scheduler fixes
Linus, please pull the latest sched-fixes git tree from:
git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip.git sched-fixes-for-linus
this fixes scheduler related sysbench (and other benchmarks) performance regressions. The sysbench throughput results compared to 2.6.24-rc4 are:
[tx/sec, higher is better]
.26-rc4 .26-rc4+sched-fixes
-------------------------------------------------
1: 687 727 +5.43%
2: 1460 1480 +1.34%
4: 2911 3080 +5.49%
8: 5496 5932 +7.35%
16: 10049 10822 +7.14%
32: 7598 10082 +24.64%
64: 7958 9331 +14.71%
128: 8169 8574 +4.73%
256: 8398 8469 +0.84%
-------------------------------------------------
SUM: 52730 58501 +9.86%with these fixes applied 2.6.26 is a net improvement over 2.6.25:
.25 .26-rc4+sched-fixes
-------------------------------------------------
1: 701 727 +3.47%
2: 1484 1480 -0.23%
4: 3050 3080 +0.96%
8: 5918 5932 +0.24%
16: 10794 10822 +0.26%
32: 10082 10082 -0.01%
64: 8894 9331 +4.68%
128: 7685 8574 +10.38%
256: 5258 8469 +37.92%
-------------------------------------------------
SUM: 53870 58501 +7.92%the group balancing revert makes the linecount rather large, but it was a relatively straightforward (albeit not automatic) revert.
it is a bit late because i was hoping for these patches from Peter to work out:
http://programming.kicks-ass.net/kernel-patches/sched-smp-group-fixes/
but they ended up being more risky than the revert, so we'll go for the revert for 2.6.26. Thanks to Peter and Mike for sorting out these issues.
Ingo
------------------>
Adrian Bunk (1):
show_schedstat(): fix memleak
Ingo Molnar (4):
revert ("sched: fair: weight calculations")
sched: cleanup
revert ("sched: fair-group: SMP-nice for group scheduling")
sched: re-tune NUMA topologies
Mike Galbraith (1):
sched: stop wake_affine from causing serious imbalance
Peter Zijlstra (1):
sched: fix sched_clock_cpu()
Roel Kluin (1):
sched: unite unlikely pairs in rt_policy() and schedule_debug()
include/linux/sched.h | 1 - include/linux/topology.h | 4 +- kernel/sched.c | 447 +++++----------------------------------------- kernel/sched_clock.c | 18 ++- kernel/sched_debug.c | 5 - kernel/sched_fair.c | 254 ++++++++++----------------- kernel/sched_rt.c | 4 - kernel/sched_stats.h | 1 + 8 files changed, 150 insertions(+), 584 deletions(-)
include/linux/sched.h +0 -1
@@ -766,7 +766,6 @@ struct sched_domain {struct sched_domain *child;/* bottom domain must be null terminated */ struct sched_group *groups;/* the balancing groups of the domain */ cpumask_t span;/* span of all CPUs in this domain */ -int first_cpu;/* cache of the first cpu in this domain */ unsigned long min_interval;/* Minimum balance interval ms */ unsigned long max_interval;/* Maximum balance interval ms */ unsigned int busy_factor;/* less balancing by factor if busy */
include/linux/topology.h +3 -1
@@ -166,7 +166,9 @@ void arch_update_cpu_topology(void); .busy_idx = 3, \ .idle_idx = 3, \ .flags =SD_LOAD_BALANCE \ - | SD_SERIALIZE, \ + |SD_BALANCE_NEWIDLE \ + |SD_WAKE_AFFINE \ + | SD_SERIALIZE, \ .last_balance = jiffies, \ .balance_interval = 64, \ }
kernel/sched.c +41 -406
613 lines not on this page · raw message
kernel/sched_clock.c +14 -4
@@ -59,22 +59,26 @@ static inline struct sched_clock_data *cpu_sdc(int cpu)return &per_cpu (sched_clock_data, cpu); } +static __read_mostly int sched_clock_running; +void sched_clock_init (void ) {u64 ktime_now =ktime_to_ns (ktime_get ()); -u64 now =0 ; +unsigned long now_jiffies = jiffies;int cpu;for_each_possible_cpu (cpu) {struct sched_clock_data *scd =cpu_sdc (cpu); scd->lock = (raw_spinlock_t )__RAW_SPIN_LOCK_UNLOCKED; - scd->prev_jiffies = jiffies; - scd->prev_raw = now; - scd->tick_raw = now; + scd->prev_jiffies = now_jiffies; + scd->prev_raw =0 ; + scd->tick_raw =0 ; scd->tick_gtod = ktime_now; scd->clock = ktime_now; } + + sched_clock_running =1 ; }/*
@@ -136,6 +140,9 @@ u64 sched_clock_cpu(int cpu)struct sched_clock_data *scd =cpu_sdc (cpu);u64 now, clock; +if (unlikely (!sched_clock_running)) +return 0ull ; +WARN_ON_ONCE (!irqs_disabled ()); now =sched_clock ();
@@ -174,6 +181,9 @@ void sched_clock_tick(void)struct sched_clock_data *scd =this_scd ();u64 now, now_gtod; +if (unlikely (!sched_clock_running)) +return ; +WARN_ON_ONCE (!irqs_disabled ()); now =sched_clock ();
kernel/sched_debug.c +0 -5
@@ -167,11 +167,6 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)#endif SEQ_printf (m," .%-30s: %ld\n" ,"nr_spread_over" , cfs_rq->nr_spread_over); -#ifdef CONFIG_FAIR_GROUP_SCHED -#ifdef CONFIG_SMP -SEQ_printf (m," .%-30s: %lu\n" ,"shares" , cfs_rq->shares); -#endif -#endif }static void print_cpu (struct seq_file *m,int cpu)
kernel/sched_fair.c +91 -163
372 lines not on this page · raw message
kernel/sched_rt.c +0 -4
@@ -513,8 +513,6 @@ static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)*/ for_each_sched_rt_entity (rt_se)enqueue_rt_entity (rt_se); - -inc_cpu_load (rq, p->se.load.weight); }static void dequeue_task_rt (struct rq *rq,struct task_struct *p,int sleep)
@@ -534,8 +532,6 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)if (rt_rq && rt_rq->rt_nr_running)enqueue_rt_entity (rt_se); } - -dec_cpu_load (rq, p->se.load.weight); }/*
kernel/sched_stats.h +1 -0
@@ -67,6 +67,7 @@ static int show_schedstat(struct seq_file *seq, void *v)preempt_enable ();#endif } +kfree (mask_str);return 0 ; }