diff options
Diffstat (limited to 'kernel/sched.c')
-rw-r--r-- | kernel/sched.c | 165 |
1 files changed, 13 insertions, 152 deletions
diff --git a/kernel/sched.c b/kernel/sched.c index 9457106b18af..58fb8af15776 100644 --- a/kernel/sched.c +++ b/kernel/sched.c | |||
@@ -75,16 +75,6 @@ | |||
75 | #include <asm/irq_regs.h> | 75 | #include <asm/irq_regs.h> |
76 | 76 | ||
77 | /* | 77 | /* |
78 | * Scheduler clock - returns current time in nanosec units. | ||
79 | * This is default implementation. | ||
80 | * Architectures and sub-architectures can override this. | ||
81 | */ | ||
82 | unsigned long long __attribute__((weak)) sched_clock(void) | ||
83 | { | ||
84 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); | ||
85 | } | ||
86 | |||
87 | /* | ||
88 | * Convert user-nice values [ -20 ... 0 ... 19 ] | 78 | * Convert user-nice values [ -20 ... 0 ... 19 ] |
89 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | 79 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], |
90 | * and back. | 80 | * and back. |
@@ -557,13 +547,7 @@ struct rq { | |||
557 | unsigned long next_balance; | 547 | unsigned long next_balance; |
558 | struct mm_struct *prev_mm; | 548 | struct mm_struct *prev_mm; |
559 | 549 | ||
560 | u64 clock, prev_clock_raw; | 550 | u64 clock; |
561 | s64 clock_max_delta; | ||
562 | |||
563 | unsigned int clock_warps, clock_overflows, clock_underflows; | ||
564 | u64 idle_clock; | ||
565 | unsigned int clock_deep_idle_events; | ||
566 | u64 tick_timestamp; | ||
567 | 551 | ||
568 | atomic_t nr_iowait; | 552 | atomic_t nr_iowait; |
569 | 553 | ||
@@ -628,82 +612,6 @@ static inline int cpu_of(struct rq *rq) | |||
628 | #endif | 612 | #endif |
629 | } | 613 | } |
630 | 614 | ||
631 | #ifdef CONFIG_NO_HZ | ||
632 | static inline bool nohz_on(int cpu) | ||
633 | { | ||
634 | return tick_get_tick_sched(cpu)->nohz_mode != NOHZ_MODE_INACTIVE; | ||
635 | } | ||
636 | |||
637 | static inline u64 max_skipped_ticks(struct rq *rq) | ||
638 | { | ||
639 | return nohz_on(cpu_of(rq)) ? jiffies - rq->last_tick_seen + 2 : 1; | ||
640 | } | ||
641 | |||
642 | static inline void update_last_tick_seen(struct rq *rq) | ||
643 | { | ||
644 | rq->last_tick_seen = jiffies; | ||
645 | } | ||
646 | #else | ||
647 | static inline u64 max_skipped_ticks(struct rq *rq) | ||
648 | { | ||
649 | return 1; | ||
650 | } | ||
651 | |||
652 | static inline void update_last_tick_seen(struct rq *rq) | ||
653 | { | ||
654 | } | ||
655 | #endif | ||
656 | |||
657 | /* | ||
658 | * Update the per-runqueue clock, as finegrained as the platform can give | ||
659 | * us, but without assuming monotonicity, etc.: | ||
660 | */ | ||
661 | static void __update_rq_clock(struct rq *rq) | ||
662 | { | ||
663 | u64 prev_raw = rq->prev_clock_raw; | ||
664 | u64 now = sched_clock(); | ||
665 | s64 delta = now - prev_raw; | ||
666 | u64 clock = rq->clock; | ||
667 | |||
668 | #ifdef CONFIG_SCHED_DEBUG | ||
669 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | ||
670 | #endif | ||
671 | /* | ||
672 | * Protect against sched_clock() occasionally going backwards: | ||
673 | */ | ||
674 | if (unlikely(delta < 0)) { | ||
675 | clock++; | ||
676 | rq->clock_warps++; | ||
677 | } else { | ||
678 | /* | ||
679 | * Catch too large forward jumps too: | ||
680 | */ | ||
681 | u64 max_jump = max_skipped_ticks(rq) * TICK_NSEC; | ||
682 | u64 max_time = rq->tick_timestamp + max_jump; | ||
683 | |||
684 | if (unlikely(clock + delta > max_time)) { | ||
685 | if (clock < max_time) | ||
686 | clock = max_time; | ||
687 | else | ||
688 | clock++; | ||
689 | rq->clock_overflows++; | ||
690 | } else { | ||
691 | if (unlikely(delta > rq->clock_max_delta)) | ||
692 | rq->clock_max_delta = delta; | ||
693 | clock += delta; | ||
694 | } | ||
695 | } | ||
696 | |||
697 | rq->prev_clock_raw = now; | ||
698 | rq->clock = clock; | ||
699 | } | ||
700 | |||
701 | static void update_rq_clock(struct rq *rq) | ||
702 | { | ||
703 | if (likely(smp_processor_id() == cpu_of(rq))) | ||
704 | __update_rq_clock(rq); | ||
705 | } | ||
706 | |||
707 | /* | 615 | /* |
708 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | 616 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. |
709 | * See detach_destroy_domains: synchronize_sched for details. | 617 | * See detach_destroy_domains: synchronize_sched for details. |
@@ -719,6 +627,11 @@ static void update_rq_clock(struct rq *rq) | |||
719 | #define task_rq(p) cpu_rq(task_cpu(p)) | 627 | #define task_rq(p) cpu_rq(task_cpu(p)) |
720 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | 628 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) |
721 | 629 | ||
630 | static inline void update_rq_clock(struct rq *rq) | ||
631 | { | ||
632 | rq->clock = sched_clock_cpu(cpu_of(rq)); | ||
633 | } | ||
634 | |||
722 | /* | 635 | /* |
723 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | 636 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: |
724 | */ | 637 | */ |
@@ -935,7 +848,6 @@ static unsigned long long __sync_cpu_clock(unsigned long long time, int cpu) | |||
935 | static unsigned long long __cpu_clock(int cpu) | 848 | static unsigned long long __cpu_clock(int cpu) |
936 | { | 849 | { |
937 | unsigned long long now; | 850 | unsigned long long now; |
938 | struct rq *rq; | ||
939 | 851 | ||
940 | /* | 852 | /* |
941 | * Only call sched_clock() if the scheduler has already been | 853 | * Only call sched_clock() if the scheduler has already been |
@@ -944,9 +856,7 @@ static unsigned long long __cpu_clock(int cpu) | |||
944 | if (unlikely(!scheduler_running)) | 856 | if (unlikely(!scheduler_running)) |
945 | return 0; | 857 | return 0; |
946 | 858 | ||
947 | rq = cpu_rq(cpu); | 859 | now = sched_clock_cpu(cpu); |
948 | update_rq_clock(rq); | ||
949 | now = rq->clock; | ||
950 | 860 | ||
951 | return now; | 861 | return now; |
952 | } | 862 | } |
@@ -1120,45 +1030,6 @@ static struct rq *this_rq_lock(void) | |||
1120 | return rq; | 1030 | return rq; |
1121 | } | 1031 | } |
1122 | 1032 | ||
1123 | /* | ||
1124 | * We are going deep-idle (irqs are disabled): | ||
1125 | */ | ||
1126 | void sched_clock_idle_sleep_event(void) | ||
1127 | { | ||
1128 | struct rq *rq = cpu_rq(smp_processor_id()); | ||
1129 | |||
1130 | WARN_ON(!irqs_disabled()); | ||
1131 | spin_lock(&rq->lock); | ||
1132 | __update_rq_clock(rq); | ||
1133 | spin_unlock(&rq->lock); | ||
1134 | rq->clock_deep_idle_events++; | ||
1135 | } | ||
1136 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | ||
1137 | |||
1138 | /* | ||
1139 | * We just idled delta nanoseconds (called with irqs disabled): | ||
1140 | */ | ||
1141 | void sched_clock_idle_wakeup_event(u64 delta_ns) | ||
1142 | { | ||
1143 | struct rq *rq = cpu_rq(smp_processor_id()); | ||
1144 | u64 now = sched_clock(); | ||
1145 | |||
1146 | WARN_ON(!irqs_disabled()); | ||
1147 | rq->idle_clock += delta_ns; | ||
1148 | /* | ||
1149 | * Override the previous timestamp and ignore all | ||
1150 | * sched_clock() deltas that occured while we idled, | ||
1151 | * and use the PM-provided delta_ns to advance the | ||
1152 | * rq clock: | ||
1153 | */ | ||
1154 | spin_lock(&rq->lock); | ||
1155 | rq->prev_clock_raw = now; | ||
1156 | rq->clock += delta_ns; | ||
1157 | spin_unlock(&rq->lock); | ||
1158 | touch_softlockup_watchdog(); | ||
1159 | } | ||
1160 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); | ||
1161 | |||
1162 | static void __resched_task(struct task_struct *p, int tif_bit); | 1033 | static void __resched_task(struct task_struct *p, int tif_bit); |
1163 | 1034 | ||
1164 | static inline void resched_task(struct task_struct *p) | 1035 | static inline void resched_task(struct task_struct *p) |
@@ -1283,7 +1154,7 @@ static enum hrtimer_restart hrtick(struct hrtimer *timer) | |||
1283 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | 1154 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); |
1284 | 1155 | ||
1285 | spin_lock(&rq->lock); | 1156 | spin_lock(&rq->lock); |
1286 | __update_rq_clock(rq); | 1157 | update_rq_clock(rq); |
1287 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | 1158 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1288 | spin_unlock(&rq->lock); | 1159 | spin_unlock(&rq->lock); |
1289 | 1160 | ||
@@ -4476,19 +4347,11 @@ void scheduler_tick(void) | |||
4476 | int cpu = smp_processor_id(); | 4347 | int cpu = smp_processor_id(); |
4477 | struct rq *rq = cpu_rq(cpu); | 4348 | struct rq *rq = cpu_rq(cpu); |
4478 | struct task_struct *curr = rq->curr; | 4349 | struct task_struct *curr = rq->curr; |
4479 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; | 4350 | |
4351 | sched_clock_tick(); | ||
4480 | 4352 | ||
4481 | spin_lock(&rq->lock); | 4353 | spin_lock(&rq->lock); |
4482 | __update_rq_clock(rq); | 4354 | update_rq_clock(rq); |
4483 | /* | ||
4484 | * Let rq->clock advance by at least TICK_NSEC: | ||
4485 | */ | ||
4486 | if (unlikely(rq->clock < next_tick)) { | ||
4487 | rq->clock = next_tick; | ||
4488 | rq->clock_underflows++; | ||
4489 | } | ||
4490 | rq->tick_timestamp = rq->clock; | ||
4491 | update_last_tick_seen(rq); | ||
4492 | update_cpu_load(rq); | 4355 | update_cpu_load(rq); |
4493 | curr->sched_class->task_tick(rq, curr, 0); | 4356 | curr->sched_class->task_tick(rq, curr, 0); |
4494 | spin_unlock(&rq->lock); | 4357 | spin_unlock(&rq->lock); |
@@ -4642,7 +4505,7 @@ need_resched_nonpreemptible: | |||
4642 | * Do the rq-clock update outside the rq lock: | 4505 | * Do the rq-clock update outside the rq lock: |
4643 | */ | 4506 | */ |
4644 | local_irq_disable(); | 4507 | local_irq_disable(); |
4645 | __update_rq_clock(rq); | 4508 | update_rq_clock(rq); |
4646 | spin_lock(&rq->lock); | 4509 | spin_lock(&rq->lock); |
4647 | clear_tsk_need_resched(prev); | 4510 | clear_tsk_need_resched(prev); |
4648 | 4511 | ||
@@ -8226,8 +8089,6 @@ void __init sched_init(void) | |||
8226 | spin_lock_init(&rq->lock); | 8089 | spin_lock_init(&rq->lock); |
8227 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); | 8090 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
8228 | rq->nr_running = 0; | 8091 | rq->nr_running = 0; |
8229 | rq->clock = 1; | ||
8230 | update_last_tick_seen(rq); | ||
8231 | init_cfs_rq(&rq->cfs, rq); | 8092 | init_cfs_rq(&rq->cfs, rq); |
8232 | init_rt_rq(&rq->rt, rq); | 8093 | init_rt_rq(&rq->rt, rq); |
8233 | #ifdef CONFIG_FAIR_GROUP_SCHED | 8094 | #ifdef CONFIG_FAIR_GROUP_SCHED |
@@ -8371,6 +8232,7 @@ EXPORT_SYMBOL(__might_sleep); | |||
8371 | static void normalize_task(struct rq *rq, struct task_struct *p) | 8232 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8372 | { | 8233 | { |
8373 | int on_rq; | 8234 | int on_rq; |
8235 | |||
8374 | update_rq_clock(rq); | 8236 | update_rq_clock(rq); |
8375 | on_rq = p->se.on_rq; | 8237 | on_rq = p->se.on_rq; |
8376 | if (on_rq) | 8238 | if (on_rq) |
@@ -8402,7 +8264,6 @@ void normalize_rt_tasks(void) | |||
8402 | p->se.sleep_start = 0; | 8264 | p->se.sleep_start = 0; |
8403 | p->se.block_start = 0; | 8265 | p->se.block_start = 0; |
8404 | #endif | 8266 | #endif |
8405 | task_rq(p)->clock = 0; | ||
8406 | 8267 | ||
8407 | if (!rt_task(p)) { | 8268 | if (!rt_task(p)) { |
8408 | /* | 8269 | /* |