[lkml]   [2014]   [Feb]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Patch in this message
Subject[PATCH 1/2] timer: Spare IPI when deferrable timer is queued on idle remote targets
From: Viresh Kumar <>

When a timer is enqueued or modified on a remote target, the latter is
expected to see and handle this timer on its next tick. However if the
target is idle and CONFIG_NO_HZ_IDLE=y, the CPU may be sleeping tickless
and the timer may be ignored.

wake_up_nohz_cpu() takes care of that by setting TIF_NEED_RESCHED and
sending an IPI to idle targets so that the tick is reevaluated on the
idle loop through the tick_nohz_idle_*() APIs.

Now this is all performed regardless of the power properties of the
timer. If the timer is deferrable, idle targets don't need to be woken
up. Only the next buzy tick needs to care about it, and no IPI kick
is needed for that to happen.

So lets spare the IPI on idle targets when the timer is deferrable.

Meanwhile we keep the current behaviour on full dynticks targets. We can
spare IPIs on idle full dynticks targets as well but some tricky races
against idle_cpu() must be dealt all along to make sure that the timer
is well handled after idle exit. We can deal with that later since
NO_HZ_FULL already has more important powersaving issues.

Reported-by: Thomas Gleixner <>
Signed-off-by: Viresh Kumar <>
Cc: Ingo Molnar <>
Cc: Paul Gortmaker <>
Cc: Paul E. McKenney <>
Cc: Peter Zijlstra <>
Cc: Steven Rostedt <>
Cc: Thomas Gleixner <>
Signed-off-by: Frederic Weisbecker <>
kernel/timer.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/kernel/timer.c b/kernel/timer.c
index accfd24..881f883 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -939,8 +939,15 @@ void add_timer_on(struct timer_list *timer, int cpu)
* with the timer by holding the timer base lock. This also
* makes sure that a CPU on the way to stop its tick can not
* evaluate the timer wheel.
+ *
+ * Spare the IPI for deferrable timers on idle targets though.
+ * The next buzy ticks will take care of it. Except full dynticks
+ * require special care against races with idle_cpu(), lets deal
+ * with that later.
- wake_up_nohz_cpu(cpu);
+ if (!tbase_get_deferrable(timer->base) || tick_nohz_full_cpu(cpu))
+ wake_up_nohz_cpu(cpu);
spin_unlock_irqrestore(&base->lock, flags);

 \ /
  Last update: 2014-02-14 13:41    [W:0.028 / U:0.604 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site