Re: [PATCH] watchdog: Respect watchdog cpumask on CPU hotplug

[Oleg Nesterov]

On 03/26, Thomas Gleixner wrote:
>
> The rework of the watchdog core to use cpu_stop_work broke the watchdog
> cpumask on CPU hotplug.
>
> The watchdog_enable/disable() functions are now called unconditionally from
> the hotplug callback, i.e. even on CPUs which are not in the watchdog
> cpumask.
>
> Only invoke them when the plugged CPU is in the watchdog cpumask.
>
> Fixes: 9cf57731b63e ("watchdog/softlockup: Replace "watchdog/%u" threads with cpu_stop_work")
> Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> ---
>  kernel/watchdog.c |    6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> --- a/kernel/watchdog.c
> +++ b/kernel/watchdog.c
> @@ -547,13 +547,15 @@ static void softlockup_start_all(void)
>
>  int lockup_detector_online_cpu(unsigned int cpu)
>  {
> -	watchdog_enable(cpu);
> +	if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
> +		watchdog_enable(cpu);
>  	return 0;
>  }
>
>  int lockup_detector_offline_cpu(unsigned int cpu)
>  {
> -	watchdog_disable(cpu);
> +	if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
> +		watchdog_disable(cpu);
>  	return 0;
>  }

IIUC without this fix an NMI watchdog can too be enabled at boot time even
if the initial watchdog_cpumask = housekeeping_cpumask(HK_FLAG_TIMER) doesn't
include the plugged CPU.

And after that writing 0 to /proc/sys/kernel/nmi_watchdog clears
NMI_WATCHDOG_ENABLED but this can't disable NMI watchdog's outside of
watchdog_allowed_mask.

So may be this can explain the problem reported by Maxime ?
See https://lore.kernel.org/lkml/b99c5a25-a5fe-18dd-2f1d-bdd6834f03e5@xxxxxxxxxx/

Oleg.