linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Frederic Weisbecker <frederic@kernel.org>
To: Anna-Maria Gleixner <anna-maria@linutronix.de>,
	Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@redhat.com>,
	keescook@chromium.org, Christoph Hellwig <hch@lst.de>,
	John Stultz <john.stultz@linaro.org>
Subject: Re: [PATCH v4 01/36] timers: Use static keys for migrate_enable/nohz_active
Date: Thu, 11 Jan 2018 05:25:05 +0100	[thread overview]
Message-ID: <20180111042504.GB11633@lerouge> (raw)
In-Reply-To: <20171221104205.7269-2-anna-maria@linutronix.de>

On Thu, Dec 21, 2017 at 11:41:30AM +0100, Anna-Maria Gleixner wrote:
> From: Thomas Gleixner <tglx@linutronix.de>
> 
> The members migrate_enable and nohz_active in the timer/hrtimer per CPU
> bases have been introduced to avoid accessing global variables for these
> decisions.
> 
> Still that results in a (cache hot) load and conditional branch, which can
> be avoided by using static keys.
> 
> Implement it with static keys and optimize for the most critical case of
> high performance networking which tends to disable the timer migration
> functionality.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
> ---
>  include/linux/hrtimer.h     |  4 --
>  kernel/time/hrtimer.c       | 17 +++------
>  kernel/time/tick-internal.h | 19 ++++++----
>  kernel/time/tick-sched.c    |  2 +-
>  kernel/time/timer.c         | 91 +++++++++++++++++++++++----------------------
>  5 files changed, 65 insertions(+), 68 deletions(-)
> 
> diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> index f8e1845aa464..4ac74dff59f0 100644
> --- a/kernel/time/tick-internal.h
> +++ b/kernel/time/tick-internal.h
> @@ -150,14 +150,19 @@ static inline void tick_nohz_init(void) { }
>  
>  #ifdef CONFIG_NO_HZ_COMMON
>  extern unsigned long tick_nohz_active;
> -#else
> +extern void timers_update_nohz(void);
> +extern struct static_key_false timers_nohz_active;
> +static inline bool is_timers_nohz_active(void)
> +{
> +	return static_branch_unlikely(&timers_nohz_active);

Shouldn't we expect instead that timers_nohz_active is a likely scenario?
I guess deactivating nohz is for specific workloads that can't stand the
deep state wake up latency.

Also perhaps the above symbols should be harmonized, something like:

timers_nohz_update()
timers_nohz_active_key
timers_nohz_active()

> diff --git a/kernel/time/timer.c b/kernel/time/timer.c
> index ffebcf878fba..1e2140a23044 100644
> --- a/kernel/time/timer.c
> +++ b/kernel/time/timer.c
> @@ -200,8 +200,6 @@ struct timer_base {
>  	unsigned long		clk;
>  	unsigned long		next_expiry;
>  	unsigned int		cpu;
> -	bool			migration_enabled;
> -	bool			nohz_active;
>  	bool			is_idle;
>  	bool			must_forward_clk;
>  	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
> @@ -210,45 +208,59 @@ struct timer_base {
>  
>  static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
>  
> -#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
> +#ifdef CONFIG_NO_HZ_COMMON
> +
> +DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
> +static DEFINE_MUTEX(timer_keys_mutex);
> +
> +static void timer_update_keys(struct work_struct *work);
> +static DECLARE_WORK(timer_update_work, timer_update_keys);
> +
> +#ifdef CONFIG_SMP
>  unsigned int sysctl_timer_migration = 1;
>  
> -void timers_update_migration(bool update_nohz)
> +DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
> +
> +static void timers_update_migration(void)
>  {
>  	bool on = sysctl_timer_migration && tick_nohz_active;
> -	unsigned int cpu;
>  
> -	/* Avoid the loop, if nothing to update */
> -	if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on)
> -		return;
> +	if (on)

You may as well put the condition directly ;)

> +		static_branch_enable(&timers_migration_enabled);
> +	else
> +		static_branch_disable(&timers_migration_enabled);
> +}

Thanks!

  parent reply	other threads:[~2018-01-11  4:25 UTC|newest]

Thread overview: 78+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-12-21 10:41 [PATCH v4 00/36] hrtimer: Provide softirq context hrtimers Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 01/36] timers: Use static keys for migrate_enable/nohz_active Anna-Maria Gleixner
2017-12-22 15:45   ` [PATCH v5 " Sebastian Andrzej Siewior
2018-01-14 22:13     ` Thomas Gleixner
2018-01-14 22:30       ` [PATCH v6 " Thomas Gleixner
2018-01-11  4:25   ` Frederic Weisbecker [this message]
2018-01-16  3:40   ` [tip:timers/core] hrtimer: Optimize the hrtimer code by using static keys for migration_enable/nohz_active tip-bot for Thomas Gleixner
2017-12-21 10:41 ` [PATCH v4 02/36] hrtimer: Correct blantanly wrong comment Anna-Maria Gleixner
2018-01-11 18:28   ` Frederic Weisbecker
2018-01-16  1:02   ` Ingo Molnar
2018-01-16  3:40   ` [tip:timers/core] hrtimer: Correct blatantly incorrect comment tip-bot for Thomas Gleixner
2017-12-21 10:41 ` [PATCH v4 03/36] hrtimer: Fix kerneldoc for struct hrtimer_cpu_base Anna-Maria Gleixner
2018-01-16  3:41   ` [tip:timers/core] hrtimer: Fix kerneldoc syntax for 'struct hrtimer_cpu_base' tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 04/36] hrtimer: Cleanup clock argument in schedule_hrtimeout_range_clock() Anna-Maria Gleixner
2018-01-16  3:41   ` [tip:timers/core] hrtimer: Clean up the 'int clock' parameter of schedule_hrtimeout_range_clock() tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 05/36] hrtimer: Fix hrtimer function description Anna-Maria Gleixner
2018-01-16  3:42   ` [tip:timers/core] hrtimer: Fix hrtimer_start[_range_ns]() function descriptions tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 06/36] hrtimer: Ensure POSIX compliance (relative CLOCK_REALTIME hrtimers) Anna-Maria Gleixner
2018-01-16  3:42   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 07/36] hrtimer: Cleanup hrtimer_mode enum Anna-Maria Gleixner
2018-01-16  3:43   ` [tip:timers/core] hrtimer: Clean up 'enum hrtimer_mode' tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 08/36] tracing/hrtimer: Take all clock bases and modes into account Anna-Maria Gleixner
2018-01-16  3:43   ` [tip:timers/core] tracing/hrtimer: Fix tracing bugs by taking " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 09/36] tracing/hrtimer: Print hrtimer mode in hrtimer_start tracepoint Anna-Maria Gleixner
2018-01-16  3:43   ` [tip:timers/core] tracing/hrtimer: Print the hrtimer mode in the 'hrtimer_start' tracepoint tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 10/36] hrtimer: Switch for loop to _ffs() evaluation Anna-Maria Gleixner
2018-01-16  3:44   ` [tip:timers/core] hrtimer: Switch 'for' " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 11/36] hrtimer: Store running timer in hrtimer_clock_base Anna-Maria Gleixner
2018-01-16  3:44   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 12/36] hrtimer: Make room in struct hrtimer_cpu_base Anna-Maria Gleixner
2018-01-16  3:45   ` [tip:timers/core] hrtimer: Make room in 'struct hrtimer_cpu_base' tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 13/36] hrtimer: Reduce conditional code (hres_active) Anna-Maria Gleixner
2018-01-16  3:45   ` [tip:timers/core] hrtimer: Make the hrtimer_cpu_base::hres_active field unconditional, to simplify the code tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 14/36] hrtimer: Use accesor functions instead of direct access Anna-Maria Gleixner
2018-01-16  3:46   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 15/36] hrtimer: Make the remote enqueue check unconditional Anna-Maria Gleixner
2018-01-16  3:46   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 16/36] hrtimer: Make hrtimer_cpu_base.next_timer handling unconditional Anna-Maria Gleixner
2018-01-16  3:46   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 17/36] hrtimer: Make hrtimer_reprogramm() unconditional Anna-Maria Gleixner
2018-01-16  3:47   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 18/36] hrtimer: Make hrtimer_force_reprogramm() unconditionally available Anna-Maria Gleixner
2018-01-16  3:47   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 19/36] hrtimer: Unify handling of hrtimer remove Anna-Maria Gleixner
2018-01-16  3:48   ` [tip:timers/core] hrtimer: Unify hrtimer removal handling tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 20/36] hrtimer: Unify handling of remote enqueue Anna-Maria Gleixner
2018-01-16  3:48   ` [tip:timers/core] hrtimer: Unify remote enqueue handling tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 21/36] hrtimer: Make remote enqueue decision less restrictive Anna-Maria Gleixner
2018-01-16  3:49   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 22/36] hrtimer: Remove base argument from hrtimer_reprogram() Anna-Maria Gleixner
2018-01-16  3:49   ` [tip:timers/core] hrtimer: Remove the 'base' parameter " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 23/36] hrtimer: Split hrtimer_start_range_ns() Anna-Maria Gleixner
2018-01-16  3:49   ` [tip:timers/core] hrtimer: Factor out __hrtimer_start_range_ns() tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 24/36] hrtimer: Split __hrtimer_get_next_event() Anna-Maria Gleixner
2018-01-16  3:50   ` [tip:timers/core] hrtimer: Factor out __hrtimer_next_event_base() tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 25/36] hrtimer: Use irqsave/irqrestore around __run_hrtimer() Anna-Maria Gleixner
2018-01-16  3:50   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 26/36] hrtimer: Add clock bases and hrtimer mode for soft irq context Anna-Maria Gleixner
2018-01-16  3:51   ` [tip:timers/core] hrtimer: Add clock bases and hrtimer mode for softirq context tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 27/36] hrtimer: Prepare handling of hard and softirq based hrtimers Anna-Maria Gleixner
2018-01-16  3:51   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 28/36] hrtimer: Implement support for " Anna-Maria Gleixner
2018-01-16 10:22   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 29/36] hrtimer: Implement SOFT/HARD clock base selection Anna-Maria Gleixner
2018-01-16 10:22   ` [tip:timers/core] " tip-bot for Anna-Maria Gleixner
2017-12-21 10:41 ` [PATCH v4 30/36] can/bcm: Replace hrtimer_tasklet with softirq based hrtimer Anna-Maria Gleixner
2017-12-21 10:42 ` [PATCH v4 31/36] mac80211_hwsim: Replace hrtimer tasklet with softirq hrtimer Anna-Maria Gleixner
2018-01-04 15:12   ` Johannes Berg
2018-01-04 15:18     ` Thomas Gleixner
2017-12-21 10:42 ` [PATCH v4 32/36] xfrm: " Anna-Maria Gleixner
2017-12-21 10:42 ` [PATCH v4 33/36] softirq: Remove tasklet_hrtimer Anna-Maria Gleixner
2017-12-21 10:42 ` [PATCH v4 34/36] ALSA/dummy: Replace tasklet with softirq hrtimer Anna-Maria Gleixner
2018-01-16 10:23   ` [tip:timers/core] " tip-bot for Thomas Gleixner
2017-12-21 10:42 ` [PATCH v4 35/36] usb/gadget/NCM: " Anna-Maria Gleixner
2018-01-16 10:23   ` [tip:timers/core] " tip-bot for Thomas Gleixner
2017-12-21 10:42 ` [PATCH v4 36/36] net/mvpp2: " Anna-Maria Gleixner
2018-01-16  1:39 ` [PATCH v4 00/36] hrtimer: Provide softirq context hrtimers Ingo Molnar
2018-01-16  2:03   ` Ingo Molnar

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20180111042504.GB11633@lerouge \
    --to=frederic@kernel.org \
    --cc=anna-maria@linutronix.de \
    --cc=bigeasy@linutronix.de \
    --cc=hch@lst.de \
    --cc=john.stultz@linaro.org \
    --cc=keescook@chromium.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@redhat.com \
    --cc=peterz@infradead.org \
    --cc=tglx@linutronix.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).