aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@woody.linux-foundation.org>2007-08-09 11:23:31 -0400
committerLinus Torvalds <torvalds@woody.linux-foundation.org>2007-08-09 11:23:31 -0400
commitbe12014dd7750648fde33e1e45cac24dc9a8be6d (patch)
treecaf9715d2c37f3c08c25e51c9f416c2bbf956236 /kernel
parente3bcf5e2785aa49f75f36a8d27d601891a7ff12b (diff)
parent7cff8cf61cac15fa29a1ca802826d2bcbca66152 (diff)
Merge git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched
* git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched: (61 commits) sched: refine negative nice level granularity sched: fix update_stats_enqueue() reniced codepath sched: round a bit better sched: make the multiplication table more accurate sched: optimize update_rq_clock() calls in the load-balancer sched: optimize activate_task() sched: clean up set_curr_task_fair() sched: remove __update_rq_clock() call from entity_tick() sched: move the __update_rq_clock() call to scheduler_tick() sched debug: remove the 'u64 now' parameter from print_task()/_rq() sched: remove the 'u64 now' local variables sched: remove the 'u64 now' parameter from deactivate_task() sched: remove the 'u64 now' parameter from dequeue_task() sched: remove the 'u64 now' parameter from enqueue_task() sched: remove the 'u64 now' parameter from dec_nr_running() sched: remove the 'u64 now' parameter from inc_nr_running() sched: remove the 'u64 now' parameter from dec_load() sched: remove the 'u64 now' parameter from inc_load() sched: remove the 'u64 now' parameter from update_curr_load() sched: remove the 'u64 now' parameter from ->task_new() ...
Diffstat (limited to 'kernel')
-rw-r--r--kernel/sched.c339
-rw-r--r--kernel/sched_debug.c16
-rw-r--r--kernel/sched_fair.c212
-rw-r--r--kernel/sched_idletask.c10
-rw-r--r--kernel/sched_rt.c48
5 files changed, 303 insertions, 322 deletions
diff --git a/kernel/sched.c b/kernel/sched.c
index 72bb9483d949..b0afd8db1396 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -318,15 +318,19 @@ static inline int cpu_of(struct rq *rq)
318} 318}
319 319
320/* 320/*
321 * Per-runqueue clock, as finegrained as the platform can give us: 321 * Update the per-runqueue clock, as finegrained as the platform can give
322 * us, but without assuming monotonicity, etc.:
322 */ 323 */
323static unsigned long long __rq_clock(struct rq *rq) 324static void __update_rq_clock(struct rq *rq)
324{ 325{
325 u64 prev_raw = rq->prev_clock_raw; 326 u64 prev_raw = rq->prev_clock_raw;
326 u64 now = sched_clock(); 327 u64 now = sched_clock();
327 s64 delta = now - prev_raw; 328 s64 delta = now - prev_raw;
328 u64 clock = rq->clock; 329 u64 clock = rq->clock;
329 330
331#ifdef CONFIG_SCHED_DEBUG
332 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
333#endif
330 /* 334 /*
331 * Protect against sched_clock() occasionally going backwards: 335 * Protect against sched_clock() occasionally going backwards:
332 */ 336 */
@@ -349,18 +353,12 @@ static unsigned long long __rq_clock(struct rq *rq)
349 353
350 rq->prev_clock_raw = now; 354 rq->prev_clock_raw = now;
351 rq->clock = clock; 355 rq->clock = clock;
352
353 return clock;
354} 356}
355 357
356static inline unsigned long long rq_clock(struct rq *rq) 358static void update_rq_clock(struct rq *rq)
357{ 359{
358 int this_cpu = smp_processor_id(); 360 if (likely(smp_processor_id() == cpu_of(rq)))
359 361 __update_rq_clock(rq);
360 if (this_cpu == cpu_of(rq))
361 return __rq_clock(rq);
362
363 return rq->clock;
364} 362}
365 363
366/* 364/*
@@ -386,9 +384,12 @@ unsigned long long cpu_clock(int cpu)
386{ 384{
387 unsigned long long now; 385 unsigned long long now;
388 unsigned long flags; 386 unsigned long flags;
387 struct rq *rq;
389 388
390 local_irq_save(flags); 389 local_irq_save(flags);
391 now = rq_clock(cpu_rq(cpu)); 390 rq = cpu_rq(cpu);
391 update_rq_clock(rq);
392 now = rq->clock;
392 local_irq_restore(flags); 393 local_irq_restore(flags);
393 394
394 return now; 395 return now;
@@ -637,6 +638,11 @@ static u64 div64_likely32(u64 divident, unsigned long divisor)
637 638
638#define WMULT_SHIFT 32 639#define WMULT_SHIFT 32
639 640
641/*
642 * Shift right and round:
643 */
644#define RSR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
645
640static unsigned long 646static unsigned long
641calc_delta_mine(unsigned long delta_exec, unsigned long weight, 647calc_delta_mine(unsigned long delta_exec, unsigned long weight,
642 struct load_weight *lw) 648 struct load_weight *lw)
@@ -644,18 +650,17 @@ calc_delta_mine(unsigned long delta_exec, unsigned long weight,
644 u64 tmp; 650 u64 tmp;
645 651
646 if (unlikely(!lw->inv_weight)) 652 if (unlikely(!lw->inv_weight))
647 lw->inv_weight = WMULT_CONST / lw->weight; 653 lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1;
648 654
649 tmp = (u64)delta_exec * weight; 655 tmp = (u64)delta_exec * weight;
650 /* 656 /*
651 * Check whether we'd overflow the 64-bit multiplication: 657 * Check whether we'd overflow the 64-bit multiplication:
652 */ 658 */
653 if (unlikely(tmp > WMULT_CONST)) { 659 if (unlikely(tmp > WMULT_CONST))
654 tmp = ((tmp >> WMULT_SHIFT/2) * lw->inv_weight) 660 tmp = RSR(RSR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
655 >> (WMULT_SHIFT/2); 661 WMULT_SHIFT/2);
656 } else { 662 else
657 tmp = (tmp * lw->inv_weight) >> WMULT_SHIFT; 663 tmp = RSR(tmp * lw->inv_weight, WMULT_SHIFT);
658 }
659 664
660 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); 665 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
661} 666}
@@ -703,11 +708,14 @@ static void update_load_sub(struct load_weight *lw, unsigned long dec)
703 * the relative distance between them is ~25%.) 708 * the relative distance between them is ~25%.)
704 */ 709 */
705static const int prio_to_weight[40] = { 710static const int prio_to_weight[40] = {
706/* -20 */ 88818, 71054, 56843, 45475, 36380, 29104, 23283, 18626, 14901, 11921, 711 /* -20 */ 88761, 71755, 56483, 46273, 36291,
707/* -10 */ 9537, 7629, 6103, 4883, 3906, 3125, 2500, 2000, 1600, 1280, 712 /* -15 */ 29154, 23254, 18705, 14949, 11916,
708/* 0 */ NICE_0_LOAD /* 1024 */, 713 /* -10 */ 9548, 7620, 6100, 4904, 3906,
709/* 1 */ 819, 655, 524, 419, 336, 268, 215, 172, 137, 714 /* -5 */ 3121, 2501, 1991, 1586, 1277,
710/* 10 */ 110, 87, 70, 56, 45, 36, 29, 23, 18, 15, 715 /* 0 */ 1024, 820, 655, 526, 423,
716 /* 5 */ 335, 272, 215, 172, 137,
717 /* 10 */ 110, 87, 70, 56, 45,
718 /* 15 */ 36, 29, 23, 18, 15,
711}; 719};
712 720
713/* 721/*
@@ -718,14 +726,14 @@ static const int prio_to_weight[40] = {
718 * into multiplications: 726 * into multiplications:
719 */ 727 */
720static const u32 prio_to_wmult[40] = { 728static const u32 prio_to_wmult[40] = {
721/* -20 */ 48356, 60446, 75558, 94446, 118058, 729 /* -20 */ 48388, 59856, 76040, 92818, 118348,
722/* -15 */ 147573, 184467, 230589, 288233, 360285, 730 /* -15 */ 147320, 184698, 229616, 287308, 360437,
723/* -10 */ 450347, 562979, 703746, 879575, 1099582, 731 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
724/* -5 */ 1374389, 1717986, 2147483, 2684354, 3355443, 732 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
725/* 0 */ 4194304, 5244160, 6557201, 8196502, 10250518, 733 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
726/* 5 */ 12782640, 16025997, 19976592, 24970740, 31350126, 734 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
727/* 10 */ 39045157, 49367440, 61356675, 76695844, 95443717, 735 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
728/* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, 736 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
729}; 737};
730 738
731static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); 739static void activate_task(struct rq *rq, struct task_struct *p, int wakeup);
@@ -745,8 +753,7 @@ static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
745 unsigned long max_nr_move, unsigned long max_load_move, 753 unsigned long max_nr_move, unsigned long max_load_move,
746 struct sched_domain *sd, enum cpu_idle_type idle, 754 struct sched_domain *sd, enum cpu_idle_type idle,
747 int *all_pinned, unsigned long *load_moved, 755 int *all_pinned, unsigned long *load_moved,
748 int this_best_prio, int best_prio, int best_prio_seen, 756 int *this_best_prio, struct rq_iterator *iterator);
749 struct rq_iterator *iterator);
750 757
751#include "sched_stats.h" 758#include "sched_stats.h"
752#include "sched_rt.c" 759#include "sched_rt.c"
@@ -782,14 +789,14 @@ static void __update_curr_load(struct rq *rq, struct load_stat *ls)
782 * This function is called /before/ updating rq->ls.load 789 * This function is called /before/ updating rq->ls.load
783 * and when switching tasks. 790 * and when switching tasks.
784 */ 791 */
785static void update_curr_load(struct rq *rq, u64 now) 792static void update_curr_load(struct rq *rq)
786{ 793{
787 struct load_stat *ls = &rq->ls; 794 struct load_stat *ls = &rq->ls;
788 u64 start; 795 u64 start;
789 796
790 start = ls->load_update_start; 797 start = ls->load_update_start;
791 ls->load_update_start = now; 798 ls->load_update_start = rq->clock;
792 ls->delta_stat += now - start; 799 ls->delta_stat += rq->clock - start;
793 /* 800 /*
794 * Stagger updates to ls->delta_fair. Very frequent updates 801 * Stagger updates to ls->delta_fair. Very frequent updates
795 * can be expensive. 802 * can be expensive.
@@ -798,30 +805,28 @@ static void update_curr_load(struct rq *rq, u64 now)
798 __update_curr_load(rq, ls); 805 __update_curr_load(rq, ls);
799} 806}
800 807
801static inline void 808static inline void inc_load(struct rq *rq, const struct task_struct *p)
802inc_load(struct rq *rq, const struct task_struct *p, u64 now)
803{ 809{
804 update_curr_load(rq, now); 810 update_curr_load(rq);
805 update_load_add(&rq->ls.load, p->se.load.weight); 811 update_load_add(&rq->ls.load, p->se.load.weight);
806} 812}
807 813
808static inline void 814static inline void dec_load(struct rq *rq, const struct task_struct *p)
809dec_load(struct rq *rq, const struct task_struct *p, u64 now)
810{ 815{
811 update_curr_load(rq, now); 816 update_curr_load(rq);
812 update_load_sub(&rq->ls.load, p->se.load.weight); 817 update_load_sub(&rq->ls.load, p->se.load.weight);
813} 818}
814 819
815static void inc_nr_running(struct task_struct *p, struct rq *rq, u64 now) 820static void inc_nr_running(struct task_struct *p, struct rq *rq)
816{ 821{
817 rq->nr_running++; 822 rq->nr_running++;
818 inc_load(rq, p, now); 823 inc_load(rq, p);
819} 824}
820 825
821static void dec_nr_running(struct task_struct *p, struct rq *rq, u64 now) 826static void dec_nr_running(struct task_struct *p, struct rq *rq)
822{ 827{
823 rq->nr_running--; 828 rq->nr_running--;
824 dec_load(rq, p, now); 829 dec_load(rq, p);
825} 830}
826 831
827static void set_load_weight(struct task_struct *p) 832static void set_load_weight(struct task_struct *p)
@@ -848,18 +853,16 @@ static void set_load_weight(struct task_struct *p)
848 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; 853 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
849} 854}
850 855
851static void 856static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
852enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
853{ 857{
854 sched_info_queued(p); 858 sched_info_queued(p);
855 p->sched_class->enqueue_task(rq, p, wakeup, now); 859 p->sched_class->enqueue_task(rq, p, wakeup);
856 p->se.on_rq = 1; 860 p->se.on_rq = 1;
857} 861}
858 862
859static void 863static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
860dequeue_task(struct rq *rq, struct task_struct *p, int sleep, u64 now)
861{ 864{
862 p->sched_class->dequeue_task(rq, p, sleep, now); 865 p->sched_class->dequeue_task(rq, p, sleep);
863 p->se.on_rq = 0; 866 p->se.on_rq = 0;
864} 867}
865 868
@@ -914,13 +917,11 @@ static int effective_prio(struct task_struct *p)
914 */ 917 */
915static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) 918static void activate_task(struct rq *rq, struct task_struct *p, int wakeup)
916{ 919{
917 u64 now = rq_clock(rq);
918
919 if (p->state == TASK_UNINTERRUPTIBLE) 920 if (p->state == TASK_UNINTERRUPTIBLE)
920 rq->nr_uninterruptible--; 921 rq->nr_uninterruptible--;
921 922
922 enqueue_task(rq, p, wakeup, now); 923 enqueue_task(rq, p, wakeup);
923 inc_nr_running(p, rq, now); 924 inc_nr_running(p, rq);
924} 925}
925 926
926/* 927/*
@@ -928,13 +929,13 @@ static void activate_task(struct rq *rq, struct task_struct *p, int wakeup)
928 */ 929 */
929static inline void activate_idle_task(struct task_struct *p, struct rq *rq) 930static inline void activate_idle_task(struct task_struct *p, struct rq *rq)
930{ 931{
931 u64 now = rq_clock(rq); 932 update_rq_clock(rq);
932 933
933 if (p->state == TASK_UNINTERRUPTIBLE) 934 if (p->state == TASK_UNINTERRUPTIBLE)
934 rq->nr_uninterruptible--; 935 rq->nr_uninterruptible--;
935 936
936 enqueue_task(rq, p, 0, now); 937 enqueue_task(rq, p, 0);
937 inc_nr_running(p, rq, now); 938 inc_nr_running(p, rq);
938} 939}
939 940
940/* 941/*
@@ -942,13 +943,11 @@ static inline void activate_idle_task(struct task_struct *p, struct rq *rq)
942 */ 943 */
943static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) 944static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep)
944{ 945{
945 u64 now = rq_clock(rq);
946
947 if (p->state == TASK_UNINTERRUPTIBLE) 946 if (p->state == TASK_UNINTERRUPTIBLE)
948 rq->nr_uninterruptible++; 947 rq->nr_uninterruptible++;
949 948
950 dequeue_task(rq, p, sleep, now); 949 dequeue_task(rq, p, sleep);
951 dec_nr_running(p, rq, now); 950 dec_nr_running(p, rq);
952} 951}
953 952
954/** 953/**
@@ -1516,6 +1515,7 @@ out_set_cpu:
1516 1515
1517out_activate: 1516out_activate:
1518#endif /* CONFIG_SMP */ 1517#endif /* CONFIG_SMP */
1518 update_rq_clock(rq);
1519 activate_task(rq, p, 1); 1519 activate_task(rq, p, 1);
1520 /* 1520 /*
1521 * Sync wakeups (i.e. those types of wakeups where the waker 1521 * Sync wakeups (i.e. those types of wakeups where the waker
@@ -1647,12 +1647,11 @@ void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
1647 unsigned long flags; 1647 unsigned long flags;
1648 struct rq *rq; 1648 struct rq *rq;
1649 int this_cpu; 1649 int this_cpu;
1650 u64 now;
1651 1650
1652 rq = task_rq_lock(p, &flags); 1651 rq = task_rq_lock(p, &flags);
1653 BUG_ON(p->state != TASK_RUNNING); 1652 BUG_ON(p->state != TASK_RUNNING);
1654 this_cpu = smp_processor_id(); /* parent's CPU */ 1653 this_cpu = smp_processor_id(); /* parent's CPU */
1655 now = rq_clock(rq); 1654 update_rq_clock(rq);
1656 1655
1657 p->prio = effective_prio(p); 1656 p->prio = effective_prio(p);
1658 1657
@@ -1666,8 +1665,8 @@ void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
1666 * Let the scheduling class do new task startup 1665 * Let the scheduling class do new task startup
1667 * management (if any): 1666 * management (if any):
1668 */ 1667 */
1669 p->sched_class->task_new(rq, p, now); 1668 p->sched_class->task_new(rq, p);
1670 inc_nr_running(p, rq, now); 1669 inc_nr_running(p, rq);
1671 } 1670 }
1672 check_preempt_curr(rq, p); 1671 check_preempt_curr(rq, p);
1673 task_rq_unlock(rq, &flags); 1672 task_rq_unlock(rq, &flags);
@@ -1954,7 +1953,6 @@ static void update_cpu_load(struct rq *this_rq)
1954 unsigned long total_load = this_rq->ls.load.weight; 1953 unsigned long total_load = this_rq->ls.load.weight;
1955 unsigned long this_load = total_load; 1954 unsigned long this_load = total_load;
1956 struct load_stat *ls = &this_rq->ls; 1955 struct load_stat *ls = &this_rq->ls;
1957 u64 now = __rq_clock(this_rq);
1958 int i, scale; 1956 int i, scale;
1959 1957
1960 this_rq->nr_load_updates++; 1958 this_rq->nr_load_updates++;
@@ -1962,7 +1960,7 @@ static void update_cpu_load(struct rq *this_rq)
1962 goto do_avg; 1960 goto do_avg;
1963 1961
1964 /* Update delta_fair/delta_exec fields first */ 1962 /* Update delta_fair/delta_exec fields first */
1965 update_curr_load(this_rq, now); 1963 update_curr_load(this_rq);
1966 1964
1967 fair_delta64 = ls->delta_fair + 1; 1965 fair_delta64 = ls->delta_fair + 1;
1968 ls->delta_fair = 0; 1966 ls->delta_fair = 0;
@@ -1970,8 +1968,8 @@ static void update_cpu_load(struct rq *this_rq)
1970 exec_delta64 = ls->delta_exec + 1; 1968 exec_delta64 = ls->delta_exec + 1;
1971 ls->delta_exec = 0; 1969 ls->delta_exec = 0;
1972 1970
1973 sample_interval64 = now - ls->load_update_last; 1971 sample_interval64 = this_rq->clock - ls->load_update_last;
1974 ls->load_update_last = now; 1972 ls->load_update_last = this_rq->clock;
1975 1973
1976 if ((s64)sample_interval64 < (s64)TICK_NSEC) 1974 if ((s64)sample_interval64 < (s64)TICK_NSEC)
1977 sample_interval64 = TICK_NSEC; 1975 sample_interval64 = TICK_NSEC;
@@ -2026,6 +2024,8 @@ static void double_rq_lock(struct rq *rq1, struct rq *rq2)
2026 spin_lock(&rq1->lock); 2024 spin_lock(&rq1->lock);
2027 } 2025 }
2028 } 2026 }
2027 update_rq_clock(rq1);
2028 update_rq_clock(rq2);
2029} 2029}
2030 2030
2031/* 2031/*
@@ -2166,8 +2166,7 @@ static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
2166 unsigned long max_nr_move, unsigned long max_load_move, 2166 unsigned long max_nr_move, unsigned long max_load_move,
2167 struct sched_domain *sd, enum cpu_idle_type idle, 2167 struct sched_domain *sd, enum cpu_idle_type idle,
2168 int *all_pinned, unsigned long *load_moved, 2168 int *all_pinned, unsigned long *load_moved,
2169 int this_best_prio, int best_prio, int best_prio_seen, 2169 int *this_best_prio, struct rq_iterator *iterator)
2170 struct rq_iterator *iterator)
2171{ 2170{
2172 int pulled = 0, pinned = 0, skip_for_load; 2171 int pulled = 0, pinned = 0, skip_for_load;
2173 struct task_struct *p; 2172 struct task_struct *p;
@@ -2192,12 +2191,8 @@ next:
2192 */ 2191 */
2193 skip_for_load = (p->se.load.weight >> 1) > rem_load_move + 2192 skip_for_load = (p->se.load.weight >> 1) > rem_load_move +
2194 SCHED_LOAD_SCALE_FUZZ; 2193 SCHED_LOAD_SCALE_FUZZ;
2195 if (skip_for_load && p->prio < this_best_prio) 2194 if ((skip_for_load && p->prio >= *this_best_prio) ||
2196 skip_for_load = !best_prio_seen && p->prio == best_prio;
2197 if (skip_for_load ||
2198 !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { 2195 !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
2199
2200 best_prio_seen |= p->prio == best_prio;
2201 p = iterator->next(iterator->arg); 2196 p = iterator->next(iterator->arg);
2202 goto next; 2197 goto next;
2203 } 2198 }
@@ -2211,8 +2206,8 @@ next:
2211 * and the prescribed amount of weighted load. 2206 * and the prescribed amount of weighted load.
2212 */ 2207 */
2213 if (pulled < max_nr_move && rem_load_move > 0) { 2208 if (pulled < max_nr_move && rem_load_move > 0) {
2214 if (p->prio < this_best_prio) 2209 if (p->prio < *this_best_prio)
2215 this_best_prio = p->prio; 2210 *this_best_prio = p->prio;
2216 p = iterator->next(iterator->arg); 2211 p = iterator->next(iterator->arg);
2217 goto next; 2212 goto next;
2218 } 2213 }
@@ -2231,32 +2226,52 @@ out:
2231} 2226}
2232 2227
2233/* 2228/*
2234 * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted 2229 * move_tasks tries to move up to max_load_move weighted load from busiest to
2235 * load from busiest to this_rq, as part of a balancing operation within 2230 * this_rq, as part of a balancing operation within domain "sd".
2236 * "domain". Returns the number of tasks moved. 2231 * Returns 1 if successful and 0 otherwise.
2237 * 2232 *
2238 * Called with both runqueues locked. 2233 * Called with both runqueues locked.
2239 */ 2234 */
2240static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, 2235static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
2241 unsigned long max_nr_move, unsigned long max_load_move, 2236 unsigned long max_load_move,
2242 struct sched_domain *sd, enum cpu_idle_type idle, 2237 struct sched_domain *sd, enum cpu_idle_type idle,
2243 int *all_pinned) 2238 int *all_pinned)
2244{ 2239{
2245 struct sched_class *class = sched_class_highest; 2240 struct sched_class *class = sched_class_highest;
2246 unsigned long load_moved, total_nr_moved = 0, nr_moved; 2241 unsigned long total_load_moved = 0;
2247 long rem_load_move = max_load_move; 2242 int this_best_prio = this_rq->curr->prio;
2248 2243
2249 do { 2244 do {
2250 nr_moved = class->load_balance(this_rq, this_cpu, busiest, 2245 total_load_moved +=
2251 max_nr_move, (unsigned long)rem_load_move, 2246 class->load_balance(this_rq, this_cpu, busiest,
2252 sd, idle, all_pinned, &load_moved); 2247 ULONG_MAX, max_load_move - total_load_moved,
2253 total_nr_moved += nr_moved; 2248 sd, idle, all_pinned, &this_best_prio);
2254 max_nr_move -= nr_moved;
2255 rem_load_move -= load_moved;
2256 class = class->next; 2249 class = class->next;
2257 } while (class && max_nr_move && rem_load_move > 0); 2250 } while (class && max_load_move > total_load_moved);
2258 2251
2259 return total_nr_moved; 2252 return total_load_moved > 0;
2253}
2254
2255/*
2256 * move_one_task tries to move exactly one task from busiest to this_rq, as
2257 * part of active balancing operations within "domain".
2258 * Returns 1 if successful and 0 otherwise.
2259 *
2260 * Called with both runqueues locked.
2261 */
2262static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
2263 struct sched_domain *sd, enum cpu_idle_type idle)
2264{
2265 struct sched_class *class;
2266 int this_best_prio = MAX_PRIO;
2267
2268 for (class = sched_class_highest; class; class = class->next)
2269 if (class->load_balance(this_rq, this_cpu, busiest,
2270 1, ULONG_MAX, sd, idle, NULL,
2271 &this_best_prio))
2272 return 1;
2273
2274 return 0;
2260} 2275}
2261 2276
2262/* 2277/*
@@ -2588,11 +2603,6 @@ find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
2588 */ 2603 */
2589#define MAX_PINNED_INTERVAL 512 2604#define MAX_PINNED_INTERVAL 512
2590 2605
2591static inline unsigned long minus_1_or_zero(unsigned long n)
2592{
2593 return n > 0 ? n - 1 : 0;
2594}
2595
2596/* 2606/*
2597 * Check this_cpu to ensure it is balanced within domain. Attempt to move 2607 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2598 * tasks if there is an imbalance. 2608 * tasks if there is an imbalance.
@@ -2601,7 +2611,7 @@ static int load_balance(int this_cpu, struct rq *this_rq,
2601 struct sched_domain *sd, enum cpu_idle_type idle, 2611 struct sched_domain *sd, enum cpu_idle_type idle,
2602 int *balance) 2612 int *balance)
2603{ 2613{
2604 int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; 2614 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
2605 struct sched_group *group; 2615 struct sched_group *group;
2606 unsigned long imbalance; 2616 unsigned long imbalance;
2607 struct rq *busiest; 2617 struct rq *busiest;
@@ -2642,18 +2652,17 @@ redo:
2642 2652
2643 schedstat_add(sd, lb_imbalance[idle], imbalance); 2653 schedstat_add(sd, lb_imbalance[idle], imbalance);
2644 2654
2645 nr_moved = 0; 2655 ld_moved = 0;
2646 if (busiest->nr_running > 1) { 2656 if (busiest->nr_running > 1) {
2647 /* 2657 /*
2648 * Attempt to move tasks. If find_busiest_group has found 2658 * Attempt to move tasks. If find_busiest_group has found
2649 * an imbalance but busiest->nr_running <= 1, the group is 2659 * an imbalance but busiest->nr_running <= 1, the group is
2650 * still unbalanced. nr_moved simply stays zero, so it is 2660 * still unbalanced. ld_moved simply stays zero, so it is
2651 * correctly treated as an imbalance. 2661 * correctly treated as an imbalance.
2652 */ 2662 */
2653 local_irq_save(flags); 2663 local_irq_save(flags);
2654 double_rq_lock(this_rq, busiest); 2664 double_rq_lock(this_rq, busiest);
2655 nr_moved = move_tasks(this_rq, this_cpu, busiest, 2665 ld_moved = move_tasks(this_rq, this_cpu, busiest,
2656 minus_1_or_zero(busiest->nr_running),
2657 imbalance, sd, idle, &all_pinned); 2666 imbalance, sd, idle, &all_pinned);
2658 double_rq_unlock(this_rq, busiest); 2667 double_rq_unlock(this_rq, busiest);
2659 local_irq_restore(flags); 2668 local_irq_restore(flags);
@@ -2661,7 +2670,7 @@ redo:
2661 /* 2670 /*
2662 * some other cpu did the load balance for us. 2671 * some other cpu did the load balance for us.
2663 */ 2672 */
2664 if (nr_moved && this_cpu != smp_processor_id()) 2673 if (ld_moved && this_cpu != smp_processor_id())
2665 resched_cpu(this_cpu); 2674 resched_cpu(this_cpu);
2666 2675
2667 /* All tasks on this runqueue were pinned by CPU affinity */ 2676 /* All tasks on this runqueue were pinned by CPU affinity */
@@ -2673,7 +2682,7 @@ redo:
2673 } 2682 }
2674 } 2683 }
2675 2684
2676 if (!nr_moved) { 2685 if (!ld_moved) {
2677 schedstat_inc(sd, lb_failed[idle]); 2686 schedstat_inc(sd, lb_failed[idle]);
2678 sd->nr_balance_failed++; 2687 sd->nr_balance_failed++;
2679 2688
@@ -2722,10 +2731,10 @@ redo:
2722 sd->balance_interval *= 2; 2731 sd->balance_interval *= 2;
2723 } 2732 }
2724 2733
2725 if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && 2734 if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
2726 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 2735 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
2727 return -1; 2736 return -1;
2728 return nr_moved; 2737 return ld_moved;
2729 2738
2730out_balanced: 2739out_balanced:
2731 schedstat_inc(sd, lb_balanced[idle]); 2740 schedstat_inc(sd, lb_balanced[idle]);
@@ -2757,7 +2766,7 @@ load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd)
2757 struct sched_group *group; 2766 struct sched_group *group;
2758 struct rq *busiest = NULL; 2767 struct rq *busiest = NULL;
2759 unsigned long imbalance; 2768 unsigned long imbalance;
2760 int nr_moved = 0; 2769 int ld_moved = 0;
2761 int sd_idle = 0; 2770 int sd_idle = 0;
2762 int all_pinned = 0; 2771 int all_pinned = 0;
2763 cpumask_t cpus = CPU_MASK_ALL; 2772 cpumask_t cpus = CPU_MASK_ALL;
@@ -2792,12 +2801,13 @@ redo:
2792 2801
2793 schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); 2802 schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance);
2794 2803
2795 nr_moved = 0; 2804 ld_moved = 0;
2796 if (busiest->nr_running > 1) { 2805 if (busiest->nr_running > 1) {
2797 /* Attempt to move tasks */ 2806 /* Attempt to move tasks */
2798 double_lock_balance(this_rq, busiest); 2807 double_lock_balance(this_rq, busiest);
2799 nr_moved = move_tasks(this_rq, this_cpu, busiest, 2808 /* this_rq->clock is already updated */
2800 minus_1_or_zero(busiest->nr_running), 2809 update_rq_clock(busiest);
2810 ld_moved = move_tasks(this_rq, this_cpu, busiest,
2801 imbalance, sd, CPU_NEWLY_IDLE, 2811 imbalance, sd, CPU_NEWLY_IDLE,
2802 &all_pinned); 2812 &all_pinned);
2803 spin_unlock(&busiest->lock); 2813 spin_unlock(&busiest->lock);
@@ -2809,7 +2819,7 @@ redo:
2809 } 2819 }
2810 } 2820 }
2811 2821
2812 if (!nr_moved) { 2822 if (!ld_moved) {
2813 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); 2823 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]);
2814 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && 2824 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
2815 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 2825 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
@@ -2817,7 +2827,7 @@ redo:
2817 } else 2827 } else
2818 sd->nr_balance_failed = 0; 2828 sd->nr_balance_failed = 0;
2819 2829
2820 return nr_moved; 2830 return ld_moved;
2821 2831
2822out_balanced: 2832out_balanced:
2823 schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); 2833 schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]);
@@ -2894,6 +2904,8 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
2894 2904
2895 /* move a task from busiest_rq to target_rq */ 2905 /* move a task from busiest_rq to target_rq */
2896 double_lock_balance(busiest_rq, target_rq); 2906 double_lock_balance(busiest_rq, target_rq);
2907 update_rq_clock(busiest_rq);
2908 update_rq_clock(target_rq);
2897 2909
2898 /* Search for an sd spanning us and the target CPU. */ 2910 /* Search for an sd spanning us and the target CPU. */
2899 for_each_domain(target_cpu, sd) { 2911 for_each_domain(target_cpu, sd) {
@@ -2905,8 +2917,8 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
2905 if (likely(sd)) { 2917 if (likely(sd)) {
2906 schedstat_inc(sd, alb_cnt); 2918 schedstat_inc(sd, alb_cnt);
2907 2919
2908 if (move_tasks(target_rq, target_cpu, busiest_rq, 1, 2920 if (move_one_task(target_rq, target_cpu, busiest_rq,
2909 ULONG_MAX, sd, CPU_IDLE, NULL)) 2921 sd, CPU_IDLE))
2910 schedstat_inc(sd, alb_pushed); 2922 schedstat_inc(sd, alb_pushed);
2911 else 2923 else
2912 schedstat_inc(sd, alb_failed); 2924 schedstat_inc(sd, alb_failed);
@@ -3175,8 +3187,7 @@ static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
3175 unsigned long max_nr_move, unsigned long max_load_move, 3187 unsigned long max_nr_move, unsigned long max_load_move,
3176 struct sched_domain *sd, enum cpu_idle_type idle, 3188 struct sched_domain *sd, enum cpu_idle_type idle,
3177 int *all_pinned, unsigned long *load_moved, 3189 int *all_pinned, unsigned long *load_moved,
3178 int this_best_prio, int best_prio, int best_prio_seen, 3190 int *this_best_prio, struct rq_iterator *iterator)
3179 struct rq_iterator *iterator)
3180{ 3191{
3181 *load_moved = 0; 3192 *load_moved = 0;
3182 3193
@@ -3202,7 +3213,8 @@ unsigned long long task_sched_runtime(struct task_struct *p)
3202 rq = task_rq_lock(p, &flags); 3213 rq = task_rq_lock(p, &flags);
3203 ns = p->se.sum_exec_runtime; 3214 ns = p->se.sum_exec_runtime;
3204 if (rq->curr == p) { 3215 if (rq->curr == p) {
3205 delta_exec = rq_clock(rq) - p->se.exec_start; 3216 update_rq_clock(rq);
3217 delta_exec = rq->clock - p->se.exec_start;
3206 if ((s64)delta_exec > 0) 3218 if ((s64)delta_exec > 0)
3207 ns += delta_exec; 3219 ns += delta_exec;
3208 } 3220 }
@@ -3298,9 +3310,10 @@ void scheduler_tick(void)
3298 struct task_struct *curr = rq->curr; 3310 struct task_struct *curr = rq->curr;
3299 3311
3300 spin_lock(&rq->lock); 3312 spin_lock(&rq->lock);
3313 __update_rq_clock(rq);
3314 update_cpu_load(rq);
3301 if (curr != rq->idle) /* FIXME: needed? */ 3315 if (curr != rq->idle) /* FIXME: needed? */
3302 curr->sched_class->task_tick(rq, curr); 3316 curr->sched_class->task_tick(rq, curr);
3303 update_cpu_load(rq);
3304 spin_unlock(&rq->lock); 3317 spin_unlock(&rq->lock);
3305 3318
3306#ifdef CONFIG_SMP 3319#ifdef CONFIG_SMP
@@ -3382,7 +3395,7 @@ static inline void schedule_debug(struct task_struct *prev)
3382 * Pick up the highest-prio task: 3395 * Pick up the highest-prio task:
3383 */ 3396 */
3384static inline struct task_struct * 3397static inline struct task_struct *
3385pick_next_task(struct rq *rq, struct task_struct *prev, u64 now) 3398pick_next_task(struct rq *rq, struct task_struct *prev)
3386{ 3399{
3387 struct sched_class *class; 3400 struct sched_class *class;
3388 struct task_struct *p; 3401 struct task_struct *p;
@@ -3392,14 +3405,14 @@ pick_next_task(struct rq *rq, struct task_struct *prev, u64 now)
3392 * the fair class we can call that function directly: 3405 * the fair class we can call that function directly:
3393 */ 3406 */
3394 if (likely(rq->nr_running == rq->cfs.nr_running)) { 3407 if (likely(rq->nr_running == rq->cfs.nr_running)) {
3395 p = fair_sched_class.pick_next_task(rq, now); 3408 p = fair_sched_class.pick_next_task(rq);
3396 if (likely(p)) 3409 if (likely(p))
3397 return p; 3410 return p;
3398 } 3411 }
3399 3412
3400 class = sched_class_highest; 3413 class = sched_class_highest;
3401 for ( ; ; ) { 3414 for ( ; ; ) {
3402 p = class->pick_next_task(rq, now); 3415 p = class->pick_next_task(rq);
3403 if (p) 3416 if (p)
3404 return p; 3417 return p;
3405 /* 3418 /*
@@ -3418,7 +3431,6 @@ asmlinkage void __sched schedule(void)
3418 struct task_struct *prev, *next; 3431 struct task_struct *prev, *next;
3419 long *switch_count; 3432 long *switch_count;
3420 struct rq *rq; 3433 struct rq *rq;
3421 u64 now;
3422 int cpu; 3434 int cpu;
3423 3435
3424need_resched: 3436need_resched:
@@ -3436,6 +3448,7 @@ need_resched_nonpreemptible:
3436 3448
3437 spin_lock_irq(&rq->lock); 3449 spin_lock_irq(&rq->lock);
3438 clear_tsk_need_resched(prev); 3450 clear_tsk_need_resched(prev);
3451 __update_rq_clock(rq);
3439 3452
3440 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { 3453 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
3441 if (unlikely((prev->state & TASK_INTERRUPTIBLE) && 3454 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
@@ -3450,9 +3463,8 @@ need_resched_nonpreemptible:
3450 if (unlikely(!rq->nr_running)) 3463 if (unlikely(!rq->nr_running))
3451 idle_balance(cpu, rq); 3464 idle_balance(cpu, rq);
3452 3465
3453 now = __rq_clock(rq); 3466 prev->sched_class->put_prev_task(rq, prev);
3454 prev->sched_class->put_prev_task(rq, prev, now); 3467 next = pick_next_task(rq, prev);
3455 next = pick_next_task(rq, prev, now);
3456 3468
3457 sched_info_switch(prev, next); 3469 sched_info_switch(prev, next);
3458 3470
@@ -3895,17 +3907,16 @@ void rt_mutex_setprio(struct task_struct *p, int prio)
3895 unsigned long flags; 3907 unsigned long flags;
3896 int oldprio, on_rq; 3908 int oldprio, on_rq;
3897 struct rq *rq; 3909 struct rq *rq;
3898 u64 now;
3899 3910
3900 BUG_ON(prio < 0 || prio > MAX_PRIO); 3911 BUG_ON(prio < 0 || prio > MAX_PRIO);
3901 3912
3902 rq = task_rq_lock(p, &flags); 3913 rq = task_rq_lock(p, &flags);
3903 now = rq_clock(rq); 3914 update_rq_clock(rq);
3904 3915
3905 oldprio = p->prio; 3916 oldprio = p->prio;
3906 on_rq = p->se.on_rq; 3917 on_rq = p->se.on_rq;
3907 if (on_rq) 3918 if (on_rq)
3908 dequeue_task(rq, p, 0, now); 3919 dequeue_task(rq, p, 0);
3909 3920
3910 if (rt_prio(prio)) 3921 if (rt_prio(prio))
3911 p->sched_class = &rt_sched_class; 3922 p->sched_class = &rt_sched_class;
@@ -3915,7 +3926,7 @@ void rt_mutex_setprio(struct task_struct *p, int prio)
3915 p->prio = prio; 3926 p->prio = prio;
3916 3927
3917 if (on_rq) { 3928 if (on_rq) {
3918 enqueue_task(rq, p, 0, now); 3929 enqueue_task(rq, p, 0);
3919 /* 3930 /*
3920 * Reschedule if we are currently running on this runqueue and 3931 * Reschedule if we are currently running on this runqueue and
3921 * our priority decreased, or if we are not currently running on 3932 * our priority decreased, or if we are not currently running on
@@ -3938,7 +3949,6 @@ void set_user_nice(struct task_struct *p, long nice)
3938 int old_prio, delta, on_rq; 3949 int old_prio, delta, on_rq;
3939 unsigned long flags; 3950 unsigned long flags;
3940 struct rq *rq; 3951 struct rq *rq;
3941 u64 now;
3942 3952
3943 if (TASK_NICE(p) == nice || nice < -20 || nice > 19) 3953 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
3944 return; 3954 return;
@@ -3947,7 +3957,7 @@ void set_user_nice(struct task_struct *p, long nice)
3947 * the task might be in the middle of scheduling on another CPU. 3957 * the task might be in the middle of scheduling on another CPU.
3948 */ 3958 */
3949 rq = task_rq_lock(p, &flags); 3959 rq = task_rq_lock(p, &flags);
3950 now = rq_clock(rq); 3960 update_rq_clock(rq);
3951 /* 3961 /*
3952 * The RT priorities are set via sched_setscheduler(), but we still 3962 * The RT priorities are set via sched_setscheduler(), but we still
3953 * allow the 'normal' nice value to be set - but as expected 3963 * allow the 'normal' nice value to be set - but as expected
@@ -3960,8 +3970,8 @@ void set_user_nice(struct task_struct *p, long nice)
3960 } 3970 }
3961 on_rq = p->se.on_rq; 3971 on_rq = p->se.on_rq;
3962 if (on_rq) { 3972 if (on_rq) {
3963 dequeue_task(rq, p, 0, now); 3973 dequeue_task(rq, p, 0);
3964 dec_load(rq, p, now); 3974 dec_load(rq, p);
3965 } 3975 }
3966 3976
3967 p->static_prio = NICE_TO_PRIO(nice); 3977 p->static_prio = NICE_TO_PRIO(nice);
@@ -3971,8 +3981,8 @@ void set_user_nice(struct task_struct *p, long nice)
3971 delta = p->prio - old_prio; 3981 delta = p->prio - old_prio;
3972 3982
3973 if (on_rq) { 3983 if (on_rq) {
3974 enqueue_task(rq, p, 0, now); 3984 enqueue_task(rq, p, 0);
3975 inc_load(rq, p, now); 3985 inc_load(rq, p);
3976 /* 3986 /*
3977 * If the task increased its priority or is running and 3987 * If the task increased its priority or is running and
3978 * lowered its priority, then reschedule its CPU: 3988 * lowered its priority, then reschedule its CPU:
@@ -4208,6 +4218,7 @@ recheck:
4208 spin_unlock_irqrestore(&p->pi_lock, flags); 4218 spin_unlock_irqrestore(&p->pi_lock, flags);
4209 goto recheck; 4219 goto recheck;
4210 } 4220 }
4221 update_rq_clock(rq);
4211 on_rq = p->se.on_rq; 4222 on_rq = p->se.on_rq;
4212 if (on_rq) 4223 if (on_rq)
4213 deactivate_task(rq, p, 0); 4224 deactivate_task(rq, p, 0);
@@ -4463,10 +4474,8 @@ long sched_getaffinity(pid_t pid, cpumask_t *mask)
4463out_unlock: 4474out_unlock:
4464 read_unlock(&tasklist_lock); 4475 read_unlock(&tasklist_lock);
4465 mutex_unlock(&sched_hotcpu_mutex); 4476 mutex_unlock(&sched_hotcpu_mutex);
4466 if (retval)
4467 return retval;
4468 4477
4469 return 0; 4478 return retval;
4470} 4479}
4471 4480
4472/** 4481/**
@@ -4966,6 +4975,7 @@ static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
4966 on_rq = p->se.on_rq; 4975 on_rq = p->se.on_rq;
4967 if (on_rq) 4976 if (on_rq)
4968 deactivate_task(rq_src, p, 0); 4977 deactivate_task(rq_src, p, 0);
4978
4969 set_task_cpu(p, dest_cpu); 4979 set_task_cpu(p, dest_cpu);
4970 if (on_rq) { 4980 if (on_rq) {
4971 activate_task(rq_dest, p, 0); 4981 activate_task(rq_dest, p, 0);
@@ -5198,7 +5208,8 @@ static void migrate_dead_tasks(unsigned int dead_cpu)
5198 for ( ; ; ) { 5208 for ( ; ; ) {
5199 if (!rq->nr_running) 5209 if (!rq->nr_running)
5200 break; 5210 break;
5201 next = pick_next_task(rq, rq->curr, rq_clock(rq)); 5211 update_rq_clock(rq);
5212 next = pick_next_task(rq, rq->curr);
5202 if (!next) 5213 if (!next)
5203 break; 5214 break;
5204 migrate_dead(dead_cpu, next); 5215 migrate_dead(dead_cpu, next);
@@ -5210,12 +5221,19 @@ static void migrate_dead_tasks(unsigned int dead_cpu)
5210#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 5221#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5211 5222
5212static struct ctl_table sd_ctl_dir[] = { 5223static struct ctl_table sd_ctl_dir[] = {
5213 {CTL_UNNUMBERED, "sched_domain", NULL, 0, 0755, NULL, }, 5224 {
5225 .procname = "sched_domain",
5226 .mode = 0755,
5227 },
5214 {0,}, 5228 {0,},
5215}; 5229};
5216 5230
5217static struct ctl_table sd_ctl_root[] = { 5231static struct ctl_table sd_ctl_root[] = {
5218 {CTL_UNNUMBERED, "kernel", NULL, 0, 0755, sd_ctl_dir, }, 5232 {
5233 .procname = "kernel",
5234 .mode = 0755,
5235 .child = sd_ctl_dir,
5236 },
5219 {0,}, 5237 {0,},
5220}; 5238};
5221 5239
@@ -5231,11 +5249,10 @@ static struct ctl_table *sd_alloc_ctl_entry(int n)
5231} 5249}
5232 5250
5233static void 5251static void
5234set_table_entry(struct ctl_table *entry, int ctl_name, 5252set_table_entry(struct ctl_table *entry,
5235 const char *procname, void *data, int maxlen, 5253 const char *procname, void *data, int maxlen,
5236 mode_t mode, proc_handler *proc_handler) 5254 mode_t mode, proc_handler *proc_handler)
5237{ 5255{
5238 entry->ctl_name = ctl_name;
5239 entry->procname = procname; 5256 entry->procname = procname;
5240 entry->data = data; 5257 entry->data = data;
5241 entry->maxlen = maxlen; 5258 entry->maxlen = maxlen;
@@ -5248,28 +5265,28 @@ sd_alloc_ctl_domain_table(struct sched_domain *sd)
5248{ 5265{
5249 struct ctl_table *table = sd_alloc_ctl_entry(14); 5266 struct ctl_table *table = sd_alloc_ctl_entry(14);
5250 5267
5251 set_table_entry(&table[0], 1, "min_interval", &sd->min_interval, 5268 set_table_entry(&table[0], "min_interval", &sd->min_interval,
5252 sizeof(long), 0644, proc_doulongvec_minmax); 5269 sizeof(long), 0644, proc_doulongvec_minmax);
5253 set_table_entry(&table[1], 2, "max_interval", &sd->max_interval, 5270 set_table_entry(&table[1], "max_interval", &sd->max_interval,
5254 sizeof(long), 0644, proc_doulongvec_minmax); 5271 sizeof(long), 0644, proc_doulongvec_minmax);
5255 set_table_entry(&table[2], 3, "busy_idx", &sd->busy_idx, 5272 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
5256 sizeof(int), 0644, proc_dointvec_minmax); 5273 sizeof(int), 0644, proc_dointvec_minmax);
5257 set_table_entry(&table[3], 4, "idle_idx", &sd->idle_idx, 5274 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
5258 sizeof(int), 0644, proc_dointvec_minmax); 5275 sizeof(int), 0644, proc_dointvec_minmax);
5259 set_table_entry(&table[4], 5, "newidle_idx", &sd->newidle_idx, 5276 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
5260 sizeof(int), 0644, proc_dointvec_minmax); 5277 sizeof(int), 0644, proc_dointvec_minmax);
5261 set_table_entry(&table[5], 6, "wake_idx", &sd->wake_idx, 5278 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
5262 sizeof(int), 0644, proc_dointvec_minmax); 5279 sizeof(int), 0644, proc_dointvec_minmax);
5263 set_table_entry(&table[6], 7, "forkexec_idx", &sd->forkexec_idx, 5280 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
5264 sizeof(int), 0644, proc_dointvec_minmax); 5281 sizeof(int), 0644, proc_dointvec_minmax);
5265 set_table_entry(&table[7], 8, "busy_factor", &sd->busy_factor, 5282 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
5266 sizeof(int), 0644, proc_dointvec_minmax); 5283 sizeof(int), 0644, proc_dointvec_minmax);
5267 set_table_entry(&table[8], 9, "imbalance_pct", &sd->imbalance_pct, 5284 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
5268 sizeof(int), 0644, proc_dointvec_minmax); 5285 sizeof(int), 0644, proc_dointvec_minmax);
5269 set_table_entry(&table[10], 11, "cache_nice_tries", 5286 set_table_entry(&table[10], "cache_nice_tries",
5270 &sd->cache_nice_tries, 5287 &sd->cache_nice_tries,
5271 sizeof(int), 0644, proc_dointvec_minmax); 5288 sizeof(int), 0644, proc_dointvec_minmax);
5272 set_table_entry(&table[12], 13, "flags", &sd->flags, 5289 set_table_entry(&table[12], "flags", &sd->flags,
5273 sizeof(int), 0644, proc_dointvec_minmax); 5290 sizeof(int), 0644, proc_dointvec_minmax);
5274 5291
5275 return table; 5292 return table;
@@ -5289,7 +5306,6 @@ static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
5289 i = 0; 5306 i = 0;
5290 for_each_domain(cpu, sd) { 5307 for_each_domain(cpu, sd) {
5291 snprintf(buf, 32, "domain%d", i); 5308 snprintf(buf, 32, "domain%d", i);
5292 entry->ctl_name = i + 1;
5293 entry->procname = kstrdup(buf, GFP_KERNEL); 5309 entry->procname = kstrdup(buf, GFP_KERNEL);
5294 entry->mode = 0755; 5310 entry->mode = 0755;
5295 entry->child = sd_alloc_ctl_domain_table(sd); 5311 entry->child = sd_alloc_ctl_domain_table(sd);
@@ -5310,7 +5326,6 @@ static void init_sched_domain_sysctl(void)
5310 5326
5311 for (i = 0; i < cpu_num; i++, entry++) { 5327 for (i = 0; i < cpu_num; i++, entry++) {
5312 snprintf(buf, 32, "cpu%d", i); 5328 snprintf(buf, 32, "cpu%d", i);
5313 entry->ctl_name = i + 1;
5314 entry->procname = kstrdup(buf, GFP_KERNEL); 5329 entry->procname = kstrdup(buf, GFP_KERNEL);
5315 entry->mode = 0755; 5330 entry->mode = 0755;
5316 entry->child = sd_alloc_ctl_cpu_table(i); 5331 entry->child = sd_alloc_ctl_cpu_table(i);
@@ -5379,6 +5394,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
5379 rq->migration_thread = NULL; 5394 rq->migration_thread = NULL;
5380 /* Idle task back to normal (off runqueue, low prio) */ 5395 /* Idle task back to normal (off runqueue, low prio) */
5381 rq = task_rq_lock(rq->idle, &flags); 5396 rq = task_rq_lock(rq->idle, &flags);
5397 update_rq_clock(rq);
5382 deactivate_task(rq, rq->idle, 0); 5398 deactivate_task(rq, rq->idle, 0);
5383 rq->idle->static_prio = MAX_PRIO; 5399 rq->idle->static_prio = MAX_PRIO;
5384 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); 5400 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
@@ -6616,12 +6632,13 @@ void normalize_rt_tasks(void)
6616 goto out_unlock; 6632 goto out_unlock;
6617#endif 6633#endif
6618 6634
6635 update_rq_clock(rq);
6619 on_rq = p->se.on_rq; 6636 on_rq = p->se.on_rq;
6620 if (on_rq) 6637 if (on_rq)
6621 deactivate_task(task_rq(p), p, 0); 6638 deactivate_task(rq, p, 0);
6622 __setscheduler(rq, p, SCHED_NORMAL, 0); 6639 __setscheduler(rq, p, SCHED_NORMAL, 0);
6623 if (on_rq) { 6640 if (on_rq) {
6624 activate_task(task_rq(p), p, 0); 6641 activate_task(rq, p, 0);
6625 resched_task(rq->curr); 6642 resched_task(rq->curr);
6626 } 6643 }
6627#ifdef CONFIG_SMP 6644#ifdef CONFIG_SMP
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
index 8421b9399e10..3da32156394e 100644
--- a/kernel/sched_debug.c
+++ b/kernel/sched_debug.c
@@ -29,7 +29,7 @@
29 } while (0) 29 } while (0)
30 30
31static void 31static void
32print_task(struct seq_file *m, struct rq *rq, struct task_struct *p, u64 now) 32print_task(struct seq_file *m, struct rq *rq, struct task_struct *p)
33{ 33{
34 if (rq->curr == p) 34 if (rq->curr == p)
35 SEQ_printf(m, "R"); 35 SEQ_printf(m, "R");
@@ -56,7 +56,7 @@ print_task(struct seq_file *m, struct rq *rq, struct task_struct *p, u64 now)
56#endif 56#endif
57} 57}
58 58
59static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu, u64 now) 59static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu)
60{ 60{
61 struct task_struct *g, *p; 61 struct task_struct *g, *p;
62 62
@@ -77,7 +77,7 @@ static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu, u64 now)
77 if (!p->se.on_rq || task_cpu(p) != rq_cpu) 77 if (!p->se.on_rq || task_cpu(p) != rq_cpu)
78 continue; 78 continue;
79 79
80 print_task(m, rq, p, now); 80 print_task(m, rq, p);
81 } while_each_thread(g, p); 81 } while_each_thread(g, p);
82 82
83 read_unlock_irq(&tasklist_lock); 83 read_unlock_irq(&tasklist_lock);
@@ -106,7 +106,7 @@ print_cfs_rq_runtime_sum(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
106 (long long)wait_runtime_rq_sum); 106 (long long)wait_runtime_rq_sum);
107} 107}
108 108
109void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq, u64 now) 109void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
110{ 110{
111 SEQ_printf(m, "\ncfs_rq %p\n", cfs_rq); 111 SEQ_printf(m, "\ncfs_rq %p\n", cfs_rq);
112 112
@@ -124,7 +124,7 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq, u64 now)
124 print_cfs_rq_runtime_sum(m, cpu, cfs_rq); 124 print_cfs_rq_runtime_sum(m, cpu, cfs_rq);
125} 125}
126 126
127static void print_cpu(struct seq_file *m, int cpu, u64 now) 127static void print_cpu(struct seq_file *m, int cpu)
128{ 128{
129 struct rq *rq = &per_cpu(runqueues, cpu); 129 struct rq *rq = &per_cpu(runqueues, cpu);
130 130
@@ -166,9 +166,9 @@ static void print_cpu(struct seq_file *m, int cpu, u64 now)
166 P(cpu_load[4]); 166 P(cpu_load[4]);
167#undef P 167#undef P
168 168
169 print_cfs_stats(m, cpu, now); 169 print_cfs_stats(m, cpu);
170 170
171 print_rq(m, rq, cpu, now); 171 print_rq(m, rq, cpu);
172} 172}
173 173
174static int sched_debug_show(struct seq_file *m, void *v) 174static int sched_debug_show(struct seq_file *m, void *v)
@@ -184,7 +184,7 @@ static int sched_debug_show(struct seq_file *m, void *v)
184 SEQ_printf(m, "now at %Lu nsecs\n", (unsigned long long)now); 184 SEQ_printf(m, "now at %Lu nsecs\n", (unsigned long long)now);
185 185
186 for_each_online_cpu(cpu) 186 for_each_online_cpu(cpu)
187 print_cpu(m, cpu, now); 187 print_cpu(m, cpu);
188 188
189 SEQ_printf(m, "\n"); 189 SEQ_printf(m, "\n");
190 190
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index 6f579ff5a9bc..e91db32cadfd 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -222,21 +222,25 @@ niced_granularity(struct sched_entity *curr, unsigned long granularity)
222{ 222{
223 u64 tmp; 223 u64 tmp;
224 224
225 if (likely(curr->load.weight == NICE_0_LOAD))
226 return granularity;
225 /* 227 /*
226 * Negative nice levels get the same granularity as nice-0: 228 * Positive nice levels get the same granularity as nice-0:
227 */ 229 */
228 if (likely(curr->load.weight >= NICE_0_LOAD)) 230 if (likely(curr->load.weight < NICE_0_LOAD)) {
229 return granularity; 231 tmp = curr->load.weight * (u64)granularity;
232 return (long) (tmp >> NICE_0_SHIFT);
233 }
230 /* 234 /*
231 * Positive nice level tasks get linearly finer 235 * Negative nice level tasks get linearly finer
232 * granularity: 236 * granularity:
233 */ 237 */
234 tmp = curr->load.weight * (u64)granularity; 238 tmp = curr->load.inv_weight * (u64)granularity;
235 239
236 /* 240 /*
237 * It will always fit into 'long': 241 * It will always fit into 'long':
238 */ 242 */
239 return (long) (tmp >> NICE_0_SHIFT); 243 return (long) (tmp >> WMULT_SHIFT);
240} 244}
241 245
242static inline void 246static inline void
@@ -281,26 +285,25 @@ add_wait_runtime(struct cfs_rq *cfs_rq, struct sched_entity *se, long delta)
281 * are not in our scheduling class. 285 * are not in our scheduling class.
282 */ 286 */
283static inline void 287static inline void
284__update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, u64 now) 288__update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr)
285{ 289{
286 unsigned long delta, delta_exec, delta_fair; 290 unsigned long delta, delta_exec, delta_fair, delta_mine;
287 long delta_mine;
288 struct load_weight *lw = &cfs_rq->load; 291 struct load_weight *lw = &cfs_rq->load;
289 unsigned long load = lw->weight; 292 unsigned long load = lw->weight;
290 293
291 if (unlikely(!load))
292 return;
293
294 delta_exec = curr->delta_exec; 294 delta_exec = curr->delta_exec;
295 schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max)); 295 schedstat_set(curr->exec_max, max((u64)delta_exec, curr->exec_max));
296 296
297 curr->sum_exec_runtime += delta_exec; 297 curr->sum_exec_runtime += delta_exec;
298 cfs_rq->exec_clock += delta_exec; 298 cfs_rq->exec_clock += delta_exec;
299 299
300 if (unlikely(!load))
301 return;
302
300 delta_fair = calc_delta_fair(delta_exec, lw); 303 delta_fair = calc_delta_fair(delta_exec, lw);
301 delta_mine = calc_delta_mine(delta_exec, curr->load.weight, lw); 304 delta_mine = calc_delta_mine(delta_exec, curr->load.weight, lw);
302 305
303 if (cfs_rq->sleeper_bonus > sysctl_sched_stat_granularity) { 306 if (cfs_rq->sleeper_bonus > sysctl_sched_granularity) {
304 delta = calc_delta_mine(cfs_rq->sleeper_bonus, 307 delta = calc_delta_mine(cfs_rq->sleeper_bonus,
305 curr->load.weight, lw); 308 curr->load.weight, lw);
306 if (unlikely(delta > cfs_rq->sleeper_bonus)) 309 if (unlikely(delta > cfs_rq->sleeper_bonus))
@@ -321,7 +324,7 @@ __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr, u64 now)
321 add_wait_runtime(cfs_rq, curr, delta_mine - delta_exec); 324 add_wait_runtime(cfs_rq, curr, delta_mine - delta_exec);
322} 325}
323 326
324static void update_curr(struct cfs_rq *cfs_rq, u64 now) 327static void update_curr(struct cfs_rq *cfs_rq)
325{ 328{
326 struct sched_entity *curr = cfs_rq_curr(cfs_rq); 329 struct sched_entity *curr = cfs_rq_curr(cfs_rq);
327 unsigned long delta_exec; 330 unsigned long delta_exec;
@@ -334,22 +337,22 @@ static void update_curr(struct cfs_rq *cfs_rq, u64 now)
334 * since the last time we changed load (this cannot 337 * since the last time we changed load (this cannot
335 * overflow on 32 bits): 338 * overflow on 32 bits):
336 */ 339 */
337 delta_exec = (unsigned long)(now - curr->exec_start); 340 delta_exec = (unsigned long)(rq_of(cfs_rq)->clock - curr->exec_start);
338 341
339 curr->delta_exec += delta_exec; 342 curr->delta_exec += delta_exec;
340 343
341 if (unlikely(curr->delta_exec > sysctl_sched_stat_granularity)) { 344 if (unlikely(curr->delta_exec > sysctl_sched_stat_granularity)) {
342 __update_curr(cfs_rq, curr, now); 345 __update_curr(cfs_rq, curr);
343 curr->delta_exec = 0; 346 curr->delta_exec = 0;
344 } 347 }
345 curr->exec_start = now; 348 curr->exec_start = rq_of(cfs_rq)->clock;
346} 349}
347 350
348static inline void 351static inline void
349update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 352update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
350{ 353{
351 se->wait_start_fair = cfs_rq->fair_clock; 354 se->wait_start_fair = cfs_rq->fair_clock;
352 schedstat_set(se->wait_start, now); 355 schedstat_set(se->wait_start, rq_of(cfs_rq)->clock);
353} 356}
354 357
355/* 358/*
@@ -377,8 +380,7 @@ calc_weighted(unsigned long delta, unsigned long weight, int shift)
377/* 380/*
378 * Task is being enqueued - update stats: 381 * Task is being enqueued - update stats:
379 */ 382 */
380static void 383static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
381update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
382{ 384{
383 s64 key; 385 s64 key;
384 386
@@ -387,7 +389,7 @@ update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
387 * a dequeue/enqueue event is a NOP) 389 * a dequeue/enqueue event is a NOP)
388 */ 390 */
389 if (se != cfs_rq_curr(cfs_rq)) 391 if (se != cfs_rq_curr(cfs_rq))
390 update_stats_wait_start(cfs_rq, se, now); 392 update_stats_wait_start(cfs_rq, se);
391 /* 393 /*
392 * Update the key: 394 * Update the key:
393 */ 395 */
@@ -407,7 +409,8 @@ update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
407 (WMULT_SHIFT - NICE_0_SHIFT); 409 (WMULT_SHIFT - NICE_0_SHIFT);
408 } else { 410 } else {
409 tmp = se->wait_runtime; 411 tmp = se->wait_runtime;
410 key -= (tmp * se->load.weight) >> NICE_0_SHIFT; 412 key -= (tmp * se->load.inv_weight) >>
413 (WMULT_SHIFT - NICE_0_SHIFT);
411 } 414 }
412 } 415 }
413 416
@@ -418,11 +421,12 @@ update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
418 * Note: must be called with a freshly updated rq->fair_clock. 421 * Note: must be called with a freshly updated rq->fair_clock.
419 */ 422 */
420static inline void 423static inline void
421__update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 424__update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
422{ 425{
423 unsigned long delta_fair = se->delta_fair_run; 426 unsigned long delta_fair = se->delta_fair_run;
424 427
425 schedstat_set(se->wait_max, max(se->wait_max, now - se->wait_start)); 428 schedstat_set(se->wait_max, max(se->wait_max,
429 rq_of(cfs_rq)->clock - se->wait_start));
426 430
427 if (unlikely(se->load.weight != NICE_0_LOAD)) 431 if (unlikely(se->load.weight != NICE_0_LOAD))
428 delta_fair = calc_weighted(delta_fair, se->load.weight, 432 delta_fair = calc_weighted(delta_fair, se->load.weight,
@@ -432,7 +436,7 @@ __update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
432} 436}
433 437
434static void 438static void
435update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 439update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
436{ 440{
437 unsigned long delta_fair; 441 unsigned long delta_fair;
438 442
@@ -442,7 +446,7 @@ update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
442 se->delta_fair_run += delta_fair; 446 se->delta_fair_run += delta_fair;
443 if (unlikely(abs(se->delta_fair_run) >= 447 if (unlikely(abs(se->delta_fair_run) >=
444 sysctl_sched_stat_granularity)) { 448 sysctl_sched_stat_granularity)) {
445 __update_stats_wait_end(cfs_rq, se, now); 449 __update_stats_wait_end(cfs_rq, se);
446 se->delta_fair_run = 0; 450 se->delta_fair_run = 0;
447 } 451 }
448 452
@@ -451,34 +455,34 @@ update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
451} 455}
452 456
453static inline void 457static inline void
454update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 458update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
455{ 459{
456 update_curr(cfs_rq, now); 460 update_curr(cfs_rq);
457 /* 461 /*
458 * Mark the end of the wait period if dequeueing a 462 * Mark the end of the wait period if dequeueing a
459 * waiting task: 463 * waiting task:
460 */ 464 */
461 if (se != cfs_rq_curr(cfs_rq)) 465 if (se != cfs_rq_curr(cfs_rq))
462 update_stats_wait_end(cfs_rq, se, now); 466 update_stats_wait_end(cfs_rq, se);
463} 467}
464 468
465/* 469/*
466 * We are picking a new current task - update its stats: 470 * We are picking a new current task - update its stats:
467 */ 471 */
468static inline void 472static inline void
469update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 473update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
470{ 474{
471 /* 475 /*
472 * We are starting a new run period: 476 * We are starting a new run period:
473 */ 477 */
474 se->exec_start = now; 478 se->exec_start = rq_of(cfs_rq)->clock;
475} 479}
476 480
477/* 481/*
478 * We are descheduling a task - update its stats: 482 * We are descheduling a task - update its stats:
479 */ 483 */
480static inline void 484static inline void
481update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 485update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
482{ 486{
483 se->exec_start = 0; 487 se->exec_start = 0;
484} 488}
@@ -487,8 +491,7 @@ update_stats_curr_end(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
487 * Scheduling class queueing methods: 491 * Scheduling class queueing methods:
488 */ 492 */
489 493
490static void 494static void __enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
491__enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
492{ 495{
493 unsigned long load = cfs_rq->load.weight, delta_fair; 496 unsigned long load = cfs_rq->load.weight, delta_fair;
494 long prev_runtime; 497 long prev_runtime;
@@ -522,8 +525,7 @@ __enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
522 schedstat_add(cfs_rq, wait_runtime, se->wait_runtime); 525 schedstat_add(cfs_rq, wait_runtime, se->wait_runtime);
523} 526}
524 527
525static void 528static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
526enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
527{ 529{
528 struct task_struct *tsk = task_of(se); 530 struct task_struct *tsk = task_of(se);
529 unsigned long delta_fair; 531 unsigned long delta_fair;
@@ -538,7 +540,7 @@ enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
538 se->delta_fair_sleep += delta_fair; 540 se->delta_fair_sleep += delta_fair;
539 if (unlikely(abs(se->delta_fair_sleep) >= 541 if (unlikely(abs(se->delta_fair_sleep) >=
540 sysctl_sched_stat_granularity)) { 542 sysctl_sched_stat_granularity)) {
541 __enqueue_sleeper(cfs_rq, se, now); 543 __enqueue_sleeper(cfs_rq, se);
542 se->delta_fair_sleep = 0; 544 se->delta_fair_sleep = 0;
543 } 545 }
544 546
@@ -546,7 +548,7 @@ enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
546 548
547#ifdef CONFIG_SCHEDSTATS 549#ifdef CONFIG_SCHEDSTATS
548 if (se->sleep_start) { 550 if (se->sleep_start) {
549 u64 delta = now - se->sleep_start; 551 u64 delta = rq_of(cfs_rq)->clock - se->sleep_start;
550 552
551 if ((s64)delta < 0) 553 if ((s64)delta < 0)
552 delta = 0; 554 delta = 0;
@@ -558,7 +560,7 @@ enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
558 se->sum_sleep_runtime += delta; 560 se->sum_sleep_runtime += delta;
559 } 561 }
560 if (se->block_start) { 562 if (se->block_start) {
561 u64 delta = now - se->block_start; 563 u64 delta = rq_of(cfs_rq)->clock - se->block_start;
562 564
563 if ((s64)delta < 0) 565 if ((s64)delta < 0)
564 delta = 0; 566 delta = 0;
@@ -573,26 +575,24 @@ enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
573} 575}
574 576
575static void 577static void
576enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, 578enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int wakeup)
577 int wakeup, u64 now)
578{ 579{
579 /* 580 /*
580 * Update the fair clock. 581 * Update the fair clock.
581 */ 582 */
582 update_curr(cfs_rq, now); 583 update_curr(cfs_rq);
583 584
584 if (wakeup) 585 if (wakeup)
585 enqueue_sleeper(cfs_rq, se, now); 586 enqueue_sleeper(cfs_rq, se);
586 587
587 update_stats_enqueue(cfs_rq, se, now); 588 update_stats_enqueue(cfs_rq, se);
588 __enqueue_entity(cfs_rq, se); 589 __enqueue_entity(cfs_rq, se);
589} 590}
590 591
591static void 592static void
592dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, 593dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep)
593 int sleep, u64 now)
594{ 594{
595 update_stats_dequeue(cfs_rq, se, now); 595 update_stats_dequeue(cfs_rq, se);
596 if (sleep) { 596 if (sleep) {
597 se->sleep_start_fair = cfs_rq->fair_clock; 597 se->sleep_start_fair = cfs_rq->fair_clock;
598#ifdef CONFIG_SCHEDSTATS 598#ifdef CONFIG_SCHEDSTATS
@@ -600,9 +600,9 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
600 struct task_struct *tsk = task_of(se); 600 struct task_struct *tsk = task_of(se);
601 601
602 if (tsk->state & TASK_INTERRUPTIBLE) 602 if (tsk->state & TASK_INTERRUPTIBLE)
603 se->sleep_start = now; 603 se->sleep_start = rq_of(cfs_rq)->clock;
604 if (tsk->state & TASK_UNINTERRUPTIBLE) 604 if (tsk->state & TASK_UNINTERRUPTIBLE)
605 se->block_start = now; 605 se->block_start = rq_of(cfs_rq)->clock;
606 } 606 }
607 cfs_rq->wait_runtime -= se->wait_runtime; 607 cfs_rq->wait_runtime -= se->wait_runtime;
608#endif 608#endif
@@ -629,7 +629,7 @@ __check_preempt_curr_fair(struct cfs_rq *cfs_rq, struct sched_entity *se,
629} 629}
630 630
631static inline void 631static inline void
632set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now) 632set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
633{ 633{
634 /* 634 /*
635 * Any task has to be enqueued before it get to execute on 635 * Any task has to be enqueued before it get to execute on
@@ -638,49 +638,46 @@ set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, u64 now)
638 * done a put_prev_task_fair() shortly before this, which 638 * done a put_prev_task_fair() shortly before this, which
639 * updated rq->fair_clock - used by update_stats_wait_end()) 639 * updated rq->fair_clock - used by update_stats_wait_end())
640 */ 640 */
641 update_stats_wait_end(cfs_rq, se, now); 641 update_stats_wait_end(cfs_rq, se);
642 update_stats_curr_start(cfs_rq, se, now); 642 update_stats_curr_start(cfs_rq, se);
643 set_cfs_rq_curr(cfs_rq, se); 643 set_cfs_rq_curr(cfs_rq, se);
644} 644}
645 645
646static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq, u64 now) 646static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
647{ 647{
648 struct sched_entity *se = __pick_next_entity(cfs_rq); 648 struct sched_entity *se = __pick_next_entity(cfs_rq);
649 649
650 set_next_entity(cfs_rq, se, now); 650 set_next_entity(cfs_rq, se);
651 651
652 return se; 652 return se;
653} 653}
654 654
655static void 655static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
656put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev, u64 now)
657{ 656{
658 /* 657 /*
659 * If still on the runqueue then deactivate_task() 658 * If still on the runqueue then deactivate_task()
660 * was not called and update_curr() has to be done: 659 * was not called and update_curr() has to be done:
661 */ 660 */
662 if (prev->on_rq) 661 if (prev->on_rq)
663 update_curr(cfs_rq, now); 662 update_curr(cfs_rq);
664 663
665 update_stats_curr_end(cfs_rq, prev, now); 664 update_stats_curr_end(cfs_rq, prev);
666 665
667 if (prev->on_rq) 666 if (prev->on_rq)
668 update_stats_wait_start(cfs_rq, prev, now); 667 update_stats_wait_start(cfs_rq, prev);
669 set_cfs_rq_curr(cfs_rq, NULL); 668 set_cfs_rq_curr(cfs_rq, NULL);
670} 669}
671 670
672static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr) 671static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
673{ 672{
674 struct rq *rq = rq_of(cfs_rq);
675 struct sched_entity *next; 673 struct sched_entity *next;
676 u64 now = __rq_clock(rq);
677 674
678 /* 675 /*
679 * Dequeue and enqueue the task to update its 676 * Dequeue and enqueue the task to update its
680 * position within the tree: 677 * position within the tree:
681 */ 678 */
682 dequeue_entity(cfs_rq, curr, 0, now); 679 dequeue_entity(cfs_rq, curr, 0);
683 enqueue_entity(cfs_rq, curr, 0, now); 680 enqueue_entity(cfs_rq, curr, 0);
684 681
685 /* 682 /*
686 * Reschedule if another task tops the current one. 683 * Reschedule if another task tops the current one.
@@ -785,8 +782,7 @@ static inline int is_same_group(struct task_struct *curr, struct task_struct *p)
785 * increased. Here we update the fair scheduling stats and 782 * increased. Here we update the fair scheduling stats and
786 * then put the task into the rbtree: 783 * then put the task into the rbtree:
787 */ 784 */
788static void 785static void enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup)
789enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
790{ 786{
791 struct cfs_rq *cfs_rq; 787 struct cfs_rq *cfs_rq;
792 struct sched_entity *se = &p->se; 788 struct sched_entity *se = &p->se;
@@ -795,7 +791,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
795 if (se->on_rq) 791 if (se->on_rq)
796 break; 792 break;
797 cfs_rq = cfs_rq_of(se); 793 cfs_rq = cfs_rq_of(se);
798 enqueue_entity(cfs_rq, se, wakeup, now); 794 enqueue_entity(cfs_rq, se, wakeup);
799 } 795 }
800} 796}
801 797
@@ -804,15 +800,14 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
804 * decreased. We remove the task from the rbtree and 800 * decreased. We remove the task from the rbtree and
805 * update the fair scheduling stats: 801 * update the fair scheduling stats:
806 */ 802 */
807static void 803static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep)
808dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep, u64 now)
809{ 804{
810 struct cfs_rq *cfs_rq; 805 struct cfs_rq *cfs_rq;
811 struct sched_entity *se = &p->se; 806 struct sched_entity *se = &p->se;
812 807
813 for_each_sched_entity(se) { 808 for_each_sched_entity(se) {
814 cfs_rq = cfs_rq_of(se); 809 cfs_rq = cfs_rq_of(se);
815 dequeue_entity(cfs_rq, se, sleep, now); 810 dequeue_entity(cfs_rq, se, sleep);
816 /* Don't dequeue parent if it has other entities besides us */ 811 /* Don't dequeue parent if it has other entities besides us */
817 if (cfs_rq->load.weight) 812 if (cfs_rq->load.weight)
818 break; 813 break;
@@ -825,14 +820,14 @@ dequeue_task_fair(struct rq *rq, struct task_struct *p, int sleep, u64 now)
825static void yield_task_fair(struct rq *rq, struct task_struct *p) 820static void yield_task_fair(struct rq *rq, struct task_struct *p)
826{ 821{
827 struct cfs_rq *cfs_rq = task_cfs_rq(p); 822 struct cfs_rq *cfs_rq = task_cfs_rq(p);
828 u64 now = __rq_clock(rq);
829 823
824 __update_rq_clock(rq);
830 /* 825 /*
831 * Dequeue and enqueue the task to update its 826 * Dequeue and enqueue the task to update its
832 * position within the tree: 827 * position within the tree:
833 */ 828 */
834 dequeue_entity(cfs_rq, &p->se, 0, now); 829 dequeue_entity(cfs_rq, &p->se, 0);
835 enqueue_entity(cfs_rq, &p->se, 0, now); 830 enqueue_entity(cfs_rq, &p->se, 0);
836} 831}
837 832
838/* 833/*
@@ -845,7 +840,8 @@ static void check_preempt_curr_fair(struct rq *rq, struct task_struct *p)
845 unsigned long gran; 840 unsigned long gran;
846 841
847 if (unlikely(rt_prio(p->prio))) { 842 if (unlikely(rt_prio(p->prio))) {
848 update_curr(cfs_rq, rq_clock(rq)); 843 update_rq_clock(rq);
844 update_curr(cfs_rq);
849 resched_task(curr); 845 resched_task(curr);
850 return; 846 return;
851 } 847 }
@@ -861,7 +857,7 @@ static void check_preempt_curr_fair(struct rq *rq, struct task_struct *p)
861 __check_preempt_curr_fair(cfs_rq, &p->se, &curr->se, gran); 857 __check_preempt_curr_fair(cfs_rq, &p->se, &curr->se, gran);
862} 858}
863 859
864static struct task_struct *pick_next_task_fair(struct rq *rq, u64 now) 860static struct task_struct *pick_next_task_fair(struct rq *rq)
865{ 861{
866 struct cfs_rq *cfs_rq = &rq->cfs; 862 struct cfs_rq *cfs_rq = &rq->cfs;
867 struct sched_entity *se; 863 struct sched_entity *se;
@@ -870,7 +866,7 @@ static struct task_struct *pick_next_task_fair(struct rq *rq, u64 now)
870 return NULL; 866 return NULL;
871 867
872 do { 868 do {
873 se = pick_next_entity(cfs_rq, now); 869 se = pick_next_entity(cfs_rq);
874 cfs_rq = group_cfs_rq(se); 870 cfs_rq = group_cfs_rq(se);
875 } while (cfs_rq); 871 } while (cfs_rq);
876 872
@@ -880,14 +876,14 @@ static struct task_struct *pick_next_task_fair(struct rq *rq, u64 now)
880/* 876/*
881 * Account for a descheduled task: 877 * Account for a descheduled task:
882 */ 878 */
883static void put_prev_task_fair(struct rq *rq, struct task_struct *prev, u64 now) 879static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
884{ 880{
885 struct sched_entity *se = &prev->se; 881 struct sched_entity *se = &prev->se;
886 struct cfs_rq *cfs_rq; 882 struct cfs_rq *cfs_rq;
887 883
888 for_each_sched_entity(se) { 884 for_each_sched_entity(se) {
889 cfs_rq = cfs_rq_of(se); 885 cfs_rq = cfs_rq_of(se);
890 put_prev_entity(cfs_rq, se, now); 886 put_prev_entity(cfs_rq, se);
891 } 887 }
892} 888}
893 889
@@ -930,6 +926,7 @@ static struct task_struct *load_balance_next_fair(void *arg)
930 return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr); 926 return __load_balance_iterator(cfs_rq, cfs_rq->rb_load_balance_curr);
931} 927}
932 928
929#ifdef CONFIG_FAIR_GROUP_SCHED
933static int cfs_rq_best_prio(struct cfs_rq *cfs_rq) 930static int cfs_rq_best_prio(struct cfs_rq *cfs_rq)
934{ 931{
935 struct sched_entity *curr; 932 struct sched_entity *curr;
@@ -943,12 +940,13 @@ static int cfs_rq_best_prio(struct cfs_rq *cfs_rq)
943 940
944 return p->prio; 941 return p->prio;
945} 942}
943#endif
946 944
947static int 945static unsigned long
948load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, 946load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
949 unsigned long max_nr_move, unsigned long max_load_move, 947 unsigned long max_nr_move, unsigned long max_load_move,
950 struct sched_domain *sd, enum cpu_idle_type idle, 948 struct sched_domain *sd, enum cpu_idle_type idle,
951 int *all_pinned, unsigned long *total_load_moved) 949 int *all_pinned, int *this_best_prio)
952{ 950{
953 struct cfs_rq *busy_cfs_rq; 951 struct cfs_rq *busy_cfs_rq;
954 unsigned long load_moved, total_nr_moved = 0, nr_moved; 952 unsigned long load_moved, total_nr_moved = 0, nr_moved;
@@ -959,10 +957,10 @@ load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
959 cfs_rq_iterator.next = load_balance_next_fair; 957 cfs_rq_iterator.next = load_balance_next_fair;
960 958
961 for_each_leaf_cfs_rq(busiest, busy_cfs_rq) { 959 for_each_leaf_cfs_rq(busiest, busy_cfs_rq) {
960#ifdef CONFIG_FAIR_GROUP_SCHED
962 struct cfs_rq *this_cfs_rq; 961 struct cfs_rq *this_cfs_rq;
963 long imbalance; 962 long imbalances;
964 unsigned long maxload; 963 unsigned long maxload;
965 int this_best_prio, best_prio, best_prio_seen = 0;
966 964
967 this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu); 965 this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu);
968 966
@@ -976,27 +974,17 @@ load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
976 imbalance /= 2; 974 imbalance /= 2;
977 maxload = min(rem_load_move, imbalance); 975 maxload = min(rem_load_move, imbalance);
978 976
979 this_best_prio = cfs_rq_best_prio(this_cfs_rq); 977 *this_best_prio = cfs_rq_best_prio(this_cfs_rq);
980 best_prio = cfs_rq_best_prio(busy_cfs_rq); 978#else
981 979#define maxload rem_load_move
982 /* 980#endif
983 * Enable handling of the case where there is more than one task
984 * with the best priority. If the current running task is one
985 * of those with prio==best_prio we know it won't be moved
986 * and therefore it's safe to override the skip (based on load)
987 * of any task we find with that prio.
988 */
989 if (cfs_rq_curr(busy_cfs_rq) == &busiest->curr->se)
990 best_prio_seen = 1;
991
992 /* pass busy_cfs_rq argument into 981 /* pass busy_cfs_rq argument into
993 * load_balance_[start|next]_fair iterators 982 * load_balance_[start|next]_fair iterators
994 */ 983 */
995 cfs_rq_iterator.arg = busy_cfs_rq; 984 cfs_rq_iterator.arg = busy_cfs_rq;
996 nr_moved = balance_tasks(this_rq, this_cpu, busiest, 985 nr_moved = balance_tasks(this_rq, this_cpu, busiest,
997 max_nr_move, maxload, sd, idle, all_pinned, 986 max_nr_move, maxload, sd, idle, all_pinned,
998 &load_moved, this_best_prio, best_prio, 987 &load_moved, this_best_prio, &cfs_rq_iterator);
999 best_prio_seen, &cfs_rq_iterator);
1000 988
1001 total_nr_moved += nr_moved; 989 total_nr_moved += nr_moved;
1002 max_nr_move -= nr_moved; 990 max_nr_move -= nr_moved;
@@ -1006,9 +994,7 @@ load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1006 break; 994 break;
1007 } 995 }
1008 996
1009 *total_load_moved = max_load_move - rem_load_move; 997 return max_load_move - rem_load_move;
1010
1011 return total_nr_moved;
1012} 998}
1013 999
1014/* 1000/*
@@ -1032,14 +1018,14 @@ static void task_tick_fair(struct rq *rq, struct task_struct *curr)
1032 * monopolize the CPU. Note: the parent runqueue is locked, 1018 * monopolize the CPU. Note: the parent runqueue is locked,
1033 * the child is not running yet. 1019 * the child is not running yet.
1034 */ 1020 */
1035static void task_new_fair(struct rq *rq, struct task_struct *p, u64 now) 1021static void task_new_fair(struct rq *rq, struct task_struct *p)
1036{ 1022{
1037 struct cfs_rq *cfs_rq = task_cfs_rq(p); 1023 struct cfs_rq *cfs_rq = task_cfs_rq(p);
1038 struct sched_entity *se = &p->se; 1024 struct sched_entity *se = &p->se;
1039 1025
1040 sched_info_queued(p); 1026 sched_info_queued(p);
1041 1027
1042 update_stats_enqueue(cfs_rq, se, now); 1028 update_stats_enqueue(cfs_rq, se);
1043 /* 1029 /*
1044 * Child runs first: we let it run before the parent 1030 * Child runs first: we let it run before the parent
1045 * until it reschedules once. We set up the key so that 1031 * until it reschedules once. We set up the key so that
@@ -1072,15 +1058,10 @@ static void task_new_fair(struct rq *rq, struct task_struct *p, u64 now)
1072 */ 1058 */
1073static void set_curr_task_fair(struct rq *rq) 1059static void set_curr_task_fair(struct rq *rq)
1074{ 1060{
1075 struct task_struct *curr = rq->curr; 1061 struct sched_entity *se = &rq->curr.se;
1076 struct sched_entity *se = &curr->se;
1077 u64 now = rq_clock(rq);
1078 struct cfs_rq *cfs_rq;
1079 1062
1080 for_each_sched_entity(se) { 1063 for_each_sched_entity(se)
1081 cfs_rq = cfs_rq_of(se); 1064 set_next_entity(cfs_rq_of(se), se);
1082 set_next_entity(cfs_rq, se, now);
1083 }
1084} 1065}
1085#else 1066#else
1086static void set_curr_task_fair(struct rq *rq) 1067static void set_curr_task_fair(struct rq *rq)
@@ -1109,12 +1090,11 @@ struct sched_class fair_sched_class __read_mostly = {
1109}; 1090};
1110 1091
1111#ifdef CONFIG_SCHED_DEBUG 1092#ifdef CONFIG_SCHED_DEBUG
1112void print_cfs_stats(struct seq_file *m, int cpu, u64 now) 1093static void print_cfs_stats(struct seq_file *m, int cpu)
1113{ 1094{
1114 struct rq *rq = cpu_rq(cpu);
1115 struct cfs_rq *cfs_rq; 1095 struct cfs_rq *cfs_rq;
1116 1096
1117 for_each_leaf_cfs_rq(rq, cfs_rq) 1097 for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
1118 print_cfs_rq(m, cpu, cfs_rq, now); 1098 print_cfs_rq(m, cpu, cfs_rq);
1119} 1099}
1120#endif 1100#endif
diff --git a/kernel/sched_idletask.c b/kernel/sched_idletask.c
index 41841e741c4a..3503fb2d9f96 100644
--- a/kernel/sched_idletask.c
+++ b/kernel/sched_idletask.c
@@ -13,7 +13,7 @@ static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p)
13 resched_task(rq->idle); 13 resched_task(rq->idle);
14} 14}
15 15
16static struct task_struct *pick_next_task_idle(struct rq *rq, u64 now) 16static struct task_struct *pick_next_task_idle(struct rq *rq)
17{ 17{
18 schedstat_inc(rq, sched_goidle); 18 schedstat_inc(rq, sched_goidle);
19 19
@@ -25,7 +25,7 @@ static struct task_struct *pick_next_task_idle(struct rq *rq, u64 now)
25 * message if some code attempts to do it: 25 * message if some code attempts to do it:
26 */ 26 */
27static void 27static void
28dequeue_task_idle(struct rq *rq, struct task_struct *p, int sleep, u64 now) 28dequeue_task_idle(struct rq *rq, struct task_struct *p, int sleep)
29{ 29{
30 spin_unlock_irq(&rq->lock); 30 spin_unlock_irq(&rq->lock);
31 printk(KERN_ERR "bad: scheduling from the idle thread!\n"); 31 printk(KERN_ERR "bad: scheduling from the idle thread!\n");
@@ -33,15 +33,15 @@ dequeue_task_idle(struct rq *rq, struct task_struct *p, int sleep, u64 now)
33 spin_lock_irq(&rq->lock); 33 spin_lock_irq(&rq->lock);
34} 34}
35 35
36static void put_prev_task_idle(struct rq *rq, struct task_struct *prev, u64 now) 36static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
37{ 37{
38} 38}
39 39
40static int 40static unsigned long
41load_balance_idle(struct rq *this_rq, int this_cpu, struct rq *busiest, 41load_balance_idle(struct rq *this_rq, int this_cpu, struct rq *busiest,
42 unsigned long max_nr_move, unsigned long max_load_move, 42 unsigned long max_nr_move, unsigned long max_load_move,
43 struct sched_domain *sd, enum cpu_idle_type idle, 43 struct sched_domain *sd, enum cpu_idle_type idle,
44 int *all_pinned, unsigned long *total_load_moved) 44 int *all_pinned, int *this_best_prio)
45{ 45{
46 return 0; 46 return 0;
47} 47}
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index 002fcf8d3f64..dcdcad632fd9 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -7,7 +7,7 @@
7 * Update the current task's runtime statistics. Skip current tasks that 7 * Update the current task's runtime statistics. Skip current tasks that
8 * are not in our scheduling class. 8 * are not in our scheduling class.
9 */ 9 */
10static inline void update_curr_rt(struct rq *rq, u64 now) 10static inline void update_curr_rt(struct rq *rq)
11{ 11{
12 struct task_struct *curr = rq->curr; 12 struct task_struct *curr = rq->curr;
13 u64 delta_exec; 13 u64 delta_exec;
@@ -15,18 +15,17 @@ static inline void update_curr_rt(struct rq *rq, u64 now)
15 if (!task_has_rt_policy(curr)) 15 if (!task_has_rt_policy(curr))
16 return; 16 return;
17 17
18 delta_exec = now - curr->se.exec_start; 18 delta_exec = rq->clock - curr->se.exec_start;
19 if (unlikely((s64)delta_exec < 0)) 19 if (unlikely((s64)delta_exec < 0))
20 delta_exec = 0; 20 delta_exec = 0;
21 21
22 schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec)); 22 schedstat_set(curr->se.exec_max, max(curr->se.exec_max, delta_exec));
23 23
24 curr->se.sum_exec_runtime += delta_exec; 24 curr->se.sum_exec_runtime += delta_exec;
25 curr->se.exec_start = now; 25 curr->se.exec_start = rq->clock;
26} 26}
27 27
28static void 28static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
29enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
30{ 29{
31 struct rt_prio_array *array = &rq->rt.active; 30 struct rt_prio_array *array = &rq->rt.active;
32 31
@@ -37,12 +36,11 @@ enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
37/* 36/*
38 * Adding/removing a task to/from a priority array: 37 * Adding/removing a task to/from a priority array:
39 */ 38 */
40static void 39static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
41dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep, u64 now)
42{ 40{
43 struct rt_prio_array *array = &rq->rt.active; 41 struct rt_prio_array *array = &rq->rt.active;
44 42
45 update_curr_rt(rq, now); 43 update_curr_rt(rq);
46 44
47 list_del(&p->run_list); 45 list_del(&p->run_list);
48 if (list_empty(array->queue + p->prio)) 46 if (list_empty(array->queue + p->prio))
@@ -75,7 +73,7 @@ static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p)
75 resched_task(rq->curr); 73 resched_task(rq->curr);
76} 74}
77 75
78static struct task_struct *pick_next_task_rt(struct rq *rq, u64 now) 76static struct task_struct *pick_next_task_rt(struct rq *rq)
79{ 77{
80 struct rt_prio_array *array = &rq->rt.active; 78 struct rt_prio_array *array = &rq->rt.active;
81 struct task_struct *next; 79 struct task_struct *next;
@@ -89,14 +87,14 @@ static struct task_struct *pick_next_task_rt(struct rq *rq, u64 now)
89 queue = array->queue + idx; 87 queue = array->queue + idx;
90 next = list_entry(queue->next, struct task_struct, run_list); 88 next = list_entry(queue->next, struct task_struct, run_list);
91 89
92 next->se.exec_start = now; 90 next->se.exec_start = rq->clock;
93 91
94 return next; 92 return next;
95} 93}
96 94
97static void put_prev_task_rt(struct rq *rq, struct task_struct *p, u64 now) 95static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
98{ 96{
99 update_curr_rt(rq, now); 97 update_curr_rt(rq);
100 p->se.exec_start = 0; 98 p->se.exec_start = 0;
101} 99}
102 100
@@ -172,28 +170,15 @@ static struct task_struct *load_balance_next_rt(void *arg)
172 return p; 170 return p;
173} 171}
174 172
175static int 173static unsigned long
176load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest, 174load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
177 unsigned long max_nr_move, unsigned long max_load_move, 175 unsigned long max_nr_move, unsigned long max_load_move,
178 struct sched_domain *sd, enum cpu_idle_type idle, 176 struct sched_domain *sd, enum cpu_idle_type idle,
179 int *all_pinned, unsigned long *load_moved) 177 int *all_pinned, int *this_best_prio)
180{ 178{
181 int this_best_prio, best_prio, best_prio_seen = 0;
182 int nr_moved; 179 int nr_moved;
183 struct rq_iterator rt_rq_iterator; 180 struct rq_iterator rt_rq_iterator;
184 181 unsigned long load_moved;
185 best_prio = sched_find_first_bit(busiest->rt.active.bitmap);
186 this_best_prio = sched_find_first_bit(this_rq->rt.active.bitmap);
187
188 /*
189 * Enable handling of the case where there is more than one task
190 * with the best priority. If the current running task is one
191 * of those with prio==best_prio we know it won't be moved
192 * and therefore it's safe to override the skip (based on load)
193 * of any task we find with that prio.
194 */
195 if (busiest->curr->prio == best_prio)
196 best_prio_seen = 1;
197 182
198 rt_rq_iterator.start = load_balance_start_rt; 183 rt_rq_iterator.start = load_balance_start_rt;
199 rt_rq_iterator.next = load_balance_next_rt; 184 rt_rq_iterator.next = load_balance_next_rt;
@@ -203,11 +188,10 @@ load_balance_rt(struct rq *this_rq, int this_cpu, struct rq *busiest,
203 rt_rq_iterator.arg = busiest; 188 rt_rq_iterator.arg = busiest;
204 189
205 nr_moved = balance_tasks(this_rq, this_cpu, busiest, max_nr_move, 190 nr_moved = balance_tasks(this_rq, this_cpu, busiest, max_nr_move,
206 max_load_move, sd, idle, all_pinned, load_moved, 191 max_load_move, sd, idle, all_pinned, &load_moved,
207 this_best_prio, best_prio, best_prio_seen, 192 this_best_prio, &rt_rq_iterator);
208 &rt_rq_iterator);
209 193
210 return nr_moved; 194 return load_moved;
211} 195}
212 196
213static void task_tick_rt(struct rq *rq, struct task_struct *p) 197static void task_tick_rt(struct rq *rq, struct task_struct *p)
generated by cgit v1.2.2 (git 2.25.0) at 2026-05-28 00:44:09 -0400