Commit aa483808516ca5cacfa0e5849691f64fec25828e

Authored by Venkatesh Pallipadi
Committed by Ingo Molnar
1 parent 305e6835e0

sched: Remove irq time from available CPU power

The idea was suggested by Peter Zijlstra here:

  http://marc.info/?l=linux-kernel&m=127476934517534&w=2

irq time is technically not available to the tasks running on the CPU.
This patch removes irq time from CPU power piggybacking on
sched_rt_avg_update().

Tested this by keeping CPU X busy with a network intensive task having 75%
oa a single CPU irq processing (hard+soft) on a 4-way system. And start seven
cycle soakers on the system. Without this change, there will be two tasks on
each CPU. With this change, there is a single task on irq busy CPU X and
remaining 7 tasks are spread around among other 3 CPUs.

Signed-off-by: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286237003-12406-8-git-send-email-venki@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

Showing 3 changed files with 30 additions and 1 deletions Side-by-side Diff

... ... @@ -519,6 +519,10 @@
519 519 u64 avg_idle;
520 520 #endif
521 521  
  522 +#ifdef CONFIG_IRQ_TIME_ACCOUNTING
  523 + u64 prev_irq_time;
  524 +#endif
  525 +
522 526 /* calc_load related fields */
523 527 unsigned long calc_load_update;
524 528 long calc_load_active;
... ... @@ -643,6 +647,7 @@
643 647 #endif /* CONFIG_CGROUP_SCHED */
644 648  
645 649 static u64 irq_time_cpu(int cpu);
  650 +static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time);
646 651  
647 652 inline void update_rq_clock(struct rq *rq)
648 653 {
... ... @@ -654,6 +659,8 @@
654 659 irq_time = irq_time_cpu(cpu);
655 660 if (rq->clock - irq_time > rq->clock_task)
656 661 rq->clock_task = rq->clock - irq_time;
  662 +
  663 + sched_irq_time_avg_update(rq, irq_time);
657 664 }
658 665 }
659 666  
660 667  
... ... @@ -1985,12 +1992,23 @@
1985 1992 local_irq_restore(flags);
1986 1993 }
1987 1994  
  1995 +static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time)
  1996 +{
  1997 + if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) {
  1998 + u64 delta_irq = curr_irq_time - rq->prev_irq_time;
  1999 + rq->prev_irq_time = curr_irq_time;
  2000 + sched_rt_avg_update(rq, delta_irq);
  2001 + }
  2002 +}
  2003 +
1988 2004 #else
1989 2005  
1990 2006 static u64 irq_time_cpu(int cpu)
1991 2007 {
1992 2008 return 0;
1993 2009 }
  2010 +
  2011 +static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { }
1994 2012  
1995 2013 #endif
1996 2014  
... ... @@ -2275,7 +2275,13 @@
2275 2275 u64 total, available;
2276 2276  
2277 2277 total = sched_avg_period() + (rq->clock - rq->age_stamp);
2278   - available = total - rq->rt_avg;
  2278 +
  2279 + if (unlikely(total < rq->rt_avg)) {
  2280 + /* Ensures that power won't end up being negative */
  2281 + available = 0;
  2282 + } else {
  2283 + available = total - rq->rt_avg;
  2284 + }
2279 2285  
2280 2286 if (unlikely((s64)total < SCHED_LOAD_SCALE))
2281 2287 total = SCHED_LOAD_SCALE;
kernel/sched_features.h
... ... @@ -61,4 +61,9 @@
61 61 * release the lock. Decreases scheduling overhead.
62 62 */
63 63 SCHED_FEAT(OWNER_SPIN, 1)
  64 +
  65 +/*
  66 + * Decrement CPU power based on irq activity
  67 + */
  68 +SCHED_FEAT(NONIRQ_POWER, 1)