Re: RCU: Call touch_nmi_watchdog() while printing stall warnings

[Tejun Heo]

Hello,

On Wed, Jan 10, 2018 at 02:34:50PM +0900, Sergey Senozhatsky wrote:
> Tejun, while we are at it, shall we also touch NMI in show_workqueue_state()?
> I suppose that show_pwq() can be quite verbose on some systems, can't it?
> 
> for instance: sysrq-t [irq] -> show_workqueue_state()

Ah, it's less likely but definitely possible.

> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 9960f8842b31..649c00a9af12 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -48,6 +48,7 @@
>  #include <linux/moduleparam.h>
>  #include <linux/uaccess.h>
>  #include <linux/sched/isolation.h>
> +#include <linux/nmi.h>
>  
>  #include "workqueue_internal.h"
>  
> @@ -4473,6 +4474,8 @@ void show_workqueue_state(void)
>  			if (pwq->nr_active || !list_empty(&pwq->delayed_works))
>  				show_pwq(pwq);
>  			spin_unlock_irqrestore(&pwq->pool->lock, flags);
> +
> +			touch_nmi_watchdog();

Can you send a patch w/ signed-off-by adding touch_nmi_watchdog() to
both outer loops in that function.  Theoretically, we can have a lot
of pools too.

Thanks.

-- 
tejun