Re: [PATCH V1] nohz: exit nohz idle before invoking softirq

[Frederic Weisbecker]

On Mon, Dec 28, 2020 at 11:47:48AM +0800, chenshiyan wrote:
> From: "shiyan.csy" <shiyan.csy@xxxxxxxxxxxxxxx>
> 
> exit nohz idle before invoking softirq, or it maymiss
> some ticks during softirq.
> 
> Signed-off-by: Shiyan Chen <chenshiyan@xxxxxxxxxxxxxxxxx>
> ---
>  kernel/softirq.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/softirq.c b/kernel/softirq.c
> index 9d71046..59bd6fe 100644
> --- a/kernel/softirq.c
> +++ b/kernel/softirq.c
> @@ -404,6 +404,10 @@ static inline void tick_irq_exit(void)
>  		if (!in_irq())
>  			tick_nohz_irq_exit();
>  	}
> +
> +	/* need to exit nohz idle if there's a softirq */
> +	if (in_serving_softirq() || local_softirq_pending())
> +		tick_nohz_idle_exit();
>  #endif
>  }
>  
> @@ -416,10 +420,11 @@ static inline void __irq_exit_rcu(void)
>  #endif
>  	account_hardirq_exit(current);
>  	preempt_count_sub(HARDIRQ_OFFSET);
> -	if (!in_interrupt() && local_softirq_pending())
> -		invoke_softirq();
>  
> +	/* must call before invoke_softirq */
>  	tick_irq_exit();
> +	if (!in_interrupt() && local_softirq_pending())
> +		invoke_softirq();

You can't reverse the order here because the softirqs may update
the next tick expiry. And that needs to be taken into account by
tick_nohz_full_update_tick().

But what issue are you trying to solve exactly?

Thanks.

>  }
>  
>  /**
> -- 
> 1.8.3.1
>