Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1668723 > unrolled thread

[PATCH 0/3] sched: A few nohz_full improvements

Started byFrederic Weisbecker <fweisbec@gmail.com>
First post2017-06-19 04:20 +0200
Last post2017-06-19 04:20 +0200
Articles 2 — 1 participant

Back to article view | Back to linux.kernel


Contents

  [PATCH 0/3] sched: A few nohz_full improvements Frederic Weisbecker <fweisbec@gmail.com> - 2017-06-19 04:20 +0200
    [PATCH 2/3] nohz: Move idle balancer registration to idle path Frederic Weisbecker <fweisbec@gmail.com> - 2017-06-19 04:20 +0200

#1668723 — [PATCH 0/3] sched: A few nohz_full improvements

FromFrederic Weisbecker <fweisbec@gmail.com>
Date2017-06-19 04:20 +0200
Subject[PATCH 0/3] sched: A few nohz_full improvements
Message-ID<tTUsh-4VS-3@gated-at.bofh.it>
Hi,

As I was working on removing the remaining 1Hz, I stepped in the last
(AFAIK) remaining buggy bits in scheduler_tick() on nohz_full.

Hopefully I can finally offload this tick in the next patchsets.

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
	nohz/0hz

HEAD: c1811204f1fd1d170223c0d548a57456685e831d

Thanks,
	Frederic
---

Frederic Weisbecker (3):
      sched/loadavg: Generalize idle naming to nohz
      nohz: Move idle balancer registration to idle path
      sched: Spare idle load balancing on nohz_full CPUs


 Documentation/trace/ftrace.txt |  2 +-
 include/linux/sched/nohz.h     |  8 +++----
 kernel/sched/fair.c            |  4 ++++
 kernel/sched/loadavg.c         | 51 +++++++++++++++++++++---------------------
 kernel/time/tick-sched.c       |  9 ++++----
 5 files changed, 40 insertions(+), 34 deletions(-)

[toc] | [next] | [standalone]


#1668725 — [PATCH 2/3] nohz: Move idle balancer registration to idle path

FromFrederic Weisbecker <fweisbec@gmail.com>
Date2017-06-19 04:20 +0200
Subject[PATCH 2/3] nohz: Move idle balancer registration to idle path
Message-ID<tTUsi-4VS-19@gated-at.bofh.it>
In reply to#1668723
The idle load balancing registration path assumes that we only stop the
tick when the CPU is idle, ignoring the nohz full case. As a result, a
nohz full CPU that is running a task may be chosen to perform idle load
balancing.

Lets make sure that only CPUs in dynticks idle mode can be picked as
idle load balancers.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 kernel/time/tick-sched.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 1c9a508..77c3be5 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -785,7 +785,6 @@ static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts,
 	 * the scheduler tick in nohz_restart_sched_tick.
 	 */
 	if (!ts->tick_stopped) {
-		nohz_balance_enter_idle(cpu);
 		calc_load_nohz_start();
 		cpu_load_update_nohz_start();
 
@@ -938,8 +937,10 @@ static void __tick_nohz_idle_enter(struct tick_sched *ts)
 			ts->idle_expires = expires;
 		}
 
-		if (!was_stopped && ts->tick_stopped)
+		if (!was_stopped && ts->tick_stopped) {
 			ts->idle_jiffies = ts->last_jiffies;
+			nohz_balance_enter_idle(cpu);
+		}
 	}
 }
 
-- 
2.7.4

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web