context, attempting to finish handler.\n");
> +	        __sdei_handler_abort();
> +	        __this_cpu_write(sdei_active_critical_event, NULL);
> +	}
> +	if (__this_cpu_read(sdei_active_normal_event)) {
> +	        pr_warn("still in SDEI normal event context, attempting to finish handler.\n");
> +	        __sdei_handler_abort();
> +	        __this_cpu_write(sdei_active_normal_event, NULL);
> +	}
> +}

I'm not sure why this moved out to drivers/firmware when the only caller is the arch code,
but it doesn't matter...

Reviewed-by: James Morse <james.morse@xxxxxxx>


Thanks,

James