sched: Mark autogroup_init() __init
[deliverable/linux.git] / kernel / sched.c
CommitLineData
1da177e4
LT
1/*
2 * kernel/sched.c
3 *
4 * Kernel scheduler and related syscalls
5 *
6 * Copyright (C) 1991-2002 Linus Torvalds
7 *
8 * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
9 * make semaphores SMP safe
10 * 1998-11-19 Implemented schedule_timeout() and related stuff
11 * by Andrea Arcangeli
12 * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
13 * hybrid priority-list and round-robin design with
14 * an array-switch method of distributing timeslices
15 * and per-CPU runqueues. Cleanups and useful suggestions
16 * by Davide Libenzi, preemptible kernel bits by Robert Love.
17 * 2003-09-03 Interactivity tuning by Con Kolivas.
18 * 2004-04-02 Scheduler domains code by Nick Piggin
c31f2e8a
IM
19 * 2007-04-15 Work begun on replacing all interactivity tuning with a
20 * fair scheduling design by Con Kolivas.
21 * 2007-05-05 Load balancing (smp-nice) and other improvements
22 * by Peter Williams
23 * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
24 * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
b9131769
IM
25 * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins,
26 * Thomas Gleixner, Mike Kravetz
1da177e4
LT
27 */
28
29#include <linux/mm.h>
30#include <linux/module.h>
31#include <linux/nmi.h>
32#include <linux/init.h>
dff06c15 33#include <linux/uaccess.h>
1da177e4
LT
34#include <linux/highmem.h>
35#include <linux/smp_lock.h>
36#include <asm/mmu_context.h>
37#include <linux/interrupt.h>
c59ede7b 38#include <linux/capability.h>
1da177e4
LT
39#include <linux/completion.h>
40#include <linux/kernel_stat.h>
9a11b49a 41#include <linux/debug_locks.h>
cdd6c482 42#include <linux/perf_event.h>
1da177e4
LT
43#include <linux/security.h>
44#include <linux/notifier.h>
45#include <linux/profile.h>
7dfb7103 46#include <linux/freezer.h>
198e2f18 47#include <linux/vmalloc.h>
1da177e4
LT
48#include <linux/blkdev.h>
49#include <linux/delay.h>
b488893a 50#include <linux/pid_namespace.h>
1da177e4
LT
51#include <linux/smp.h>
52#include <linux/threads.h>
53#include <linux/timer.h>
54#include <linux/rcupdate.h>
55#include <linux/cpu.h>
56#include <linux/cpuset.h>
57#include <linux/percpu.h>
b5aadf7f 58#include <linux/proc_fs.h>
1da177e4 59#include <linux/seq_file.h>
969c7921 60#include <linux/stop_machine.h>
e692ab53 61#include <linux/sysctl.h>
1da177e4
LT
62#include <linux/syscalls.h>
63#include <linux/times.h>
8f0ab514 64#include <linux/tsacct_kern.h>
c6fd91f0 65#include <linux/kprobes.h>
0ff92245 66#include <linux/delayacct.h>
dff06c15 67#include <linux/unistd.h>
f5ff8422 68#include <linux/pagemap.h>
8f4d37ec 69#include <linux/hrtimer.h>
30914a58 70#include <linux/tick.h>
f00b45c1
PZ
71#include <linux/debugfs.h>
72#include <linux/ctype.h>
6cd8a4bb 73#include <linux/ftrace.h>
5a0e3ad6 74#include <linux/slab.h>
1da177e4 75
5517d86b 76#include <asm/tlb.h>
838225b4 77#include <asm/irq_regs.h>
335d7afb 78#include <asm/mutex.h>
1da177e4 79
6e0534f2 80#include "sched_cpupri.h"
21aa9af0 81#include "workqueue_sched.h"
5091faa4 82#include "sched_autogroup.h"
6e0534f2 83
a8d154b0 84#define CREATE_TRACE_POINTS
ad8d75ff 85#include <trace/events/sched.h>
a8d154b0 86
1da177e4
LT
87/*
88 * Convert user-nice values [ -20 ... 0 ... 19 ]
89 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
90 * and back.
91 */
92#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
93#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
94#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
95
96/*
97 * 'User priority' is the nice value converted to something we
98 * can work with better when scaling various scheduler parameters,
99 * it's a [ 0 ... 39 ] range.
100 */
101#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
102#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
103#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
104
105/*
d7876a08 106 * Helpers for converting nanosecond timing to jiffy resolution
1da177e4 107 */
d6322faf 108#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
1da177e4 109
6aa645ea
IM
110#define NICE_0_LOAD SCHED_LOAD_SCALE
111#define NICE_0_SHIFT SCHED_LOAD_SHIFT
112
1da177e4
LT
113/*
114 * These are the 'tuning knobs' of the scheduler:
115 *
a4ec24b4 116 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
1da177e4
LT
117 * Timeslices get refilled after they expire.
118 */
1da177e4 119#define DEF_TIMESLICE (100 * HZ / 1000)
2dd73a4f 120
d0b27fa7
PZ
121/*
122 * single value that denotes runtime == period, ie unlimited time.
123 */
124#define RUNTIME_INF ((u64)~0ULL)
125
e05606d3
IM
126static inline int rt_policy(int policy)
127{
3f33a7ce 128 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
e05606d3
IM
129 return 1;
130 return 0;
131}
132
133static inline int task_has_rt_policy(struct task_struct *p)
134{
135 return rt_policy(p->policy);
136}
137
1da177e4 138/*
6aa645ea 139 * This is the priority-queue data structure of the RT scheduling class:
1da177e4 140 */
6aa645ea
IM
141struct rt_prio_array {
142 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
143 struct list_head queue[MAX_RT_PRIO];
144};
145
d0b27fa7 146struct rt_bandwidth {
ea736ed5 147 /* nests inside the rq lock: */
0986b11b 148 raw_spinlock_t rt_runtime_lock;
ea736ed5
IM
149 ktime_t rt_period;
150 u64 rt_runtime;
151 struct hrtimer rt_period_timer;
d0b27fa7
PZ
152};
153
154static struct rt_bandwidth def_rt_bandwidth;
155
156static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
157
158static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
159{
160 struct rt_bandwidth *rt_b =
161 container_of(timer, struct rt_bandwidth, rt_period_timer);
162 ktime_t now;
163 int overrun;
164 int idle = 0;
165
166 for (;;) {
167 now = hrtimer_cb_get_time(timer);
168 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
169
170 if (!overrun)
171 break;
172
173 idle = do_sched_rt_period_timer(rt_b, overrun);
174 }
175
176 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
177}
178
179static
180void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
181{
182 rt_b->rt_period = ns_to_ktime(period);
183 rt_b->rt_runtime = runtime;
184
0986b11b 185 raw_spin_lock_init(&rt_b->rt_runtime_lock);
ac086bc2 186
d0b27fa7
PZ
187 hrtimer_init(&rt_b->rt_period_timer,
188 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
189 rt_b->rt_period_timer.function = sched_rt_period_timer;
d0b27fa7
PZ
190}
191
c8bfff6d
KH
192static inline int rt_bandwidth_enabled(void)
193{
194 return sysctl_sched_rt_runtime >= 0;
d0b27fa7
PZ
195}
196
197static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
198{
199 ktime_t now;
200
cac64d00 201 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
d0b27fa7
PZ
202 return;
203
204 if (hrtimer_active(&rt_b->rt_period_timer))
205 return;
206
0986b11b 207 raw_spin_lock(&rt_b->rt_runtime_lock);
d0b27fa7 208 for (;;) {
7f1e2ca9
PZ
209 unsigned long delta;
210 ktime_t soft, hard;
211
d0b27fa7
PZ
212 if (hrtimer_active(&rt_b->rt_period_timer))
213 break;
214
215 now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
216 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
7f1e2ca9
PZ
217
218 soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
219 hard = hrtimer_get_expires(&rt_b->rt_period_timer);
220 delta = ktime_to_ns(ktime_sub(hard, soft));
221 __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
5c333864 222 HRTIMER_MODE_ABS_PINNED, 0);
d0b27fa7 223 }
0986b11b 224 raw_spin_unlock(&rt_b->rt_runtime_lock);
d0b27fa7
PZ
225}
226
227#ifdef CONFIG_RT_GROUP_SCHED
228static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
229{
230 hrtimer_cancel(&rt_b->rt_period_timer);
231}
232#endif
233
712555ee
HC
234/*
235 * sched_domains_mutex serializes calls to arch_init_sched_domains,
236 * detach_destroy_domains and partition_sched_domains.
237 */
238static DEFINE_MUTEX(sched_domains_mutex);
239
7c941438 240#ifdef CONFIG_CGROUP_SCHED
29f59db3 241
68318b8e
SV
242#include <linux/cgroup.h>
243
29f59db3
SV
244struct cfs_rq;
245
6f505b16
PZ
246static LIST_HEAD(task_groups);
247
29f59db3 248/* task group related information */
4cf86d77 249struct task_group {
68318b8e 250 struct cgroup_subsys_state css;
6c415b92 251
052f1dc7 252#ifdef CONFIG_FAIR_GROUP_SCHED
29f59db3
SV
253 /* schedulable entities of this group on each cpu */
254 struct sched_entity **se;
255 /* runqueue "owned" by this group on each cpu */
256 struct cfs_rq **cfs_rq;
257 unsigned long shares;
2069dd75
PZ
258
259 atomic_t load_weight;
052f1dc7
PZ
260#endif
261
262#ifdef CONFIG_RT_GROUP_SCHED
263 struct sched_rt_entity **rt_se;
264 struct rt_rq **rt_rq;
265
d0b27fa7 266 struct rt_bandwidth rt_bandwidth;
052f1dc7 267#endif
6b2d7700 268
ae8393e5 269 struct rcu_head rcu;
6f505b16 270 struct list_head list;
f473aa5e
PZ
271
272 struct task_group *parent;
273 struct list_head siblings;
274 struct list_head children;
5091faa4
MG
275
276#ifdef CONFIG_SCHED_AUTOGROUP
277 struct autogroup *autogroup;
278#endif
29f59db3
SV
279};
280
3d4b47b4 281/* task_group_lock serializes the addition/removal of task groups */
8ed36996 282static DEFINE_SPINLOCK(task_group_lock);
ec2c507f 283
e9036b36
CG
284#ifdef CONFIG_FAIR_GROUP_SCHED
285
07e06b01 286# define ROOT_TASK_GROUP_LOAD NICE_0_LOAD
052f1dc7 287
cb4ad1ff 288/*
2e084786
LJ
289 * A weight of 0 or 1 can cause arithmetics problems.
290 * A weight of a cfs_rq is the sum of weights of which entities
291 * are queued on this cfs_rq, so a weight of a entity should not be
292 * too large, so as the shares value of a task group.
cb4ad1ff
MX
293 * (The default weight is 1024 - so there's no practical
294 * limitation from this.)
295 */
18d95a28 296#define MIN_SHARES 2
2e084786 297#define MAX_SHARES (1UL << 18)
18d95a28 298
07e06b01 299static int root_task_group_load = ROOT_TASK_GROUP_LOAD;
052f1dc7
PZ
300#endif
301
29f59db3 302/* Default task group.
3a252015 303 * Every task in system belong to this group at bootup.
29f59db3 304 */
07e06b01 305struct task_group root_task_group;
29f59db3 306
7c941438 307#endif /* CONFIG_CGROUP_SCHED */
29f59db3 308
6aa645ea
IM
309/* CFS-related fields in a runqueue */
310struct cfs_rq {
311 struct load_weight load;
312 unsigned long nr_running;
313
6aa645ea 314 u64 exec_clock;
e9acbff6 315 u64 min_vruntime;
6aa645ea
IM
316
317 struct rb_root tasks_timeline;
318 struct rb_node *rb_leftmost;
4a55bd5e
PZ
319
320 struct list_head tasks;
321 struct list_head *balance_iterator;
322
323 /*
324 * 'curr' points to currently running entity on this cfs_rq.
6aa645ea
IM
325 * It is set to NULL otherwise (i.e when none are currently running).
326 */
4793241b 327 struct sched_entity *curr, *next, *last;
ddc97297 328
5ac5c4d6 329 unsigned int nr_spread_over;
ddc97297 330
62160e3f 331#ifdef CONFIG_FAIR_GROUP_SCHED
6aa645ea
IM
332 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
333
41a2d6cf
IM
334 /*
335 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
6aa645ea
IM
336 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
337 * (like users, containers etc.)
338 *
339 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
340 * list is used during load balance.
341 */
3d4b47b4 342 int on_list;
41a2d6cf
IM
343 struct list_head leaf_cfs_rq_list;
344 struct task_group *tg; /* group that "owns" this runqueue */
c09595f6
PZ
345
346#ifdef CONFIG_SMP
c09595f6 347 /*
c8cba857 348 * the part of load.weight contributed by tasks
c09595f6 349 */
c8cba857 350 unsigned long task_weight;
c09595f6 351
c8cba857
PZ
352 /*
353 * h_load = weight * f(tg)
354 *
355 * Where f(tg) is the recursive weight fraction assigned to
356 * this group.
357 */
358 unsigned long h_load;
c09595f6 359
c8cba857 360 /*
3b3d190e
PT
361 * Maintaining per-cpu shares distribution for group scheduling
362 *
363 * load_stamp is the last time we updated the load average
364 * load_last is the last time we updated the load average and saw load
365 * load_unacc_exec_time is currently unaccounted execution time
c8cba857 366 */
2069dd75
PZ
367 u64 load_avg;
368 u64 load_period;
3b3d190e 369 u64 load_stamp, load_last, load_unacc_exec_time;
f1d239f7 370
2069dd75 371 unsigned long load_contribution;
c09595f6 372#endif
6aa645ea
IM
373#endif
374};
1da177e4 375
6aa645ea
IM
376/* Real-Time classes' related field in a runqueue: */
377struct rt_rq {
378 struct rt_prio_array active;
63489e45 379 unsigned long rt_nr_running;
052f1dc7 380#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
e864c499
GH
381 struct {
382 int curr; /* highest queued rt task prio */
398a153b 383#ifdef CONFIG_SMP
e864c499 384 int next; /* next highest */
398a153b 385#endif
e864c499 386 } highest_prio;
6f505b16 387#endif
fa85ae24 388#ifdef CONFIG_SMP
73fe6aae 389 unsigned long rt_nr_migratory;
a1ba4d8b 390 unsigned long rt_nr_total;
a22d7fc1 391 int overloaded;
917b627d 392 struct plist_head pushable_tasks;
fa85ae24 393#endif
6f505b16 394 int rt_throttled;
fa85ae24 395 u64 rt_time;
ac086bc2 396 u64 rt_runtime;
ea736ed5 397 /* Nests inside the rq lock: */
0986b11b 398 raw_spinlock_t rt_runtime_lock;
6f505b16 399
052f1dc7 400#ifdef CONFIG_RT_GROUP_SCHED
23b0fdfc
PZ
401 unsigned long rt_nr_boosted;
402
6f505b16
PZ
403 struct rq *rq;
404 struct list_head leaf_rt_rq_list;
405 struct task_group *tg;
6f505b16 406#endif
6aa645ea
IM
407};
408
57d885fe
GH
409#ifdef CONFIG_SMP
410
411/*
412 * We add the notion of a root-domain which will be used to define per-domain
0eab9146
IM
413 * variables. Each exclusive cpuset essentially defines an island domain by
414 * fully partitioning the member cpus from any other cpuset. Whenever a new
57d885fe
GH
415 * exclusive cpuset is created, we also create and attach a new root-domain
416 * object.
417 *
57d885fe
GH
418 */
419struct root_domain {
420 atomic_t refcount;
c6c4927b
RR
421 cpumask_var_t span;
422 cpumask_var_t online;
637f5085 423
0eab9146 424 /*
637f5085
GH
425 * The "RT overload" flag: it gets set if a CPU has more than
426 * one runnable RT task.
427 */
c6c4927b 428 cpumask_var_t rto_mask;
0eab9146 429 atomic_t rto_count;
6e0534f2 430 struct cpupri cpupri;
57d885fe
GH
431};
432
dc938520
GH
433/*
434 * By default the system creates a single root-domain with all cpus as
435 * members (mimicking the global state we have today).
436 */
57d885fe
GH
437static struct root_domain def_root_domain;
438
ed2d372c 439#endif /* CONFIG_SMP */
57d885fe 440
1da177e4
LT
441/*
442 * This is the main, per-CPU runqueue data structure.
443 *
444 * Locking rule: those places that want to lock multiple runqueues
445 * (such as the load balancing or the thread migration code), lock
446 * acquire operations must be ordered by ascending &runqueue.
447 */
70b97a7f 448struct rq {
d8016491 449 /* runqueue lock: */
05fa785c 450 raw_spinlock_t lock;
1da177e4
LT
451
452 /*
453 * nr_running and cpu_load should be in the same cacheline because
454 * remote CPUs use both these fields when doing load calculation.
455 */
456 unsigned long nr_running;
6aa645ea
IM
457 #define CPU_LOAD_IDX_MAX 5
458 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
fdf3e95d 459 unsigned long last_load_update_tick;
46cb4b7c 460#ifdef CONFIG_NO_HZ
39c0cbe2 461 u64 nohz_stamp;
83cd4fe2 462 unsigned char nohz_balance_kick;
46cb4b7c 463#endif
a64692a3
MG
464 unsigned int skip_clock_update;
465
d8016491
IM
466 /* capture load from *all* tasks on this cpu: */
467 struct load_weight load;
6aa645ea
IM
468 unsigned long nr_load_updates;
469 u64 nr_switches;
470
471 struct cfs_rq cfs;
6f505b16 472 struct rt_rq rt;
6f505b16 473
6aa645ea 474#ifdef CONFIG_FAIR_GROUP_SCHED
d8016491
IM
475 /* list of leaf cfs_rq on this cpu: */
476 struct list_head leaf_cfs_rq_list;
052f1dc7
PZ
477#endif
478#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 479 struct list_head leaf_rt_rq_list;
1da177e4 480#endif
1da177e4
LT
481
482 /*
483 * This is part of a global counter where only the total sum
484 * over all CPUs matters. A task can increase this counter on
485 * one CPU and if it got migrated afterwards it may decrease
486 * it on another CPU. Always updated under the runqueue lock:
487 */
488 unsigned long nr_uninterruptible;
489
34f971f6 490 struct task_struct *curr, *idle, *stop;
c9819f45 491 unsigned long next_balance;
1da177e4 492 struct mm_struct *prev_mm;
6aa645ea 493
3e51f33f 494 u64 clock;
305e6835 495 u64 clock_task;
6aa645ea 496
1da177e4
LT
497 atomic_t nr_iowait;
498
499#ifdef CONFIG_SMP
0eab9146 500 struct root_domain *rd;
1da177e4
LT
501 struct sched_domain *sd;
502
e51fd5e2
PZ
503 unsigned long cpu_power;
504
a0a522ce 505 unsigned char idle_at_tick;
1da177e4 506 /* For active balancing */
3f029d3c 507 int post_schedule;
1da177e4
LT
508 int active_balance;
509 int push_cpu;
969c7921 510 struct cpu_stop_work active_balance_work;
d8016491
IM
511 /* cpu of this runqueue: */
512 int cpu;
1f11eb6a 513 int online;
1da177e4 514
a8a51d5e 515 unsigned long avg_load_per_task;
1da177e4 516
e9e9250b
PZ
517 u64 rt_avg;
518 u64 age_stamp;
1b9508f6
MG
519 u64 idle_stamp;
520 u64 avg_idle;
1da177e4
LT
521#endif
522
aa483808
VP
523#ifdef CONFIG_IRQ_TIME_ACCOUNTING
524 u64 prev_irq_time;
525#endif
526
dce48a84
TG
527 /* calc_load related fields */
528 unsigned long calc_load_update;
529 long calc_load_active;
530
8f4d37ec 531#ifdef CONFIG_SCHED_HRTICK
31656519
PZ
532#ifdef CONFIG_SMP
533 int hrtick_csd_pending;
534 struct call_single_data hrtick_csd;
535#endif
8f4d37ec
PZ
536 struct hrtimer hrtick_timer;
537#endif
538
1da177e4
LT
539#ifdef CONFIG_SCHEDSTATS
540 /* latency stats */
541 struct sched_info rq_sched_info;
9c2c4802
KC
542 unsigned long long rq_cpu_time;
543 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1da177e4
LT
544
545 /* sys_sched_yield() stats */
480b9434 546 unsigned int yld_count;
1da177e4
LT
547
548 /* schedule() stats */
480b9434
KC
549 unsigned int sched_switch;
550 unsigned int sched_count;
551 unsigned int sched_goidle;
1da177e4
LT
552
553 /* try_to_wake_up() stats */
480b9434
KC
554 unsigned int ttwu_count;
555 unsigned int ttwu_local;
b8efb561
IM
556
557 /* BKL stats */
480b9434 558 unsigned int bkl_count;
1da177e4
LT
559#endif
560};
561
f34e3b61 562static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1da177e4 563
a64692a3 564
1e5a7405 565static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags);
dd41f596 566
0a2966b4
CL
567static inline int cpu_of(struct rq *rq)
568{
569#ifdef CONFIG_SMP
570 return rq->cpu;
571#else
572 return 0;
573#endif
574}
575
497f0ab3 576#define rcu_dereference_check_sched_domain(p) \
d11c563d
PM
577 rcu_dereference_check((p), \
578 rcu_read_lock_sched_held() || \
579 lockdep_is_held(&sched_domains_mutex))
580
674311d5
NP
581/*
582 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1a20ff27 583 * See detach_destroy_domains: synchronize_sched for details.
674311d5
NP
584 *
585 * The domain tree of any CPU may only be accessed from within
586 * preempt-disabled sections.
587 */
48f24c4d 588#define for_each_domain(cpu, __sd) \
497f0ab3 589 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
1da177e4
LT
590
591#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
592#define this_rq() (&__get_cpu_var(runqueues))
593#define task_rq(p) cpu_rq(task_cpu(p))
594#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
54d35f29 595#define raw_rq() (&__raw_get_cpu_var(runqueues))
1da177e4 596
dc61b1d6
PZ
597#ifdef CONFIG_CGROUP_SCHED
598
599/*
600 * Return the group to which this tasks belongs.
601 *
602 * We use task_subsys_state_check() and extend the RCU verification
603 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
604 * holds that lock for each task it moves into the cgroup. Therefore
605 * by holding that lock, we pin the task to the current cgroup.
606 */
607static inline struct task_group *task_group(struct task_struct *p)
608{
5091faa4 609 struct task_group *tg;
dc61b1d6
PZ
610 struct cgroup_subsys_state *css;
611
612 css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
613 lockdep_is_held(&task_rq(p)->lock));
5091faa4
MG
614 tg = container_of(css, struct task_group, css);
615
616 return autogroup_task_group(p, tg);
dc61b1d6
PZ
617}
618
619/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
620static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
621{
622#ifdef CONFIG_FAIR_GROUP_SCHED
623 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
624 p->se.parent = task_group(p)->se[cpu];
625#endif
626
627#ifdef CONFIG_RT_GROUP_SCHED
628 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
629 p->rt.parent = task_group(p)->rt_se[cpu];
630#endif
631}
632
633#else /* CONFIG_CGROUP_SCHED */
634
635static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
636static inline struct task_group *task_group(struct task_struct *p)
637{
638 return NULL;
639}
640
641#endif /* CONFIG_CGROUP_SCHED */
642
fe44d621 643static void update_rq_clock_task(struct rq *rq, s64 delta);
305e6835 644
fe44d621 645static void update_rq_clock(struct rq *rq)
3e51f33f 646{
fe44d621 647 s64 delta;
305e6835 648
f26f9aff
MG
649 if (rq->skip_clock_update)
650 return;
aa483808 651
fe44d621
PZ
652 delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
653 rq->clock += delta;
654 update_rq_clock_task(rq, delta);
3e51f33f
PZ
655}
656
bf5c91ba
IM
657/*
658 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
659 */
660#ifdef CONFIG_SCHED_DEBUG
661# define const_debug __read_mostly
662#else
663# define const_debug static const
664#endif
665
017730c1
IM
666/**
667 * runqueue_is_locked
e17b38bf 668 * @cpu: the processor in question.
017730c1
IM
669 *
670 * Returns true if the current cpu runqueue is locked.
671 * This interface allows printk to be called with the runqueue lock
672 * held and know whether or not it is OK to wake up the klogd.
673 */
89f19f04 674int runqueue_is_locked(int cpu)
017730c1 675{
05fa785c 676 return raw_spin_is_locked(&cpu_rq(cpu)->lock);
017730c1
IM
677}
678
bf5c91ba
IM
679/*
680 * Debugging: various feature bits
681 */
f00b45c1
PZ
682
683#define SCHED_FEAT(name, enabled) \
684 __SCHED_FEAT_##name ,
685
bf5c91ba 686enum {
f00b45c1 687#include "sched_features.h"
bf5c91ba
IM
688};
689
f00b45c1
PZ
690#undef SCHED_FEAT
691
692#define SCHED_FEAT(name, enabled) \
693 (1UL << __SCHED_FEAT_##name) * enabled |
694
bf5c91ba 695const_debug unsigned int sysctl_sched_features =
f00b45c1
PZ
696#include "sched_features.h"
697 0;
698
699#undef SCHED_FEAT
700
701#ifdef CONFIG_SCHED_DEBUG
702#define SCHED_FEAT(name, enabled) \
703 #name ,
704
983ed7a6 705static __read_mostly char *sched_feat_names[] = {
f00b45c1
PZ
706#include "sched_features.h"
707 NULL
708};
709
710#undef SCHED_FEAT
711
34f3a814 712static int sched_feat_show(struct seq_file *m, void *v)
f00b45c1 713{
f00b45c1
PZ
714 int i;
715
716 for (i = 0; sched_feat_names[i]; i++) {
34f3a814
LZ
717 if (!(sysctl_sched_features & (1UL << i)))
718 seq_puts(m, "NO_");
719 seq_printf(m, "%s ", sched_feat_names[i]);
f00b45c1 720 }
34f3a814 721 seq_puts(m, "\n");
f00b45c1 722
34f3a814 723 return 0;
f00b45c1
PZ
724}
725
726static ssize_t
727sched_feat_write(struct file *filp, const char __user *ubuf,
728 size_t cnt, loff_t *ppos)
729{
730 char buf[64];
7740191c 731 char *cmp;
f00b45c1
PZ
732 int neg = 0;
733 int i;
734
735 if (cnt > 63)
736 cnt = 63;
737
738 if (copy_from_user(&buf, ubuf, cnt))
739 return -EFAULT;
740
741 buf[cnt] = 0;
7740191c 742 cmp = strstrip(buf);
f00b45c1 743
c24b7c52 744 if (strncmp(buf, "NO_", 3) == 0) {
f00b45c1
PZ
745 neg = 1;
746 cmp += 3;
747 }
748
749 for (i = 0; sched_feat_names[i]; i++) {
7740191c 750 if (strcmp(cmp, sched_feat_names[i]) == 0) {
f00b45c1
PZ
751 if (neg)
752 sysctl_sched_features &= ~(1UL << i);
753 else
754 sysctl_sched_features |= (1UL << i);
755 break;
756 }
757 }
758
759 if (!sched_feat_names[i])
760 return -EINVAL;
761
42994724 762 *ppos += cnt;
f00b45c1
PZ
763
764 return cnt;
765}
766
34f3a814
LZ
767static int sched_feat_open(struct inode *inode, struct file *filp)
768{
769 return single_open(filp, sched_feat_show, NULL);
770}
771
828c0950 772static const struct file_operations sched_feat_fops = {
34f3a814
LZ
773 .open = sched_feat_open,
774 .write = sched_feat_write,
775 .read = seq_read,
776 .llseek = seq_lseek,
777 .release = single_release,
f00b45c1
PZ
778};
779
780static __init int sched_init_debug(void)
781{
f00b45c1
PZ
782 debugfs_create_file("sched_features", 0644, NULL, NULL,
783 &sched_feat_fops);
784
785 return 0;
786}
787late_initcall(sched_init_debug);
788
789#endif
790
791#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
bf5c91ba 792
b82d9fdd
PZ
793/*
794 * Number of tasks to iterate in a single balance run.
795 * Limited because this is done with IRQs disabled.
796 */
797const_debug unsigned int sysctl_sched_nr_migrate = 32;
798
e9e9250b
PZ
799/*
800 * period over which we average the RT time consumption, measured
801 * in ms.
802 *
803 * default: 1s
804 */
805const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
806
fa85ae24 807/*
9f0c1e56 808 * period over which we measure -rt task cpu usage in us.
fa85ae24
PZ
809 * default: 1s
810 */
9f0c1e56 811unsigned int sysctl_sched_rt_period = 1000000;
fa85ae24 812
6892b75e
IM
813static __read_mostly int scheduler_running;
814
9f0c1e56
PZ
815/*
816 * part of the period that we allow rt tasks to run in us.
817 * default: 0.95s
818 */
819int sysctl_sched_rt_runtime = 950000;
fa85ae24 820
d0b27fa7
PZ
821static inline u64 global_rt_period(void)
822{
823 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
824}
825
826static inline u64 global_rt_runtime(void)
827{
e26873bb 828 if (sysctl_sched_rt_runtime < 0)
d0b27fa7
PZ
829 return RUNTIME_INF;
830
831 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
832}
fa85ae24 833
1da177e4 834#ifndef prepare_arch_switch
4866cde0
NP
835# define prepare_arch_switch(next) do { } while (0)
836#endif
837#ifndef finish_arch_switch
838# define finish_arch_switch(prev) do { } while (0)
839#endif
840
051a1d1a
DA
841static inline int task_current(struct rq *rq, struct task_struct *p)
842{
843 return rq->curr == p;
844}
845
4866cde0 846#ifndef __ARCH_WANT_UNLOCKED_CTXSW
70b97a7f 847static inline int task_running(struct rq *rq, struct task_struct *p)
4866cde0 848{
051a1d1a 849 return task_current(rq, p);
4866cde0
NP
850}
851
70b97a7f 852static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
4866cde0
NP
853{
854}
855
70b97a7f 856static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
4866cde0 857{
da04c035
IM
858#ifdef CONFIG_DEBUG_SPINLOCK
859 /* this is a valid case when another task releases the spinlock */
860 rq->lock.owner = current;
861#endif
8a25d5de
IM
862 /*
863 * If we are tracking spinlock dependencies then we have to
864 * fix up the runqueue lock - which gets 'carried over' from
865 * prev into current:
866 */
867 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
868
05fa785c 869 raw_spin_unlock_irq(&rq->lock);
4866cde0
NP
870}
871
872#else /* __ARCH_WANT_UNLOCKED_CTXSW */
70b97a7f 873static inline int task_running(struct rq *rq, struct task_struct *p)
4866cde0
NP
874{
875#ifdef CONFIG_SMP
876 return p->oncpu;
877#else
051a1d1a 878 return task_current(rq, p);
4866cde0
NP
879#endif
880}
881
70b97a7f 882static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
4866cde0
NP
883{
884#ifdef CONFIG_SMP
885 /*
886 * We can optimise this out completely for !SMP, because the
887 * SMP rebalancing from interrupt is the only thing that cares
888 * here.
889 */
890 next->oncpu = 1;
891#endif
892#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
05fa785c 893 raw_spin_unlock_irq(&rq->lock);
4866cde0 894#else
05fa785c 895 raw_spin_unlock(&rq->lock);
4866cde0
NP
896#endif
897}
898
70b97a7f 899static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
4866cde0
NP
900{
901#ifdef CONFIG_SMP
902 /*
903 * After ->oncpu is cleared, the task can be moved to a different CPU.
904 * We must ensure this doesn't happen until the switch is completely
905 * finished.
906 */
907 smp_wmb();
908 prev->oncpu = 0;
909#endif
910#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
911 local_irq_enable();
1da177e4 912#endif
4866cde0
NP
913}
914#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
1da177e4 915
0970d299 916/*
65cc8e48
PZ
917 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
918 * against ttwu().
0970d299
PZ
919 */
920static inline int task_is_waking(struct task_struct *p)
921{
0017d735 922 return unlikely(p->state == TASK_WAKING);
0970d299
PZ
923}
924
b29739f9
IM
925/*
926 * __task_rq_lock - lock the runqueue a given task resides on.
927 * Must be called interrupts disabled.
928 */
70b97a7f 929static inline struct rq *__task_rq_lock(struct task_struct *p)
b29739f9
IM
930 __acquires(rq->lock)
931{
0970d299
PZ
932 struct rq *rq;
933
3a5c359a 934 for (;;) {
0970d299 935 rq = task_rq(p);
05fa785c 936 raw_spin_lock(&rq->lock);
65cc8e48 937 if (likely(rq == task_rq(p)))
3a5c359a 938 return rq;
05fa785c 939 raw_spin_unlock(&rq->lock);
b29739f9 940 }
b29739f9
IM
941}
942
1da177e4
LT
943/*
944 * task_rq_lock - lock the runqueue a given task resides on and disable
41a2d6cf 945 * interrupts. Note the ordering: we can safely lookup the task_rq without
1da177e4
LT
946 * explicitly disabling preemption.
947 */
70b97a7f 948static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1da177e4
LT
949 __acquires(rq->lock)
950{
70b97a7f 951 struct rq *rq;
1da177e4 952
3a5c359a
AK
953 for (;;) {
954 local_irq_save(*flags);
955 rq = task_rq(p);
05fa785c 956 raw_spin_lock(&rq->lock);
65cc8e48 957 if (likely(rq == task_rq(p)))
3a5c359a 958 return rq;
05fa785c 959 raw_spin_unlock_irqrestore(&rq->lock, *flags);
1da177e4 960 }
1da177e4
LT
961}
962
a9957449 963static void __task_rq_unlock(struct rq *rq)
b29739f9
IM
964 __releases(rq->lock)
965{
05fa785c 966 raw_spin_unlock(&rq->lock);
b29739f9
IM
967}
968
70b97a7f 969static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
1da177e4
LT
970 __releases(rq->lock)
971{
05fa785c 972 raw_spin_unlock_irqrestore(&rq->lock, *flags);
1da177e4
LT
973}
974
1da177e4 975/*
cc2a73b5 976 * this_rq_lock - lock this runqueue and disable interrupts.
1da177e4 977 */
a9957449 978static struct rq *this_rq_lock(void)
1da177e4
LT
979 __acquires(rq->lock)
980{
70b97a7f 981 struct rq *rq;
1da177e4
LT
982
983 local_irq_disable();
984 rq = this_rq();
05fa785c 985 raw_spin_lock(&rq->lock);
1da177e4
LT
986
987 return rq;
988}
989
8f4d37ec
PZ
990#ifdef CONFIG_SCHED_HRTICK
991/*
992 * Use HR-timers to deliver accurate preemption points.
993 *
994 * Its all a bit involved since we cannot program an hrt while holding the
995 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
996 * reschedule event.
997 *
998 * When we get rescheduled we reprogram the hrtick_timer outside of the
999 * rq->lock.
1000 */
8f4d37ec
PZ
1001
1002/*
1003 * Use hrtick when:
1004 * - enabled by features
1005 * - hrtimer is actually high res
1006 */
1007static inline int hrtick_enabled(struct rq *rq)
1008{
1009 if (!sched_feat(HRTICK))
1010 return 0;
ba42059f 1011 if (!cpu_active(cpu_of(rq)))
b328ca18 1012 return 0;
8f4d37ec
PZ
1013 return hrtimer_is_hres_active(&rq->hrtick_timer);
1014}
1015
8f4d37ec
PZ
1016static void hrtick_clear(struct rq *rq)
1017{
1018 if (hrtimer_active(&rq->hrtick_timer))
1019 hrtimer_cancel(&rq->hrtick_timer);
1020}
1021
8f4d37ec
PZ
1022/*
1023 * High-resolution timer tick.
1024 * Runs from hardirq context with interrupts disabled.
1025 */
1026static enum hrtimer_restart hrtick(struct hrtimer *timer)
1027{
1028 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1029
1030 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1031
05fa785c 1032 raw_spin_lock(&rq->lock);
3e51f33f 1033 update_rq_clock(rq);
8f4d37ec 1034 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
05fa785c 1035 raw_spin_unlock(&rq->lock);
8f4d37ec
PZ
1036
1037 return HRTIMER_NORESTART;
1038}
1039
95e904c7 1040#ifdef CONFIG_SMP
31656519
PZ
1041/*
1042 * called from hardirq (IPI) context
1043 */
1044static void __hrtick_start(void *arg)
b328ca18 1045{
31656519 1046 struct rq *rq = arg;
b328ca18 1047
05fa785c 1048 raw_spin_lock(&rq->lock);
31656519
PZ
1049 hrtimer_restart(&rq->hrtick_timer);
1050 rq->hrtick_csd_pending = 0;
05fa785c 1051 raw_spin_unlock(&rq->lock);
b328ca18
PZ
1052}
1053
31656519
PZ
1054/*
1055 * Called to set the hrtick timer state.
1056 *
1057 * called with rq->lock held and irqs disabled
1058 */
1059static void hrtick_start(struct rq *rq, u64 delay)
b328ca18 1060{
31656519
PZ
1061 struct hrtimer *timer = &rq->hrtick_timer;
1062 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
b328ca18 1063
cc584b21 1064 hrtimer_set_expires(timer, time);
31656519
PZ
1065
1066 if (rq == this_rq()) {
1067 hrtimer_restart(timer);
1068 } else if (!rq->hrtick_csd_pending) {
6e275637 1069 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
31656519
PZ
1070 rq->hrtick_csd_pending = 1;
1071 }
b328ca18
PZ
1072}
1073
1074static int
1075hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1076{
1077 int cpu = (int)(long)hcpu;
1078
1079 switch (action) {
1080 case CPU_UP_CANCELED:
1081 case CPU_UP_CANCELED_FROZEN:
1082 case CPU_DOWN_PREPARE:
1083 case CPU_DOWN_PREPARE_FROZEN:
1084 case CPU_DEAD:
1085 case CPU_DEAD_FROZEN:
31656519 1086 hrtick_clear(cpu_rq(cpu));
b328ca18
PZ
1087 return NOTIFY_OK;
1088 }
1089
1090 return NOTIFY_DONE;
1091}
1092
fa748203 1093static __init void init_hrtick(void)
b328ca18
PZ
1094{
1095 hotcpu_notifier(hotplug_hrtick, 0);
1096}
31656519
PZ
1097#else
1098/*
1099 * Called to set the hrtick timer state.
1100 *
1101 * called with rq->lock held and irqs disabled
1102 */
1103static void hrtick_start(struct rq *rq, u64 delay)
1104{
7f1e2ca9 1105 __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
5c333864 1106 HRTIMER_MODE_REL_PINNED, 0);
31656519 1107}
b328ca18 1108
006c75f1 1109static inline void init_hrtick(void)
8f4d37ec 1110{
8f4d37ec 1111}
31656519 1112#endif /* CONFIG_SMP */
8f4d37ec 1113
31656519 1114static void init_rq_hrtick(struct rq *rq)
8f4d37ec 1115{
31656519
PZ
1116#ifdef CONFIG_SMP
1117 rq->hrtick_csd_pending = 0;
8f4d37ec 1118
31656519
PZ
1119 rq->hrtick_csd.flags = 0;
1120 rq->hrtick_csd.func = __hrtick_start;
1121 rq->hrtick_csd.info = rq;
1122#endif
8f4d37ec 1123
31656519
PZ
1124 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1125 rq->hrtick_timer.function = hrtick;
8f4d37ec 1126}
006c75f1 1127#else /* CONFIG_SCHED_HRTICK */
8f4d37ec
PZ
1128static inline void hrtick_clear(struct rq *rq)
1129{
1130}
1131
8f4d37ec
PZ
1132static inline void init_rq_hrtick(struct rq *rq)
1133{
1134}
1135
b328ca18
PZ
1136static inline void init_hrtick(void)
1137{
1138}
006c75f1 1139#endif /* CONFIG_SCHED_HRTICK */
8f4d37ec 1140
c24d20db
IM
1141/*
1142 * resched_task - mark a task 'to be rescheduled now'.
1143 *
1144 * On UP this means the setting of the need_resched flag, on SMP it
1145 * might also involve a cross-CPU call to trigger the scheduler on
1146 * the target CPU.
1147 */
1148#ifdef CONFIG_SMP
1149
1150#ifndef tsk_is_polling
1151#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1152#endif
1153
31656519 1154static void resched_task(struct task_struct *p)
c24d20db
IM
1155{
1156 int cpu;
1157
05fa785c 1158 assert_raw_spin_locked(&task_rq(p)->lock);
c24d20db 1159
5ed0cec0 1160 if (test_tsk_need_resched(p))
c24d20db
IM
1161 return;
1162
5ed0cec0 1163 set_tsk_need_resched(p);
c24d20db
IM
1164
1165 cpu = task_cpu(p);
1166 if (cpu == smp_processor_id())
1167 return;
1168
1169 /* NEED_RESCHED must be visible before we test polling */
1170 smp_mb();
1171 if (!tsk_is_polling(p))
1172 smp_send_reschedule(cpu);
1173}
1174
1175static void resched_cpu(int cpu)
1176{
1177 struct rq *rq = cpu_rq(cpu);
1178 unsigned long flags;
1179
05fa785c 1180 if (!raw_spin_trylock_irqsave(&rq->lock, flags))
c24d20db
IM
1181 return;
1182 resched_task(cpu_curr(cpu));
05fa785c 1183 raw_spin_unlock_irqrestore(&rq->lock, flags);
c24d20db 1184}
06d8308c
TG
1185
1186#ifdef CONFIG_NO_HZ
83cd4fe2
VP
1187/*
1188 * In the semi idle case, use the nearest busy cpu for migrating timers
1189 * from an idle cpu. This is good for power-savings.
1190 *
1191 * We don't do similar optimization for completely idle system, as
1192 * selecting an idle cpu will add more delays to the timers than intended
1193 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1194 */
1195int get_nohz_timer_target(void)
1196{
1197 int cpu = smp_processor_id();
1198 int i;
1199 struct sched_domain *sd;
1200
1201 for_each_domain(cpu, sd) {
1202 for_each_cpu(i, sched_domain_span(sd))
1203 if (!idle_cpu(i))
1204 return i;
1205 }
1206 return cpu;
1207}
06d8308c
TG
1208/*
1209 * When add_timer_on() enqueues a timer into the timer wheel of an
1210 * idle CPU then this timer might expire before the next timer event
1211 * which is scheduled to wake up that CPU. In case of a completely
1212 * idle system the next event might even be infinite time into the
1213 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1214 * leaves the inner idle loop so the newly added timer is taken into
1215 * account when the CPU goes back to idle and evaluates the timer
1216 * wheel for the next timer event.
1217 */
1218void wake_up_idle_cpu(int cpu)
1219{
1220 struct rq *rq = cpu_rq(cpu);
1221
1222 if (cpu == smp_processor_id())
1223 return;
1224
1225 /*
1226 * This is safe, as this function is called with the timer
1227 * wheel base lock of (cpu) held. When the CPU is on the way
1228 * to idle and has not yet set rq->curr to idle then it will
1229 * be serialized on the timer wheel base lock and take the new
1230 * timer into account automatically.
1231 */
1232 if (rq->curr != rq->idle)
1233 return;
1234
1235 /*
1236 * We can set TIF_RESCHED on the idle task of the other CPU
1237 * lockless. The worst case is that the other CPU runs the
1238 * idle task through an additional NOOP schedule()
1239 */
5ed0cec0 1240 set_tsk_need_resched(rq->idle);
06d8308c
TG
1241
1242 /* NEED_RESCHED must be visible before we test polling */
1243 smp_mb();
1244 if (!tsk_is_polling(rq->idle))
1245 smp_send_reschedule(cpu);
1246}
39c0cbe2 1247
6d6bc0ad 1248#endif /* CONFIG_NO_HZ */
06d8308c 1249
e9e9250b
PZ
1250static u64 sched_avg_period(void)
1251{
1252 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1253}
1254
1255static void sched_avg_update(struct rq *rq)
1256{
1257 s64 period = sched_avg_period();
1258
1259 while ((s64)(rq->clock - rq->age_stamp) > period) {
0d98bb26
WD
1260 /*
1261 * Inline assembly required to prevent the compiler
1262 * optimising this loop into a divmod call.
1263 * See __iter_div_u64_rem() for another example of this.
1264 */
1265 asm("" : "+rm" (rq->age_stamp));
e9e9250b
PZ
1266 rq->age_stamp += period;
1267 rq->rt_avg /= 2;
1268 }
1269}
1270
1271static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1272{
1273 rq->rt_avg += rt_delta;
1274 sched_avg_update(rq);
1275}
1276
6d6bc0ad 1277#else /* !CONFIG_SMP */
31656519 1278static void resched_task(struct task_struct *p)
c24d20db 1279{
05fa785c 1280 assert_raw_spin_locked(&task_rq(p)->lock);
31656519 1281 set_tsk_need_resched(p);
c24d20db 1282}
e9e9250b
PZ
1283
1284static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1285{
1286}
da2b71ed
SS
1287
1288static void sched_avg_update(struct rq *rq)
1289{
1290}
6d6bc0ad 1291#endif /* CONFIG_SMP */
c24d20db 1292
45bf76df
IM
1293#if BITS_PER_LONG == 32
1294# define WMULT_CONST (~0UL)
1295#else
1296# define WMULT_CONST (1UL << 32)
1297#endif
1298
1299#define WMULT_SHIFT 32
1300
194081eb
IM
1301/*
1302 * Shift right and round:
1303 */
cf2ab469 1304#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
194081eb 1305
a7be37ac
PZ
1306/*
1307 * delta *= weight / lw
1308 */
cb1c4fc9 1309static unsigned long
45bf76df
IM
1310calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1311 struct load_weight *lw)
1312{
1313 u64 tmp;
1314
7a232e03
LJ
1315 if (!lw->inv_weight) {
1316 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1317 lw->inv_weight = 1;
1318 else
1319 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1320 / (lw->weight+1);
1321 }
45bf76df
IM
1322
1323 tmp = (u64)delta_exec * weight;
1324 /*
1325 * Check whether we'd overflow the 64-bit multiplication:
1326 */
194081eb 1327 if (unlikely(tmp > WMULT_CONST))
cf2ab469 1328 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
194081eb
IM
1329 WMULT_SHIFT/2);
1330 else
cf2ab469 1331 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
45bf76df 1332
ecf691da 1333 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
45bf76df
IM
1334}
1335
1091985b 1336static inline void update_load_add(struct load_weight *lw, unsigned long inc)
45bf76df
IM
1337{
1338 lw->weight += inc;
e89996ae 1339 lw->inv_weight = 0;
45bf76df
IM
1340}
1341
1091985b 1342static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
45bf76df
IM
1343{
1344 lw->weight -= dec;
e89996ae 1345 lw->inv_weight = 0;
45bf76df
IM
1346}
1347
2069dd75
PZ
1348static inline void update_load_set(struct load_weight *lw, unsigned long w)
1349{
1350 lw->weight = w;
1351 lw->inv_weight = 0;
1352}
1353
2dd73a4f
PW
1354/*
1355 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1356 * of tasks with abnormal "nice" values across CPUs the contribution that
1357 * each task makes to its run queue's load is weighted according to its
41a2d6cf 1358 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2dd73a4f
PW
1359 * scaled version of the new time slice allocation that they receive on time
1360 * slice expiry etc.
1361 */
1362
cce7ade8
PZ
1363#define WEIGHT_IDLEPRIO 3
1364#define WMULT_IDLEPRIO 1431655765
dd41f596
IM
1365
1366/*
1367 * Nice levels are multiplicative, with a gentle 10% change for every
1368 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1369 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1370 * that remained on nice 0.
1371 *
1372 * The "10% effect" is relative and cumulative: from _any_ nice level,
1373 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
f9153ee6
IM
1374 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1375 * If a task goes up by ~10% and another task goes down by ~10% then
1376 * the relative distance between them is ~25%.)
dd41f596
IM
1377 */
1378static const int prio_to_weight[40] = {
254753dc
IM
1379 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1380 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1381 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1382 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1383 /* 0 */ 1024, 820, 655, 526, 423,
1384 /* 5 */ 335, 272, 215, 172, 137,
1385 /* 10 */ 110, 87, 70, 56, 45,
1386 /* 15 */ 36, 29, 23, 18, 15,
dd41f596
IM
1387};
1388
5714d2de
IM
1389/*
1390 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1391 *
1392 * In cases where the weight does not change often, we can use the
1393 * precalculated inverse to speed up arithmetics by turning divisions
1394 * into multiplications:
1395 */
dd41f596 1396static const u32 prio_to_wmult[40] = {
254753dc
IM
1397 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1398 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1399 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1400 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1401 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1402 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1403 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1404 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
dd41f596 1405};
2dd73a4f 1406
ef12fefa
BR
1407/* Time spent by the tasks of the cpu accounting group executing in ... */
1408enum cpuacct_stat_index {
1409 CPUACCT_STAT_USER, /* ... user mode */
1410 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1411
1412 CPUACCT_STAT_NSTATS,
1413};
1414
d842de87
SV
1415#ifdef CONFIG_CGROUP_CPUACCT
1416static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
ef12fefa
BR
1417static void cpuacct_update_stats(struct task_struct *tsk,
1418 enum cpuacct_stat_index idx, cputime_t val);
d842de87
SV
1419#else
1420static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
ef12fefa
BR
1421static inline void cpuacct_update_stats(struct task_struct *tsk,
1422 enum cpuacct_stat_index idx, cputime_t val) {}
d842de87
SV
1423#endif
1424
18d95a28
PZ
1425static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1426{
1427 update_load_add(&rq->load, load);
1428}
1429
1430static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1431{
1432 update_load_sub(&rq->load, load);
1433}
1434
7940ca36 1435#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
eb755805 1436typedef int (*tg_visitor)(struct task_group *, void *);
c09595f6
PZ
1437
1438/*
1439 * Iterate the full tree, calling @down when first entering a node and @up when
1440 * leaving it for the final time.
1441 */
eb755805 1442static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
c09595f6
PZ
1443{
1444 struct task_group *parent, *child;
eb755805 1445 int ret;
c09595f6
PZ
1446
1447 rcu_read_lock();
1448 parent = &root_task_group;
1449down:
eb755805
PZ
1450 ret = (*down)(parent, data);
1451 if (ret)
1452 goto out_unlock;
c09595f6
PZ
1453 list_for_each_entry_rcu(child, &parent->children, siblings) {
1454 parent = child;
1455 goto down;
1456
1457up:
1458 continue;
1459 }
eb755805
PZ
1460 ret = (*up)(parent, data);
1461 if (ret)
1462 goto out_unlock;
c09595f6
PZ
1463
1464 child = parent;
1465 parent = parent->parent;
1466 if (parent)
1467 goto up;
eb755805 1468out_unlock:
c09595f6 1469 rcu_read_unlock();
eb755805
PZ
1470
1471 return ret;
c09595f6
PZ
1472}
1473
eb755805
PZ
1474static int tg_nop(struct task_group *tg, void *data)
1475{
1476 return 0;
c09595f6 1477}
eb755805
PZ
1478#endif
1479
1480#ifdef CONFIG_SMP
f5f08f39
PZ
1481/* Used instead of source_load when we know the type == 0 */
1482static unsigned long weighted_cpuload(const int cpu)
1483{
1484 return cpu_rq(cpu)->load.weight;
1485}
1486
1487/*
1488 * Return a low guess at the load of a migration-source cpu weighted
1489 * according to the scheduling class and "nice" value.
1490 *
1491 * We want to under-estimate the load of migration sources, to
1492 * balance conservatively.
1493 */
1494static unsigned long source_load(int cpu, int type)
1495{
1496 struct rq *rq = cpu_rq(cpu);
1497 unsigned long total = weighted_cpuload(cpu);
1498
1499 if (type == 0 || !sched_feat(LB_BIAS))
1500 return total;
1501
1502 return min(rq->cpu_load[type-1], total);
1503}
1504
1505/*
1506 * Return a high guess at the load of a migration-target cpu weighted
1507 * according to the scheduling class and "nice" value.
1508 */
1509static unsigned long target_load(int cpu, int type)
1510{
1511 struct rq *rq = cpu_rq(cpu);
1512 unsigned long total = weighted_cpuload(cpu);
1513
1514 if (type == 0 || !sched_feat(LB_BIAS))
1515 return total;
1516
1517 return max(rq->cpu_load[type-1], total);
1518}
1519
ae154be1
PZ
1520static unsigned long power_of(int cpu)
1521{
e51fd5e2 1522 return cpu_rq(cpu)->cpu_power;
ae154be1
PZ
1523}
1524
eb755805
PZ
1525static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1526
1527static unsigned long cpu_avg_load_per_task(int cpu)
1528{
1529 struct rq *rq = cpu_rq(cpu);
af6d596f 1530 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
eb755805 1531
4cd42620
SR
1532 if (nr_running)
1533 rq->avg_load_per_task = rq->load.weight / nr_running;
a2d47777
BS
1534 else
1535 rq->avg_load_per_task = 0;
eb755805
PZ
1536
1537 return rq->avg_load_per_task;
1538}
1539
1540#ifdef CONFIG_FAIR_GROUP_SCHED
c09595f6 1541
c09595f6 1542/*
c8cba857
PZ
1543 * Compute the cpu's hierarchical load factor for each task group.
1544 * This needs to be done in a top-down fashion because the load of a child
1545 * group is a fraction of its parents load.
c09595f6 1546 */
eb755805 1547static int tg_load_down(struct task_group *tg, void *data)
c09595f6 1548{
c8cba857 1549 unsigned long load;
eb755805 1550 long cpu = (long)data;
c09595f6 1551
c8cba857
PZ
1552 if (!tg->parent) {
1553 load = cpu_rq(cpu)->load.weight;
1554 } else {
1555 load = tg->parent->cfs_rq[cpu]->h_load;
2069dd75 1556 load *= tg->se[cpu]->load.weight;
c8cba857
PZ
1557 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1558 }
c09595f6 1559
c8cba857 1560 tg->cfs_rq[cpu]->h_load = load;
c09595f6 1561
eb755805 1562 return 0;
c09595f6
PZ
1563}
1564
eb755805 1565static void update_h_load(long cpu)
c09595f6 1566{
eb755805 1567 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
c09595f6
PZ
1568}
1569
18d95a28
PZ
1570#endif
1571
8f45e2b5
GH
1572#ifdef CONFIG_PREEMPT
1573
b78bb868
PZ
1574static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1575
70574a99 1576/*
8f45e2b5
GH
1577 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1578 * way at the expense of forcing extra atomic operations in all
1579 * invocations. This assures that the double_lock is acquired using the
1580 * same underlying policy as the spinlock_t on this architecture, which
1581 * reduces latency compared to the unfair variant below. However, it
1582 * also adds more overhead and therefore may reduce throughput.
70574a99 1583 */
8f45e2b5
GH
1584static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1585 __releases(this_rq->lock)
1586 __acquires(busiest->lock)
1587 __acquires(this_rq->lock)
1588{
05fa785c 1589 raw_spin_unlock(&this_rq->lock);
8f45e2b5
GH
1590 double_rq_lock(this_rq, busiest);
1591
1592 return 1;
1593}
1594
1595#else
1596/*
1597 * Unfair double_lock_balance: Optimizes throughput at the expense of
1598 * latency by eliminating extra atomic operations when the locks are
1599 * already in proper order on entry. This favors lower cpu-ids and will
1600 * grant the double lock to lower cpus over higher ids under contention,
1601 * regardless of entry order into the function.
1602 */
1603static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
70574a99
AD
1604 __releases(this_rq->lock)
1605 __acquires(busiest->lock)
1606 __acquires(this_rq->lock)
1607{
1608 int ret = 0;
1609
05fa785c 1610 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
70574a99 1611 if (busiest < this_rq) {
05fa785c
TG
1612 raw_spin_unlock(&this_rq->lock);
1613 raw_spin_lock(&busiest->lock);
1614 raw_spin_lock_nested(&this_rq->lock,
1615 SINGLE_DEPTH_NESTING);
70574a99
AD
1616 ret = 1;
1617 } else
05fa785c
TG
1618 raw_spin_lock_nested(&busiest->lock,
1619 SINGLE_DEPTH_NESTING);
70574a99
AD
1620 }
1621 return ret;
1622}
1623
8f45e2b5
GH
1624#endif /* CONFIG_PREEMPT */
1625
1626/*
1627 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1628 */
1629static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1630{
1631 if (unlikely(!irqs_disabled())) {
1632 /* printk() doesn't work good under rq->lock */
05fa785c 1633 raw_spin_unlock(&this_rq->lock);
8f45e2b5
GH
1634 BUG_ON(1);
1635 }
1636
1637 return _double_lock_balance(this_rq, busiest);
1638}
1639
70574a99
AD
1640static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1641 __releases(busiest->lock)
1642{
05fa785c 1643 raw_spin_unlock(&busiest->lock);
70574a99
AD
1644 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1645}
1e3c88bd
PZ
1646
1647/*
1648 * double_rq_lock - safely lock two runqueues
1649 *
1650 * Note this does not disable interrupts like task_rq_lock,
1651 * you need to do so manually before calling.
1652 */
1653static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1654 __acquires(rq1->lock)
1655 __acquires(rq2->lock)
1656{
1657 BUG_ON(!irqs_disabled());
1658 if (rq1 == rq2) {
1659 raw_spin_lock(&rq1->lock);
1660 __acquire(rq2->lock); /* Fake it out ;) */
1661 } else {
1662 if (rq1 < rq2) {
1663 raw_spin_lock(&rq1->lock);
1664 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1665 } else {
1666 raw_spin_lock(&rq2->lock);
1667 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1668 }
1669 }
1e3c88bd
PZ
1670}
1671
1672/*
1673 * double_rq_unlock - safely unlock two runqueues
1674 *
1675 * Note this does not restore interrupts like task_rq_unlock,
1676 * you need to do so manually after calling.
1677 */
1678static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1679 __releases(rq1->lock)
1680 __releases(rq2->lock)
1681{
1682 raw_spin_unlock(&rq1->lock);
1683 if (rq1 != rq2)
1684 raw_spin_unlock(&rq2->lock);
1685 else
1686 __release(rq2->lock);
1687}
1688
18d95a28
PZ
1689#endif
1690
74f5187a 1691static void calc_load_account_idle(struct rq *this_rq);
0bcdcf28 1692static void update_sysctl(void);
acb4a848 1693static int get_update_sysctl_factor(void);
fdf3e95d 1694static void update_cpu_load(struct rq *this_rq);
dce48a84 1695
cd29fe6f
PZ
1696static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1697{
1698 set_task_rq(p, cpu);
1699#ifdef CONFIG_SMP
1700 /*
1701 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1702 * successfuly executed on another CPU. We must ensure that updates of
1703 * per-task data have been completed by this moment.
1704 */
1705 smp_wmb();
1706 task_thread_info(p)->cpu = cpu;
1707#endif
1708}
dce48a84 1709
1e3c88bd 1710static const struct sched_class rt_sched_class;
dd41f596 1711
34f971f6 1712#define sched_class_highest (&stop_sched_class)
1f11eb6a
GH
1713#define for_each_class(class) \
1714 for (class = sched_class_highest; class; class = class->next)
dd41f596 1715
1e3c88bd
PZ
1716#include "sched_stats.h"
1717
c09595f6 1718static void inc_nr_running(struct rq *rq)
9c217245
IM
1719{
1720 rq->nr_running++;
9c217245
IM
1721}
1722
c09595f6 1723static void dec_nr_running(struct rq *rq)
9c217245
IM
1724{
1725 rq->nr_running--;
9c217245
IM
1726}
1727
45bf76df
IM
1728static void set_load_weight(struct task_struct *p)
1729{
dd41f596
IM
1730 /*
1731 * SCHED_IDLE tasks get minimal weight:
1732 */
1733 if (p->policy == SCHED_IDLE) {
1734 p->se.load.weight = WEIGHT_IDLEPRIO;
1735 p->se.load.inv_weight = WMULT_IDLEPRIO;
1736 return;
1737 }
71f8bd46 1738
dd41f596
IM
1739 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1740 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
71f8bd46
IM
1741}
1742
371fd7e7 1743static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
2087a1ad 1744{
a64692a3 1745 update_rq_clock(rq);
dd41f596 1746 sched_info_queued(p);
371fd7e7 1747 p->sched_class->enqueue_task(rq, p, flags);
dd41f596 1748 p->se.on_rq = 1;
71f8bd46
IM
1749}
1750
371fd7e7 1751static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
71f8bd46 1752{
a64692a3 1753 update_rq_clock(rq);
46ac22ba 1754 sched_info_dequeued(p);
371fd7e7 1755 p->sched_class->dequeue_task(rq, p, flags);
dd41f596 1756 p->se.on_rq = 0;
71f8bd46
IM
1757}
1758
1e3c88bd
PZ
1759/*
1760 * activate_task - move a task to the runqueue.
1761 */
371fd7e7 1762static void activate_task(struct rq *rq, struct task_struct *p, int flags)
1e3c88bd
PZ
1763{
1764 if (task_contributes_to_load(p))
1765 rq->nr_uninterruptible--;
1766
371fd7e7 1767 enqueue_task(rq, p, flags);
1e3c88bd
PZ
1768 inc_nr_running(rq);
1769}
1770
1771/*
1772 * deactivate_task - remove a task from the runqueue.
1773 */
371fd7e7 1774static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1e3c88bd
PZ
1775{
1776 if (task_contributes_to_load(p))
1777 rq->nr_uninterruptible++;
1778
371fd7e7 1779 dequeue_task(rq, p, flags);
1e3c88bd
PZ
1780 dec_nr_running(rq);
1781}
1782
b52bfee4
VP
1783#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1784
305e6835
VP
1785/*
1786 * There are no locks covering percpu hardirq/softirq time.
1787 * They are only modified in account_system_vtime, on corresponding CPU
1788 * with interrupts disabled. So, writes are safe.
1789 * They are read and saved off onto struct rq in update_rq_clock().
1790 * This may result in other CPU reading this CPU's irq time and can
1791 * race with irq/account_system_vtime on this CPU. We would either get old
8e92c201
PZ
1792 * or new value with a side effect of accounting a slice of irq time to wrong
1793 * task when irq is in progress while we read rq->clock. That is a worthy
1794 * compromise in place of having locks on each irq in account_system_time.
305e6835 1795 */
b52bfee4
VP
1796static DEFINE_PER_CPU(u64, cpu_hardirq_time);
1797static DEFINE_PER_CPU(u64, cpu_softirq_time);
1798
1799static DEFINE_PER_CPU(u64, irq_start_time);
1800static int sched_clock_irqtime;
1801
1802void enable_sched_clock_irqtime(void)
1803{
1804 sched_clock_irqtime = 1;
1805}
1806
1807void disable_sched_clock_irqtime(void)
1808{
1809 sched_clock_irqtime = 0;
1810}
1811
8e92c201
PZ
1812#ifndef CONFIG_64BIT
1813static DEFINE_PER_CPU(seqcount_t, irq_time_seq);
1814
1815static inline void irq_time_write_begin(void)
1816{
1817 __this_cpu_inc(irq_time_seq.sequence);
1818 smp_wmb();
1819}
1820
1821static inline void irq_time_write_end(void)
1822{
1823 smp_wmb();
1824 __this_cpu_inc(irq_time_seq.sequence);
1825}
1826
1827static inline u64 irq_time_read(int cpu)
1828{
1829 u64 irq_time;
1830 unsigned seq;
1831
1832 do {
1833 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1834 irq_time = per_cpu(cpu_softirq_time, cpu) +
1835 per_cpu(cpu_hardirq_time, cpu);
1836 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1837
1838 return irq_time;
1839}
1840#else /* CONFIG_64BIT */
1841static inline void irq_time_write_begin(void)
1842{
1843}
1844
1845static inline void irq_time_write_end(void)
1846{
1847}
1848
1849static inline u64 irq_time_read(int cpu)
305e6835 1850{
305e6835
VP
1851 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1852}
8e92c201 1853#endif /* CONFIG_64BIT */
305e6835 1854
fe44d621
PZ
1855/*
1856 * Called before incrementing preempt_count on {soft,}irq_enter
1857 * and before decrementing preempt_count on {soft,}irq_exit.
1858 */
b52bfee4
VP
1859void account_system_vtime(struct task_struct *curr)
1860{
1861 unsigned long flags;
fe44d621 1862 s64 delta;
b52bfee4 1863 int cpu;
b52bfee4
VP
1864
1865 if (!sched_clock_irqtime)
1866 return;
1867
1868 local_irq_save(flags);
1869
b52bfee4 1870 cpu = smp_processor_id();
fe44d621
PZ
1871 delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
1872 __this_cpu_add(irq_start_time, delta);
1873
8e92c201 1874 irq_time_write_begin();
b52bfee4
VP
1875 /*
1876 * We do not account for softirq time from ksoftirqd here.
1877 * We want to continue accounting softirq time to ksoftirqd thread
1878 * in that case, so as not to confuse scheduler with a special task
1879 * that do not consume any time, but still wants to run.
1880 */
1881 if (hardirq_count())
fe44d621 1882 __this_cpu_add(cpu_hardirq_time, delta);
b52bfee4 1883 else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD))
fe44d621 1884 __this_cpu_add(cpu_softirq_time, delta);
b52bfee4 1885
8e92c201 1886 irq_time_write_end();
b52bfee4
VP
1887 local_irq_restore(flags);
1888}
b7dadc38 1889EXPORT_SYMBOL_GPL(account_system_vtime);
b52bfee4 1890
fe44d621 1891static void update_rq_clock_task(struct rq *rq, s64 delta)
aa483808 1892{
fe44d621
PZ
1893 s64 irq_delta;
1894
8e92c201 1895 irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
fe44d621
PZ
1896
1897 /*
1898 * Since irq_time is only updated on {soft,}irq_exit, we might run into
1899 * this case when a previous update_rq_clock() happened inside a
1900 * {soft,}irq region.
1901 *
1902 * When this happens, we stop ->clock_task and only update the
1903 * prev_irq_time stamp to account for the part that fit, so that a next
1904 * update will consume the rest. This ensures ->clock_task is
1905 * monotonic.
1906 *
1907 * It does however cause some slight miss-attribution of {soft,}irq
1908 * time, a more accurate solution would be to update the irq_time using
1909 * the current rq->clock timestamp, except that would require using
1910 * atomic ops.
1911 */
1912 if (irq_delta > delta)
1913 irq_delta = delta;
1914
1915 rq->prev_irq_time += irq_delta;
1916 delta -= irq_delta;
1917 rq->clock_task += delta;
1918
1919 if (irq_delta && sched_feat(NONIRQ_POWER))
1920 sched_rt_avg_update(rq, irq_delta);
aa483808
VP
1921}
1922
fe44d621 1923#else /* CONFIG_IRQ_TIME_ACCOUNTING */
305e6835 1924
fe44d621 1925static void update_rq_clock_task(struct rq *rq, s64 delta)
305e6835 1926{
fe44d621 1927 rq->clock_task += delta;
305e6835
VP
1928}
1929
fe44d621 1930#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
b52bfee4 1931
1e3c88bd
PZ
1932#include "sched_idletask.c"
1933#include "sched_fair.c"
1934#include "sched_rt.c"
5091faa4 1935#include "sched_autogroup.c"
34f971f6 1936#include "sched_stoptask.c"
1e3c88bd
PZ
1937#ifdef CONFIG_SCHED_DEBUG
1938# include "sched_debug.c"
1939#endif
1940
34f971f6
PZ
1941void sched_set_stop_task(int cpu, struct task_struct *stop)
1942{
1943 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
1944 struct task_struct *old_stop = cpu_rq(cpu)->stop;
1945
1946 if (stop) {
1947 /*
1948 * Make it appear like a SCHED_FIFO task, its something
1949 * userspace knows about and won't get confused about.
1950 *
1951 * Also, it will make PI more or less work without too
1952 * much confusion -- but then, stop work should not
1953 * rely on PI working anyway.
1954 */
1955 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
1956
1957 stop->sched_class = &stop_sched_class;
1958 }
1959
1960 cpu_rq(cpu)->stop = stop;
1961
1962 if (old_stop) {
1963 /*
1964 * Reset it back to a normal scheduling class so that
1965 * it can die in pieces.
1966 */
1967 old_stop->sched_class = &rt_sched_class;
1968 }
1969}
1970
14531189 1971/*
dd41f596 1972 * __normal_prio - return the priority that is based on the static prio
14531189 1973 */
14531189
IM
1974static inline int __normal_prio(struct task_struct *p)
1975{
dd41f596 1976 return p->static_prio;
14531189
IM
1977}
1978
b29739f9
IM
1979/*
1980 * Calculate the expected normal priority: i.e. priority
1981 * without taking RT-inheritance into account. Might be
1982 * boosted by interactivity modifiers. Changes upon fork,
1983 * setprio syscalls, and whenever the interactivity
1984 * estimator recalculates.
1985 */
36c8b586 1986static inline int normal_prio(struct task_struct *p)
b29739f9
IM
1987{
1988 int prio;
1989
e05606d3 1990 if (task_has_rt_policy(p))
b29739f9
IM
1991 prio = MAX_RT_PRIO-1 - p->rt_priority;
1992 else
1993 prio = __normal_prio(p);
1994 return prio;
1995}
1996
1997/*
1998 * Calculate the current priority, i.e. the priority
1999 * taken into account by the scheduler. This value might
2000 * be boosted by RT tasks, or might be boosted by
2001 * interactivity modifiers. Will be RT if the task got
2002 * RT-boosted. If not then it returns p->normal_prio.
2003 */
36c8b586 2004static int effective_prio(struct task_struct *p)
b29739f9
IM
2005{
2006 p->normal_prio = normal_prio(p);
2007 /*
2008 * If we are RT tasks or we were boosted to RT priority,
2009 * keep the priority unchanged. Otherwise, update priority
2010 * to the normal priority:
2011 */
2012 if (!rt_prio(p->prio))
2013 return p->normal_prio;
2014 return p->prio;
2015}
2016
1da177e4
LT
2017/**
2018 * task_curr - is this task currently executing on a CPU?
2019 * @p: the task in question.
2020 */
36c8b586 2021inline int task_curr(const struct task_struct *p)
1da177e4
LT
2022{
2023 return cpu_curr(task_cpu(p)) == p;
2024}
2025
cb469845
SR
2026static inline void check_class_changed(struct rq *rq, struct task_struct *p,
2027 const struct sched_class *prev_class,
2028 int oldprio, int running)
2029{
2030 if (prev_class != p->sched_class) {
2031 if (prev_class->switched_from)
2032 prev_class->switched_from(rq, p, running);
2033 p->sched_class->switched_to(rq, p, running);
2034 } else
2035 p->sched_class->prio_changed(rq, p, oldprio, running);
2036}
2037
1e5a7405
PZ
2038static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
2039{
2040 const struct sched_class *class;
2041
2042 if (p->sched_class == rq->curr->sched_class) {
2043 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
2044 } else {
2045 for_each_class(class) {
2046 if (class == rq->curr->sched_class)
2047 break;
2048 if (class == p->sched_class) {
2049 resched_task(rq->curr);
2050 break;
2051 }
2052 }
2053 }
2054
2055 /*
2056 * A queue event has occurred, and we're going to schedule. In
2057 * this case, we can save a useless back to back clock update.
2058 */
f26f9aff 2059 if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr))
1e5a7405
PZ
2060 rq->skip_clock_update = 1;
2061}
2062
1da177e4 2063#ifdef CONFIG_SMP
cc367732
IM
2064/*
2065 * Is this task likely cache-hot:
2066 */
e7693a36 2067static int
cc367732
IM
2068task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2069{
2070 s64 delta;
2071
e6c8fba7
PZ
2072 if (p->sched_class != &fair_sched_class)
2073 return 0;
2074
ef8002f6
NR
2075 if (unlikely(p->policy == SCHED_IDLE))
2076 return 0;
2077
f540a608
IM
2078 /*
2079 * Buddy candidates are cache hot:
2080 */
f685ceac 2081 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
4793241b
PZ
2082 (&p->se == cfs_rq_of(&p->se)->next ||
2083 &p->se == cfs_rq_of(&p->se)->last))
f540a608
IM
2084 return 1;
2085
6bc1665b
IM
2086 if (sysctl_sched_migration_cost == -1)
2087 return 1;
2088 if (sysctl_sched_migration_cost == 0)
2089 return 0;
2090
cc367732
IM
2091 delta = now - p->se.exec_start;
2092
2093 return delta < (s64)sysctl_sched_migration_cost;
2094}
2095
dd41f596 2096void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
c65cc870 2097{
e2912009
PZ
2098#ifdef CONFIG_SCHED_DEBUG
2099 /*
2100 * We should never call set_task_cpu() on a blocked task,
2101 * ttwu() will sort out the placement.
2102 */
077614ee
PZ
2103 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2104 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
e2912009
PZ
2105#endif
2106
de1d7286 2107 trace_sched_migrate_task(p, new_cpu);
cbc34ed1 2108
0c69774e
PZ
2109 if (task_cpu(p) != new_cpu) {
2110 p->se.nr_migrations++;
2111 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2112 }
dd41f596
IM
2113
2114 __set_task_cpu(p, new_cpu);
c65cc870
IM
2115}
2116
969c7921 2117struct migration_arg {
36c8b586 2118 struct task_struct *task;
1da177e4 2119 int dest_cpu;
70b97a7f 2120};
1da177e4 2121
969c7921
TH
2122static int migration_cpu_stop(void *data);
2123
1da177e4
LT
2124/*
2125 * The task's runqueue lock must be held.
2126 * Returns true if you have to wait for migration thread.
2127 */
b7a2b39d 2128static bool migrate_task(struct task_struct *p, struct rq *rq)
1da177e4 2129{
1da177e4
LT
2130 /*
2131 * If the task is not on a runqueue (and not running), then
e2912009 2132 * the next wake-up will properly place the task.
1da177e4 2133 */
969c7921 2134 return p->se.on_rq || task_running(rq, p);
1da177e4
LT
2135}
2136
2137/*
2138 * wait_task_inactive - wait for a thread to unschedule.
2139 *
85ba2d86
RM
2140 * If @match_state is nonzero, it's the @p->state value just checked and
2141 * not expected to change. If it changes, i.e. @p might have woken up,
2142 * then return zero. When we succeed in waiting for @p to be off its CPU,
2143 * we return a positive number (its total switch count). If a second call
2144 * a short while later returns the same number, the caller can be sure that
2145 * @p has remained unscheduled the whole time.
2146 *
1da177e4
LT
2147 * The caller must ensure that the task *will* unschedule sometime soon,
2148 * else this function might spin for a *long* time. This function can't
2149 * be called with interrupts off, or it may introduce deadlock with
2150 * smp_call_function() if an IPI is sent by the same process we are
2151 * waiting to become inactive.
2152 */
85ba2d86 2153unsigned long wait_task_inactive(struct task_struct *p, long match_state)
1da177e4
LT
2154{
2155 unsigned long flags;
dd41f596 2156 int running, on_rq;
85ba2d86 2157 unsigned long ncsw;
70b97a7f 2158 struct rq *rq;
1da177e4 2159
3a5c359a
AK
2160 for (;;) {
2161 /*
2162 * We do the initial early heuristics without holding
2163 * any task-queue locks at all. We'll only try to get
2164 * the runqueue lock when things look like they will
2165 * work out!
2166 */
2167 rq = task_rq(p);
fa490cfd 2168
3a5c359a
AK
2169 /*
2170 * If the task is actively running on another CPU
2171 * still, just relax and busy-wait without holding
2172 * any locks.
2173 *
2174 * NOTE! Since we don't hold any locks, it's not
2175 * even sure that "rq" stays as the right runqueue!
2176 * But we don't care, since "task_running()" will
2177 * return false if the runqueue has changed and p
2178 * is actually now running somewhere else!
2179 */
85ba2d86
RM
2180 while (task_running(rq, p)) {
2181 if (match_state && unlikely(p->state != match_state))
2182 return 0;
3a5c359a 2183 cpu_relax();
85ba2d86 2184 }
fa490cfd 2185
3a5c359a
AK
2186 /*
2187 * Ok, time to look more closely! We need the rq
2188 * lock now, to be *sure*. If we're wrong, we'll
2189 * just go back and repeat.
2190 */
2191 rq = task_rq_lock(p, &flags);
27a9da65 2192 trace_sched_wait_task(p);
3a5c359a
AK
2193 running = task_running(rq, p);
2194 on_rq = p->se.on_rq;
85ba2d86 2195 ncsw = 0;
f31e11d8 2196 if (!match_state || p->state == match_state)
93dcf55f 2197 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
3a5c359a 2198 task_rq_unlock(rq, &flags);
fa490cfd 2199
85ba2d86
RM
2200 /*
2201 * If it changed from the expected state, bail out now.
2202 */
2203 if (unlikely(!ncsw))
2204 break;
2205
3a5c359a
AK
2206 /*
2207 * Was it really running after all now that we
2208 * checked with the proper locks actually held?
2209 *
2210 * Oops. Go back and try again..
2211 */
2212 if (unlikely(running)) {
2213 cpu_relax();
2214 continue;
2215 }
fa490cfd 2216
3a5c359a
AK
2217 /*
2218 * It's not enough that it's not actively running,
2219 * it must be off the runqueue _entirely_, and not
2220 * preempted!
2221 *
80dd99b3 2222 * So if it was still runnable (but just not actively
3a5c359a
AK
2223 * running right now), it's preempted, and we should
2224 * yield - it could be a while.
2225 */
2226 if (unlikely(on_rq)) {
2227 schedule_timeout_uninterruptible(1);
2228 continue;
2229 }
fa490cfd 2230
3a5c359a
AK
2231 /*
2232 * Ahh, all good. It wasn't running, and it wasn't
2233 * runnable, which means that it will never become
2234 * running in the future either. We're all done!
2235 */
2236 break;
2237 }
85ba2d86
RM
2238
2239 return ncsw;
1da177e4
LT
2240}
2241
2242/***
2243 * kick_process - kick a running thread to enter/exit the kernel
2244 * @p: the to-be-kicked thread
2245 *
2246 * Cause a process which is running on another CPU to enter
2247 * kernel-mode, without any delay. (to get signals handled.)
2248 *
2249 * NOTE: this function doesnt have to take the runqueue lock,
2250 * because all it wants to ensure is that the remote task enters
2251 * the kernel. If the IPI races and the task has been migrated
2252 * to another CPU then no harm is done and the purpose has been
2253 * achieved as well.
2254 */
36c8b586 2255void kick_process(struct task_struct *p)
1da177e4
LT
2256{
2257 int cpu;
2258
2259 preempt_disable();
2260 cpu = task_cpu(p);
2261 if ((cpu != smp_processor_id()) && task_curr(p))
2262 smp_send_reschedule(cpu);
2263 preempt_enable();
2264}
b43e3521 2265EXPORT_SYMBOL_GPL(kick_process);
476d139c 2266#endif /* CONFIG_SMP */
1da177e4 2267
0793a61d
TG
2268/**
2269 * task_oncpu_function_call - call a function on the cpu on which a task runs
2270 * @p: the task to evaluate
2271 * @func: the function to be called
2272 * @info: the function call argument
2273 *
2274 * Calls the function @func when the task is currently running. This might
2275 * be on the current CPU, which just calls the function directly
2276 */
2277void task_oncpu_function_call(struct task_struct *p,
2278 void (*func) (void *info), void *info)
2279{
2280 int cpu;
2281
2282 preempt_disable();
2283 cpu = task_cpu(p);
2284 if (task_curr(p))
2285 smp_call_function_single(cpu, func, info, 1);
2286 preempt_enable();
2287}
2288
970b13ba 2289#ifdef CONFIG_SMP
30da688e
ON
2290/*
2291 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2292 */
5da9a0fb
PZ
2293static int select_fallback_rq(int cpu, struct task_struct *p)
2294{
2295 int dest_cpu;
2296 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2297
2298 /* Look for allowed, online CPU in same node. */
2299 for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2300 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2301 return dest_cpu;
2302
2303 /* Any allowed, online CPU? */
2304 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2305 if (dest_cpu < nr_cpu_ids)
2306 return dest_cpu;
2307
2308 /* No more Mr. Nice Guy. */
48c5ccae
PZ
2309 dest_cpu = cpuset_cpus_allowed_fallback(p);
2310 /*
2311 * Don't tell them about moving exiting tasks or
2312 * kernel threads (both mm NULL), since they never
2313 * leave kernel.
2314 */
2315 if (p->mm && printk_ratelimit()) {
2316 printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n",
2317 task_pid_nr(p), p->comm, cpu);
5da9a0fb
PZ
2318 }
2319
2320 return dest_cpu;
2321}
2322
e2912009 2323/*
30da688e 2324 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
e2912009 2325 */
970b13ba 2326static inline
0017d735 2327int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
970b13ba 2328{
0017d735 2329 int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
e2912009
PZ
2330
2331 /*
2332 * In order not to call set_task_cpu() on a blocking task we need
2333 * to rely on ttwu() to place the task on a valid ->cpus_allowed
2334 * cpu.
2335 *
2336 * Since this is common to all placement strategies, this lives here.
2337 *
2338 * [ this allows ->select_task() to simply return task_cpu(p) and
2339 * not worry about this generic constraint ]
2340 */
2341 if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
70f11205 2342 !cpu_online(cpu)))
5da9a0fb 2343 cpu = select_fallback_rq(task_cpu(p), p);
e2912009
PZ
2344
2345 return cpu;
970b13ba 2346}
09a40af5
MG
2347
2348static void update_avg(u64 *avg, u64 sample)
2349{
2350 s64 diff = sample - *avg;
2351 *avg += diff >> 3;
2352}
970b13ba
PZ
2353#endif
2354
9ed3811a
TH
2355static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2356 bool is_sync, bool is_migrate, bool is_local,
2357 unsigned long en_flags)
2358{
2359 schedstat_inc(p, se.statistics.nr_wakeups);
2360 if (is_sync)
2361 schedstat_inc(p, se.statistics.nr_wakeups_sync);
2362 if (is_migrate)
2363 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2364 if (is_local)
2365 schedstat_inc(p, se.statistics.nr_wakeups_local);
2366 else
2367 schedstat_inc(p, se.statistics.nr_wakeups_remote);
2368
2369 activate_task(rq, p, en_flags);
2370}
2371
2372static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2373 int wake_flags, bool success)
2374{
2375 trace_sched_wakeup(p, success);
2376 check_preempt_curr(rq, p, wake_flags);
2377
2378 p->state = TASK_RUNNING;
2379#ifdef CONFIG_SMP
2380 if (p->sched_class->task_woken)
2381 p->sched_class->task_woken(rq, p);
2382
2383 if (unlikely(rq->idle_stamp)) {
2384 u64 delta = rq->clock - rq->idle_stamp;
2385 u64 max = 2*sysctl_sched_migration_cost;
2386
2387 if (delta > max)
2388 rq->avg_idle = max;
2389 else
2390 update_avg(&rq->avg_idle, delta);
2391 rq->idle_stamp = 0;
2392 }
2393#endif
21aa9af0
TH
2394 /* if a worker is waking up, notify workqueue */
2395 if ((p->flags & PF_WQ_WORKER) && success)
2396 wq_worker_waking_up(p, cpu_of(rq));
9ed3811a
TH
2397}
2398
2399/**
1da177e4 2400 * try_to_wake_up - wake up a thread
9ed3811a 2401 * @p: the thread to be awakened
1da177e4 2402 * @state: the mask of task states that can be woken
9ed3811a 2403 * @wake_flags: wake modifier flags (WF_*)
1da177e4
LT
2404 *
2405 * Put it on the run-queue if it's not already there. The "current"
2406 * thread is always on the run-queue (except when the actual
2407 * re-schedule is in progress), and as such you're allowed to do
2408 * the simpler "current->state = TASK_RUNNING" to mark yourself
2409 * runnable without the overhead of this.
2410 *
9ed3811a
TH
2411 * Returns %true if @p was woken up, %false if it was already running
2412 * or @state didn't match @p's state.
1da177e4 2413 */
7d478721
PZ
2414static int try_to_wake_up(struct task_struct *p, unsigned int state,
2415 int wake_flags)
1da177e4 2416{
cc367732 2417 int cpu, orig_cpu, this_cpu, success = 0;
1da177e4 2418 unsigned long flags;
371fd7e7 2419 unsigned long en_flags = ENQUEUE_WAKEUP;
ab3b3aa5 2420 struct rq *rq;
1da177e4 2421
e9c84311 2422 this_cpu = get_cpu();
2398f2c6 2423
04e2f174 2424 smp_wmb();
ab3b3aa5 2425 rq = task_rq_lock(p, &flags);
e9c84311 2426 if (!(p->state & state))
1da177e4
LT
2427 goto out;
2428
dd41f596 2429 if (p->se.on_rq)
1da177e4
LT
2430 goto out_running;
2431
2432 cpu = task_cpu(p);
cc367732 2433 orig_cpu = cpu;
1da177e4
LT
2434
2435#ifdef CONFIG_SMP
2436 if (unlikely(task_running(rq, p)))
2437 goto out_activate;
2438
e9c84311
PZ
2439 /*
2440 * In order to handle concurrent wakeups and release the rq->lock
2441 * we put the task in TASK_WAKING state.
eb24073b
IM
2442 *
2443 * First fix up the nr_uninterruptible count:
e9c84311 2444 */
cc87f76a
PZ
2445 if (task_contributes_to_load(p)) {
2446 if (likely(cpu_online(orig_cpu)))
2447 rq->nr_uninterruptible--;
2448 else
2449 this_rq()->nr_uninterruptible--;
2450 }
e9c84311 2451 p->state = TASK_WAKING;
efbbd05a 2452
371fd7e7 2453 if (p->sched_class->task_waking) {
efbbd05a 2454 p->sched_class->task_waking(rq, p);
371fd7e7
PZ
2455 en_flags |= ENQUEUE_WAKING;
2456 }
efbbd05a 2457
0017d735
PZ
2458 cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2459 if (cpu != orig_cpu)
5d2f5a61 2460 set_task_cpu(p, cpu);
0017d735 2461 __task_rq_unlock(rq);
ab19cb23 2462
0970d299
PZ
2463 rq = cpu_rq(cpu);
2464 raw_spin_lock(&rq->lock);
f5dc3753 2465
0970d299
PZ
2466 /*
2467 * We migrated the task without holding either rq->lock, however
2468 * since the task is not on the task list itself, nobody else
2469 * will try and migrate the task, hence the rq should match the
2470 * cpu we just moved it to.
2471 */
2472 WARN_ON(task_cpu(p) != cpu);
e9c84311 2473 WARN_ON(p->state != TASK_WAKING);
1da177e4 2474
e7693a36
GH
2475#ifdef CONFIG_SCHEDSTATS
2476 schedstat_inc(rq, ttwu_count);
2477 if (cpu == this_cpu)
2478 schedstat_inc(rq, ttwu_local);
2479 else {
2480 struct sched_domain *sd;
2481 for_each_domain(this_cpu, sd) {
758b2cdc 2482 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
e7693a36
GH
2483 schedstat_inc(sd, ttwu_wake_remote);
2484 break;
2485 }
2486 }
2487 }
6d6bc0ad 2488#endif /* CONFIG_SCHEDSTATS */
e7693a36 2489
1da177e4
LT
2490out_activate:
2491#endif /* CONFIG_SMP */
9ed3811a
TH
2492 ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2493 cpu == this_cpu, en_flags);
1da177e4 2494 success = 1;
1da177e4 2495out_running:
9ed3811a 2496 ttwu_post_activation(p, rq, wake_flags, success);
1da177e4
LT
2497out:
2498 task_rq_unlock(rq, &flags);
e9c84311 2499 put_cpu();
1da177e4
LT
2500
2501 return success;
2502}
2503
21aa9af0
TH
2504/**
2505 * try_to_wake_up_local - try to wake up a local task with rq lock held
2506 * @p: the thread to be awakened
2507 *
2508 * Put @p on the run-queue if it's not alredy there. The caller must
2509 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2510 * the current task. this_rq() stays locked over invocation.
2511 */
2512static void try_to_wake_up_local(struct task_struct *p)
2513{
2514 struct rq *rq = task_rq(p);
2515 bool success = false;
2516
2517 BUG_ON(rq != this_rq());
2518 BUG_ON(p == current);
2519 lockdep_assert_held(&rq->lock);
2520
2521 if (!(p->state & TASK_NORMAL))
2522 return;
2523
2524 if (!p->se.on_rq) {
2525 if (likely(!task_running(rq, p))) {
2526 schedstat_inc(rq, ttwu_count);
2527 schedstat_inc(rq, ttwu_local);
2528 }
2529 ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2530 success = true;
2531 }
2532 ttwu_post_activation(p, rq, 0, success);
2533}
2534
50fa610a
DH
2535/**
2536 * wake_up_process - Wake up a specific process
2537 * @p: The process to be woken up.
2538 *
2539 * Attempt to wake up the nominated process and move it to the set of runnable
2540 * processes. Returns 1 if the process was woken up, 0 if it was already
2541 * running.
2542 *
2543 * It may be assumed that this function implies a write memory barrier before
2544 * changing the task state if and only if any tasks are woken up.
2545 */
7ad5b3a5 2546int wake_up_process(struct task_struct *p)
1da177e4 2547{
d9514f6c 2548 return try_to_wake_up(p, TASK_ALL, 0);
1da177e4 2549}
1da177e4
LT
2550EXPORT_SYMBOL(wake_up_process);
2551
7ad5b3a5 2552int wake_up_state(struct task_struct *p, unsigned int state)
1da177e4
LT
2553{
2554 return try_to_wake_up(p, state, 0);
2555}
2556
1da177e4
LT
2557/*
2558 * Perform scheduler related setup for a newly forked process p.
2559 * p is forked by current.
dd41f596
IM
2560 *
2561 * __sched_fork() is basic setup used by init_idle() too:
2562 */
2563static void __sched_fork(struct task_struct *p)
2564{
dd41f596
IM
2565 p->se.exec_start = 0;
2566 p->se.sum_exec_runtime = 0;
f6cf891c 2567 p->se.prev_sum_exec_runtime = 0;
6c594c21 2568 p->se.nr_migrations = 0;
6cfb0d5d
IM
2569
2570#ifdef CONFIG_SCHEDSTATS
41acab88 2571 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
6cfb0d5d 2572#endif
476d139c 2573
fa717060 2574 INIT_LIST_HEAD(&p->rt.run_list);
dd41f596 2575 p->se.on_rq = 0;
4a55bd5e 2576 INIT_LIST_HEAD(&p->se.group_node);
476d139c 2577
e107be36
AK
2578#ifdef CONFIG_PREEMPT_NOTIFIERS
2579 INIT_HLIST_HEAD(&p->preempt_notifiers);
2580#endif
dd41f596
IM
2581}
2582
2583/*
2584 * fork()/clone()-time setup:
2585 */
2586void sched_fork(struct task_struct *p, int clone_flags)
2587{
2588 int cpu = get_cpu();
2589
2590 __sched_fork(p);
06b83b5f 2591 /*
0017d735 2592 * We mark the process as running here. This guarantees that
06b83b5f
PZ
2593 * nobody will actually run it, and a signal or other external
2594 * event cannot wake it up and insert it on the runqueue either.
2595 */
0017d735 2596 p->state = TASK_RUNNING;
dd41f596 2597
b9dc29e7
MG
2598 /*
2599 * Revert to default priority/policy on fork if requested.
2600 */
2601 if (unlikely(p->sched_reset_on_fork)) {
f83f9ac2 2602 if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
b9dc29e7 2603 p->policy = SCHED_NORMAL;
f83f9ac2
PW
2604 p->normal_prio = p->static_prio;
2605 }
b9dc29e7 2606
6c697bdf
MG
2607 if (PRIO_TO_NICE(p->static_prio) < 0) {
2608 p->static_prio = NICE_TO_PRIO(0);
f83f9ac2 2609 p->normal_prio = p->static_prio;
6c697bdf
MG
2610 set_load_weight(p);
2611 }
2612
b9dc29e7
MG
2613 /*
2614 * We don't need the reset flag anymore after the fork. It has
2615 * fulfilled its duty:
2616 */
2617 p->sched_reset_on_fork = 0;
2618 }
ca94c442 2619
f83f9ac2
PW
2620 /*
2621 * Make sure we do not leak PI boosting priority to the child.
2622 */
2623 p->prio = current->normal_prio;
2624
2ddbf952
HS
2625 if (!rt_prio(p->prio))
2626 p->sched_class = &fair_sched_class;
b29739f9 2627
cd29fe6f
PZ
2628 if (p->sched_class->task_fork)
2629 p->sched_class->task_fork(p);
2630
86951599
PZ
2631 /*
2632 * The child is not yet in the pid-hash so no cgroup attach races,
2633 * and the cgroup is pinned to this child due to cgroup_fork()
2634 * is ran before sched_fork().
2635 *
2636 * Silence PROVE_RCU.
2637 */
2638 rcu_read_lock();
5f3edc1b 2639 set_task_cpu(p, cpu);
86951599 2640 rcu_read_unlock();
5f3edc1b 2641
52f17b6c 2642#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
dd41f596 2643 if (likely(sched_info_on()))
52f17b6c 2644 memset(&p->sched_info, 0, sizeof(p->sched_info));
1da177e4 2645#endif
d6077cb8 2646#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
4866cde0
NP
2647 p->oncpu = 0;
2648#endif
1da177e4 2649#ifdef CONFIG_PREEMPT
4866cde0 2650 /* Want to start with kernel preemption disabled. */
a1261f54 2651 task_thread_info(p)->preempt_count = 1;
1da177e4 2652#endif
806c09a7 2653#ifdef CONFIG_SMP
917b627d 2654 plist_node_init(&p->pushable_tasks, MAX_PRIO);
806c09a7 2655#endif
917b627d 2656
476d139c 2657 put_cpu();
1da177e4
LT
2658}
2659
2660/*
2661 * wake_up_new_task - wake up a newly created task for the first time.
2662 *
2663 * This function will do some initial scheduler statistics housekeeping
2664 * that must be done for every newly created context, then puts the task
2665 * on the runqueue and wakes it.
2666 */
7ad5b3a5 2667void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
1da177e4
LT
2668{
2669 unsigned long flags;
dd41f596 2670 struct rq *rq;
c890692b 2671 int cpu __maybe_unused = get_cpu();
fabf318e
PZ
2672
2673#ifdef CONFIG_SMP
0017d735
PZ
2674 rq = task_rq_lock(p, &flags);
2675 p->state = TASK_WAKING;
2676
fabf318e
PZ
2677 /*
2678 * Fork balancing, do it here and not earlier because:
2679 * - cpus_allowed can change in the fork path
2680 * - any previously selected cpu might disappear through hotplug
2681 *
0017d735
PZ
2682 * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2683 * without people poking at ->cpus_allowed.
fabf318e 2684 */
0017d735 2685 cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
fabf318e 2686 set_task_cpu(p, cpu);
1da177e4 2687
06b83b5f 2688 p->state = TASK_RUNNING;
0017d735
PZ
2689 task_rq_unlock(rq, &flags);
2690#endif
2691
2692 rq = task_rq_lock(p, &flags);
cd29fe6f 2693 activate_task(rq, p, 0);
27a9da65 2694 trace_sched_wakeup_new(p, 1);
a7558e01 2695 check_preempt_curr(rq, p, WF_FORK);
9a897c5a 2696#ifdef CONFIG_SMP
efbbd05a
PZ
2697 if (p->sched_class->task_woken)
2698 p->sched_class->task_woken(rq, p);
9a897c5a 2699#endif
dd41f596 2700 task_rq_unlock(rq, &flags);
fabf318e 2701 put_cpu();
1da177e4
LT
2702}
2703
e107be36
AK
2704#ifdef CONFIG_PREEMPT_NOTIFIERS
2705
2706/**
80dd99b3 2707 * preempt_notifier_register - tell me when current is being preempted & rescheduled
421cee29 2708 * @notifier: notifier struct to register
e107be36
AK
2709 */
2710void preempt_notifier_register(struct preempt_notifier *notifier)
2711{
2712 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2713}
2714EXPORT_SYMBOL_GPL(preempt_notifier_register);
2715
2716/**
2717 * preempt_notifier_unregister - no longer interested in preemption notifications
421cee29 2718 * @notifier: notifier struct to unregister
e107be36
AK
2719 *
2720 * This is safe to call from within a preemption notifier.
2721 */
2722void preempt_notifier_unregister(struct preempt_notifier *notifier)
2723{
2724 hlist_del(&notifier->link);
2725}
2726EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2727
2728static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2729{
2730 struct preempt_notifier *notifier;
2731 struct hlist_node *node;
2732
2733 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2734 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2735}
2736
2737static void
2738fire_sched_out_preempt_notifiers(struct task_struct *curr,
2739 struct task_struct *next)
2740{
2741 struct preempt_notifier *notifier;
2742 struct hlist_node *node;
2743
2744 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2745 notifier->ops->sched_out(notifier, next);
2746}
2747
6d6bc0ad 2748#else /* !CONFIG_PREEMPT_NOTIFIERS */
e107be36
AK
2749
2750static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2751{
2752}
2753
2754static void
2755fire_sched_out_preempt_notifiers(struct task_struct *curr,
2756 struct task_struct *next)
2757{
2758}
2759
6d6bc0ad 2760#endif /* CONFIG_PREEMPT_NOTIFIERS */
e107be36 2761
4866cde0
NP
2762/**
2763 * prepare_task_switch - prepare to switch tasks
2764 * @rq: the runqueue preparing to switch
421cee29 2765 * @prev: the current task that is being switched out
4866cde0
NP
2766 * @next: the task we are going to switch to.
2767 *
2768 * This is called with the rq lock held and interrupts off. It must
2769 * be paired with a subsequent finish_task_switch after the context
2770 * switch.
2771 *
2772 * prepare_task_switch sets up locking and calls architecture specific
2773 * hooks.
2774 */
e107be36
AK
2775static inline void
2776prepare_task_switch(struct rq *rq, struct task_struct *prev,
2777 struct task_struct *next)
4866cde0 2778{
e107be36 2779 fire_sched_out_preempt_notifiers(prev, next);
4866cde0
NP
2780 prepare_lock_switch(rq, next);
2781 prepare_arch_switch(next);
2782}
2783
1da177e4
LT
2784/**
2785 * finish_task_switch - clean up after a task-switch
344babaa 2786 * @rq: runqueue associated with task-switch
1da177e4
LT
2787 * @prev: the thread we just switched away from.
2788 *
4866cde0
NP
2789 * finish_task_switch must be called after the context switch, paired
2790 * with a prepare_task_switch call before the context switch.
2791 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2792 * and do any other architecture-specific cleanup actions.
1da177e4
LT
2793 *
2794 * Note that we may have delayed dropping an mm in context_switch(). If
41a2d6cf 2795 * so, we finish that here outside of the runqueue lock. (Doing it
1da177e4
LT
2796 * with the lock held can cause deadlocks; see schedule() for
2797 * details.)
2798 */
a9957449 2799static void finish_task_switch(struct rq *rq, struct task_struct *prev)
1da177e4
LT
2800 __releases(rq->lock)
2801{
1da177e4 2802 struct mm_struct *mm = rq->prev_mm;
55a101f8 2803 long prev_state;
1da177e4
LT
2804
2805 rq->prev_mm = NULL;
2806
2807 /*
2808 * A task struct has one reference for the use as "current".
c394cc9f 2809 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
55a101f8
ON
2810 * schedule one last time. The schedule call will never return, and
2811 * the scheduled task must drop that reference.
c394cc9f 2812 * The test for TASK_DEAD must occur while the runqueue locks are
1da177e4
LT
2813 * still held, otherwise prev could be scheduled on another cpu, die
2814 * there before we look at prev->state, and then the reference would
2815 * be dropped twice.
2816 * Manfred Spraul <manfred@colorfullife.com>
2817 */
55a101f8 2818 prev_state = prev->state;
4866cde0 2819 finish_arch_switch(prev);
8381f65d
JI
2820#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2821 local_irq_disable();
2822#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
49f47433 2823 perf_event_task_sched_in(current);
8381f65d
JI
2824#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2825 local_irq_enable();
2826#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
4866cde0 2827 finish_lock_switch(rq, prev);
e8fa1362 2828
e107be36 2829 fire_sched_in_preempt_notifiers(current);
1da177e4
LT
2830 if (mm)
2831 mmdrop(mm);
c394cc9f 2832 if (unlikely(prev_state == TASK_DEAD)) {
c6fd91f0 2833 /*
2834 * Remove function-return probe instances associated with this
2835 * task and put them back on the free list.
9761eea8 2836 */
c6fd91f0 2837 kprobe_flush_task(prev);
1da177e4 2838 put_task_struct(prev);
c6fd91f0 2839 }
1da177e4
LT
2840}
2841
3f029d3c
GH
2842#ifdef CONFIG_SMP
2843
2844/* assumes rq->lock is held */
2845static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2846{
2847 if (prev->sched_class->pre_schedule)
2848 prev->sched_class->pre_schedule(rq, prev);
2849}
2850
2851/* rq->lock is NOT held, but preemption is disabled */
2852static inline void post_schedule(struct rq *rq)
2853{
2854 if (rq->post_schedule) {
2855 unsigned long flags;
2856
05fa785c 2857 raw_spin_lock_irqsave(&rq->lock, flags);
3f029d3c
GH
2858 if (rq->curr->sched_class->post_schedule)
2859 rq->curr->sched_class->post_schedule(rq);
05fa785c 2860 raw_spin_unlock_irqrestore(&rq->lock, flags);
3f029d3c
GH
2861
2862 rq->post_schedule = 0;
2863 }
2864}
2865
2866#else
da19ab51 2867
3f029d3c
GH
2868static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2869{
2870}
2871
2872static inline void post_schedule(struct rq *rq)
2873{
1da177e4
LT
2874}
2875
3f029d3c
GH
2876#endif
2877
1da177e4
LT
2878/**
2879 * schedule_tail - first thing a freshly forked thread must call.
2880 * @prev: the thread we just switched away from.
2881 */
36c8b586 2882asmlinkage void schedule_tail(struct task_struct *prev)
1da177e4
LT
2883 __releases(rq->lock)
2884{
70b97a7f
IM
2885 struct rq *rq = this_rq();
2886
4866cde0 2887 finish_task_switch(rq, prev);
da19ab51 2888
3f029d3c
GH
2889 /*
2890 * FIXME: do we need to worry about rq being invalidated by the
2891 * task_switch?
2892 */
2893 post_schedule(rq);
70b97a7f 2894
4866cde0
NP
2895#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2896 /* In this case, finish_task_switch does not reenable preemption */
2897 preempt_enable();
2898#endif
1da177e4 2899 if (current->set_child_tid)
b488893a 2900 put_user(task_pid_vnr(current), current->set_child_tid);
1da177e4
LT
2901}
2902
2903/*
2904 * context_switch - switch to the new MM and the new
2905 * thread's register state.
2906 */
dd41f596 2907static inline void
70b97a7f 2908context_switch(struct rq *rq, struct task_struct *prev,
36c8b586 2909 struct task_struct *next)
1da177e4 2910{
dd41f596 2911 struct mm_struct *mm, *oldmm;
1da177e4 2912
e107be36 2913 prepare_task_switch(rq, prev, next);
27a9da65 2914 trace_sched_switch(prev, next);
dd41f596
IM
2915 mm = next->mm;
2916 oldmm = prev->active_mm;
9226d125
ZA
2917 /*
2918 * For paravirt, this is coupled with an exit in switch_to to
2919 * combine the page table reload and the switch backend into
2920 * one hypercall.
2921 */
224101ed 2922 arch_start_context_switch(prev);
9226d125 2923
31915ab4 2924 if (!mm) {
1da177e4
LT
2925 next->active_mm = oldmm;
2926 atomic_inc(&oldmm->mm_count);
2927 enter_lazy_tlb(oldmm, next);
2928 } else
2929 switch_mm(oldmm, mm, next);
2930
31915ab4 2931 if (!prev->mm) {
1da177e4 2932 prev->active_mm = NULL;
1da177e4
LT
2933 rq->prev_mm = oldmm;
2934 }
3a5f5e48
IM
2935 /*
2936 * Since the runqueue lock will be released by the next
2937 * task (which is an invalid locking op but in the case
2938 * of the scheduler it's an obvious special-case), so we
2939 * do an early lockdep release here:
2940 */
2941#ifndef __ARCH_WANT_UNLOCKED_CTXSW
8a25d5de 2942 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
3a5f5e48 2943#endif
1da177e4
LT
2944
2945 /* Here we just switch the register state and the stack. */
2946 switch_to(prev, next, prev);
2947
dd41f596
IM
2948 barrier();
2949 /*
2950 * this_rq must be evaluated again because prev may have moved
2951 * CPUs since it called schedule(), thus the 'rq' on its stack
2952 * frame will be invalid.
2953 */
2954 finish_task_switch(this_rq(), prev);
1da177e4
LT
2955}
2956
2957/*
2958 * nr_running, nr_uninterruptible and nr_context_switches:
2959 *
2960 * externally visible scheduler statistics: current number of runnable
2961 * threads, current number of uninterruptible-sleeping threads, total
2962 * number of context switches performed since bootup.
2963 */
2964unsigned long nr_running(void)
2965{
2966 unsigned long i, sum = 0;
2967
2968 for_each_online_cpu(i)
2969 sum += cpu_rq(i)->nr_running;
2970
2971 return sum;
f711f609 2972}
1da177e4
LT
2973
2974unsigned long nr_uninterruptible(void)
f711f609 2975{
1da177e4 2976 unsigned long i, sum = 0;
f711f609 2977
0a945022 2978 for_each_possible_cpu(i)
1da177e4 2979 sum += cpu_rq(i)->nr_uninterruptible;
f711f609
GS
2980
2981 /*
1da177e4
LT
2982 * Since we read the counters lockless, it might be slightly
2983 * inaccurate. Do not allow it to go below zero though:
f711f609 2984 */
1da177e4
LT
2985 if (unlikely((long)sum < 0))
2986 sum = 0;
f711f609 2987
1da177e4 2988 return sum;
f711f609 2989}
f711f609 2990
1da177e4 2991unsigned long long nr_context_switches(void)
46cb4b7c 2992{
cc94abfc
SR
2993 int i;
2994 unsigned long long sum = 0;
46cb4b7c 2995
0a945022 2996 for_each_possible_cpu(i)
1da177e4 2997 sum += cpu_rq(i)->nr_switches;
46cb4b7c 2998
1da177e4
LT
2999 return sum;
3000}
483b4ee6 3001
1da177e4
LT
3002unsigned long nr_iowait(void)
3003{
3004 unsigned long i, sum = 0;
483b4ee6 3005
0a945022 3006 for_each_possible_cpu(i)
1da177e4 3007 sum += atomic_read(&cpu_rq(i)->nr_iowait);
46cb4b7c 3008
1da177e4
LT
3009 return sum;
3010}
483b4ee6 3011
8c215bd3 3012unsigned long nr_iowait_cpu(int cpu)
69d25870 3013{
8c215bd3 3014 struct rq *this = cpu_rq(cpu);
69d25870
AV
3015 return atomic_read(&this->nr_iowait);
3016}
46cb4b7c 3017
69d25870
AV
3018unsigned long this_cpu_load(void)
3019{
3020 struct rq *this = this_rq();
3021 return this->cpu_load[0];
3022}
e790fb0b 3023
46cb4b7c 3024
dce48a84
TG
3025/* Variables and functions for calc_load */
3026static atomic_long_t calc_load_tasks;
3027static unsigned long calc_load_update;
3028unsigned long avenrun[3];
3029EXPORT_SYMBOL(avenrun);
46cb4b7c 3030
74f5187a
PZ
3031static long calc_load_fold_active(struct rq *this_rq)
3032{
3033 long nr_active, delta = 0;
3034
3035 nr_active = this_rq->nr_running;
3036 nr_active += (long) this_rq->nr_uninterruptible;
3037
3038 if (nr_active != this_rq->calc_load_active) {
3039 delta = nr_active - this_rq->calc_load_active;
3040 this_rq->calc_load_active = nr_active;
3041 }
3042
3043 return delta;
3044}
3045
0f004f5a
PZ
3046static unsigned long
3047calc_load(unsigned long load, unsigned long exp, unsigned long active)
3048{
3049 load *= exp;
3050 load += active * (FIXED_1 - exp);
3051 load += 1UL << (FSHIFT - 1);
3052 return load >> FSHIFT;
3053}
3054
74f5187a
PZ
3055#ifdef CONFIG_NO_HZ
3056/*
3057 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
3058 *
3059 * When making the ILB scale, we should try to pull this in as well.
3060 */
3061static atomic_long_t calc_load_tasks_idle;
3062
3063static void calc_load_account_idle(struct rq *this_rq)
3064{
3065 long delta;
3066
3067 delta = calc_load_fold_active(this_rq);
3068 if (delta)
3069 atomic_long_add(delta, &calc_load_tasks_idle);
3070}
3071
3072static long calc_load_fold_idle(void)
3073{
3074 long delta = 0;
3075
3076 /*
3077 * Its got a race, we don't care...
3078 */
3079 if (atomic_long_read(&calc_load_tasks_idle))
3080 delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3081
3082 return delta;
3083}
0f004f5a
PZ
3084
3085/**
3086 * fixed_power_int - compute: x^n, in O(log n) time
3087 *
3088 * @x: base of the power
3089 * @frac_bits: fractional bits of @x
3090 * @n: power to raise @x to.
3091 *
3092 * By exploiting the relation between the definition of the natural power
3093 * function: x^n := x*x*...*x (x multiplied by itself for n times), and
3094 * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i,
3095 * (where: n_i \elem {0, 1}, the binary vector representing n),
3096 * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is
3097 * of course trivially computable in O(log_2 n), the length of our binary
3098 * vector.
3099 */
3100static unsigned long
3101fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n)
3102{
3103 unsigned long result = 1UL << frac_bits;
3104
3105 if (n) for (;;) {
3106 if (n & 1) {
3107 result *= x;
3108 result += 1UL << (frac_bits - 1);
3109 result >>= frac_bits;
3110 }
3111 n >>= 1;
3112 if (!n)
3113 break;
3114 x *= x;
3115 x += 1UL << (frac_bits - 1);
3116 x >>= frac_bits;
3117 }
3118
3119 return result;
3120}
3121
3122/*
3123 * a1 = a0 * e + a * (1 - e)
3124 *
3125 * a2 = a1 * e + a * (1 - e)
3126 * = (a0 * e + a * (1 - e)) * e + a * (1 - e)
3127 * = a0 * e^2 + a * (1 - e) * (1 + e)
3128 *
3129 * a3 = a2 * e + a * (1 - e)
3130 * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e)
3131 * = a0 * e^3 + a * (1 - e) * (1 + e + e^2)
3132 *
3133 * ...
3134 *
3135 * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1]
3136 * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e)
3137 * = a0 * e^n + a * (1 - e^n)
3138 *
3139 * [1] application of the geometric series:
3140 *
3141 * n 1 - x^(n+1)
3142 * S_n := \Sum x^i = -------------
3143 * i=0 1 - x
3144 */
3145static unsigned long
3146calc_load_n(unsigned long load, unsigned long exp,
3147 unsigned long active, unsigned int n)
3148{
3149
3150 return calc_load(load, fixed_power_int(exp, FSHIFT, n), active);
3151}
3152
3153/*
3154 * NO_HZ can leave us missing all per-cpu ticks calling
3155 * calc_load_account_active(), but since an idle CPU folds its delta into
3156 * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold
3157 * in the pending idle delta if our idle period crossed a load cycle boundary.
3158 *
3159 * Once we've updated the global active value, we need to apply the exponential
3160 * weights adjusted to the number of cycles missed.
3161 */
3162static void calc_global_nohz(unsigned long ticks)
3163{
3164 long delta, active, n;
3165
3166 if (time_before(jiffies, calc_load_update))
3167 return;
3168
3169 /*
3170 * If we crossed a calc_load_update boundary, make sure to fold
3171 * any pending idle changes, the respective CPUs might have
3172 * missed the tick driven calc_load_account_active() update
3173 * due to NO_HZ.
3174 */
3175 delta = calc_load_fold_idle();
3176 if (delta)
3177 atomic_long_add(delta, &calc_load_tasks);
3178
3179 /*
3180 * If we were idle for multiple load cycles, apply them.
3181 */
3182 if (ticks >= LOAD_FREQ) {
3183 n = ticks / LOAD_FREQ;
3184
3185 active = atomic_long_read(&calc_load_tasks);
3186 active = active > 0 ? active * FIXED_1 : 0;
3187
3188 avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n);
3189 avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n);
3190 avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n);
3191
3192 calc_load_update += n * LOAD_FREQ;
3193 }
3194
3195 /*
3196 * Its possible the remainder of the above division also crosses
3197 * a LOAD_FREQ period, the regular check in calc_global_load()
3198 * which comes after this will take care of that.
3199 *
3200 * Consider us being 11 ticks before a cycle completion, and us
3201 * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will
3202 * age us 4 cycles, and the test in calc_global_load() will
3203 * pick up the final one.
3204 */
3205}
74f5187a
PZ
3206#else
3207static void calc_load_account_idle(struct rq *this_rq)
3208{
3209}
3210
3211static inline long calc_load_fold_idle(void)
3212{
3213 return 0;
3214}
0f004f5a
PZ
3215
3216static void calc_global_nohz(unsigned long ticks)
3217{
3218}
74f5187a
PZ
3219#endif
3220
2d02494f
TG
3221/**
3222 * get_avenrun - get the load average array
3223 * @loads: pointer to dest load array
3224 * @offset: offset to add
3225 * @shift: shift count to shift the result left
3226 *
3227 * These values are estimates at best, so no need for locking.
3228 */
3229void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3230{
3231 loads[0] = (avenrun[0] + offset) << shift;
3232 loads[1] = (avenrun[1] + offset) << shift;
3233 loads[2] = (avenrun[2] + offset) << shift;
46cb4b7c 3234}
46cb4b7c 3235
46cb4b7c 3236/*
dce48a84
TG
3237 * calc_load - update the avenrun load estimates 10 ticks after the
3238 * CPUs have updated calc_load_tasks.
7835b98b 3239 */
0f004f5a 3240void calc_global_load(unsigned long ticks)
7835b98b 3241{
dce48a84 3242 long active;
1da177e4 3243
0f004f5a
PZ
3244 calc_global_nohz(ticks);
3245
3246 if (time_before(jiffies, calc_load_update + 10))
dce48a84 3247 return;
1da177e4 3248
dce48a84
TG
3249 active = atomic_long_read(&calc_load_tasks);
3250 active = active > 0 ? active * FIXED_1 : 0;
1da177e4 3251
dce48a84
TG
3252 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3253 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3254 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
dd41f596 3255
dce48a84
TG
3256 calc_load_update += LOAD_FREQ;
3257}
1da177e4 3258
dce48a84 3259/*
74f5187a
PZ
3260 * Called from update_cpu_load() to periodically update this CPU's
3261 * active count.
dce48a84
TG
3262 */
3263static void calc_load_account_active(struct rq *this_rq)
3264{
74f5187a 3265 long delta;
08c183f3 3266
74f5187a
PZ
3267 if (time_before(jiffies, this_rq->calc_load_update))
3268 return;
783609c6 3269
74f5187a
PZ
3270 delta = calc_load_fold_active(this_rq);
3271 delta += calc_load_fold_idle();
3272 if (delta)
dce48a84 3273 atomic_long_add(delta, &calc_load_tasks);
74f5187a
PZ
3274
3275 this_rq->calc_load_update += LOAD_FREQ;
46cb4b7c
SS
3276}
3277
fdf3e95d
VP
3278/*
3279 * The exact cpuload at various idx values, calculated at every tick would be
3280 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3281 *
3282 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3283 * on nth tick when cpu may be busy, then we have:
3284 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3285 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3286 *
3287 * decay_load_missed() below does efficient calculation of
3288 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3289 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3290 *
3291 * The calculation is approximated on a 128 point scale.
3292 * degrade_zero_ticks is the number of ticks after which load at any
3293 * particular idx is approximated to be zero.
3294 * degrade_factor is a precomputed table, a row for each load idx.
3295 * Each column corresponds to degradation factor for a power of two ticks,
3296 * based on 128 point scale.
3297 * Example:
3298 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3299 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3300 *
3301 * With this power of 2 load factors, we can degrade the load n times
3302 * by looking at 1 bits in n and doing as many mult/shift instead of
3303 * n mult/shifts needed by the exact degradation.
3304 */
3305#define DEGRADE_SHIFT 7
3306static const unsigned char
3307 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3308static const unsigned char
3309 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3310 {0, 0, 0, 0, 0, 0, 0, 0},
3311 {64, 32, 8, 0, 0, 0, 0, 0},
3312 {96, 72, 40, 12, 1, 0, 0},
3313 {112, 98, 75, 43, 15, 1, 0},
3314 {120, 112, 98, 76, 45, 16, 2} };
3315
3316/*
3317 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3318 * would be when CPU is idle and so we just decay the old load without
3319 * adding any new load.
3320 */
3321static unsigned long
3322decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3323{
3324 int j = 0;
3325
3326 if (!missed_updates)
3327 return load;
3328
3329 if (missed_updates >= degrade_zero_ticks[idx])
3330 return 0;
3331
3332 if (idx == 1)
3333 return load >> missed_updates;
3334
3335 while (missed_updates) {
3336 if (missed_updates % 2)
3337 load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3338
3339 missed_updates >>= 1;
3340 j++;
3341 }
3342 return load;
3343}
3344
46cb4b7c 3345/*
dd41f596 3346 * Update rq->cpu_load[] statistics. This function is usually called every
fdf3e95d
VP
3347 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3348 * every tick. We fix it up based on jiffies.
46cb4b7c 3349 */
dd41f596 3350static void update_cpu_load(struct rq *this_rq)
46cb4b7c 3351{
495eca49 3352 unsigned long this_load = this_rq->load.weight;
fdf3e95d
VP
3353 unsigned long curr_jiffies = jiffies;
3354 unsigned long pending_updates;
dd41f596 3355 int i, scale;
46cb4b7c 3356
dd41f596 3357 this_rq->nr_load_updates++;
46cb4b7c 3358
fdf3e95d
VP
3359 /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3360 if (curr_jiffies == this_rq->last_load_update_tick)
3361 return;
3362
3363 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3364 this_rq->last_load_update_tick = curr_jiffies;
3365
dd41f596 3366 /* Update our load: */
fdf3e95d
VP
3367 this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3368 for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
dd41f596 3369 unsigned long old_load, new_load;
7d1e6a9b 3370
dd41f596 3371 /* scale is effectively 1 << i now, and >> i divides by scale */
46cb4b7c 3372
dd41f596 3373 old_load = this_rq->cpu_load[i];
fdf3e95d 3374 old_load = decay_load_missed(old_load, pending_updates - 1, i);
dd41f596 3375 new_load = this_load;
a25707f3
IM
3376 /*
3377 * Round up the averaging division if load is increasing. This
3378 * prevents us from getting stuck on 9 if the load is 10, for
3379 * example.
3380 */
3381 if (new_load > old_load)
fdf3e95d
VP
3382 new_load += scale - 1;
3383
3384 this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
dd41f596 3385 }
da2b71ed
SS
3386
3387 sched_avg_update(this_rq);
fdf3e95d
VP
3388}
3389
3390static void update_cpu_load_active(struct rq *this_rq)
3391{
3392 update_cpu_load(this_rq);
46cb4b7c 3393
74f5187a 3394 calc_load_account_active(this_rq);
46cb4b7c
SS
3395}
3396
dd41f596 3397#ifdef CONFIG_SMP
8a0be9ef 3398
46cb4b7c 3399/*
38022906
PZ
3400 * sched_exec - execve() is a valuable balancing opportunity, because at
3401 * this point the task has the smallest effective memory and cache footprint.
46cb4b7c 3402 */
38022906 3403void sched_exec(void)
46cb4b7c 3404{
38022906 3405 struct task_struct *p = current;
1da177e4 3406 unsigned long flags;
70b97a7f 3407 struct rq *rq;
0017d735 3408 int dest_cpu;
46cb4b7c 3409
1da177e4 3410 rq = task_rq_lock(p, &flags);
0017d735
PZ
3411 dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3412 if (dest_cpu == smp_processor_id())
3413 goto unlock;
38022906 3414
46cb4b7c 3415 /*
38022906 3416 * select_task_rq() can race against ->cpus_allowed
46cb4b7c 3417 */
30da688e 3418 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
b7a2b39d 3419 likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) {
969c7921 3420 struct migration_arg arg = { p, dest_cpu };
46cb4b7c 3421
1da177e4 3422 task_rq_unlock(rq, &flags);
969c7921 3423 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1da177e4
LT
3424 return;
3425 }
0017d735 3426unlock:
1da177e4 3427 task_rq_unlock(rq, &flags);
1da177e4 3428}
dd41f596 3429
1da177e4
LT
3430#endif
3431
1da177e4
LT
3432DEFINE_PER_CPU(struct kernel_stat, kstat);
3433
3434EXPORT_PER_CPU_SYMBOL(kstat);
3435
3436/*
c5f8d995 3437 * Return any ns on the sched_clock that have not yet been accounted in
f06febc9 3438 * @p in case that task is currently running.
c5f8d995
HS
3439 *
3440 * Called with task_rq_lock() held on @rq.
1da177e4 3441 */
c5f8d995
HS
3442static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3443{
3444 u64 ns = 0;
3445
3446 if (task_current(rq, p)) {
3447 update_rq_clock(rq);
305e6835 3448 ns = rq->clock_task - p->se.exec_start;
c5f8d995
HS
3449 if ((s64)ns < 0)
3450 ns = 0;
3451 }
3452
3453 return ns;
3454}
3455
bb34d92f 3456unsigned long long task_delta_exec(struct task_struct *p)
1da177e4 3457{
1da177e4 3458 unsigned long flags;
41b86e9c 3459 struct rq *rq;
bb34d92f 3460 u64 ns = 0;
48f24c4d 3461
41b86e9c 3462 rq = task_rq_lock(p, &flags);
c5f8d995
HS
3463 ns = do_task_delta_exec(p, rq);
3464 task_rq_unlock(rq, &flags);
1508487e 3465
c5f8d995
HS
3466 return ns;
3467}
f06febc9 3468
c5f8d995
HS
3469/*
3470 * Return accounted runtime for the task.
3471 * In case the task is currently running, return the runtime plus current's
3472 * pending runtime that have not been accounted yet.
3473 */
3474unsigned long long task_sched_runtime(struct task_struct *p)
3475{
3476 unsigned long flags;
3477 struct rq *rq;
3478 u64 ns = 0;
3479
3480 rq = task_rq_lock(p, &flags);
3481 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3482 task_rq_unlock(rq, &flags);
3483
3484 return ns;
3485}
48f24c4d 3486
c5f8d995
HS
3487/*
3488 * Return sum_exec_runtime for the thread group.
3489 * In case the task is currently running, return the sum plus current's
3490 * pending runtime that have not been accounted yet.
3491 *
3492 * Note that the thread group might have other running tasks as well,
3493 * so the return value not includes other pending runtime that other
3494 * running tasks might have.
3495 */
3496unsigned long long thread_group_sched_runtime(struct task_struct *p)
3497{
3498 struct task_cputime totals;
3499 unsigned long flags;
3500 struct rq *rq;
3501 u64 ns;
3502
3503 rq = task_rq_lock(p, &flags);
3504 thread_group_cputime(p, &totals);
3505 ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
41b86e9c 3506 task_rq_unlock(rq, &flags);
48f24c4d 3507
1da177e4
LT
3508 return ns;
3509}
3510
1da177e4
LT
3511/*
3512 * Account user cpu time to a process.
3513 * @p: the process that the cpu time gets accounted to
1da177e4 3514 * @cputime: the cpu time spent in user space since the last update
457533a7 3515 * @cputime_scaled: cputime scaled by cpu frequency
1da177e4 3516 */
457533a7
MS
3517void account_user_time(struct task_struct *p, cputime_t cputime,
3518 cputime_t cputime_scaled)
1da177e4
LT
3519{
3520 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3521 cputime64_t tmp;
3522
457533a7 3523 /* Add user time to process. */
1da177e4 3524 p->utime = cputime_add(p->utime, cputime);
457533a7 3525 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
f06febc9 3526 account_group_user_time(p, cputime);
1da177e4
LT
3527
3528 /* Add user time to cpustat. */
3529 tmp = cputime_to_cputime64(cputime);
3530 if (TASK_NICE(p) > 0)
3531 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3532 else
3533 cpustat->user = cputime64_add(cpustat->user, tmp);
ef12fefa
BR
3534
3535 cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
49b5cf34
JL
3536 /* Account for user time used */
3537 acct_update_integrals(p);
1da177e4
LT
3538}
3539
94886b84
LV
3540/*
3541 * Account guest cpu time to a process.
3542 * @p: the process that the cpu time gets accounted to
3543 * @cputime: the cpu time spent in virtual machine since the last update
457533a7 3544 * @cputime_scaled: cputime scaled by cpu frequency
94886b84 3545 */
457533a7
MS
3546static void account_guest_time(struct task_struct *p, cputime_t cputime,
3547 cputime_t cputime_scaled)
94886b84
LV
3548{
3549 cputime64_t tmp;
3550 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3551
3552 tmp = cputime_to_cputime64(cputime);
3553
457533a7 3554 /* Add guest time to process. */
94886b84 3555 p->utime = cputime_add(p->utime, cputime);
457533a7 3556 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
f06febc9 3557 account_group_user_time(p, cputime);
94886b84
LV
3558 p->gtime = cputime_add(p->gtime, cputime);
3559
457533a7 3560 /* Add guest time to cpustat. */
ce0e7b28
RO
3561 if (TASK_NICE(p) > 0) {
3562 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3563 cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3564 } else {
3565 cpustat->user = cputime64_add(cpustat->user, tmp);
3566 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3567 }
94886b84
LV
3568}
3569
1da177e4
LT
3570/*
3571 * Account system cpu time to a process.
3572 * @p: the process that the cpu time gets accounted to
3573 * @hardirq_offset: the offset to subtract from hardirq_count()
3574 * @cputime: the cpu time spent in kernel space since the last update
457533a7 3575 * @cputime_scaled: cputime scaled by cpu frequency
1da177e4
LT
3576 */
3577void account_system_time(struct task_struct *p, int hardirq_offset,
457533a7 3578 cputime_t cputime, cputime_t cputime_scaled)
1da177e4
LT
3579{
3580 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
1da177e4
LT
3581 cputime64_t tmp;
3582
983ed7a6 3583 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
457533a7 3584 account_guest_time(p, cputime, cputime_scaled);
983ed7a6
HH
3585 return;
3586 }
94886b84 3587
457533a7 3588 /* Add system time to process. */
1da177e4 3589 p->stime = cputime_add(p->stime, cputime);
457533a7 3590 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
f06febc9 3591 account_group_system_time(p, cputime);
1da177e4
LT
3592
3593 /* Add system time to cpustat. */
3594 tmp = cputime_to_cputime64(cputime);
3595 if (hardirq_count() - hardirq_offset)
3596 cpustat->irq = cputime64_add(cpustat->irq, tmp);
75e1056f 3597 else if (in_serving_softirq())
1da177e4 3598 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
1da177e4 3599 else
79741dd3
MS
3600 cpustat->system = cputime64_add(cpustat->system, tmp);
3601
ef12fefa
BR
3602 cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3603
1da177e4
LT
3604 /* Account for system time used */
3605 acct_update_integrals(p);
1da177e4
LT
3606}
3607
c66f08be 3608/*
1da177e4 3609 * Account for involuntary wait time.
1da177e4 3610 * @steal: the cpu time spent in involuntary wait
c66f08be 3611 */
79741dd3 3612void account_steal_time(cputime_t cputime)
c66f08be 3613{
79741dd3
MS
3614 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3615 cputime64_t cputime64 = cputime_to_cputime64(cputime);
3616
3617 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
c66f08be
MN
3618}
3619
1da177e4 3620/*
79741dd3
MS
3621 * Account for idle time.
3622 * @cputime: the cpu time spent in idle wait
1da177e4 3623 */
79741dd3 3624void account_idle_time(cputime_t cputime)
1da177e4
LT
3625{
3626 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
79741dd3 3627 cputime64_t cputime64 = cputime_to_cputime64(cputime);
70b97a7f 3628 struct rq *rq = this_rq();
1da177e4 3629
79741dd3
MS
3630 if (atomic_read(&rq->nr_iowait) > 0)
3631 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3632 else
3633 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
1da177e4
LT
3634}
3635
79741dd3
MS
3636#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3637
3638/*
3639 * Account a single tick of cpu time.
3640 * @p: the process that the cpu time gets accounted to
3641 * @user_tick: indicates if the tick is a user or a system tick
3642 */
3643void account_process_tick(struct task_struct *p, int user_tick)
3644{
a42548a1 3645 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
79741dd3
MS
3646 struct rq *rq = this_rq();
3647
3648 if (user_tick)
a42548a1 3649 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
f5f293a4 3650 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
a42548a1 3651 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
79741dd3
MS
3652 one_jiffy_scaled);
3653 else
a42548a1 3654 account_idle_time(cputime_one_jiffy);
79741dd3
MS
3655}
3656
3657/*
3658 * Account multiple ticks of steal time.
3659 * @p: the process from which the cpu time has been stolen
3660 * @ticks: number of stolen ticks
3661 */
3662void account_steal_ticks(unsigned long ticks)
3663{
3664 account_steal_time(jiffies_to_cputime(ticks));
3665}
3666
3667/*
3668 * Account multiple ticks of idle time.
3669 * @ticks: number of stolen ticks
3670 */
3671void account_idle_ticks(unsigned long ticks)
3672{
3673 account_idle_time(jiffies_to_cputime(ticks));
1da177e4
LT
3674}
3675
79741dd3
MS
3676#endif
3677
49048622
BS
3678/*
3679 * Use precise platform statistics if available:
3680 */
3681#ifdef CONFIG_VIRT_CPU_ACCOUNTING
d180c5bc 3682void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3683{
d99ca3b9
HS
3684 *ut = p->utime;
3685 *st = p->stime;
49048622
BS
3686}
3687
0cf55e1e 3688void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3689{
0cf55e1e
HS
3690 struct task_cputime cputime;
3691
3692 thread_group_cputime(p, &cputime);
3693
3694 *ut = cputime.utime;
3695 *st = cputime.stime;
49048622
BS
3696}
3697#else
761b1d26
HS
3698
3699#ifndef nsecs_to_cputime
b7b20df9 3700# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
761b1d26
HS
3701#endif
3702
d180c5bc 3703void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3704{
d99ca3b9 3705 cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
49048622
BS
3706
3707 /*
3708 * Use CFS's precise accounting:
3709 */
d180c5bc 3710 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
49048622
BS
3711
3712 if (total) {
e75e863d 3713 u64 temp = rtime;
d180c5bc 3714
e75e863d 3715 temp *= utime;
49048622 3716 do_div(temp, total);
d180c5bc
HS
3717 utime = (cputime_t)temp;
3718 } else
3719 utime = rtime;
49048622 3720
d180c5bc
HS
3721 /*
3722 * Compare with previous values, to keep monotonicity:
3723 */
761b1d26 3724 p->prev_utime = max(p->prev_utime, utime);
d99ca3b9 3725 p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
49048622 3726
d99ca3b9
HS
3727 *ut = p->prev_utime;
3728 *st = p->prev_stime;
49048622
BS
3729}
3730
0cf55e1e
HS
3731/*
3732 * Must be called with siglock held.
3733 */
3734void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3735{
0cf55e1e
HS
3736 struct signal_struct *sig = p->signal;
3737 struct task_cputime cputime;
3738 cputime_t rtime, utime, total;
49048622 3739
0cf55e1e 3740 thread_group_cputime(p, &cputime);
49048622 3741
0cf55e1e
HS
3742 total = cputime_add(cputime.utime, cputime.stime);
3743 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
49048622 3744
0cf55e1e 3745 if (total) {
e75e863d 3746 u64 temp = rtime;
49048622 3747
e75e863d 3748 temp *= cputime.utime;
0cf55e1e
HS
3749 do_div(temp, total);
3750 utime = (cputime_t)temp;
3751 } else
3752 utime = rtime;
3753
3754 sig->prev_utime = max(sig->prev_utime, utime);
3755 sig->prev_stime = max(sig->prev_stime,
3756 cputime_sub(rtime, sig->prev_utime));
3757
3758 *ut = sig->prev_utime;
3759 *st = sig->prev_stime;
49048622 3760}
49048622 3761#endif
49048622 3762
7835b98b
CL
3763/*
3764 * This function gets called by the timer code, with HZ frequency.
3765 * We call it with interrupts disabled.
3766 *
3767 * It also gets called by the fork code, when changing the parent's
3768 * timeslices.
3769 */
3770void scheduler_tick(void)
3771{
7835b98b
CL
3772 int cpu = smp_processor_id();
3773 struct rq *rq = cpu_rq(cpu);
dd41f596 3774 struct task_struct *curr = rq->curr;
3e51f33f
PZ
3775
3776 sched_clock_tick();
dd41f596 3777
05fa785c 3778 raw_spin_lock(&rq->lock);
3e51f33f 3779 update_rq_clock(rq);
fdf3e95d 3780 update_cpu_load_active(rq);
fa85ae24 3781 curr->sched_class->task_tick(rq, curr, 0);
05fa785c 3782 raw_spin_unlock(&rq->lock);
7835b98b 3783
e9d2b064 3784 perf_event_task_tick();
e220d2dc 3785
e418e1c2 3786#ifdef CONFIG_SMP
dd41f596
IM
3787 rq->idle_at_tick = idle_cpu(cpu);
3788 trigger_load_balance(rq, cpu);
e418e1c2 3789#endif
1da177e4
LT
3790}
3791
132380a0 3792notrace unsigned long get_parent_ip(unsigned long addr)
6cd8a4bb
SR
3793{
3794 if (in_lock_functions(addr)) {
3795 addr = CALLER_ADDR2;
3796 if (in_lock_functions(addr))
3797 addr = CALLER_ADDR3;
3798 }
3799 return addr;
3800}
1da177e4 3801
7e49fcce
SR
3802#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3803 defined(CONFIG_PREEMPT_TRACER))
3804
43627582 3805void __kprobes add_preempt_count(int val)
1da177e4 3806{
6cd8a4bb 3807#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3808 /*
3809 * Underflow?
3810 */
9a11b49a
IM
3811 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3812 return;
6cd8a4bb 3813#endif
1da177e4 3814 preempt_count() += val;
6cd8a4bb 3815#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3816 /*
3817 * Spinlock count overflowing soon?
3818 */
33859f7f
MOS
3819 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3820 PREEMPT_MASK - 10);
6cd8a4bb
SR
3821#endif
3822 if (preempt_count() == val)
3823 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
1da177e4
LT
3824}
3825EXPORT_SYMBOL(add_preempt_count);
3826
43627582 3827void __kprobes sub_preempt_count(int val)
1da177e4 3828{
6cd8a4bb 3829#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3830 /*
3831 * Underflow?
3832 */
01e3eb82 3833 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
9a11b49a 3834 return;
1da177e4
LT
3835 /*
3836 * Is the spinlock portion underflowing?
3837 */
9a11b49a
IM
3838 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3839 !(preempt_count() & PREEMPT_MASK)))
3840 return;
6cd8a4bb 3841#endif
9a11b49a 3842
6cd8a4bb
SR
3843 if (preempt_count() == val)
3844 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
1da177e4
LT
3845 preempt_count() -= val;
3846}
3847EXPORT_SYMBOL(sub_preempt_count);
3848
3849#endif
3850
3851/*
dd41f596 3852 * Print scheduling while atomic bug:
1da177e4 3853 */
dd41f596 3854static noinline void __schedule_bug(struct task_struct *prev)
1da177e4 3855{
838225b4
SS
3856 struct pt_regs *regs = get_irq_regs();
3857
3df0fc5b
PZ
3858 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3859 prev->comm, prev->pid, preempt_count());
838225b4 3860
dd41f596 3861 debug_show_held_locks(prev);
e21f5b15 3862 print_modules();
dd41f596
IM
3863 if (irqs_disabled())
3864 print_irqtrace_events(prev);
838225b4
SS
3865
3866 if (regs)
3867 show_regs(regs);
3868 else
3869 dump_stack();
dd41f596 3870}
1da177e4 3871
dd41f596
IM
3872/*
3873 * Various schedule()-time debugging checks and statistics:
3874 */
3875static inline void schedule_debug(struct task_struct *prev)
3876{
1da177e4 3877 /*
41a2d6cf 3878 * Test if we are atomic. Since do_exit() needs to call into
1da177e4
LT
3879 * schedule() atomically, we ignore that path for now.
3880 * Otherwise, whine if we are scheduling when we should not be.
3881 */
3f33a7ce 3882 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
dd41f596
IM
3883 __schedule_bug(prev);
3884
1da177e4
LT
3885 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3886
2d72376b 3887 schedstat_inc(this_rq(), sched_count);
b8efb561
IM
3888#ifdef CONFIG_SCHEDSTATS
3889 if (unlikely(prev->lock_depth >= 0)) {
2d72376b
IM
3890 schedstat_inc(this_rq(), bkl_count);
3891 schedstat_inc(prev, sched_info.bkl_count);
b8efb561
IM
3892 }
3893#endif
dd41f596
IM
3894}
3895
6cecd084 3896static void put_prev_task(struct rq *rq, struct task_struct *prev)
df1c99d4 3897{
a64692a3
MG
3898 if (prev->se.on_rq)
3899 update_rq_clock(rq);
6cecd084 3900 prev->sched_class->put_prev_task(rq, prev);
df1c99d4
MG
3901}
3902
dd41f596
IM
3903/*
3904 * Pick up the highest-prio task:
3905 */
3906static inline struct task_struct *
b67802ea 3907pick_next_task(struct rq *rq)
dd41f596 3908{
5522d5d5 3909 const struct sched_class *class;
dd41f596 3910 struct task_struct *p;
1da177e4
LT
3911
3912 /*
dd41f596
IM
3913 * Optimization: we know that if all tasks are in
3914 * the fair class we can call that function directly:
1da177e4 3915 */
dd41f596 3916 if (likely(rq->nr_running == rq->cfs.nr_running)) {
fb8d4724 3917 p = fair_sched_class.pick_next_task(rq);
dd41f596
IM
3918 if (likely(p))
3919 return p;
1da177e4
LT
3920 }
3921
34f971f6 3922 for_each_class(class) {
fb8d4724 3923 p = class->pick_next_task(rq);
dd41f596
IM
3924 if (p)
3925 return p;
dd41f596 3926 }
34f971f6
PZ
3927
3928 BUG(); /* the idle class will always have a runnable task */
dd41f596 3929}
1da177e4 3930
dd41f596
IM
3931/*
3932 * schedule() is the main scheduler function.
3933 */
ff743345 3934asmlinkage void __sched schedule(void)
dd41f596
IM
3935{
3936 struct task_struct *prev, *next;
67ca7bde 3937 unsigned long *switch_count;
dd41f596 3938 struct rq *rq;
31656519 3939 int cpu;
dd41f596 3940
ff743345
PZ
3941need_resched:
3942 preempt_disable();
dd41f596
IM
3943 cpu = smp_processor_id();
3944 rq = cpu_rq(cpu);
25502a6c 3945 rcu_note_context_switch(cpu);
dd41f596 3946 prev = rq->curr;
dd41f596
IM
3947
3948 release_kernel_lock(prev);
3949need_resched_nonpreemptible:
3950
3951 schedule_debug(prev);
1da177e4 3952
31656519 3953 if (sched_feat(HRTICK))
f333fdc9 3954 hrtick_clear(rq);
8f4d37ec 3955
05fa785c 3956 raw_spin_lock_irq(&rq->lock);
1da177e4 3957
246d86b5 3958 switch_count = &prev->nivcsw;
1da177e4 3959 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
21aa9af0 3960 if (unlikely(signal_pending_state(prev->state, prev))) {
1da177e4 3961 prev->state = TASK_RUNNING;
21aa9af0
TH
3962 } else {
3963 /*
3964 * If a worker is going to sleep, notify and
3965 * ask workqueue whether it wants to wake up a
3966 * task to maintain concurrency. If so, wake
3967 * up the task.
3968 */
3969 if (prev->flags & PF_WQ_WORKER) {
3970 struct task_struct *to_wakeup;
3971
3972 to_wakeup = wq_worker_sleeping(prev, cpu);
3973 if (to_wakeup)
3974 try_to_wake_up_local(to_wakeup);
3975 }
371fd7e7 3976 deactivate_task(rq, prev, DEQUEUE_SLEEP);
21aa9af0 3977 }
dd41f596 3978 switch_count = &prev->nvcsw;
1da177e4
LT
3979 }
3980
3f029d3c 3981 pre_schedule(rq, prev);
f65eda4f 3982
dd41f596 3983 if (unlikely(!rq->nr_running))
1da177e4 3984 idle_balance(cpu, rq);
1da177e4 3985
df1c99d4 3986 put_prev_task(rq, prev);
b67802ea 3987 next = pick_next_task(rq);
f26f9aff
MG
3988 clear_tsk_need_resched(prev);
3989 rq->skip_clock_update = 0;
1da177e4 3990
1da177e4 3991 if (likely(prev != next)) {
673a90a1 3992 sched_info_switch(prev, next);
49f47433 3993 perf_event_task_sched_out(prev, next);
673a90a1 3994
1da177e4
LT
3995 rq->nr_switches++;
3996 rq->curr = next;
3997 ++*switch_count;
3998
dd41f596 3999 context_switch(rq, prev, next); /* unlocks the rq */
8f4d37ec 4000 /*
246d86b5
ON
4001 * The context switch have flipped the stack from under us
4002 * and restored the local variables which were saved when
4003 * this task called schedule() in the past. prev == current
4004 * is still correct, but it can be moved to another cpu/rq.
8f4d37ec
PZ
4005 */
4006 cpu = smp_processor_id();
4007 rq = cpu_rq(cpu);
1da177e4 4008 } else
05fa785c 4009 raw_spin_unlock_irq(&rq->lock);
1da177e4 4010
3f029d3c 4011 post_schedule(rq);
1da177e4 4012
246d86b5 4013 if (unlikely(reacquire_kernel_lock(prev)))
1da177e4 4014 goto need_resched_nonpreemptible;
8f4d37ec 4015
1da177e4 4016 preempt_enable_no_resched();
ff743345 4017 if (need_resched())
1da177e4
LT
4018 goto need_resched;
4019}
1da177e4
LT
4020EXPORT_SYMBOL(schedule);
4021
c08f7829 4022#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
0d66bf6d
PZ
4023/*
4024 * Look out! "owner" is an entirely speculative pointer
4025 * access and not reliable.
4026 */
4027int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
4028{
4029 unsigned int cpu;
4030 struct rq *rq;
4031
4032 if (!sched_feat(OWNER_SPIN))
4033 return 0;
4034
4035#ifdef CONFIG_DEBUG_PAGEALLOC
4036 /*
4037 * Need to access the cpu field knowing that
4038 * DEBUG_PAGEALLOC could have unmapped it if
4039 * the mutex owner just released it and exited.
4040 */
4041 if (probe_kernel_address(&owner->cpu, cpu))
4b402210 4042 return 0;
0d66bf6d
PZ
4043#else
4044 cpu = owner->cpu;
4045#endif
4046
4047 /*
4048 * Even if the access succeeded (likely case),
4049 * the cpu field may no longer be valid.
4050 */
4051 if (cpu >= nr_cpumask_bits)
4b402210 4052 return 0;
0d66bf6d
PZ
4053
4054 /*
4055 * We need to validate that we can do a
4056 * get_cpu() and that we have the percpu area.
4057 */
4058 if (!cpu_online(cpu))
4b402210 4059 return 0;
0d66bf6d
PZ
4060
4061 rq = cpu_rq(cpu);
4062
4063 for (;;) {
4064 /*
4065 * Owner changed, break to re-assess state.
4066 */
9d0f4dcc
TC
4067 if (lock->owner != owner) {
4068 /*
4069 * If the lock has switched to a different owner,
4070 * we likely have heavy contention. Return 0 to quit
4071 * optimistic spinning and not contend further:
4072 */
4073 if (lock->owner)
4074 return 0;
0d66bf6d 4075 break;
9d0f4dcc 4076 }
0d66bf6d
PZ
4077
4078 /*
4079 * Is that owner really running on that cpu?
4080 */
4081 if (task_thread_info(rq->curr) != owner || need_resched())
4082 return 0;
4083
335d7afb 4084 arch_mutex_cpu_relax();
0d66bf6d 4085 }
4b402210 4086
0d66bf6d
PZ
4087 return 1;
4088}
4089#endif
4090
1da177e4
LT
4091#ifdef CONFIG_PREEMPT
4092/*
2ed6e34f 4093 * this is the entry point to schedule() from in-kernel preemption
41a2d6cf 4094 * off of preempt_enable. Kernel preemptions off return from interrupt
1da177e4
LT
4095 * occur there and call schedule directly.
4096 */
d1f74e20 4097asmlinkage void __sched notrace preempt_schedule(void)
1da177e4
LT
4098{
4099 struct thread_info *ti = current_thread_info();
6478d880 4100
1da177e4
LT
4101 /*
4102 * If there is a non-zero preempt_count or interrupts are disabled,
41a2d6cf 4103 * we do not want to preempt the current task. Just return..
1da177e4 4104 */
beed33a8 4105 if (likely(ti->preempt_count || irqs_disabled()))
1da177e4
LT
4106 return;
4107
3a5c359a 4108 do {
d1f74e20 4109 add_preempt_count_notrace(PREEMPT_ACTIVE);
3a5c359a 4110 schedule();
d1f74e20 4111 sub_preempt_count_notrace(PREEMPT_ACTIVE);
1da177e4 4112
3a5c359a
AK
4113 /*
4114 * Check again in case we missed a preemption opportunity
4115 * between schedule and now.
4116 */
4117 barrier();
5ed0cec0 4118 } while (need_resched());
1da177e4 4119}
1da177e4
LT
4120EXPORT_SYMBOL(preempt_schedule);
4121
4122/*
2ed6e34f 4123 * this is the entry point to schedule() from kernel preemption
1da177e4
LT
4124 * off of irq context.
4125 * Note, that this is called and return with irqs disabled. This will
4126 * protect us against recursive calling from irq.
4127 */
4128asmlinkage void __sched preempt_schedule_irq(void)
4129{
4130 struct thread_info *ti = current_thread_info();
6478d880 4131
2ed6e34f 4132 /* Catch callers which need to be fixed */
1da177e4
LT
4133 BUG_ON(ti->preempt_count || !irqs_disabled());
4134
3a5c359a
AK
4135 do {
4136 add_preempt_count(PREEMPT_ACTIVE);
3a5c359a
AK
4137 local_irq_enable();
4138 schedule();
4139 local_irq_disable();
3a5c359a 4140 sub_preempt_count(PREEMPT_ACTIVE);
1da177e4 4141
3a5c359a
AK
4142 /*
4143 * Check again in case we missed a preemption opportunity
4144 * between schedule and now.
4145 */
4146 barrier();
5ed0cec0 4147 } while (need_resched());
1da177e4
LT
4148}
4149
4150#endif /* CONFIG_PREEMPT */
4151
63859d4f 4152int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
95cdf3b7 4153 void *key)
1da177e4 4154{
63859d4f 4155 return try_to_wake_up(curr->private, mode, wake_flags);
1da177e4 4156}
1da177e4
LT
4157EXPORT_SYMBOL(default_wake_function);
4158
4159/*
41a2d6cf
IM
4160 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4161 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
1da177e4
LT
4162 * number) then we wake all the non-exclusive tasks and one exclusive task.
4163 *
4164 * There are circumstances in which we can try to wake a task which has already
41a2d6cf 4165 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
1da177e4
LT
4166 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4167 */
78ddb08f 4168static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
63859d4f 4169 int nr_exclusive, int wake_flags, void *key)
1da177e4 4170{
2e45874c 4171 wait_queue_t *curr, *next;
1da177e4 4172
2e45874c 4173 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
48f24c4d
IM
4174 unsigned flags = curr->flags;
4175
63859d4f 4176 if (curr->func(curr, mode, wake_flags, key) &&
48f24c4d 4177 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
1da177e4
LT
4178 break;
4179 }
4180}
4181
4182/**
4183 * __wake_up - wake up threads blocked on a waitqueue.
4184 * @q: the waitqueue
4185 * @mode: which threads
4186 * @nr_exclusive: how many wake-one or wake-many threads to wake up
67be2dd1 4187 * @key: is directly passed to the wakeup function
50fa610a
DH
4188 *
4189 * It may be assumed that this function implies a write memory barrier before
4190 * changing the task state if and only if any tasks are woken up.
1da177e4 4191 */
7ad5b3a5 4192void __wake_up(wait_queue_head_t *q, unsigned int mode,
95cdf3b7 4193 int nr_exclusive, void *key)
1da177e4
LT
4194{
4195 unsigned long flags;
4196
4197 spin_lock_irqsave(&q->lock, flags);
4198 __wake_up_common(q, mode, nr_exclusive, 0, key);
4199 spin_unlock_irqrestore(&q->lock, flags);
4200}
1da177e4
LT
4201EXPORT_SYMBOL(__wake_up);
4202
4203/*
4204 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4205 */
7ad5b3a5 4206void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
1da177e4
LT
4207{
4208 __wake_up_common(q, mode, 1, 0, NULL);
4209}
22c43c81 4210EXPORT_SYMBOL_GPL(__wake_up_locked);
1da177e4 4211
4ede816a
DL
4212void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4213{
4214 __wake_up_common(q, mode, 1, 0, key);
4215}
4216
1da177e4 4217/**
4ede816a 4218 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
1da177e4
LT
4219 * @q: the waitqueue
4220 * @mode: which threads
4221 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4ede816a 4222 * @key: opaque value to be passed to wakeup targets
1da177e4
LT
4223 *
4224 * The sync wakeup differs that the waker knows that it will schedule
4225 * away soon, so while the target thread will be woken up, it will not
4226 * be migrated to another CPU - ie. the two threads are 'synchronized'
4227 * with each other. This can prevent needless bouncing between CPUs.
4228 *
4229 * On UP it can prevent extra preemption.
50fa610a
DH
4230 *
4231 * It may be assumed that this function implies a write memory barrier before
4232 * changing the task state if and only if any tasks are woken up.
1da177e4 4233 */
4ede816a
DL
4234void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4235 int nr_exclusive, void *key)
1da177e4
LT
4236{
4237 unsigned long flags;
7d478721 4238 int wake_flags = WF_SYNC;
1da177e4
LT
4239
4240 if (unlikely(!q))
4241 return;
4242
4243 if (unlikely(!nr_exclusive))
7d478721 4244 wake_flags = 0;
1da177e4
LT
4245
4246 spin_lock_irqsave(&q->lock, flags);
7d478721 4247 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
1da177e4
LT
4248 spin_unlock_irqrestore(&q->lock, flags);
4249}
4ede816a
DL
4250EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4251
4252/*
4253 * __wake_up_sync - see __wake_up_sync_key()
4254 */
4255void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4256{
4257 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4258}
1da177e4
LT
4259EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4260
65eb3dc6
KD
4261/**
4262 * complete: - signals a single thread waiting on this completion
4263 * @x: holds the state of this particular completion
4264 *
4265 * This will wake up a single thread waiting on this completion. Threads will be
4266 * awakened in the same order in which they were queued.
4267 *
4268 * See also complete_all(), wait_for_completion() and related routines.
50fa610a
DH
4269 *
4270 * It may be assumed that this function implies a write memory barrier before
4271 * changing the task state if and only if any tasks are woken up.
65eb3dc6 4272 */
b15136e9 4273void complete(struct completion *x)
1da177e4
LT
4274{
4275 unsigned long flags;
4276
4277 spin_lock_irqsave(&x->wait.lock, flags);
4278 x->done++;
d9514f6c 4279 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
1da177e4
LT
4280 spin_unlock_irqrestore(&x->wait.lock, flags);
4281}
4282EXPORT_SYMBOL(complete);
4283
65eb3dc6
KD
4284/**
4285 * complete_all: - signals all threads waiting on this completion
4286 * @x: holds the state of this particular completion
4287 *
4288 * This will wake up all threads waiting on this particular completion event.
50fa610a
DH
4289 *
4290 * It may be assumed that this function implies a write memory barrier before
4291 * changing the task state if and only if any tasks are woken up.
65eb3dc6 4292 */
b15136e9 4293void complete_all(struct completion *x)
1da177e4
LT
4294{
4295 unsigned long flags;
4296
4297 spin_lock_irqsave(&x->wait.lock, flags);
4298 x->done += UINT_MAX/2;
d9514f6c 4299 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
1da177e4
LT
4300 spin_unlock_irqrestore(&x->wait.lock, flags);
4301}
4302EXPORT_SYMBOL(complete_all);
4303
8cbbe86d
AK
4304static inline long __sched
4305do_wait_for_common(struct completion *x, long timeout, int state)
1da177e4 4306{
1da177e4
LT
4307 if (!x->done) {
4308 DECLARE_WAITQUEUE(wait, current);
4309
a93d2f17 4310 __add_wait_queue_tail_exclusive(&x->wait, &wait);
1da177e4 4311 do {
94d3d824 4312 if (signal_pending_state(state, current)) {
ea71a546
ON
4313 timeout = -ERESTARTSYS;
4314 break;
8cbbe86d
AK
4315 }
4316 __set_current_state(state);
1da177e4
LT
4317 spin_unlock_irq(&x->wait.lock);
4318 timeout = schedule_timeout(timeout);
4319 spin_lock_irq(&x->wait.lock);
ea71a546 4320 } while (!x->done && timeout);
1da177e4 4321 __remove_wait_queue(&x->wait, &wait);
ea71a546
ON
4322 if (!x->done)
4323 return timeout;
1da177e4
LT
4324 }
4325 x->done--;
ea71a546 4326 return timeout ?: 1;
1da177e4 4327}
1da177e4 4328
8cbbe86d
AK
4329static long __sched
4330wait_for_common(struct completion *x, long timeout, int state)
1da177e4 4331{
1da177e4
LT
4332 might_sleep();
4333
4334 spin_lock_irq(&x->wait.lock);
8cbbe86d 4335 timeout = do_wait_for_common(x, timeout, state);
1da177e4 4336 spin_unlock_irq(&x->wait.lock);
8cbbe86d
AK
4337 return timeout;
4338}
1da177e4 4339
65eb3dc6
KD
4340/**
4341 * wait_for_completion: - waits for completion of a task
4342 * @x: holds the state of this particular completion
4343 *
4344 * This waits to be signaled for completion of a specific task. It is NOT
4345 * interruptible and there is no timeout.
4346 *
4347 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4348 * and interrupt capability. Also see complete().
4349 */
b15136e9 4350void __sched wait_for_completion(struct completion *x)
8cbbe86d
AK
4351{
4352 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
1da177e4 4353}
8cbbe86d 4354EXPORT_SYMBOL(wait_for_completion);
1da177e4 4355
65eb3dc6
KD
4356/**
4357 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4358 * @x: holds the state of this particular completion
4359 * @timeout: timeout value in jiffies
4360 *
4361 * This waits for either a completion of a specific task to be signaled or for a
4362 * specified timeout to expire. The timeout is in jiffies. It is not
4363 * interruptible.
4364 */
b15136e9 4365unsigned long __sched
8cbbe86d 4366wait_for_completion_timeout(struct completion *x, unsigned long timeout)
1da177e4 4367{
8cbbe86d 4368 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
1da177e4 4369}
8cbbe86d 4370EXPORT_SYMBOL(wait_for_completion_timeout);
1da177e4 4371
65eb3dc6
KD
4372/**
4373 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4374 * @x: holds the state of this particular completion
4375 *
4376 * This waits for completion of a specific task to be signaled. It is
4377 * interruptible.
4378 */
8cbbe86d 4379int __sched wait_for_completion_interruptible(struct completion *x)
0fec171c 4380{
51e97990
AK
4381 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4382 if (t == -ERESTARTSYS)
4383 return t;
4384 return 0;
0fec171c 4385}
8cbbe86d 4386EXPORT_SYMBOL(wait_for_completion_interruptible);
1da177e4 4387
65eb3dc6
KD
4388/**
4389 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4390 * @x: holds the state of this particular completion
4391 * @timeout: timeout value in jiffies
4392 *
4393 * This waits for either a completion of a specific task to be signaled or for a
4394 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4395 */
6bf41237 4396long __sched
8cbbe86d
AK
4397wait_for_completion_interruptible_timeout(struct completion *x,
4398 unsigned long timeout)
0fec171c 4399{
8cbbe86d 4400 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
0fec171c 4401}
8cbbe86d 4402EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
1da177e4 4403
65eb3dc6
KD
4404/**
4405 * wait_for_completion_killable: - waits for completion of a task (killable)
4406 * @x: holds the state of this particular completion
4407 *
4408 * This waits to be signaled for completion of a specific task. It can be
4409 * interrupted by a kill signal.
4410 */
009e577e
MW
4411int __sched wait_for_completion_killable(struct completion *x)
4412{
4413 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4414 if (t == -ERESTARTSYS)
4415 return t;
4416 return 0;
4417}
4418EXPORT_SYMBOL(wait_for_completion_killable);
4419
0aa12fb4
SW
4420/**
4421 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4422 * @x: holds the state of this particular completion
4423 * @timeout: timeout value in jiffies
4424 *
4425 * This waits for either a completion of a specific task to be
4426 * signaled or for a specified timeout to expire. It can be
4427 * interrupted by a kill signal. The timeout is in jiffies.
4428 */
6bf41237 4429long __sched
0aa12fb4
SW
4430wait_for_completion_killable_timeout(struct completion *x,
4431 unsigned long timeout)
4432{
4433 return wait_for_common(x, timeout, TASK_KILLABLE);
4434}
4435EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4436
be4de352
DC
4437/**
4438 * try_wait_for_completion - try to decrement a completion without blocking
4439 * @x: completion structure
4440 *
4441 * Returns: 0 if a decrement cannot be done without blocking
4442 * 1 if a decrement succeeded.
4443 *
4444 * If a completion is being used as a counting completion,
4445 * attempt to decrement the counter without blocking. This
4446 * enables us to avoid waiting if the resource the completion
4447 * is protecting is not available.
4448 */
4449bool try_wait_for_completion(struct completion *x)
4450{
7539a3b3 4451 unsigned long flags;
be4de352
DC
4452 int ret = 1;
4453
7539a3b3 4454 spin_lock_irqsave(&x->wait.lock, flags);
be4de352
DC
4455 if (!x->done)
4456 ret = 0;
4457 else
4458 x->done--;
7539a3b3 4459 spin_unlock_irqrestore(&x->wait.lock, flags);
be4de352
DC
4460 return ret;
4461}
4462EXPORT_SYMBOL(try_wait_for_completion);
4463
4464/**
4465 * completion_done - Test to see if a completion has any waiters
4466 * @x: completion structure
4467 *
4468 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4469 * 1 if there are no waiters.
4470 *
4471 */
4472bool completion_done(struct completion *x)
4473{
7539a3b3 4474 unsigned long flags;
be4de352
DC
4475 int ret = 1;
4476
7539a3b3 4477 spin_lock_irqsave(&x->wait.lock, flags);
be4de352
DC
4478 if (!x->done)
4479 ret = 0;
7539a3b3 4480 spin_unlock_irqrestore(&x->wait.lock, flags);
be4de352
DC
4481 return ret;
4482}
4483EXPORT_SYMBOL(completion_done);
4484
8cbbe86d
AK
4485static long __sched
4486sleep_on_common(wait_queue_head_t *q, int state, long timeout)
1da177e4 4487{
0fec171c
IM
4488 unsigned long flags;
4489 wait_queue_t wait;
4490
4491 init_waitqueue_entry(&wait, current);
1da177e4 4492
8cbbe86d 4493 __set_current_state(state);
1da177e4 4494
8cbbe86d
AK
4495 spin_lock_irqsave(&q->lock, flags);
4496 __add_wait_queue(q, &wait);
4497 spin_unlock(&q->lock);
4498 timeout = schedule_timeout(timeout);
4499 spin_lock_irq(&q->lock);
4500 __remove_wait_queue(q, &wait);
4501 spin_unlock_irqrestore(&q->lock, flags);
4502
4503 return timeout;
4504}
4505
4506void __sched interruptible_sleep_on(wait_queue_head_t *q)
4507{
4508 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
1da177e4 4509}
1da177e4
LT
4510EXPORT_SYMBOL(interruptible_sleep_on);
4511
0fec171c 4512long __sched
95cdf3b7 4513interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
1da177e4 4514{
8cbbe86d 4515 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
1da177e4 4516}
1da177e4
LT
4517EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4518
0fec171c 4519void __sched sleep_on(wait_queue_head_t *q)
1da177e4 4520{
8cbbe86d 4521 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
1da177e4 4522}
1da177e4
LT
4523EXPORT_SYMBOL(sleep_on);
4524
0fec171c 4525long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
1da177e4 4526{
8cbbe86d 4527 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
1da177e4 4528}
1da177e4
LT
4529EXPORT_SYMBOL(sleep_on_timeout);
4530
b29739f9
IM
4531#ifdef CONFIG_RT_MUTEXES
4532
4533/*
4534 * rt_mutex_setprio - set the current priority of a task
4535 * @p: task
4536 * @prio: prio value (kernel-internal form)
4537 *
4538 * This function changes the 'effective' priority of a task. It does
4539 * not touch ->normal_prio like __setscheduler().
4540 *
4541 * Used by the rt_mutex code to implement priority inheritance logic.
4542 */
36c8b586 4543void rt_mutex_setprio(struct task_struct *p, int prio)
b29739f9
IM
4544{
4545 unsigned long flags;
83b699ed 4546 int oldprio, on_rq, running;
70b97a7f 4547 struct rq *rq;
83ab0aa0 4548 const struct sched_class *prev_class;
b29739f9
IM
4549
4550 BUG_ON(prio < 0 || prio > MAX_PRIO);
4551
4552 rq = task_rq_lock(p, &flags);
4553
a8027073 4554 trace_sched_pi_setprio(p, prio);
d5f9f942 4555 oldprio = p->prio;
83ab0aa0 4556 prev_class = p->sched_class;
dd41f596 4557 on_rq = p->se.on_rq;
051a1d1a 4558 running = task_current(rq, p);
0e1f3483 4559 if (on_rq)
69be72c1 4560 dequeue_task(rq, p, 0);
0e1f3483
HS
4561 if (running)
4562 p->sched_class->put_prev_task(rq, p);
dd41f596
IM
4563
4564 if (rt_prio(prio))
4565 p->sched_class = &rt_sched_class;
4566 else
4567 p->sched_class = &fair_sched_class;
4568
b29739f9
IM
4569 p->prio = prio;
4570
0e1f3483
HS
4571 if (running)
4572 p->sched_class->set_curr_task(rq);
dd41f596 4573 if (on_rq) {
371fd7e7 4574 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
cb469845
SR
4575
4576 check_class_changed(rq, p, prev_class, oldprio, running);
b29739f9
IM
4577 }
4578 task_rq_unlock(rq, &flags);
4579}
4580
4581#endif
4582
36c8b586 4583void set_user_nice(struct task_struct *p, long nice)
1da177e4 4584{
dd41f596 4585 int old_prio, delta, on_rq;
1da177e4 4586 unsigned long flags;
70b97a7f 4587 struct rq *rq;
1da177e4
LT
4588
4589 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4590 return;
4591 /*
4592 * We have to be careful, if called from sys_setpriority(),
4593 * the task might be in the middle of scheduling on another CPU.
4594 */
4595 rq = task_rq_lock(p, &flags);
4596 /*
4597 * The RT priorities are set via sched_setscheduler(), but we still
4598 * allow the 'normal' nice value to be set - but as expected
4599 * it wont have any effect on scheduling until the task is
dd41f596 4600 * SCHED_FIFO/SCHED_RR:
1da177e4 4601 */
e05606d3 4602 if (task_has_rt_policy(p)) {
1da177e4
LT
4603 p->static_prio = NICE_TO_PRIO(nice);
4604 goto out_unlock;
4605 }
dd41f596 4606 on_rq = p->se.on_rq;
c09595f6 4607 if (on_rq)
69be72c1 4608 dequeue_task(rq, p, 0);
1da177e4 4609
1da177e4 4610 p->static_prio = NICE_TO_PRIO(nice);
2dd73a4f 4611 set_load_weight(p);
b29739f9
IM
4612 old_prio = p->prio;
4613 p->prio = effective_prio(p);
4614 delta = p->prio - old_prio;
1da177e4 4615
dd41f596 4616 if (on_rq) {
371fd7e7 4617 enqueue_task(rq, p, 0);
1da177e4 4618 /*
d5f9f942
AM
4619 * If the task increased its priority or is running and
4620 * lowered its priority, then reschedule its CPU:
1da177e4 4621 */
d5f9f942 4622 if (delta < 0 || (delta > 0 && task_running(rq, p)))
1da177e4
LT
4623 resched_task(rq->curr);
4624 }
4625out_unlock:
4626 task_rq_unlock(rq, &flags);
4627}
1da177e4
LT
4628EXPORT_SYMBOL(set_user_nice);
4629
e43379f1
MM
4630/*
4631 * can_nice - check if a task can reduce its nice value
4632 * @p: task
4633 * @nice: nice value
4634 */
36c8b586 4635int can_nice(const struct task_struct *p, const int nice)
e43379f1 4636{
024f4747
MM
4637 /* convert nice value [19,-20] to rlimit style value [1,40] */
4638 int nice_rlim = 20 - nice;
48f24c4d 4639
78d7d407 4640 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
e43379f1
MM
4641 capable(CAP_SYS_NICE));
4642}
4643
1da177e4
LT
4644#ifdef __ARCH_WANT_SYS_NICE
4645
4646/*
4647 * sys_nice - change the priority of the current process.
4648 * @increment: priority increment
4649 *
4650 * sys_setpriority is a more generic, but much slower function that
4651 * does similar things.
4652 */
5add95d4 4653SYSCALL_DEFINE1(nice, int, increment)
1da177e4 4654{
48f24c4d 4655 long nice, retval;
1da177e4
LT
4656
4657 /*
4658 * Setpriority might change our priority at the same moment.
4659 * We don't have to worry. Conceptually one call occurs first
4660 * and we have a single winner.
4661 */
e43379f1
MM
4662 if (increment < -40)
4663 increment = -40;
1da177e4
LT
4664 if (increment > 40)
4665 increment = 40;
4666
2b8f836f 4667 nice = TASK_NICE(current) + increment;
1da177e4
LT
4668 if (nice < -20)
4669 nice = -20;
4670 if (nice > 19)
4671 nice = 19;
4672
e43379f1
MM
4673 if (increment < 0 && !can_nice(current, nice))
4674 return -EPERM;
4675
1da177e4
LT
4676 retval = security_task_setnice(current, nice);
4677 if (retval)
4678 return retval;
4679
4680 set_user_nice(current, nice);
4681 return 0;
4682}
4683
4684#endif
4685
4686/**
4687 * task_prio - return the priority value of a given task.
4688 * @p: the task in question.
4689 *
4690 * This is the priority value as seen by users in /proc.
4691 * RT tasks are offset by -200. Normal tasks are centered
4692 * around 0, value goes from -16 to +15.
4693 */
36c8b586 4694int task_prio(const struct task_struct *p)
1da177e4
LT
4695{
4696 return p->prio - MAX_RT_PRIO;
4697}
4698
4699/**
4700 * task_nice - return the nice value of a given task.
4701 * @p: the task in question.
4702 */
36c8b586 4703int task_nice(const struct task_struct *p)
1da177e4
LT
4704{
4705 return TASK_NICE(p);
4706}
150d8bed 4707EXPORT_SYMBOL(task_nice);
1da177e4
LT
4708
4709/**
4710 * idle_cpu - is a given cpu idle currently?
4711 * @cpu: the processor in question.
4712 */
4713int idle_cpu(int cpu)
4714{
4715 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4716}
4717
1da177e4
LT
4718/**
4719 * idle_task - return the idle task for a given cpu.
4720 * @cpu: the processor in question.
4721 */
36c8b586 4722struct task_struct *idle_task(int cpu)
1da177e4
LT
4723{
4724 return cpu_rq(cpu)->idle;
4725}
4726
4727/**
4728 * find_process_by_pid - find a process with a matching PID value.
4729 * @pid: the pid in question.
4730 */
a9957449 4731static struct task_struct *find_process_by_pid(pid_t pid)
1da177e4 4732{
228ebcbe 4733 return pid ? find_task_by_vpid(pid) : current;
1da177e4
LT
4734}
4735
4736/* Actually do priority change: must hold rq lock. */
dd41f596
IM
4737static void
4738__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
1da177e4 4739{
dd41f596 4740 BUG_ON(p->se.on_rq);
48f24c4d 4741
1da177e4
LT
4742 p->policy = policy;
4743 p->rt_priority = prio;
b29739f9
IM
4744 p->normal_prio = normal_prio(p);
4745 /* we are holding p->pi_lock already */
4746 p->prio = rt_mutex_getprio(p);
ffd44db5
PZ
4747 if (rt_prio(p->prio))
4748 p->sched_class = &rt_sched_class;
4749 else
4750 p->sched_class = &fair_sched_class;
2dd73a4f 4751 set_load_weight(p);
1da177e4
LT
4752}
4753
c69e8d9c
DH
4754/*
4755 * check the target process has a UID that matches the current process's
4756 */
4757static bool check_same_owner(struct task_struct *p)
4758{
4759 const struct cred *cred = current_cred(), *pcred;
4760 bool match;
4761
4762 rcu_read_lock();
4763 pcred = __task_cred(p);
4764 match = (cred->euid == pcred->euid ||
4765 cred->euid == pcred->uid);
4766 rcu_read_unlock();
4767 return match;
4768}
4769
961ccddd 4770static int __sched_setscheduler(struct task_struct *p, int policy,
fe7de49f 4771 const struct sched_param *param, bool user)
1da177e4 4772{
83b699ed 4773 int retval, oldprio, oldpolicy = -1, on_rq, running;
1da177e4 4774 unsigned long flags;
83ab0aa0 4775 const struct sched_class *prev_class;
70b97a7f 4776 struct rq *rq;
ca94c442 4777 int reset_on_fork;
1da177e4 4778
66e5393a
SR
4779 /* may grab non-irq protected spin_locks */
4780 BUG_ON(in_interrupt());
1da177e4
LT
4781recheck:
4782 /* double check policy once rq lock held */
ca94c442
LP
4783 if (policy < 0) {
4784 reset_on_fork = p->sched_reset_on_fork;
1da177e4 4785 policy = oldpolicy = p->policy;
ca94c442
LP
4786 } else {
4787 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4788 policy &= ~SCHED_RESET_ON_FORK;
4789
4790 if (policy != SCHED_FIFO && policy != SCHED_RR &&
4791 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4792 policy != SCHED_IDLE)
4793 return -EINVAL;
4794 }
4795
1da177e4
LT
4796 /*
4797 * Valid priorities for SCHED_FIFO and SCHED_RR are
dd41f596
IM
4798 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4799 * SCHED_BATCH and SCHED_IDLE is 0.
1da177e4
LT
4800 */
4801 if (param->sched_priority < 0 ||
95cdf3b7 4802 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
d46523ea 4803 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
1da177e4 4804 return -EINVAL;
e05606d3 4805 if (rt_policy(policy) != (param->sched_priority != 0))
1da177e4
LT
4806 return -EINVAL;
4807
37e4ab3f
OC
4808 /*
4809 * Allow unprivileged RT tasks to decrease priority:
4810 */
961ccddd 4811 if (user && !capable(CAP_SYS_NICE)) {
e05606d3 4812 if (rt_policy(policy)) {
a44702e8
ON
4813 unsigned long rlim_rtprio =
4814 task_rlimit(p, RLIMIT_RTPRIO);
8dc3e909
ON
4815
4816 /* can't set/change the rt policy */
4817 if (policy != p->policy && !rlim_rtprio)
4818 return -EPERM;
4819
4820 /* can't increase priority */
4821 if (param->sched_priority > p->rt_priority &&
4822 param->sched_priority > rlim_rtprio)
4823 return -EPERM;
4824 }
dd41f596
IM
4825 /*
4826 * Like positive nice levels, dont allow tasks to
4827 * move out of SCHED_IDLE either:
4828 */
4829 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4830 return -EPERM;
5fe1d75f 4831
37e4ab3f 4832 /* can't change other user's priorities */
c69e8d9c 4833 if (!check_same_owner(p))
37e4ab3f 4834 return -EPERM;
ca94c442
LP
4835
4836 /* Normal users shall not reset the sched_reset_on_fork flag */
4837 if (p->sched_reset_on_fork && !reset_on_fork)
4838 return -EPERM;
37e4ab3f 4839 }
1da177e4 4840
725aad24 4841 if (user) {
b0ae1981 4842 retval = security_task_setscheduler(p);
725aad24
JF
4843 if (retval)
4844 return retval;
4845 }
4846
b29739f9
IM
4847 /*
4848 * make sure no PI-waiters arrive (or leave) while we are
4849 * changing the priority of the task:
4850 */
1d615482 4851 raw_spin_lock_irqsave(&p->pi_lock, flags);
1da177e4
LT
4852 /*
4853 * To be able to change p->policy safely, the apropriate
4854 * runqueue lock must be held.
4855 */
b29739f9 4856 rq = __task_rq_lock(p);
dc61b1d6 4857
34f971f6
PZ
4858 /*
4859 * Changing the policy of the stop threads its a very bad idea
4860 */
4861 if (p == rq->stop) {
4862 __task_rq_unlock(rq);
4863 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4864 return -EINVAL;
4865 }
4866
dc61b1d6
PZ
4867#ifdef CONFIG_RT_GROUP_SCHED
4868 if (user) {
4869 /*
4870 * Do not allow realtime tasks into groups that have no runtime
4871 * assigned.
4872 */
4873 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4874 task_group(p)->rt_bandwidth.rt_runtime == 0) {
4875 __task_rq_unlock(rq);
4876 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4877 return -EPERM;
4878 }
4879 }
4880#endif
4881
1da177e4
LT
4882 /* recheck policy now with rq lock held */
4883 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4884 policy = oldpolicy = -1;
b29739f9 4885 __task_rq_unlock(rq);
1d615482 4886 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
1da177e4
LT
4887 goto recheck;
4888 }
dd41f596 4889 on_rq = p->se.on_rq;
051a1d1a 4890 running = task_current(rq, p);
0e1f3483 4891 if (on_rq)
2e1cb74a 4892 deactivate_task(rq, p, 0);
0e1f3483
HS
4893 if (running)
4894 p->sched_class->put_prev_task(rq, p);
f6b53205 4895
ca94c442
LP
4896 p->sched_reset_on_fork = reset_on_fork;
4897
1da177e4 4898 oldprio = p->prio;
83ab0aa0 4899 prev_class = p->sched_class;
dd41f596 4900 __setscheduler(rq, p, policy, param->sched_priority);
f6b53205 4901
0e1f3483
HS
4902 if (running)
4903 p->sched_class->set_curr_task(rq);
dd41f596
IM
4904 if (on_rq) {
4905 activate_task(rq, p, 0);
cb469845
SR
4906
4907 check_class_changed(rq, p, prev_class, oldprio, running);
1da177e4 4908 }
b29739f9 4909 __task_rq_unlock(rq);
1d615482 4910 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
b29739f9 4911
95e02ca9
TG
4912 rt_mutex_adjust_pi(p);
4913
1da177e4
LT
4914 return 0;
4915}
961ccddd
RR
4916
4917/**
4918 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4919 * @p: the task in question.
4920 * @policy: new policy.
4921 * @param: structure containing the new RT priority.
4922 *
4923 * NOTE that the task may be already dead.
4924 */
4925int sched_setscheduler(struct task_struct *p, int policy,
fe7de49f 4926 const struct sched_param *param)
961ccddd
RR
4927{
4928 return __sched_setscheduler(p, policy, param, true);
4929}
1da177e4
LT
4930EXPORT_SYMBOL_GPL(sched_setscheduler);
4931
961ccddd
RR
4932/**
4933 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4934 * @p: the task in question.
4935 * @policy: new policy.
4936 * @param: structure containing the new RT priority.
4937 *
4938 * Just like sched_setscheduler, only don't bother checking if the
4939 * current context has permission. For example, this is needed in
4940 * stop_machine(): we create temporary high priority worker threads,
4941 * but our caller might not have that capability.
4942 */
4943int sched_setscheduler_nocheck(struct task_struct *p, int policy,
fe7de49f 4944 const struct sched_param *param)
961ccddd
RR
4945{
4946 return __sched_setscheduler(p, policy, param, false);
4947}
4948
95cdf3b7
IM
4949static int
4950do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
1da177e4 4951{
1da177e4
LT
4952 struct sched_param lparam;
4953 struct task_struct *p;
36c8b586 4954 int retval;
1da177e4
LT
4955
4956 if (!param || pid < 0)
4957 return -EINVAL;
4958 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4959 return -EFAULT;
5fe1d75f
ON
4960
4961 rcu_read_lock();
4962 retval = -ESRCH;
1da177e4 4963 p = find_process_by_pid(pid);
5fe1d75f
ON
4964 if (p != NULL)
4965 retval = sched_setscheduler(p, policy, &lparam);
4966 rcu_read_unlock();
36c8b586 4967
1da177e4
LT
4968 return retval;
4969}
4970
4971/**
4972 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4973 * @pid: the pid in question.
4974 * @policy: new policy.
4975 * @param: structure containing the new RT priority.
4976 */
5add95d4
HC
4977SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4978 struct sched_param __user *, param)
1da177e4 4979{
c21761f1
JB
4980 /* negative values for policy are not valid */
4981 if (policy < 0)
4982 return -EINVAL;
4983
1da177e4
LT
4984 return do_sched_setscheduler(pid, policy, param);
4985}
4986
4987/**
4988 * sys_sched_setparam - set/change the RT priority of a thread
4989 * @pid: the pid in question.
4990 * @param: structure containing the new RT priority.
4991 */
5add95d4 4992SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
1da177e4
LT
4993{
4994 return do_sched_setscheduler(pid, -1, param);
4995}
4996
4997/**
4998 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4999 * @pid: the pid in question.
5000 */
5add95d4 5001SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
1da177e4 5002{
36c8b586 5003 struct task_struct *p;
3a5c359a 5004 int retval;
1da177e4
LT
5005
5006 if (pid < 0)
3a5c359a 5007 return -EINVAL;
1da177e4
LT
5008
5009 retval = -ESRCH;
5fe85be0 5010 rcu_read_lock();
1da177e4
LT
5011 p = find_process_by_pid(pid);
5012 if (p) {
5013 retval = security_task_getscheduler(p);
5014 if (!retval)
ca94c442
LP
5015 retval = p->policy
5016 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
1da177e4 5017 }
5fe85be0 5018 rcu_read_unlock();
1da177e4
LT
5019 return retval;
5020}
5021
5022/**
ca94c442 5023 * sys_sched_getparam - get the RT priority of a thread
1da177e4
LT
5024 * @pid: the pid in question.
5025 * @param: structure containing the RT priority.
5026 */
5add95d4 5027SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
1da177e4
LT
5028{
5029 struct sched_param lp;
36c8b586 5030 struct task_struct *p;
3a5c359a 5031 int retval;
1da177e4
LT
5032
5033 if (!param || pid < 0)
3a5c359a 5034 return -EINVAL;
1da177e4 5035
5fe85be0 5036 rcu_read_lock();
1da177e4
LT
5037 p = find_process_by_pid(pid);
5038 retval = -ESRCH;
5039 if (!p)
5040 goto out_unlock;
5041
5042 retval = security_task_getscheduler(p);
5043 if (retval)
5044 goto out_unlock;
5045
5046 lp.sched_priority = p->rt_priority;
5fe85be0 5047 rcu_read_unlock();
1da177e4
LT
5048
5049 /*
5050 * This one might sleep, we cannot do it with a spinlock held ...
5051 */
5052 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
5053
1da177e4
LT
5054 return retval;
5055
5056out_unlock:
5fe85be0 5057 rcu_read_unlock();
1da177e4
LT
5058 return retval;
5059}
5060
96f874e2 5061long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
1da177e4 5062{
5a16f3d3 5063 cpumask_var_t cpus_allowed, new_mask;
36c8b586
IM
5064 struct task_struct *p;
5065 int retval;
1da177e4 5066
95402b38 5067 get_online_cpus();
23f5d142 5068 rcu_read_lock();
1da177e4
LT
5069
5070 p = find_process_by_pid(pid);
5071 if (!p) {
23f5d142 5072 rcu_read_unlock();
95402b38 5073 put_online_cpus();
1da177e4
LT
5074 return -ESRCH;
5075 }
5076
23f5d142 5077 /* Prevent p going away */
1da177e4 5078 get_task_struct(p);
23f5d142 5079 rcu_read_unlock();
1da177e4 5080
5a16f3d3
RR
5081 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5082 retval = -ENOMEM;
5083 goto out_put_task;
5084 }
5085 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5086 retval = -ENOMEM;
5087 goto out_free_cpus_allowed;
5088 }
1da177e4 5089 retval = -EPERM;
c69e8d9c 5090 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
1da177e4
LT
5091 goto out_unlock;
5092
b0ae1981 5093 retval = security_task_setscheduler(p);
e7834f8f
DQ
5094 if (retval)
5095 goto out_unlock;
5096
5a16f3d3
RR
5097 cpuset_cpus_allowed(p, cpus_allowed);
5098 cpumask_and(new_mask, in_mask, cpus_allowed);
49246274 5099again:
5a16f3d3 5100 retval = set_cpus_allowed_ptr(p, new_mask);
1da177e4 5101
8707d8b8 5102 if (!retval) {
5a16f3d3
RR
5103 cpuset_cpus_allowed(p, cpus_allowed);
5104 if (!cpumask_subset(new_mask, cpus_allowed)) {
8707d8b8
PM
5105 /*
5106 * We must have raced with a concurrent cpuset
5107 * update. Just reset the cpus_allowed to the
5108 * cpuset's cpus_allowed
5109 */
5a16f3d3 5110 cpumask_copy(new_mask, cpus_allowed);
8707d8b8
PM
5111 goto again;
5112 }
5113 }
1da177e4 5114out_unlock:
5a16f3d3
RR
5115 free_cpumask_var(new_mask);
5116out_free_cpus_allowed:
5117 free_cpumask_var(cpus_allowed);
5118out_put_task:
1da177e4 5119 put_task_struct(p);
95402b38 5120 put_online_cpus();
1da177e4
LT
5121 return retval;
5122}
5123
5124static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
96f874e2 5125 struct cpumask *new_mask)
1da177e4 5126{
96f874e2
RR
5127 if (len < cpumask_size())
5128 cpumask_clear(new_mask);
5129 else if (len > cpumask_size())
5130 len = cpumask_size();
5131
1da177e4
LT
5132 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5133}
5134
5135/**
5136 * sys_sched_setaffinity - set the cpu affinity of a process
5137 * @pid: pid of the process
5138 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5139 * @user_mask_ptr: user-space pointer to the new cpu mask
5140 */
5add95d4
HC
5141SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
5142 unsigned long __user *, user_mask_ptr)
1da177e4 5143{
5a16f3d3 5144 cpumask_var_t new_mask;
1da177e4
LT
5145 int retval;
5146
5a16f3d3
RR
5147 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5148 return -ENOMEM;
1da177e4 5149
5a16f3d3
RR
5150 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5151 if (retval == 0)
5152 retval = sched_setaffinity(pid, new_mask);
5153 free_cpumask_var(new_mask);
5154 return retval;
1da177e4
LT
5155}
5156
96f874e2 5157long sched_getaffinity(pid_t pid, struct cpumask *mask)
1da177e4 5158{
36c8b586 5159 struct task_struct *p;
31605683
TG
5160 unsigned long flags;
5161 struct rq *rq;
1da177e4 5162 int retval;
1da177e4 5163
95402b38 5164 get_online_cpus();
23f5d142 5165 rcu_read_lock();
1da177e4
LT
5166
5167 retval = -ESRCH;
5168 p = find_process_by_pid(pid);
5169 if (!p)
5170 goto out_unlock;
5171
e7834f8f
DQ
5172 retval = security_task_getscheduler(p);
5173 if (retval)
5174 goto out_unlock;
5175
31605683 5176 rq = task_rq_lock(p, &flags);
96f874e2 5177 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
31605683 5178 task_rq_unlock(rq, &flags);
1da177e4
LT
5179
5180out_unlock:
23f5d142 5181 rcu_read_unlock();
95402b38 5182 put_online_cpus();
1da177e4 5183
9531b62f 5184 return retval;
1da177e4
LT
5185}
5186
5187/**
5188 * sys_sched_getaffinity - get the cpu affinity of a process
5189 * @pid: pid of the process
5190 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5191 * @user_mask_ptr: user-space pointer to hold the current cpu mask
5192 */
5add95d4
HC
5193SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
5194 unsigned long __user *, user_mask_ptr)
1da177e4
LT
5195{
5196 int ret;
f17c8607 5197 cpumask_var_t mask;
1da177e4 5198
84fba5ec 5199 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
cd3d8031
KM
5200 return -EINVAL;
5201 if (len & (sizeof(unsigned long)-1))
1da177e4
LT
5202 return -EINVAL;
5203
f17c8607
RR
5204 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5205 return -ENOMEM;
1da177e4 5206
f17c8607
RR
5207 ret = sched_getaffinity(pid, mask);
5208 if (ret == 0) {
8bc037fb 5209 size_t retlen = min_t(size_t, len, cpumask_size());
cd3d8031
KM
5210
5211 if (copy_to_user(user_mask_ptr, mask, retlen))
f17c8607
RR
5212 ret = -EFAULT;
5213 else
cd3d8031 5214 ret = retlen;
f17c8607
RR
5215 }
5216 free_cpumask_var(mask);
1da177e4 5217
f17c8607 5218 return ret;
1da177e4
LT
5219}
5220
5221/**
5222 * sys_sched_yield - yield the current processor to other threads.
5223 *
dd41f596
IM
5224 * This function yields the current CPU to other tasks. If there are no
5225 * other threads running on this CPU then this function will return.
1da177e4 5226 */
5add95d4 5227SYSCALL_DEFINE0(sched_yield)
1da177e4 5228{
70b97a7f 5229 struct rq *rq = this_rq_lock();
1da177e4 5230
2d72376b 5231 schedstat_inc(rq, yld_count);
4530d7ab 5232 current->sched_class->yield_task(rq);
1da177e4
LT
5233
5234 /*
5235 * Since we are going to call schedule() anyway, there's
5236 * no need to preempt or enable interrupts:
5237 */
5238 __release(rq->lock);
8a25d5de 5239 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
9828ea9d 5240 do_raw_spin_unlock(&rq->lock);
1da177e4
LT
5241 preempt_enable_no_resched();
5242
5243 schedule();
5244
5245 return 0;
5246}
5247
d86ee480
PZ
5248static inline int should_resched(void)
5249{
5250 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5251}
5252
e7b38404 5253static void __cond_resched(void)
1da177e4 5254{
e7aaaa69
FW
5255 add_preempt_count(PREEMPT_ACTIVE);
5256 schedule();
5257 sub_preempt_count(PREEMPT_ACTIVE);
1da177e4
LT
5258}
5259
02b67cc3 5260int __sched _cond_resched(void)
1da177e4 5261{
d86ee480 5262 if (should_resched()) {
1da177e4
LT
5263 __cond_resched();
5264 return 1;
5265 }
5266 return 0;
5267}
02b67cc3 5268EXPORT_SYMBOL(_cond_resched);
1da177e4
LT
5269
5270/*
613afbf8 5271 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
1da177e4
LT
5272 * call schedule, and on return reacquire the lock.
5273 *
41a2d6cf 5274 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
1da177e4
LT
5275 * operations here to prevent schedule() from being called twice (once via
5276 * spin_unlock(), once by hand).
5277 */
613afbf8 5278int __cond_resched_lock(spinlock_t *lock)
1da177e4 5279{
d86ee480 5280 int resched = should_resched();
6df3cecb
JK
5281 int ret = 0;
5282
f607c668
PZ
5283 lockdep_assert_held(lock);
5284
95c354fe 5285 if (spin_needbreak(lock) || resched) {
1da177e4 5286 spin_unlock(lock);
d86ee480 5287 if (resched)
95c354fe
NP
5288 __cond_resched();
5289 else
5290 cpu_relax();
6df3cecb 5291 ret = 1;
1da177e4 5292 spin_lock(lock);
1da177e4 5293 }
6df3cecb 5294 return ret;
1da177e4 5295}
613afbf8 5296EXPORT_SYMBOL(__cond_resched_lock);
1da177e4 5297
613afbf8 5298int __sched __cond_resched_softirq(void)
1da177e4
LT
5299{
5300 BUG_ON(!in_softirq());
5301
d86ee480 5302 if (should_resched()) {
98d82567 5303 local_bh_enable();
1da177e4
LT
5304 __cond_resched();
5305 local_bh_disable();
5306 return 1;
5307 }
5308 return 0;
5309}
613afbf8 5310EXPORT_SYMBOL(__cond_resched_softirq);
1da177e4 5311
1da177e4
LT
5312/**
5313 * yield - yield the current processor to other threads.
5314 *
72fd4a35 5315 * This is a shortcut for kernel-space yielding - it marks the
1da177e4
LT
5316 * thread runnable and calls sys_sched_yield().
5317 */
5318void __sched yield(void)
5319{
5320 set_current_state(TASK_RUNNING);
5321 sys_sched_yield();
5322}
1da177e4
LT
5323EXPORT_SYMBOL(yield);
5324
5325/*
41a2d6cf 5326 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
1da177e4 5327 * that process accounting knows that this is a task in IO wait state.
1da177e4
LT
5328 */
5329void __sched io_schedule(void)
5330{
54d35f29 5331 struct rq *rq = raw_rq();
1da177e4 5332
0ff92245 5333 delayacct_blkio_start();
1da177e4 5334 atomic_inc(&rq->nr_iowait);
8f0dfc34 5335 current->in_iowait = 1;
1da177e4 5336 schedule();
8f0dfc34 5337 current->in_iowait = 0;
1da177e4 5338 atomic_dec(&rq->nr_iowait);
0ff92245 5339 delayacct_blkio_end();
1da177e4 5340}
1da177e4
LT
5341EXPORT_SYMBOL(io_schedule);
5342
5343long __sched io_schedule_timeout(long timeout)
5344{
54d35f29 5345 struct rq *rq = raw_rq();
1da177e4
LT
5346 long ret;
5347
0ff92245 5348 delayacct_blkio_start();
1da177e4 5349 atomic_inc(&rq->nr_iowait);
8f0dfc34 5350 current->in_iowait = 1;
1da177e4 5351 ret = schedule_timeout(timeout);
8f0dfc34 5352 current->in_iowait = 0;
1da177e4 5353 atomic_dec(&rq->nr_iowait);
0ff92245 5354 delayacct_blkio_end();
1da177e4
LT
5355 return ret;
5356}
5357
5358/**
5359 * sys_sched_get_priority_max - return maximum RT priority.
5360 * @policy: scheduling class.
5361 *
5362 * this syscall returns the maximum rt_priority that can be used
5363 * by a given scheduling class.
5364 */
5add95d4 5365SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
1da177e4
LT
5366{
5367 int ret = -EINVAL;
5368
5369 switch (policy) {
5370 case SCHED_FIFO:
5371 case SCHED_RR:
5372 ret = MAX_USER_RT_PRIO-1;
5373 break;
5374 case SCHED_NORMAL:
b0a9499c 5375 case SCHED_BATCH:
dd41f596 5376 case SCHED_IDLE:
1da177e4
LT
5377 ret = 0;
5378 break;
5379 }
5380 return ret;
5381}
5382
5383/**
5384 * sys_sched_get_priority_min - return minimum RT priority.
5385 * @policy: scheduling class.
5386 *
5387 * this syscall returns the minimum rt_priority that can be used
5388 * by a given scheduling class.
5389 */
5add95d4 5390SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
1da177e4
LT
5391{
5392 int ret = -EINVAL;
5393
5394 switch (policy) {
5395 case SCHED_FIFO:
5396 case SCHED_RR:
5397 ret = 1;
5398 break;
5399 case SCHED_NORMAL:
b0a9499c 5400 case SCHED_BATCH:
dd41f596 5401 case SCHED_IDLE:
1da177e4
LT
5402 ret = 0;
5403 }
5404 return ret;
5405}
5406
5407/**
5408 * sys_sched_rr_get_interval - return the default timeslice of a process.
5409 * @pid: pid of the process.
5410 * @interval: userspace pointer to the timeslice value.
5411 *
5412 * this syscall writes the default timeslice value of a given process
5413 * into the user-space timespec buffer. A value of '0' means infinity.
5414 */
17da2bd9 5415SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
754fe8d2 5416 struct timespec __user *, interval)
1da177e4 5417{
36c8b586 5418 struct task_struct *p;
a4ec24b4 5419 unsigned int time_slice;
dba091b9
TG
5420 unsigned long flags;
5421 struct rq *rq;
3a5c359a 5422 int retval;
1da177e4 5423 struct timespec t;
1da177e4
LT
5424
5425 if (pid < 0)
3a5c359a 5426 return -EINVAL;
1da177e4
LT
5427
5428 retval = -ESRCH;
1a551ae7 5429 rcu_read_lock();
1da177e4
LT
5430 p = find_process_by_pid(pid);
5431 if (!p)
5432 goto out_unlock;
5433
5434 retval = security_task_getscheduler(p);
5435 if (retval)
5436 goto out_unlock;
5437
dba091b9
TG
5438 rq = task_rq_lock(p, &flags);
5439 time_slice = p->sched_class->get_rr_interval(rq, p);
5440 task_rq_unlock(rq, &flags);
a4ec24b4 5441
1a551ae7 5442 rcu_read_unlock();
a4ec24b4 5443 jiffies_to_timespec(time_slice, &t);
1da177e4 5444 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
1da177e4 5445 return retval;
3a5c359a 5446
1da177e4 5447out_unlock:
1a551ae7 5448 rcu_read_unlock();
1da177e4
LT
5449 return retval;
5450}
5451
7c731e0a 5452static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
36c8b586 5453
82a1fcb9 5454void sched_show_task(struct task_struct *p)
1da177e4 5455{
1da177e4 5456 unsigned long free = 0;
36c8b586 5457 unsigned state;
1da177e4 5458
1da177e4 5459 state = p->state ? __ffs(p->state) + 1 : 0;
28d0686c 5460 printk(KERN_INFO "%-15.15s %c", p->comm,
2ed6e34f 5461 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
4bd77321 5462#if BITS_PER_LONG == 32
1da177e4 5463 if (state == TASK_RUNNING)
3df0fc5b 5464 printk(KERN_CONT " running ");
1da177e4 5465 else
3df0fc5b 5466 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
1da177e4
LT
5467#else
5468 if (state == TASK_RUNNING)
3df0fc5b 5469 printk(KERN_CONT " running task ");
1da177e4 5470 else
3df0fc5b 5471 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
1da177e4
LT
5472#endif
5473#ifdef CONFIG_DEBUG_STACK_USAGE
7c9f8861 5474 free = stack_not_used(p);
1da177e4 5475#endif
3df0fc5b 5476 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
aa47b7e0
DR
5477 task_pid_nr(p), task_pid_nr(p->real_parent),
5478 (unsigned long)task_thread_info(p)->flags);
1da177e4 5479
5fb5e6de 5480 show_stack(p, NULL);
1da177e4
LT
5481}
5482
e59e2ae2 5483void show_state_filter(unsigned long state_filter)
1da177e4 5484{
36c8b586 5485 struct task_struct *g, *p;
1da177e4 5486
4bd77321 5487#if BITS_PER_LONG == 32
3df0fc5b
PZ
5488 printk(KERN_INFO
5489 " task PC stack pid father\n");
1da177e4 5490#else
3df0fc5b
PZ
5491 printk(KERN_INFO
5492 " task PC stack pid father\n");
1da177e4
LT
5493#endif
5494 read_lock(&tasklist_lock);
5495 do_each_thread(g, p) {
5496 /*
5497 * reset the NMI-timeout, listing all files on a slow
5498 * console might take alot of time:
5499 */
5500 touch_nmi_watchdog();
39bc89fd 5501 if (!state_filter || (p->state & state_filter))
82a1fcb9 5502 sched_show_task(p);
1da177e4
LT
5503 } while_each_thread(g, p);
5504
04c9167f
JF
5505 touch_all_softlockup_watchdogs();
5506
dd41f596
IM
5507#ifdef CONFIG_SCHED_DEBUG
5508 sysrq_sched_debug_show();
5509#endif
1da177e4 5510 read_unlock(&tasklist_lock);
e59e2ae2
IM
5511 /*
5512 * Only show locks if all tasks are dumped:
5513 */
93335a21 5514 if (!state_filter)
e59e2ae2 5515 debug_show_all_locks();
1da177e4
LT
5516}
5517
1df21055
IM
5518void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5519{
dd41f596 5520 idle->sched_class = &idle_sched_class;
1df21055
IM
5521}
5522
f340c0d1
IM
5523/**
5524 * init_idle - set up an idle thread for a given CPU
5525 * @idle: task in question
5526 * @cpu: cpu the idle task belongs to
5527 *
5528 * NOTE: this function does not set the idle thread's NEED_RESCHED
5529 * flag, to make booting more robust.
5530 */
5c1e1767 5531void __cpuinit init_idle(struct task_struct *idle, int cpu)
1da177e4 5532{
70b97a7f 5533 struct rq *rq = cpu_rq(cpu);
1da177e4
LT
5534 unsigned long flags;
5535
05fa785c 5536 raw_spin_lock_irqsave(&rq->lock, flags);
5cbd54ef 5537
dd41f596 5538 __sched_fork(idle);
06b83b5f 5539 idle->state = TASK_RUNNING;
dd41f596
IM
5540 idle->se.exec_start = sched_clock();
5541
96f874e2 5542 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
6506cf6c
PZ
5543 /*
5544 * We're having a chicken and egg problem, even though we are
5545 * holding rq->lock, the cpu isn't yet set to this cpu so the
5546 * lockdep check in task_group() will fail.
5547 *
5548 * Similar case to sched_fork(). / Alternatively we could
5549 * use task_rq_lock() here and obtain the other rq->lock.
5550 *
5551 * Silence PROVE_RCU
5552 */
5553 rcu_read_lock();
dd41f596 5554 __set_task_cpu(idle, cpu);
6506cf6c 5555 rcu_read_unlock();
1da177e4 5556
1da177e4 5557 rq->curr = rq->idle = idle;
4866cde0
NP
5558#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5559 idle->oncpu = 1;
5560#endif
05fa785c 5561 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4
LT
5562
5563 /* Set the preempt count _outside_ the spinlocks! */
8e3e076c
LT
5564#if defined(CONFIG_PREEMPT)
5565 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5566#else
a1261f54 5567 task_thread_info(idle)->preempt_count = 0;
8e3e076c 5568#endif
dd41f596
IM
5569 /*
5570 * The idle tasks have their own, simple scheduling class:
5571 */
5572 idle->sched_class = &idle_sched_class;
fb52607a 5573 ftrace_graph_init_task(idle);
1da177e4
LT
5574}
5575
5576/*
5577 * In a system that switches off the HZ timer nohz_cpu_mask
5578 * indicates which cpus entered this state. This is used
5579 * in the rcu update to wait only for active cpus. For system
5580 * which do not switch off the HZ timer nohz_cpu_mask should
6a7b3dc3 5581 * always be CPU_BITS_NONE.
1da177e4 5582 */
6a7b3dc3 5583cpumask_var_t nohz_cpu_mask;
1da177e4 5584
19978ca6
IM
5585/*
5586 * Increase the granularity value when there are more CPUs,
5587 * because with more CPUs the 'effective latency' as visible
5588 * to users decreases. But the relationship is not linear,
5589 * so pick a second-best guess by going with the log2 of the
5590 * number of CPUs.
5591 *
5592 * This idea comes from the SD scheduler of Con Kolivas:
5593 */
acb4a848 5594static int get_update_sysctl_factor(void)
19978ca6 5595{
4ca3ef71 5596 unsigned int cpus = min_t(int, num_online_cpus(), 8);
1983a922
CE
5597 unsigned int factor;
5598
5599 switch (sysctl_sched_tunable_scaling) {
5600 case SCHED_TUNABLESCALING_NONE:
5601 factor = 1;
5602 break;
5603 case SCHED_TUNABLESCALING_LINEAR:
5604 factor = cpus;
5605 break;
5606 case SCHED_TUNABLESCALING_LOG:
5607 default:
5608 factor = 1 + ilog2(cpus);
5609 break;
5610 }
19978ca6 5611
acb4a848
CE
5612 return factor;
5613}
19978ca6 5614
acb4a848
CE
5615static void update_sysctl(void)
5616{
5617 unsigned int factor = get_update_sysctl_factor();
19978ca6 5618
0bcdcf28
CE
5619#define SET_SYSCTL(name) \
5620 (sysctl_##name = (factor) * normalized_sysctl_##name)
5621 SET_SYSCTL(sched_min_granularity);
5622 SET_SYSCTL(sched_latency);
5623 SET_SYSCTL(sched_wakeup_granularity);
0bcdcf28
CE
5624#undef SET_SYSCTL
5625}
55cd5340 5626
0bcdcf28
CE
5627static inline void sched_init_granularity(void)
5628{
5629 update_sysctl();
19978ca6
IM
5630}
5631
1da177e4
LT
5632#ifdef CONFIG_SMP
5633/*
5634 * This is how migration works:
5635 *
969c7921
TH
5636 * 1) we invoke migration_cpu_stop() on the target CPU using
5637 * stop_one_cpu().
5638 * 2) stopper starts to run (implicitly forcing the migrated thread
5639 * off the CPU)
5640 * 3) it checks whether the migrated task is still in the wrong runqueue.
5641 * 4) if it's in the wrong runqueue then the migration thread removes
1da177e4 5642 * it and puts it into the right queue.
969c7921
TH
5643 * 5) stopper completes and stop_one_cpu() returns and the migration
5644 * is done.
1da177e4
LT
5645 */
5646
5647/*
5648 * Change a given task's CPU affinity. Migrate the thread to a
5649 * proper CPU and schedule it away if the CPU it's executing on
5650 * is removed from the allowed bitmask.
5651 *
5652 * NOTE: the caller must have a valid reference to the task, the
41a2d6cf 5653 * task must not exit() & deallocate itself prematurely. The
1da177e4
LT
5654 * call is not atomic; no spinlocks may be held.
5655 */
96f874e2 5656int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1da177e4
LT
5657{
5658 unsigned long flags;
70b97a7f 5659 struct rq *rq;
969c7921 5660 unsigned int dest_cpu;
48f24c4d 5661 int ret = 0;
1da177e4 5662
65cc8e48
PZ
5663 /*
5664 * Serialize against TASK_WAKING so that ttwu() and wunt() can
5665 * drop the rq->lock and still rely on ->cpus_allowed.
5666 */
5667again:
5668 while (task_is_waking(p))
5669 cpu_relax();
1da177e4 5670 rq = task_rq_lock(p, &flags);
65cc8e48
PZ
5671 if (task_is_waking(p)) {
5672 task_rq_unlock(rq, &flags);
5673 goto again;
5674 }
e2912009 5675
6ad4c188 5676 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
1da177e4
LT
5677 ret = -EINVAL;
5678 goto out;
5679 }
5680
9985b0ba 5681 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
96f874e2 5682 !cpumask_equal(&p->cpus_allowed, new_mask))) {
9985b0ba
DR
5683 ret = -EINVAL;
5684 goto out;
5685 }
5686
73fe6aae 5687 if (p->sched_class->set_cpus_allowed)
cd8ba7cd 5688 p->sched_class->set_cpus_allowed(p, new_mask);
73fe6aae 5689 else {
96f874e2
RR
5690 cpumask_copy(&p->cpus_allowed, new_mask);
5691 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
73fe6aae
GH
5692 }
5693
1da177e4 5694 /* Can the task run on the task's current CPU? If so, we're done */
96f874e2 5695 if (cpumask_test_cpu(task_cpu(p), new_mask))
1da177e4
LT
5696 goto out;
5697
969c7921 5698 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
b7a2b39d 5699 if (migrate_task(p, rq)) {
969c7921 5700 struct migration_arg arg = { p, dest_cpu };
1da177e4
LT
5701 /* Need help from migration thread: drop lock and wait. */
5702 task_rq_unlock(rq, &flags);
969c7921 5703 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1da177e4
LT
5704 tlb_migrate_finish(p->mm);
5705 return 0;
5706 }
5707out:
5708 task_rq_unlock(rq, &flags);
48f24c4d 5709
1da177e4
LT
5710 return ret;
5711}
cd8ba7cd 5712EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
1da177e4
LT
5713
5714/*
41a2d6cf 5715 * Move (not current) task off this cpu, onto dest cpu. We're doing
1da177e4
LT
5716 * this because either it can't run here any more (set_cpus_allowed()
5717 * away from this CPU, or CPU going down), or because we're
5718 * attempting to rebalance this task on exec (sched_exec).
5719 *
5720 * So we race with normal scheduler movements, but that's OK, as long
5721 * as the task is no longer on this CPU.
efc30814
KK
5722 *
5723 * Returns non-zero if task was successfully migrated.
1da177e4 5724 */
efc30814 5725static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
1da177e4 5726{
70b97a7f 5727 struct rq *rq_dest, *rq_src;
e2912009 5728 int ret = 0;
1da177e4 5729
e761b772 5730 if (unlikely(!cpu_active(dest_cpu)))
efc30814 5731 return ret;
1da177e4
LT
5732
5733 rq_src = cpu_rq(src_cpu);
5734 rq_dest = cpu_rq(dest_cpu);
5735
5736 double_rq_lock(rq_src, rq_dest);
5737 /* Already moved. */
5738 if (task_cpu(p) != src_cpu)
b1e38734 5739 goto done;
1da177e4 5740 /* Affinity changed (again). */
96f874e2 5741 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
b1e38734 5742 goto fail;
1da177e4 5743
e2912009
PZ
5744 /*
5745 * If we're not on a rq, the next wake-up will ensure we're
5746 * placed properly.
5747 */
5748 if (p->se.on_rq) {
2e1cb74a 5749 deactivate_task(rq_src, p, 0);
e2912009 5750 set_task_cpu(p, dest_cpu);
dd41f596 5751 activate_task(rq_dest, p, 0);
15afe09b 5752 check_preempt_curr(rq_dest, p, 0);
1da177e4 5753 }
b1e38734 5754done:
efc30814 5755 ret = 1;
b1e38734 5756fail:
1da177e4 5757 double_rq_unlock(rq_src, rq_dest);
efc30814 5758 return ret;
1da177e4
LT
5759}
5760
5761/*
969c7921
TH
5762 * migration_cpu_stop - this will be executed by a highprio stopper thread
5763 * and performs thread migration by bumping thread off CPU then
5764 * 'pushing' onto another runqueue.
1da177e4 5765 */
969c7921 5766static int migration_cpu_stop(void *data)
1da177e4 5767{
969c7921 5768 struct migration_arg *arg = data;
f7b4cddc 5769
969c7921
TH
5770 /*
5771 * The original target cpu might have gone down and we might
5772 * be on another cpu but it doesn't matter.
5773 */
f7b4cddc 5774 local_irq_disable();
969c7921 5775 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
f7b4cddc 5776 local_irq_enable();
1da177e4 5777 return 0;
f7b4cddc
ON
5778}
5779
1da177e4 5780#ifdef CONFIG_HOTPLUG_CPU
48c5ccae 5781
054b9108 5782/*
48c5ccae
PZ
5783 * Ensures that the idle task is using init_mm right before its cpu goes
5784 * offline.
054b9108 5785 */
48c5ccae 5786void idle_task_exit(void)
1da177e4 5787{
48c5ccae 5788 struct mm_struct *mm = current->active_mm;
e76bd8d9 5789
48c5ccae 5790 BUG_ON(cpu_online(smp_processor_id()));
e76bd8d9 5791
48c5ccae
PZ
5792 if (mm != &init_mm)
5793 switch_mm(mm, &init_mm, current);
5794 mmdrop(mm);
1da177e4
LT
5795}
5796
5797/*
5798 * While a dead CPU has no uninterruptible tasks queued at this point,
5799 * it might still have a nonzero ->nr_uninterruptible counter, because
5800 * for performance reasons the counter is not stricly tracking tasks to
5801 * their home CPUs. So we just add the counter to another CPU's counter,
5802 * to keep the global sum constant after CPU-down:
5803 */
70b97a7f 5804static void migrate_nr_uninterruptible(struct rq *rq_src)
1da177e4 5805{
6ad4c188 5806 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
1da177e4 5807
1da177e4
LT
5808 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5809 rq_src->nr_uninterruptible = 0;
1da177e4
LT
5810}
5811
dd41f596 5812/*
48c5ccae 5813 * remove the tasks which were accounted by rq from calc_load_tasks.
1da177e4 5814 */
48c5ccae 5815static void calc_global_load_remove(struct rq *rq)
1da177e4 5816{
48c5ccae
PZ
5817 atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
5818 rq->calc_load_active = 0;
1da177e4
LT
5819}
5820
48f24c4d 5821/*
48c5ccae
PZ
5822 * Migrate all tasks from the rq, sleeping tasks will be migrated by
5823 * try_to_wake_up()->select_task_rq().
5824 *
5825 * Called with rq->lock held even though we'er in stop_machine() and
5826 * there's no concurrency possible, we hold the required locks anyway
5827 * because of lock validation efforts.
1da177e4 5828 */
48c5ccae 5829static void migrate_tasks(unsigned int dead_cpu)
1da177e4 5830{
70b97a7f 5831 struct rq *rq = cpu_rq(dead_cpu);
48c5ccae
PZ
5832 struct task_struct *next, *stop = rq->stop;
5833 int dest_cpu;
1da177e4
LT
5834
5835 /*
48c5ccae
PZ
5836 * Fudge the rq selection such that the below task selection loop
5837 * doesn't get stuck on the currently eligible stop task.
5838 *
5839 * We're currently inside stop_machine() and the rq is either stuck
5840 * in the stop_machine_cpu_stop() loop, or we're executing this code,
5841 * either way we should never end up calling schedule() until we're
5842 * done here.
1da177e4 5843 */
48c5ccae 5844 rq->stop = NULL;
48f24c4d 5845
dd41f596 5846 for ( ; ; ) {
48c5ccae
PZ
5847 /*
5848 * There's this thread running, bail when that's the only
5849 * remaining thread.
5850 */
5851 if (rq->nr_running == 1)
dd41f596 5852 break;
48c5ccae 5853
b67802ea 5854 next = pick_next_task(rq);
48c5ccae 5855 BUG_ON(!next);
79c53799 5856 next->sched_class->put_prev_task(rq, next);
e692ab53 5857
48c5ccae
PZ
5858 /* Find suitable destination for @next, with force if needed. */
5859 dest_cpu = select_fallback_rq(dead_cpu, next);
5860 raw_spin_unlock(&rq->lock);
5861
5862 __migrate_task(next, dead_cpu, dest_cpu);
5863
5864 raw_spin_lock(&rq->lock);
1da177e4 5865 }
dce48a84 5866
48c5ccae 5867 rq->stop = stop;
dce48a84 5868}
48c5ccae 5869
1da177e4
LT
5870#endif /* CONFIG_HOTPLUG_CPU */
5871
e692ab53
NP
5872#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5873
5874static struct ctl_table sd_ctl_dir[] = {
e0361851
AD
5875 {
5876 .procname = "sched_domain",
c57baf1e 5877 .mode = 0555,
e0361851 5878 },
56992309 5879 {}
e692ab53
NP
5880};
5881
5882static struct ctl_table sd_ctl_root[] = {
e0361851
AD
5883 {
5884 .procname = "kernel",
c57baf1e 5885 .mode = 0555,
e0361851
AD
5886 .child = sd_ctl_dir,
5887 },
56992309 5888 {}
e692ab53
NP
5889};
5890
5891static struct ctl_table *sd_alloc_ctl_entry(int n)
5892{
5893 struct ctl_table *entry =
5cf9f062 5894 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
e692ab53 5895
e692ab53
NP
5896 return entry;
5897}
5898
6382bc90
MM
5899static void sd_free_ctl_entry(struct ctl_table **tablep)
5900{
cd790076 5901 struct ctl_table *entry;
6382bc90 5902
cd790076
MM
5903 /*
5904 * In the intermediate directories, both the child directory and
5905 * procname are dynamically allocated and could fail but the mode
41a2d6cf 5906 * will always be set. In the lowest directory the names are
cd790076
MM
5907 * static strings and all have proc handlers.
5908 */
5909 for (entry = *tablep; entry->mode; entry++) {
6382bc90
MM
5910 if (entry->child)
5911 sd_free_ctl_entry(&entry->child);
cd790076
MM
5912 if (entry->proc_handler == NULL)
5913 kfree(entry->procname);
5914 }
6382bc90
MM
5915
5916 kfree(*tablep);
5917 *tablep = NULL;
5918}
5919
e692ab53 5920static void
e0361851 5921set_table_entry(struct ctl_table *entry,
e692ab53
NP
5922 const char *procname, void *data, int maxlen,
5923 mode_t mode, proc_handler *proc_handler)
5924{
e692ab53
NP
5925 entry->procname = procname;
5926 entry->data = data;
5927 entry->maxlen = maxlen;
5928 entry->mode = mode;
5929 entry->proc_handler = proc_handler;
5930}
5931
5932static struct ctl_table *
5933sd_alloc_ctl_domain_table(struct sched_domain *sd)
5934{
a5d8c348 5935 struct ctl_table *table = sd_alloc_ctl_entry(13);
e692ab53 5936
ad1cdc1d
MM
5937 if (table == NULL)
5938 return NULL;
5939
e0361851 5940 set_table_entry(&table[0], "min_interval", &sd->min_interval,
e692ab53 5941 sizeof(long), 0644, proc_doulongvec_minmax);
e0361851 5942 set_table_entry(&table[1], "max_interval", &sd->max_interval,
e692ab53 5943 sizeof(long), 0644, proc_doulongvec_minmax);
e0361851 5944 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
e692ab53 5945 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5946 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
e692ab53 5947 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5948 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
e692ab53 5949 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5950 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
e692ab53 5951 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5952 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
e692ab53 5953 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5954 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
e692ab53 5955 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5956 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
e692ab53 5957 sizeof(int), 0644, proc_dointvec_minmax);
ace8b3d6 5958 set_table_entry(&table[9], "cache_nice_tries",
e692ab53
NP
5959 &sd->cache_nice_tries,
5960 sizeof(int), 0644, proc_dointvec_minmax);
ace8b3d6 5961 set_table_entry(&table[10], "flags", &sd->flags,
e692ab53 5962 sizeof(int), 0644, proc_dointvec_minmax);
a5d8c348
IM
5963 set_table_entry(&table[11], "name", sd->name,
5964 CORENAME_MAX_SIZE, 0444, proc_dostring);
5965 /* &table[12] is terminator */
e692ab53
NP
5966
5967 return table;
5968}
5969
9a4e7159 5970static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
e692ab53
NP
5971{
5972 struct ctl_table *entry, *table;
5973 struct sched_domain *sd;
5974 int domain_num = 0, i;
5975 char buf[32];
5976
5977 for_each_domain(cpu, sd)
5978 domain_num++;
5979 entry = table = sd_alloc_ctl_entry(domain_num + 1);
ad1cdc1d
MM
5980 if (table == NULL)
5981 return NULL;
e692ab53
NP
5982
5983 i = 0;
5984 for_each_domain(cpu, sd) {
5985 snprintf(buf, 32, "domain%d", i);
e692ab53 5986 entry->procname = kstrdup(buf, GFP_KERNEL);
c57baf1e 5987 entry->mode = 0555;
e692ab53
NP
5988 entry->child = sd_alloc_ctl_domain_table(sd);
5989 entry++;
5990 i++;
5991 }
5992 return table;
5993}
5994
5995static struct ctl_table_header *sd_sysctl_header;
6382bc90 5996static void register_sched_domain_sysctl(void)
e692ab53 5997{
6ad4c188 5998 int i, cpu_num = num_possible_cpus();
e692ab53
NP
5999 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
6000 char buf[32];
6001
7378547f
MM
6002 WARN_ON(sd_ctl_dir[0].child);
6003 sd_ctl_dir[0].child = entry;
6004
ad1cdc1d
MM
6005 if (entry == NULL)
6006 return;
6007
6ad4c188 6008 for_each_possible_cpu(i) {
e692ab53 6009 snprintf(buf, 32, "cpu%d", i);
e692ab53 6010 entry->procname = kstrdup(buf, GFP_KERNEL);
c57baf1e 6011 entry->mode = 0555;
e692ab53 6012 entry->child = sd_alloc_ctl_cpu_table(i);
97b6ea7b 6013 entry++;
e692ab53 6014 }
7378547f
MM
6015
6016 WARN_ON(sd_sysctl_header);
e692ab53
NP
6017 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
6018}
6382bc90 6019
7378547f 6020/* may be called multiple times per register */
6382bc90
MM
6021static void unregister_sched_domain_sysctl(void)
6022{
7378547f
MM
6023 if (sd_sysctl_header)
6024 unregister_sysctl_table(sd_sysctl_header);
6382bc90 6025 sd_sysctl_header = NULL;
7378547f
MM
6026 if (sd_ctl_dir[0].child)
6027 sd_free_ctl_entry(&sd_ctl_dir[0].child);
6382bc90 6028}
e692ab53 6029#else
6382bc90
MM
6030static void register_sched_domain_sysctl(void)
6031{
6032}
6033static void unregister_sched_domain_sysctl(void)
e692ab53
NP
6034{
6035}
6036#endif
6037
1f11eb6a
GH
6038static void set_rq_online(struct rq *rq)
6039{
6040 if (!rq->online) {
6041 const struct sched_class *class;
6042
c6c4927b 6043 cpumask_set_cpu(rq->cpu, rq->rd->online);
1f11eb6a
GH
6044 rq->online = 1;
6045
6046 for_each_class(class) {
6047 if (class->rq_online)
6048 class->rq_online(rq);
6049 }
6050 }
6051}
6052
6053static void set_rq_offline(struct rq *rq)
6054{
6055 if (rq->online) {
6056 const struct sched_class *class;
6057
6058 for_each_class(class) {
6059 if (class->rq_offline)
6060 class->rq_offline(rq);
6061 }
6062
c6c4927b 6063 cpumask_clear_cpu(rq->cpu, rq->rd->online);
1f11eb6a
GH
6064 rq->online = 0;
6065 }
6066}
6067
1da177e4
LT
6068/*
6069 * migration_call - callback that gets triggered when a CPU is added.
6070 * Here we can start up the necessary migration thread for the new CPU.
6071 */
48f24c4d
IM
6072static int __cpuinit
6073migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
1da177e4 6074{
48f24c4d 6075 int cpu = (long)hcpu;
1da177e4 6076 unsigned long flags;
969c7921 6077 struct rq *rq = cpu_rq(cpu);
1da177e4 6078
48c5ccae 6079 switch (action & ~CPU_TASKS_FROZEN) {
5be9361c 6080
1da177e4 6081 case CPU_UP_PREPARE:
a468d389 6082 rq->calc_load_update = calc_load_update;
1da177e4 6083 break;
48f24c4d 6084
1da177e4 6085 case CPU_ONLINE:
1f94ef59 6086 /* Update our root-domain */
05fa785c 6087 raw_spin_lock_irqsave(&rq->lock, flags);
1f94ef59 6088 if (rq->rd) {
c6c4927b 6089 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
1f11eb6a
GH
6090
6091 set_rq_online(rq);
1f94ef59 6092 }
05fa785c 6093 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4 6094 break;
48f24c4d 6095
1da177e4 6096#ifdef CONFIG_HOTPLUG_CPU
08f503b0 6097 case CPU_DYING:
57d885fe 6098 /* Update our root-domain */
05fa785c 6099 raw_spin_lock_irqsave(&rq->lock, flags);
57d885fe 6100 if (rq->rd) {
c6c4927b 6101 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
1f11eb6a 6102 set_rq_offline(rq);
57d885fe 6103 }
48c5ccae
PZ
6104 migrate_tasks(cpu);
6105 BUG_ON(rq->nr_running != 1); /* the migration thread */
05fa785c 6106 raw_spin_unlock_irqrestore(&rq->lock, flags);
48c5ccae
PZ
6107
6108 migrate_nr_uninterruptible(rq);
6109 calc_global_load_remove(rq);
57d885fe 6110 break;
1da177e4
LT
6111#endif
6112 }
6113 return NOTIFY_OK;
6114}
6115
f38b0820
PM
6116/*
6117 * Register at high priority so that task migration (migrate_all_tasks)
6118 * happens before everything else. This has to be lower priority than
cdd6c482 6119 * the notifier in the perf_event subsystem, though.
1da177e4 6120 */
26c2143b 6121static struct notifier_block __cpuinitdata migration_notifier = {
1da177e4 6122 .notifier_call = migration_call,
50a323b7 6123 .priority = CPU_PRI_MIGRATION,
1da177e4
LT
6124};
6125
3a101d05
TH
6126static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
6127 unsigned long action, void *hcpu)
6128{
6129 switch (action & ~CPU_TASKS_FROZEN) {
6130 case CPU_ONLINE:
6131 case CPU_DOWN_FAILED:
6132 set_cpu_active((long)hcpu, true);
6133 return NOTIFY_OK;
6134 default:
6135 return NOTIFY_DONE;
6136 }
6137}
6138
6139static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
6140 unsigned long action, void *hcpu)
6141{
6142 switch (action & ~CPU_TASKS_FROZEN) {
6143 case CPU_DOWN_PREPARE:
6144 set_cpu_active((long)hcpu, false);
6145 return NOTIFY_OK;
6146 default:
6147 return NOTIFY_DONE;
6148 }
6149}
6150
7babe8db 6151static int __init migration_init(void)
1da177e4
LT
6152{
6153 void *cpu = (void *)(long)smp_processor_id();
07dccf33 6154 int err;
48f24c4d 6155
3a101d05 6156 /* Initialize migration for the boot CPU */
07dccf33
AM
6157 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6158 BUG_ON(err == NOTIFY_BAD);
1da177e4
LT
6159 migration_call(&migration_notifier, CPU_ONLINE, cpu);
6160 register_cpu_notifier(&migration_notifier);
7babe8db 6161
3a101d05
TH
6162 /* Register cpu active notifiers */
6163 cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
6164 cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
6165
a004cd42 6166 return 0;
1da177e4 6167}
7babe8db 6168early_initcall(migration_init);
1da177e4
LT
6169#endif
6170
6171#ifdef CONFIG_SMP
476f3534 6172
3e9830dc 6173#ifdef CONFIG_SCHED_DEBUG
4dcf6aff 6174
f6630114
MT
6175static __read_mostly int sched_domain_debug_enabled;
6176
6177static int __init sched_domain_debug_setup(char *str)
6178{
6179 sched_domain_debug_enabled = 1;
6180
6181 return 0;
6182}
6183early_param("sched_debug", sched_domain_debug_setup);
6184
7c16ec58 6185static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
96f874e2 6186 struct cpumask *groupmask)
1da177e4 6187{
4dcf6aff 6188 struct sched_group *group = sd->groups;
434d53b0 6189 char str[256];
1da177e4 6190
968ea6d8 6191 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
96f874e2 6192 cpumask_clear(groupmask);
4dcf6aff
IM
6193
6194 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6195
6196 if (!(sd->flags & SD_LOAD_BALANCE)) {
3df0fc5b 6197 printk("does not load-balance\n");
4dcf6aff 6198 if (sd->parent)
3df0fc5b
PZ
6199 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6200 " has parent");
4dcf6aff 6201 return -1;
41c7ce9a
NP
6202 }
6203
3df0fc5b 6204 printk(KERN_CONT "span %s level %s\n", str, sd->name);
4dcf6aff 6205
758b2cdc 6206 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
3df0fc5b
PZ
6207 printk(KERN_ERR "ERROR: domain->span does not contain "
6208 "CPU%d\n", cpu);
4dcf6aff 6209 }
758b2cdc 6210 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
3df0fc5b
PZ
6211 printk(KERN_ERR "ERROR: domain->groups does not contain"
6212 " CPU%d\n", cpu);
4dcf6aff 6213 }
1da177e4 6214
4dcf6aff 6215 printk(KERN_DEBUG "%*s groups:", level + 1, "");
1da177e4 6216 do {
4dcf6aff 6217 if (!group) {
3df0fc5b
PZ
6218 printk("\n");
6219 printk(KERN_ERR "ERROR: group is NULL\n");
1da177e4
LT
6220 break;
6221 }
6222
18a3885f 6223 if (!group->cpu_power) {
3df0fc5b
PZ
6224 printk(KERN_CONT "\n");
6225 printk(KERN_ERR "ERROR: domain->cpu_power not "
6226 "set\n");
4dcf6aff
IM
6227 break;
6228 }
1da177e4 6229
758b2cdc 6230 if (!cpumask_weight(sched_group_cpus(group))) {
3df0fc5b
PZ
6231 printk(KERN_CONT "\n");
6232 printk(KERN_ERR "ERROR: empty group\n");
4dcf6aff
IM
6233 break;
6234 }
1da177e4 6235
758b2cdc 6236 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
3df0fc5b
PZ
6237 printk(KERN_CONT "\n");
6238 printk(KERN_ERR "ERROR: repeated CPUs\n");
4dcf6aff
IM
6239 break;
6240 }
1da177e4 6241
758b2cdc 6242 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
1da177e4 6243
968ea6d8 6244 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
381512cf 6245
3df0fc5b 6246 printk(KERN_CONT " %s", str);
18a3885f 6247 if (group->cpu_power != SCHED_LOAD_SCALE) {
3df0fc5b
PZ
6248 printk(KERN_CONT " (cpu_power = %d)",
6249 group->cpu_power);
381512cf 6250 }
1da177e4 6251
4dcf6aff
IM
6252 group = group->next;
6253 } while (group != sd->groups);
3df0fc5b 6254 printk(KERN_CONT "\n");
1da177e4 6255
758b2cdc 6256 if (!cpumask_equal(sched_domain_span(sd), groupmask))
3df0fc5b 6257 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
1da177e4 6258
758b2cdc
RR
6259 if (sd->parent &&
6260 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
3df0fc5b
PZ
6261 printk(KERN_ERR "ERROR: parent span is not a superset "
6262 "of domain->span\n");
4dcf6aff
IM
6263 return 0;
6264}
1da177e4 6265
4dcf6aff
IM
6266static void sched_domain_debug(struct sched_domain *sd, int cpu)
6267{
d5dd3db1 6268 cpumask_var_t groupmask;
4dcf6aff 6269 int level = 0;
1da177e4 6270
f6630114
MT
6271 if (!sched_domain_debug_enabled)
6272 return;
6273
4dcf6aff
IM
6274 if (!sd) {
6275 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6276 return;
6277 }
1da177e4 6278
4dcf6aff
IM
6279 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6280
d5dd3db1 6281 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
7c16ec58
MT
6282 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6283 return;
6284 }
6285
4dcf6aff 6286 for (;;) {
7c16ec58 6287 if (sched_domain_debug_one(sd, cpu, level, groupmask))
4dcf6aff 6288 break;
1da177e4
LT
6289 level++;
6290 sd = sd->parent;
33859f7f 6291 if (!sd)
4dcf6aff
IM
6292 break;
6293 }
d5dd3db1 6294 free_cpumask_var(groupmask);
1da177e4 6295}
6d6bc0ad 6296#else /* !CONFIG_SCHED_DEBUG */
48f24c4d 6297# define sched_domain_debug(sd, cpu) do { } while (0)
6d6bc0ad 6298#endif /* CONFIG_SCHED_DEBUG */
1da177e4 6299
1a20ff27 6300static int sd_degenerate(struct sched_domain *sd)
245af2c7 6301{
758b2cdc 6302 if (cpumask_weight(sched_domain_span(sd)) == 1)
245af2c7
SS
6303 return 1;
6304
6305 /* Following flags need at least 2 groups */
6306 if (sd->flags & (SD_LOAD_BALANCE |
6307 SD_BALANCE_NEWIDLE |
6308 SD_BALANCE_FORK |
89c4710e
SS
6309 SD_BALANCE_EXEC |
6310 SD_SHARE_CPUPOWER |
6311 SD_SHARE_PKG_RESOURCES)) {
245af2c7
SS
6312 if (sd->groups != sd->groups->next)
6313 return 0;
6314 }
6315
6316 /* Following flags don't use groups */
c88d5910 6317 if (sd->flags & (SD_WAKE_AFFINE))
245af2c7
SS
6318 return 0;
6319
6320 return 1;
6321}
6322
48f24c4d
IM
6323static int
6324sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
245af2c7
SS
6325{
6326 unsigned long cflags = sd->flags, pflags = parent->flags;
6327
6328 if (sd_degenerate(parent))
6329 return 1;
6330
758b2cdc 6331 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
245af2c7
SS
6332 return 0;
6333
245af2c7
SS
6334 /* Flags needing groups don't count if only 1 group in parent */
6335 if (parent->groups == parent->groups->next) {
6336 pflags &= ~(SD_LOAD_BALANCE |
6337 SD_BALANCE_NEWIDLE |
6338 SD_BALANCE_FORK |
89c4710e
SS
6339 SD_BALANCE_EXEC |
6340 SD_SHARE_CPUPOWER |
6341 SD_SHARE_PKG_RESOURCES);
5436499e
KC
6342 if (nr_node_ids == 1)
6343 pflags &= ~SD_SERIALIZE;
245af2c7
SS
6344 }
6345 if (~cflags & pflags)
6346 return 0;
6347
6348 return 1;
6349}
6350
c6c4927b
RR
6351static void free_rootdomain(struct root_domain *rd)
6352{
047106ad
PZ
6353 synchronize_sched();
6354
68e74568
RR
6355 cpupri_cleanup(&rd->cpupri);
6356
c6c4927b
RR
6357 free_cpumask_var(rd->rto_mask);
6358 free_cpumask_var(rd->online);
6359 free_cpumask_var(rd->span);
6360 kfree(rd);
6361}
6362
57d885fe
GH
6363static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6364{
a0490fa3 6365 struct root_domain *old_rd = NULL;
57d885fe 6366 unsigned long flags;
57d885fe 6367
05fa785c 6368 raw_spin_lock_irqsave(&rq->lock, flags);
57d885fe
GH
6369
6370 if (rq->rd) {
a0490fa3 6371 old_rd = rq->rd;
57d885fe 6372
c6c4927b 6373 if (cpumask_test_cpu(rq->cpu, old_rd->online))
1f11eb6a 6374 set_rq_offline(rq);
57d885fe 6375
c6c4927b 6376 cpumask_clear_cpu(rq->cpu, old_rd->span);
dc938520 6377
a0490fa3
IM
6378 /*
6379 * If we dont want to free the old_rt yet then
6380 * set old_rd to NULL to skip the freeing later
6381 * in this function:
6382 */
6383 if (!atomic_dec_and_test(&old_rd->refcount))
6384 old_rd = NULL;
57d885fe
GH
6385 }
6386
6387 atomic_inc(&rd->refcount);
6388 rq->rd = rd;
6389
c6c4927b 6390 cpumask_set_cpu(rq->cpu, rd->span);
00aec93d 6391 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
1f11eb6a 6392 set_rq_online(rq);
57d885fe 6393
05fa785c 6394 raw_spin_unlock_irqrestore(&rq->lock, flags);
a0490fa3
IM
6395
6396 if (old_rd)
6397 free_rootdomain(old_rd);
57d885fe
GH
6398}
6399
68c38fc3 6400static int init_rootdomain(struct root_domain *rd)
57d885fe
GH
6401{
6402 memset(rd, 0, sizeof(*rd));
6403
68c38fc3 6404 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
0c910d28 6405 goto out;
68c38fc3 6406 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
c6c4927b 6407 goto free_span;
68c38fc3 6408 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
c6c4927b 6409 goto free_online;
6e0534f2 6410
68c38fc3 6411 if (cpupri_init(&rd->cpupri) != 0)
68e74568 6412 goto free_rto_mask;
c6c4927b 6413 return 0;
6e0534f2 6414
68e74568
RR
6415free_rto_mask:
6416 free_cpumask_var(rd->rto_mask);
c6c4927b
RR
6417free_online:
6418 free_cpumask_var(rd->online);
6419free_span:
6420 free_cpumask_var(rd->span);
0c910d28 6421out:
c6c4927b 6422 return -ENOMEM;
57d885fe
GH
6423}
6424
6425static void init_defrootdomain(void)
6426{
68c38fc3 6427 init_rootdomain(&def_root_domain);
c6c4927b 6428
57d885fe
GH
6429 atomic_set(&def_root_domain.refcount, 1);
6430}
6431
dc938520 6432static struct root_domain *alloc_rootdomain(void)
57d885fe
GH
6433{
6434 struct root_domain *rd;
6435
6436 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6437 if (!rd)
6438 return NULL;
6439
68c38fc3 6440 if (init_rootdomain(rd) != 0) {
c6c4927b
RR
6441 kfree(rd);
6442 return NULL;
6443 }
57d885fe
GH
6444
6445 return rd;
6446}
6447
1da177e4 6448/*
0eab9146 6449 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
1da177e4
LT
6450 * hold the hotplug lock.
6451 */
0eab9146
IM
6452static void
6453cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
1da177e4 6454{
70b97a7f 6455 struct rq *rq = cpu_rq(cpu);
245af2c7
SS
6456 struct sched_domain *tmp;
6457
669c55e9
PZ
6458 for (tmp = sd; tmp; tmp = tmp->parent)
6459 tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
6460
245af2c7 6461 /* Remove the sched domains which do not contribute to scheduling. */
f29c9b1c 6462 for (tmp = sd; tmp; ) {
245af2c7
SS
6463 struct sched_domain *parent = tmp->parent;
6464 if (!parent)
6465 break;
f29c9b1c 6466
1a848870 6467 if (sd_parent_degenerate(tmp, parent)) {
245af2c7 6468 tmp->parent = parent->parent;
1a848870
SS
6469 if (parent->parent)
6470 parent->parent->child = tmp;
f29c9b1c
LZ
6471 } else
6472 tmp = tmp->parent;
245af2c7
SS
6473 }
6474
1a848870 6475 if (sd && sd_degenerate(sd)) {
245af2c7 6476 sd = sd->parent;
1a848870
SS
6477 if (sd)
6478 sd->child = NULL;
6479 }
1da177e4
LT
6480
6481 sched_domain_debug(sd, cpu);
6482
57d885fe 6483 rq_attach_root(rq, rd);
674311d5 6484 rcu_assign_pointer(rq->sd, sd);
1da177e4
LT
6485}
6486
6487/* cpus with isolated domains */
dcc30a35 6488static cpumask_var_t cpu_isolated_map;
1da177e4
LT
6489
6490/* Setup the mask of cpus configured for isolated domains */
6491static int __init isolated_cpu_setup(char *str)
6492{
bdddd296 6493 alloc_bootmem_cpumask_var(&cpu_isolated_map);
968ea6d8 6494 cpulist_parse(str, cpu_isolated_map);
1da177e4
LT
6495 return 1;
6496}
6497
8927f494 6498__setup("isolcpus=", isolated_cpu_setup);
1da177e4
LT
6499
6500/*
6711cab4
SS
6501 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6502 * to a function which identifies what group(along with sched group) a CPU
96f874e2
RR
6503 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6504 * (due to the fact that we keep track of groups covered with a struct cpumask).
1da177e4
LT
6505 *
6506 * init_sched_build_groups will build a circular linked list of the groups
6507 * covered by the given span, and will set each group's ->cpumask correctly,
6508 * and ->cpu_power to 0.
6509 */
a616058b 6510static void
96f874e2
RR
6511init_sched_build_groups(const struct cpumask *span,
6512 const struct cpumask *cpu_map,
6513 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
7c16ec58 6514 struct sched_group **sg,
96f874e2
RR
6515 struct cpumask *tmpmask),
6516 struct cpumask *covered, struct cpumask *tmpmask)
1da177e4
LT
6517{
6518 struct sched_group *first = NULL, *last = NULL;
1da177e4
LT
6519 int i;
6520
96f874e2 6521 cpumask_clear(covered);
7c16ec58 6522
abcd083a 6523 for_each_cpu(i, span) {
6711cab4 6524 struct sched_group *sg;
7c16ec58 6525 int group = group_fn(i, cpu_map, &sg, tmpmask);
1da177e4
LT
6526 int j;
6527
758b2cdc 6528 if (cpumask_test_cpu(i, covered))
1da177e4
LT
6529 continue;
6530
758b2cdc 6531 cpumask_clear(sched_group_cpus(sg));
18a3885f 6532 sg->cpu_power = 0;
1da177e4 6533
abcd083a 6534 for_each_cpu(j, span) {
7c16ec58 6535 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
1da177e4
LT
6536 continue;
6537
96f874e2 6538 cpumask_set_cpu(j, covered);
758b2cdc 6539 cpumask_set_cpu(j, sched_group_cpus(sg));
1da177e4
LT
6540 }
6541 if (!first)
6542 first = sg;
6543 if (last)
6544 last->next = sg;
6545 last = sg;
6546 }
6547 last->next = first;
6548}
6549
9c1cfda2 6550#define SD_NODES_PER_DOMAIN 16
1da177e4 6551
9c1cfda2 6552#ifdef CONFIG_NUMA
198e2f18 6553
9c1cfda2
JH
6554/**
6555 * find_next_best_node - find the next node to include in a sched_domain
6556 * @node: node whose sched_domain we're building
6557 * @used_nodes: nodes already in the sched_domain
6558 *
41a2d6cf 6559 * Find the next node to include in a given scheduling domain. Simply
9c1cfda2
JH
6560 * finds the closest node not already in the @used_nodes map.
6561 *
6562 * Should use nodemask_t.
6563 */
c5f59f08 6564static int find_next_best_node(int node, nodemask_t *used_nodes)
9c1cfda2
JH
6565{
6566 int i, n, val, min_val, best_node = 0;
6567
6568 min_val = INT_MAX;
6569
076ac2af 6570 for (i = 0; i < nr_node_ids; i++) {
9c1cfda2 6571 /* Start at @node */
076ac2af 6572 n = (node + i) % nr_node_ids;
9c1cfda2
JH
6573
6574 if (!nr_cpus_node(n))
6575 continue;
6576
6577 /* Skip already used nodes */
c5f59f08 6578 if (node_isset(n, *used_nodes))
9c1cfda2
JH
6579 continue;
6580
6581 /* Simple min distance search */
6582 val = node_distance(node, n);
6583
6584 if (val < min_val) {
6585 min_val = val;
6586 best_node = n;
6587 }
6588 }
6589
c5f59f08 6590 node_set(best_node, *used_nodes);
9c1cfda2
JH
6591 return best_node;
6592}
6593
6594/**
6595 * sched_domain_node_span - get a cpumask for a node's sched_domain
6596 * @node: node whose cpumask we're constructing
73486722 6597 * @span: resulting cpumask
9c1cfda2 6598 *
41a2d6cf 6599 * Given a node, construct a good cpumask for its sched_domain to span. It
9c1cfda2
JH
6600 * should be one that prevents unnecessary balancing, but also spreads tasks
6601 * out optimally.
6602 */
96f874e2 6603static void sched_domain_node_span(int node, struct cpumask *span)
9c1cfda2 6604{
c5f59f08 6605 nodemask_t used_nodes;
48f24c4d 6606 int i;
9c1cfda2 6607
6ca09dfc 6608 cpumask_clear(span);
c5f59f08 6609 nodes_clear(used_nodes);
9c1cfda2 6610
6ca09dfc 6611 cpumask_or(span, span, cpumask_of_node(node));
c5f59f08 6612 node_set(node, used_nodes);
9c1cfda2
JH
6613
6614 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
c5f59f08 6615 int next_node = find_next_best_node(node, &used_nodes);
48f24c4d 6616
6ca09dfc 6617 cpumask_or(span, span, cpumask_of_node(next_node));
9c1cfda2 6618 }
9c1cfda2 6619}
6d6bc0ad 6620#endif /* CONFIG_NUMA */
9c1cfda2 6621
5c45bf27 6622int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
48f24c4d 6623
6c99e9ad
RR
6624/*
6625 * The cpus mask in sched_group and sched_domain hangs off the end.
4200efd9
IM
6626 *
6627 * ( See the the comments in include/linux/sched.h:struct sched_group
6628 * and struct sched_domain. )
6c99e9ad
RR
6629 */
6630struct static_sched_group {
6631 struct sched_group sg;
6632 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
6633};
6634
6635struct static_sched_domain {
6636 struct sched_domain sd;
6637 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
6638};
6639
49a02c51
AH
6640struct s_data {
6641#ifdef CONFIG_NUMA
6642 int sd_allnodes;
6643 cpumask_var_t domainspan;
6644 cpumask_var_t covered;
6645 cpumask_var_t notcovered;
6646#endif
6647 cpumask_var_t nodemask;
6648 cpumask_var_t this_sibling_map;
6649 cpumask_var_t this_core_map;
01a08546 6650 cpumask_var_t this_book_map;
49a02c51
AH
6651 cpumask_var_t send_covered;
6652 cpumask_var_t tmpmask;
6653 struct sched_group **sched_group_nodes;
6654 struct root_domain *rd;
6655};
6656
2109b99e
AH
6657enum s_alloc {
6658 sa_sched_groups = 0,
6659 sa_rootdomain,
6660 sa_tmpmask,
6661 sa_send_covered,
01a08546 6662 sa_this_book_map,
2109b99e
AH
6663 sa_this_core_map,
6664 sa_this_sibling_map,
6665 sa_nodemask,
6666 sa_sched_group_nodes,
6667#ifdef CONFIG_NUMA
6668 sa_notcovered,
6669 sa_covered,
6670 sa_domainspan,
6671#endif
6672 sa_none,
6673};
6674
9c1cfda2 6675/*
48f24c4d 6676 * SMT sched-domains:
9c1cfda2 6677 */
1da177e4 6678#ifdef CONFIG_SCHED_SMT
6c99e9ad 6679static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
1871e52c 6680static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
48f24c4d 6681
41a2d6cf 6682static int
96f874e2
RR
6683cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6684 struct sched_group **sg, struct cpumask *unused)
1da177e4 6685{
6711cab4 6686 if (sg)
1871e52c 6687 *sg = &per_cpu(sched_groups, cpu).sg;
1da177e4
LT
6688 return cpu;
6689}
6d6bc0ad 6690#endif /* CONFIG_SCHED_SMT */
1da177e4 6691
48f24c4d
IM
6692/*
6693 * multi-core sched-domains:
6694 */
1e9f28fa 6695#ifdef CONFIG_SCHED_MC
6c99e9ad
RR
6696static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6697static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
1e9f28fa 6698
41a2d6cf 6699static int
96f874e2
RR
6700cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6701 struct sched_group **sg, struct cpumask *mask)
1e9f28fa 6702{
6711cab4 6703 int group;
f269893c 6704#ifdef CONFIG_SCHED_SMT
c69fc56d 6705 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
96f874e2 6706 group = cpumask_first(mask);
f269893c
HC
6707#else
6708 group = cpu;
6709#endif
6711cab4 6710 if (sg)
6c99e9ad 6711 *sg = &per_cpu(sched_group_core, group).sg;
6711cab4 6712 return group;
1e9f28fa 6713}
f269893c 6714#endif /* CONFIG_SCHED_MC */
1e9f28fa 6715
01a08546
HC
6716/*
6717 * book sched-domains:
6718 */
6719#ifdef CONFIG_SCHED_BOOK
6720static DEFINE_PER_CPU(struct static_sched_domain, book_domains);
6721static DEFINE_PER_CPU(struct static_sched_group, sched_group_book);
6722
41a2d6cf 6723static int
01a08546
HC
6724cpu_to_book_group(int cpu, const struct cpumask *cpu_map,
6725 struct sched_group **sg, struct cpumask *mask)
1e9f28fa 6726{
01a08546
HC
6727 int group = cpu;
6728#ifdef CONFIG_SCHED_MC
6729 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
6730 group = cpumask_first(mask);
6731#elif defined(CONFIG_SCHED_SMT)
6732 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
6733 group = cpumask_first(mask);
6734#endif
6711cab4 6735 if (sg)
01a08546
HC
6736 *sg = &per_cpu(sched_group_book, group).sg;
6737 return group;
1e9f28fa 6738}
01a08546 6739#endif /* CONFIG_SCHED_BOOK */
1e9f28fa 6740
6c99e9ad
RR
6741static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6742static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
48f24c4d 6743
41a2d6cf 6744static int
96f874e2
RR
6745cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6746 struct sched_group **sg, struct cpumask *mask)
1da177e4 6747{
6711cab4 6748 int group;
01a08546
HC
6749#ifdef CONFIG_SCHED_BOOK
6750 cpumask_and(mask, cpu_book_mask(cpu), cpu_map);
6751 group = cpumask_first(mask);
6752#elif defined(CONFIG_SCHED_MC)
6ca09dfc 6753 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
96f874e2 6754 group = cpumask_first(mask);
1e9f28fa 6755#elif defined(CONFIG_SCHED_SMT)
c69fc56d 6756 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
96f874e2 6757 group = cpumask_first(mask);
1da177e4 6758#else
6711cab4 6759 group = cpu;
1da177e4 6760#endif
6711cab4 6761 if (sg)
6c99e9ad 6762 *sg = &per_cpu(sched_group_phys, group).sg;
6711cab4 6763 return group;
1da177e4
LT
6764}
6765
6766#ifdef CONFIG_NUMA
1da177e4 6767/*
9c1cfda2
JH
6768 * The init_sched_build_groups can't handle what we want to do with node
6769 * groups, so roll our own. Now each node has its own list of groups which
6770 * gets dynamically allocated.
1da177e4 6771 */
62ea9ceb 6772static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
434d53b0 6773static struct sched_group ***sched_group_nodes_bycpu;
1da177e4 6774
62ea9ceb 6775static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
6c99e9ad 6776static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
9c1cfda2 6777
96f874e2
RR
6778static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
6779 struct sched_group **sg,
6780 struct cpumask *nodemask)
9c1cfda2 6781{
6711cab4
SS
6782 int group;
6783
6ca09dfc 6784 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
96f874e2 6785 group = cpumask_first(nodemask);
6711cab4
SS
6786
6787 if (sg)
6c99e9ad 6788 *sg = &per_cpu(sched_group_allnodes, group).sg;
6711cab4 6789 return group;
1da177e4 6790}
6711cab4 6791
08069033
SS
6792static void init_numa_sched_groups_power(struct sched_group *group_head)
6793{
6794 struct sched_group *sg = group_head;
6795 int j;
6796
6797 if (!sg)
6798 return;
3a5c359a 6799 do {
758b2cdc 6800 for_each_cpu(j, sched_group_cpus(sg)) {
3a5c359a 6801 struct sched_domain *sd;
08069033 6802
6c99e9ad 6803 sd = &per_cpu(phys_domains, j).sd;
13318a71 6804 if (j != group_first_cpu(sd->groups)) {
3a5c359a
AK
6805 /*
6806 * Only add "power" once for each
6807 * physical package.
6808 */
6809 continue;
6810 }
08069033 6811
18a3885f 6812 sg->cpu_power += sd->groups->cpu_power;
3a5c359a
AK
6813 }
6814 sg = sg->next;
6815 } while (sg != group_head);
08069033 6816}
0601a88d
AH
6817
6818static int build_numa_sched_groups(struct s_data *d,
6819 const struct cpumask *cpu_map, int num)
6820{
6821 struct sched_domain *sd;
6822 struct sched_group *sg, *prev;
6823 int n, j;
6824
6825 cpumask_clear(d->covered);
6826 cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
6827 if (cpumask_empty(d->nodemask)) {
6828 d->sched_group_nodes[num] = NULL;
6829 goto out;
6830 }
6831
6832 sched_domain_node_span(num, d->domainspan);
6833 cpumask_and(d->domainspan, d->domainspan, cpu_map);
6834
6835 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6836 GFP_KERNEL, num);
6837 if (!sg) {
3df0fc5b
PZ
6838 printk(KERN_WARNING "Can not alloc domain group for node %d\n",
6839 num);
0601a88d
AH
6840 return -ENOMEM;
6841 }
6842 d->sched_group_nodes[num] = sg;
6843
6844 for_each_cpu(j, d->nodemask) {
6845 sd = &per_cpu(node_domains, j).sd;
6846 sd->groups = sg;
6847 }
6848
18a3885f 6849 sg->cpu_power = 0;
0601a88d
AH
6850 cpumask_copy(sched_group_cpus(sg), d->nodemask);
6851 sg->next = sg;
6852 cpumask_or(d->covered, d->covered, d->nodemask);
6853
6854 prev = sg;
6855 for (j = 0; j < nr_node_ids; j++) {
6856 n = (num + j) % nr_node_ids;
6857 cpumask_complement(d->notcovered, d->covered);
6858 cpumask_and(d->tmpmask, d->notcovered, cpu_map);
6859 cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
6860 if (cpumask_empty(d->tmpmask))
6861 break;
6862 cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
6863 if (cpumask_empty(d->tmpmask))
6864 continue;
6865 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6866 GFP_KERNEL, num);
6867 if (!sg) {
3df0fc5b
PZ
6868 printk(KERN_WARNING
6869 "Can not alloc domain group for node %d\n", j);
0601a88d
AH
6870 return -ENOMEM;
6871 }
18a3885f 6872 sg->cpu_power = 0;
0601a88d
AH
6873 cpumask_copy(sched_group_cpus(sg), d->tmpmask);
6874 sg->next = prev->next;
6875 cpumask_or(d->covered, d->covered, d->tmpmask);
6876 prev->next = sg;
6877 prev = sg;
6878 }
6879out:
6880 return 0;
6881}
6d6bc0ad 6882#endif /* CONFIG_NUMA */
1da177e4 6883
a616058b 6884#ifdef CONFIG_NUMA
51888ca2 6885/* Free memory allocated for various sched_group structures */
96f874e2
RR
6886static void free_sched_groups(const struct cpumask *cpu_map,
6887 struct cpumask *nodemask)
51888ca2 6888{
a616058b 6889 int cpu, i;
51888ca2 6890
abcd083a 6891 for_each_cpu(cpu, cpu_map) {
51888ca2
SV
6892 struct sched_group **sched_group_nodes
6893 = sched_group_nodes_bycpu[cpu];
6894
51888ca2
SV
6895 if (!sched_group_nodes)
6896 continue;
6897
076ac2af 6898 for (i = 0; i < nr_node_ids; i++) {
51888ca2
SV
6899 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6900
6ca09dfc 6901 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
96f874e2 6902 if (cpumask_empty(nodemask))
51888ca2
SV
6903 continue;
6904
6905 if (sg == NULL)
6906 continue;
6907 sg = sg->next;
6908next_sg:
6909 oldsg = sg;
6910 sg = sg->next;
6911 kfree(oldsg);
6912 if (oldsg != sched_group_nodes[i])
6913 goto next_sg;
6914 }
6915 kfree(sched_group_nodes);
6916 sched_group_nodes_bycpu[cpu] = NULL;
6917 }
51888ca2 6918}
6d6bc0ad 6919#else /* !CONFIG_NUMA */
96f874e2
RR
6920static void free_sched_groups(const struct cpumask *cpu_map,
6921 struct cpumask *nodemask)
a616058b
SS
6922{
6923}
6d6bc0ad 6924#endif /* CONFIG_NUMA */
51888ca2 6925
89c4710e
SS
6926/*
6927 * Initialize sched groups cpu_power.
6928 *
6929 * cpu_power indicates the capacity of sched group, which is used while
6930 * distributing the load between different sched groups in a sched domain.
6931 * Typically cpu_power for all the groups in a sched domain will be same unless
6932 * there are asymmetries in the topology. If there are asymmetries, group
6933 * having more cpu_power will pickup more load compared to the group having
6934 * less cpu_power.
89c4710e
SS
6935 */
6936static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6937{
6938 struct sched_domain *child;
6939 struct sched_group *group;
f93e65c1
PZ
6940 long power;
6941 int weight;
89c4710e
SS
6942
6943 WARN_ON(!sd || !sd->groups);
6944
13318a71 6945 if (cpu != group_first_cpu(sd->groups))
89c4710e
SS
6946 return;
6947
aae6d3dd
SS
6948 sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups));
6949
89c4710e
SS
6950 child = sd->child;
6951
18a3885f 6952 sd->groups->cpu_power = 0;
5517d86b 6953
f93e65c1
PZ
6954 if (!child) {
6955 power = SCHED_LOAD_SCALE;
6956 weight = cpumask_weight(sched_domain_span(sd));
6957 /*
6958 * SMT siblings share the power of a single core.
a52bfd73
PZ
6959 * Usually multiple threads get a better yield out of
6960 * that one core than a single thread would have,
6961 * reflect that in sd->smt_gain.
f93e65c1 6962 */
a52bfd73
PZ
6963 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
6964 power *= sd->smt_gain;
f93e65c1 6965 power /= weight;
a52bfd73
PZ
6966 power >>= SCHED_LOAD_SHIFT;
6967 }
18a3885f 6968 sd->groups->cpu_power += power;
89c4710e
SS
6969 return;
6970 }
6971
89c4710e 6972 /*
f93e65c1 6973 * Add cpu_power of each child group to this groups cpu_power.
89c4710e
SS
6974 */
6975 group = child->groups;
6976 do {
18a3885f 6977 sd->groups->cpu_power += group->cpu_power;
89c4710e
SS
6978 group = group->next;
6979 } while (group != child->groups);
6980}
6981
7c16ec58
MT
6982/*
6983 * Initializers for schedule domains
6984 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
6985 */
6986
a5d8c348
IM
6987#ifdef CONFIG_SCHED_DEBUG
6988# define SD_INIT_NAME(sd, type) sd->name = #type
6989#else
6990# define SD_INIT_NAME(sd, type) do { } while (0)
6991#endif
6992
7c16ec58 6993#define SD_INIT(sd, type) sd_init_##type(sd)
a5d8c348 6994
7c16ec58
MT
6995#define SD_INIT_FUNC(type) \
6996static noinline void sd_init_##type(struct sched_domain *sd) \
6997{ \
6998 memset(sd, 0, sizeof(*sd)); \
6999 *sd = SD_##type##_INIT; \
1d3504fc 7000 sd->level = SD_LV_##type; \
a5d8c348 7001 SD_INIT_NAME(sd, type); \
7c16ec58
MT
7002}
7003
7004SD_INIT_FUNC(CPU)
7005#ifdef CONFIG_NUMA
7006 SD_INIT_FUNC(ALLNODES)
7007 SD_INIT_FUNC(NODE)
7008#endif
7009#ifdef CONFIG_SCHED_SMT
7010 SD_INIT_FUNC(SIBLING)
7011#endif
7012#ifdef CONFIG_SCHED_MC
7013 SD_INIT_FUNC(MC)
7014#endif
01a08546
HC
7015#ifdef CONFIG_SCHED_BOOK
7016 SD_INIT_FUNC(BOOK)
7017#endif
7c16ec58 7018
1d3504fc
HS
7019static int default_relax_domain_level = -1;
7020
7021static int __init setup_relax_domain_level(char *str)
7022{
30e0e178
LZ
7023 unsigned long val;
7024
7025 val = simple_strtoul(str, NULL, 0);
7026 if (val < SD_LV_MAX)
7027 default_relax_domain_level = val;
7028
1d3504fc
HS
7029 return 1;
7030}
7031__setup("relax_domain_level=", setup_relax_domain_level);
7032
7033static void set_domain_attribute(struct sched_domain *sd,
7034 struct sched_domain_attr *attr)
7035{
7036 int request;
7037
7038 if (!attr || attr->relax_domain_level < 0) {
7039 if (default_relax_domain_level < 0)
7040 return;
7041 else
7042 request = default_relax_domain_level;
7043 } else
7044 request = attr->relax_domain_level;
7045 if (request < sd->level) {
7046 /* turn off idle balance on this domain */
c88d5910 7047 sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
1d3504fc
HS
7048 } else {
7049 /* turn on idle balance on this domain */
c88d5910 7050 sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
1d3504fc
HS
7051 }
7052}
7053
2109b99e
AH
7054static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
7055 const struct cpumask *cpu_map)
7056{
7057 switch (what) {
7058 case sa_sched_groups:
7059 free_sched_groups(cpu_map, d->tmpmask); /* fall through */
7060 d->sched_group_nodes = NULL;
7061 case sa_rootdomain:
7062 free_rootdomain(d->rd); /* fall through */
7063 case sa_tmpmask:
7064 free_cpumask_var(d->tmpmask); /* fall through */
7065 case sa_send_covered:
7066 free_cpumask_var(d->send_covered); /* fall through */
01a08546
HC
7067 case sa_this_book_map:
7068 free_cpumask_var(d->this_book_map); /* fall through */
2109b99e
AH
7069 case sa_this_core_map:
7070 free_cpumask_var(d->this_core_map); /* fall through */
7071 case sa_this_sibling_map:
7072 free_cpumask_var(d->this_sibling_map); /* fall through */
7073 case sa_nodemask:
7074 free_cpumask_var(d->nodemask); /* fall through */
7075 case sa_sched_group_nodes:
d1b55138 7076#ifdef CONFIG_NUMA
2109b99e
AH
7077 kfree(d->sched_group_nodes); /* fall through */
7078 case sa_notcovered:
7079 free_cpumask_var(d->notcovered); /* fall through */
7080 case sa_covered:
7081 free_cpumask_var(d->covered); /* fall through */
7082 case sa_domainspan:
7083 free_cpumask_var(d->domainspan); /* fall through */
3404c8d9 7084#endif
2109b99e
AH
7085 case sa_none:
7086 break;
7087 }
7088}
3404c8d9 7089
2109b99e
AH
7090static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
7091 const struct cpumask *cpu_map)
7092{
3404c8d9 7093#ifdef CONFIG_NUMA
2109b99e
AH
7094 if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
7095 return sa_none;
7096 if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
7097 return sa_domainspan;
7098 if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
7099 return sa_covered;
7100 /* Allocate the per-node list of sched groups */
7101 d->sched_group_nodes = kcalloc(nr_node_ids,
7102 sizeof(struct sched_group *), GFP_KERNEL);
7103 if (!d->sched_group_nodes) {
3df0fc5b 7104 printk(KERN_WARNING "Can not alloc sched group node list\n");
2109b99e 7105 return sa_notcovered;
d1b55138 7106 }
2109b99e 7107 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
d1b55138 7108#endif
2109b99e
AH
7109 if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
7110 return sa_sched_group_nodes;
7111 if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
7112 return sa_nodemask;
7113 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
7114 return sa_this_sibling_map;
01a08546 7115 if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL))
2109b99e 7116 return sa_this_core_map;
01a08546
HC
7117 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
7118 return sa_this_book_map;
2109b99e
AH
7119 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
7120 return sa_send_covered;
7121 d->rd = alloc_rootdomain();
7122 if (!d->rd) {
3df0fc5b 7123 printk(KERN_WARNING "Cannot alloc root domain\n");
2109b99e 7124 return sa_tmpmask;
57d885fe 7125 }
2109b99e
AH
7126 return sa_rootdomain;
7127}
57d885fe 7128
7f4588f3
AH
7129static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
7130 const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
7131{
7132 struct sched_domain *sd = NULL;
7c16ec58 7133#ifdef CONFIG_NUMA
7f4588f3 7134 struct sched_domain *parent;
1da177e4 7135
7f4588f3
AH
7136 d->sd_allnodes = 0;
7137 if (cpumask_weight(cpu_map) >
7138 SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
7139 sd = &per_cpu(allnodes_domains, i).sd;
7140 SD_INIT(sd, ALLNODES);
1d3504fc 7141 set_domain_attribute(sd, attr);
7f4588f3
AH
7142 cpumask_copy(sched_domain_span(sd), cpu_map);
7143 cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
7144 d->sd_allnodes = 1;
7145 }
7146 parent = sd;
7147
7148 sd = &per_cpu(node_domains, i).sd;
7149 SD_INIT(sd, NODE);
7150 set_domain_attribute(sd, attr);
7151 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
7152 sd->parent = parent;
7153 if (parent)
7154 parent->child = sd;
7155 cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
1da177e4 7156#endif
7f4588f3
AH
7157 return sd;
7158}
1da177e4 7159
87cce662
AH
7160static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
7161 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7162 struct sched_domain *parent, int i)
7163{
7164 struct sched_domain *sd;
7165 sd = &per_cpu(phys_domains, i).sd;
7166 SD_INIT(sd, CPU);
7167 set_domain_attribute(sd, attr);
7168 cpumask_copy(sched_domain_span(sd), d->nodemask);
7169 sd->parent = parent;
7170 if (parent)
7171 parent->child = sd;
7172 cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
7173 return sd;
7174}
1da177e4 7175
01a08546
HC
7176static struct sched_domain *__build_book_sched_domain(struct s_data *d,
7177 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7178 struct sched_domain *parent, int i)
7179{
7180 struct sched_domain *sd = parent;
7181#ifdef CONFIG_SCHED_BOOK
7182 sd = &per_cpu(book_domains, i).sd;
7183 SD_INIT(sd, BOOK);
7184 set_domain_attribute(sd, attr);
7185 cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i));
7186 sd->parent = parent;
7187 parent->child = sd;
7188 cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask);
7189#endif
7190 return sd;
7191}
7192
410c4081
AH
7193static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
7194 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7195 struct sched_domain *parent, int i)
7196{
7197 struct sched_domain *sd = parent;
1e9f28fa 7198#ifdef CONFIG_SCHED_MC
410c4081
AH
7199 sd = &per_cpu(core_domains, i).sd;
7200 SD_INIT(sd, MC);
7201 set_domain_attribute(sd, attr);
7202 cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
7203 sd->parent = parent;
7204 parent->child = sd;
7205 cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
1e9f28fa 7206#endif
410c4081
AH
7207 return sd;
7208}
1e9f28fa 7209
d8173535
AH
7210static struct sched_domain *__build_smt_sched_domain(struct s_data *d,
7211 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7212 struct sched_domain *parent, int i)
7213{
7214 struct sched_domain *sd = parent;
1da177e4 7215#ifdef CONFIG_SCHED_SMT
d8173535
AH
7216 sd = &per_cpu(cpu_domains, i).sd;
7217 SD_INIT(sd, SIBLING);
7218 set_domain_attribute(sd, attr);
7219 cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
7220 sd->parent = parent;
7221 parent->child = sd;
7222 cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
1da177e4 7223#endif
d8173535
AH
7224 return sd;
7225}
1da177e4 7226
0e8e85c9
AH
7227static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
7228 const struct cpumask *cpu_map, int cpu)
7229{
7230 switch (l) {
1da177e4 7231#ifdef CONFIG_SCHED_SMT
0e8e85c9
AH
7232 case SD_LV_SIBLING: /* set up CPU (sibling) groups */
7233 cpumask_and(d->this_sibling_map, cpu_map,
7234 topology_thread_cpumask(cpu));
7235 if (cpu == cpumask_first(d->this_sibling_map))
7236 init_sched_build_groups(d->this_sibling_map, cpu_map,
7237 &cpu_to_cpu_group,
7238 d->send_covered, d->tmpmask);
7239 break;
1da177e4 7240#endif
1e9f28fa 7241#ifdef CONFIG_SCHED_MC
a2af04cd
AH
7242 case SD_LV_MC: /* set up multi-core groups */
7243 cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
7244 if (cpu == cpumask_first(d->this_core_map))
7245 init_sched_build_groups(d->this_core_map, cpu_map,
7246 &cpu_to_core_group,
7247 d->send_covered, d->tmpmask);
7248 break;
01a08546
HC
7249#endif
7250#ifdef CONFIG_SCHED_BOOK
7251 case SD_LV_BOOK: /* set up book groups */
7252 cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu));
7253 if (cpu == cpumask_first(d->this_book_map))
7254 init_sched_build_groups(d->this_book_map, cpu_map,
7255 &cpu_to_book_group,
7256 d->send_covered, d->tmpmask);
7257 break;
1e9f28fa 7258#endif
86548096
AH
7259 case SD_LV_CPU: /* set up physical groups */
7260 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
7261 if (!cpumask_empty(d->nodemask))
7262 init_sched_build_groups(d->nodemask, cpu_map,
7263 &cpu_to_phys_group,
7264 d->send_covered, d->tmpmask);
7265 break;
1da177e4 7266#ifdef CONFIG_NUMA
de616e36
AH
7267 case SD_LV_ALLNODES:
7268 init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
7269 d->send_covered, d->tmpmask);
7270 break;
7271#endif
0e8e85c9
AH
7272 default:
7273 break;
7c16ec58 7274 }
0e8e85c9 7275}
9c1cfda2 7276
2109b99e
AH
7277/*
7278 * Build sched domains for a given set of cpus and attach the sched domains
7279 * to the individual cpus
7280 */
7281static int __build_sched_domains(const struct cpumask *cpu_map,
7282 struct sched_domain_attr *attr)
7283{
7284 enum s_alloc alloc_state = sa_none;
7285 struct s_data d;
294b0c96 7286 struct sched_domain *sd;
2109b99e 7287 int i;
7c16ec58 7288#ifdef CONFIG_NUMA
2109b99e 7289 d.sd_allnodes = 0;
7c16ec58 7290#endif
9c1cfda2 7291
2109b99e
AH
7292 alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
7293 if (alloc_state != sa_rootdomain)
7294 goto error;
7295 alloc_state = sa_sched_groups;
9c1cfda2 7296
1da177e4 7297 /*
1a20ff27 7298 * Set up domains for cpus specified by the cpu_map.
1da177e4 7299 */
abcd083a 7300 for_each_cpu(i, cpu_map) {
49a02c51
AH
7301 cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
7302 cpu_map);
9761eea8 7303
7f4588f3 7304 sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
87cce662 7305 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
01a08546 7306 sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
410c4081 7307 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
d8173535 7308 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
1da177e4 7309 }
9c1cfda2 7310
abcd083a 7311 for_each_cpu(i, cpu_map) {
0e8e85c9 7312 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
01a08546 7313 build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
a2af04cd 7314 build_sched_groups(&d, SD_LV_MC, cpu_map, i);
1da177e4 7315 }
9c1cfda2 7316
1da177e4 7317 /* Set up physical groups */
86548096
AH
7318 for (i = 0; i < nr_node_ids; i++)
7319 build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
9c1cfda2 7320
1da177e4
LT
7321#ifdef CONFIG_NUMA
7322 /* Set up node groups */
de616e36
AH
7323 if (d.sd_allnodes)
7324 build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
9c1cfda2 7325
0601a88d
AH
7326 for (i = 0; i < nr_node_ids; i++)
7327 if (build_numa_sched_groups(&d, cpu_map, i))
51888ca2 7328 goto error;
1da177e4
LT
7329#endif
7330
7331 /* Calculate CPU power for physical packages and nodes */
5c45bf27 7332#ifdef CONFIG_SCHED_SMT
abcd083a 7333 for_each_cpu(i, cpu_map) {
294b0c96 7334 sd = &per_cpu(cpu_domains, i).sd;
89c4710e 7335 init_sched_groups_power(i, sd);
5c45bf27 7336 }
1da177e4 7337#endif
1e9f28fa 7338#ifdef CONFIG_SCHED_MC
abcd083a 7339 for_each_cpu(i, cpu_map) {
294b0c96 7340 sd = &per_cpu(core_domains, i).sd;
89c4710e 7341 init_sched_groups_power(i, sd);
5c45bf27
SS
7342 }
7343#endif
01a08546
HC
7344#ifdef CONFIG_SCHED_BOOK
7345 for_each_cpu(i, cpu_map) {
7346 sd = &per_cpu(book_domains, i).sd;
7347 init_sched_groups_power(i, sd);
7348 }
7349#endif
1e9f28fa 7350
abcd083a 7351 for_each_cpu(i, cpu_map) {
294b0c96 7352 sd = &per_cpu(phys_domains, i).sd;
89c4710e 7353 init_sched_groups_power(i, sd);
1da177e4
LT
7354 }
7355
9c1cfda2 7356#ifdef CONFIG_NUMA
076ac2af 7357 for (i = 0; i < nr_node_ids; i++)
49a02c51 7358 init_numa_sched_groups_power(d.sched_group_nodes[i]);
9c1cfda2 7359
49a02c51 7360 if (d.sd_allnodes) {
6711cab4 7361 struct sched_group *sg;
f712c0c7 7362
96f874e2 7363 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
49a02c51 7364 d.tmpmask);
f712c0c7
SS
7365 init_numa_sched_groups_power(sg);
7366 }
9c1cfda2
JH
7367#endif
7368
1da177e4 7369 /* Attach the domains */
abcd083a 7370 for_each_cpu(i, cpu_map) {
1da177e4 7371#ifdef CONFIG_SCHED_SMT
6c99e9ad 7372 sd = &per_cpu(cpu_domains, i).sd;
1e9f28fa 7373#elif defined(CONFIG_SCHED_MC)
6c99e9ad 7374 sd = &per_cpu(core_domains, i).sd;
01a08546
HC
7375#elif defined(CONFIG_SCHED_BOOK)
7376 sd = &per_cpu(book_domains, i).sd;
1da177e4 7377#else
6c99e9ad 7378 sd = &per_cpu(phys_domains, i).sd;
1da177e4 7379#endif
49a02c51 7380 cpu_attach_domain(sd, d.rd, i);
1da177e4 7381 }
51888ca2 7382
2109b99e
AH
7383 d.sched_group_nodes = NULL; /* don't free this we still need it */
7384 __free_domain_allocs(&d, sa_tmpmask, cpu_map);
7385 return 0;
51888ca2 7386
51888ca2 7387error:
2109b99e
AH
7388 __free_domain_allocs(&d, alloc_state, cpu_map);
7389 return -ENOMEM;
1da177e4 7390}
029190c5 7391
96f874e2 7392static int build_sched_domains(const struct cpumask *cpu_map)
1d3504fc
HS
7393{
7394 return __build_sched_domains(cpu_map, NULL);
7395}
7396
acc3f5d7 7397static cpumask_var_t *doms_cur; /* current sched domains */
029190c5 7398static int ndoms_cur; /* number of sched domains in 'doms_cur' */
4285f594
IM
7399static struct sched_domain_attr *dattr_cur;
7400 /* attribues of custom domains in 'doms_cur' */
029190c5
PJ
7401
7402/*
7403 * Special case: If a kmalloc of a doms_cur partition (array of
4212823f
RR
7404 * cpumask) fails, then fallback to a single sched domain,
7405 * as determined by the single cpumask fallback_doms.
029190c5 7406 */
4212823f 7407static cpumask_var_t fallback_doms;
029190c5 7408
ee79d1bd
HC
7409/*
7410 * arch_update_cpu_topology lets virtualized architectures update the
7411 * cpu core maps. It is supposed to return 1 if the topology changed
7412 * or 0 if it stayed the same.
7413 */
7414int __attribute__((weak)) arch_update_cpu_topology(void)
22e52b07 7415{
ee79d1bd 7416 return 0;
22e52b07
HC
7417}
7418
acc3f5d7
RR
7419cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
7420{
7421 int i;
7422 cpumask_var_t *doms;
7423
7424 doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
7425 if (!doms)
7426 return NULL;
7427 for (i = 0; i < ndoms; i++) {
7428 if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
7429 free_sched_domains(doms, i);
7430 return NULL;
7431 }
7432 }
7433 return doms;
7434}
7435
7436void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
7437{
7438 unsigned int i;
7439 for (i = 0; i < ndoms; i++)
7440 free_cpumask_var(doms[i]);
7441 kfree(doms);
7442}
7443
1a20ff27 7444/*
41a2d6cf 7445 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
029190c5
PJ
7446 * For now this just excludes isolated cpus, but could be used to
7447 * exclude other special cases in the future.
1a20ff27 7448 */
96f874e2 7449static int arch_init_sched_domains(const struct cpumask *cpu_map)
1a20ff27 7450{
7378547f
MM
7451 int err;
7452
22e52b07 7453 arch_update_cpu_topology();
029190c5 7454 ndoms_cur = 1;
acc3f5d7 7455 doms_cur = alloc_sched_domains(ndoms_cur);
029190c5 7456 if (!doms_cur)
acc3f5d7
RR
7457 doms_cur = &fallback_doms;
7458 cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
1d3504fc 7459 dattr_cur = NULL;
acc3f5d7 7460 err = build_sched_domains(doms_cur[0]);
6382bc90 7461 register_sched_domain_sysctl();
7378547f
MM
7462
7463 return err;
1a20ff27
DG
7464}
7465
96f874e2
RR
7466static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7467 struct cpumask *tmpmask)
1da177e4 7468{
7c16ec58 7469 free_sched_groups(cpu_map, tmpmask);
9c1cfda2 7470}
1da177e4 7471
1a20ff27
DG
7472/*
7473 * Detach sched domains from a group of cpus specified in cpu_map
7474 * These cpus will now be attached to the NULL domain
7475 */
96f874e2 7476static void detach_destroy_domains(const struct cpumask *cpu_map)
1a20ff27 7477{
96f874e2
RR
7478 /* Save because hotplug lock held. */
7479 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
1a20ff27
DG
7480 int i;
7481
abcd083a 7482 for_each_cpu(i, cpu_map)
57d885fe 7483 cpu_attach_domain(NULL, &def_root_domain, i);
1a20ff27 7484 synchronize_sched();
96f874e2 7485 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
1a20ff27
DG
7486}
7487
1d3504fc
HS
7488/* handle null as "default" */
7489static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7490 struct sched_domain_attr *new, int idx_new)
7491{
7492 struct sched_domain_attr tmp;
7493
7494 /* fast path */
7495 if (!new && !cur)
7496 return 1;
7497
7498 tmp = SD_ATTR_INIT;
7499 return !memcmp(cur ? (cur + idx_cur) : &tmp,
7500 new ? (new + idx_new) : &tmp,
7501 sizeof(struct sched_domain_attr));
7502}
7503
029190c5
PJ
7504/*
7505 * Partition sched domains as specified by the 'ndoms_new'
41a2d6cf 7506 * cpumasks in the array doms_new[] of cpumasks. This compares
029190c5
PJ
7507 * doms_new[] to the current sched domain partitioning, doms_cur[].
7508 * It destroys each deleted domain and builds each new domain.
7509 *
acc3f5d7 7510 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
41a2d6cf
IM
7511 * The masks don't intersect (don't overlap.) We should setup one
7512 * sched domain for each mask. CPUs not in any of the cpumasks will
7513 * not be load balanced. If the same cpumask appears both in the
029190c5
PJ
7514 * current 'doms_cur' domains and in the new 'doms_new', we can leave
7515 * it as it is.
7516 *
acc3f5d7
RR
7517 * The passed in 'doms_new' should be allocated using
7518 * alloc_sched_domains. This routine takes ownership of it and will
7519 * free_sched_domains it when done with it. If the caller failed the
7520 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
7521 * and partition_sched_domains() will fallback to the single partition
7522 * 'fallback_doms', it also forces the domains to be rebuilt.
029190c5 7523 *
96f874e2 7524 * If doms_new == NULL it will be replaced with cpu_online_mask.
700018e0
LZ
7525 * ndoms_new == 0 is a special case for destroying existing domains,
7526 * and it will not create the default domain.
dfb512ec 7527 *
029190c5
PJ
7528 * Call with hotplug lock held
7529 */
acc3f5d7 7530void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
1d3504fc 7531 struct sched_domain_attr *dattr_new)
029190c5 7532{
dfb512ec 7533 int i, j, n;
d65bd5ec 7534 int new_topology;
029190c5 7535
712555ee 7536 mutex_lock(&sched_domains_mutex);
a1835615 7537
7378547f
MM
7538 /* always unregister in case we don't destroy any domains */
7539 unregister_sched_domain_sysctl();
7540
d65bd5ec
HC
7541 /* Let architecture update cpu core mappings. */
7542 new_topology = arch_update_cpu_topology();
7543
dfb512ec 7544 n = doms_new ? ndoms_new : 0;
029190c5
PJ
7545
7546 /* Destroy deleted domains */
7547 for (i = 0; i < ndoms_cur; i++) {
d65bd5ec 7548 for (j = 0; j < n && !new_topology; j++) {
acc3f5d7 7549 if (cpumask_equal(doms_cur[i], doms_new[j])
1d3504fc 7550 && dattrs_equal(dattr_cur, i, dattr_new, j))
029190c5
PJ
7551 goto match1;
7552 }
7553 /* no match - a current sched domain not in new doms_new[] */
acc3f5d7 7554 detach_destroy_domains(doms_cur[i]);
029190c5
PJ
7555match1:
7556 ;
7557 }
7558
e761b772
MK
7559 if (doms_new == NULL) {
7560 ndoms_cur = 0;
acc3f5d7 7561 doms_new = &fallback_doms;
6ad4c188 7562 cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
faa2f98f 7563 WARN_ON_ONCE(dattr_new);
e761b772
MK
7564 }
7565
029190c5
PJ
7566 /* Build new domains */
7567 for (i = 0; i < ndoms_new; i++) {
d65bd5ec 7568 for (j = 0; j < ndoms_cur && !new_topology; j++) {
acc3f5d7 7569 if (cpumask_equal(doms_new[i], doms_cur[j])
1d3504fc 7570 && dattrs_equal(dattr_new, i, dattr_cur, j))
029190c5
PJ
7571 goto match2;
7572 }
7573 /* no match - add a new doms_new */
acc3f5d7 7574 __build_sched_domains(doms_new[i],
1d3504fc 7575 dattr_new ? dattr_new + i : NULL);
029190c5
PJ
7576match2:
7577 ;
7578 }
7579
7580 /* Remember the new sched domains */
acc3f5d7
RR
7581 if (doms_cur != &fallback_doms)
7582 free_sched_domains(doms_cur, ndoms_cur);
1d3504fc 7583 kfree(dattr_cur); /* kfree(NULL) is safe */
029190c5 7584 doms_cur = doms_new;
1d3504fc 7585 dattr_cur = dattr_new;
029190c5 7586 ndoms_cur = ndoms_new;
7378547f
MM
7587
7588 register_sched_domain_sysctl();
a1835615 7589
712555ee 7590 mutex_unlock(&sched_domains_mutex);
029190c5
PJ
7591}
7592
5c45bf27 7593#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
c70f22d2 7594static void arch_reinit_sched_domains(void)
5c45bf27 7595{
95402b38 7596 get_online_cpus();
dfb512ec
MK
7597
7598 /* Destroy domains first to force the rebuild */
7599 partition_sched_domains(0, NULL, NULL);
7600
e761b772 7601 rebuild_sched_domains();
95402b38 7602 put_online_cpus();
5c45bf27
SS
7603}
7604
7605static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7606{
afb8a9b7 7607 unsigned int level = 0;
5c45bf27 7608
afb8a9b7
GS
7609 if (sscanf(buf, "%u", &level) != 1)
7610 return -EINVAL;
7611
7612 /*
7613 * level is always be positive so don't check for
7614 * level < POWERSAVINGS_BALANCE_NONE which is 0
7615 * What happens on 0 or 1 byte write,
7616 * need to check for count as well?
7617 */
7618
7619 if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS)
5c45bf27
SS
7620 return -EINVAL;
7621
7622 if (smt)
afb8a9b7 7623 sched_smt_power_savings = level;
5c45bf27 7624 else
afb8a9b7 7625 sched_mc_power_savings = level;
5c45bf27 7626
c70f22d2 7627 arch_reinit_sched_domains();
5c45bf27 7628
c70f22d2 7629 return count;
5c45bf27
SS
7630}
7631
5c45bf27 7632#ifdef CONFIG_SCHED_MC
f718cd4a 7633static ssize_t sched_mc_power_savings_show(struct sysdev_class *class,
c9be0a36 7634 struct sysdev_class_attribute *attr,
f718cd4a 7635 char *page)
5c45bf27
SS
7636{
7637 return sprintf(page, "%u\n", sched_mc_power_savings);
7638}
f718cd4a 7639static ssize_t sched_mc_power_savings_store(struct sysdev_class *class,
c9be0a36 7640 struct sysdev_class_attribute *attr,
48f24c4d 7641 const char *buf, size_t count)
5c45bf27
SS
7642{
7643 return sched_power_savings_store(buf, count, 0);
7644}
f718cd4a
AK
7645static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644,
7646 sched_mc_power_savings_show,
7647 sched_mc_power_savings_store);
5c45bf27
SS
7648#endif
7649
7650#ifdef CONFIG_SCHED_SMT
f718cd4a 7651static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev,
c9be0a36 7652 struct sysdev_class_attribute *attr,
f718cd4a 7653 char *page)
5c45bf27
SS
7654{
7655 return sprintf(page, "%u\n", sched_smt_power_savings);
7656}
f718cd4a 7657static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev,
c9be0a36 7658 struct sysdev_class_attribute *attr,
48f24c4d 7659 const char *buf, size_t count)
5c45bf27
SS
7660{
7661 return sched_power_savings_store(buf, count, 1);
7662}
f718cd4a
AK
7663static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644,
7664 sched_smt_power_savings_show,
6707de00
AB
7665 sched_smt_power_savings_store);
7666#endif
7667
39aac648 7668int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
6707de00
AB
7669{
7670 int err = 0;
7671
7672#ifdef CONFIG_SCHED_SMT
7673 if (smt_capable())
7674 err = sysfs_create_file(&cls->kset.kobj,
7675 &attr_sched_smt_power_savings.attr);
7676#endif
7677#ifdef CONFIG_SCHED_MC
7678 if (!err && mc_capable())
7679 err = sysfs_create_file(&cls->kset.kobj,
7680 &attr_sched_mc_power_savings.attr);
7681#endif
7682 return err;
7683}
6d6bc0ad 7684#endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
5c45bf27 7685
1da177e4 7686/*
3a101d05
TH
7687 * Update cpusets according to cpu_active mask. If cpusets are
7688 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
7689 * around partition_sched_domains().
1da177e4 7690 */
0b2e918a
TH
7691static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
7692 void *hcpu)
e761b772 7693{
3a101d05 7694 switch (action & ~CPU_TASKS_FROZEN) {
e761b772 7695 case CPU_ONLINE:
6ad4c188 7696 case CPU_DOWN_FAILED:
3a101d05 7697 cpuset_update_active_cpus();
e761b772 7698 return NOTIFY_OK;
3a101d05
TH
7699 default:
7700 return NOTIFY_DONE;
7701 }
7702}
e761b772 7703
0b2e918a
TH
7704static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
7705 void *hcpu)
3a101d05
TH
7706{
7707 switch (action & ~CPU_TASKS_FROZEN) {
7708 case CPU_DOWN_PREPARE:
7709 cpuset_update_active_cpus();
7710 return NOTIFY_OK;
e761b772
MK
7711 default:
7712 return NOTIFY_DONE;
7713 }
7714}
e761b772
MK
7715
7716static int update_runtime(struct notifier_block *nfb,
7717 unsigned long action, void *hcpu)
1da177e4 7718{
7def2be1
PZ
7719 int cpu = (int)(long)hcpu;
7720
1da177e4 7721 switch (action) {
1da177e4 7722 case CPU_DOWN_PREPARE:
8bb78442 7723 case CPU_DOWN_PREPARE_FROZEN:
7def2be1 7724 disable_runtime(cpu_rq(cpu));
1da177e4
LT
7725 return NOTIFY_OK;
7726
1da177e4 7727 case CPU_DOWN_FAILED:
8bb78442 7728 case CPU_DOWN_FAILED_FROZEN:
1da177e4 7729 case CPU_ONLINE:
8bb78442 7730 case CPU_ONLINE_FROZEN:
7def2be1 7731 enable_runtime(cpu_rq(cpu));
e761b772
MK
7732 return NOTIFY_OK;
7733
1da177e4
LT
7734 default:
7735 return NOTIFY_DONE;
7736 }
1da177e4 7737}
1da177e4
LT
7738
7739void __init sched_init_smp(void)
7740{
dcc30a35
RR
7741 cpumask_var_t non_isolated_cpus;
7742
7743 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
cb5fd13f 7744 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
5c1e1767 7745
434d53b0
MT
7746#if defined(CONFIG_NUMA)
7747 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
7748 GFP_KERNEL);
7749 BUG_ON(sched_group_nodes_bycpu == NULL);
7750#endif
95402b38 7751 get_online_cpus();
712555ee 7752 mutex_lock(&sched_domains_mutex);
6ad4c188 7753 arch_init_sched_domains(cpu_active_mask);
dcc30a35
RR
7754 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
7755 if (cpumask_empty(non_isolated_cpus))
7756 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
712555ee 7757 mutex_unlock(&sched_domains_mutex);
95402b38 7758 put_online_cpus();
e761b772 7759
3a101d05
TH
7760 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
7761 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
e761b772
MK
7762
7763 /* RT runtime code needs to handle some hotplug events */
7764 hotcpu_notifier(update_runtime, 0);
7765
b328ca18 7766 init_hrtick();
5c1e1767
NP
7767
7768 /* Move init over to a non-isolated CPU */
dcc30a35 7769 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
5c1e1767 7770 BUG();
19978ca6 7771 sched_init_granularity();
dcc30a35 7772 free_cpumask_var(non_isolated_cpus);
4212823f 7773
0e3900e6 7774 init_sched_rt_class();
1da177e4
LT
7775}
7776#else
7777void __init sched_init_smp(void)
7778{
19978ca6 7779 sched_init_granularity();
1da177e4
LT
7780}
7781#endif /* CONFIG_SMP */
7782
cd1bb94b
AB
7783const_debug unsigned int sysctl_timer_migration = 1;
7784
1da177e4
LT
7785int in_sched_functions(unsigned long addr)
7786{
1da177e4
LT
7787 return in_lock_functions(addr) ||
7788 (addr >= (unsigned long)__sched_text_start
7789 && addr < (unsigned long)__sched_text_end);
7790}
7791
a9957449 7792static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
dd41f596
IM
7793{
7794 cfs_rq->tasks_timeline = RB_ROOT;
4a55bd5e 7795 INIT_LIST_HEAD(&cfs_rq->tasks);
dd41f596
IM
7796#ifdef CONFIG_FAIR_GROUP_SCHED
7797 cfs_rq->rq = rq;
7798#endif
67e9fb2a 7799 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
dd41f596
IM
7800}
7801
fa85ae24
PZ
7802static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
7803{
7804 struct rt_prio_array *array;
7805 int i;
7806
7807 array = &rt_rq->active;
7808 for (i = 0; i < MAX_RT_PRIO; i++) {
7809 INIT_LIST_HEAD(array->queue + i);
7810 __clear_bit(i, array->bitmap);
7811 }
7812 /* delimiter for bitsearch: */
7813 __set_bit(MAX_RT_PRIO, array->bitmap);
7814
052f1dc7 7815#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
e864c499 7816 rt_rq->highest_prio.curr = MAX_RT_PRIO;
398a153b 7817#ifdef CONFIG_SMP
e864c499 7818 rt_rq->highest_prio.next = MAX_RT_PRIO;
48d5e258 7819#endif
48d5e258 7820#endif
fa85ae24
PZ
7821#ifdef CONFIG_SMP
7822 rt_rq->rt_nr_migratory = 0;
fa85ae24 7823 rt_rq->overloaded = 0;
05fa785c 7824 plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock);
fa85ae24
PZ
7825#endif
7826
7827 rt_rq->rt_time = 0;
7828 rt_rq->rt_throttled = 0;
ac086bc2 7829 rt_rq->rt_runtime = 0;
0986b11b 7830 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
6f505b16 7831
052f1dc7 7832#ifdef CONFIG_RT_GROUP_SCHED
23b0fdfc 7833 rt_rq->rt_nr_boosted = 0;
6f505b16
PZ
7834 rt_rq->rq = rq;
7835#endif
fa85ae24
PZ
7836}
7837
6f505b16 7838#ifdef CONFIG_FAIR_GROUP_SCHED
ec7dc8ac 7839static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
3d4b47b4 7840 struct sched_entity *se, int cpu,
ec7dc8ac 7841 struct sched_entity *parent)
6f505b16 7842{
ec7dc8ac 7843 struct rq *rq = cpu_rq(cpu);
6f505b16
PZ
7844 tg->cfs_rq[cpu] = cfs_rq;
7845 init_cfs_rq(cfs_rq, rq);
7846 cfs_rq->tg = tg;
6f505b16
PZ
7847
7848 tg->se[cpu] = se;
07e06b01 7849 /* se could be NULL for root_task_group */
354d60c2
DG
7850 if (!se)
7851 return;
7852
ec7dc8ac
DG
7853 if (!parent)
7854 se->cfs_rq = &rq->cfs;
7855 else
7856 se->cfs_rq = parent->my_q;
7857
6f505b16 7858 se->my_q = cfs_rq;
9437178f 7859 update_load_set(&se->load, 0);
ec7dc8ac 7860 se->parent = parent;
6f505b16 7861}
052f1dc7 7862#endif
6f505b16 7863
052f1dc7 7864#ifdef CONFIG_RT_GROUP_SCHED
ec7dc8ac 7865static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
3d4b47b4 7866 struct sched_rt_entity *rt_se, int cpu,
ec7dc8ac 7867 struct sched_rt_entity *parent)
6f505b16 7868{
ec7dc8ac
DG
7869 struct rq *rq = cpu_rq(cpu);
7870
6f505b16
PZ
7871 tg->rt_rq[cpu] = rt_rq;
7872 init_rt_rq(rt_rq, rq);
7873 rt_rq->tg = tg;
ac086bc2 7874 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
6f505b16
PZ
7875
7876 tg->rt_se[cpu] = rt_se;
354d60c2
DG
7877 if (!rt_se)
7878 return;
7879
ec7dc8ac
DG
7880 if (!parent)
7881 rt_se->rt_rq = &rq->rt;
7882 else
7883 rt_se->rt_rq = parent->my_q;
7884
6f505b16 7885 rt_se->my_q = rt_rq;
ec7dc8ac 7886 rt_se->parent = parent;
6f505b16
PZ
7887 INIT_LIST_HEAD(&rt_se->run_list);
7888}
7889#endif
7890
1da177e4
LT
7891void __init sched_init(void)
7892{
dd41f596 7893 int i, j;
434d53b0
MT
7894 unsigned long alloc_size = 0, ptr;
7895
7896#ifdef CONFIG_FAIR_GROUP_SCHED
7897 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7898#endif
7899#ifdef CONFIG_RT_GROUP_SCHED
7900 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
eff766a6 7901#endif
df7c8e84 7902#ifdef CONFIG_CPUMASK_OFFSTACK
8c083f08 7903 alloc_size += num_possible_cpus() * cpumask_size();
434d53b0 7904#endif
434d53b0 7905 if (alloc_size) {
36b7b6d4 7906 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
434d53b0
MT
7907
7908#ifdef CONFIG_FAIR_GROUP_SCHED
07e06b01 7909 root_task_group.se = (struct sched_entity **)ptr;
434d53b0
MT
7910 ptr += nr_cpu_ids * sizeof(void **);
7911
07e06b01 7912 root_task_group.cfs_rq = (struct cfs_rq **)ptr;
434d53b0 7913 ptr += nr_cpu_ids * sizeof(void **);
eff766a6 7914
6d6bc0ad 7915#endif /* CONFIG_FAIR_GROUP_SCHED */
434d53b0 7916#ifdef CONFIG_RT_GROUP_SCHED
07e06b01 7917 root_task_group.rt_se = (struct sched_rt_entity **)ptr;
434d53b0
MT
7918 ptr += nr_cpu_ids * sizeof(void **);
7919
07e06b01 7920 root_task_group.rt_rq = (struct rt_rq **)ptr;
eff766a6
PZ
7921 ptr += nr_cpu_ids * sizeof(void **);
7922
6d6bc0ad 7923#endif /* CONFIG_RT_GROUP_SCHED */
df7c8e84
RR
7924#ifdef CONFIG_CPUMASK_OFFSTACK
7925 for_each_possible_cpu(i) {
7926 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
7927 ptr += cpumask_size();
7928 }
7929#endif /* CONFIG_CPUMASK_OFFSTACK */
434d53b0 7930 }
dd41f596 7931
57d885fe
GH
7932#ifdef CONFIG_SMP
7933 init_defrootdomain();
7934#endif
7935
d0b27fa7
PZ
7936 init_rt_bandwidth(&def_rt_bandwidth,
7937 global_rt_period(), global_rt_runtime());
7938
7939#ifdef CONFIG_RT_GROUP_SCHED
07e06b01 7940 init_rt_bandwidth(&root_task_group.rt_bandwidth,
d0b27fa7 7941 global_rt_period(), global_rt_runtime());
6d6bc0ad 7942#endif /* CONFIG_RT_GROUP_SCHED */
d0b27fa7 7943
7c941438 7944#ifdef CONFIG_CGROUP_SCHED
07e06b01
YZ
7945 list_add(&root_task_group.list, &task_groups);
7946 INIT_LIST_HEAD(&root_task_group.children);
5091faa4 7947 autogroup_init(&init_task);
7c941438 7948#endif /* CONFIG_CGROUP_SCHED */
6f505b16 7949
0a945022 7950 for_each_possible_cpu(i) {
70b97a7f 7951 struct rq *rq;
1da177e4
LT
7952
7953 rq = cpu_rq(i);
05fa785c 7954 raw_spin_lock_init(&rq->lock);
7897986b 7955 rq->nr_running = 0;
dce48a84
TG
7956 rq->calc_load_active = 0;
7957 rq->calc_load_update = jiffies + LOAD_FREQ;
dd41f596 7958 init_cfs_rq(&rq->cfs, rq);
6f505b16 7959 init_rt_rq(&rq->rt, rq);
dd41f596 7960#ifdef CONFIG_FAIR_GROUP_SCHED
07e06b01 7961 root_task_group.shares = root_task_group_load;
6f505b16 7962 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
354d60c2 7963 /*
07e06b01 7964 * How much cpu bandwidth does root_task_group get?
354d60c2
DG
7965 *
7966 * In case of task-groups formed thr' the cgroup filesystem, it
7967 * gets 100% of the cpu resources in the system. This overall
7968 * system cpu resource is divided among the tasks of
07e06b01 7969 * root_task_group and its child task-groups in a fair manner,
354d60c2
DG
7970 * based on each entity's (task or task-group's) weight
7971 * (se->load.weight).
7972 *
07e06b01 7973 * In other words, if root_task_group has 10 tasks of weight
354d60c2
DG
7974 * 1024) and two child groups A0 and A1 (of weight 1024 each),
7975 * then A0's share of the cpu resource is:
7976 *
0d905bca 7977 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
354d60c2 7978 *
07e06b01
YZ
7979 * We achieve this by letting root_task_group's tasks sit
7980 * directly in rq->cfs (i.e root_task_group->se[] = NULL).
354d60c2 7981 */
07e06b01 7982 init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
354d60c2
DG
7983#endif /* CONFIG_FAIR_GROUP_SCHED */
7984
7985 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
052f1dc7 7986#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 7987 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
07e06b01 7988 init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
dd41f596 7989#endif
1da177e4 7990
dd41f596
IM
7991 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
7992 rq->cpu_load[j] = 0;
fdf3e95d
VP
7993
7994 rq->last_load_update_tick = jiffies;
7995
1da177e4 7996#ifdef CONFIG_SMP
41c7ce9a 7997 rq->sd = NULL;
57d885fe 7998 rq->rd = NULL;
e51fd5e2 7999 rq->cpu_power = SCHED_LOAD_SCALE;
3f029d3c 8000 rq->post_schedule = 0;
1da177e4 8001 rq->active_balance = 0;
dd41f596 8002 rq->next_balance = jiffies;
1da177e4 8003 rq->push_cpu = 0;
0a2966b4 8004 rq->cpu = i;
1f11eb6a 8005 rq->online = 0;
eae0c9df
MG
8006 rq->idle_stamp = 0;
8007 rq->avg_idle = 2*sysctl_sched_migration_cost;
dc938520 8008 rq_attach_root(rq, &def_root_domain);
83cd4fe2
VP
8009#ifdef CONFIG_NO_HZ
8010 rq->nohz_balance_kick = 0;
8011 init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i));
8012#endif
1da177e4 8013#endif
8f4d37ec 8014 init_rq_hrtick(rq);
1da177e4 8015 atomic_set(&rq->nr_iowait, 0);
1da177e4
LT
8016 }
8017
2dd73a4f 8018 set_load_weight(&init_task);
b50f60ce 8019
e107be36
AK
8020#ifdef CONFIG_PREEMPT_NOTIFIERS
8021 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
8022#endif
8023
c9819f45 8024#ifdef CONFIG_SMP
962cf36c 8025 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
c9819f45
CL
8026#endif
8027
b50f60ce 8028#ifdef CONFIG_RT_MUTEXES
1d615482 8029 plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock);
b50f60ce
HC
8030#endif
8031
1da177e4
LT
8032 /*
8033 * The boot idle thread does lazy MMU switching as well:
8034 */
8035 atomic_inc(&init_mm.mm_count);
8036 enter_lazy_tlb(&init_mm, current);
8037
8038 /*
8039 * Make us the idle thread. Technically, schedule() should not be
8040 * called from this thread, however somewhere below it might be,
8041 * but because we are the idle thread, we just pick up running again
8042 * when this runqueue becomes "idle".
8043 */
8044 init_idle(current, smp_processor_id());
dce48a84
TG
8045
8046 calc_load_update = jiffies + LOAD_FREQ;
8047
dd41f596
IM
8048 /*
8049 * During early bootup we pretend to be a normal task:
8050 */
8051 current->sched_class = &fair_sched_class;
6892b75e 8052
6a7b3dc3 8053 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
49557e62 8054 zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT);
bf4d83f6 8055#ifdef CONFIG_SMP
7d1e6a9b 8056#ifdef CONFIG_NO_HZ
83cd4fe2
VP
8057 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
8058 alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT);
8059 atomic_set(&nohz.load_balancer, nr_cpu_ids);
8060 atomic_set(&nohz.first_pick_cpu, nr_cpu_ids);
8061 atomic_set(&nohz.second_pick_cpu, nr_cpu_ids);
7d1e6a9b 8062#endif
bdddd296
RR
8063 /* May be allocated at isolcpus cmdline parse time */
8064 if (cpu_isolated_map == NULL)
8065 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
bf4d83f6 8066#endif /* SMP */
6a7b3dc3 8067
6892b75e 8068 scheduler_running = 1;
1da177e4
LT
8069}
8070
8071#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
e4aafea2
FW
8072static inline int preempt_count_equals(int preempt_offset)
8073{
234da7bc 8074 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
e4aafea2
FW
8075
8076 return (nested == PREEMPT_INATOMIC_BASE + preempt_offset);
8077}
8078
d894837f 8079void __might_sleep(const char *file, int line, int preempt_offset)
1da177e4 8080{
48f24c4d 8081#ifdef in_atomic
1da177e4
LT
8082 static unsigned long prev_jiffy; /* ratelimiting */
8083
e4aafea2
FW
8084 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
8085 system_state != SYSTEM_RUNNING || oops_in_progress)
aef745fc
IM
8086 return;
8087 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
8088 return;
8089 prev_jiffy = jiffies;
8090
3df0fc5b
PZ
8091 printk(KERN_ERR
8092 "BUG: sleeping function called from invalid context at %s:%d\n",
8093 file, line);
8094 printk(KERN_ERR
8095 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
8096 in_atomic(), irqs_disabled(),
8097 current->pid, current->comm);
aef745fc
IM
8098
8099 debug_show_held_locks(current);
8100 if (irqs_disabled())
8101 print_irqtrace_events(current);
8102 dump_stack();
1da177e4
LT
8103#endif
8104}
8105EXPORT_SYMBOL(__might_sleep);
8106#endif
8107
8108#ifdef CONFIG_MAGIC_SYSRQ
3a5e4dc1
AK
8109static void normalize_task(struct rq *rq, struct task_struct *p)
8110{
8111 int on_rq;
3e51f33f 8112
3a5e4dc1
AK
8113 on_rq = p->se.on_rq;
8114 if (on_rq)
8115 deactivate_task(rq, p, 0);
8116 __setscheduler(rq, p, SCHED_NORMAL, 0);
8117 if (on_rq) {
8118 activate_task(rq, p, 0);
8119 resched_task(rq->curr);
8120 }
8121}
8122
1da177e4
LT
8123void normalize_rt_tasks(void)
8124{
a0f98a1c 8125 struct task_struct *g, *p;
1da177e4 8126 unsigned long flags;
70b97a7f 8127 struct rq *rq;
1da177e4 8128
4cf5d77a 8129 read_lock_irqsave(&tasklist_lock, flags);
a0f98a1c 8130 do_each_thread(g, p) {
178be793
IM
8131 /*
8132 * Only normalize user tasks:
8133 */
8134 if (!p->mm)
8135 continue;
8136
6cfb0d5d 8137 p->se.exec_start = 0;
6cfb0d5d 8138#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
8139 p->se.statistics.wait_start = 0;
8140 p->se.statistics.sleep_start = 0;
8141 p->se.statistics.block_start = 0;
6cfb0d5d 8142#endif
dd41f596
IM
8143
8144 if (!rt_task(p)) {
8145 /*
8146 * Renice negative nice level userspace
8147 * tasks back to 0:
8148 */
8149 if (TASK_NICE(p) < 0 && p->mm)
8150 set_user_nice(p, 0);
1da177e4 8151 continue;
dd41f596 8152 }
1da177e4 8153
1d615482 8154 raw_spin_lock(&p->pi_lock);
b29739f9 8155 rq = __task_rq_lock(p);
1da177e4 8156
178be793 8157 normalize_task(rq, p);
3a5e4dc1 8158
b29739f9 8159 __task_rq_unlock(rq);
1d615482 8160 raw_spin_unlock(&p->pi_lock);
a0f98a1c
IM
8161 } while_each_thread(g, p);
8162
4cf5d77a 8163 read_unlock_irqrestore(&tasklist_lock, flags);
1da177e4
LT
8164}
8165
8166#endif /* CONFIG_MAGIC_SYSRQ */
1df5c10a 8167
67fc4e0c 8168#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
1df5c10a 8169/*
67fc4e0c 8170 * These functions are only useful for the IA64 MCA handling, or kdb.
1df5c10a
LT
8171 *
8172 * They can only be called when the whole system has been
8173 * stopped - every CPU needs to be quiescent, and no scheduling
8174 * activity can take place. Using them for anything else would
8175 * be a serious bug, and as a result, they aren't even visible
8176 * under any other configuration.
8177 */
8178
8179/**
8180 * curr_task - return the current task for a given cpu.
8181 * @cpu: the processor in question.
8182 *
8183 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8184 */
36c8b586 8185struct task_struct *curr_task(int cpu)
1df5c10a
LT
8186{
8187 return cpu_curr(cpu);
8188}
8189
67fc4e0c
JW
8190#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
8191
8192#ifdef CONFIG_IA64
1df5c10a
LT
8193/**
8194 * set_curr_task - set the current task for a given cpu.
8195 * @cpu: the processor in question.
8196 * @p: the task pointer to set.
8197 *
8198 * Description: This function must only be used when non-maskable interrupts
41a2d6cf
IM
8199 * are serviced on a separate stack. It allows the architecture to switch the
8200 * notion of the current task on a cpu in a non-blocking manner. This function
1df5c10a
LT
8201 * must be called with all CPU's synchronized, and interrupts disabled, the
8202 * and caller must save the original value of the current task (see
8203 * curr_task() above) and restore that value before reenabling interrupts and
8204 * re-starting the system.
8205 *
8206 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8207 */
36c8b586 8208void set_curr_task(int cpu, struct task_struct *p)
1df5c10a
LT
8209{
8210 cpu_curr(cpu) = p;
8211}
8212
8213#endif
29f59db3 8214
bccbe08a
PZ
8215#ifdef CONFIG_FAIR_GROUP_SCHED
8216static void free_fair_sched_group(struct task_group *tg)
6f505b16
PZ
8217{
8218 int i;
8219
8220 for_each_possible_cpu(i) {
8221 if (tg->cfs_rq)
8222 kfree(tg->cfs_rq[i]);
8223 if (tg->se)
8224 kfree(tg->se[i]);
6f505b16
PZ
8225 }
8226
8227 kfree(tg->cfs_rq);
8228 kfree(tg->se);
6f505b16
PZ
8229}
8230
ec7dc8ac
DG
8231static
8232int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
29f59db3 8233{
29f59db3 8234 struct cfs_rq *cfs_rq;
eab17229 8235 struct sched_entity *se;
9b5b7751 8236 struct rq *rq;
29f59db3
SV
8237 int i;
8238
434d53b0 8239 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
29f59db3
SV
8240 if (!tg->cfs_rq)
8241 goto err;
434d53b0 8242 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
29f59db3
SV
8243 if (!tg->se)
8244 goto err;
052f1dc7
PZ
8245
8246 tg->shares = NICE_0_LOAD;
29f59db3
SV
8247
8248 for_each_possible_cpu(i) {
9b5b7751 8249 rq = cpu_rq(i);
29f59db3 8250
eab17229
LZ
8251 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8252 GFP_KERNEL, cpu_to_node(i));
29f59db3
SV
8253 if (!cfs_rq)
8254 goto err;
8255
eab17229
LZ
8256 se = kzalloc_node(sizeof(struct sched_entity),
8257 GFP_KERNEL, cpu_to_node(i));
29f59db3 8258 if (!se)
dfc12eb2 8259 goto err_free_rq;
29f59db3 8260
3d4b47b4 8261 init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
bccbe08a
PZ
8262 }
8263
8264 return 1;
8265
49246274 8266err_free_rq:
dfc12eb2 8267 kfree(cfs_rq);
49246274 8268err:
bccbe08a
PZ
8269 return 0;
8270}
8271
bccbe08a
PZ
8272static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8273{
3d4b47b4
PZ
8274 struct rq *rq = cpu_rq(cpu);
8275 unsigned long flags;
3d4b47b4
PZ
8276
8277 /*
8278 * Only empty task groups can be destroyed; so we can speculatively
8279 * check on_list without danger of it being re-added.
8280 */
8281 if (!tg->cfs_rq[cpu]->on_list)
8282 return;
8283
8284 raw_spin_lock_irqsave(&rq->lock, flags);
822bc180 8285 list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
3d4b47b4 8286 raw_spin_unlock_irqrestore(&rq->lock, flags);
bccbe08a 8287}
6d6bc0ad 8288#else /* !CONFG_FAIR_GROUP_SCHED */
bccbe08a
PZ
8289static inline void free_fair_sched_group(struct task_group *tg)
8290{
8291}
8292
ec7dc8ac
DG
8293static inline
8294int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8295{
8296 return 1;
8297}
8298
bccbe08a
PZ
8299static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8300{
8301}
6d6bc0ad 8302#endif /* CONFIG_FAIR_GROUP_SCHED */
052f1dc7
PZ
8303
8304#ifdef CONFIG_RT_GROUP_SCHED
bccbe08a
PZ
8305static void free_rt_sched_group(struct task_group *tg)
8306{
8307 int i;
8308
d0b27fa7
PZ
8309 destroy_rt_bandwidth(&tg->rt_bandwidth);
8310
bccbe08a
PZ
8311 for_each_possible_cpu(i) {
8312 if (tg->rt_rq)
8313 kfree(tg->rt_rq[i]);
8314 if (tg->rt_se)
8315 kfree(tg->rt_se[i]);
8316 }
8317
8318 kfree(tg->rt_rq);
8319 kfree(tg->rt_se);
8320}
8321
ec7dc8ac
DG
8322static
8323int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8324{
8325 struct rt_rq *rt_rq;
eab17229 8326 struct sched_rt_entity *rt_se;
bccbe08a
PZ
8327 struct rq *rq;
8328 int i;
8329
434d53b0 8330 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
bccbe08a
PZ
8331 if (!tg->rt_rq)
8332 goto err;
434d53b0 8333 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
bccbe08a
PZ
8334 if (!tg->rt_se)
8335 goto err;
8336
d0b27fa7
PZ
8337 init_rt_bandwidth(&tg->rt_bandwidth,
8338 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
bccbe08a
PZ
8339
8340 for_each_possible_cpu(i) {
8341 rq = cpu_rq(i);
8342
eab17229
LZ
8343 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8344 GFP_KERNEL, cpu_to_node(i));
6f505b16
PZ
8345 if (!rt_rq)
8346 goto err;
29f59db3 8347
eab17229
LZ
8348 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8349 GFP_KERNEL, cpu_to_node(i));
6f505b16 8350 if (!rt_se)
dfc12eb2 8351 goto err_free_rq;
29f59db3 8352
3d4b47b4 8353 init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]);
29f59db3
SV
8354 }
8355
bccbe08a
PZ
8356 return 1;
8357
49246274 8358err_free_rq:
dfc12eb2 8359 kfree(rt_rq);
49246274 8360err:
bccbe08a
PZ
8361 return 0;
8362}
6d6bc0ad 8363#else /* !CONFIG_RT_GROUP_SCHED */
bccbe08a
PZ
8364static inline void free_rt_sched_group(struct task_group *tg)
8365{
8366}
8367
ec7dc8ac
DG
8368static inline
8369int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8370{
8371 return 1;
8372}
6d6bc0ad 8373#endif /* CONFIG_RT_GROUP_SCHED */
bccbe08a 8374
7c941438 8375#ifdef CONFIG_CGROUP_SCHED
bccbe08a
PZ
8376static void free_sched_group(struct task_group *tg)
8377{
8378 free_fair_sched_group(tg);
8379 free_rt_sched_group(tg);
8380 kfree(tg);
8381}
8382
8383/* allocate runqueue etc for a new task group */
ec7dc8ac 8384struct task_group *sched_create_group(struct task_group *parent)
bccbe08a
PZ
8385{
8386 struct task_group *tg;
8387 unsigned long flags;
bccbe08a
PZ
8388
8389 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8390 if (!tg)
8391 return ERR_PTR(-ENOMEM);
8392
ec7dc8ac 8393 if (!alloc_fair_sched_group(tg, parent))
bccbe08a
PZ
8394 goto err;
8395
ec7dc8ac 8396 if (!alloc_rt_sched_group(tg, parent))
bccbe08a
PZ
8397 goto err;
8398
8ed36996 8399 spin_lock_irqsave(&task_group_lock, flags);
6f505b16 8400 list_add_rcu(&tg->list, &task_groups);
f473aa5e
PZ
8401
8402 WARN_ON(!parent); /* root should already exist */
8403
8404 tg->parent = parent;
f473aa5e 8405 INIT_LIST_HEAD(&tg->children);
09f2724a 8406 list_add_rcu(&tg->siblings, &parent->children);
8ed36996 8407 spin_unlock_irqrestore(&task_group_lock, flags);
29f59db3 8408
9b5b7751 8409 return tg;
29f59db3
SV
8410
8411err:
6f505b16 8412 free_sched_group(tg);
29f59db3
SV
8413 return ERR_PTR(-ENOMEM);
8414}
8415
9b5b7751 8416/* rcu callback to free various structures associated with a task group */
6f505b16 8417static void free_sched_group_rcu(struct rcu_head *rhp)
29f59db3 8418{
29f59db3 8419 /* now it should be safe to free those cfs_rqs */
6f505b16 8420 free_sched_group(container_of(rhp, struct task_group, rcu));
29f59db3
SV
8421}
8422
9b5b7751 8423/* Destroy runqueue etc associated with a task group */
4cf86d77 8424void sched_destroy_group(struct task_group *tg)
29f59db3 8425{
8ed36996 8426 unsigned long flags;
9b5b7751 8427 int i;
29f59db3 8428
3d4b47b4
PZ
8429 /* end participation in shares distribution */
8430 for_each_possible_cpu(i)
bccbe08a 8431 unregister_fair_sched_group(tg, i);
3d4b47b4
PZ
8432
8433 spin_lock_irqsave(&task_group_lock, flags);
6f505b16 8434 list_del_rcu(&tg->list);
f473aa5e 8435 list_del_rcu(&tg->siblings);
8ed36996 8436 spin_unlock_irqrestore(&task_group_lock, flags);
9b5b7751 8437
9b5b7751 8438 /* wait for possible concurrent references to cfs_rqs complete */
6f505b16 8439 call_rcu(&tg->rcu, free_sched_group_rcu);
29f59db3
SV
8440}
8441
9b5b7751 8442/* change task's runqueue when it moves between groups.
3a252015
IM
8443 * The caller of this function should have put the task in its new group
8444 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8445 * reflect its new group.
9b5b7751
SV
8446 */
8447void sched_move_task(struct task_struct *tsk)
29f59db3
SV
8448{
8449 int on_rq, running;
8450 unsigned long flags;
8451 struct rq *rq;
8452
8453 rq = task_rq_lock(tsk, &flags);
8454
051a1d1a 8455 running = task_current(rq, tsk);
29f59db3
SV
8456 on_rq = tsk->se.on_rq;
8457
0e1f3483 8458 if (on_rq)
29f59db3 8459 dequeue_task(rq, tsk, 0);
0e1f3483
HS
8460 if (unlikely(running))
8461 tsk->sched_class->put_prev_task(rq, tsk);
29f59db3 8462
810b3817 8463#ifdef CONFIG_FAIR_GROUP_SCHED
b2b5ce02
PZ
8464 if (tsk->sched_class->task_move_group)
8465 tsk->sched_class->task_move_group(tsk, on_rq);
8466 else
810b3817 8467#endif
b2b5ce02 8468 set_task_rq(tsk, task_cpu(tsk));
810b3817 8469
0e1f3483
HS
8470 if (unlikely(running))
8471 tsk->sched_class->set_curr_task(rq);
8472 if (on_rq)
371fd7e7 8473 enqueue_task(rq, tsk, 0);
29f59db3 8474
29f59db3
SV
8475 task_rq_unlock(rq, &flags);
8476}
7c941438 8477#endif /* CONFIG_CGROUP_SCHED */
29f59db3 8478
052f1dc7 8479#ifdef CONFIG_FAIR_GROUP_SCHED
8ed36996
PZ
8480static DEFINE_MUTEX(shares_mutex);
8481
4cf86d77 8482int sched_group_set_shares(struct task_group *tg, unsigned long shares)
29f59db3
SV
8483{
8484 int i;
8ed36996 8485 unsigned long flags;
c61935fd 8486
ec7dc8ac
DG
8487 /*
8488 * We can't change the weight of the root cgroup.
8489 */
8490 if (!tg->se[0])
8491 return -EINVAL;
8492
18d95a28
PZ
8493 if (shares < MIN_SHARES)
8494 shares = MIN_SHARES;
cb4ad1ff
MX
8495 else if (shares > MAX_SHARES)
8496 shares = MAX_SHARES;
62fb1851 8497
8ed36996 8498 mutex_lock(&shares_mutex);
9b5b7751 8499 if (tg->shares == shares)
5cb350ba 8500 goto done;
29f59db3 8501
9b5b7751 8502 tg->shares = shares;
c09595f6 8503 for_each_possible_cpu(i) {
9437178f
PT
8504 struct rq *rq = cpu_rq(i);
8505 struct sched_entity *se;
8506
8507 se = tg->se[i];
8508 /* Propagate contribution to hierarchy */
8509 raw_spin_lock_irqsave(&rq->lock, flags);
8510 for_each_sched_entity(se)
8511 update_cfs_shares(group_cfs_rq(se), 0);
8512 raw_spin_unlock_irqrestore(&rq->lock, flags);
c09595f6 8513 }
29f59db3 8514
5cb350ba 8515done:
8ed36996 8516 mutex_unlock(&shares_mutex);
9b5b7751 8517 return 0;
29f59db3
SV
8518}
8519
5cb350ba
DG
8520unsigned long sched_group_shares(struct task_group *tg)
8521{
8522 return tg->shares;
8523}
052f1dc7 8524#endif
5cb350ba 8525
052f1dc7 8526#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 8527/*
9f0c1e56 8528 * Ensure that the real time constraints are schedulable.
6f505b16 8529 */
9f0c1e56
PZ
8530static DEFINE_MUTEX(rt_constraints_mutex);
8531
8532static unsigned long to_ratio(u64 period, u64 runtime)
8533{
8534 if (runtime == RUNTIME_INF)
9a7e0b18 8535 return 1ULL << 20;
9f0c1e56 8536
9a7e0b18 8537 return div64_u64(runtime << 20, period);
9f0c1e56
PZ
8538}
8539
9a7e0b18
PZ
8540/* Must be called with tasklist_lock held */
8541static inline int tg_has_rt_tasks(struct task_group *tg)
b40b2e8e 8542{
9a7e0b18 8543 struct task_struct *g, *p;
b40b2e8e 8544
9a7e0b18
PZ
8545 do_each_thread(g, p) {
8546 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8547 return 1;
8548 } while_each_thread(g, p);
b40b2e8e 8549
9a7e0b18
PZ
8550 return 0;
8551}
b40b2e8e 8552
9a7e0b18
PZ
8553struct rt_schedulable_data {
8554 struct task_group *tg;
8555 u64 rt_period;
8556 u64 rt_runtime;
8557};
b40b2e8e 8558
9a7e0b18
PZ
8559static int tg_schedulable(struct task_group *tg, void *data)
8560{
8561 struct rt_schedulable_data *d = data;
8562 struct task_group *child;
8563 unsigned long total, sum = 0;
8564 u64 period, runtime;
b40b2e8e 8565
9a7e0b18
PZ
8566 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8567 runtime = tg->rt_bandwidth.rt_runtime;
b40b2e8e 8568
9a7e0b18
PZ
8569 if (tg == d->tg) {
8570 period = d->rt_period;
8571 runtime = d->rt_runtime;
b40b2e8e 8572 }
b40b2e8e 8573
4653f803
PZ
8574 /*
8575 * Cannot have more runtime than the period.
8576 */
8577 if (runtime > period && runtime != RUNTIME_INF)
8578 return -EINVAL;
6f505b16 8579
4653f803
PZ
8580 /*
8581 * Ensure we don't starve existing RT tasks.
8582 */
9a7e0b18
PZ
8583 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8584 return -EBUSY;
6f505b16 8585
9a7e0b18 8586 total = to_ratio(period, runtime);
6f505b16 8587
4653f803
PZ
8588 /*
8589 * Nobody can have more than the global setting allows.
8590 */
8591 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8592 return -EINVAL;
6f505b16 8593
4653f803
PZ
8594 /*
8595 * The sum of our children's runtime should not exceed our own.
8596 */
9a7e0b18
PZ
8597 list_for_each_entry_rcu(child, &tg->children, siblings) {
8598 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8599 runtime = child->rt_bandwidth.rt_runtime;
6f505b16 8600
9a7e0b18
PZ
8601 if (child == d->tg) {
8602 period = d->rt_period;
8603 runtime = d->rt_runtime;
8604 }
6f505b16 8605
9a7e0b18 8606 sum += to_ratio(period, runtime);
9f0c1e56 8607 }
6f505b16 8608
9a7e0b18
PZ
8609 if (sum > total)
8610 return -EINVAL;
8611
8612 return 0;
6f505b16
PZ
8613}
8614
9a7e0b18 8615static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
521f1a24 8616{
9a7e0b18
PZ
8617 struct rt_schedulable_data data = {
8618 .tg = tg,
8619 .rt_period = period,
8620 .rt_runtime = runtime,
8621 };
8622
8623 return walk_tg_tree(tg_schedulable, tg_nop, &data);
521f1a24
DG
8624}
8625
d0b27fa7
PZ
8626static int tg_set_bandwidth(struct task_group *tg,
8627 u64 rt_period, u64 rt_runtime)
6f505b16 8628{
ac086bc2 8629 int i, err = 0;
9f0c1e56 8630
9f0c1e56 8631 mutex_lock(&rt_constraints_mutex);
521f1a24 8632 read_lock(&tasklist_lock);
9a7e0b18
PZ
8633 err = __rt_schedulable(tg, rt_period, rt_runtime);
8634 if (err)
9f0c1e56 8635 goto unlock;
ac086bc2 8636
0986b11b 8637 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
d0b27fa7
PZ
8638 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8639 tg->rt_bandwidth.rt_runtime = rt_runtime;
ac086bc2
PZ
8640
8641 for_each_possible_cpu(i) {
8642 struct rt_rq *rt_rq = tg->rt_rq[i];
8643
0986b11b 8644 raw_spin_lock(&rt_rq->rt_runtime_lock);
ac086bc2 8645 rt_rq->rt_runtime = rt_runtime;
0986b11b 8646 raw_spin_unlock(&rt_rq->rt_runtime_lock);
ac086bc2 8647 }
0986b11b 8648 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
49246274 8649unlock:
521f1a24 8650 read_unlock(&tasklist_lock);
9f0c1e56
PZ
8651 mutex_unlock(&rt_constraints_mutex);
8652
8653 return err;
6f505b16
PZ
8654}
8655
d0b27fa7
PZ
8656int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8657{
8658 u64 rt_runtime, rt_period;
8659
8660 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8661 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8662 if (rt_runtime_us < 0)
8663 rt_runtime = RUNTIME_INF;
8664
8665 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8666}
8667
9f0c1e56
PZ
8668long sched_group_rt_runtime(struct task_group *tg)
8669{
8670 u64 rt_runtime_us;
8671
d0b27fa7 8672 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
9f0c1e56
PZ
8673 return -1;
8674
d0b27fa7 8675 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
9f0c1e56
PZ
8676 do_div(rt_runtime_us, NSEC_PER_USEC);
8677 return rt_runtime_us;
8678}
d0b27fa7
PZ
8679
8680int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8681{
8682 u64 rt_runtime, rt_period;
8683
8684 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8685 rt_runtime = tg->rt_bandwidth.rt_runtime;
8686
619b0488
R
8687 if (rt_period == 0)
8688 return -EINVAL;
8689
d0b27fa7
PZ
8690 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8691}
8692
8693long sched_group_rt_period(struct task_group *tg)
8694{
8695 u64 rt_period_us;
8696
8697 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8698 do_div(rt_period_us, NSEC_PER_USEC);
8699 return rt_period_us;
8700}
8701
8702static int sched_rt_global_constraints(void)
8703{
4653f803 8704 u64 runtime, period;
d0b27fa7
PZ
8705 int ret = 0;
8706
ec5d4989
HS
8707 if (sysctl_sched_rt_period <= 0)
8708 return -EINVAL;
8709
4653f803
PZ
8710 runtime = global_rt_runtime();
8711 period = global_rt_period();
8712
8713 /*
8714 * Sanity check on the sysctl variables.
8715 */
8716 if (runtime > period && runtime != RUNTIME_INF)
8717 return -EINVAL;
10b612f4 8718
d0b27fa7 8719 mutex_lock(&rt_constraints_mutex);
9a7e0b18 8720 read_lock(&tasklist_lock);
4653f803 8721 ret = __rt_schedulable(NULL, 0, 0);
9a7e0b18 8722 read_unlock(&tasklist_lock);
d0b27fa7
PZ
8723 mutex_unlock(&rt_constraints_mutex);
8724
8725 return ret;
8726}
54e99124
DG
8727
8728int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8729{
8730 /* Don't accept realtime tasks when there is no way for them to run */
8731 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
8732 return 0;
8733
8734 return 1;
8735}
8736
6d6bc0ad 8737#else /* !CONFIG_RT_GROUP_SCHED */
d0b27fa7
PZ
8738static int sched_rt_global_constraints(void)
8739{
ac086bc2
PZ
8740 unsigned long flags;
8741 int i;
8742
ec5d4989
HS
8743 if (sysctl_sched_rt_period <= 0)
8744 return -EINVAL;
8745
60aa605d
PZ
8746 /*
8747 * There's always some RT tasks in the root group
8748 * -- migration, kstopmachine etc..
8749 */
8750 if (sysctl_sched_rt_runtime == 0)
8751 return -EBUSY;
8752
0986b11b 8753 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
ac086bc2
PZ
8754 for_each_possible_cpu(i) {
8755 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
8756
0986b11b 8757 raw_spin_lock(&rt_rq->rt_runtime_lock);
ac086bc2 8758 rt_rq->rt_runtime = global_rt_runtime();
0986b11b 8759 raw_spin_unlock(&rt_rq->rt_runtime_lock);
ac086bc2 8760 }
0986b11b 8761 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
ac086bc2 8762
d0b27fa7
PZ
8763 return 0;
8764}
6d6bc0ad 8765#endif /* CONFIG_RT_GROUP_SCHED */
d0b27fa7
PZ
8766
8767int sched_rt_handler(struct ctl_table *table, int write,
8d65af78 8768 void __user *buffer, size_t *lenp,
d0b27fa7
PZ
8769 loff_t *ppos)
8770{
8771 int ret;
8772 int old_period, old_runtime;
8773 static DEFINE_MUTEX(mutex);
8774
8775 mutex_lock(&mutex);
8776 old_period = sysctl_sched_rt_period;
8777 old_runtime = sysctl_sched_rt_runtime;
8778
8d65af78 8779 ret = proc_dointvec(table, write, buffer, lenp, ppos);
d0b27fa7
PZ
8780
8781 if (!ret && write) {
8782 ret = sched_rt_global_constraints();
8783 if (ret) {
8784 sysctl_sched_rt_period = old_period;
8785 sysctl_sched_rt_runtime = old_runtime;
8786 } else {
8787 def_rt_bandwidth.rt_runtime = global_rt_runtime();
8788 def_rt_bandwidth.rt_period =
8789 ns_to_ktime(global_rt_period());
8790 }
8791 }
8792 mutex_unlock(&mutex);
8793
8794 return ret;
8795}
68318b8e 8796
052f1dc7 8797#ifdef CONFIG_CGROUP_SCHED
68318b8e
SV
8798
8799/* return corresponding task_group object of a cgroup */
2b01dfe3 8800static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
68318b8e 8801{
2b01dfe3
PM
8802 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
8803 struct task_group, css);
68318b8e
SV
8804}
8805
8806static struct cgroup_subsys_state *
2b01dfe3 8807cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
68318b8e 8808{
ec7dc8ac 8809 struct task_group *tg, *parent;
68318b8e 8810
2b01dfe3 8811 if (!cgrp->parent) {
68318b8e 8812 /* This is early initialization for the top cgroup */
07e06b01 8813 return &root_task_group.css;
68318b8e
SV
8814 }
8815
ec7dc8ac
DG
8816 parent = cgroup_tg(cgrp->parent);
8817 tg = sched_create_group(parent);
68318b8e
SV
8818 if (IS_ERR(tg))
8819 return ERR_PTR(-ENOMEM);
8820
68318b8e
SV
8821 return &tg->css;
8822}
8823
41a2d6cf
IM
8824static void
8825cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
68318b8e 8826{
2b01dfe3 8827 struct task_group *tg = cgroup_tg(cgrp);
68318b8e
SV
8828
8829 sched_destroy_group(tg);
8830}
8831
41a2d6cf 8832static int
be367d09 8833cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
68318b8e 8834{
b68aa230 8835#ifdef CONFIG_RT_GROUP_SCHED
54e99124 8836 if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk))
b68aa230
PZ
8837 return -EINVAL;
8838#else
68318b8e
SV
8839 /* We don't support RT-tasks being in separate groups */
8840 if (tsk->sched_class != &fair_sched_class)
8841 return -EINVAL;
b68aa230 8842#endif
be367d09
BB
8843 return 0;
8844}
68318b8e 8845
be367d09
BB
8846static int
8847cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
8848 struct task_struct *tsk, bool threadgroup)
8849{
8850 int retval = cpu_cgroup_can_attach_task(cgrp, tsk);
8851 if (retval)
8852 return retval;
8853 if (threadgroup) {
8854 struct task_struct *c;
8855 rcu_read_lock();
8856 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8857 retval = cpu_cgroup_can_attach_task(cgrp, c);
8858 if (retval) {
8859 rcu_read_unlock();
8860 return retval;
8861 }
8862 }
8863 rcu_read_unlock();
8864 }
68318b8e
SV
8865 return 0;
8866}
8867
8868static void
2b01dfe3 8869cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
be367d09
BB
8870 struct cgroup *old_cont, struct task_struct *tsk,
8871 bool threadgroup)
68318b8e
SV
8872{
8873 sched_move_task(tsk);
be367d09
BB
8874 if (threadgroup) {
8875 struct task_struct *c;
8876 rcu_read_lock();
8877 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8878 sched_move_task(c);
8879 }
8880 rcu_read_unlock();
8881 }
68318b8e
SV
8882}
8883
052f1dc7 8884#ifdef CONFIG_FAIR_GROUP_SCHED
f4c753b7 8885static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
2b01dfe3 8886 u64 shareval)
68318b8e 8887{
2b01dfe3 8888 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
68318b8e
SV
8889}
8890
f4c753b7 8891static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
68318b8e 8892{
2b01dfe3 8893 struct task_group *tg = cgroup_tg(cgrp);
68318b8e
SV
8894
8895 return (u64) tg->shares;
8896}
6d6bc0ad 8897#endif /* CONFIG_FAIR_GROUP_SCHED */
68318b8e 8898
052f1dc7 8899#ifdef CONFIG_RT_GROUP_SCHED
0c70814c 8900static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
06ecb27c 8901 s64 val)
6f505b16 8902{
06ecb27c 8903 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
6f505b16
PZ
8904}
8905
06ecb27c 8906static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
6f505b16 8907{
06ecb27c 8908 return sched_group_rt_runtime(cgroup_tg(cgrp));
6f505b16 8909}
d0b27fa7
PZ
8910
8911static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
8912 u64 rt_period_us)
8913{
8914 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
8915}
8916
8917static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
8918{
8919 return sched_group_rt_period(cgroup_tg(cgrp));
8920}
6d6bc0ad 8921#endif /* CONFIG_RT_GROUP_SCHED */
6f505b16 8922
fe5c7cc2 8923static struct cftype cpu_files[] = {
052f1dc7 8924#ifdef CONFIG_FAIR_GROUP_SCHED
fe5c7cc2
PM
8925 {
8926 .name = "shares",
f4c753b7
PM
8927 .read_u64 = cpu_shares_read_u64,
8928 .write_u64 = cpu_shares_write_u64,
fe5c7cc2 8929 },
052f1dc7
PZ
8930#endif
8931#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 8932 {
9f0c1e56 8933 .name = "rt_runtime_us",
06ecb27c
PM
8934 .read_s64 = cpu_rt_runtime_read,
8935 .write_s64 = cpu_rt_runtime_write,
6f505b16 8936 },
d0b27fa7
PZ
8937 {
8938 .name = "rt_period_us",
f4c753b7
PM
8939 .read_u64 = cpu_rt_period_read_uint,
8940 .write_u64 = cpu_rt_period_write_uint,
d0b27fa7 8941 },
052f1dc7 8942#endif
68318b8e
SV
8943};
8944
8945static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
8946{
fe5c7cc2 8947 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
68318b8e
SV
8948}
8949
8950struct cgroup_subsys cpu_cgroup_subsys = {
38605cae
IM
8951 .name = "cpu",
8952 .create = cpu_cgroup_create,
8953 .destroy = cpu_cgroup_destroy,
8954 .can_attach = cpu_cgroup_can_attach,
8955 .attach = cpu_cgroup_attach,
8956 .populate = cpu_cgroup_populate,
8957 .subsys_id = cpu_cgroup_subsys_id,
68318b8e
SV
8958 .early_init = 1,
8959};
8960
052f1dc7 8961#endif /* CONFIG_CGROUP_SCHED */
d842de87
SV
8962
8963#ifdef CONFIG_CGROUP_CPUACCT
8964
8965/*
8966 * CPU accounting code for task groups.
8967 *
8968 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
8969 * (balbir@in.ibm.com).
8970 */
8971
934352f2 8972/* track cpu usage of a group of tasks and its child groups */
d842de87
SV
8973struct cpuacct {
8974 struct cgroup_subsys_state css;
8975 /* cpuusage holds pointer to a u64-type object on every cpu */
43cf38eb 8976 u64 __percpu *cpuusage;
ef12fefa 8977 struct percpu_counter cpustat[CPUACCT_STAT_NSTATS];
934352f2 8978 struct cpuacct *parent;
d842de87
SV
8979};
8980
8981struct cgroup_subsys cpuacct_subsys;
8982
8983/* return cpu accounting group corresponding to this container */
32cd756a 8984static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
d842de87 8985{
32cd756a 8986 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
d842de87
SV
8987 struct cpuacct, css);
8988}
8989
8990/* return cpu accounting group to which this task belongs */
8991static inline struct cpuacct *task_ca(struct task_struct *tsk)
8992{
8993 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
8994 struct cpuacct, css);
8995}
8996
8997/* create a new cpu accounting group */
8998static struct cgroup_subsys_state *cpuacct_create(
32cd756a 8999 struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87
SV
9000{
9001 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
ef12fefa 9002 int i;
d842de87
SV
9003
9004 if (!ca)
ef12fefa 9005 goto out;
d842de87
SV
9006
9007 ca->cpuusage = alloc_percpu(u64);
ef12fefa
BR
9008 if (!ca->cpuusage)
9009 goto out_free_ca;
9010
9011 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
9012 if (percpu_counter_init(&ca->cpustat[i], 0))
9013 goto out_free_counters;
d842de87 9014
934352f2
BR
9015 if (cgrp->parent)
9016 ca->parent = cgroup_ca(cgrp->parent);
9017
d842de87 9018 return &ca->css;
ef12fefa
BR
9019
9020out_free_counters:
9021 while (--i >= 0)
9022 percpu_counter_destroy(&ca->cpustat[i]);
9023 free_percpu(ca->cpuusage);
9024out_free_ca:
9025 kfree(ca);
9026out:
9027 return ERR_PTR(-ENOMEM);
d842de87
SV
9028}
9029
9030/* destroy an existing cpu accounting group */
41a2d6cf 9031static void
32cd756a 9032cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87 9033{
32cd756a 9034 struct cpuacct *ca = cgroup_ca(cgrp);
ef12fefa 9035 int i;
d842de87 9036
ef12fefa
BR
9037 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
9038 percpu_counter_destroy(&ca->cpustat[i]);
d842de87
SV
9039 free_percpu(ca->cpuusage);
9040 kfree(ca);
9041}
9042
720f5498
KC
9043static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
9044{
b36128c8 9045 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
720f5498
KC
9046 u64 data;
9047
9048#ifndef CONFIG_64BIT
9049 /*
9050 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
9051 */
05fa785c 9052 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
720f5498 9053 data = *cpuusage;
05fa785c 9054 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
720f5498
KC
9055#else
9056 data = *cpuusage;
9057#endif
9058
9059 return data;
9060}
9061
9062static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
9063{
b36128c8 9064 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
720f5498
KC
9065
9066#ifndef CONFIG_64BIT
9067 /*
9068 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
9069 */
05fa785c 9070 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
720f5498 9071 *cpuusage = val;
05fa785c 9072 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
720f5498
KC
9073#else
9074 *cpuusage = val;
9075#endif
9076}
9077
d842de87 9078/* return total cpu usage (in nanoseconds) of a group */
32cd756a 9079static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
d842de87 9080{
32cd756a 9081 struct cpuacct *ca = cgroup_ca(cgrp);
d842de87
SV
9082 u64 totalcpuusage = 0;
9083 int i;
9084
720f5498
KC
9085 for_each_present_cpu(i)
9086 totalcpuusage += cpuacct_cpuusage_read(ca, i);
d842de87
SV
9087
9088 return totalcpuusage;
9089}
9090
0297b803
DG
9091static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
9092 u64 reset)
9093{
9094 struct cpuacct *ca = cgroup_ca(cgrp);
9095 int err = 0;
9096 int i;
9097
9098 if (reset) {
9099 err = -EINVAL;
9100 goto out;
9101 }
9102
720f5498
KC
9103 for_each_present_cpu(i)
9104 cpuacct_cpuusage_write(ca, i, 0);
0297b803 9105
0297b803
DG
9106out:
9107 return err;
9108}
9109
e9515c3c
KC
9110static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
9111 struct seq_file *m)
9112{
9113 struct cpuacct *ca = cgroup_ca(cgroup);
9114 u64 percpu;
9115 int i;
9116
9117 for_each_present_cpu(i) {
9118 percpu = cpuacct_cpuusage_read(ca, i);
9119 seq_printf(m, "%llu ", (unsigned long long) percpu);
9120 }
9121 seq_printf(m, "\n");
9122 return 0;
9123}
9124
ef12fefa
BR
9125static const char *cpuacct_stat_desc[] = {
9126 [CPUACCT_STAT_USER] = "user",
9127 [CPUACCT_STAT_SYSTEM] = "system",
9128};
9129
9130static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
9131 struct cgroup_map_cb *cb)
9132{
9133 struct cpuacct *ca = cgroup_ca(cgrp);
9134 int i;
9135
9136 for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
9137 s64 val = percpu_counter_read(&ca->cpustat[i]);
9138 val = cputime64_to_clock_t(val);
9139 cb->fill(cb, cpuacct_stat_desc[i], val);
9140 }
9141 return 0;
9142}
9143
d842de87
SV
9144static struct cftype files[] = {
9145 {
9146 .name = "usage",
f4c753b7
PM
9147 .read_u64 = cpuusage_read,
9148 .write_u64 = cpuusage_write,
d842de87 9149 },
e9515c3c
KC
9150 {
9151 .name = "usage_percpu",
9152 .read_seq_string = cpuacct_percpu_seq_read,
9153 },
ef12fefa
BR
9154 {
9155 .name = "stat",
9156 .read_map = cpuacct_stats_show,
9157 },
d842de87
SV
9158};
9159
32cd756a 9160static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87 9161{
32cd756a 9162 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
d842de87
SV
9163}
9164
9165/*
9166 * charge this task's execution time to its accounting group.
9167 *
9168 * called with rq->lock held.
9169 */
9170static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9171{
9172 struct cpuacct *ca;
934352f2 9173 int cpu;
d842de87 9174
c40c6f85 9175 if (unlikely(!cpuacct_subsys.active))
d842de87
SV
9176 return;
9177
934352f2 9178 cpu = task_cpu(tsk);
a18b83b7
BR
9179
9180 rcu_read_lock();
9181
d842de87 9182 ca = task_ca(tsk);
d842de87 9183
934352f2 9184 for (; ca; ca = ca->parent) {
b36128c8 9185 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
d842de87
SV
9186 *cpuusage += cputime;
9187 }
a18b83b7
BR
9188
9189 rcu_read_unlock();
d842de87
SV
9190}
9191
fa535a77
AB
9192/*
9193 * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large
9194 * in cputime_t units. As a result, cpuacct_update_stats calls
9195 * percpu_counter_add with values large enough to always overflow the
9196 * per cpu batch limit causing bad SMP scalability.
9197 *
9198 * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we
9199 * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled
9200 * and enabled. We cap it at INT_MAX which is the largest allowed batch value.
9201 */
9202#ifdef CONFIG_SMP
9203#define CPUACCT_BATCH \
9204 min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX)
9205#else
9206#define CPUACCT_BATCH 0
9207#endif
9208
ef12fefa
BR
9209/*
9210 * Charge the system/user time to the task's accounting group.
9211 */
9212static void cpuacct_update_stats(struct task_struct *tsk,
9213 enum cpuacct_stat_index idx, cputime_t val)
9214{
9215 struct cpuacct *ca;
fa535a77 9216 int batch = CPUACCT_BATCH;
ef12fefa
BR
9217
9218 if (unlikely(!cpuacct_subsys.active))
9219 return;
9220
9221 rcu_read_lock();
9222 ca = task_ca(tsk);
9223
9224 do {
fa535a77 9225 __percpu_counter_add(&ca->cpustat[idx], val, batch);
ef12fefa
BR
9226 ca = ca->parent;
9227 } while (ca);
9228 rcu_read_unlock();
9229}
9230
d842de87
SV
9231struct cgroup_subsys cpuacct_subsys = {
9232 .name = "cpuacct",
9233 .create = cpuacct_create,
9234 .destroy = cpuacct_destroy,
9235 .populate = cpuacct_populate,
9236 .subsys_id = cpuacct_subsys_id,
9237};
9238#endif /* CONFIG_CGROUP_CPUACCT */
03b042bf 9239
This page took 1.965534 seconds and 5 git commands to generate.