Received-SPF: pass (google.com: domain of linux-kernel+bounces-195677-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: Oleg Nesterov <onestero@redhat.com>
Date: Thu, 30 May 2024 12:52:38 -0400
To: Frederic Weisbecker <frederic@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>, Nicholas Piggin <npiggin@gmail.com>,
	Peter Zijlstra <peterz@infradead.org>, Phil Auld <pauld@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Chris von Recklinghausen <crecklin@redhat.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] tick/nohz_full: don't abuse smp_call_function_single()
 in tick_setup_device()
Message-ID: <Zliu1qVoUB3Y1rTE@redhat.com>
References: <20240522151742.GA10400@redhat.com>
 <20240528122019.GA28794@redhat.com>
 <ZliSt-RDyxf1bZ_t@localhost.localdomain>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZliSt-RDyxf1bZ_t@localhost.localdomain>

Frederic,

Thanks for review.

On 05/30, Frederic Weisbecker wrote:
>
> Looks good, but can we have a WARN_ONCE(tick_do_timer_cpu != tick_do_timer_boot_cpu)
> right before that, just to make sure our assumptions above are right forever and
> the boot CPU hasn't stopped the tick up to that point?

Sure, I thought about the additional sanity checks too. Although I had something
different in mind.

Frederic, et al, I am on private trip again without my working laptop, can't read
the code. I'll reply on Saturday, OK?

Oleg.

> 
> And after all, pushing a bit further your subsequent patch, can we get rid of
> tick_do_timer_boot_cpu and ifdefery altogether? Such as:
> 
> diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
> index fb0fdec8719a..63a7bd405de7 100644
> --- a/kernel/time/tick-common.c
> +++ b/kernel/time/tick-common.c
> @@ -48,14 +48,6 @@ ktime_t tick_next_period;
>   *    procedure also covers cpu hotplug.
>   */
>  int tick_do_timer_cpu __read_mostly = TICK_DO_TIMER_BOOT;
> -#ifdef CONFIG_NO_HZ_FULL
> -/*
> - * tick_do_timer_boot_cpu indicates the boot CPU temporarily owns
> - * tick_do_timer_cpu and it should be taken over by an eligible secondary
> - * when one comes online.
> - */
> -static int tick_do_timer_boot_cpu __read_mostly = -1;
> -#endif
>  
>  /*
>   * Debugging: see timer_list.c
> @@ -177,26 +169,6 @@ void tick_setup_periodic(struct clock_event_device *dev, int broadcast)
>  	}
>  }
>  
> -#ifdef CONFIG_NO_HZ_FULL
> -static void giveup_do_timer(void *info)
> -{
> -	int cpu = *(unsigned int *)info;
> -
> -	WARN_ON(tick_do_timer_cpu != smp_processor_id());
> -
> -	tick_do_timer_cpu = cpu;
> -}
> -
> -static void tick_take_do_timer_from_boot(void)
> -{
> -	int cpu = smp_processor_id();
> -	int from = tick_do_timer_boot_cpu;
> -
> -	if (from >= 0 && from != cpu)
> -		smp_call_function_single(from, giveup_do_timer, &cpu, 1);
> -}
> -#endif
> -
>  /*
>   * Setup the tick device
>   */
> @@ -211,29 +183,28 @@ static void tick_setup_device(struct tick_device *td,
>  	 * First device setup ?
>  	 */
>  	if (!td->evtdev) {
> +		int timekeeper = READ_ONCE(tick_do_timer_cpu);
>  		/*
>  		 * If no cpu took the do_timer update, assign it to
>  		 * this cpu:
>  		 */
> -		if (tick_do_timer_cpu == TICK_DO_TIMER_BOOT) {
> +		if (timekeeper == TICK_DO_TIMER_BOOT) {
>  			tick_do_timer_cpu = cpu;
>  			tick_next_period = ktime_get();
> -#ifdef CONFIG_NO_HZ_FULL
> +		} else if (timekeeper == TICK_DO_TIMER_NONE) {
> +			if (WARN_ON_ONCE(tick_nohz_full_enabled()))
> +				WRITE_ONCE(tick_do_timer_cpu, cpu);
> +		} else if (tick_nohz_full_cpu(timekeeper) && !tick_nohz_full_cpu(cpu)) {
>  			/*
> -			 * The boot CPU may be nohz_full, in which case set
> -			 * tick_do_timer_boot_cpu so the first housekeeping
> -			 * secondary that comes up will take do_timer from
> -			 * us.
> +			 * The boot CPU will stay in periodic (NOHZ disabled)
> +			 * mode until clocksource_done_booting() called after
> +			 * smp_init() selects a high resolution clocksource and
> +			 * timekeeping_notify() kicks the NOHZ stuff alive.
> +			 *
> +			 * So this WRITE_ONCE can only race with the READ_ONCE
> +			 * check in tick_periodic() but this race is harmless.
>  			 */
> -			if (tick_nohz_full_cpu(cpu))
> -				tick_do_timer_boot_cpu = cpu;
> -
> -		} else if (tick_do_timer_boot_cpu != -1 &&
> -						!tick_nohz_full_cpu(cpu)) {
> -			tick_take_do_timer_from_boot();
> -			tick_do_timer_boot_cpu = -1;
> -			WARN_ON(tick_do_timer_cpu != cpu);
> -#endif
> +			WRITE_ONCE(tick_do_timer_cpu, cpu);
>  		}
>  
>  		/*
>