]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - kernel/sched.c
sched: Remove irq time from available CPU power
[mirror_ubuntu-zesty-kernel.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>
1da177e4 78
6e0534f2 79#include "sched_cpupri.h"
21aa9af0 80#include "workqueue_sched.h"
6e0534f2 81
a8d154b0 82#define CREATE_TRACE_POINTS
ad8d75ff 83#include <trace/events/sched.h>
a8d154b0 84
1da177e4
LT
85/*
86 * Convert user-nice values [ -20 ... 0 ... 19 ]
87 * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
88 * and back.
89 */
90#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
91#define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
92#define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
93
94/*
95 * 'User priority' is the nice value converted to something we
96 * can work with better when scaling various scheduler parameters,
97 * it's a [ 0 ... 39 ] range.
98 */
99#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
100#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
101#define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
102
103/*
d7876a08 104 * Helpers for converting nanosecond timing to jiffy resolution
1da177e4 105 */
d6322faf 106#define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ))
1da177e4 107
6aa645ea
IM
108#define NICE_0_LOAD SCHED_LOAD_SCALE
109#define NICE_0_SHIFT SCHED_LOAD_SHIFT
110
1da177e4
LT
111/*
112 * These are the 'tuning knobs' of the scheduler:
113 *
a4ec24b4 114 * default timeslice is 100 msecs (used only for SCHED_RR tasks).
1da177e4
LT
115 * Timeslices get refilled after they expire.
116 */
1da177e4 117#define DEF_TIMESLICE (100 * HZ / 1000)
2dd73a4f 118
d0b27fa7
PZ
119/*
120 * single value that denotes runtime == period, ie unlimited time.
121 */
122#define RUNTIME_INF ((u64)~0ULL)
123
e05606d3
IM
124static inline int rt_policy(int policy)
125{
3f33a7ce 126 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
e05606d3
IM
127 return 1;
128 return 0;
129}
130
131static inline int task_has_rt_policy(struct task_struct *p)
132{
133 return rt_policy(p->policy);
134}
135
1da177e4 136/*
6aa645ea 137 * This is the priority-queue data structure of the RT scheduling class:
1da177e4 138 */
6aa645ea
IM
139struct rt_prio_array {
140 DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
141 struct list_head queue[MAX_RT_PRIO];
142};
143
d0b27fa7 144struct rt_bandwidth {
ea736ed5 145 /* nests inside the rq lock: */
0986b11b 146 raw_spinlock_t rt_runtime_lock;
ea736ed5
IM
147 ktime_t rt_period;
148 u64 rt_runtime;
149 struct hrtimer rt_period_timer;
d0b27fa7
PZ
150};
151
152static struct rt_bandwidth def_rt_bandwidth;
153
154static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
155
156static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
157{
158 struct rt_bandwidth *rt_b =
159 container_of(timer, struct rt_bandwidth, rt_period_timer);
160 ktime_t now;
161 int overrun;
162 int idle = 0;
163
164 for (;;) {
165 now = hrtimer_cb_get_time(timer);
166 overrun = hrtimer_forward(timer, now, rt_b->rt_period);
167
168 if (!overrun)
169 break;
170
171 idle = do_sched_rt_period_timer(rt_b, overrun);
172 }
173
174 return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
175}
176
177static
178void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
179{
180 rt_b->rt_period = ns_to_ktime(period);
181 rt_b->rt_runtime = runtime;
182
0986b11b 183 raw_spin_lock_init(&rt_b->rt_runtime_lock);
ac086bc2 184
d0b27fa7
PZ
185 hrtimer_init(&rt_b->rt_period_timer,
186 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
187 rt_b->rt_period_timer.function = sched_rt_period_timer;
d0b27fa7
PZ
188}
189
c8bfff6d
KH
190static inline int rt_bandwidth_enabled(void)
191{
192 return sysctl_sched_rt_runtime >= 0;
d0b27fa7
PZ
193}
194
195static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
196{
197 ktime_t now;
198
cac64d00 199 if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF)
d0b27fa7
PZ
200 return;
201
202 if (hrtimer_active(&rt_b->rt_period_timer))
203 return;
204
0986b11b 205 raw_spin_lock(&rt_b->rt_runtime_lock);
d0b27fa7 206 for (;;) {
7f1e2ca9
PZ
207 unsigned long delta;
208 ktime_t soft, hard;
209
d0b27fa7
PZ
210 if (hrtimer_active(&rt_b->rt_period_timer))
211 break;
212
213 now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
214 hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
7f1e2ca9
PZ
215
216 soft = hrtimer_get_softexpires(&rt_b->rt_period_timer);
217 hard = hrtimer_get_expires(&rt_b->rt_period_timer);
218 delta = ktime_to_ns(ktime_sub(hard, soft));
219 __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta,
5c333864 220 HRTIMER_MODE_ABS_PINNED, 0);
d0b27fa7 221 }
0986b11b 222 raw_spin_unlock(&rt_b->rt_runtime_lock);
d0b27fa7
PZ
223}
224
225#ifdef CONFIG_RT_GROUP_SCHED
226static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
227{
228 hrtimer_cancel(&rt_b->rt_period_timer);
229}
230#endif
231
712555ee
HC
232/*
233 * sched_domains_mutex serializes calls to arch_init_sched_domains,
234 * detach_destroy_domains and partition_sched_domains.
235 */
236static DEFINE_MUTEX(sched_domains_mutex);
237
7c941438 238#ifdef CONFIG_CGROUP_SCHED
29f59db3 239
68318b8e
SV
240#include <linux/cgroup.h>
241
29f59db3
SV
242struct cfs_rq;
243
6f505b16
PZ
244static LIST_HEAD(task_groups);
245
29f59db3 246/* task group related information */
4cf86d77 247struct task_group {
68318b8e 248 struct cgroup_subsys_state css;
6c415b92 249
052f1dc7 250#ifdef CONFIG_FAIR_GROUP_SCHED
29f59db3
SV
251 /* schedulable entities of this group on each cpu */
252 struct sched_entity **se;
253 /* runqueue "owned" by this group on each cpu */
254 struct cfs_rq **cfs_rq;
255 unsigned long shares;
052f1dc7
PZ
256#endif
257
258#ifdef CONFIG_RT_GROUP_SCHED
259 struct sched_rt_entity **rt_se;
260 struct rt_rq **rt_rq;
261
d0b27fa7 262 struct rt_bandwidth rt_bandwidth;
052f1dc7 263#endif
6b2d7700 264
ae8393e5 265 struct rcu_head rcu;
6f505b16 266 struct list_head list;
f473aa5e
PZ
267
268 struct task_group *parent;
269 struct list_head siblings;
270 struct list_head children;
29f59db3
SV
271};
272
eff766a6 273#define root_task_group init_task_group
6f505b16 274
8ed36996 275/* task_group_lock serializes add/remove of task groups and also changes to
ec2c507f
SV
276 * a task group's cpu shares.
277 */
8ed36996 278static DEFINE_SPINLOCK(task_group_lock);
ec2c507f 279
e9036b36
CG
280#ifdef CONFIG_FAIR_GROUP_SCHED
281
57310a98
PZ
282#ifdef CONFIG_SMP
283static int root_task_group_empty(void)
284{
285 return list_empty(&root_task_group.children);
286}
287#endif
288
052f1dc7 289# define INIT_TASK_GROUP_LOAD NICE_0_LOAD
052f1dc7 290
cb4ad1ff 291/*
2e084786
LJ
292 * A weight of 0 or 1 can cause arithmetics problems.
293 * A weight of a cfs_rq is the sum of weights of which entities
294 * are queued on this cfs_rq, so a weight of a entity should not be
295 * too large, so as the shares value of a task group.
cb4ad1ff
MX
296 * (The default weight is 1024 - so there's no practical
297 * limitation from this.)
298 */
18d95a28 299#define MIN_SHARES 2
2e084786 300#define MAX_SHARES (1UL << 18)
18d95a28 301
052f1dc7
PZ
302static int init_task_group_load = INIT_TASK_GROUP_LOAD;
303#endif
304
29f59db3 305/* Default task group.
3a252015 306 * Every task in system belong to this group at bootup.
29f59db3 307 */
434d53b0 308struct task_group init_task_group;
29f59db3 309
7c941438 310#endif /* CONFIG_CGROUP_SCHED */
29f59db3 311
6aa645ea
IM
312/* CFS-related fields in a runqueue */
313struct cfs_rq {
314 struct load_weight load;
315 unsigned long nr_running;
316
6aa645ea 317 u64 exec_clock;
e9acbff6 318 u64 min_vruntime;
6aa645ea
IM
319
320 struct rb_root tasks_timeline;
321 struct rb_node *rb_leftmost;
4a55bd5e
PZ
322
323 struct list_head tasks;
324 struct list_head *balance_iterator;
325
326 /*
327 * 'curr' points to currently running entity on this cfs_rq.
6aa645ea
IM
328 * It is set to NULL otherwise (i.e when none are currently running).
329 */
4793241b 330 struct sched_entity *curr, *next, *last;
ddc97297 331
5ac5c4d6 332 unsigned int nr_spread_over;
ddc97297 333
62160e3f 334#ifdef CONFIG_FAIR_GROUP_SCHED
6aa645ea
IM
335 struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
336
41a2d6cf
IM
337 /*
338 * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
6aa645ea
IM
339 * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
340 * (like users, containers etc.)
341 *
342 * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
343 * list is used during load balance.
344 */
41a2d6cf
IM
345 struct list_head leaf_cfs_rq_list;
346 struct task_group *tg; /* group that "owns" this runqueue */
c09595f6
PZ
347
348#ifdef CONFIG_SMP
c09595f6 349 /*
c8cba857 350 * the part of load.weight contributed by tasks
c09595f6 351 */
c8cba857 352 unsigned long task_weight;
c09595f6 353
c8cba857
PZ
354 /*
355 * h_load = weight * f(tg)
356 *
357 * Where f(tg) is the recursive weight fraction assigned to
358 * this group.
359 */
360 unsigned long h_load;
c09595f6 361
c8cba857
PZ
362 /*
363 * this cpu's part of tg->shares
364 */
365 unsigned long shares;
f1d239f7
PZ
366
367 /*
368 * load.weight at the time we set shares
369 */
370 unsigned long rq_weight;
c09595f6 371#endif
6aa645ea
IM
372#endif
373};
1da177e4 374
6aa645ea
IM
375/* Real-Time classes' related field in a runqueue: */
376struct rt_rq {
377 struct rt_prio_array active;
63489e45 378 unsigned long rt_nr_running;
052f1dc7 379#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
e864c499
GH
380 struct {
381 int curr; /* highest queued rt task prio */
398a153b 382#ifdef CONFIG_SMP
e864c499 383 int next; /* next highest */
398a153b 384#endif
e864c499 385 } highest_prio;
6f505b16 386#endif
fa85ae24 387#ifdef CONFIG_SMP
73fe6aae 388 unsigned long rt_nr_migratory;
a1ba4d8b 389 unsigned long rt_nr_total;
a22d7fc1 390 int overloaded;
917b627d 391 struct plist_head pushable_tasks;
fa85ae24 392#endif
6f505b16 393 int rt_throttled;
fa85ae24 394 u64 rt_time;
ac086bc2 395 u64 rt_runtime;
ea736ed5 396 /* Nests inside the rq lock: */
0986b11b 397 raw_spinlock_t rt_runtime_lock;
6f505b16 398
052f1dc7 399#ifdef CONFIG_RT_GROUP_SCHED
23b0fdfc
PZ
400 unsigned long rt_nr_boosted;
401
6f505b16
PZ
402 struct rq *rq;
403 struct list_head leaf_rt_rq_list;
404 struct task_group *tg;
6f505b16 405#endif
6aa645ea
IM
406};
407
57d885fe
GH
408#ifdef CONFIG_SMP
409
410/*
411 * We add the notion of a root-domain which will be used to define per-domain
0eab9146
IM
412 * variables. Each exclusive cpuset essentially defines an island domain by
413 * fully partitioning the member cpus from any other cpuset. Whenever a new
57d885fe
GH
414 * exclusive cpuset is created, we also create and attach a new root-domain
415 * object.
416 *
57d885fe
GH
417 */
418struct root_domain {
419 atomic_t refcount;
c6c4927b
RR
420 cpumask_var_t span;
421 cpumask_var_t online;
637f5085 422
0eab9146 423 /*
637f5085
GH
424 * The "RT overload" flag: it gets set if a CPU has more than
425 * one runnable RT task.
426 */
c6c4927b 427 cpumask_var_t rto_mask;
0eab9146 428 atomic_t rto_count;
6e0534f2 429 struct cpupri cpupri;
57d885fe
GH
430};
431
dc938520
GH
432/*
433 * By default the system creates a single root-domain with all cpus as
434 * members (mimicking the global state we have today).
435 */
57d885fe
GH
436static struct root_domain def_root_domain;
437
ed2d372c 438#endif /* CONFIG_SMP */
57d885fe 439
1da177e4
LT
440/*
441 * This is the main, per-CPU runqueue data structure.
442 *
443 * Locking rule: those places that want to lock multiple runqueues
444 * (such as the load balancing or the thread migration code), lock
445 * acquire operations must be ordered by ascending &runqueue.
446 */
70b97a7f 447struct rq {
d8016491 448 /* runqueue lock: */
05fa785c 449 raw_spinlock_t lock;
1da177e4
LT
450
451 /*
452 * nr_running and cpu_load should be in the same cacheline because
453 * remote CPUs use both these fields when doing load calculation.
454 */
455 unsigned long nr_running;
6aa645ea
IM
456 #define CPU_LOAD_IDX_MAX 5
457 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
fdf3e95d 458 unsigned long last_load_update_tick;
46cb4b7c 459#ifdef CONFIG_NO_HZ
39c0cbe2 460 u64 nohz_stamp;
83cd4fe2 461 unsigned char nohz_balance_kick;
46cb4b7c 462#endif
a64692a3
MG
463 unsigned int skip_clock_update;
464
d8016491
IM
465 /* capture load from *all* tasks on this cpu: */
466 struct load_weight load;
6aa645ea
IM
467 unsigned long nr_load_updates;
468 u64 nr_switches;
469
470 struct cfs_rq cfs;
6f505b16 471 struct rt_rq rt;
6f505b16 472
6aa645ea 473#ifdef CONFIG_FAIR_GROUP_SCHED
d8016491
IM
474 /* list of leaf cfs_rq on this cpu: */
475 struct list_head leaf_cfs_rq_list;
052f1dc7
PZ
476#endif
477#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 478 struct list_head leaf_rt_rq_list;
1da177e4 479#endif
1da177e4
LT
480
481 /*
482 * This is part of a global counter where only the total sum
483 * over all CPUs matters. A task can increase this counter on
484 * one CPU and if it got migrated afterwards it may decrease
485 * it on another CPU. Always updated under the runqueue lock:
486 */
487 unsigned long nr_uninterruptible;
488
34f971f6 489 struct task_struct *curr, *idle, *stop;
c9819f45 490 unsigned long next_balance;
1da177e4 491 struct mm_struct *prev_mm;
6aa645ea 492
3e51f33f 493 u64 clock;
305e6835 494 u64 clock_task;
6aa645ea 495
1da177e4
LT
496 atomic_t nr_iowait;
497
498#ifdef CONFIG_SMP
0eab9146 499 struct root_domain *rd;
1da177e4
LT
500 struct sched_domain *sd;
501
e51fd5e2
PZ
502 unsigned long cpu_power;
503
a0a522ce 504 unsigned char idle_at_tick;
1da177e4 505 /* For active balancing */
3f029d3c 506 int post_schedule;
1da177e4
LT
507 int active_balance;
508 int push_cpu;
969c7921 509 struct cpu_stop_work active_balance_work;
d8016491
IM
510 /* cpu of this runqueue: */
511 int cpu;
1f11eb6a 512 int online;
1da177e4 513
a8a51d5e 514 unsigned long avg_load_per_task;
1da177e4 515
e9e9250b
PZ
516 u64 rt_avg;
517 u64 age_stamp;
1b9508f6
MG
518 u64 idle_stamp;
519 u64 avg_idle;
1da177e4
LT
520#endif
521
aa483808
VP
522#ifdef CONFIG_IRQ_TIME_ACCOUNTING
523 u64 prev_irq_time;
524#endif
525
dce48a84
TG
526 /* calc_load related fields */
527 unsigned long calc_load_update;
528 long calc_load_active;
529
8f4d37ec 530#ifdef CONFIG_SCHED_HRTICK
31656519
PZ
531#ifdef CONFIG_SMP
532 int hrtick_csd_pending;
533 struct call_single_data hrtick_csd;
534#endif
8f4d37ec
PZ
535 struct hrtimer hrtick_timer;
536#endif
537
1da177e4
LT
538#ifdef CONFIG_SCHEDSTATS
539 /* latency stats */
540 struct sched_info rq_sched_info;
9c2c4802
KC
541 unsigned long long rq_cpu_time;
542 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
1da177e4
LT
543
544 /* sys_sched_yield() stats */
480b9434 545 unsigned int yld_count;
1da177e4
LT
546
547 /* schedule() stats */
480b9434
KC
548 unsigned int sched_switch;
549 unsigned int sched_count;
550 unsigned int sched_goidle;
1da177e4
LT
551
552 /* try_to_wake_up() stats */
480b9434
KC
553 unsigned int ttwu_count;
554 unsigned int ttwu_local;
b8efb561
IM
555
556 /* BKL stats */
480b9434 557 unsigned int bkl_count;
1da177e4
LT
558#endif
559};
560
f34e3b61 561static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
1da177e4 562
7d478721
PZ
563static inline
564void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
dd41f596 565{
7d478721 566 rq->curr->sched_class->check_preempt_curr(rq, p, flags);
a64692a3
MG
567
568 /*
569 * A queue event has occurred, and we're going to schedule. In
570 * this case, we can save a useless back to back clock update.
571 */
572 if (test_tsk_need_resched(p))
573 rq->skip_clock_update = 1;
dd41f596
IM
574}
575
0a2966b4
CL
576static inline int cpu_of(struct rq *rq)
577{
578#ifdef CONFIG_SMP
579 return rq->cpu;
580#else
581 return 0;
582#endif
583}
584
497f0ab3 585#define rcu_dereference_check_sched_domain(p) \
d11c563d
PM
586 rcu_dereference_check((p), \
587 rcu_read_lock_sched_held() || \
588 lockdep_is_held(&sched_domains_mutex))
589
674311d5
NP
590/*
591 * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
1a20ff27 592 * See detach_destroy_domains: synchronize_sched for details.
674311d5
NP
593 *
594 * The domain tree of any CPU may only be accessed from within
595 * preempt-disabled sections.
596 */
48f24c4d 597#define for_each_domain(cpu, __sd) \
497f0ab3 598 for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
1da177e4
LT
599
600#define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
601#define this_rq() (&__get_cpu_var(runqueues))
602#define task_rq(p) cpu_rq(task_cpu(p))
603#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
54d35f29 604#define raw_rq() (&__raw_get_cpu_var(runqueues))
1da177e4 605
dc61b1d6
PZ
606#ifdef CONFIG_CGROUP_SCHED
607
608/*
609 * Return the group to which this tasks belongs.
610 *
611 * We use task_subsys_state_check() and extend the RCU verification
612 * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach()
613 * holds that lock for each task it moves into the cgroup. Therefore
614 * by holding that lock, we pin the task to the current cgroup.
615 */
616static inline struct task_group *task_group(struct task_struct *p)
617{
618 struct cgroup_subsys_state *css;
619
620 css = task_subsys_state_check(p, cpu_cgroup_subsys_id,
621 lockdep_is_held(&task_rq(p)->lock));
622 return container_of(css, struct task_group, css);
623}
624
625/* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
626static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
627{
628#ifdef CONFIG_FAIR_GROUP_SCHED
629 p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
630 p->se.parent = task_group(p)->se[cpu];
631#endif
632
633#ifdef CONFIG_RT_GROUP_SCHED
634 p->rt.rt_rq = task_group(p)->rt_rq[cpu];
635 p->rt.parent = task_group(p)->rt_se[cpu];
636#endif
637}
638
639#else /* CONFIG_CGROUP_SCHED */
640
641static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
642static inline struct task_group *task_group(struct task_struct *p)
643{
644 return NULL;
645}
646
647#endif /* CONFIG_CGROUP_SCHED */
648
305e6835 649static u64 irq_time_cpu(int cpu);
aa483808 650static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time);
305e6835 651
aa9c4c0f 652inline void update_rq_clock(struct rq *rq)
3e51f33f 653{
305e6835
VP
654 if (!rq->skip_clock_update) {
655 int cpu = cpu_of(rq);
656 u64 irq_time;
657
658 rq->clock = sched_clock_cpu(cpu);
659 irq_time = irq_time_cpu(cpu);
660 if (rq->clock - irq_time > rq->clock_task)
661 rq->clock_task = rq->clock - irq_time;
aa483808
VP
662
663 sched_irq_time_avg_update(rq, irq_time);
305e6835 664 }
3e51f33f
PZ
665}
666
bf5c91ba
IM
667/*
668 * Tunables that become constants when CONFIG_SCHED_DEBUG is off:
669 */
670#ifdef CONFIG_SCHED_DEBUG
671# define const_debug __read_mostly
672#else
673# define const_debug static const
674#endif
675
017730c1
IM
676/**
677 * runqueue_is_locked
e17b38bf 678 * @cpu: the processor in question.
017730c1
IM
679 *
680 * Returns true if the current cpu runqueue is locked.
681 * This interface allows printk to be called with the runqueue lock
682 * held and know whether or not it is OK to wake up the klogd.
683 */
89f19f04 684int runqueue_is_locked(int cpu)
017730c1 685{
05fa785c 686 return raw_spin_is_locked(&cpu_rq(cpu)->lock);
017730c1
IM
687}
688
bf5c91ba
IM
689/*
690 * Debugging: various feature bits
691 */
f00b45c1
PZ
692
693#define SCHED_FEAT(name, enabled) \
694 __SCHED_FEAT_##name ,
695
bf5c91ba 696enum {
f00b45c1 697#include "sched_features.h"
bf5c91ba
IM
698};
699
f00b45c1
PZ
700#undef SCHED_FEAT
701
702#define SCHED_FEAT(name, enabled) \
703 (1UL << __SCHED_FEAT_##name) * enabled |
704
bf5c91ba 705const_debug unsigned int sysctl_sched_features =
f00b45c1
PZ
706#include "sched_features.h"
707 0;
708
709#undef SCHED_FEAT
710
711#ifdef CONFIG_SCHED_DEBUG
712#define SCHED_FEAT(name, enabled) \
713 #name ,
714
983ed7a6 715static __read_mostly char *sched_feat_names[] = {
f00b45c1
PZ
716#include "sched_features.h"
717 NULL
718};
719
720#undef SCHED_FEAT
721
34f3a814 722static int sched_feat_show(struct seq_file *m, void *v)
f00b45c1 723{
f00b45c1
PZ
724 int i;
725
726 for (i = 0; sched_feat_names[i]; i++) {
34f3a814
LZ
727 if (!(sysctl_sched_features & (1UL << i)))
728 seq_puts(m, "NO_");
729 seq_printf(m, "%s ", sched_feat_names[i]);
f00b45c1 730 }
34f3a814 731 seq_puts(m, "\n");
f00b45c1 732
34f3a814 733 return 0;
f00b45c1
PZ
734}
735
736static ssize_t
737sched_feat_write(struct file *filp, const char __user *ubuf,
738 size_t cnt, loff_t *ppos)
739{
740 char buf[64];
7740191c 741 char *cmp;
f00b45c1
PZ
742 int neg = 0;
743 int i;
744
745 if (cnt > 63)
746 cnt = 63;
747
748 if (copy_from_user(&buf, ubuf, cnt))
749 return -EFAULT;
750
751 buf[cnt] = 0;
7740191c 752 cmp = strstrip(buf);
f00b45c1 753
c24b7c52 754 if (strncmp(buf, "NO_", 3) == 0) {
f00b45c1
PZ
755 neg = 1;
756 cmp += 3;
757 }
758
759 for (i = 0; sched_feat_names[i]; i++) {
7740191c 760 if (strcmp(cmp, sched_feat_names[i]) == 0) {
f00b45c1
PZ
761 if (neg)
762 sysctl_sched_features &= ~(1UL << i);
763 else
764 sysctl_sched_features |= (1UL << i);
765 break;
766 }
767 }
768
769 if (!sched_feat_names[i])
770 return -EINVAL;
771
42994724 772 *ppos += cnt;
f00b45c1
PZ
773
774 return cnt;
775}
776
34f3a814
LZ
777static int sched_feat_open(struct inode *inode, struct file *filp)
778{
779 return single_open(filp, sched_feat_show, NULL);
780}
781
828c0950 782static const struct file_operations sched_feat_fops = {
34f3a814
LZ
783 .open = sched_feat_open,
784 .write = sched_feat_write,
785 .read = seq_read,
786 .llseek = seq_lseek,
787 .release = single_release,
f00b45c1
PZ
788};
789
790static __init int sched_init_debug(void)
791{
f00b45c1
PZ
792 debugfs_create_file("sched_features", 0644, NULL, NULL,
793 &sched_feat_fops);
794
795 return 0;
796}
797late_initcall(sched_init_debug);
798
799#endif
800
801#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
bf5c91ba 802
b82d9fdd
PZ
803/*
804 * Number of tasks to iterate in a single balance run.
805 * Limited because this is done with IRQs disabled.
806 */
807const_debug unsigned int sysctl_sched_nr_migrate = 32;
808
2398f2c6
PZ
809/*
810 * ratelimit for updating the group shares.
55cd5340 811 * default: 0.25ms
2398f2c6 812 */
55cd5340 813unsigned int sysctl_sched_shares_ratelimit = 250000;
0bcdcf28 814unsigned int normalized_sysctl_sched_shares_ratelimit = 250000;
2398f2c6 815
ffda12a1
PZ
816/*
817 * Inject some fuzzyness into changing the per-cpu group shares
818 * this avoids remote rq-locks at the expense of fairness.
819 * default: 4
820 */
821unsigned int sysctl_sched_shares_thresh = 4;
822
e9e9250b
PZ
823/*
824 * period over which we average the RT time consumption, measured
825 * in ms.
826 *
827 * default: 1s
828 */
829const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;
830
fa85ae24 831/*
9f0c1e56 832 * period over which we measure -rt task cpu usage in us.
fa85ae24
PZ
833 * default: 1s
834 */
9f0c1e56 835unsigned int sysctl_sched_rt_period = 1000000;
fa85ae24 836
6892b75e
IM
837static __read_mostly int scheduler_running;
838
9f0c1e56
PZ
839/*
840 * part of the period that we allow rt tasks to run in us.
841 * default: 0.95s
842 */
843int sysctl_sched_rt_runtime = 950000;
fa85ae24 844
d0b27fa7
PZ
845static inline u64 global_rt_period(void)
846{
847 return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
848}
849
850static inline u64 global_rt_runtime(void)
851{
e26873bb 852 if (sysctl_sched_rt_runtime < 0)
d0b27fa7
PZ
853 return RUNTIME_INF;
854
855 return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
856}
fa85ae24 857
1da177e4 858#ifndef prepare_arch_switch
4866cde0
NP
859# define prepare_arch_switch(next) do { } while (0)
860#endif
861#ifndef finish_arch_switch
862# define finish_arch_switch(prev) do { } while (0)
863#endif
864
051a1d1a
DA
865static inline int task_current(struct rq *rq, struct task_struct *p)
866{
867 return rq->curr == p;
868}
869
4866cde0 870#ifndef __ARCH_WANT_UNLOCKED_CTXSW
70b97a7f 871static inline int task_running(struct rq *rq, struct task_struct *p)
4866cde0 872{
051a1d1a 873 return task_current(rq, p);
4866cde0
NP
874}
875
70b97a7f 876static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
4866cde0
NP
877{
878}
879
70b97a7f 880static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
4866cde0 881{
da04c035
IM
882#ifdef CONFIG_DEBUG_SPINLOCK
883 /* this is a valid case when another task releases the spinlock */
884 rq->lock.owner = current;
885#endif
8a25d5de
IM
886 /*
887 * If we are tracking spinlock dependencies then we have to
888 * fix up the runqueue lock - which gets 'carried over' from
889 * prev into current:
890 */
891 spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
892
05fa785c 893 raw_spin_unlock_irq(&rq->lock);
4866cde0
NP
894}
895
896#else /* __ARCH_WANT_UNLOCKED_CTXSW */
70b97a7f 897static inline int task_running(struct rq *rq, struct task_struct *p)
4866cde0
NP
898{
899#ifdef CONFIG_SMP
900 return p->oncpu;
901#else
051a1d1a 902 return task_current(rq, p);
4866cde0
NP
903#endif
904}
905
70b97a7f 906static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
4866cde0
NP
907{
908#ifdef CONFIG_SMP
909 /*
910 * We can optimise this out completely for !SMP, because the
911 * SMP rebalancing from interrupt is the only thing that cares
912 * here.
913 */
914 next->oncpu = 1;
915#endif
916#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
05fa785c 917 raw_spin_unlock_irq(&rq->lock);
4866cde0 918#else
05fa785c 919 raw_spin_unlock(&rq->lock);
4866cde0
NP
920#endif
921}
922
70b97a7f 923static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
4866cde0
NP
924{
925#ifdef CONFIG_SMP
926 /*
927 * After ->oncpu is cleared, the task can be moved to a different CPU.
928 * We must ensure this doesn't happen until the switch is completely
929 * finished.
930 */
931 smp_wmb();
932 prev->oncpu = 0;
933#endif
934#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
935 local_irq_enable();
1da177e4 936#endif
4866cde0
NP
937}
938#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
1da177e4 939
0970d299 940/*
65cc8e48
PZ
941 * Check whether the task is waking, we use this to synchronize ->cpus_allowed
942 * against ttwu().
0970d299
PZ
943 */
944static inline int task_is_waking(struct task_struct *p)
945{
0017d735 946 return unlikely(p->state == TASK_WAKING);
0970d299
PZ
947}
948
b29739f9
IM
949/*
950 * __task_rq_lock - lock the runqueue a given task resides on.
951 * Must be called interrupts disabled.
952 */
70b97a7f 953static inline struct rq *__task_rq_lock(struct task_struct *p)
b29739f9
IM
954 __acquires(rq->lock)
955{
0970d299
PZ
956 struct rq *rq;
957
3a5c359a 958 for (;;) {
0970d299 959 rq = task_rq(p);
05fa785c 960 raw_spin_lock(&rq->lock);
65cc8e48 961 if (likely(rq == task_rq(p)))
3a5c359a 962 return rq;
05fa785c 963 raw_spin_unlock(&rq->lock);
b29739f9 964 }
b29739f9
IM
965}
966
1da177e4
LT
967/*
968 * task_rq_lock - lock the runqueue a given task resides on and disable
41a2d6cf 969 * interrupts. Note the ordering: we can safely lookup the task_rq without
1da177e4
LT
970 * explicitly disabling preemption.
971 */
70b97a7f 972static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
1da177e4
LT
973 __acquires(rq->lock)
974{
70b97a7f 975 struct rq *rq;
1da177e4 976
3a5c359a
AK
977 for (;;) {
978 local_irq_save(*flags);
979 rq = task_rq(p);
05fa785c 980 raw_spin_lock(&rq->lock);
65cc8e48 981 if (likely(rq == task_rq(p)))
3a5c359a 982 return rq;
05fa785c 983 raw_spin_unlock_irqrestore(&rq->lock, *flags);
1da177e4 984 }
1da177e4
LT
985}
986
a9957449 987static void __task_rq_unlock(struct rq *rq)
b29739f9
IM
988 __releases(rq->lock)
989{
05fa785c 990 raw_spin_unlock(&rq->lock);
b29739f9
IM
991}
992
70b97a7f 993static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
1da177e4
LT
994 __releases(rq->lock)
995{
05fa785c 996 raw_spin_unlock_irqrestore(&rq->lock, *flags);
1da177e4
LT
997}
998
1da177e4 999/*
cc2a73b5 1000 * this_rq_lock - lock this runqueue and disable interrupts.
1da177e4 1001 */
a9957449 1002static struct rq *this_rq_lock(void)
1da177e4
LT
1003 __acquires(rq->lock)
1004{
70b97a7f 1005 struct rq *rq;
1da177e4
LT
1006
1007 local_irq_disable();
1008 rq = this_rq();
05fa785c 1009 raw_spin_lock(&rq->lock);
1da177e4
LT
1010
1011 return rq;
1012}
1013
8f4d37ec
PZ
1014#ifdef CONFIG_SCHED_HRTICK
1015/*
1016 * Use HR-timers to deliver accurate preemption points.
1017 *
1018 * Its all a bit involved since we cannot program an hrt while holding the
1019 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
1020 * reschedule event.
1021 *
1022 * When we get rescheduled we reprogram the hrtick_timer outside of the
1023 * rq->lock.
1024 */
8f4d37ec
PZ
1025
1026/*
1027 * Use hrtick when:
1028 * - enabled by features
1029 * - hrtimer is actually high res
1030 */
1031static inline int hrtick_enabled(struct rq *rq)
1032{
1033 if (!sched_feat(HRTICK))
1034 return 0;
ba42059f 1035 if (!cpu_active(cpu_of(rq)))
b328ca18 1036 return 0;
8f4d37ec
PZ
1037 return hrtimer_is_hres_active(&rq->hrtick_timer);
1038}
1039
8f4d37ec
PZ
1040static void hrtick_clear(struct rq *rq)
1041{
1042 if (hrtimer_active(&rq->hrtick_timer))
1043 hrtimer_cancel(&rq->hrtick_timer);
1044}
1045
8f4d37ec
PZ
1046/*
1047 * High-resolution timer tick.
1048 * Runs from hardirq context with interrupts disabled.
1049 */
1050static enum hrtimer_restart hrtick(struct hrtimer *timer)
1051{
1052 struct rq *rq = container_of(timer, struct rq, hrtick_timer);
1053
1054 WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());
1055
05fa785c 1056 raw_spin_lock(&rq->lock);
3e51f33f 1057 update_rq_clock(rq);
8f4d37ec 1058 rq->curr->sched_class->task_tick(rq, rq->curr, 1);
05fa785c 1059 raw_spin_unlock(&rq->lock);
8f4d37ec
PZ
1060
1061 return HRTIMER_NORESTART;
1062}
1063
95e904c7 1064#ifdef CONFIG_SMP
31656519
PZ
1065/*
1066 * called from hardirq (IPI) context
1067 */
1068static void __hrtick_start(void *arg)
b328ca18 1069{
31656519 1070 struct rq *rq = arg;
b328ca18 1071
05fa785c 1072 raw_spin_lock(&rq->lock);
31656519
PZ
1073 hrtimer_restart(&rq->hrtick_timer);
1074 rq->hrtick_csd_pending = 0;
05fa785c 1075 raw_spin_unlock(&rq->lock);
b328ca18
PZ
1076}
1077
31656519
PZ
1078/*
1079 * Called to set the hrtick timer state.
1080 *
1081 * called with rq->lock held and irqs disabled
1082 */
1083static void hrtick_start(struct rq *rq, u64 delay)
b328ca18 1084{
31656519
PZ
1085 struct hrtimer *timer = &rq->hrtick_timer;
1086 ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
b328ca18 1087
cc584b21 1088 hrtimer_set_expires(timer, time);
31656519
PZ
1089
1090 if (rq == this_rq()) {
1091 hrtimer_restart(timer);
1092 } else if (!rq->hrtick_csd_pending) {
6e275637 1093 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
31656519
PZ
1094 rq->hrtick_csd_pending = 1;
1095 }
b328ca18
PZ
1096}
1097
1098static int
1099hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
1100{
1101 int cpu = (int)(long)hcpu;
1102
1103 switch (action) {
1104 case CPU_UP_CANCELED:
1105 case CPU_UP_CANCELED_FROZEN:
1106 case CPU_DOWN_PREPARE:
1107 case CPU_DOWN_PREPARE_FROZEN:
1108 case CPU_DEAD:
1109 case CPU_DEAD_FROZEN:
31656519 1110 hrtick_clear(cpu_rq(cpu));
b328ca18
PZ
1111 return NOTIFY_OK;
1112 }
1113
1114 return NOTIFY_DONE;
1115}
1116
fa748203 1117static __init void init_hrtick(void)
b328ca18
PZ
1118{
1119 hotcpu_notifier(hotplug_hrtick, 0);
1120}
31656519
PZ
1121#else
1122/*
1123 * Called to set the hrtick timer state.
1124 *
1125 * called with rq->lock held and irqs disabled
1126 */
1127static void hrtick_start(struct rq *rq, u64 delay)
1128{
7f1e2ca9 1129 __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
5c333864 1130 HRTIMER_MODE_REL_PINNED, 0);
31656519 1131}
b328ca18 1132
006c75f1 1133static inline void init_hrtick(void)
8f4d37ec 1134{
8f4d37ec 1135}
31656519 1136#endif /* CONFIG_SMP */
8f4d37ec 1137
31656519 1138static void init_rq_hrtick(struct rq *rq)
8f4d37ec 1139{
31656519
PZ
1140#ifdef CONFIG_SMP
1141 rq->hrtick_csd_pending = 0;
8f4d37ec 1142
31656519
PZ
1143 rq->hrtick_csd.flags = 0;
1144 rq->hrtick_csd.func = __hrtick_start;
1145 rq->hrtick_csd.info = rq;
1146#endif
8f4d37ec 1147
31656519
PZ
1148 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1149 rq->hrtick_timer.function = hrtick;
8f4d37ec 1150}
006c75f1 1151#else /* CONFIG_SCHED_HRTICK */
8f4d37ec
PZ
1152static inline void hrtick_clear(struct rq *rq)
1153{
1154}
1155
8f4d37ec
PZ
1156static inline void init_rq_hrtick(struct rq *rq)
1157{
1158}
1159
b328ca18
PZ
1160static inline void init_hrtick(void)
1161{
1162}
006c75f1 1163#endif /* CONFIG_SCHED_HRTICK */
8f4d37ec 1164
c24d20db
IM
1165/*
1166 * resched_task - mark a task 'to be rescheduled now'.
1167 *
1168 * On UP this means the setting of the need_resched flag, on SMP it
1169 * might also involve a cross-CPU call to trigger the scheduler on
1170 * the target CPU.
1171 */
1172#ifdef CONFIG_SMP
1173
1174#ifndef tsk_is_polling
1175#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
1176#endif
1177
31656519 1178static void resched_task(struct task_struct *p)
c24d20db
IM
1179{
1180 int cpu;
1181
05fa785c 1182 assert_raw_spin_locked(&task_rq(p)->lock);
c24d20db 1183
5ed0cec0 1184 if (test_tsk_need_resched(p))
c24d20db
IM
1185 return;
1186
5ed0cec0 1187 set_tsk_need_resched(p);
c24d20db
IM
1188
1189 cpu = task_cpu(p);
1190 if (cpu == smp_processor_id())
1191 return;
1192
1193 /* NEED_RESCHED must be visible before we test polling */
1194 smp_mb();
1195 if (!tsk_is_polling(p))
1196 smp_send_reschedule(cpu);
1197}
1198
1199static void resched_cpu(int cpu)
1200{
1201 struct rq *rq = cpu_rq(cpu);
1202 unsigned long flags;
1203
05fa785c 1204 if (!raw_spin_trylock_irqsave(&rq->lock, flags))
c24d20db
IM
1205 return;
1206 resched_task(cpu_curr(cpu));
05fa785c 1207 raw_spin_unlock_irqrestore(&rq->lock, flags);
c24d20db 1208}
06d8308c
TG
1209
1210#ifdef CONFIG_NO_HZ
83cd4fe2
VP
1211/*
1212 * In the semi idle case, use the nearest busy cpu for migrating timers
1213 * from an idle cpu. This is good for power-savings.
1214 *
1215 * We don't do similar optimization for completely idle system, as
1216 * selecting an idle cpu will add more delays to the timers than intended
1217 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
1218 */
1219int get_nohz_timer_target(void)
1220{
1221 int cpu = smp_processor_id();
1222 int i;
1223 struct sched_domain *sd;
1224
1225 for_each_domain(cpu, sd) {
1226 for_each_cpu(i, sched_domain_span(sd))
1227 if (!idle_cpu(i))
1228 return i;
1229 }
1230 return cpu;
1231}
06d8308c
TG
1232/*
1233 * When add_timer_on() enqueues a timer into the timer wheel of an
1234 * idle CPU then this timer might expire before the next timer event
1235 * which is scheduled to wake up that CPU. In case of a completely
1236 * idle system the next event might even be infinite time into the
1237 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
1238 * leaves the inner idle loop so the newly added timer is taken into
1239 * account when the CPU goes back to idle and evaluates the timer
1240 * wheel for the next timer event.
1241 */
1242void wake_up_idle_cpu(int cpu)
1243{
1244 struct rq *rq = cpu_rq(cpu);
1245
1246 if (cpu == smp_processor_id())
1247 return;
1248
1249 /*
1250 * This is safe, as this function is called with the timer
1251 * wheel base lock of (cpu) held. When the CPU is on the way
1252 * to idle and has not yet set rq->curr to idle then it will
1253 * be serialized on the timer wheel base lock and take the new
1254 * timer into account automatically.
1255 */
1256 if (rq->curr != rq->idle)
1257 return;
1258
1259 /*
1260 * We can set TIF_RESCHED on the idle task of the other CPU
1261 * lockless. The worst case is that the other CPU runs the
1262 * idle task through an additional NOOP schedule()
1263 */
5ed0cec0 1264 set_tsk_need_resched(rq->idle);
06d8308c
TG
1265
1266 /* NEED_RESCHED must be visible before we test polling */
1267 smp_mb();
1268 if (!tsk_is_polling(rq->idle))
1269 smp_send_reschedule(cpu);
1270}
39c0cbe2 1271
6d6bc0ad 1272#endif /* CONFIG_NO_HZ */
06d8308c 1273
e9e9250b
PZ
1274static u64 sched_avg_period(void)
1275{
1276 return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2;
1277}
1278
1279static void sched_avg_update(struct rq *rq)
1280{
1281 s64 period = sched_avg_period();
1282
1283 while ((s64)(rq->clock - rq->age_stamp) > period) {
0d98bb26
WD
1284 /*
1285 * Inline assembly required to prevent the compiler
1286 * optimising this loop into a divmod call.
1287 * See __iter_div_u64_rem() for another example of this.
1288 */
1289 asm("" : "+rm" (rq->age_stamp));
e9e9250b
PZ
1290 rq->age_stamp += period;
1291 rq->rt_avg /= 2;
1292 }
1293}
1294
1295static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1296{
1297 rq->rt_avg += rt_delta;
1298 sched_avg_update(rq);
1299}
1300
6d6bc0ad 1301#else /* !CONFIG_SMP */
31656519 1302static void resched_task(struct task_struct *p)
c24d20db 1303{
05fa785c 1304 assert_raw_spin_locked(&task_rq(p)->lock);
31656519 1305 set_tsk_need_resched(p);
c24d20db 1306}
e9e9250b
PZ
1307
1308static void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
1309{
1310}
da2b71ed
SS
1311
1312static void sched_avg_update(struct rq *rq)
1313{
1314}
6d6bc0ad 1315#endif /* CONFIG_SMP */
c24d20db 1316
45bf76df
IM
1317#if BITS_PER_LONG == 32
1318# define WMULT_CONST (~0UL)
1319#else
1320# define WMULT_CONST (1UL << 32)
1321#endif
1322
1323#define WMULT_SHIFT 32
1324
194081eb
IM
1325/*
1326 * Shift right and round:
1327 */
cf2ab469 1328#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
194081eb 1329
a7be37ac
PZ
1330/*
1331 * delta *= weight / lw
1332 */
cb1c4fc9 1333static unsigned long
45bf76df
IM
1334calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1335 struct load_weight *lw)
1336{
1337 u64 tmp;
1338
7a232e03
LJ
1339 if (!lw->inv_weight) {
1340 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1341 lw->inv_weight = 1;
1342 else
1343 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1344 / (lw->weight+1);
1345 }
45bf76df
IM
1346
1347 tmp = (u64)delta_exec * weight;
1348 /*
1349 * Check whether we'd overflow the 64-bit multiplication:
1350 */
194081eb 1351 if (unlikely(tmp > WMULT_CONST))
cf2ab469 1352 tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
194081eb
IM
1353 WMULT_SHIFT/2);
1354 else
cf2ab469 1355 tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);
45bf76df 1356
ecf691da 1357 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
45bf76df
IM
1358}
1359
1091985b 1360static inline void update_load_add(struct load_weight *lw, unsigned long inc)
45bf76df
IM
1361{
1362 lw->weight += inc;
e89996ae 1363 lw->inv_weight = 0;
45bf76df
IM
1364}
1365
1091985b 1366static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
45bf76df
IM
1367{
1368 lw->weight -= dec;
e89996ae 1369 lw->inv_weight = 0;
45bf76df
IM
1370}
1371
2dd73a4f
PW
1372/*
1373 * To aid in avoiding the subversion of "niceness" due to uneven distribution
1374 * of tasks with abnormal "nice" values across CPUs the contribution that
1375 * each task makes to its run queue's load is weighted according to its
41a2d6cf 1376 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
2dd73a4f
PW
1377 * scaled version of the new time slice allocation that they receive on time
1378 * slice expiry etc.
1379 */
1380
cce7ade8
PZ
1381#define WEIGHT_IDLEPRIO 3
1382#define WMULT_IDLEPRIO 1431655765
dd41f596
IM
1383
1384/*
1385 * Nice levels are multiplicative, with a gentle 10% change for every
1386 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
1387 * nice 1, it will get ~10% less CPU time than another CPU-bound task
1388 * that remained on nice 0.
1389 *
1390 * The "10% effect" is relative and cumulative: from _any_ nice level,
1391 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
f9153ee6
IM
1392 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
1393 * If a task goes up by ~10% and another task goes down by ~10% then
1394 * the relative distance between them is ~25%.)
dd41f596
IM
1395 */
1396static const int prio_to_weight[40] = {
254753dc
IM
1397 /* -20 */ 88761, 71755, 56483, 46273, 36291,
1398 /* -15 */ 29154, 23254, 18705, 14949, 11916,
1399 /* -10 */ 9548, 7620, 6100, 4904, 3906,
1400 /* -5 */ 3121, 2501, 1991, 1586, 1277,
1401 /* 0 */ 1024, 820, 655, 526, 423,
1402 /* 5 */ 335, 272, 215, 172, 137,
1403 /* 10 */ 110, 87, 70, 56, 45,
1404 /* 15 */ 36, 29, 23, 18, 15,
dd41f596
IM
1405};
1406
5714d2de
IM
1407/*
1408 * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
1409 *
1410 * In cases where the weight does not change often, we can use the
1411 * precalculated inverse to speed up arithmetics by turning divisions
1412 * into multiplications:
1413 */
dd41f596 1414static const u32 prio_to_wmult[40] = {
254753dc
IM
1415 /* -20 */ 48388, 59856, 76040, 92818, 118348,
1416 /* -15 */ 147320, 184698, 229616, 287308, 360437,
1417 /* -10 */ 449829, 563644, 704093, 875809, 1099582,
1418 /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326,
1419 /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587,
1420 /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126,
1421 /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717,
1422 /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
dd41f596 1423};
2dd73a4f 1424
ef12fefa
BR
1425/* Time spent by the tasks of the cpu accounting group executing in ... */
1426enum cpuacct_stat_index {
1427 CPUACCT_STAT_USER, /* ... user mode */
1428 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
1429
1430 CPUACCT_STAT_NSTATS,
1431};
1432
d842de87
SV
1433#ifdef CONFIG_CGROUP_CPUACCT
1434static void cpuacct_charge(struct task_struct *tsk, u64 cputime);
ef12fefa
BR
1435static void cpuacct_update_stats(struct task_struct *tsk,
1436 enum cpuacct_stat_index idx, cputime_t val);
d842de87
SV
1437#else
1438static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
ef12fefa
BR
1439static inline void cpuacct_update_stats(struct task_struct *tsk,
1440 enum cpuacct_stat_index idx, cputime_t val) {}
d842de87
SV
1441#endif
1442
18d95a28
PZ
1443static inline void inc_cpu_load(struct rq *rq, unsigned long load)
1444{
1445 update_load_add(&rq->load, load);
1446}
1447
1448static inline void dec_cpu_load(struct rq *rq, unsigned long load)
1449{
1450 update_load_sub(&rq->load, load);
1451}
1452
7940ca36 1453#if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED)
eb755805 1454typedef int (*tg_visitor)(struct task_group *, void *);
c09595f6
PZ
1455
1456/*
1457 * Iterate the full tree, calling @down when first entering a node and @up when
1458 * leaving it for the final time.
1459 */
eb755805 1460static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data)
c09595f6
PZ
1461{
1462 struct task_group *parent, *child;
eb755805 1463 int ret;
c09595f6
PZ
1464
1465 rcu_read_lock();
1466 parent = &root_task_group;
1467down:
eb755805
PZ
1468 ret = (*down)(parent, data);
1469 if (ret)
1470 goto out_unlock;
c09595f6
PZ
1471 list_for_each_entry_rcu(child, &parent->children, siblings) {
1472 parent = child;
1473 goto down;
1474
1475up:
1476 continue;
1477 }
eb755805
PZ
1478 ret = (*up)(parent, data);
1479 if (ret)
1480 goto out_unlock;
c09595f6
PZ
1481
1482 child = parent;
1483 parent = parent->parent;
1484 if (parent)
1485 goto up;
eb755805 1486out_unlock:
c09595f6 1487 rcu_read_unlock();
eb755805
PZ
1488
1489 return ret;
c09595f6
PZ
1490}
1491
eb755805
PZ
1492static int tg_nop(struct task_group *tg, void *data)
1493{
1494 return 0;
c09595f6 1495}
eb755805
PZ
1496#endif
1497
1498#ifdef CONFIG_SMP
f5f08f39
PZ
1499/* Used instead of source_load when we know the type == 0 */
1500static unsigned long weighted_cpuload(const int cpu)
1501{
1502 return cpu_rq(cpu)->load.weight;
1503}
1504
1505/*
1506 * Return a low guess at the load of a migration-source cpu weighted
1507 * according to the scheduling class and "nice" value.
1508 *
1509 * We want to under-estimate the load of migration sources, to
1510 * balance conservatively.
1511 */
1512static unsigned long source_load(int cpu, int type)
1513{
1514 struct rq *rq = cpu_rq(cpu);
1515 unsigned long total = weighted_cpuload(cpu);
1516
1517 if (type == 0 || !sched_feat(LB_BIAS))
1518 return total;
1519
1520 return min(rq->cpu_load[type-1], total);
1521}
1522
1523/*
1524 * Return a high guess at the load of a migration-target cpu weighted
1525 * according to the scheduling class and "nice" value.
1526 */
1527static unsigned long target_load(int cpu, int type)
1528{
1529 struct rq *rq = cpu_rq(cpu);
1530 unsigned long total = weighted_cpuload(cpu);
1531
1532 if (type == 0 || !sched_feat(LB_BIAS))
1533 return total;
1534
1535 return max(rq->cpu_load[type-1], total);
1536}
1537
ae154be1
PZ
1538static unsigned long power_of(int cpu)
1539{
e51fd5e2 1540 return cpu_rq(cpu)->cpu_power;
ae154be1
PZ
1541}
1542
eb755805
PZ
1543static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1544
1545static unsigned long cpu_avg_load_per_task(int cpu)
1546{
1547 struct rq *rq = cpu_rq(cpu);
af6d596f 1548 unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
eb755805 1549
4cd42620
SR
1550 if (nr_running)
1551 rq->avg_load_per_task = rq->load.weight / nr_running;
a2d47777
BS
1552 else
1553 rq->avg_load_per_task = 0;
eb755805
PZ
1554
1555 return rq->avg_load_per_task;
1556}
1557
1558#ifdef CONFIG_FAIR_GROUP_SCHED
c09595f6 1559
43cf38eb 1560static __read_mostly unsigned long __percpu *update_shares_data;
34d76c41 1561
c09595f6
PZ
1562static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1563
1564/*
1565 * Calculate and set the cpu's group shares.
1566 */
34d76c41
PZ
1567static void update_group_shares_cpu(struct task_group *tg, int cpu,
1568 unsigned long sd_shares,
1569 unsigned long sd_rq_weight,
4a6cc4bd 1570 unsigned long *usd_rq_weight)
18d95a28 1571{
34d76c41 1572 unsigned long shares, rq_weight;
a5004278 1573 int boost = 0;
c09595f6 1574
4a6cc4bd 1575 rq_weight = usd_rq_weight[cpu];
a5004278
PZ
1576 if (!rq_weight) {
1577 boost = 1;
1578 rq_weight = NICE_0_LOAD;
1579 }
c8cba857 1580
c09595f6 1581 /*
a8af7246
PZ
1582 * \Sum_j shares_j * rq_weight_i
1583 * shares_i = -----------------------------
1584 * \Sum_j rq_weight_j
c09595f6 1585 */
ec4e0e2f 1586 shares = (sd_shares * rq_weight) / sd_rq_weight;
ffda12a1 1587 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
c09595f6 1588
ffda12a1
PZ
1589 if (abs(shares - tg->se[cpu]->load.weight) >
1590 sysctl_sched_shares_thresh) {
1591 struct rq *rq = cpu_rq(cpu);
1592 unsigned long flags;
c09595f6 1593
05fa785c 1594 raw_spin_lock_irqsave(&rq->lock, flags);
34d76c41 1595 tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight;
a5004278 1596 tg->cfs_rq[cpu]->shares = boost ? 0 : shares;
ffda12a1 1597 __set_se_shares(tg->se[cpu], shares);
05fa785c 1598 raw_spin_unlock_irqrestore(&rq->lock, flags);
ffda12a1 1599 }
18d95a28 1600}
c09595f6
PZ
1601
1602/*
c8cba857
PZ
1603 * Re-compute the task group their per cpu shares over the given domain.
1604 * This needs to be done in a bottom-up fashion because the rq weight of a
1605 * parent group depends on the shares of its child groups.
c09595f6 1606 */
eb755805 1607static int tg_shares_up(struct task_group *tg, void *data)
c09595f6 1608{
cd8ad40d 1609 unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0;
4a6cc4bd 1610 unsigned long *usd_rq_weight;
eb755805 1611 struct sched_domain *sd = data;
34d76c41 1612 unsigned long flags;
c8cba857 1613 int i;
c09595f6 1614
34d76c41
PZ
1615 if (!tg->se[0])
1616 return 0;
1617
1618 local_irq_save(flags);
4a6cc4bd 1619 usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id());
34d76c41 1620
758b2cdc 1621 for_each_cpu(i, sched_domain_span(sd)) {
34d76c41 1622 weight = tg->cfs_rq[i]->load.weight;
4a6cc4bd 1623 usd_rq_weight[i] = weight;
34d76c41 1624
cd8ad40d 1625 rq_weight += weight;
ec4e0e2f
KC
1626 /*
1627 * If there are currently no tasks on the cpu pretend there
1628 * is one of average load so that when a new task gets to
1629 * run here it will not get delayed by group starvation.
1630 */
ec4e0e2f
KC
1631 if (!weight)
1632 weight = NICE_0_LOAD;
1633
cd8ad40d 1634 sum_weight += weight;
c8cba857 1635 shares += tg->cfs_rq[i]->shares;
c09595f6 1636 }
c09595f6 1637
cd8ad40d
PZ
1638 if (!rq_weight)
1639 rq_weight = sum_weight;
1640
c8cba857
PZ
1641 if ((!shares && rq_weight) || shares > tg->shares)
1642 shares = tg->shares;
1643
1644 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1645 shares = tg->shares;
c09595f6 1646
758b2cdc 1647 for_each_cpu(i, sched_domain_span(sd))
4a6cc4bd 1648 update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight);
34d76c41
PZ
1649
1650 local_irq_restore(flags);
eb755805
PZ
1651
1652 return 0;
c09595f6
PZ
1653}
1654
1655/*
c8cba857
PZ
1656 * Compute the cpu's hierarchical load factor for each task group.
1657 * This needs to be done in a top-down fashion because the load of a child
1658 * group is a fraction of its parents load.
c09595f6 1659 */
eb755805 1660static int tg_load_down(struct task_group *tg, void *data)
c09595f6 1661{
c8cba857 1662 unsigned long load;
eb755805 1663 long cpu = (long)data;
c09595f6 1664
c8cba857
PZ
1665 if (!tg->parent) {
1666 load = cpu_rq(cpu)->load.weight;
1667 } else {
1668 load = tg->parent->cfs_rq[cpu]->h_load;
1669 load *= tg->cfs_rq[cpu]->shares;
1670 load /= tg->parent->cfs_rq[cpu]->load.weight + 1;
1671 }
c09595f6 1672
c8cba857 1673 tg->cfs_rq[cpu]->h_load = load;
c09595f6 1674
eb755805 1675 return 0;
c09595f6
PZ
1676}
1677
c8cba857 1678static void update_shares(struct sched_domain *sd)
4d8d595d 1679{
e7097159
PZ
1680 s64 elapsed;
1681 u64 now;
1682
1683 if (root_task_group_empty())
1684 return;
1685
c676329a 1686 now = local_clock();
e7097159 1687 elapsed = now - sd->last_update;
2398f2c6
PZ
1688
1689 if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) {
1690 sd->last_update = now;
eb755805 1691 walk_tg_tree(tg_nop, tg_shares_up, sd);
2398f2c6 1692 }
4d8d595d
PZ
1693}
1694
eb755805 1695static void update_h_load(long cpu)
c09595f6 1696{
eb755805 1697 walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
c09595f6
PZ
1698}
1699
c09595f6
PZ
1700#else
1701
c8cba857 1702static inline void update_shares(struct sched_domain *sd)
4d8d595d
PZ
1703{
1704}
1705
18d95a28
PZ
1706#endif
1707
8f45e2b5
GH
1708#ifdef CONFIG_PREEMPT
1709
b78bb868
PZ
1710static void double_rq_lock(struct rq *rq1, struct rq *rq2);
1711
70574a99 1712/*
8f45e2b5
GH
1713 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1714 * way at the expense of forcing extra atomic operations in all
1715 * invocations. This assures that the double_lock is acquired using the
1716 * same underlying policy as the spinlock_t on this architecture, which
1717 * reduces latency compared to the unfair variant below. However, it
1718 * also adds more overhead and therefore may reduce throughput.
70574a99 1719 */
8f45e2b5
GH
1720static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1721 __releases(this_rq->lock)
1722 __acquires(busiest->lock)
1723 __acquires(this_rq->lock)
1724{
05fa785c 1725 raw_spin_unlock(&this_rq->lock);
8f45e2b5
GH
1726 double_rq_lock(this_rq, busiest);
1727
1728 return 1;
1729}
1730
1731#else
1732/*
1733 * Unfair double_lock_balance: Optimizes throughput at the expense of
1734 * latency by eliminating extra atomic operations when the locks are
1735 * already in proper order on entry. This favors lower cpu-ids and will
1736 * grant the double lock to lower cpus over higher ids under contention,
1737 * regardless of entry order into the function.
1738 */
1739static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
70574a99
AD
1740 __releases(this_rq->lock)
1741 __acquires(busiest->lock)
1742 __acquires(this_rq->lock)
1743{
1744 int ret = 0;
1745
05fa785c 1746 if (unlikely(!raw_spin_trylock(&busiest->lock))) {
70574a99 1747 if (busiest < this_rq) {
05fa785c
TG
1748 raw_spin_unlock(&this_rq->lock);
1749 raw_spin_lock(&busiest->lock);
1750 raw_spin_lock_nested(&this_rq->lock,
1751 SINGLE_DEPTH_NESTING);
70574a99
AD
1752 ret = 1;
1753 } else
05fa785c
TG
1754 raw_spin_lock_nested(&busiest->lock,
1755 SINGLE_DEPTH_NESTING);
70574a99
AD
1756 }
1757 return ret;
1758}
1759
8f45e2b5
GH
1760#endif /* CONFIG_PREEMPT */
1761
1762/*
1763 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1764 */
1765static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1766{
1767 if (unlikely(!irqs_disabled())) {
1768 /* printk() doesn't work good under rq->lock */
05fa785c 1769 raw_spin_unlock(&this_rq->lock);
8f45e2b5
GH
1770 BUG_ON(1);
1771 }
1772
1773 return _double_lock_balance(this_rq, busiest);
1774}
1775
70574a99
AD
1776static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1777 __releases(busiest->lock)
1778{
05fa785c 1779 raw_spin_unlock(&busiest->lock);
70574a99
AD
1780 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1781}
1e3c88bd
PZ
1782
1783/*
1784 * double_rq_lock - safely lock two runqueues
1785 *
1786 * Note this does not disable interrupts like task_rq_lock,
1787 * you need to do so manually before calling.
1788 */
1789static void double_rq_lock(struct rq *rq1, struct rq *rq2)
1790 __acquires(rq1->lock)
1791 __acquires(rq2->lock)
1792{
1793 BUG_ON(!irqs_disabled());
1794 if (rq1 == rq2) {
1795 raw_spin_lock(&rq1->lock);
1796 __acquire(rq2->lock); /* Fake it out ;) */
1797 } else {
1798 if (rq1 < rq2) {
1799 raw_spin_lock(&rq1->lock);
1800 raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING);
1801 } else {
1802 raw_spin_lock(&rq2->lock);
1803 raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING);
1804 }
1805 }
1e3c88bd
PZ
1806}
1807
1808/*
1809 * double_rq_unlock - safely unlock two runqueues
1810 *
1811 * Note this does not restore interrupts like task_rq_unlock,
1812 * you need to do so manually after calling.
1813 */
1814static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
1815 __releases(rq1->lock)
1816 __releases(rq2->lock)
1817{
1818 raw_spin_unlock(&rq1->lock);
1819 if (rq1 != rq2)
1820 raw_spin_unlock(&rq2->lock);
1821 else
1822 __release(rq2->lock);
1823}
1824
18d95a28
PZ
1825#endif
1826
30432094 1827#ifdef CONFIG_FAIR_GROUP_SCHED
34e83e85
IM
1828static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1829{
30432094 1830#ifdef CONFIG_SMP
34e83e85
IM
1831 cfs_rq->shares = shares;
1832#endif
1833}
30432094 1834#endif
e7693a36 1835
74f5187a 1836static void calc_load_account_idle(struct rq *this_rq);
0bcdcf28 1837static void update_sysctl(void);
acb4a848 1838static int get_update_sysctl_factor(void);
fdf3e95d 1839static void update_cpu_load(struct rq *this_rq);
dce48a84 1840
cd29fe6f
PZ
1841static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1842{
1843 set_task_rq(p, cpu);
1844#ifdef CONFIG_SMP
1845 /*
1846 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
1847 * successfuly executed on another CPU. We must ensure that updates of
1848 * per-task data have been completed by this moment.
1849 */
1850 smp_wmb();
1851 task_thread_info(p)->cpu = cpu;
1852#endif
1853}
dce48a84 1854
1e3c88bd 1855static const struct sched_class rt_sched_class;
dd41f596 1856
34f971f6 1857#define sched_class_highest (&stop_sched_class)
1f11eb6a
GH
1858#define for_each_class(class) \
1859 for (class = sched_class_highest; class; class = class->next)
dd41f596 1860
1e3c88bd
PZ
1861#include "sched_stats.h"
1862
c09595f6 1863static void inc_nr_running(struct rq *rq)
9c217245
IM
1864{
1865 rq->nr_running++;
9c217245
IM
1866}
1867
c09595f6 1868static void dec_nr_running(struct rq *rq)
9c217245
IM
1869{
1870 rq->nr_running--;
9c217245
IM
1871}
1872
45bf76df
IM
1873static void set_load_weight(struct task_struct *p)
1874{
dd41f596
IM
1875 /*
1876 * SCHED_IDLE tasks get minimal weight:
1877 */
1878 if (p->policy == SCHED_IDLE) {
1879 p->se.load.weight = WEIGHT_IDLEPRIO;
1880 p->se.load.inv_weight = WMULT_IDLEPRIO;
1881 return;
1882 }
71f8bd46 1883
dd41f596
IM
1884 p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
1885 p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
71f8bd46
IM
1886}
1887
371fd7e7 1888static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
2087a1ad 1889{
a64692a3 1890 update_rq_clock(rq);
dd41f596 1891 sched_info_queued(p);
371fd7e7 1892 p->sched_class->enqueue_task(rq, p, flags);
dd41f596 1893 p->se.on_rq = 1;
71f8bd46
IM
1894}
1895
371fd7e7 1896static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
71f8bd46 1897{
a64692a3 1898 update_rq_clock(rq);
46ac22ba 1899 sched_info_dequeued(p);
371fd7e7 1900 p->sched_class->dequeue_task(rq, p, flags);
dd41f596 1901 p->se.on_rq = 0;
71f8bd46
IM
1902}
1903
1e3c88bd
PZ
1904/*
1905 * activate_task - move a task to the runqueue.
1906 */
371fd7e7 1907static void activate_task(struct rq *rq, struct task_struct *p, int flags)
1e3c88bd
PZ
1908{
1909 if (task_contributes_to_load(p))
1910 rq->nr_uninterruptible--;
1911
371fd7e7 1912 enqueue_task(rq, p, flags);
1e3c88bd
PZ
1913 inc_nr_running(rq);
1914}
1915
1916/*
1917 * deactivate_task - remove a task from the runqueue.
1918 */
371fd7e7 1919static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1e3c88bd
PZ
1920{
1921 if (task_contributes_to_load(p))
1922 rq->nr_uninterruptible++;
1923
371fd7e7 1924 dequeue_task(rq, p, flags);
1e3c88bd
PZ
1925 dec_nr_running(rq);
1926}
1927
b52bfee4
VP
1928#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1929
305e6835
VP
1930/*
1931 * There are no locks covering percpu hardirq/softirq time.
1932 * They are only modified in account_system_vtime, on corresponding CPU
1933 * with interrupts disabled. So, writes are safe.
1934 * They are read and saved off onto struct rq in update_rq_clock().
1935 * This may result in other CPU reading this CPU's irq time and can
1936 * race with irq/account_system_vtime on this CPU. We would either get old
1937 * or new value (or semi updated value on 32 bit) with a side effect of
1938 * accounting a slice of irq time to wrong task when irq is in progress
1939 * while we read rq->clock. That is a worthy compromise in place of having
1940 * locks on each irq in account_system_time.
1941 */
b52bfee4
VP
1942static DEFINE_PER_CPU(u64, cpu_hardirq_time);
1943static DEFINE_PER_CPU(u64, cpu_softirq_time);
1944
1945static DEFINE_PER_CPU(u64, irq_start_time);
1946static int sched_clock_irqtime;
1947
1948void enable_sched_clock_irqtime(void)
1949{
1950 sched_clock_irqtime = 1;
1951}
1952
1953void disable_sched_clock_irqtime(void)
1954{
1955 sched_clock_irqtime = 0;
1956}
1957
305e6835
VP
1958static u64 irq_time_cpu(int cpu)
1959{
1960 if (!sched_clock_irqtime)
1961 return 0;
1962
1963 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1964}
1965
b52bfee4
VP
1966void account_system_vtime(struct task_struct *curr)
1967{
1968 unsigned long flags;
1969 int cpu;
1970 u64 now, delta;
1971
1972 if (!sched_clock_irqtime)
1973 return;
1974
1975 local_irq_save(flags);
1976
1977 now = sched_clock();
1978 cpu = smp_processor_id();
1979 delta = now - per_cpu(irq_start_time, cpu);
1980 per_cpu(irq_start_time, cpu) = now;
1981 /*
1982 * We do not account for softirq time from ksoftirqd here.
1983 * We want to continue accounting softirq time to ksoftirqd thread
1984 * in that case, so as not to confuse scheduler with a special task
1985 * that do not consume any time, but still wants to run.
1986 */
1987 if (hardirq_count())
1988 per_cpu(cpu_hardirq_time, cpu) += delta;
1989 else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD))
1990 per_cpu(cpu_softirq_time, cpu) += delta;
1991
1992 local_irq_restore(flags);
1993}
1994
aa483808
VP
1995static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time)
1996{
1997 if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) {
1998 u64 delta_irq = curr_irq_time - rq->prev_irq_time;
1999 rq->prev_irq_time = curr_irq_time;
2000 sched_rt_avg_update(rq, delta_irq);
2001 }
2002}
2003
305e6835
VP
2004#else
2005
2006static u64 irq_time_cpu(int cpu)
2007{
2008 return 0;
2009}
2010
aa483808
VP
2011static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { }
2012
b52bfee4
VP
2013#endif
2014
1e3c88bd
PZ
2015#include "sched_idletask.c"
2016#include "sched_fair.c"
2017#include "sched_rt.c"
34f971f6 2018#include "sched_stoptask.c"
1e3c88bd
PZ
2019#ifdef CONFIG_SCHED_DEBUG
2020# include "sched_debug.c"
2021#endif
2022
34f971f6
PZ
2023void sched_set_stop_task(int cpu, struct task_struct *stop)
2024{
2025 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
2026 struct task_struct *old_stop = cpu_rq(cpu)->stop;
2027
2028 if (stop) {
2029 /*
2030 * Make it appear like a SCHED_FIFO task, its something
2031 * userspace knows about and won't get confused about.
2032 *
2033 * Also, it will make PI more or less work without too
2034 * much confusion -- but then, stop work should not
2035 * rely on PI working anyway.
2036 */
2037 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
2038
2039 stop->sched_class = &stop_sched_class;
2040 }
2041
2042 cpu_rq(cpu)->stop = stop;
2043
2044 if (old_stop) {
2045 /*
2046 * Reset it back to a normal scheduling class so that
2047 * it can die in pieces.
2048 */
2049 old_stop->sched_class = &rt_sched_class;
2050 }
2051}
2052
14531189 2053/*
dd41f596 2054 * __normal_prio - return the priority that is based on the static prio
14531189 2055 */
14531189
IM
2056static inline int __normal_prio(struct task_struct *p)
2057{
dd41f596 2058 return p->static_prio;
14531189
IM
2059}
2060
b29739f9
IM
2061/*
2062 * Calculate the expected normal priority: i.e. priority
2063 * without taking RT-inheritance into account. Might be
2064 * boosted by interactivity modifiers. Changes upon fork,
2065 * setprio syscalls, and whenever the interactivity
2066 * estimator recalculates.
2067 */
36c8b586 2068static inline int normal_prio(struct task_struct *p)
b29739f9
IM
2069{
2070 int prio;
2071
e05606d3 2072 if (task_has_rt_policy(p))
b29739f9
IM
2073 prio = MAX_RT_PRIO-1 - p->rt_priority;
2074 else
2075 prio = __normal_prio(p);
2076 return prio;
2077}
2078
2079/*
2080 * Calculate the current priority, i.e. the priority
2081 * taken into account by the scheduler. This value might
2082 * be boosted by RT tasks, or might be boosted by
2083 * interactivity modifiers. Will be RT if the task got
2084 * RT-boosted. If not then it returns p->normal_prio.
2085 */
36c8b586 2086static int effective_prio(struct task_struct *p)
b29739f9
IM
2087{
2088 p->normal_prio = normal_prio(p);
2089 /*
2090 * If we are RT tasks or we were boosted to RT priority,
2091 * keep the priority unchanged. Otherwise, update priority
2092 * to the normal priority:
2093 */
2094 if (!rt_prio(p->prio))
2095 return p->normal_prio;
2096 return p->prio;
2097}
2098
1da177e4
LT
2099/**
2100 * task_curr - is this task currently executing on a CPU?
2101 * @p: the task in question.
2102 */
36c8b586 2103inline int task_curr(const struct task_struct *p)
1da177e4
LT
2104{
2105 return cpu_curr(task_cpu(p)) == p;
2106}
2107
cb469845
SR
2108static inline void check_class_changed(struct rq *rq, struct task_struct *p,
2109 const struct sched_class *prev_class,
2110 int oldprio, int running)
2111{
2112 if (prev_class != p->sched_class) {
2113 if (prev_class->switched_from)
2114 prev_class->switched_from(rq, p, running);
2115 p->sched_class->switched_to(rq, p, running);
2116 } else
2117 p->sched_class->prio_changed(rq, p, oldprio, running);
2118}
2119
1da177e4 2120#ifdef CONFIG_SMP
cc367732
IM
2121/*
2122 * Is this task likely cache-hot:
2123 */
e7693a36 2124static int
cc367732
IM
2125task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2126{
2127 s64 delta;
2128
e6c8fba7
PZ
2129 if (p->sched_class != &fair_sched_class)
2130 return 0;
2131
ef8002f6
NR
2132 if (unlikely(p->policy == SCHED_IDLE))
2133 return 0;
2134
f540a608
IM
2135 /*
2136 * Buddy candidates are cache hot:
2137 */
f685ceac 2138 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
4793241b
PZ
2139 (&p->se == cfs_rq_of(&p->se)->next ||
2140 &p->se == cfs_rq_of(&p->se)->last))
f540a608
IM
2141 return 1;
2142
6bc1665b
IM
2143 if (sysctl_sched_migration_cost == -1)
2144 return 1;
2145 if (sysctl_sched_migration_cost == 0)
2146 return 0;
2147
cc367732
IM
2148 delta = now - p->se.exec_start;
2149
2150 return delta < (s64)sysctl_sched_migration_cost;
2151}
2152
dd41f596 2153void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
c65cc870 2154{
e2912009
PZ
2155#ifdef CONFIG_SCHED_DEBUG
2156 /*
2157 * We should never call set_task_cpu() on a blocked task,
2158 * ttwu() will sort out the placement.
2159 */
077614ee
PZ
2160 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2161 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
e2912009
PZ
2162#endif
2163
de1d7286 2164 trace_sched_migrate_task(p, new_cpu);
cbc34ed1 2165
0c69774e
PZ
2166 if (task_cpu(p) != new_cpu) {
2167 p->se.nr_migrations++;
2168 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2169 }
dd41f596
IM
2170
2171 __set_task_cpu(p, new_cpu);
c65cc870
IM
2172}
2173
969c7921 2174struct migration_arg {
36c8b586 2175 struct task_struct *task;
1da177e4 2176 int dest_cpu;
70b97a7f 2177};
1da177e4 2178
969c7921
TH
2179static int migration_cpu_stop(void *data);
2180
1da177e4
LT
2181/*
2182 * The task's runqueue lock must be held.
2183 * Returns true if you have to wait for migration thread.
2184 */
969c7921 2185static bool migrate_task(struct task_struct *p, int dest_cpu)
1da177e4 2186{
70b97a7f 2187 struct rq *rq = task_rq(p);
1da177e4
LT
2188
2189 /*
2190 * If the task is not on a runqueue (and not running), then
e2912009 2191 * the next wake-up will properly place the task.
1da177e4 2192 */
969c7921 2193 return p->se.on_rq || task_running(rq, p);
1da177e4
LT
2194}
2195
2196/*
2197 * wait_task_inactive - wait for a thread to unschedule.
2198 *
85ba2d86
RM
2199 * If @match_state is nonzero, it's the @p->state value just checked and
2200 * not expected to change. If it changes, i.e. @p might have woken up,
2201 * then return zero. When we succeed in waiting for @p to be off its CPU,
2202 * we return a positive number (its total switch count). If a second call
2203 * a short while later returns the same number, the caller can be sure that
2204 * @p has remained unscheduled the whole time.
2205 *
1da177e4
LT
2206 * The caller must ensure that the task *will* unschedule sometime soon,
2207 * else this function might spin for a *long* time. This function can't
2208 * be called with interrupts off, or it may introduce deadlock with
2209 * smp_call_function() if an IPI is sent by the same process we are
2210 * waiting to become inactive.
2211 */
85ba2d86 2212unsigned long wait_task_inactive(struct task_struct *p, long match_state)
1da177e4
LT
2213{
2214 unsigned long flags;
dd41f596 2215 int running, on_rq;
85ba2d86 2216 unsigned long ncsw;
70b97a7f 2217 struct rq *rq;
1da177e4 2218
3a5c359a
AK
2219 for (;;) {
2220 /*
2221 * We do the initial early heuristics without holding
2222 * any task-queue locks at all. We'll only try to get
2223 * the runqueue lock when things look like they will
2224 * work out!
2225 */
2226 rq = task_rq(p);
fa490cfd 2227
3a5c359a
AK
2228 /*
2229 * If the task is actively running on another CPU
2230 * still, just relax and busy-wait without holding
2231 * any locks.
2232 *
2233 * NOTE! Since we don't hold any locks, it's not
2234 * even sure that "rq" stays as the right runqueue!
2235 * But we don't care, since "task_running()" will
2236 * return false if the runqueue has changed and p
2237 * is actually now running somewhere else!
2238 */
85ba2d86
RM
2239 while (task_running(rq, p)) {
2240 if (match_state && unlikely(p->state != match_state))
2241 return 0;
3a5c359a 2242 cpu_relax();
85ba2d86 2243 }
fa490cfd 2244
3a5c359a
AK
2245 /*
2246 * Ok, time to look more closely! We need the rq
2247 * lock now, to be *sure*. If we're wrong, we'll
2248 * just go back and repeat.
2249 */
2250 rq = task_rq_lock(p, &flags);
27a9da65 2251 trace_sched_wait_task(p);
3a5c359a
AK
2252 running = task_running(rq, p);
2253 on_rq = p->se.on_rq;
85ba2d86 2254 ncsw = 0;
f31e11d8 2255 if (!match_state || p->state == match_state)
93dcf55f 2256 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
3a5c359a 2257 task_rq_unlock(rq, &flags);
fa490cfd 2258
85ba2d86
RM
2259 /*
2260 * If it changed from the expected state, bail out now.
2261 */
2262 if (unlikely(!ncsw))
2263 break;
2264
3a5c359a
AK
2265 /*
2266 * Was it really running after all now that we
2267 * checked with the proper locks actually held?
2268 *
2269 * Oops. Go back and try again..
2270 */
2271 if (unlikely(running)) {
2272 cpu_relax();
2273 continue;
2274 }
fa490cfd 2275
3a5c359a
AK
2276 /*
2277 * It's not enough that it's not actively running,
2278 * it must be off the runqueue _entirely_, and not
2279 * preempted!
2280 *
80dd99b3 2281 * So if it was still runnable (but just not actively
3a5c359a
AK
2282 * running right now), it's preempted, and we should
2283 * yield - it could be a while.
2284 */
2285 if (unlikely(on_rq)) {
2286 schedule_timeout_uninterruptible(1);
2287 continue;
2288 }
fa490cfd 2289
3a5c359a
AK
2290 /*
2291 * Ahh, all good. It wasn't running, and it wasn't
2292 * runnable, which means that it will never become
2293 * running in the future either. We're all done!
2294 */
2295 break;
2296 }
85ba2d86
RM
2297
2298 return ncsw;
1da177e4
LT
2299}
2300
2301/***
2302 * kick_process - kick a running thread to enter/exit the kernel
2303 * @p: the to-be-kicked thread
2304 *
2305 * Cause a process which is running on another CPU to enter
2306 * kernel-mode, without any delay. (to get signals handled.)
2307 *
2308 * NOTE: this function doesnt have to take the runqueue lock,
2309 * because all it wants to ensure is that the remote task enters
2310 * the kernel. If the IPI races and the task has been migrated
2311 * to another CPU then no harm is done and the purpose has been
2312 * achieved as well.
2313 */
36c8b586 2314void kick_process(struct task_struct *p)
1da177e4
LT
2315{
2316 int cpu;
2317
2318 preempt_disable();
2319 cpu = task_cpu(p);
2320 if ((cpu != smp_processor_id()) && task_curr(p))
2321 smp_send_reschedule(cpu);
2322 preempt_enable();
2323}
b43e3521 2324EXPORT_SYMBOL_GPL(kick_process);
476d139c 2325#endif /* CONFIG_SMP */
1da177e4 2326
0793a61d
TG
2327/**
2328 * task_oncpu_function_call - call a function on the cpu on which a task runs
2329 * @p: the task to evaluate
2330 * @func: the function to be called
2331 * @info: the function call argument
2332 *
2333 * Calls the function @func when the task is currently running. This might
2334 * be on the current CPU, which just calls the function directly
2335 */
2336void task_oncpu_function_call(struct task_struct *p,
2337 void (*func) (void *info), void *info)
2338{
2339 int cpu;
2340
2341 preempt_disable();
2342 cpu = task_cpu(p);
2343 if (task_curr(p))
2344 smp_call_function_single(cpu, func, info, 1);
2345 preempt_enable();
2346}
2347
970b13ba 2348#ifdef CONFIG_SMP
30da688e
ON
2349/*
2350 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2351 */
5da9a0fb
PZ
2352static int select_fallback_rq(int cpu, struct task_struct *p)
2353{
2354 int dest_cpu;
2355 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2356
2357 /* Look for allowed, online CPU in same node. */
2358 for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2359 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2360 return dest_cpu;
2361
2362 /* Any allowed, online CPU? */
2363 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2364 if (dest_cpu < nr_cpu_ids)
2365 return dest_cpu;
2366
2367 /* No more Mr. Nice Guy. */
897f0b3c 2368 if (unlikely(dest_cpu >= nr_cpu_ids)) {
9084bb82 2369 dest_cpu = cpuset_cpus_allowed_fallback(p);
5da9a0fb
PZ
2370 /*
2371 * Don't tell them about moving exiting tasks or
2372 * kernel threads (both mm NULL), since they never
2373 * leave kernel.
2374 */
2375 if (p->mm && printk_ratelimit()) {
2376 printk(KERN_INFO "process %d (%s) no "
2377 "longer affine to cpu%d\n",
2378 task_pid_nr(p), p->comm, cpu);
2379 }
2380 }
2381
2382 return dest_cpu;
2383}
2384
e2912009 2385/*
30da688e 2386 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
e2912009 2387 */
970b13ba 2388static inline
0017d735 2389int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
970b13ba 2390{
0017d735 2391 int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
e2912009
PZ
2392
2393 /*
2394 * In order not to call set_task_cpu() on a blocking task we need
2395 * to rely on ttwu() to place the task on a valid ->cpus_allowed
2396 * cpu.
2397 *
2398 * Since this is common to all placement strategies, this lives here.
2399 *
2400 * [ this allows ->select_task() to simply return task_cpu(p) and
2401 * not worry about this generic constraint ]
2402 */
2403 if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
70f11205 2404 !cpu_online(cpu)))
5da9a0fb 2405 cpu = select_fallback_rq(task_cpu(p), p);
e2912009
PZ
2406
2407 return cpu;
970b13ba 2408}
09a40af5
MG
2409
2410static void update_avg(u64 *avg, u64 sample)
2411{
2412 s64 diff = sample - *avg;
2413 *avg += diff >> 3;
2414}
970b13ba
PZ
2415#endif
2416
9ed3811a
TH
2417static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2418 bool is_sync, bool is_migrate, bool is_local,
2419 unsigned long en_flags)
2420{
2421 schedstat_inc(p, se.statistics.nr_wakeups);
2422 if (is_sync)
2423 schedstat_inc(p, se.statistics.nr_wakeups_sync);
2424 if (is_migrate)
2425 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2426 if (is_local)
2427 schedstat_inc(p, se.statistics.nr_wakeups_local);
2428 else
2429 schedstat_inc(p, se.statistics.nr_wakeups_remote);
2430
2431 activate_task(rq, p, en_flags);
2432}
2433
2434static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2435 int wake_flags, bool success)
2436{
2437 trace_sched_wakeup(p, success);
2438 check_preempt_curr(rq, p, wake_flags);
2439
2440 p->state = TASK_RUNNING;
2441#ifdef CONFIG_SMP
2442 if (p->sched_class->task_woken)
2443 p->sched_class->task_woken(rq, p);
2444
2445 if (unlikely(rq->idle_stamp)) {
2446 u64 delta = rq->clock - rq->idle_stamp;
2447 u64 max = 2*sysctl_sched_migration_cost;
2448
2449 if (delta > max)
2450 rq->avg_idle = max;
2451 else
2452 update_avg(&rq->avg_idle, delta);
2453 rq->idle_stamp = 0;
2454 }
2455#endif
21aa9af0
TH
2456 /* if a worker is waking up, notify workqueue */
2457 if ((p->flags & PF_WQ_WORKER) && success)
2458 wq_worker_waking_up(p, cpu_of(rq));
9ed3811a
TH
2459}
2460
2461/**
1da177e4 2462 * try_to_wake_up - wake up a thread
9ed3811a 2463 * @p: the thread to be awakened
1da177e4 2464 * @state: the mask of task states that can be woken
9ed3811a 2465 * @wake_flags: wake modifier flags (WF_*)
1da177e4
LT
2466 *
2467 * Put it on the run-queue if it's not already there. The "current"
2468 * thread is always on the run-queue (except when the actual
2469 * re-schedule is in progress), and as such you're allowed to do
2470 * the simpler "current->state = TASK_RUNNING" to mark yourself
2471 * runnable without the overhead of this.
2472 *
9ed3811a
TH
2473 * Returns %true if @p was woken up, %false if it was already running
2474 * or @state didn't match @p's state.
1da177e4 2475 */
7d478721
PZ
2476static int try_to_wake_up(struct task_struct *p, unsigned int state,
2477 int wake_flags)
1da177e4 2478{
cc367732 2479 int cpu, orig_cpu, this_cpu, success = 0;
1da177e4 2480 unsigned long flags;
371fd7e7 2481 unsigned long en_flags = ENQUEUE_WAKEUP;
ab3b3aa5 2482 struct rq *rq;
1da177e4 2483
e9c84311 2484 this_cpu = get_cpu();
2398f2c6 2485
04e2f174 2486 smp_wmb();
ab3b3aa5 2487 rq = task_rq_lock(p, &flags);
e9c84311 2488 if (!(p->state & state))
1da177e4
LT
2489 goto out;
2490
dd41f596 2491 if (p->se.on_rq)
1da177e4
LT
2492 goto out_running;
2493
2494 cpu = task_cpu(p);
cc367732 2495 orig_cpu = cpu;
1da177e4
LT
2496
2497#ifdef CONFIG_SMP
2498 if (unlikely(task_running(rq, p)))
2499 goto out_activate;
2500
e9c84311
PZ
2501 /*
2502 * In order to handle concurrent wakeups and release the rq->lock
2503 * we put the task in TASK_WAKING state.
eb24073b
IM
2504 *
2505 * First fix up the nr_uninterruptible count:
e9c84311 2506 */
cc87f76a
PZ
2507 if (task_contributes_to_load(p)) {
2508 if (likely(cpu_online(orig_cpu)))
2509 rq->nr_uninterruptible--;
2510 else
2511 this_rq()->nr_uninterruptible--;
2512 }
e9c84311 2513 p->state = TASK_WAKING;
efbbd05a 2514
371fd7e7 2515 if (p->sched_class->task_waking) {
efbbd05a 2516 p->sched_class->task_waking(rq, p);
371fd7e7
PZ
2517 en_flags |= ENQUEUE_WAKING;
2518 }
efbbd05a 2519
0017d735
PZ
2520 cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2521 if (cpu != orig_cpu)
5d2f5a61 2522 set_task_cpu(p, cpu);
0017d735 2523 __task_rq_unlock(rq);
ab19cb23 2524
0970d299
PZ
2525 rq = cpu_rq(cpu);
2526 raw_spin_lock(&rq->lock);
f5dc3753 2527
0970d299
PZ
2528 /*
2529 * We migrated the task without holding either rq->lock, however
2530 * since the task is not on the task list itself, nobody else
2531 * will try and migrate the task, hence the rq should match the
2532 * cpu we just moved it to.
2533 */
2534 WARN_ON(task_cpu(p) != cpu);
e9c84311 2535 WARN_ON(p->state != TASK_WAKING);
1da177e4 2536
e7693a36
GH
2537#ifdef CONFIG_SCHEDSTATS
2538 schedstat_inc(rq, ttwu_count);
2539 if (cpu == this_cpu)
2540 schedstat_inc(rq, ttwu_local);
2541 else {
2542 struct sched_domain *sd;
2543 for_each_domain(this_cpu, sd) {
758b2cdc 2544 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
e7693a36
GH
2545 schedstat_inc(sd, ttwu_wake_remote);
2546 break;
2547 }
2548 }
2549 }
6d6bc0ad 2550#endif /* CONFIG_SCHEDSTATS */
e7693a36 2551
1da177e4
LT
2552out_activate:
2553#endif /* CONFIG_SMP */
9ed3811a
TH
2554 ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2555 cpu == this_cpu, en_flags);
1da177e4 2556 success = 1;
1da177e4 2557out_running:
9ed3811a 2558 ttwu_post_activation(p, rq, wake_flags, success);
1da177e4
LT
2559out:
2560 task_rq_unlock(rq, &flags);
e9c84311 2561 put_cpu();
1da177e4
LT
2562
2563 return success;
2564}
2565
21aa9af0
TH
2566/**
2567 * try_to_wake_up_local - try to wake up a local task with rq lock held
2568 * @p: the thread to be awakened
2569 *
2570 * Put @p on the run-queue if it's not alredy there. The caller must
2571 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2572 * the current task. this_rq() stays locked over invocation.
2573 */
2574static void try_to_wake_up_local(struct task_struct *p)
2575{
2576 struct rq *rq = task_rq(p);
2577 bool success = false;
2578
2579 BUG_ON(rq != this_rq());
2580 BUG_ON(p == current);
2581 lockdep_assert_held(&rq->lock);
2582
2583 if (!(p->state & TASK_NORMAL))
2584 return;
2585
2586 if (!p->se.on_rq) {
2587 if (likely(!task_running(rq, p))) {
2588 schedstat_inc(rq, ttwu_count);
2589 schedstat_inc(rq, ttwu_local);
2590 }
2591 ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2592 success = true;
2593 }
2594 ttwu_post_activation(p, rq, 0, success);
2595}
2596
50fa610a
DH
2597/**
2598 * wake_up_process - Wake up a specific process
2599 * @p: The process to be woken up.
2600 *
2601 * Attempt to wake up the nominated process and move it to the set of runnable
2602 * processes. Returns 1 if the process was woken up, 0 if it was already
2603 * running.
2604 *
2605 * It may be assumed that this function implies a write memory barrier before
2606 * changing the task state if and only if any tasks are woken up.
2607 */
7ad5b3a5 2608int wake_up_process(struct task_struct *p)
1da177e4 2609{
d9514f6c 2610 return try_to_wake_up(p, TASK_ALL, 0);
1da177e4 2611}
1da177e4
LT
2612EXPORT_SYMBOL(wake_up_process);
2613
7ad5b3a5 2614int wake_up_state(struct task_struct *p, unsigned int state)
1da177e4
LT
2615{
2616 return try_to_wake_up(p, state, 0);
2617}
2618
1da177e4
LT
2619/*
2620 * Perform scheduler related setup for a newly forked process p.
2621 * p is forked by current.
dd41f596
IM
2622 *
2623 * __sched_fork() is basic setup used by init_idle() too:
2624 */
2625static void __sched_fork(struct task_struct *p)
2626{
dd41f596
IM
2627 p->se.exec_start = 0;
2628 p->se.sum_exec_runtime = 0;
f6cf891c 2629 p->se.prev_sum_exec_runtime = 0;
6c594c21 2630 p->se.nr_migrations = 0;
6cfb0d5d
IM
2631
2632#ifdef CONFIG_SCHEDSTATS
41acab88 2633 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
6cfb0d5d 2634#endif
476d139c 2635
fa717060 2636 INIT_LIST_HEAD(&p->rt.run_list);
dd41f596 2637 p->se.on_rq = 0;
4a55bd5e 2638 INIT_LIST_HEAD(&p->se.group_node);
476d139c 2639
e107be36
AK
2640#ifdef CONFIG_PREEMPT_NOTIFIERS
2641 INIT_HLIST_HEAD(&p->preempt_notifiers);
2642#endif
dd41f596
IM
2643}
2644
2645/*
2646 * fork()/clone()-time setup:
2647 */
2648void sched_fork(struct task_struct *p, int clone_flags)
2649{
2650 int cpu = get_cpu();
2651
2652 __sched_fork(p);
06b83b5f 2653 /*
0017d735 2654 * We mark the process as running here. This guarantees that
06b83b5f
PZ
2655 * nobody will actually run it, and a signal or other external
2656 * event cannot wake it up and insert it on the runqueue either.
2657 */
0017d735 2658 p->state = TASK_RUNNING;
dd41f596 2659
b9dc29e7
MG
2660 /*
2661 * Revert to default priority/policy on fork if requested.
2662 */
2663 if (unlikely(p->sched_reset_on_fork)) {
f83f9ac2 2664 if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
b9dc29e7 2665 p->policy = SCHED_NORMAL;
f83f9ac2
PW
2666 p->normal_prio = p->static_prio;
2667 }
b9dc29e7 2668
6c697bdf
MG
2669 if (PRIO_TO_NICE(p->static_prio) < 0) {
2670 p->static_prio = NICE_TO_PRIO(0);
f83f9ac2 2671 p->normal_prio = p->static_prio;
6c697bdf
MG
2672 set_load_weight(p);
2673 }
2674
b9dc29e7
MG
2675 /*
2676 * We don't need the reset flag anymore after the fork. It has
2677 * fulfilled its duty:
2678 */
2679 p->sched_reset_on_fork = 0;
2680 }
ca94c442 2681
f83f9ac2
PW
2682 /*
2683 * Make sure we do not leak PI boosting priority to the child.
2684 */
2685 p->prio = current->normal_prio;
2686
2ddbf952
HS
2687 if (!rt_prio(p->prio))
2688 p->sched_class = &fair_sched_class;
b29739f9 2689
cd29fe6f
PZ
2690 if (p->sched_class->task_fork)
2691 p->sched_class->task_fork(p);
2692
86951599
PZ
2693 /*
2694 * The child is not yet in the pid-hash so no cgroup attach races,
2695 * and the cgroup is pinned to this child due to cgroup_fork()
2696 * is ran before sched_fork().
2697 *
2698 * Silence PROVE_RCU.
2699 */
2700 rcu_read_lock();
5f3edc1b 2701 set_task_cpu(p, cpu);
86951599 2702 rcu_read_unlock();
5f3edc1b 2703
52f17b6c 2704#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
dd41f596 2705 if (likely(sched_info_on()))
52f17b6c 2706 memset(&p->sched_info, 0, sizeof(p->sched_info));
1da177e4 2707#endif
d6077cb8 2708#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
4866cde0
NP
2709 p->oncpu = 0;
2710#endif
1da177e4 2711#ifdef CONFIG_PREEMPT
4866cde0 2712 /* Want to start with kernel preemption disabled. */
a1261f54 2713 task_thread_info(p)->preempt_count = 1;
1da177e4 2714#endif
917b627d
GH
2715 plist_node_init(&p->pushable_tasks, MAX_PRIO);
2716
476d139c 2717 put_cpu();
1da177e4
LT
2718}
2719
2720/*
2721 * wake_up_new_task - wake up a newly created task for the first time.
2722 *
2723 * This function will do some initial scheduler statistics housekeeping
2724 * that must be done for every newly created context, then puts the task
2725 * on the runqueue and wakes it.
2726 */
7ad5b3a5 2727void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
1da177e4
LT
2728{
2729 unsigned long flags;
dd41f596 2730 struct rq *rq;
c890692b 2731 int cpu __maybe_unused = get_cpu();
fabf318e
PZ
2732
2733#ifdef CONFIG_SMP
0017d735
PZ
2734 rq = task_rq_lock(p, &flags);
2735 p->state = TASK_WAKING;
2736
fabf318e
PZ
2737 /*
2738 * Fork balancing, do it here and not earlier because:
2739 * - cpus_allowed can change in the fork path
2740 * - any previously selected cpu might disappear through hotplug
2741 *
0017d735
PZ
2742 * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2743 * without people poking at ->cpus_allowed.
fabf318e 2744 */
0017d735 2745 cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
fabf318e 2746 set_task_cpu(p, cpu);
1da177e4 2747
06b83b5f 2748 p->state = TASK_RUNNING;
0017d735
PZ
2749 task_rq_unlock(rq, &flags);
2750#endif
2751
2752 rq = task_rq_lock(p, &flags);
cd29fe6f 2753 activate_task(rq, p, 0);
27a9da65 2754 trace_sched_wakeup_new(p, 1);
a7558e01 2755 check_preempt_curr(rq, p, WF_FORK);
9a897c5a 2756#ifdef CONFIG_SMP
efbbd05a
PZ
2757 if (p->sched_class->task_woken)
2758 p->sched_class->task_woken(rq, p);
9a897c5a 2759#endif
dd41f596 2760 task_rq_unlock(rq, &flags);
fabf318e 2761 put_cpu();
1da177e4
LT
2762}
2763
e107be36
AK
2764#ifdef CONFIG_PREEMPT_NOTIFIERS
2765
2766/**
80dd99b3 2767 * preempt_notifier_register - tell me when current is being preempted & rescheduled
421cee29 2768 * @notifier: notifier struct to register
e107be36
AK
2769 */
2770void preempt_notifier_register(struct preempt_notifier *notifier)
2771{
2772 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2773}
2774EXPORT_SYMBOL_GPL(preempt_notifier_register);
2775
2776/**
2777 * preempt_notifier_unregister - no longer interested in preemption notifications
421cee29 2778 * @notifier: notifier struct to unregister
e107be36
AK
2779 *
2780 * This is safe to call from within a preemption notifier.
2781 */
2782void preempt_notifier_unregister(struct preempt_notifier *notifier)
2783{
2784 hlist_del(&notifier->link);
2785}
2786EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2787
2788static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2789{
2790 struct preempt_notifier *notifier;
2791 struct hlist_node *node;
2792
2793 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2794 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2795}
2796
2797static void
2798fire_sched_out_preempt_notifiers(struct task_struct *curr,
2799 struct task_struct *next)
2800{
2801 struct preempt_notifier *notifier;
2802 struct hlist_node *node;
2803
2804 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2805 notifier->ops->sched_out(notifier, next);
2806}
2807
6d6bc0ad 2808#else /* !CONFIG_PREEMPT_NOTIFIERS */
e107be36
AK
2809
2810static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2811{
2812}
2813
2814static void
2815fire_sched_out_preempt_notifiers(struct task_struct *curr,
2816 struct task_struct *next)
2817{
2818}
2819
6d6bc0ad 2820#endif /* CONFIG_PREEMPT_NOTIFIERS */
e107be36 2821
4866cde0
NP
2822/**
2823 * prepare_task_switch - prepare to switch tasks
2824 * @rq: the runqueue preparing to switch
421cee29 2825 * @prev: the current task that is being switched out
4866cde0
NP
2826 * @next: the task we are going to switch to.
2827 *
2828 * This is called with the rq lock held and interrupts off. It must
2829 * be paired with a subsequent finish_task_switch after the context
2830 * switch.
2831 *
2832 * prepare_task_switch sets up locking and calls architecture specific
2833 * hooks.
2834 */
e107be36
AK
2835static inline void
2836prepare_task_switch(struct rq *rq, struct task_struct *prev,
2837 struct task_struct *next)
4866cde0 2838{
e107be36 2839 fire_sched_out_preempt_notifiers(prev, next);
4866cde0
NP
2840 prepare_lock_switch(rq, next);
2841 prepare_arch_switch(next);
2842}
2843
1da177e4
LT
2844/**
2845 * finish_task_switch - clean up after a task-switch
344babaa 2846 * @rq: runqueue associated with task-switch
1da177e4
LT
2847 * @prev: the thread we just switched away from.
2848 *
4866cde0
NP
2849 * finish_task_switch must be called after the context switch, paired
2850 * with a prepare_task_switch call before the context switch.
2851 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2852 * and do any other architecture-specific cleanup actions.
1da177e4
LT
2853 *
2854 * Note that we may have delayed dropping an mm in context_switch(). If
41a2d6cf 2855 * so, we finish that here outside of the runqueue lock. (Doing it
1da177e4
LT
2856 * with the lock held can cause deadlocks; see schedule() for
2857 * details.)
2858 */
a9957449 2859static void finish_task_switch(struct rq *rq, struct task_struct *prev)
1da177e4
LT
2860 __releases(rq->lock)
2861{
1da177e4 2862 struct mm_struct *mm = rq->prev_mm;
55a101f8 2863 long prev_state;
1da177e4
LT
2864
2865 rq->prev_mm = NULL;
2866
2867 /*
2868 * A task struct has one reference for the use as "current".
c394cc9f 2869 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
55a101f8
ON
2870 * schedule one last time. The schedule call will never return, and
2871 * the scheduled task must drop that reference.
c394cc9f 2872 * The test for TASK_DEAD must occur while the runqueue locks are
1da177e4
LT
2873 * still held, otherwise prev could be scheduled on another cpu, die
2874 * there before we look at prev->state, and then the reference would
2875 * be dropped twice.
2876 * Manfred Spraul <manfred@colorfullife.com>
2877 */
55a101f8 2878 prev_state = prev->state;
4866cde0 2879 finish_arch_switch(prev);
8381f65d
JI
2880#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2881 local_irq_disable();
2882#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
49f47433 2883 perf_event_task_sched_in(current);
8381f65d
JI
2884#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2885 local_irq_enable();
2886#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
4866cde0 2887 finish_lock_switch(rq, prev);
e8fa1362 2888
e107be36 2889 fire_sched_in_preempt_notifiers(current);
1da177e4
LT
2890 if (mm)
2891 mmdrop(mm);
c394cc9f 2892 if (unlikely(prev_state == TASK_DEAD)) {
c6fd91f0 2893 /*
2894 * Remove function-return probe instances associated with this
2895 * task and put them back on the free list.
9761eea8 2896 */
c6fd91f0 2897 kprobe_flush_task(prev);
1da177e4 2898 put_task_struct(prev);
c6fd91f0 2899 }
1da177e4
LT
2900}
2901
3f029d3c
GH
2902#ifdef CONFIG_SMP
2903
2904/* assumes rq->lock is held */
2905static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2906{
2907 if (prev->sched_class->pre_schedule)
2908 prev->sched_class->pre_schedule(rq, prev);
2909}
2910
2911/* rq->lock is NOT held, but preemption is disabled */
2912static inline void post_schedule(struct rq *rq)
2913{
2914 if (rq->post_schedule) {
2915 unsigned long flags;
2916
05fa785c 2917 raw_spin_lock_irqsave(&rq->lock, flags);
3f029d3c
GH
2918 if (rq->curr->sched_class->post_schedule)
2919 rq->curr->sched_class->post_schedule(rq);
05fa785c 2920 raw_spin_unlock_irqrestore(&rq->lock, flags);
3f029d3c
GH
2921
2922 rq->post_schedule = 0;
2923 }
2924}
2925
2926#else
da19ab51 2927
3f029d3c
GH
2928static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2929{
2930}
2931
2932static inline void post_schedule(struct rq *rq)
2933{
1da177e4
LT
2934}
2935
3f029d3c
GH
2936#endif
2937
1da177e4
LT
2938/**
2939 * schedule_tail - first thing a freshly forked thread must call.
2940 * @prev: the thread we just switched away from.
2941 */
36c8b586 2942asmlinkage void schedule_tail(struct task_struct *prev)
1da177e4
LT
2943 __releases(rq->lock)
2944{
70b97a7f
IM
2945 struct rq *rq = this_rq();
2946
4866cde0 2947 finish_task_switch(rq, prev);
da19ab51 2948
3f029d3c
GH
2949 /*
2950 * FIXME: do we need to worry about rq being invalidated by the
2951 * task_switch?
2952 */
2953 post_schedule(rq);
70b97a7f 2954
4866cde0
NP
2955#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2956 /* In this case, finish_task_switch does not reenable preemption */
2957 preempt_enable();
2958#endif
1da177e4 2959 if (current->set_child_tid)
b488893a 2960 put_user(task_pid_vnr(current), current->set_child_tid);
1da177e4
LT
2961}
2962
2963/*
2964 * context_switch - switch to the new MM and the new
2965 * thread's register state.
2966 */
dd41f596 2967static inline void
70b97a7f 2968context_switch(struct rq *rq, struct task_struct *prev,
36c8b586 2969 struct task_struct *next)
1da177e4 2970{
dd41f596 2971 struct mm_struct *mm, *oldmm;
1da177e4 2972
e107be36 2973 prepare_task_switch(rq, prev, next);
27a9da65 2974 trace_sched_switch(prev, next);
dd41f596
IM
2975 mm = next->mm;
2976 oldmm = prev->active_mm;
9226d125
ZA
2977 /*
2978 * For paravirt, this is coupled with an exit in switch_to to
2979 * combine the page table reload and the switch backend into
2980 * one hypercall.
2981 */
224101ed 2982 arch_start_context_switch(prev);
9226d125 2983
31915ab4 2984 if (!mm) {
1da177e4
LT
2985 next->active_mm = oldmm;
2986 atomic_inc(&oldmm->mm_count);
2987 enter_lazy_tlb(oldmm, next);
2988 } else
2989 switch_mm(oldmm, mm, next);
2990
31915ab4 2991 if (!prev->mm) {
1da177e4 2992 prev->active_mm = NULL;
1da177e4
LT
2993 rq->prev_mm = oldmm;
2994 }
3a5f5e48
IM
2995 /*
2996 * Since the runqueue lock will be released by the next
2997 * task (which is an invalid locking op but in the case
2998 * of the scheduler it's an obvious special-case), so we
2999 * do an early lockdep release here:
3000 */
3001#ifndef __ARCH_WANT_UNLOCKED_CTXSW
8a25d5de 3002 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
3a5f5e48 3003#endif
1da177e4
LT
3004
3005 /* Here we just switch the register state and the stack. */
3006 switch_to(prev, next, prev);
3007
dd41f596
IM
3008 barrier();
3009 /*
3010 * this_rq must be evaluated again because prev may have moved
3011 * CPUs since it called schedule(), thus the 'rq' on its stack
3012 * frame will be invalid.
3013 */
3014 finish_task_switch(this_rq(), prev);
1da177e4
LT
3015}
3016
3017/*
3018 * nr_running, nr_uninterruptible and nr_context_switches:
3019 *
3020 * externally visible scheduler statistics: current number of runnable
3021 * threads, current number of uninterruptible-sleeping threads, total
3022 * number of context switches performed since bootup.
3023 */
3024unsigned long nr_running(void)
3025{
3026 unsigned long i, sum = 0;
3027
3028 for_each_online_cpu(i)
3029 sum += cpu_rq(i)->nr_running;
3030
3031 return sum;
f711f609 3032}
1da177e4
LT
3033
3034unsigned long nr_uninterruptible(void)
f711f609 3035{
1da177e4 3036 unsigned long i, sum = 0;
f711f609 3037
0a945022 3038 for_each_possible_cpu(i)
1da177e4 3039 sum += cpu_rq(i)->nr_uninterruptible;
f711f609
GS
3040
3041 /*
1da177e4
LT
3042 * Since we read the counters lockless, it might be slightly
3043 * inaccurate. Do not allow it to go below zero though:
f711f609 3044 */
1da177e4
LT
3045 if (unlikely((long)sum < 0))
3046 sum = 0;
f711f609 3047
1da177e4 3048 return sum;
f711f609 3049}
f711f609 3050
1da177e4 3051unsigned long long nr_context_switches(void)
46cb4b7c 3052{
cc94abfc
SR
3053 int i;
3054 unsigned long long sum = 0;
46cb4b7c 3055
0a945022 3056 for_each_possible_cpu(i)
1da177e4 3057 sum += cpu_rq(i)->nr_switches;
46cb4b7c 3058
1da177e4
LT
3059 return sum;
3060}
483b4ee6 3061
1da177e4
LT
3062unsigned long nr_iowait(void)
3063{
3064 unsigned long i, sum = 0;
483b4ee6 3065
0a945022 3066 for_each_possible_cpu(i)
1da177e4 3067 sum += atomic_read(&cpu_rq(i)->nr_iowait);
46cb4b7c 3068
1da177e4
LT
3069 return sum;
3070}
483b4ee6 3071
8c215bd3 3072unsigned long nr_iowait_cpu(int cpu)
69d25870 3073{
8c215bd3 3074 struct rq *this = cpu_rq(cpu);
69d25870
AV
3075 return atomic_read(&this->nr_iowait);
3076}
46cb4b7c 3077
69d25870
AV
3078unsigned long this_cpu_load(void)
3079{
3080 struct rq *this = this_rq();
3081 return this->cpu_load[0];
3082}
e790fb0b 3083
46cb4b7c 3084
dce48a84
TG
3085/* Variables and functions for calc_load */
3086static atomic_long_t calc_load_tasks;
3087static unsigned long calc_load_update;
3088unsigned long avenrun[3];
3089EXPORT_SYMBOL(avenrun);
46cb4b7c 3090
74f5187a
PZ
3091static long calc_load_fold_active(struct rq *this_rq)
3092{
3093 long nr_active, delta = 0;
3094
3095 nr_active = this_rq->nr_running;
3096 nr_active += (long) this_rq->nr_uninterruptible;
3097
3098 if (nr_active != this_rq->calc_load_active) {
3099 delta = nr_active - this_rq->calc_load_active;
3100 this_rq->calc_load_active = nr_active;
3101 }
3102
3103 return delta;
3104}
3105
3106#ifdef CONFIG_NO_HZ
3107/*
3108 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
3109 *
3110 * When making the ILB scale, we should try to pull this in as well.
3111 */
3112static atomic_long_t calc_load_tasks_idle;
3113
3114static void calc_load_account_idle(struct rq *this_rq)
3115{
3116 long delta;
3117
3118 delta = calc_load_fold_active(this_rq);
3119 if (delta)
3120 atomic_long_add(delta, &calc_load_tasks_idle);
3121}
3122
3123static long calc_load_fold_idle(void)
3124{
3125 long delta = 0;
3126
3127 /*
3128 * Its got a race, we don't care...
3129 */
3130 if (atomic_long_read(&calc_load_tasks_idle))
3131 delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3132
3133 return delta;
3134}
3135#else
3136static void calc_load_account_idle(struct rq *this_rq)
3137{
3138}
3139
3140static inline long calc_load_fold_idle(void)
3141{
3142 return 0;
3143}
3144#endif
3145
2d02494f
TG
3146/**
3147 * get_avenrun - get the load average array
3148 * @loads: pointer to dest load array
3149 * @offset: offset to add
3150 * @shift: shift count to shift the result left
3151 *
3152 * These values are estimates at best, so no need for locking.
3153 */
3154void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3155{
3156 loads[0] = (avenrun[0] + offset) << shift;
3157 loads[1] = (avenrun[1] + offset) << shift;
3158 loads[2] = (avenrun[2] + offset) << shift;
46cb4b7c 3159}
46cb4b7c 3160
dce48a84
TG
3161static unsigned long
3162calc_load(unsigned long load, unsigned long exp, unsigned long active)
db1b1fef 3163{
dce48a84
TG
3164 load *= exp;
3165 load += active * (FIXED_1 - exp);
3166 return load >> FSHIFT;
3167}
46cb4b7c
SS
3168
3169/*
dce48a84
TG
3170 * calc_load - update the avenrun load estimates 10 ticks after the
3171 * CPUs have updated calc_load_tasks.
7835b98b 3172 */
dce48a84 3173void calc_global_load(void)
7835b98b 3174{
dce48a84
TG
3175 unsigned long upd = calc_load_update + 10;
3176 long active;
1da177e4 3177
dce48a84
TG
3178 if (time_before(jiffies, upd))
3179 return;
1da177e4 3180
dce48a84
TG
3181 active = atomic_long_read(&calc_load_tasks);
3182 active = active > 0 ? active * FIXED_1 : 0;
1da177e4 3183
dce48a84
TG
3184 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3185 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3186 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
dd41f596 3187
dce48a84
TG
3188 calc_load_update += LOAD_FREQ;
3189}
1da177e4 3190
dce48a84 3191/*
74f5187a
PZ
3192 * Called from update_cpu_load() to periodically update this CPU's
3193 * active count.
dce48a84
TG
3194 */
3195static void calc_load_account_active(struct rq *this_rq)
3196{
74f5187a 3197 long delta;
08c183f3 3198
74f5187a
PZ
3199 if (time_before(jiffies, this_rq->calc_load_update))
3200 return;
783609c6 3201
74f5187a
PZ
3202 delta = calc_load_fold_active(this_rq);
3203 delta += calc_load_fold_idle();
3204 if (delta)
dce48a84 3205 atomic_long_add(delta, &calc_load_tasks);
74f5187a
PZ
3206
3207 this_rq->calc_load_update += LOAD_FREQ;
46cb4b7c
SS
3208}
3209
fdf3e95d
VP
3210/*
3211 * The exact cpuload at various idx values, calculated at every tick would be
3212 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3213 *
3214 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3215 * on nth tick when cpu may be busy, then we have:
3216 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3217 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3218 *
3219 * decay_load_missed() below does efficient calculation of
3220 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3221 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3222 *
3223 * The calculation is approximated on a 128 point scale.
3224 * degrade_zero_ticks is the number of ticks after which load at any
3225 * particular idx is approximated to be zero.
3226 * degrade_factor is a precomputed table, a row for each load idx.
3227 * Each column corresponds to degradation factor for a power of two ticks,
3228 * based on 128 point scale.
3229 * Example:
3230 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3231 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3232 *
3233 * With this power of 2 load factors, we can degrade the load n times
3234 * by looking at 1 bits in n and doing as many mult/shift instead of
3235 * n mult/shifts needed by the exact degradation.
3236 */
3237#define DEGRADE_SHIFT 7
3238static const unsigned char
3239 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3240static const unsigned char
3241 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3242 {0, 0, 0, 0, 0, 0, 0, 0},
3243 {64, 32, 8, 0, 0, 0, 0, 0},
3244 {96, 72, 40, 12, 1, 0, 0},
3245 {112, 98, 75, 43, 15, 1, 0},
3246 {120, 112, 98, 76, 45, 16, 2} };
3247
3248/*
3249 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3250 * would be when CPU is idle and so we just decay the old load without
3251 * adding any new load.
3252 */
3253static unsigned long
3254decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3255{
3256 int j = 0;
3257
3258 if (!missed_updates)
3259 return load;
3260
3261 if (missed_updates >= degrade_zero_ticks[idx])
3262 return 0;
3263
3264 if (idx == 1)
3265 return load >> missed_updates;
3266
3267 while (missed_updates) {
3268 if (missed_updates % 2)
3269 load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3270
3271 missed_updates >>= 1;
3272 j++;
3273 }
3274 return load;
3275}
3276
46cb4b7c 3277/*
dd41f596 3278 * Update rq->cpu_load[] statistics. This function is usually called every
fdf3e95d
VP
3279 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3280 * every tick. We fix it up based on jiffies.
46cb4b7c 3281 */
dd41f596 3282static void update_cpu_load(struct rq *this_rq)
46cb4b7c 3283{
495eca49 3284 unsigned long this_load = this_rq->load.weight;
fdf3e95d
VP
3285 unsigned long curr_jiffies = jiffies;
3286 unsigned long pending_updates;
dd41f596 3287 int i, scale;
46cb4b7c 3288
dd41f596 3289 this_rq->nr_load_updates++;
46cb4b7c 3290
fdf3e95d
VP
3291 /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3292 if (curr_jiffies == this_rq->last_load_update_tick)
3293 return;
3294
3295 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3296 this_rq->last_load_update_tick = curr_jiffies;
3297
dd41f596 3298 /* Update our load: */
fdf3e95d
VP
3299 this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3300 for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
dd41f596 3301 unsigned long old_load, new_load;
7d1e6a9b 3302
dd41f596 3303 /* scale is effectively 1 << i now, and >> i divides by scale */
46cb4b7c 3304
dd41f596 3305 old_load = this_rq->cpu_load[i];
fdf3e95d 3306 old_load = decay_load_missed(old_load, pending_updates - 1, i);
dd41f596 3307 new_load = this_load;
a25707f3
IM
3308 /*
3309 * Round up the averaging division if load is increasing. This
3310 * prevents us from getting stuck on 9 if the load is 10, for
3311 * example.
3312 */
3313 if (new_load > old_load)
fdf3e95d
VP
3314 new_load += scale - 1;
3315
3316 this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
dd41f596 3317 }
da2b71ed
SS
3318
3319 sched_avg_update(this_rq);
fdf3e95d
VP
3320}
3321
3322static void update_cpu_load_active(struct rq *this_rq)
3323{
3324 update_cpu_load(this_rq);
46cb4b7c 3325
74f5187a 3326 calc_load_account_active(this_rq);
46cb4b7c
SS
3327}
3328
dd41f596 3329#ifdef CONFIG_SMP
8a0be9ef 3330
46cb4b7c 3331/*
38022906
PZ
3332 * sched_exec - execve() is a valuable balancing opportunity, because at
3333 * this point the task has the smallest effective memory and cache footprint.
46cb4b7c 3334 */
38022906 3335void sched_exec(void)
46cb4b7c 3336{
38022906 3337 struct task_struct *p = current;
1da177e4 3338 unsigned long flags;
70b97a7f 3339 struct rq *rq;
0017d735 3340 int dest_cpu;
46cb4b7c 3341
1da177e4 3342 rq = task_rq_lock(p, &flags);
0017d735
PZ
3343 dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3344 if (dest_cpu == smp_processor_id())
3345 goto unlock;
38022906 3346
46cb4b7c 3347 /*
38022906 3348 * select_task_rq() can race against ->cpus_allowed
46cb4b7c 3349 */
30da688e 3350 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
969c7921
TH
3351 likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) {
3352 struct migration_arg arg = { p, dest_cpu };
46cb4b7c 3353
1da177e4 3354 task_rq_unlock(rq, &flags);
969c7921 3355 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1da177e4
LT
3356 return;
3357 }
0017d735 3358unlock:
1da177e4 3359 task_rq_unlock(rq, &flags);
1da177e4 3360}
dd41f596 3361
1da177e4
LT
3362#endif
3363
1da177e4
LT
3364DEFINE_PER_CPU(struct kernel_stat, kstat);
3365
3366EXPORT_PER_CPU_SYMBOL(kstat);
3367
3368/*
c5f8d995 3369 * Return any ns on the sched_clock that have not yet been accounted in
f06febc9 3370 * @p in case that task is currently running.
c5f8d995
HS
3371 *
3372 * Called with task_rq_lock() held on @rq.
1da177e4 3373 */
c5f8d995
HS
3374static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3375{
3376 u64 ns = 0;
3377
3378 if (task_current(rq, p)) {
3379 update_rq_clock(rq);
305e6835 3380 ns = rq->clock_task - p->se.exec_start;
c5f8d995
HS
3381 if ((s64)ns < 0)
3382 ns = 0;
3383 }
3384
3385 return ns;
3386}
3387
bb34d92f 3388unsigned long long task_delta_exec(struct task_struct *p)
1da177e4 3389{
1da177e4 3390 unsigned long flags;
41b86e9c 3391 struct rq *rq;
bb34d92f 3392 u64 ns = 0;
48f24c4d 3393
41b86e9c 3394 rq = task_rq_lock(p, &flags);
c5f8d995
HS
3395 ns = do_task_delta_exec(p, rq);
3396 task_rq_unlock(rq, &flags);
1508487e 3397
c5f8d995
HS
3398 return ns;
3399}
f06febc9 3400
c5f8d995
HS
3401/*
3402 * Return accounted runtime for the task.
3403 * In case the task is currently running, return the runtime plus current's
3404 * pending runtime that have not been accounted yet.
3405 */
3406unsigned long long task_sched_runtime(struct task_struct *p)
3407{
3408 unsigned long flags;
3409 struct rq *rq;
3410 u64 ns = 0;
3411
3412 rq = task_rq_lock(p, &flags);
3413 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3414 task_rq_unlock(rq, &flags);
3415
3416 return ns;
3417}
48f24c4d 3418
c5f8d995
HS
3419/*
3420 * Return sum_exec_runtime for the thread group.
3421 * In case the task is currently running, return the sum plus current's
3422 * pending runtime that have not been accounted yet.
3423 *
3424 * Note that the thread group might have other running tasks as well,
3425 * so the return value not includes other pending runtime that other
3426 * running tasks might have.
3427 */
3428unsigned long long thread_group_sched_runtime(struct task_struct *p)
3429{
3430 struct task_cputime totals;
3431 unsigned long flags;
3432 struct rq *rq;
3433 u64 ns;
3434
3435 rq = task_rq_lock(p, &flags);
3436 thread_group_cputime(p, &totals);
3437 ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
41b86e9c 3438 task_rq_unlock(rq, &flags);
48f24c4d 3439
1da177e4
LT
3440 return ns;
3441}
3442
1da177e4
LT
3443/*
3444 * Account user cpu time to a process.
3445 * @p: the process that the cpu time gets accounted to
1da177e4 3446 * @cputime: the cpu time spent in user space since the last update
457533a7 3447 * @cputime_scaled: cputime scaled by cpu frequency
1da177e4 3448 */
457533a7
MS
3449void account_user_time(struct task_struct *p, cputime_t cputime,
3450 cputime_t cputime_scaled)
1da177e4
LT
3451{
3452 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3453 cputime64_t tmp;
3454
457533a7 3455 /* Add user time to process. */
1da177e4 3456 p->utime = cputime_add(p->utime, cputime);
457533a7 3457 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
f06febc9 3458 account_group_user_time(p, cputime);
1da177e4
LT
3459
3460 /* Add user time to cpustat. */
3461 tmp = cputime_to_cputime64(cputime);
3462 if (TASK_NICE(p) > 0)
3463 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3464 else
3465 cpustat->user = cputime64_add(cpustat->user, tmp);
ef12fefa
BR
3466
3467 cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
49b5cf34
JL
3468 /* Account for user time used */
3469 acct_update_integrals(p);
1da177e4
LT
3470}
3471
94886b84
LV
3472/*
3473 * Account guest cpu time to a process.
3474 * @p: the process that the cpu time gets accounted to
3475 * @cputime: the cpu time spent in virtual machine since the last update
457533a7 3476 * @cputime_scaled: cputime scaled by cpu frequency
94886b84 3477 */
457533a7
MS
3478static void account_guest_time(struct task_struct *p, cputime_t cputime,
3479 cputime_t cputime_scaled)
94886b84
LV
3480{
3481 cputime64_t tmp;
3482 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3483
3484 tmp = cputime_to_cputime64(cputime);
3485
457533a7 3486 /* Add guest time to process. */
94886b84 3487 p->utime = cputime_add(p->utime, cputime);
457533a7 3488 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
f06febc9 3489 account_group_user_time(p, cputime);
94886b84
LV
3490 p->gtime = cputime_add(p->gtime, cputime);
3491
457533a7 3492 /* Add guest time to cpustat. */
ce0e7b28
RO
3493 if (TASK_NICE(p) > 0) {
3494 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3495 cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3496 } else {
3497 cpustat->user = cputime64_add(cpustat->user, tmp);
3498 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3499 }
94886b84
LV
3500}
3501
1da177e4
LT
3502/*
3503 * Account system cpu time to a process.
3504 * @p: the process that the cpu time gets accounted to
3505 * @hardirq_offset: the offset to subtract from hardirq_count()
3506 * @cputime: the cpu time spent in kernel space since the last update
457533a7 3507 * @cputime_scaled: cputime scaled by cpu frequency
1da177e4
LT
3508 */
3509void account_system_time(struct task_struct *p, int hardirq_offset,
457533a7 3510 cputime_t cputime, cputime_t cputime_scaled)
1da177e4
LT
3511{
3512 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
1da177e4
LT
3513 cputime64_t tmp;
3514
983ed7a6 3515 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
457533a7 3516 account_guest_time(p, cputime, cputime_scaled);
983ed7a6
HH
3517 return;
3518 }
94886b84 3519
457533a7 3520 /* Add system time to process. */
1da177e4 3521 p->stime = cputime_add(p->stime, cputime);
457533a7 3522 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
f06febc9 3523 account_group_system_time(p, cputime);
1da177e4
LT
3524
3525 /* Add system time to cpustat. */
3526 tmp = cputime_to_cputime64(cputime);
3527 if (hardirq_count() - hardirq_offset)
3528 cpustat->irq = cputime64_add(cpustat->irq, tmp);
75e1056f 3529 else if (in_serving_softirq())
1da177e4 3530 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
1da177e4 3531 else
79741dd3
MS
3532 cpustat->system = cputime64_add(cpustat->system, tmp);
3533
ef12fefa
BR
3534 cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3535
1da177e4
LT
3536 /* Account for system time used */
3537 acct_update_integrals(p);
1da177e4
LT
3538}
3539
c66f08be 3540/*
1da177e4 3541 * Account for involuntary wait time.
1da177e4 3542 * @steal: the cpu time spent in involuntary wait
c66f08be 3543 */
79741dd3 3544void account_steal_time(cputime_t cputime)
c66f08be 3545{
79741dd3
MS
3546 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3547 cputime64_t cputime64 = cputime_to_cputime64(cputime);
3548
3549 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
c66f08be
MN
3550}
3551
1da177e4 3552/*
79741dd3
MS
3553 * Account for idle time.
3554 * @cputime: the cpu time spent in idle wait
1da177e4 3555 */
79741dd3 3556void account_idle_time(cputime_t cputime)
1da177e4
LT
3557{
3558 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
79741dd3 3559 cputime64_t cputime64 = cputime_to_cputime64(cputime);
70b97a7f 3560 struct rq *rq = this_rq();
1da177e4 3561
79741dd3
MS
3562 if (atomic_read(&rq->nr_iowait) > 0)
3563 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3564 else
3565 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
1da177e4
LT
3566}
3567
79741dd3
MS
3568#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3569
3570/*
3571 * Account a single tick of cpu time.
3572 * @p: the process that the cpu time gets accounted to
3573 * @user_tick: indicates if the tick is a user or a system tick
3574 */
3575void account_process_tick(struct task_struct *p, int user_tick)
3576{
a42548a1 3577 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
79741dd3
MS
3578 struct rq *rq = this_rq();
3579
3580 if (user_tick)
a42548a1 3581 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
f5f293a4 3582 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
a42548a1 3583 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
79741dd3
MS
3584 one_jiffy_scaled);
3585 else
a42548a1 3586 account_idle_time(cputime_one_jiffy);
79741dd3
MS
3587}
3588
3589/*
3590 * Account multiple ticks of steal time.
3591 * @p: the process from which the cpu time has been stolen
3592 * @ticks: number of stolen ticks
3593 */
3594void account_steal_ticks(unsigned long ticks)
3595{
3596 account_steal_time(jiffies_to_cputime(ticks));
3597}
3598
3599/*
3600 * Account multiple ticks of idle time.
3601 * @ticks: number of stolen ticks
3602 */
3603void account_idle_ticks(unsigned long ticks)
3604{
3605 account_idle_time(jiffies_to_cputime(ticks));
1da177e4
LT
3606}
3607
79741dd3
MS
3608#endif
3609
49048622
BS
3610/*
3611 * Use precise platform statistics if available:
3612 */
3613#ifdef CONFIG_VIRT_CPU_ACCOUNTING
d180c5bc 3614void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3615{
d99ca3b9
HS
3616 *ut = p->utime;
3617 *st = p->stime;
49048622
BS
3618}
3619
0cf55e1e 3620void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3621{
0cf55e1e
HS
3622 struct task_cputime cputime;
3623
3624 thread_group_cputime(p, &cputime);
3625
3626 *ut = cputime.utime;
3627 *st = cputime.stime;
49048622
BS
3628}
3629#else
761b1d26
HS
3630
3631#ifndef nsecs_to_cputime
b7b20df9 3632# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
761b1d26
HS
3633#endif
3634
d180c5bc 3635void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3636{
d99ca3b9 3637 cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
49048622
BS
3638
3639 /*
3640 * Use CFS's precise accounting:
3641 */
d180c5bc 3642 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
49048622
BS
3643
3644 if (total) {
e75e863d 3645 u64 temp = rtime;
d180c5bc 3646
e75e863d 3647 temp *= utime;
49048622 3648 do_div(temp, total);
d180c5bc
HS
3649 utime = (cputime_t)temp;
3650 } else
3651 utime = rtime;
49048622 3652
d180c5bc
HS
3653 /*
3654 * Compare with previous values, to keep monotonicity:
3655 */
761b1d26 3656 p->prev_utime = max(p->prev_utime, utime);
d99ca3b9 3657 p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
49048622 3658
d99ca3b9
HS
3659 *ut = p->prev_utime;
3660 *st = p->prev_stime;
49048622
BS
3661}
3662
0cf55e1e
HS
3663/*
3664 * Must be called with siglock held.
3665 */
3666void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3667{
0cf55e1e
HS
3668 struct signal_struct *sig = p->signal;
3669 struct task_cputime cputime;
3670 cputime_t rtime, utime, total;
49048622 3671
0cf55e1e 3672 thread_group_cputime(p, &cputime);
49048622 3673
0cf55e1e
HS
3674 total = cputime_add(cputime.utime, cputime.stime);
3675 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
49048622 3676
0cf55e1e 3677 if (total) {
e75e863d 3678 u64 temp = rtime;
49048622 3679
e75e863d 3680 temp *= cputime.utime;
0cf55e1e
HS
3681 do_div(temp, total);
3682 utime = (cputime_t)temp;
3683 } else
3684 utime = rtime;
3685
3686 sig->prev_utime = max(sig->prev_utime, utime);
3687 sig->prev_stime = max(sig->prev_stime,
3688 cputime_sub(rtime, sig->prev_utime));
3689
3690 *ut = sig->prev_utime;
3691 *st = sig->prev_stime;
49048622 3692}
49048622 3693#endif
49048622 3694
7835b98b
CL
3695/*
3696 * This function gets called by the timer code, with HZ frequency.
3697 * We call it with interrupts disabled.
3698 *
3699 * It also gets called by the fork code, when changing the parent's
3700 * timeslices.
3701 */
3702void scheduler_tick(void)
3703{
7835b98b
CL
3704 int cpu = smp_processor_id();
3705 struct rq *rq = cpu_rq(cpu);
dd41f596 3706 struct task_struct *curr = rq->curr;
3e51f33f
PZ
3707
3708 sched_clock_tick();
dd41f596 3709
05fa785c 3710 raw_spin_lock(&rq->lock);
3e51f33f 3711 update_rq_clock(rq);
fdf3e95d 3712 update_cpu_load_active(rq);
fa85ae24 3713 curr->sched_class->task_tick(rq, curr, 0);
05fa785c 3714 raw_spin_unlock(&rq->lock);
7835b98b 3715
49f47433 3716 perf_event_task_tick(curr);
e220d2dc 3717
e418e1c2 3718#ifdef CONFIG_SMP
dd41f596
IM
3719 rq->idle_at_tick = idle_cpu(cpu);
3720 trigger_load_balance(rq, cpu);
e418e1c2 3721#endif
1da177e4
LT
3722}
3723
132380a0 3724notrace unsigned long get_parent_ip(unsigned long addr)
6cd8a4bb
SR
3725{
3726 if (in_lock_functions(addr)) {
3727 addr = CALLER_ADDR2;
3728 if (in_lock_functions(addr))
3729 addr = CALLER_ADDR3;
3730 }
3731 return addr;
3732}
1da177e4 3733
7e49fcce
SR
3734#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3735 defined(CONFIG_PREEMPT_TRACER))
3736
43627582 3737void __kprobes add_preempt_count(int val)
1da177e4 3738{
6cd8a4bb 3739#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3740 /*
3741 * Underflow?
3742 */
9a11b49a
IM
3743 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3744 return;
6cd8a4bb 3745#endif
1da177e4 3746 preempt_count() += val;
6cd8a4bb 3747#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3748 /*
3749 * Spinlock count overflowing soon?
3750 */
33859f7f
MOS
3751 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3752 PREEMPT_MASK - 10);
6cd8a4bb
SR
3753#endif
3754 if (preempt_count() == val)
3755 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
1da177e4
LT
3756}
3757EXPORT_SYMBOL(add_preempt_count);
3758
43627582 3759void __kprobes sub_preempt_count(int val)
1da177e4 3760{
6cd8a4bb 3761#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3762 /*
3763 * Underflow?
3764 */
01e3eb82 3765 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
9a11b49a 3766 return;
1da177e4
LT
3767 /*
3768 * Is the spinlock portion underflowing?
3769 */
9a11b49a
IM
3770 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3771 !(preempt_count() & PREEMPT_MASK)))
3772 return;
6cd8a4bb 3773#endif
9a11b49a 3774
6cd8a4bb
SR
3775 if (preempt_count() == val)
3776 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
1da177e4
LT
3777 preempt_count() -= val;
3778}
3779EXPORT_SYMBOL(sub_preempt_count);
3780
3781#endif
3782
3783/*
dd41f596 3784 * Print scheduling while atomic bug:
1da177e4 3785 */
dd41f596 3786static noinline void __schedule_bug(struct task_struct *prev)
1da177e4 3787{
838225b4
SS
3788 struct pt_regs *regs = get_irq_regs();
3789
3df0fc5b
PZ
3790 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3791 prev->comm, prev->pid, preempt_count());
838225b4 3792
dd41f596 3793 debug_show_held_locks(prev);
e21f5b15 3794 print_modules();
dd41f596
IM
3795 if (irqs_disabled())
3796 print_irqtrace_events(prev);
838225b4
SS
3797
3798 if (regs)
3799 show_regs(regs);
3800 else
3801 dump_stack();
dd41f596 3802}
1da177e4 3803
dd41f596
IM
3804/*
3805 * Various schedule()-time debugging checks and statistics:
3806 */
3807static inline void schedule_debug(struct task_struct *prev)
3808{
1da177e4 3809 /*
41a2d6cf 3810 * Test if we are atomic. Since do_exit() needs to call into
1da177e4
LT
3811 * schedule() atomically, we ignore that path for now.
3812 * Otherwise, whine if we are scheduling when we should not be.
3813 */
3f33a7ce 3814 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
dd41f596
IM
3815 __schedule_bug(prev);
3816
1da177e4
LT
3817 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3818
2d72376b 3819 schedstat_inc(this_rq(), sched_count);
b8efb561
IM
3820#ifdef CONFIG_SCHEDSTATS
3821 if (unlikely(prev->lock_depth >= 0)) {
2d72376b
IM
3822 schedstat_inc(this_rq(), bkl_count);
3823 schedstat_inc(prev, sched_info.bkl_count);
b8efb561
IM
3824 }
3825#endif
dd41f596
IM
3826}
3827
6cecd084 3828static void put_prev_task(struct rq *rq, struct task_struct *prev)
df1c99d4 3829{
a64692a3
MG
3830 if (prev->se.on_rq)
3831 update_rq_clock(rq);
3832 rq->skip_clock_update = 0;
6cecd084 3833 prev->sched_class->put_prev_task(rq, prev);
df1c99d4
MG
3834}
3835
dd41f596
IM
3836/*
3837 * Pick up the highest-prio task:
3838 */
3839static inline struct task_struct *
b67802ea 3840pick_next_task(struct rq *rq)
dd41f596 3841{
5522d5d5 3842 const struct sched_class *class;
dd41f596 3843 struct task_struct *p;
1da177e4
LT
3844
3845 /*
dd41f596
IM
3846 * Optimization: we know that if all tasks are in
3847 * the fair class we can call that function directly:
1da177e4 3848 */
dd41f596 3849 if (likely(rq->nr_running == rq->cfs.nr_running)) {
fb8d4724 3850 p = fair_sched_class.pick_next_task(rq);
dd41f596
IM
3851 if (likely(p))
3852 return p;
1da177e4
LT
3853 }
3854
34f971f6 3855 for_each_class(class) {
fb8d4724 3856 p = class->pick_next_task(rq);
dd41f596
IM
3857 if (p)
3858 return p;
dd41f596 3859 }
34f971f6
PZ
3860
3861 BUG(); /* the idle class will always have a runnable task */
dd41f596 3862}
1da177e4 3863
dd41f596
IM
3864/*
3865 * schedule() is the main scheduler function.
3866 */
ff743345 3867asmlinkage void __sched schedule(void)
dd41f596
IM
3868{
3869 struct task_struct *prev, *next;
67ca7bde 3870 unsigned long *switch_count;
dd41f596 3871 struct rq *rq;
31656519 3872 int cpu;
dd41f596 3873
ff743345
PZ
3874need_resched:
3875 preempt_disable();
dd41f596
IM
3876 cpu = smp_processor_id();
3877 rq = cpu_rq(cpu);
25502a6c 3878 rcu_note_context_switch(cpu);
dd41f596 3879 prev = rq->curr;
dd41f596
IM
3880
3881 release_kernel_lock(prev);
3882need_resched_nonpreemptible:
3883
3884 schedule_debug(prev);
1da177e4 3885
31656519 3886 if (sched_feat(HRTICK))
f333fdc9 3887 hrtick_clear(rq);
8f4d37ec 3888
05fa785c 3889 raw_spin_lock_irq(&rq->lock);
1e819950 3890 clear_tsk_need_resched(prev);
1da177e4 3891
246d86b5 3892 switch_count = &prev->nivcsw;
1da177e4 3893 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
21aa9af0 3894 if (unlikely(signal_pending_state(prev->state, prev))) {
1da177e4 3895 prev->state = TASK_RUNNING;
21aa9af0
TH
3896 } else {
3897 /*
3898 * If a worker is going to sleep, notify and
3899 * ask workqueue whether it wants to wake up a
3900 * task to maintain concurrency. If so, wake
3901 * up the task.
3902 */
3903 if (prev->flags & PF_WQ_WORKER) {
3904 struct task_struct *to_wakeup;
3905
3906 to_wakeup = wq_worker_sleeping(prev, cpu);
3907 if (to_wakeup)
3908 try_to_wake_up_local(to_wakeup);
3909 }
371fd7e7 3910 deactivate_task(rq, prev, DEQUEUE_SLEEP);
21aa9af0 3911 }
dd41f596 3912 switch_count = &prev->nvcsw;
1da177e4
LT
3913 }
3914
3f029d3c 3915 pre_schedule(rq, prev);
f65eda4f 3916
dd41f596 3917 if (unlikely(!rq->nr_running))
1da177e4 3918 idle_balance(cpu, rq);
1da177e4 3919
df1c99d4 3920 put_prev_task(rq, prev);
b67802ea 3921 next = pick_next_task(rq);
1da177e4 3922
1da177e4 3923 if (likely(prev != next)) {
673a90a1 3924 sched_info_switch(prev, next);
49f47433 3925 perf_event_task_sched_out(prev, next);
673a90a1 3926
1da177e4
LT
3927 rq->nr_switches++;
3928 rq->curr = next;
3929 ++*switch_count;
3930
dd41f596 3931 context_switch(rq, prev, next); /* unlocks the rq */
8f4d37ec 3932 /*
246d86b5
ON
3933 * The context switch have flipped the stack from under us
3934 * and restored the local variables which were saved when
3935 * this task called schedule() in the past. prev == current
3936 * is still correct, but it can be moved to another cpu/rq.
8f4d37ec
PZ
3937 */
3938 cpu = smp_processor_id();
3939 rq = cpu_rq(cpu);
1da177e4 3940 } else
05fa785c 3941 raw_spin_unlock_irq(&rq->lock);
1da177e4 3942
3f029d3c 3943 post_schedule(rq);
1da177e4 3944
246d86b5 3945 if (unlikely(reacquire_kernel_lock(prev)))
1da177e4 3946 goto need_resched_nonpreemptible;
8f4d37ec 3947
1da177e4 3948 preempt_enable_no_resched();
ff743345 3949 if (need_resched())
1da177e4
LT
3950 goto need_resched;
3951}
1da177e4
LT
3952EXPORT_SYMBOL(schedule);
3953
c08f7829 3954#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
0d66bf6d
PZ
3955/*
3956 * Look out! "owner" is an entirely speculative pointer
3957 * access and not reliable.
3958 */
3959int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
3960{
3961 unsigned int cpu;
3962 struct rq *rq;
3963
3964 if (!sched_feat(OWNER_SPIN))
3965 return 0;
3966
3967#ifdef CONFIG_DEBUG_PAGEALLOC
3968 /*
3969 * Need to access the cpu field knowing that
3970 * DEBUG_PAGEALLOC could have unmapped it if
3971 * the mutex owner just released it and exited.
3972 */
3973 if (probe_kernel_address(&owner->cpu, cpu))
4b402210 3974 return 0;
0d66bf6d
PZ
3975#else
3976 cpu = owner->cpu;
3977#endif
3978
3979 /*
3980 * Even if the access succeeded (likely case),
3981 * the cpu field may no longer be valid.
3982 */
3983 if (cpu >= nr_cpumask_bits)
4b402210 3984 return 0;
0d66bf6d
PZ
3985
3986 /*
3987 * We need to validate that we can do a
3988 * get_cpu() and that we have the percpu area.
3989 */
3990 if (!cpu_online(cpu))
4b402210 3991 return 0;
0d66bf6d
PZ
3992
3993 rq = cpu_rq(cpu);
3994
3995 for (;;) {
3996 /*
3997 * Owner changed, break to re-assess state.
3998 */
9d0f4dcc
TC
3999 if (lock->owner != owner) {
4000 /*
4001 * If the lock has switched to a different owner,
4002 * we likely have heavy contention. Return 0 to quit
4003 * optimistic spinning and not contend further:
4004 */
4005 if (lock->owner)
4006 return 0;
0d66bf6d 4007 break;
9d0f4dcc 4008 }
0d66bf6d
PZ
4009
4010 /*
4011 * Is that owner really running on that cpu?
4012 */
4013 if (task_thread_info(rq->curr) != owner || need_resched())
4014 return 0;
4015
4016 cpu_relax();
4017 }
4b402210 4018
0d66bf6d
PZ
4019 return 1;
4020}
4021#endif
4022
1da177e4
LT
4023#ifdef CONFIG_PREEMPT
4024/*
2ed6e34f 4025 * this is the entry point to schedule() from in-kernel preemption
41a2d6cf 4026 * off of preempt_enable. Kernel preemptions off return from interrupt
1da177e4
LT
4027 * occur there and call schedule directly.
4028 */
d1f74e20 4029asmlinkage void __sched notrace preempt_schedule(void)
1da177e4
LT
4030{
4031 struct thread_info *ti = current_thread_info();
6478d880 4032
1da177e4
LT
4033 /*
4034 * If there is a non-zero preempt_count or interrupts are disabled,
41a2d6cf 4035 * we do not want to preempt the current task. Just return..
1da177e4 4036 */
beed33a8 4037 if (likely(ti->preempt_count || irqs_disabled()))
1da177e4
LT
4038 return;
4039
3a5c359a 4040 do {
d1f74e20 4041 add_preempt_count_notrace(PREEMPT_ACTIVE);
3a5c359a 4042 schedule();
d1f74e20 4043 sub_preempt_count_notrace(PREEMPT_ACTIVE);
1da177e4 4044
3a5c359a
AK
4045 /*
4046 * Check again in case we missed a preemption opportunity
4047 * between schedule and now.
4048 */
4049 barrier();
5ed0cec0 4050 } while (need_resched());
1da177e4 4051}
1da177e4
LT
4052EXPORT_SYMBOL(preempt_schedule);
4053
4054/*
2ed6e34f 4055 * this is the entry point to schedule() from kernel preemption
1da177e4
LT
4056 * off of irq context.
4057 * Note, that this is called and return with irqs disabled. This will
4058 * protect us against recursive calling from irq.
4059 */
4060asmlinkage void __sched preempt_schedule_irq(void)
4061{
4062 struct thread_info *ti = current_thread_info();
6478d880 4063
2ed6e34f 4064 /* Catch callers which need to be fixed */
1da177e4
LT
4065 BUG_ON(ti->preempt_count || !irqs_disabled());
4066
3a5c359a
AK
4067 do {
4068 add_preempt_count(PREEMPT_ACTIVE);
3a5c359a
AK
4069 local_irq_enable();
4070 schedule();
4071 local_irq_disable();
3a5c359a 4072 sub_preempt_count(PREEMPT_ACTIVE);
1da177e4 4073
3a5c359a
AK
4074 /*
4075 * Check again in case we missed a preemption opportunity
4076 * between schedule and now.
4077 */
4078 barrier();
5ed0cec0 4079 } while (need_resched());
1da177e4
LT
4080}
4081
4082#endif /* CONFIG_PREEMPT */
4083
63859d4f 4084int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
95cdf3b7 4085 void *key)
1da177e4 4086{
63859d4f 4087 return try_to_wake_up(curr->private, mode, wake_flags);
1da177e4 4088}
1da177e4
LT
4089EXPORT_SYMBOL(default_wake_function);
4090
4091/*
41a2d6cf
IM
4092 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4093 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
1da177e4
LT
4094 * number) then we wake all the non-exclusive tasks and one exclusive task.
4095 *
4096 * There are circumstances in which we can try to wake a task which has already
41a2d6cf 4097 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
1da177e4
LT
4098 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4099 */
78ddb08f 4100static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
63859d4f 4101 int nr_exclusive, int wake_flags, void *key)
1da177e4 4102{
2e45874c 4103 wait_queue_t *curr, *next;
1da177e4 4104
2e45874c 4105 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
48f24c4d
IM
4106 unsigned flags = curr->flags;
4107
63859d4f 4108 if (curr->func(curr, mode, wake_flags, key) &&
48f24c4d 4109 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
1da177e4
LT
4110 break;
4111 }
4112}
4113
4114/**
4115 * __wake_up - wake up threads blocked on a waitqueue.
4116 * @q: the waitqueue
4117 * @mode: which threads
4118 * @nr_exclusive: how many wake-one or wake-many threads to wake up
67be2dd1 4119 * @key: is directly passed to the wakeup function
50fa610a
DH
4120 *
4121 * It may be assumed that this function implies a write memory barrier before
4122 * changing the task state if and only if any tasks are woken up.
1da177e4 4123 */
7ad5b3a5 4124void __wake_up(wait_queue_head_t *q, unsigned int mode,
95cdf3b7 4125 int nr_exclusive, void *key)
1da177e4
LT
4126{
4127 unsigned long flags;
4128
4129 spin_lock_irqsave(&q->lock, flags);
4130 __wake_up_common(q, mode, nr_exclusive, 0, key);
4131 spin_unlock_irqrestore(&q->lock, flags);
4132}
1da177e4
LT
4133EXPORT_SYMBOL(__wake_up);
4134
4135/*
4136 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4137 */
7ad5b3a5 4138void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
1da177e4
LT
4139{
4140 __wake_up_common(q, mode, 1, 0, NULL);
4141}
22c43c81 4142EXPORT_SYMBOL_GPL(__wake_up_locked);
1da177e4 4143
4ede816a
DL
4144void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4145{
4146 __wake_up_common(q, mode, 1, 0, key);
4147}
4148
1da177e4 4149/**
4ede816a 4150 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
1da177e4
LT
4151 * @q: the waitqueue
4152 * @mode: which threads
4153 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4ede816a 4154 * @key: opaque value to be passed to wakeup targets
1da177e4
LT
4155 *
4156 * The sync wakeup differs that the waker knows that it will schedule
4157 * away soon, so while the target thread will be woken up, it will not
4158 * be migrated to another CPU - ie. the two threads are 'synchronized'
4159 * with each other. This can prevent needless bouncing between CPUs.
4160 *
4161 * On UP it can prevent extra preemption.
50fa610a
DH
4162 *
4163 * It may be assumed that this function implies a write memory barrier before
4164 * changing the task state if and only if any tasks are woken up.
1da177e4 4165 */
4ede816a
DL
4166void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4167 int nr_exclusive, void *key)
1da177e4
LT
4168{
4169 unsigned long flags;
7d478721 4170 int wake_flags = WF_SYNC;
1da177e4
LT
4171
4172 if (unlikely(!q))
4173 return;
4174
4175 if (unlikely(!nr_exclusive))
7d478721 4176 wake_flags = 0;
1da177e4
LT
4177
4178 spin_lock_irqsave(&q->lock, flags);
7d478721 4179 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
1da177e4
LT
4180 spin_unlock_irqrestore(&q->lock, flags);
4181}
4ede816a
DL
4182EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4183
4184/*
4185 * __wake_up_sync - see __wake_up_sync_key()
4186 */
4187void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4188{
4189 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4190}
1da177e4
LT
4191EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4192
65eb3dc6
KD
4193/**
4194 * complete: - signals a single thread waiting on this completion
4195 * @x: holds the state of this particular completion
4196 *
4197 * This will wake up a single thread waiting on this completion. Threads will be
4198 * awakened in the same order in which they were queued.
4199 *
4200 * See also complete_all(), wait_for_completion() and related routines.
50fa610a
DH
4201 *
4202 * It may be assumed that this function implies a write memory barrier before
4203 * changing the task state if and only if any tasks are woken up.
65eb3dc6 4204 */
b15136e9 4205void complete(struct completion *x)
1da177e4
LT
4206{
4207 unsigned long flags;
4208
4209 spin_lock_irqsave(&x->wait.lock, flags);
4210 x->done++;
d9514f6c 4211 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
1da177e4
LT
4212 spin_unlock_irqrestore(&x->wait.lock, flags);
4213}
4214EXPORT_SYMBOL(complete);
4215
65eb3dc6
KD
4216/**
4217 * complete_all: - signals all threads waiting on this completion
4218 * @x: holds the state of this particular completion
4219 *
4220 * This will wake up all threads waiting on this particular completion event.
50fa610a
DH
4221 *
4222 * It may be assumed that this function implies a write memory barrier before
4223 * changing the task state if and only if any tasks are woken up.
65eb3dc6 4224 */
b15136e9 4225void complete_all(struct completion *x)
1da177e4
LT
4226{
4227 unsigned long flags;
4228
4229 spin_lock_irqsave(&x->wait.lock, flags);
4230 x->done += UINT_MAX/2;
d9514f6c 4231 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
1da177e4
LT
4232 spin_unlock_irqrestore(&x->wait.lock, flags);
4233}
4234EXPORT_SYMBOL(complete_all);
4235
8cbbe86d
AK
4236static inline long __sched
4237do_wait_for_common(struct completion *x, long timeout, int state)
1da177e4 4238{
1da177e4
LT
4239 if (!x->done) {
4240 DECLARE_WAITQUEUE(wait, current);
4241
a93d2f17 4242 __add_wait_queue_tail_exclusive(&x->wait, &wait);
1da177e4 4243 do {
94d3d824 4244 if (signal_pending_state(state, current)) {
ea71a546
ON
4245 timeout = -ERESTARTSYS;
4246 break;
8cbbe86d
AK
4247 }
4248 __set_current_state(state);
1da177e4
LT
4249 spin_unlock_irq(&x->wait.lock);
4250 timeout = schedule_timeout(timeout);
4251 spin_lock_irq(&x->wait.lock);
ea71a546 4252 } while (!x->done && timeout);
1da177e4 4253 __remove_wait_queue(&x->wait, &wait);
ea71a546
ON
4254 if (!x->done)
4255 return timeout;
1da177e4
LT
4256 }
4257 x->done--;
ea71a546 4258 return timeout ?: 1;
1da177e4 4259}
1da177e4 4260
8cbbe86d
AK
4261static long __sched
4262wait_for_common(struct completion *x, long timeout, int state)
1da177e4 4263{
1da177e4
LT
4264 might_sleep();
4265
4266 spin_lock_irq(&x->wait.lock);
8cbbe86d 4267 timeout = do_wait_for_common(x, timeout, state);
1da177e4 4268 spin_unlock_irq(&x->wait.lock);
8cbbe86d
AK
4269 return timeout;
4270}
1da177e4 4271
65eb3dc6
KD
4272/**
4273 * wait_for_completion: - waits for completion of a task
4274 * @x: holds the state of this particular completion
4275 *
4276 * This waits to be signaled for completion of a specific task. It is NOT
4277 * interruptible and there is no timeout.
4278 *
4279 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4280 * and interrupt capability. Also see complete().
4281 */
b15136e9 4282void __sched wait_for_completion(struct completion *x)
8cbbe86d
AK
4283{
4284 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
1da177e4 4285}
8cbbe86d 4286EXPORT_SYMBOL(wait_for_completion);
1da177e4 4287
65eb3dc6
KD
4288/**
4289 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4290 * @x: holds the state of this particular completion
4291 * @timeout: timeout value in jiffies
4292 *
4293 * This waits for either a completion of a specific task to be signaled or for a
4294 * specified timeout to expire. The timeout is in jiffies. It is not
4295 * interruptible.
4296 */
b15136e9 4297unsigned long __sched
8cbbe86d 4298wait_for_completion_timeout(struct completion *x, unsigned long timeout)
1da177e4 4299{
8cbbe86d 4300 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
1da177e4 4301}
8cbbe86d 4302EXPORT_SYMBOL(wait_for_completion_timeout);
1da177e4 4303
65eb3dc6
KD
4304/**
4305 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4306 * @x: holds the state of this particular completion
4307 *
4308 * This waits for completion of a specific task to be signaled. It is
4309 * interruptible.
4310 */
8cbbe86d 4311int __sched wait_for_completion_interruptible(struct completion *x)
0fec171c 4312{
51e97990
AK
4313 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4314 if (t == -ERESTARTSYS)
4315 return t;
4316 return 0;
0fec171c 4317}
8cbbe86d 4318EXPORT_SYMBOL(wait_for_completion_interruptible);
1da177e4 4319
65eb3dc6
KD
4320/**
4321 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4322 * @x: holds the state of this particular completion
4323 * @timeout: timeout value in jiffies
4324 *
4325 * This waits for either a completion of a specific task to be signaled or for a
4326 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4327 */
b15136e9 4328unsigned long __sched
8cbbe86d
AK
4329wait_for_completion_interruptible_timeout(struct completion *x,
4330 unsigned long timeout)
0fec171c 4331{
8cbbe86d 4332 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
0fec171c 4333}
8cbbe86d 4334EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
1da177e4 4335
65eb3dc6
KD
4336/**
4337 * wait_for_completion_killable: - waits for completion of a task (killable)
4338 * @x: holds the state of this particular completion
4339 *
4340 * This waits to be signaled for completion of a specific task. It can be
4341 * interrupted by a kill signal.
4342 */
009e577e
MW
4343int __sched wait_for_completion_killable(struct completion *x)
4344{
4345 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4346 if (t == -ERESTARTSYS)
4347 return t;
4348 return 0;
4349}
4350EXPORT_SYMBOL(wait_for_completion_killable);
4351
0aa12fb4
SW
4352/**
4353 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4354 * @x: holds the state of this particular completion
4355 * @timeout: timeout value in jiffies
4356 *
4357 * This waits for either a completion of a specific task to be
4358 * signaled or for a specified timeout to expire. It can be
4359 * interrupted by a kill signal. The timeout is in jiffies.
4360 */
4361unsigned long __sched
4362wait_for_completion_killable_timeout(struct completion *x,
4363 unsigned long timeout)
4364{
4365 return wait_for_common(x, timeout, TASK_KILLABLE);
4366}
4367EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4368
be4de352
DC
4369/**
4370 * try_wait_for_completion - try to decrement a completion without blocking
4371 * @x: completion structure
4372 *
4373 * Returns: 0 if a decrement cannot be done without blocking
4374 * 1 if a decrement succeeded.
4375 *
4376 * If a completion is being used as a counting completion,
4377 * attempt to decrement the counter without blocking. This
4378 * enables us to avoid waiting if the resource the completion
4379 * is protecting is not available.
4380 */
4381bool try_wait_for_completion(struct completion *x)
4382{
7539a3b3 4383 unsigned long flags;
be4de352
DC
4384 int ret = 1;
4385
7539a3b3 4386 spin_lock_irqsave(&x->wait.lock, flags);
be4de352
DC
4387 if (!x->done)
4388 ret = 0;
4389 else
4390 x->done--;
7539a3b3 4391 spin_unlock_irqrestore(&x->wait.lock, flags);
be4de352
DC
4392 return ret;
4393}
4394EXPORT_SYMBOL(try_wait_for_completion);
4395
4396/**
4397 * completion_done - Test to see if a completion has any waiters
4398 * @x: completion structure
4399 *
4400 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4401 * 1 if there are no waiters.
4402 *
4403 */
4404bool completion_done(struct completion *x)
4405{
7539a3b3 4406 unsigned long flags;
be4de352
DC
4407 int ret = 1;
4408
7539a3b3 4409 spin_lock_irqsave(&x->wait.lock, flags);
be4de352
DC
4410 if (!x->done)
4411 ret = 0;
7539a3b3 4412 spin_unlock_irqrestore(&x->wait.lock, flags);
be4de352
DC
4413 return ret;
4414}
4415EXPORT_SYMBOL(completion_done);
4416
8cbbe86d
AK
4417static long __sched
4418sleep_on_common(wait_queue_head_t *q, int state, long timeout)
1da177e4 4419{
0fec171c
IM
4420 unsigned long flags;
4421 wait_queue_t wait;
4422
4423 init_waitqueue_entry(&wait, current);
1da177e4 4424
8cbbe86d 4425 __set_current_state(state);
1da177e4 4426
8cbbe86d
AK
4427 spin_lock_irqsave(&q->lock, flags);
4428 __add_wait_queue(q, &wait);
4429 spin_unlock(&q->lock);
4430 timeout = schedule_timeout(timeout);
4431 spin_lock_irq(&q->lock);
4432 __remove_wait_queue(q, &wait);
4433 spin_unlock_irqrestore(&q->lock, flags);
4434
4435 return timeout;
4436}
4437
4438void __sched interruptible_sleep_on(wait_queue_head_t *q)
4439{
4440 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
1da177e4 4441}
1da177e4
LT
4442EXPORT_SYMBOL(interruptible_sleep_on);
4443
0fec171c 4444long __sched
95cdf3b7 4445interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
1da177e4 4446{
8cbbe86d 4447 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
1da177e4 4448}
1da177e4
LT
4449EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4450
0fec171c 4451void __sched sleep_on(wait_queue_head_t *q)
1da177e4 4452{
8cbbe86d 4453 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
1da177e4 4454}
1da177e4
LT
4455EXPORT_SYMBOL(sleep_on);
4456
0fec171c 4457long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
1da177e4 4458{
8cbbe86d 4459 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
1da177e4 4460}
1da177e4
LT
4461EXPORT_SYMBOL(sleep_on_timeout);
4462
b29739f9
IM
4463#ifdef CONFIG_RT_MUTEXES
4464
4465/*
4466 * rt_mutex_setprio - set the current priority of a task
4467 * @p: task
4468 * @prio: prio value (kernel-internal form)
4469 *
4470 * This function changes the 'effective' priority of a task. It does
4471 * not touch ->normal_prio like __setscheduler().
4472 *
4473 * Used by the rt_mutex code to implement priority inheritance logic.
4474 */
36c8b586 4475void rt_mutex_setprio(struct task_struct *p, int prio)
b29739f9
IM
4476{
4477 unsigned long flags;
83b699ed 4478 int oldprio, on_rq, running;
70b97a7f 4479 struct rq *rq;
83ab0aa0 4480 const struct sched_class *prev_class;
b29739f9
IM
4481
4482 BUG_ON(prio < 0 || prio > MAX_PRIO);
4483
4484 rq = task_rq_lock(p, &flags);
4485
a8027073 4486 trace_sched_pi_setprio(p, prio);
d5f9f942 4487 oldprio = p->prio;
83ab0aa0 4488 prev_class = p->sched_class;
dd41f596 4489 on_rq = p->se.on_rq;
051a1d1a 4490 running = task_current(rq, p);
0e1f3483 4491 if (on_rq)
69be72c1 4492 dequeue_task(rq, p, 0);
0e1f3483
HS
4493 if (running)
4494 p->sched_class->put_prev_task(rq, p);
dd41f596
IM
4495
4496 if (rt_prio(prio))
4497 p->sched_class = &rt_sched_class;
4498 else
4499 p->sched_class = &fair_sched_class;
4500
b29739f9
IM
4501 p->prio = prio;
4502
0e1f3483
HS
4503 if (running)
4504 p->sched_class->set_curr_task(rq);
dd41f596 4505 if (on_rq) {
371fd7e7 4506 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
cb469845
SR
4507
4508 check_class_changed(rq, p, prev_class, oldprio, running);
b29739f9
IM
4509 }
4510 task_rq_unlock(rq, &flags);
4511}
4512
4513#endif
4514
36c8b586 4515void set_user_nice(struct task_struct *p, long nice)
1da177e4 4516{
dd41f596 4517 int old_prio, delta, on_rq;
1da177e4 4518 unsigned long flags;
70b97a7f 4519 struct rq *rq;
1da177e4
LT
4520
4521 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4522 return;
4523 /*
4524 * We have to be careful, if called from sys_setpriority(),
4525 * the task might be in the middle of scheduling on another CPU.
4526 */
4527 rq = task_rq_lock(p, &flags);
4528 /*
4529 * The RT priorities are set via sched_setscheduler(), but we still
4530 * allow the 'normal' nice value to be set - but as expected
4531 * it wont have any effect on scheduling until the task is
dd41f596 4532 * SCHED_FIFO/SCHED_RR:
1da177e4 4533 */
e05606d3 4534 if (task_has_rt_policy(p)) {
1da177e4
LT
4535 p->static_prio = NICE_TO_PRIO(nice);
4536 goto out_unlock;
4537 }
dd41f596 4538 on_rq = p->se.on_rq;
c09595f6 4539 if (on_rq)
69be72c1 4540 dequeue_task(rq, p, 0);
1da177e4 4541
1da177e4 4542 p->static_prio = NICE_TO_PRIO(nice);
2dd73a4f 4543 set_load_weight(p);
b29739f9
IM
4544 old_prio = p->prio;
4545 p->prio = effective_prio(p);
4546 delta = p->prio - old_prio;
1da177e4 4547
dd41f596 4548 if (on_rq) {
371fd7e7 4549 enqueue_task(rq, p, 0);
1da177e4 4550 /*
d5f9f942
AM
4551 * If the task increased its priority or is running and
4552 * lowered its priority, then reschedule its CPU:
1da177e4 4553 */
d5f9f942 4554 if (delta < 0 || (delta > 0 && task_running(rq, p)))
1da177e4
LT
4555 resched_task(rq->curr);
4556 }
4557out_unlock:
4558 task_rq_unlock(rq, &flags);
4559}
1da177e4
LT
4560EXPORT_SYMBOL(set_user_nice);
4561
e43379f1
MM
4562/*
4563 * can_nice - check if a task can reduce its nice value
4564 * @p: task
4565 * @nice: nice value
4566 */
36c8b586 4567int can_nice(const struct task_struct *p, const int nice)
e43379f1 4568{
024f4747
MM
4569 /* convert nice value [19,-20] to rlimit style value [1,40] */
4570 int nice_rlim = 20 - nice;
48f24c4d 4571
78d7d407 4572 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
e43379f1
MM
4573 capable(CAP_SYS_NICE));
4574}
4575
1da177e4
LT
4576#ifdef __ARCH_WANT_SYS_NICE
4577
4578/*
4579 * sys_nice - change the priority of the current process.
4580 * @increment: priority increment
4581 *
4582 * sys_setpriority is a more generic, but much slower function that
4583 * does similar things.
4584 */
5add95d4 4585SYSCALL_DEFINE1(nice, int, increment)
1da177e4 4586{
48f24c4d 4587 long nice, retval;
1da177e4
LT
4588
4589 /*
4590 * Setpriority might change our priority at the same moment.
4591 * We don't have to worry. Conceptually one call occurs first
4592 * and we have a single winner.
4593 */
e43379f1
MM
4594 if (increment < -40)
4595 increment = -40;
1da177e4
LT
4596 if (increment > 40)
4597 increment = 40;
4598
2b8f836f 4599 nice = TASK_NICE(current) + increment;
1da177e4
LT
4600 if (nice < -20)
4601 nice = -20;
4602 if (nice > 19)
4603 nice = 19;
4604
e43379f1
MM
4605 if (increment < 0 && !can_nice(current, nice))
4606 return -EPERM;
4607
1da177e4
LT
4608 retval = security_task_setnice(current, nice);
4609 if (retval)
4610 return retval;
4611
4612 set_user_nice(current, nice);
4613 return 0;
4614}
4615
4616#endif
4617
4618/**
4619 * task_prio - return the priority value of a given task.
4620 * @p: the task in question.
4621 *
4622 * This is the priority value as seen by users in /proc.
4623 * RT tasks are offset by -200. Normal tasks are centered
4624 * around 0, value goes from -16 to +15.
4625 */
36c8b586 4626int task_prio(const struct task_struct *p)
1da177e4
LT
4627{
4628 return p->prio - MAX_RT_PRIO;
4629}
4630
4631/**
4632 * task_nice - return the nice value of a given task.
4633 * @p: the task in question.
4634 */
36c8b586 4635int task_nice(const struct task_struct *p)
1da177e4
LT
4636{
4637 return TASK_NICE(p);
4638}
150d8bed 4639EXPORT_SYMBOL(task_nice);
1da177e4
LT
4640
4641/**
4642 * idle_cpu - is a given cpu idle currently?
4643 * @cpu: the processor in question.
4644 */
4645int idle_cpu(int cpu)
4646{
4647 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4648}
4649
1da177e4
LT
4650/**
4651 * idle_task - return the idle task for a given cpu.
4652 * @cpu: the processor in question.
4653 */
36c8b586 4654struct task_struct *idle_task(int cpu)
1da177e4
LT
4655{
4656 return cpu_rq(cpu)->idle;
4657}
4658
4659/**
4660 * find_process_by_pid - find a process with a matching PID value.
4661 * @pid: the pid in question.
4662 */
a9957449 4663static struct task_struct *find_process_by_pid(pid_t pid)
1da177e4 4664{
228ebcbe 4665 return pid ? find_task_by_vpid(pid) : current;
1da177e4
LT
4666}
4667
4668/* Actually do priority change: must hold rq lock. */
dd41f596
IM
4669static void
4670__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
1da177e4 4671{
dd41f596 4672 BUG_ON(p->se.on_rq);
48f24c4d 4673
1da177e4
LT
4674 p->policy = policy;
4675 p->rt_priority = prio;
b29739f9
IM
4676 p->normal_prio = normal_prio(p);
4677 /* we are holding p->pi_lock already */
4678 p->prio = rt_mutex_getprio(p);
ffd44db5
PZ
4679 if (rt_prio(p->prio))
4680 p->sched_class = &rt_sched_class;
4681 else
4682 p->sched_class = &fair_sched_class;
2dd73a4f 4683 set_load_weight(p);
1da177e4
LT
4684}
4685
c69e8d9c
DH
4686/*
4687 * check the target process has a UID that matches the current process's
4688 */
4689static bool check_same_owner(struct task_struct *p)
4690{
4691 const struct cred *cred = current_cred(), *pcred;
4692 bool match;
4693
4694 rcu_read_lock();
4695 pcred = __task_cred(p);
4696 match = (cred->euid == pcred->euid ||
4697 cred->euid == pcred->uid);
4698 rcu_read_unlock();
4699 return match;
4700}
4701
961ccddd
RR
4702static int __sched_setscheduler(struct task_struct *p, int policy,
4703 struct sched_param *param, bool user)
1da177e4 4704{
83b699ed 4705 int retval, oldprio, oldpolicy = -1, on_rq, running;
1da177e4 4706 unsigned long flags;
83ab0aa0 4707 const struct sched_class *prev_class;
70b97a7f 4708 struct rq *rq;
ca94c442 4709 int reset_on_fork;
1da177e4 4710
66e5393a
SR
4711 /* may grab non-irq protected spin_locks */
4712 BUG_ON(in_interrupt());
1da177e4
LT
4713recheck:
4714 /* double check policy once rq lock held */
ca94c442
LP
4715 if (policy < 0) {
4716 reset_on_fork = p->sched_reset_on_fork;
1da177e4 4717 policy = oldpolicy = p->policy;
ca94c442
LP
4718 } else {
4719 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4720 policy &= ~SCHED_RESET_ON_FORK;
4721
4722 if (policy != SCHED_FIFO && policy != SCHED_RR &&
4723 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4724 policy != SCHED_IDLE)
4725 return -EINVAL;
4726 }
4727
1da177e4
LT
4728 /*
4729 * Valid priorities for SCHED_FIFO and SCHED_RR are
dd41f596
IM
4730 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4731 * SCHED_BATCH and SCHED_IDLE is 0.
1da177e4
LT
4732 */
4733 if (param->sched_priority < 0 ||
95cdf3b7 4734 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
d46523ea 4735 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
1da177e4 4736 return -EINVAL;
e05606d3 4737 if (rt_policy(policy) != (param->sched_priority != 0))
1da177e4
LT
4738 return -EINVAL;
4739
37e4ab3f
OC
4740 /*
4741 * Allow unprivileged RT tasks to decrease priority:
4742 */
961ccddd 4743 if (user && !capable(CAP_SYS_NICE)) {
e05606d3 4744 if (rt_policy(policy)) {
a44702e8
ON
4745 unsigned long rlim_rtprio =
4746 task_rlimit(p, RLIMIT_RTPRIO);
8dc3e909
ON
4747
4748 /* can't set/change the rt policy */
4749 if (policy != p->policy && !rlim_rtprio)
4750 return -EPERM;
4751
4752 /* can't increase priority */
4753 if (param->sched_priority > p->rt_priority &&
4754 param->sched_priority > rlim_rtprio)
4755 return -EPERM;
4756 }
dd41f596
IM
4757 /*
4758 * Like positive nice levels, dont allow tasks to
4759 * move out of SCHED_IDLE either:
4760 */
4761 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4762 return -EPERM;
5fe1d75f 4763
37e4ab3f 4764 /* can't change other user's priorities */
c69e8d9c 4765 if (!check_same_owner(p))
37e4ab3f 4766 return -EPERM;
ca94c442
LP
4767
4768 /* Normal users shall not reset the sched_reset_on_fork flag */
4769 if (p->sched_reset_on_fork && !reset_on_fork)
4770 return -EPERM;
37e4ab3f 4771 }
1da177e4 4772
725aad24 4773 if (user) {
725aad24
JF
4774 retval = security_task_setscheduler(p, policy, param);
4775 if (retval)
4776 return retval;
4777 }
4778
b29739f9
IM
4779 /*
4780 * make sure no PI-waiters arrive (or leave) while we are
4781 * changing the priority of the task:
4782 */
1d615482 4783 raw_spin_lock_irqsave(&p->pi_lock, flags);
1da177e4
LT
4784 /*
4785 * To be able to change p->policy safely, the apropriate
4786 * runqueue lock must be held.
4787 */
b29739f9 4788 rq = __task_rq_lock(p);
dc61b1d6 4789
34f971f6
PZ
4790 /*
4791 * Changing the policy of the stop threads its a very bad idea
4792 */
4793 if (p == rq->stop) {
4794 __task_rq_unlock(rq);
4795 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4796 return -EINVAL;
4797 }
4798
dc61b1d6
PZ
4799#ifdef CONFIG_RT_GROUP_SCHED
4800 if (user) {
4801 /*
4802 * Do not allow realtime tasks into groups that have no runtime
4803 * assigned.
4804 */
4805 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4806 task_group(p)->rt_bandwidth.rt_runtime == 0) {
4807 __task_rq_unlock(rq);
4808 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4809 return -EPERM;
4810 }
4811 }
4812#endif
4813
1da177e4
LT
4814 /* recheck policy now with rq lock held */
4815 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4816 policy = oldpolicy = -1;
b29739f9 4817 __task_rq_unlock(rq);
1d615482 4818 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
1da177e4
LT
4819 goto recheck;
4820 }
dd41f596 4821 on_rq = p->se.on_rq;
051a1d1a 4822 running = task_current(rq, p);
0e1f3483 4823 if (on_rq)
2e1cb74a 4824 deactivate_task(rq, p, 0);
0e1f3483
HS
4825 if (running)
4826 p->sched_class->put_prev_task(rq, p);
f6b53205 4827
ca94c442
LP
4828 p->sched_reset_on_fork = reset_on_fork;
4829
1da177e4 4830 oldprio = p->prio;
83ab0aa0 4831 prev_class = p->sched_class;
dd41f596 4832 __setscheduler(rq, p, policy, param->sched_priority);
f6b53205 4833
0e1f3483
HS
4834 if (running)
4835 p->sched_class->set_curr_task(rq);
dd41f596
IM
4836 if (on_rq) {
4837 activate_task(rq, p, 0);
cb469845
SR
4838
4839 check_class_changed(rq, p, prev_class, oldprio, running);
1da177e4 4840 }
b29739f9 4841 __task_rq_unlock(rq);
1d615482 4842 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
b29739f9 4843
95e02ca9
TG
4844 rt_mutex_adjust_pi(p);
4845
1da177e4
LT
4846 return 0;
4847}
961ccddd
RR
4848
4849/**
4850 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4851 * @p: the task in question.
4852 * @policy: new policy.
4853 * @param: structure containing the new RT priority.
4854 *
4855 * NOTE that the task may be already dead.
4856 */
4857int sched_setscheduler(struct task_struct *p, int policy,
4858 struct sched_param *param)
4859{
4860 return __sched_setscheduler(p, policy, param, true);
4861}
1da177e4
LT
4862EXPORT_SYMBOL_GPL(sched_setscheduler);
4863
961ccddd
RR
4864/**
4865 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4866 * @p: the task in question.
4867 * @policy: new policy.
4868 * @param: structure containing the new RT priority.
4869 *
4870 * Just like sched_setscheduler, only don't bother checking if the
4871 * current context has permission. For example, this is needed in
4872 * stop_machine(): we create temporary high priority worker threads,
4873 * but our caller might not have that capability.
4874 */
4875int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4876 struct sched_param *param)
4877{
4878 return __sched_setscheduler(p, policy, param, false);
4879}
4880
95cdf3b7
IM
4881static int
4882do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
1da177e4 4883{
1da177e4
LT
4884 struct sched_param lparam;
4885 struct task_struct *p;
36c8b586 4886 int retval;
1da177e4
LT
4887
4888 if (!param || pid < 0)
4889 return -EINVAL;
4890 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4891 return -EFAULT;
5fe1d75f
ON
4892
4893 rcu_read_lock();
4894 retval = -ESRCH;
1da177e4 4895 p = find_process_by_pid(pid);
5fe1d75f
ON
4896 if (p != NULL)
4897 retval = sched_setscheduler(p, policy, &lparam);
4898 rcu_read_unlock();
36c8b586 4899
1da177e4
LT
4900 return retval;
4901}
4902
4903/**
4904 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4905 * @pid: the pid in question.
4906 * @policy: new policy.
4907 * @param: structure containing the new RT priority.
4908 */
5add95d4
HC
4909SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4910 struct sched_param __user *, param)
1da177e4 4911{
c21761f1
JB
4912 /* negative values for policy are not valid */
4913 if (policy < 0)
4914 return -EINVAL;
4915
1da177e4
LT
4916 return do_sched_setscheduler(pid, policy, param);
4917}
4918
4919/**
4920 * sys_sched_setparam - set/change the RT priority of a thread
4921 * @pid: the pid in question.
4922 * @param: structure containing the new RT priority.
4923 */
5add95d4 4924SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
1da177e4
LT
4925{
4926 return do_sched_setscheduler(pid, -1, param);
4927}
4928
4929/**
4930 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4931 * @pid: the pid in question.
4932 */
5add95d4 4933SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
1da177e4 4934{
36c8b586 4935 struct task_struct *p;
3a5c359a 4936 int retval;
1da177e4
LT
4937
4938 if (pid < 0)
3a5c359a 4939 return -EINVAL;
1da177e4
LT
4940
4941 retval = -ESRCH;
5fe85be0 4942 rcu_read_lock();
1da177e4
LT
4943 p = find_process_by_pid(pid);
4944 if (p) {
4945 retval = security_task_getscheduler(p);
4946 if (!retval)
ca94c442
LP
4947 retval = p->policy
4948 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
1da177e4 4949 }
5fe85be0 4950 rcu_read_unlock();
1da177e4
LT
4951 return retval;
4952}
4953
4954/**
ca94c442 4955 * sys_sched_getparam - get the RT priority of a thread
1da177e4
LT
4956 * @pid: the pid in question.
4957 * @param: structure containing the RT priority.
4958 */
5add95d4 4959SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
1da177e4
LT
4960{
4961 struct sched_param lp;
36c8b586 4962 struct task_struct *p;
3a5c359a 4963 int retval;
1da177e4
LT
4964
4965 if (!param || pid < 0)
3a5c359a 4966 return -EINVAL;
1da177e4 4967
5fe85be0 4968 rcu_read_lock();
1da177e4
LT
4969 p = find_process_by_pid(pid);
4970 retval = -ESRCH;
4971 if (!p)
4972 goto out_unlock;
4973
4974 retval = security_task_getscheduler(p);
4975 if (retval)
4976 goto out_unlock;
4977
4978 lp.sched_priority = p->rt_priority;
5fe85be0 4979 rcu_read_unlock();
1da177e4
LT
4980
4981 /*
4982 * This one might sleep, we cannot do it with a spinlock held ...
4983 */
4984 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4985
1da177e4
LT
4986 return retval;
4987
4988out_unlock:
5fe85be0 4989 rcu_read_unlock();
1da177e4
LT
4990 return retval;
4991}
4992
96f874e2 4993long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
1da177e4 4994{
5a16f3d3 4995 cpumask_var_t cpus_allowed, new_mask;
36c8b586
IM
4996 struct task_struct *p;
4997 int retval;
1da177e4 4998
95402b38 4999 get_online_cpus();
23f5d142 5000 rcu_read_lock();
1da177e4
LT
5001
5002 p = find_process_by_pid(pid);
5003 if (!p) {
23f5d142 5004 rcu_read_unlock();
95402b38 5005 put_online_cpus();
1da177e4
LT
5006 return -ESRCH;
5007 }
5008
23f5d142 5009 /* Prevent p going away */
1da177e4 5010 get_task_struct(p);
23f5d142 5011 rcu_read_unlock();
1da177e4 5012
5a16f3d3
RR
5013 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5014 retval = -ENOMEM;
5015 goto out_put_task;
5016 }
5017 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5018 retval = -ENOMEM;
5019 goto out_free_cpus_allowed;
5020 }
1da177e4 5021 retval = -EPERM;
c69e8d9c 5022 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
1da177e4
LT
5023 goto out_unlock;
5024
e7834f8f
DQ
5025 retval = security_task_setscheduler(p, 0, NULL);
5026 if (retval)
5027 goto out_unlock;
5028
5a16f3d3
RR
5029 cpuset_cpus_allowed(p, cpus_allowed);
5030 cpumask_and(new_mask, in_mask, cpus_allowed);
49246274 5031again:
5a16f3d3 5032 retval = set_cpus_allowed_ptr(p, new_mask);
1da177e4 5033
8707d8b8 5034 if (!retval) {
5a16f3d3
RR
5035 cpuset_cpus_allowed(p, cpus_allowed);
5036 if (!cpumask_subset(new_mask, cpus_allowed)) {
8707d8b8
PM
5037 /*
5038 * We must have raced with a concurrent cpuset
5039 * update. Just reset the cpus_allowed to the
5040 * cpuset's cpus_allowed
5041 */
5a16f3d3 5042 cpumask_copy(new_mask, cpus_allowed);
8707d8b8
PM
5043 goto again;
5044 }
5045 }
1da177e4 5046out_unlock:
5a16f3d3
RR
5047 free_cpumask_var(new_mask);
5048out_free_cpus_allowed:
5049 free_cpumask_var(cpus_allowed);
5050out_put_task:
1da177e4 5051 put_task_struct(p);
95402b38 5052 put_online_cpus();
1da177e4
LT
5053 return retval;
5054}
5055
5056static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
96f874e2 5057 struct cpumask *new_mask)
1da177e4 5058{
96f874e2
RR
5059 if (len < cpumask_size())
5060 cpumask_clear(new_mask);
5061 else if (len > cpumask_size())
5062 len = cpumask_size();
5063
1da177e4
LT
5064 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5065}
5066
5067/**
5068 * sys_sched_setaffinity - set the cpu affinity of a process
5069 * @pid: pid of the process
5070 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5071 * @user_mask_ptr: user-space pointer to the new cpu mask
5072 */
5add95d4
HC
5073SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
5074 unsigned long __user *, user_mask_ptr)
1da177e4 5075{
5a16f3d3 5076 cpumask_var_t new_mask;
1da177e4
LT
5077 int retval;
5078
5a16f3d3
RR
5079 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5080 return -ENOMEM;
1da177e4 5081
5a16f3d3
RR
5082 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5083 if (retval == 0)
5084 retval = sched_setaffinity(pid, new_mask);
5085 free_cpumask_var(new_mask);
5086 return retval;
1da177e4
LT
5087}
5088
96f874e2 5089long sched_getaffinity(pid_t pid, struct cpumask *mask)
1da177e4 5090{
36c8b586 5091 struct task_struct *p;
31605683
TG
5092 unsigned long flags;
5093 struct rq *rq;
1da177e4 5094 int retval;
1da177e4 5095
95402b38 5096 get_online_cpus();
23f5d142 5097 rcu_read_lock();
1da177e4
LT
5098
5099 retval = -ESRCH;
5100 p = find_process_by_pid(pid);
5101 if (!p)
5102 goto out_unlock;
5103
e7834f8f
DQ
5104 retval = security_task_getscheduler(p);
5105 if (retval)
5106 goto out_unlock;
5107
31605683 5108 rq = task_rq_lock(p, &flags);
96f874e2 5109 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
31605683 5110 task_rq_unlock(rq, &flags);
1da177e4
LT
5111
5112out_unlock:
23f5d142 5113 rcu_read_unlock();
95402b38 5114 put_online_cpus();
1da177e4 5115
9531b62f 5116 return retval;
1da177e4
LT
5117}
5118
5119/**
5120 * sys_sched_getaffinity - get the cpu affinity of a process
5121 * @pid: pid of the process
5122 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5123 * @user_mask_ptr: user-space pointer to hold the current cpu mask
5124 */
5add95d4
HC
5125SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
5126 unsigned long __user *, user_mask_ptr)
1da177e4
LT
5127{
5128 int ret;
f17c8607 5129 cpumask_var_t mask;
1da177e4 5130
84fba5ec 5131 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
cd3d8031
KM
5132 return -EINVAL;
5133 if (len & (sizeof(unsigned long)-1))
1da177e4
LT
5134 return -EINVAL;
5135
f17c8607
RR
5136 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5137 return -ENOMEM;
1da177e4 5138
f17c8607
RR
5139 ret = sched_getaffinity(pid, mask);
5140 if (ret == 0) {
8bc037fb 5141 size_t retlen = min_t(size_t, len, cpumask_size());
cd3d8031
KM
5142
5143 if (copy_to_user(user_mask_ptr, mask, retlen))
f17c8607
RR
5144 ret = -EFAULT;
5145 else
cd3d8031 5146 ret = retlen;
f17c8607
RR
5147 }
5148 free_cpumask_var(mask);
1da177e4 5149
f17c8607 5150 return ret;
1da177e4
LT
5151}
5152
5153/**
5154 * sys_sched_yield - yield the current processor to other threads.
5155 *
dd41f596
IM
5156 * This function yields the current CPU to other tasks. If there are no
5157 * other threads running on this CPU then this function will return.
1da177e4 5158 */
5add95d4 5159SYSCALL_DEFINE0(sched_yield)
1da177e4 5160{
70b97a7f 5161 struct rq *rq = this_rq_lock();
1da177e4 5162
2d72376b 5163 schedstat_inc(rq, yld_count);
4530d7ab 5164 current->sched_class->yield_task(rq);
1da177e4
LT
5165
5166 /*
5167 * Since we are going to call schedule() anyway, there's
5168 * no need to preempt or enable interrupts:
5169 */
5170 __release(rq->lock);
8a25d5de 5171 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
9828ea9d 5172 do_raw_spin_unlock(&rq->lock);
1da177e4
LT
5173 preempt_enable_no_resched();
5174
5175 schedule();
5176
5177 return 0;
5178}
5179
d86ee480
PZ
5180static inline int should_resched(void)
5181{
5182 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5183}
5184
e7b38404 5185static void __cond_resched(void)
1da177e4 5186{
e7aaaa69
FW
5187 add_preempt_count(PREEMPT_ACTIVE);
5188 schedule();
5189 sub_preempt_count(PREEMPT_ACTIVE);
1da177e4
LT
5190}
5191
02b67cc3 5192int __sched _cond_resched(void)
1da177e4 5193{
d86ee480 5194 if (should_resched()) {
1da177e4
LT
5195 __cond_resched();
5196 return 1;
5197 }
5198 return 0;
5199}
02b67cc3 5200EXPORT_SYMBOL(_cond_resched);
1da177e4
LT
5201
5202/*
613afbf8 5203 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
1da177e4
LT
5204 * call schedule, and on return reacquire the lock.
5205 *
41a2d6cf 5206 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
1da177e4
LT
5207 * operations here to prevent schedule() from being called twice (once via
5208 * spin_unlock(), once by hand).
5209 */
613afbf8 5210int __cond_resched_lock(spinlock_t *lock)
1da177e4 5211{
d86ee480 5212 int resched = should_resched();
6df3cecb
JK
5213 int ret = 0;
5214
f607c668
PZ
5215 lockdep_assert_held(lock);
5216
95c354fe 5217 if (spin_needbreak(lock) || resched) {
1da177e4 5218 spin_unlock(lock);
d86ee480 5219 if (resched)
95c354fe
NP
5220 __cond_resched();
5221 else
5222 cpu_relax();
6df3cecb 5223 ret = 1;
1da177e4 5224 spin_lock(lock);
1da177e4 5225 }
6df3cecb 5226 return ret;
1da177e4 5227}
613afbf8 5228EXPORT_SYMBOL(__cond_resched_lock);
1da177e4 5229
613afbf8 5230int __sched __cond_resched_softirq(void)
1da177e4
LT
5231{
5232 BUG_ON(!in_softirq());
5233
d86ee480 5234 if (should_resched()) {
98d82567 5235 local_bh_enable();
1da177e4
LT
5236 __cond_resched();
5237 local_bh_disable();
5238 return 1;
5239 }
5240 return 0;
5241}
613afbf8 5242EXPORT_SYMBOL(__cond_resched_softirq);
1da177e4 5243
1da177e4
LT
5244/**
5245 * yield - yield the current processor to other threads.
5246 *
72fd4a35 5247 * This is a shortcut for kernel-space yielding - it marks the
1da177e4
LT
5248 * thread runnable and calls sys_sched_yield().
5249 */
5250void __sched yield(void)
5251{
5252 set_current_state(TASK_RUNNING);
5253 sys_sched_yield();
5254}
1da177e4
LT
5255EXPORT_SYMBOL(yield);
5256
5257/*
41a2d6cf 5258 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
1da177e4 5259 * that process accounting knows that this is a task in IO wait state.
1da177e4
LT
5260 */
5261void __sched io_schedule(void)
5262{
54d35f29 5263 struct rq *rq = raw_rq();
1da177e4 5264
0ff92245 5265 delayacct_blkio_start();
1da177e4 5266 atomic_inc(&rq->nr_iowait);
8f0dfc34 5267 current->in_iowait = 1;
1da177e4 5268 schedule();
8f0dfc34 5269 current->in_iowait = 0;
1da177e4 5270 atomic_dec(&rq->nr_iowait);
0ff92245 5271 delayacct_blkio_end();
1da177e4 5272}
1da177e4
LT
5273EXPORT_SYMBOL(io_schedule);
5274
5275long __sched io_schedule_timeout(long timeout)
5276{
54d35f29 5277 struct rq *rq = raw_rq();
1da177e4
LT
5278 long ret;
5279
0ff92245 5280 delayacct_blkio_start();
1da177e4 5281 atomic_inc(&rq->nr_iowait);
8f0dfc34 5282 current->in_iowait = 1;
1da177e4 5283 ret = schedule_timeout(timeout);
8f0dfc34 5284 current->in_iowait = 0;
1da177e4 5285 atomic_dec(&rq->nr_iowait);
0ff92245 5286 delayacct_blkio_end();
1da177e4
LT
5287 return ret;
5288}
5289
5290/**
5291 * sys_sched_get_priority_max - return maximum RT priority.
5292 * @policy: scheduling class.
5293 *
5294 * this syscall returns the maximum rt_priority that can be used
5295 * by a given scheduling class.
5296 */
5add95d4 5297SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
1da177e4
LT
5298{
5299 int ret = -EINVAL;
5300
5301 switch (policy) {
5302 case SCHED_FIFO:
5303 case SCHED_RR:
5304 ret = MAX_USER_RT_PRIO-1;
5305 break;
5306 case SCHED_NORMAL:
b0a9499c 5307 case SCHED_BATCH:
dd41f596 5308 case SCHED_IDLE:
1da177e4
LT
5309 ret = 0;
5310 break;
5311 }
5312 return ret;
5313}
5314
5315/**
5316 * sys_sched_get_priority_min - return minimum RT priority.
5317 * @policy: scheduling class.
5318 *
5319 * this syscall returns the minimum rt_priority that can be used
5320 * by a given scheduling class.
5321 */
5add95d4 5322SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
1da177e4
LT
5323{
5324 int ret = -EINVAL;
5325
5326 switch (policy) {
5327 case SCHED_FIFO:
5328 case SCHED_RR:
5329 ret = 1;
5330 break;
5331 case SCHED_NORMAL:
b0a9499c 5332 case SCHED_BATCH:
dd41f596 5333 case SCHED_IDLE:
1da177e4
LT
5334 ret = 0;
5335 }
5336 return ret;
5337}
5338
5339/**
5340 * sys_sched_rr_get_interval - return the default timeslice of a process.
5341 * @pid: pid of the process.
5342 * @interval: userspace pointer to the timeslice value.
5343 *
5344 * this syscall writes the default timeslice value of a given process
5345 * into the user-space timespec buffer. A value of '0' means infinity.
5346 */
17da2bd9 5347SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
754fe8d2 5348 struct timespec __user *, interval)
1da177e4 5349{
36c8b586 5350 struct task_struct *p;
a4ec24b4 5351 unsigned int time_slice;
dba091b9
TG
5352 unsigned long flags;
5353 struct rq *rq;
3a5c359a 5354 int retval;
1da177e4 5355 struct timespec t;
1da177e4
LT
5356
5357 if (pid < 0)
3a5c359a 5358 return -EINVAL;
1da177e4
LT
5359
5360 retval = -ESRCH;
1a551ae7 5361 rcu_read_lock();
1da177e4
LT
5362 p = find_process_by_pid(pid);
5363 if (!p)
5364 goto out_unlock;
5365
5366 retval = security_task_getscheduler(p);
5367 if (retval)
5368 goto out_unlock;
5369
dba091b9
TG
5370 rq = task_rq_lock(p, &flags);
5371 time_slice = p->sched_class->get_rr_interval(rq, p);
5372 task_rq_unlock(rq, &flags);
a4ec24b4 5373
1a551ae7 5374 rcu_read_unlock();
a4ec24b4 5375 jiffies_to_timespec(time_slice, &t);
1da177e4 5376 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
1da177e4 5377 return retval;
3a5c359a 5378
1da177e4 5379out_unlock:
1a551ae7 5380 rcu_read_unlock();
1da177e4
LT
5381 return retval;
5382}
5383
7c731e0a 5384static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
36c8b586 5385
82a1fcb9 5386void sched_show_task(struct task_struct *p)
1da177e4 5387{
1da177e4 5388 unsigned long free = 0;
36c8b586 5389 unsigned state;
1da177e4 5390
1da177e4 5391 state = p->state ? __ffs(p->state) + 1 : 0;
3df0fc5b 5392 printk(KERN_INFO "%-13.13s %c", p->comm,
2ed6e34f 5393 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
4bd77321 5394#if BITS_PER_LONG == 32
1da177e4 5395 if (state == TASK_RUNNING)
3df0fc5b 5396 printk(KERN_CONT " running ");
1da177e4 5397 else
3df0fc5b 5398 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
1da177e4
LT
5399#else
5400 if (state == TASK_RUNNING)
3df0fc5b 5401 printk(KERN_CONT " running task ");
1da177e4 5402 else
3df0fc5b 5403 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
1da177e4
LT
5404#endif
5405#ifdef CONFIG_DEBUG_STACK_USAGE
7c9f8861 5406 free = stack_not_used(p);
1da177e4 5407#endif
3df0fc5b 5408 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
aa47b7e0
DR
5409 task_pid_nr(p), task_pid_nr(p->real_parent),
5410 (unsigned long)task_thread_info(p)->flags);
1da177e4 5411
5fb5e6de 5412 show_stack(p, NULL);
1da177e4
LT
5413}
5414
e59e2ae2 5415void show_state_filter(unsigned long state_filter)
1da177e4 5416{
36c8b586 5417 struct task_struct *g, *p;
1da177e4 5418
4bd77321 5419#if BITS_PER_LONG == 32
3df0fc5b
PZ
5420 printk(KERN_INFO
5421 " task PC stack pid father\n");
1da177e4 5422#else
3df0fc5b
PZ
5423 printk(KERN_INFO
5424 " task PC stack pid father\n");
1da177e4
LT
5425#endif
5426 read_lock(&tasklist_lock);
5427 do_each_thread(g, p) {
5428 /*
5429 * reset the NMI-timeout, listing all files on a slow
5430 * console might take alot of time:
5431 */
5432 touch_nmi_watchdog();
39bc89fd 5433 if (!state_filter || (p->state & state_filter))
82a1fcb9 5434 sched_show_task(p);
1da177e4
LT
5435 } while_each_thread(g, p);
5436
04c9167f
JF
5437 touch_all_softlockup_watchdogs();
5438
dd41f596
IM
5439#ifdef CONFIG_SCHED_DEBUG
5440 sysrq_sched_debug_show();
5441#endif
1da177e4 5442 read_unlock(&tasklist_lock);
e59e2ae2
IM
5443 /*
5444 * Only show locks if all tasks are dumped:
5445 */
93335a21 5446 if (!state_filter)
e59e2ae2 5447 debug_show_all_locks();
1da177e4
LT
5448}
5449
1df21055
IM
5450void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5451{
dd41f596 5452 idle->sched_class = &idle_sched_class;
1df21055
IM
5453}
5454
f340c0d1
IM
5455/**
5456 * init_idle - set up an idle thread for a given CPU
5457 * @idle: task in question
5458 * @cpu: cpu the idle task belongs to
5459 *
5460 * NOTE: this function does not set the idle thread's NEED_RESCHED
5461 * flag, to make booting more robust.
5462 */
5c1e1767 5463void __cpuinit init_idle(struct task_struct *idle, int cpu)
1da177e4 5464{
70b97a7f 5465 struct rq *rq = cpu_rq(cpu);
1da177e4
LT
5466 unsigned long flags;
5467
05fa785c 5468 raw_spin_lock_irqsave(&rq->lock, flags);
5cbd54ef 5469
dd41f596 5470 __sched_fork(idle);
06b83b5f 5471 idle->state = TASK_RUNNING;
dd41f596
IM
5472 idle->se.exec_start = sched_clock();
5473
96f874e2 5474 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
dd41f596 5475 __set_task_cpu(idle, cpu);
1da177e4 5476
1da177e4 5477 rq->curr = rq->idle = idle;
4866cde0
NP
5478#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5479 idle->oncpu = 1;
5480#endif
05fa785c 5481 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4
LT
5482
5483 /* Set the preempt count _outside_ the spinlocks! */
8e3e076c
LT
5484#if defined(CONFIG_PREEMPT)
5485 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5486#else
a1261f54 5487 task_thread_info(idle)->preempt_count = 0;
8e3e076c 5488#endif
dd41f596
IM
5489 /*
5490 * The idle tasks have their own, simple scheduling class:
5491 */
5492 idle->sched_class = &idle_sched_class;
fb52607a 5493 ftrace_graph_init_task(idle);
1da177e4
LT
5494}
5495
5496/*
5497 * In a system that switches off the HZ timer nohz_cpu_mask
5498 * indicates which cpus entered this state. This is used
5499 * in the rcu update to wait only for active cpus. For system
5500 * which do not switch off the HZ timer nohz_cpu_mask should
6a7b3dc3 5501 * always be CPU_BITS_NONE.
1da177e4 5502 */
6a7b3dc3 5503cpumask_var_t nohz_cpu_mask;
1da177e4 5504
19978ca6
IM
5505/*
5506 * Increase the granularity value when there are more CPUs,
5507 * because with more CPUs the 'effective latency' as visible
5508 * to users decreases. But the relationship is not linear,
5509 * so pick a second-best guess by going with the log2 of the
5510 * number of CPUs.
5511 *
5512 * This idea comes from the SD scheduler of Con Kolivas:
5513 */
acb4a848 5514static int get_update_sysctl_factor(void)
19978ca6 5515{
4ca3ef71 5516 unsigned int cpus = min_t(int, num_online_cpus(), 8);
1983a922
CE
5517 unsigned int factor;
5518
5519 switch (sysctl_sched_tunable_scaling) {
5520 case SCHED_TUNABLESCALING_NONE:
5521 factor = 1;
5522 break;
5523 case SCHED_TUNABLESCALING_LINEAR:
5524 factor = cpus;
5525 break;
5526 case SCHED_TUNABLESCALING_LOG:
5527 default:
5528 factor = 1 + ilog2(cpus);
5529 break;
5530 }
19978ca6 5531
acb4a848
CE
5532 return factor;
5533}
19978ca6 5534
acb4a848
CE
5535static void update_sysctl(void)
5536{
5537 unsigned int factor = get_update_sysctl_factor();
19978ca6 5538
0bcdcf28
CE
5539#define SET_SYSCTL(name) \
5540 (sysctl_##name = (factor) * normalized_sysctl_##name)
5541 SET_SYSCTL(sched_min_granularity);
5542 SET_SYSCTL(sched_latency);
5543 SET_SYSCTL(sched_wakeup_granularity);
5544 SET_SYSCTL(sched_shares_ratelimit);
5545#undef SET_SYSCTL
5546}
55cd5340 5547
0bcdcf28
CE
5548static inline void sched_init_granularity(void)
5549{
5550 update_sysctl();
19978ca6
IM
5551}
5552
1da177e4
LT
5553#ifdef CONFIG_SMP
5554/*
5555 * This is how migration works:
5556 *
969c7921
TH
5557 * 1) we invoke migration_cpu_stop() on the target CPU using
5558 * stop_one_cpu().
5559 * 2) stopper starts to run (implicitly forcing the migrated thread
5560 * off the CPU)
5561 * 3) it checks whether the migrated task is still in the wrong runqueue.
5562 * 4) if it's in the wrong runqueue then the migration thread removes
1da177e4 5563 * it and puts it into the right queue.
969c7921
TH
5564 * 5) stopper completes and stop_one_cpu() returns and the migration
5565 * is done.
1da177e4
LT
5566 */
5567
5568/*
5569 * Change a given task's CPU affinity. Migrate the thread to a
5570 * proper CPU and schedule it away if the CPU it's executing on
5571 * is removed from the allowed bitmask.
5572 *
5573 * NOTE: the caller must have a valid reference to the task, the
41a2d6cf 5574 * task must not exit() & deallocate itself prematurely. The
1da177e4
LT
5575 * call is not atomic; no spinlocks may be held.
5576 */
96f874e2 5577int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1da177e4
LT
5578{
5579 unsigned long flags;
70b97a7f 5580 struct rq *rq;
969c7921 5581 unsigned int dest_cpu;
48f24c4d 5582 int ret = 0;
1da177e4 5583
65cc8e48
PZ
5584 /*
5585 * Serialize against TASK_WAKING so that ttwu() and wunt() can
5586 * drop the rq->lock and still rely on ->cpus_allowed.
5587 */
5588again:
5589 while (task_is_waking(p))
5590 cpu_relax();
1da177e4 5591 rq = task_rq_lock(p, &flags);
65cc8e48
PZ
5592 if (task_is_waking(p)) {
5593 task_rq_unlock(rq, &flags);
5594 goto again;
5595 }
e2912009 5596
6ad4c188 5597 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
1da177e4
LT
5598 ret = -EINVAL;
5599 goto out;
5600 }
5601
9985b0ba 5602 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
96f874e2 5603 !cpumask_equal(&p->cpus_allowed, new_mask))) {
9985b0ba
DR
5604 ret = -EINVAL;
5605 goto out;
5606 }
5607
73fe6aae 5608 if (p->sched_class->set_cpus_allowed)
cd8ba7cd 5609 p->sched_class->set_cpus_allowed(p, new_mask);
73fe6aae 5610 else {
96f874e2
RR
5611 cpumask_copy(&p->cpus_allowed, new_mask);
5612 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
73fe6aae
GH
5613 }
5614
1da177e4 5615 /* Can the task run on the task's current CPU? If so, we're done */
96f874e2 5616 if (cpumask_test_cpu(task_cpu(p), new_mask))
1da177e4
LT
5617 goto out;
5618
969c7921
TH
5619 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
5620 if (migrate_task(p, dest_cpu)) {
5621 struct migration_arg arg = { p, dest_cpu };
1da177e4
LT
5622 /* Need help from migration thread: drop lock and wait. */
5623 task_rq_unlock(rq, &flags);
969c7921 5624 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1da177e4
LT
5625 tlb_migrate_finish(p->mm);
5626 return 0;
5627 }
5628out:
5629 task_rq_unlock(rq, &flags);
48f24c4d 5630
1da177e4
LT
5631 return ret;
5632}
cd8ba7cd 5633EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
1da177e4
LT
5634
5635/*
41a2d6cf 5636 * Move (not current) task off this cpu, onto dest cpu. We're doing
1da177e4
LT
5637 * this because either it can't run here any more (set_cpus_allowed()
5638 * away from this CPU, or CPU going down), or because we're
5639 * attempting to rebalance this task on exec (sched_exec).
5640 *
5641 * So we race with normal scheduler movements, but that's OK, as long
5642 * as the task is no longer on this CPU.
efc30814
KK
5643 *
5644 * Returns non-zero if task was successfully migrated.
1da177e4 5645 */
efc30814 5646static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
1da177e4 5647{
70b97a7f 5648 struct rq *rq_dest, *rq_src;
e2912009 5649 int ret = 0;
1da177e4 5650
e761b772 5651 if (unlikely(!cpu_active(dest_cpu)))
efc30814 5652 return ret;
1da177e4
LT
5653
5654 rq_src = cpu_rq(src_cpu);
5655 rq_dest = cpu_rq(dest_cpu);
5656
5657 double_rq_lock(rq_src, rq_dest);
5658 /* Already moved. */
5659 if (task_cpu(p) != src_cpu)
b1e38734 5660 goto done;
1da177e4 5661 /* Affinity changed (again). */
96f874e2 5662 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
b1e38734 5663 goto fail;
1da177e4 5664
e2912009
PZ
5665 /*
5666 * If we're not on a rq, the next wake-up will ensure we're
5667 * placed properly.
5668 */
5669 if (p->se.on_rq) {
2e1cb74a 5670 deactivate_task(rq_src, p, 0);
e2912009 5671 set_task_cpu(p, dest_cpu);
dd41f596 5672 activate_task(rq_dest, p, 0);
15afe09b 5673 check_preempt_curr(rq_dest, p, 0);
1da177e4 5674 }
b1e38734 5675done:
efc30814 5676 ret = 1;
b1e38734 5677fail:
1da177e4 5678 double_rq_unlock(rq_src, rq_dest);
efc30814 5679 return ret;
1da177e4
LT
5680}
5681
5682/*
969c7921
TH
5683 * migration_cpu_stop - this will be executed by a highprio stopper thread
5684 * and performs thread migration by bumping thread off CPU then
5685 * 'pushing' onto another runqueue.
1da177e4 5686 */
969c7921 5687static int migration_cpu_stop(void *data)
1da177e4 5688{
969c7921 5689 struct migration_arg *arg = data;
f7b4cddc 5690
969c7921
TH
5691 /*
5692 * The original target cpu might have gone down and we might
5693 * be on another cpu but it doesn't matter.
5694 */
f7b4cddc 5695 local_irq_disable();
969c7921 5696 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
f7b4cddc 5697 local_irq_enable();
1da177e4 5698 return 0;
f7b4cddc
ON
5699}
5700
1da177e4 5701#ifdef CONFIG_HOTPLUG_CPU
054b9108 5702/*
3a4fa0a2 5703 * Figure out where task on dead CPU should go, use force if necessary.
054b9108 5704 */
6a1bdc1b 5705void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
1da177e4 5706{
1445c08d
ON
5707 struct rq *rq = cpu_rq(dead_cpu);
5708 int needs_cpu, uninitialized_var(dest_cpu);
5709 unsigned long flags;
e76bd8d9 5710
1445c08d 5711 local_irq_save(flags);
e76bd8d9 5712
1445c08d
ON
5713 raw_spin_lock(&rq->lock);
5714 needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING);
5715 if (needs_cpu)
5716 dest_cpu = select_fallback_rq(dead_cpu, p);
5717 raw_spin_unlock(&rq->lock);
c1804d54
ON
5718 /*
5719 * It can only fail if we race with set_cpus_allowed(),
5720 * in the racer should migrate the task anyway.
5721 */
1445c08d 5722 if (needs_cpu)
c1804d54 5723 __migrate_task(p, dead_cpu, dest_cpu);
1445c08d 5724 local_irq_restore(flags);
1da177e4
LT
5725}
5726
5727/*
5728 * While a dead CPU has no uninterruptible tasks queued at this point,
5729 * it might still have a nonzero ->nr_uninterruptible counter, because
5730 * for performance reasons the counter is not stricly tracking tasks to
5731 * their home CPUs. So we just add the counter to another CPU's counter,
5732 * to keep the global sum constant after CPU-down:
5733 */
70b97a7f 5734static void migrate_nr_uninterruptible(struct rq *rq_src)
1da177e4 5735{
6ad4c188 5736 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
1da177e4
LT
5737 unsigned long flags;
5738
5739 local_irq_save(flags);
5740 double_rq_lock(rq_src, rq_dest);
5741 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5742 rq_src->nr_uninterruptible = 0;
5743 double_rq_unlock(rq_src, rq_dest);
5744 local_irq_restore(flags);
5745}
5746
5747/* Run through task list and migrate tasks from the dead cpu. */
5748static void migrate_live_tasks(int src_cpu)
5749{
48f24c4d 5750 struct task_struct *p, *t;
1da177e4 5751
f7b4cddc 5752 read_lock(&tasklist_lock);
1da177e4 5753
48f24c4d
IM
5754 do_each_thread(t, p) {
5755 if (p == current)
1da177e4
LT
5756 continue;
5757
48f24c4d
IM
5758 if (task_cpu(p) == src_cpu)
5759 move_task_off_dead_cpu(src_cpu, p);
5760 } while_each_thread(t, p);
1da177e4 5761
f7b4cddc 5762 read_unlock(&tasklist_lock);
1da177e4
LT
5763}
5764
dd41f596
IM
5765/*
5766 * Schedules idle task to be the next runnable task on current CPU.
94bc9a7b
DA
5767 * It does so by boosting its priority to highest possible.
5768 * Used by CPU offline code.
1da177e4
LT
5769 */
5770void sched_idle_next(void)
5771{
48f24c4d 5772 int this_cpu = smp_processor_id();
70b97a7f 5773 struct rq *rq = cpu_rq(this_cpu);
1da177e4
LT
5774 struct task_struct *p = rq->idle;
5775 unsigned long flags;
5776
5777 /* cpu has to be offline */
48f24c4d 5778 BUG_ON(cpu_online(this_cpu));
1da177e4 5779
48f24c4d
IM
5780 /*
5781 * Strictly not necessary since rest of the CPUs are stopped by now
5782 * and interrupts disabled on the current cpu.
1da177e4 5783 */
05fa785c 5784 raw_spin_lock_irqsave(&rq->lock, flags);
1da177e4 5785
dd41f596 5786 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
48f24c4d 5787
94bc9a7b 5788 activate_task(rq, p, 0);
1da177e4 5789
05fa785c 5790 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4
LT
5791}
5792
48f24c4d
IM
5793/*
5794 * Ensures that the idle task is using init_mm right before its cpu goes
1da177e4
LT
5795 * offline.
5796 */
5797void idle_task_exit(void)
5798{
5799 struct mm_struct *mm = current->active_mm;
5800
5801 BUG_ON(cpu_online(smp_processor_id()));
5802
5803 if (mm != &init_mm)
5804 switch_mm(mm, &init_mm, current);
5805 mmdrop(mm);
5806}
5807
054b9108 5808/* called under rq->lock with disabled interrupts */
36c8b586 5809static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
1da177e4 5810{
70b97a7f 5811 struct rq *rq = cpu_rq(dead_cpu);
1da177e4
LT
5812
5813 /* Must be exiting, otherwise would be on tasklist. */
270f722d 5814 BUG_ON(!p->exit_state);
1da177e4
LT
5815
5816 /* Cannot have done final schedule yet: would have vanished. */
c394cc9f 5817 BUG_ON(p->state == TASK_DEAD);
1da177e4 5818
48f24c4d 5819 get_task_struct(p);
1da177e4
LT
5820
5821 /*
5822 * Drop lock around migration; if someone else moves it,
41a2d6cf 5823 * that's OK. No task can be added to this CPU, so iteration is
1da177e4
LT
5824 * fine.
5825 */
05fa785c 5826 raw_spin_unlock_irq(&rq->lock);
48f24c4d 5827 move_task_off_dead_cpu(dead_cpu, p);
05fa785c 5828 raw_spin_lock_irq(&rq->lock);
1da177e4 5829
48f24c4d 5830 put_task_struct(p);
1da177e4
LT
5831}
5832
5833/* release_task() removes task from tasklist, so we won't find dead tasks. */
5834static void migrate_dead_tasks(unsigned int dead_cpu)
5835{
70b97a7f 5836 struct rq *rq = cpu_rq(dead_cpu);
dd41f596 5837 struct task_struct *next;
48f24c4d 5838
dd41f596
IM
5839 for ( ; ; ) {
5840 if (!rq->nr_running)
5841 break;
b67802ea 5842 next = pick_next_task(rq);
dd41f596
IM
5843 if (!next)
5844 break;
79c53799 5845 next->sched_class->put_prev_task(rq, next);
dd41f596 5846 migrate_dead(dead_cpu, next);
e692ab53 5847
1da177e4
LT
5848 }
5849}
dce48a84
TG
5850
5851/*
5852 * remove the tasks which were accounted by rq from calc_load_tasks.
5853 */
5854static void calc_global_load_remove(struct rq *rq)
5855{
5856 atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
a468d389 5857 rq->calc_load_active = 0;
dce48a84 5858}
1da177e4
LT
5859#endif /* CONFIG_HOTPLUG_CPU */
5860
e692ab53
NP
5861#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5862
5863static struct ctl_table sd_ctl_dir[] = {
e0361851
AD
5864 {
5865 .procname = "sched_domain",
c57baf1e 5866 .mode = 0555,
e0361851 5867 },
56992309 5868 {}
e692ab53
NP
5869};
5870
5871static struct ctl_table sd_ctl_root[] = {
e0361851
AD
5872 {
5873 .procname = "kernel",
c57baf1e 5874 .mode = 0555,
e0361851
AD
5875 .child = sd_ctl_dir,
5876 },
56992309 5877 {}
e692ab53
NP
5878};
5879
5880static struct ctl_table *sd_alloc_ctl_entry(int n)
5881{
5882 struct ctl_table *entry =
5cf9f062 5883 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
e692ab53 5884
e692ab53
NP
5885 return entry;
5886}
5887
6382bc90
MM
5888static void sd_free_ctl_entry(struct ctl_table **tablep)
5889{
cd790076 5890 struct ctl_table *entry;
6382bc90 5891
cd790076
MM
5892 /*
5893 * In the intermediate directories, both the child directory and
5894 * procname are dynamically allocated and could fail but the mode
41a2d6cf 5895 * will always be set. In the lowest directory the names are
cd790076
MM
5896 * static strings and all have proc handlers.
5897 */
5898 for (entry = *tablep; entry->mode; entry++) {
6382bc90
MM
5899 if (entry->child)
5900 sd_free_ctl_entry(&entry->child);
cd790076
MM
5901 if (entry->proc_handler == NULL)
5902 kfree(entry->procname);
5903 }
6382bc90
MM
5904
5905 kfree(*tablep);
5906 *tablep = NULL;
5907}
5908
e692ab53 5909static void
e0361851 5910set_table_entry(struct ctl_table *entry,
e692ab53
NP
5911 const char *procname, void *data, int maxlen,
5912 mode_t mode, proc_handler *proc_handler)
5913{
e692ab53
NP
5914 entry->procname = procname;
5915 entry->data = data;
5916 entry->maxlen = maxlen;
5917 entry->mode = mode;
5918 entry->proc_handler = proc_handler;
5919}
5920
5921static struct ctl_table *
5922sd_alloc_ctl_domain_table(struct sched_domain *sd)
5923{
a5d8c348 5924 struct ctl_table *table = sd_alloc_ctl_entry(13);
e692ab53 5925
ad1cdc1d
MM
5926 if (table == NULL)
5927 return NULL;
5928
e0361851 5929 set_table_entry(&table[0], "min_interval", &sd->min_interval,
e692ab53 5930 sizeof(long), 0644, proc_doulongvec_minmax);
e0361851 5931 set_table_entry(&table[1], "max_interval", &sd->max_interval,
e692ab53 5932 sizeof(long), 0644, proc_doulongvec_minmax);
e0361851 5933 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
e692ab53 5934 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5935 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
e692ab53 5936 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5937 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
e692ab53 5938 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5939 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
e692ab53 5940 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5941 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
e692ab53 5942 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5943 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
e692ab53 5944 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5945 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
e692ab53 5946 sizeof(int), 0644, proc_dointvec_minmax);
ace8b3d6 5947 set_table_entry(&table[9], "cache_nice_tries",
e692ab53
NP
5948 &sd->cache_nice_tries,
5949 sizeof(int), 0644, proc_dointvec_minmax);
ace8b3d6 5950 set_table_entry(&table[10], "flags", &sd->flags,
e692ab53 5951 sizeof(int), 0644, proc_dointvec_minmax);
a5d8c348
IM
5952 set_table_entry(&table[11], "name", sd->name,
5953 CORENAME_MAX_SIZE, 0444, proc_dostring);
5954 /* &table[12] is terminator */
e692ab53
NP
5955
5956 return table;
5957}
5958
9a4e7159 5959static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
e692ab53
NP
5960{
5961 struct ctl_table *entry, *table;
5962 struct sched_domain *sd;
5963 int domain_num = 0, i;
5964 char buf[32];
5965
5966 for_each_domain(cpu, sd)
5967 domain_num++;
5968 entry = table = sd_alloc_ctl_entry(domain_num + 1);
ad1cdc1d
MM
5969 if (table == NULL)
5970 return NULL;
e692ab53
NP
5971
5972 i = 0;
5973 for_each_domain(cpu, sd) {
5974 snprintf(buf, 32, "domain%d", i);
e692ab53 5975 entry->procname = kstrdup(buf, GFP_KERNEL);
c57baf1e 5976 entry->mode = 0555;
e692ab53
NP
5977 entry->child = sd_alloc_ctl_domain_table(sd);
5978 entry++;
5979 i++;
5980 }
5981 return table;
5982}
5983
5984static struct ctl_table_header *sd_sysctl_header;
6382bc90 5985static void register_sched_domain_sysctl(void)
e692ab53 5986{
6ad4c188 5987 int i, cpu_num = num_possible_cpus();
e692ab53
NP
5988 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
5989 char buf[32];
5990
7378547f
MM
5991 WARN_ON(sd_ctl_dir[0].child);
5992 sd_ctl_dir[0].child = entry;
5993
ad1cdc1d
MM
5994 if (entry == NULL)
5995 return;
5996
6ad4c188 5997 for_each_possible_cpu(i) {
e692ab53 5998 snprintf(buf, 32, "cpu%d", i);
e692ab53 5999 entry->procname = kstrdup(buf, GFP_KERNEL);
c57baf1e 6000 entry->mode = 0555;
e692ab53 6001 entry->child = sd_alloc_ctl_cpu_table(i);
97b6ea7b 6002 entry++;
e692ab53 6003 }
7378547f
MM
6004
6005 WARN_ON(sd_sysctl_header);
e692ab53
NP
6006 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
6007}
6382bc90 6008
7378547f 6009/* may be called multiple times per register */
6382bc90
MM
6010static void unregister_sched_domain_sysctl(void)
6011{
7378547f
MM
6012 if (sd_sysctl_header)
6013 unregister_sysctl_table(sd_sysctl_header);
6382bc90 6014 sd_sysctl_header = NULL;
7378547f
MM
6015 if (sd_ctl_dir[0].child)
6016 sd_free_ctl_entry(&sd_ctl_dir[0].child);
6382bc90 6017}
e692ab53 6018#else
6382bc90
MM
6019static void register_sched_domain_sysctl(void)
6020{
6021}
6022static void unregister_sched_domain_sysctl(void)
e692ab53
NP
6023{
6024}
6025#endif
6026
1f11eb6a
GH
6027static void set_rq_online(struct rq *rq)
6028{
6029 if (!rq->online) {
6030 const struct sched_class *class;
6031
c6c4927b 6032 cpumask_set_cpu(rq->cpu, rq->rd->online);
1f11eb6a
GH
6033 rq->online = 1;
6034
6035 for_each_class(class) {
6036 if (class->rq_online)
6037 class->rq_online(rq);
6038 }
6039 }
6040}
6041
6042static void set_rq_offline(struct rq *rq)
6043{
6044 if (rq->online) {
6045 const struct sched_class *class;
6046
6047 for_each_class(class) {
6048 if (class->rq_offline)
6049 class->rq_offline(rq);
6050 }
6051
c6c4927b 6052 cpumask_clear_cpu(rq->cpu, rq->rd->online);
1f11eb6a
GH
6053 rq->online = 0;
6054 }
6055}
6056
1da177e4
LT
6057/*
6058 * migration_call - callback that gets triggered when a CPU is added.
6059 * Here we can start up the necessary migration thread for the new CPU.
6060 */
48f24c4d
IM
6061static int __cpuinit
6062migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
1da177e4 6063{
48f24c4d 6064 int cpu = (long)hcpu;
1da177e4 6065 unsigned long flags;
969c7921 6066 struct rq *rq = cpu_rq(cpu);
1da177e4
LT
6067
6068 switch (action) {
5be9361c 6069
1da177e4 6070 case CPU_UP_PREPARE:
8bb78442 6071 case CPU_UP_PREPARE_FROZEN:
a468d389 6072 rq->calc_load_update = calc_load_update;
1da177e4 6073 break;
48f24c4d 6074
1da177e4 6075 case CPU_ONLINE:
8bb78442 6076 case CPU_ONLINE_FROZEN:
1f94ef59 6077 /* Update our root-domain */
05fa785c 6078 raw_spin_lock_irqsave(&rq->lock, flags);
1f94ef59 6079 if (rq->rd) {
c6c4927b 6080 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
1f11eb6a
GH
6081
6082 set_rq_online(rq);
1f94ef59 6083 }
05fa785c 6084 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4 6085 break;
48f24c4d 6086
1da177e4 6087#ifdef CONFIG_HOTPLUG_CPU
1da177e4 6088 case CPU_DEAD:
8bb78442 6089 case CPU_DEAD_FROZEN:
1da177e4 6090 migrate_live_tasks(cpu);
1da177e4 6091 /* Idle task back to normal (off runqueue, low prio) */
05fa785c 6092 raw_spin_lock_irq(&rq->lock);
2e1cb74a 6093 deactivate_task(rq, rq->idle, 0);
dd41f596
IM
6094 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
6095 rq->idle->sched_class = &idle_sched_class;
1da177e4 6096 migrate_dead_tasks(cpu);
05fa785c 6097 raw_spin_unlock_irq(&rq->lock);
1da177e4
LT
6098 migrate_nr_uninterruptible(rq);
6099 BUG_ON(rq->nr_running != 0);
dce48a84 6100 calc_global_load_remove(rq);
1da177e4 6101 break;
57d885fe 6102
08f503b0
GH
6103 case CPU_DYING:
6104 case CPU_DYING_FROZEN:
57d885fe 6105 /* Update our root-domain */
05fa785c 6106 raw_spin_lock_irqsave(&rq->lock, flags);
57d885fe 6107 if (rq->rd) {
c6c4927b 6108 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
1f11eb6a 6109 set_rq_offline(rq);
57d885fe 6110 }
05fa785c 6111 raw_spin_unlock_irqrestore(&rq->lock, flags);
57d885fe 6112 break;
1da177e4
LT
6113#endif
6114 }
6115 return NOTIFY_OK;
6116}
6117
f38b0820
PM
6118/*
6119 * Register at high priority so that task migration (migrate_all_tasks)
6120 * happens before everything else. This has to be lower priority than
cdd6c482 6121 * the notifier in the perf_event subsystem, though.
1da177e4 6122 */
26c2143b 6123static struct notifier_block __cpuinitdata migration_notifier = {
1da177e4 6124 .notifier_call = migration_call,
50a323b7 6125 .priority = CPU_PRI_MIGRATION,
1da177e4
LT
6126};
6127
3a101d05
TH
6128static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
6129 unsigned long action, void *hcpu)
6130{
6131 switch (action & ~CPU_TASKS_FROZEN) {
6132 case CPU_ONLINE:
6133 case CPU_DOWN_FAILED:
6134 set_cpu_active((long)hcpu, true);
6135 return NOTIFY_OK;
6136 default:
6137 return NOTIFY_DONE;
6138 }
6139}
6140
6141static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
6142 unsigned long action, void *hcpu)
6143{
6144 switch (action & ~CPU_TASKS_FROZEN) {
6145 case CPU_DOWN_PREPARE:
6146 set_cpu_active((long)hcpu, false);
6147 return NOTIFY_OK;
6148 default:
6149 return NOTIFY_DONE;
6150 }
6151}
6152
7babe8db 6153static int __init migration_init(void)
1da177e4
LT
6154{
6155 void *cpu = (void *)(long)smp_processor_id();
07dccf33 6156 int err;
48f24c4d 6157
3a101d05 6158 /* Initialize migration for the boot CPU */
07dccf33
AM
6159 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6160 BUG_ON(err == NOTIFY_BAD);
1da177e4
LT
6161 migration_call(&migration_notifier, CPU_ONLINE, cpu);
6162 register_cpu_notifier(&migration_notifier);
7babe8db 6163
3a101d05
TH
6164 /* Register cpu active notifiers */
6165 cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
6166 cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
6167
a004cd42 6168 return 0;
1da177e4 6169}
7babe8db 6170early_initcall(migration_init);
1da177e4
LT
6171#endif
6172
6173#ifdef CONFIG_SMP
476f3534 6174
3e9830dc 6175#ifdef CONFIG_SCHED_DEBUG
4dcf6aff 6176
f6630114
MT
6177static __read_mostly int sched_domain_debug_enabled;
6178
6179static int __init sched_domain_debug_setup(char *str)
6180{
6181 sched_domain_debug_enabled = 1;
6182
6183 return 0;
6184}
6185early_param("sched_debug", sched_domain_debug_setup);
6186
7c16ec58 6187static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
96f874e2 6188 struct cpumask *groupmask)
1da177e4 6189{
4dcf6aff 6190 struct sched_group *group = sd->groups;
434d53b0 6191 char str[256];
1da177e4 6192
968ea6d8 6193 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
96f874e2 6194 cpumask_clear(groupmask);
4dcf6aff
IM
6195
6196 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6197
6198 if (!(sd->flags & SD_LOAD_BALANCE)) {
3df0fc5b 6199 printk("does not load-balance\n");
4dcf6aff 6200 if (sd->parent)
3df0fc5b
PZ
6201 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6202 " has parent");
4dcf6aff 6203 return -1;
41c7ce9a
NP
6204 }
6205
3df0fc5b 6206 printk(KERN_CONT "span %s level %s\n", str, sd->name);
4dcf6aff 6207
758b2cdc 6208 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
3df0fc5b
PZ
6209 printk(KERN_ERR "ERROR: domain->span does not contain "
6210 "CPU%d\n", cpu);
4dcf6aff 6211 }
758b2cdc 6212 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
3df0fc5b
PZ
6213 printk(KERN_ERR "ERROR: domain->groups does not contain"
6214 " CPU%d\n", cpu);
4dcf6aff 6215 }
1da177e4 6216
4dcf6aff 6217 printk(KERN_DEBUG "%*s groups:", level + 1, "");
1da177e4 6218 do {
4dcf6aff 6219 if (!group) {
3df0fc5b
PZ
6220 printk("\n");
6221 printk(KERN_ERR "ERROR: group is NULL\n");
1da177e4
LT
6222 break;
6223 }
6224
18a3885f 6225 if (!group->cpu_power) {
3df0fc5b
PZ
6226 printk(KERN_CONT "\n");
6227 printk(KERN_ERR "ERROR: domain->cpu_power not "
6228 "set\n");
4dcf6aff
IM
6229 break;
6230 }
1da177e4 6231
758b2cdc 6232 if (!cpumask_weight(sched_group_cpus(group))) {
3df0fc5b
PZ
6233 printk(KERN_CONT "\n");
6234 printk(KERN_ERR "ERROR: empty group\n");
4dcf6aff
IM
6235 break;
6236 }
1da177e4 6237
758b2cdc 6238 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
3df0fc5b
PZ
6239 printk(KERN_CONT "\n");
6240 printk(KERN_ERR "ERROR: repeated CPUs\n");
4dcf6aff
IM
6241 break;
6242 }
1da177e4 6243
758b2cdc 6244 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
1da177e4 6245
968ea6d8 6246 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
381512cf 6247
3df0fc5b 6248 printk(KERN_CONT " %s", str);
18a3885f 6249 if (group->cpu_power != SCHED_LOAD_SCALE) {
3df0fc5b
PZ
6250 printk(KERN_CONT " (cpu_power = %d)",
6251 group->cpu_power);
381512cf 6252 }
1da177e4 6253
4dcf6aff
IM
6254 group = group->next;
6255 } while (group != sd->groups);
3df0fc5b 6256 printk(KERN_CONT "\n");
1da177e4 6257
758b2cdc 6258 if (!cpumask_equal(sched_domain_span(sd), groupmask))
3df0fc5b 6259 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
1da177e4 6260
758b2cdc
RR
6261 if (sd->parent &&
6262 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
3df0fc5b
PZ
6263 printk(KERN_ERR "ERROR: parent span is not a superset "
6264 "of domain->span\n");
4dcf6aff
IM
6265 return 0;
6266}
1da177e4 6267
4dcf6aff
IM
6268static void sched_domain_debug(struct sched_domain *sd, int cpu)
6269{
d5dd3db1 6270 cpumask_var_t groupmask;
4dcf6aff 6271 int level = 0;
1da177e4 6272
f6630114
MT
6273 if (!sched_domain_debug_enabled)
6274 return;
6275
4dcf6aff
IM
6276 if (!sd) {
6277 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6278 return;
6279 }
1da177e4 6280
4dcf6aff
IM
6281 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6282
d5dd3db1 6283 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
7c16ec58
MT
6284 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6285 return;
6286 }
6287
4dcf6aff 6288 for (;;) {
7c16ec58 6289 if (sched_domain_debug_one(sd, cpu, level, groupmask))
4dcf6aff 6290 break;
1da177e4
LT
6291 level++;
6292 sd = sd->parent;
33859f7f 6293 if (!sd)
4dcf6aff
IM
6294 break;
6295 }
d5dd3db1 6296 free_cpumask_var(groupmask);
1da177e4 6297}
6d6bc0ad 6298#else /* !CONFIG_SCHED_DEBUG */
48f24c4d 6299# define sched_domain_debug(sd, cpu) do { } while (0)
6d6bc0ad 6300#endif /* CONFIG_SCHED_DEBUG */
1da177e4 6301
1a20ff27 6302static int sd_degenerate(struct sched_domain *sd)
245af2c7 6303{
758b2cdc 6304 if (cpumask_weight(sched_domain_span(sd)) == 1)
245af2c7
SS
6305 return 1;
6306
6307 /* Following flags need at least 2 groups */
6308 if (sd->flags & (SD_LOAD_BALANCE |
6309 SD_BALANCE_NEWIDLE |
6310 SD_BALANCE_FORK |
89c4710e
SS
6311 SD_BALANCE_EXEC |
6312 SD_SHARE_CPUPOWER |
6313 SD_SHARE_PKG_RESOURCES)) {
245af2c7
SS
6314 if (sd->groups != sd->groups->next)
6315 return 0;
6316 }
6317
6318 /* Following flags don't use groups */
c88d5910 6319 if (sd->flags & (SD_WAKE_AFFINE))
245af2c7
SS
6320 return 0;
6321
6322 return 1;
6323}
6324
48f24c4d
IM
6325static int
6326sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
245af2c7
SS
6327{
6328 unsigned long cflags = sd->flags, pflags = parent->flags;
6329
6330 if (sd_degenerate(parent))
6331 return 1;
6332
758b2cdc 6333 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
245af2c7
SS
6334 return 0;
6335
245af2c7
SS
6336 /* Flags needing groups don't count if only 1 group in parent */
6337 if (parent->groups == parent->groups->next) {
6338 pflags &= ~(SD_LOAD_BALANCE |
6339 SD_BALANCE_NEWIDLE |
6340 SD_BALANCE_FORK |
89c4710e
SS
6341 SD_BALANCE_EXEC |
6342 SD_SHARE_CPUPOWER |
6343 SD_SHARE_PKG_RESOURCES);
5436499e
KC
6344 if (nr_node_ids == 1)
6345 pflags &= ~SD_SERIALIZE;
245af2c7
SS
6346 }
6347 if (~cflags & pflags)
6348 return 0;
6349
6350 return 1;
6351}
6352
c6c4927b
RR
6353static void free_rootdomain(struct root_domain *rd)
6354{
047106ad
PZ
6355 synchronize_sched();
6356
68e74568
RR
6357 cpupri_cleanup(&rd->cpupri);
6358
c6c4927b
RR
6359 free_cpumask_var(rd->rto_mask);
6360 free_cpumask_var(rd->online);
6361 free_cpumask_var(rd->span);
6362 kfree(rd);
6363}
6364
57d885fe
GH
6365static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6366{
a0490fa3 6367 struct root_domain *old_rd = NULL;
57d885fe 6368 unsigned long flags;
57d885fe 6369
05fa785c 6370 raw_spin_lock_irqsave(&rq->lock, flags);
57d885fe
GH
6371
6372 if (rq->rd) {
a0490fa3 6373 old_rd = rq->rd;
57d885fe 6374
c6c4927b 6375 if (cpumask_test_cpu(rq->cpu, old_rd->online))
1f11eb6a 6376 set_rq_offline(rq);
57d885fe 6377
c6c4927b 6378 cpumask_clear_cpu(rq->cpu, old_rd->span);
dc938520 6379
a0490fa3
IM
6380 /*
6381 * If we dont want to free the old_rt yet then
6382 * set old_rd to NULL to skip the freeing later
6383 * in this function:
6384 */
6385 if (!atomic_dec_and_test(&old_rd->refcount))
6386 old_rd = NULL;
57d885fe
GH
6387 }
6388
6389 atomic_inc(&rd->refcount);
6390 rq->rd = rd;
6391
c6c4927b 6392 cpumask_set_cpu(rq->cpu, rd->span);
00aec93d 6393 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
1f11eb6a 6394 set_rq_online(rq);
57d885fe 6395
05fa785c 6396 raw_spin_unlock_irqrestore(&rq->lock, flags);
a0490fa3
IM
6397
6398 if (old_rd)
6399 free_rootdomain(old_rd);
57d885fe
GH
6400}
6401
68c38fc3 6402static int init_rootdomain(struct root_domain *rd)
57d885fe
GH
6403{
6404 memset(rd, 0, sizeof(*rd));
6405
68c38fc3 6406 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
0c910d28 6407 goto out;
68c38fc3 6408 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
c6c4927b 6409 goto free_span;
68c38fc3 6410 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
c6c4927b 6411 goto free_online;
6e0534f2 6412
68c38fc3 6413 if (cpupri_init(&rd->cpupri) != 0)
68e74568 6414 goto free_rto_mask;
c6c4927b 6415 return 0;
6e0534f2 6416
68e74568
RR
6417free_rto_mask:
6418 free_cpumask_var(rd->rto_mask);
c6c4927b
RR
6419free_online:
6420 free_cpumask_var(rd->online);
6421free_span:
6422 free_cpumask_var(rd->span);
0c910d28 6423out:
c6c4927b 6424 return -ENOMEM;
57d885fe
GH
6425}
6426
6427static void init_defrootdomain(void)
6428{
68c38fc3 6429 init_rootdomain(&def_root_domain);
c6c4927b 6430
57d885fe
GH
6431 atomic_set(&def_root_domain.refcount, 1);
6432}
6433
dc938520 6434static struct root_domain *alloc_rootdomain(void)
57d885fe
GH
6435{
6436 struct root_domain *rd;
6437
6438 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6439 if (!rd)
6440 return NULL;
6441
68c38fc3 6442 if (init_rootdomain(rd) != 0) {
c6c4927b
RR
6443 kfree(rd);
6444 return NULL;
6445 }
57d885fe
GH
6446
6447 return rd;
6448}
6449
1da177e4 6450/*
0eab9146 6451 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
1da177e4
LT
6452 * hold the hotplug lock.
6453 */
0eab9146
IM
6454static void
6455cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
1da177e4 6456{
70b97a7f 6457 struct rq *rq = cpu_rq(cpu);
245af2c7
SS
6458 struct sched_domain *tmp;
6459
669c55e9
PZ
6460 for (tmp = sd; tmp; tmp = tmp->parent)
6461 tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
6462
245af2c7 6463 /* Remove the sched domains which do not contribute to scheduling. */
f29c9b1c 6464 for (tmp = sd; tmp; ) {
245af2c7
SS
6465 struct sched_domain *parent = tmp->parent;
6466 if (!parent)
6467 break;
f29c9b1c 6468
1a848870 6469 if (sd_parent_degenerate(tmp, parent)) {
245af2c7 6470 tmp->parent = parent->parent;
1a848870
SS
6471 if (parent->parent)
6472 parent->parent->child = tmp;
f29c9b1c
LZ
6473 } else
6474 tmp = tmp->parent;
245af2c7
SS
6475 }
6476
1a848870 6477 if (sd && sd_degenerate(sd)) {
245af2c7 6478 sd = sd->parent;
1a848870
SS
6479 if (sd)
6480 sd->child = NULL;
6481 }
1da177e4
LT
6482
6483 sched_domain_debug(sd, cpu);
6484
57d885fe 6485 rq_attach_root(rq, rd);
674311d5 6486 rcu_assign_pointer(rq->sd, sd);
1da177e4
LT
6487}
6488
6489/* cpus with isolated domains */
dcc30a35 6490static cpumask_var_t cpu_isolated_map;
1da177e4
LT
6491
6492/* Setup the mask of cpus configured for isolated domains */
6493static int __init isolated_cpu_setup(char *str)
6494{
bdddd296 6495 alloc_bootmem_cpumask_var(&cpu_isolated_map);
968ea6d8 6496 cpulist_parse(str, cpu_isolated_map);
1da177e4
LT
6497 return 1;
6498}
6499
8927f494 6500__setup("isolcpus=", isolated_cpu_setup);
1da177e4
LT
6501
6502/*
6711cab4
SS
6503 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6504 * to a function which identifies what group(along with sched group) a CPU
96f874e2
RR
6505 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6506 * (due to the fact that we keep track of groups covered with a struct cpumask).
1da177e4
LT
6507 *
6508 * init_sched_build_groups will build a circular linked list of the groups
6509 * covered by the given span, and will set each group's ->cpumask correctly,
6510 * and ->cpu_power to 0.
6511 */
a616058b 6512static void
96f874e2
RR
6513init_sched_build_groups(const struct cpumask *span,
6514 const struct cpumask *cpu_map,
6515 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
7c16ec58 6516 struct sched_group **sg,
96f874e2
RR
6517 struct cpumask *tmpmask),
6518 struct cpumask *covered, struct cpumask *tmpmask)
1da177e4
LT
6519{
6520 struct sched_group *first = NULL, *last = NULL;
1da177e4
LT
6521 int i;
6522
96f874e2 6523 cpumask_clear(covered);
7c16ec58 6524
abcd083a 6525 for_each_cpu(i, span) {
6711cab4 6526 struct sched_group *sg;
7c16ec58 6527 int group = group_fn(i, cpu_map, &sg, tmpmask);
1da177e4
LT
6528 int j;
6529
758b2cdc 6530 if (cpumask_test_cpu(i, covered))
1da177e4
LT
6531 continue;
6532
758b2cdc 6533 cpumask_clear(sched_group_cpus(sg));
18a3885f 6534 sg->cpu_power = 0;
1da177e4 6535
abcd083a 6536 for_each_cpu(j, span) {
7c16ec58 6537 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
1da177e4
LT
6538 continue;
6539
96f874e2 6540 cpumask_set_cpu(j, covered);
758b2cdc 6541 cpumask_set_cpu(j, sched_group_cpus(sg));
1da177e4
LT
6542 }
6543 if (!first)
6544 first = sg;
6545 if (last)
6546 last->next = sg;
6547 last = sg;
6548 }
6549 last->next = first;
6550}
6551
9c1cfda2 6552#define SD_NODES_PER_DOMAIN 16
1da177e4 6553
9c1cfda2 6554#ifdef CONFIG_NUMA
198e2f18 6555
9c1cfda2
JH
6556/**
6557 * find_next_best_node - find the next node to include in a sched_domain
6558 * @node: node whose sched_domain we're building
6559 * @used_nodes: nodes already in the sched_domain
6560 *
41a2d6cf 6561 * Find the next node to include in a given scheduling domain. Simply
9c1cfda2
JH
6562 * finds the closest node not already in the @used_nodes map.
6563 *
6564 * Should use nodemask_t.
6565 */
c5f59f08 6566static int find_next_best_node(int node, nodemask_t *used_nodes)
9c1cfda2
JH
6567{
6568 int i, n, val, min_val, best_node = 0;
6569
6570 min_val = INT_MAX;
6571
076ac2af 6572 for (i = 0; i < nr_node_ids; i++) {
9c1cfda2 6573 /* Start at @node */
076ac2af 6574 n = (node + i) % nr_node_ids;
9c1cfda2
JH
6575
6576 if (!nr_cpus_node(n))
6577 continue;
6578
6579 /* Skip already used nodes */
c5f59f08 6580 if (node_isset(n, *used_nodes))
9c1cfda2
JH
6581 continue;
6582
6583 /* Simple min distance search */
6584 val = node_distance(node, n);
6585
6586 if (val < min_val) {
6587 min_val = val;
6588 best_node = n;
6589 }
6590 }
6591
c5f59f08 6592 node_set(best_node, *used_nodes);
9c1cfda2
JH
6593 return best_node;
6594}
6595
6596/**
6597 * sched_domain_node_span - get a cpumask for a node's sched_domain
6598 * @node: node whose cpumask we're constructing
73486722 6599 * @span: resulting cpumask
9c1cfda2 6600 *
41a2d6cf 6601 * Given a node, construct a good cpumask for its sched_domain to span. It
9c1cfda2
JH
6602 * should be one that prevents unnecessary balancing, but also spreads tasks
6603 * out optimally.
6604 */
96f874e2 6605static void sched_domain_node_span(int node, struct cpumask *span)
9c1cfda2 6606{
c5f59f08 6607 nodemask_t used_nodes;
48f24c4d 6608 int i;
9c1cfda2 6609
6ca09dfc 6610 cpumask_clear(span);
c5f59f08 6611 nodes_clear(used_nodes);
9c1cfda2 6612
6ca09dfc 6613 cpumask_or(span, span, cpumask_of_node(node));
c5f59f08 6614 node_set(node, used_nodes);
9c1cfda2
JH
6615
6616 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
c5f59f08 6617 int next_node = find_next_best_node(node, &used_nodes);
48f24c4d 6618
6ca09dfc 6619 cpumask_or(span, span, cpumask_of_node(next_node));
9c1cfda2 6620 }
9c1cfda2 6621}
6d6bc0ad 6622#endif /* CONFIG_NUMA */
9c1cfda2 6623
5c45bf27 6624int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
48f24c4d 6625
6c99e9ad
RR
6626/*
6627 * The cpus mask in sched_group and sched_domain hangs off the end.
4200efd9
IM
6628 *
6629 * ( See the the comments in include/linux/sched.h:struct sched_group
6630 * and struct sched_domain. )
6c99e9ad
RR
6631 */
6632struct static_sched_group {
6633 struct sched_group sg;
6634 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
6635};
6636
6637struct static_sched_domain {
6638 struct sched_domain sd;
6639 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
6640};
6641
49a02c51
AH
6642struct s_data {
6643#ifdef CONFIG_NUMA
6644 int sd_allnodes;
6645 cpumask_var_t domainspan;
6646 cpumask_var_t covered;
6647 cpumask_var_t notcovered;
6648#endif
6649 cpumask_var_t nodemask;
6650 cpumask_var_t this_sibling_map;
6651 cpumask_var_t this_core_map;
01a08546 6652 cpumask_var_t this_book_map;
49a02c51
AH
6653 cpumask_var_t send_covered;
6654 cpumask_var_t tmpmask;
6655 struct sched_group **sched_group_nodes;
6656 struct root_domain *rd;
6657};
6658
2109b99e
AH
6659enum s_alloc {
6660 sa_sched_groups = 0,
6661 sa_rootdomain,
6662 sa_tmpmask,
6663 sa_send_covered,
01a08546 6664 sa_this_book_map,
2109b99e
AH
6665 sa_this_core_map,
6666 sa_this_sibling_map,
6667 sa_nodemask,
6668 sa_sched_group_nodes,
6669#ifdef CONFIG_NUMA
6670 sa_notcovered,
6671 sa_covered,
6672 sa_domainspan,
6673#endif
6674 sa_none,
6675};
6676
9c1cfda2 6677/*
48f24c4d 6678 * SMT sched-domains:
9c1cfda2 6679 */
1da177e4 6680#ifdef CONFIG_SCHED_SMT
6c99e9ad 6681static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
1871e52c 6682static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
48f24c4d 6683
41a2d6cf 6684static int
96f874e2
RR
6685cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6686 struct sched_group **sg, struct cpumask *unused)
1da177e4 6687{
6711cab4 6688 if (sg)
1871e52c 6689 *sg = &per_cpu(sched_groups, cpu).sg;
1da177e4
LT
6690 return cpu;
6691}
6d6bc0ad 6692#endif /* CONFIG_SCHED_SMT */
1da177e4 6693
48f24c4d
IM
6694/*
6695 * multi-core sched-domains:
6696 */
1e9f28fa 6697#ifdef CONFIG_SCHED_MC
6c99e9ad
RR
6698static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6699static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
1e9f28fa 6700
41a2d6cf 6701static int
96f874e2
RR
6702cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6703 struct sched_group **sg, struct cpumask *mask)
1e9f28fa 6704{
6711cab4 6705 int group;
f269893c 6706#ifdef CONFIG_SCHED_SMT
c69fc56d 6707 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
96f874e2 6708 group = cpumask_first(mask);
f269893c
HC
6709#else
6710 group = cpu;
6711#endif
6711cab4 6712 if (sg)
6c99e9ad 6713 *sg = &per_cpu(sched_group_core, group).sg;
6711cab4 6714 return group;
1e9f28fa 6715}
f269893c 6716#endif /* CONFIG_SCHED_MC */
1e9f28fa 6717
01a08546
HC
6718/*
6719 * book sched-domains:
6720 */
6721#ifdef CONFIG_SCHED_BOOK
6722static DEFINE_PER_CPU(struct static_sched_domain, book_domains);
6723static DEFINE_PER_CPU(struct static_sched_group, sched_group_book);
6724
6725static int
6726cpu_to_book_group(int cpu, const struct cpumask *cpu_map,
6727 struct sched_group **sg, struct cpumask *mask)
6728{
6729 int group = cpu;
6730#ifdef CONFIG_SCHED_MC
6731 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
6732 group = cpumask_first(mask);
6733#elif defined(CONFIG_SCHED_SMT)
6734 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
6735 group = cpumask_first(mask);
6736#endif
6737 if (sg)
6738 *sg = &per_cpu(sched_group_book, group).sg;
6739 return group;
6740}
6741#endif /* CONFIG_SCHED_BOOK */
6742
6c99e9ad
RR
6743static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6744static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
48f24c4d 6745
41a2d6cf 6746static int
96f874e2
RR
6747cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6748 struct sched_group **sg, struct cpumask *mask)
1da177e4 6749{
6711cab4 6750 int group;
01a08546
HC
6751#ifdef CONFIG_SCHED_BOOK
6752 cpumask_and(mask, cpu_book_mask(cpu), cpu_map);
6753 group = cpumask_first(mask);
6754#elif defined(CONFIG_SCHED_MC)
6ca09dfc 6755 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
96f874e2 6756 group = cpumask_first(mask);
1e9f28fa 6757#elif defined(CONFIG_SCHED_SMT)
c69fc56d 6758 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
96f874e2 6759 group = cpumask_first(mask);
1da177e4 6760#else
6711cab4 6761 group = cpu;
1da177e4 6762#endif
6711cab4 6763 if (sg)
6c99e9ad 6764 *sg = &per_cpu(sched_group_phys, group).sg;
6711cab4 6765 return group;
1da177e4
LT
6766}
6767
6768#ifdef CONFIG_NUMA
1da177e4 6769/*
9c1cfda2
JH
6770 * The init_sched_build_groups can't handle what we want to do with node
6771 * groups, so roll our own. Now each node has its own list of groups which
6772 * gets dynamically allocated.
1da177e4 6773 */
62ea9ceb 6774static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
434d53b0 6775static struct sched_group ***sched_group_nodes_bycpu;
1da177e4 6776
62ea9ceb 6777static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
6c99e9ad 6778static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
9c1cfda2 6779
96f874e2
RR
6780static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
6781 struct sched_group **sg,
6782 struct cpumask *nodemask)
9c1cfda2 6783{
6711cab4
SS
6784 int group;
6785
6ca09dfc 6786 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
96f874e2 6787 group = cpumask_first(nodemask);
6711cab4
SS
6788
6789 if (sg)
6c99e9ad 6790 *sg = &per_cpu(sched_group_allnodes, group).sg;
6711cab4 6791 return group;
1da177e4 6792}
6711cab4 6793
08069033
SS
6794static void init_numa_sched_groups_power(struct sched_group *group_head)
6795{
6796 struct sched_group *sg = group_head;
6797 int j;
6798
6799 if (!sg)
6800 return;
3a5c359a 6801 do {
758b2cdc 6802 for_each_cpu(j, sched_group_cpus(sg)) {
3a5c359a 6803 struct sched_domain *sd;
08069033 6804
6c99e9ad 6805 sd = &per_cpu(phys_domains, j).sd;
13318a71 6806 if (j != group_first_cpu(sd->groups)) {
3a5c359a
AK
6807 /*
6808 * Only add "power" once for each
6809 * physical package.
6810 */
6811 continue;
6812 }
08069033 6813
18a3885f 6814 sg->cpu_power += sd->groups->cpu_power;
3a5c359a
AK
6815 }
6816 sg = sg->next;
6817 } while (sg != group_head);
08069033 6818}
0601a88d
AH
6819
6820static int build_numa_sched_groups(struct s_data *d,
6821 const struct cpumask *cpu_map, int num)
6822{
6823 struct sched_domain *sd;
6824 struct sched_group *sg, *prev;
6825 int n, j;
6826
6827 cpumask_clear(d->covered);
6828 cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
6829 if (cpumask_empty(d->nodemask)) {
6830 d->sched_group_nodes[num] = NULL;
6831 goto out;
6832 }
6833
6834 sched_domain_node_span(num, d->domainspan);
6835 cpumask_and(d->domainspan, d->domainspan, cpu_map);
6836
6837 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6838 GFP_KERNEL, num);
6839 if (!sg) {
3df0fc5b
PZ
6840 printk(KERN_WARNING "Can not alloc domain group for node %d\n",
6841 num);
0601a88d
AH
6842 return -ENOMEM;
6843 }
6844 d->sched_group_nodes[num] = sg;
6845
6846 for_each_cpu(j, d->nodemask) {
6847 sd = &per_cpu(node_domains, j).sd;
6848 sd->groups = sg;
6849 }
6850
18a3885f 6851 sg->cpu_power = 0;
0601a88d
AH
6852 cpumask_copy(sched_group_cpus(sg), d->nodemask);
6853 sg->next = sg;
6854 cpumask_or(d->covered, d->covered, d->nodemask);
6855
6856 prev = sg;
6857 for (j = 0; j < nr_node_ids; j++) {
6858 n = (num + j) % nr_node_ids;
6859 cpumask_complement(d->notcovered, d->covered);
6860 cpumask_and(d->tmpmask, d->notcovered, cpu_map);
6861 cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
6862 if (cpumask_empty(d->tmpmask))
6863 break;
6864 cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
6865 if (cpumask_empty(d->tmpmask))
6866 continue;
6867 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6868 GFP_KERNEL, num);
6869 if (!sg) {
3df0fc5b
PZ
6870 printk(KERN_WARNING
6871 "Can not alloc domain group for node %d\n", j);
0601a88d
AH
6872 return -ENOMEM;
6873 }
18a3885f 6874 sg->cpu_power = 0;
0601a88d
AH
6875 cpumask_copy(sched_group_cpus(sg), d->tmpmask);
6876 sg->next = prev->next;
6877 cpumask_or(d->covered, d->covered, d->tmpmask);
6878 prev->next = sg;
6879 prev = sg;
6880 }
6881out:
6882 return 0;
6883}
6d6bc0ad 6884#endif /* CONFIG_NUMA */
1da177e4 6885
a616058b 6886#ifdef CONFIG_NUMA
51888ca2 6887/* Free memory allocated for various sched_group structures */
96f874e2
RR
6888static void free_sched_groups(const struct cpumask *cpu_map,
6889 struct cpumask *nodemask)
51888ca2 6890{
a616058b 6891 int cpu, i;
51888ca2 6892
abcd083a 6893 for_each_cpu(cpu, cpu_map) {
51888ca2
SV
6894 struct sched_group **sched_group_nodes
6895 = sched_group_nodes_bycpu[cpu];
6896
51888ca2
SV
6897 if (!sched_group_nodes)
6898 continue;
6899
076ac2af 6900 for (i = 0; i < nr_node_ids; i++) {
51888ca2
SV
6901 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6902
6ca09dfc 6903 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
96f874e2 6904 if (cpumask_empty(nodemask))
51888ca2
SV
6905 continue;
6906
6907 if (sg == NULL)
6908 continue;
6909 sg = sg->next;
6910next_sg:
6911 oldsg = sg;
6912 sg = sg->next;
6913 kfree(oldsg);
6914 if (oldsg != sched_group_nodes[i])
6915 goto next_sg;
6916 }
6917 kfree(sched_group_nodes);
6918 sched_group_nodes_bycpu[cpu] = NULL;
6919 }
51888ca2 6920}
6d6bc0ad 6921#else /* !CONFIG_NUMA */
96f874e2
RR
6922static void free_sched_groups(const struct cpumask *cpu_map,
6923 struct cpumask *nodemask)
a616058b
SS
6924{
6925}
6d6bc0ad 6926#endif /* CONFIG_NUMA */
51888ca2 6927
89c4710e
SS
6928/*
6929 * Initialize sched groups cpu_power.
6930 *
6931 * cpu_power indicates the capacity of sched group, which is used while
6932 * distributing the load between different sched groups in a sched domain.
6933 * Typically cpu_power for all the groups in a sched domain will be same unless
6934 * there are asymmetries in the topology. If there are asymmetries, group
6935 * having more cpu_power will pickup more load compared to the group having
6936 * less cpu_power.
89c4710e
SS
6937 */
6938static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6939{
6940 struct sched_domain *child;
6941 struct sched_group *group;
f93e65c1
PZ
6942 long power;
6943 int weight;
89c4710e
SS
6944
6945 WARN_ON(!sd || !sd->groups);
6946
13318a71 6947 if (cpu != group_first_cpu(sd->groups))
89c4710e
SS
6948 return;
6949
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
DG
7839static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
7840 struct sched_entity *se, int cpu, int add,
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;
7847 if (add)
7848 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
7849
7850 tg->se[cpu] = se;
354d60c2
DG
7851 /* se could be NULL for init_task_group */
7852 if (!se)
7853 return;
7854
ec7dc8ac
DG
7855 if (!parent)
7856 se->cfs_rq = &rq->cfs;
7857 else
7858 se->cfs_rq = parent->my_q;
7859
6f505b16
PZ
7860 se->my_q = cfs_rq;
7861 se->load.weight = tg->shares;
e05510d0 7862 se->load.inv_weight = 0;
ec7dc8ac 7863 se->parent = parent;
6f505b16 7864}
052f1dc7 7865#endif
6f505b16 7866
052f1dc7 7867#ifdef CONFIG_RT_GROUP_SCHED
ec7dc8ac
DG
7868static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
7869 struct sched_rt_entity *rt_se, int cpu, int add,
7870 struct sched_rt_entity *parent)
6f505b16 7871{
ec7dc8ac
DG
7872 struct rq *rq = cpu_rq(cpu);
7873
6f505b16
PZ
7874 tg->rt_rq[cpu] = rt_rq;
7875 init_rt_rq(rt_rq, rq);
7876 rt_rq->tg = tg;
ac086bc2 7877 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
6f505b16
PZ
7878 if (add)
7879 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
7880
7881 tg->rt_se[cpu] = rt_se;
354d60c2
DG
7882 if (!rt_se)
7883 return;
7884
ec7dc8ac
DG
7885 if (!parent)
7886 rt_se->rt_rq = &rq->rt;
7887 else
7888 rt_se->rt_rq = parent->my_q;
7889
6f505b16 7890 rt_se->my_q = rt_rq;
ec7dc8ac 7891 rt_se->parent = parent;
6f505b16
PZ
7892 INIT_LIST_HEAD(&rt_se->run_list);
7893}
7894#endif
7895
1da177e4
LT
7896void __init sched_init(void)
7897{
dd41f596 7898 int i, j;
434d53b0
MT
7899 unsigned long alloc_size = 0, ptr;
7900
7901#ifdef CONFIG_FAIR_GROUP_SCHED
7902 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7903#endif
7904#ifdef CONFIG_RT_GROUP_SCHED
7905 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
eff766a6 7906#endif
df7c8e84 7907#ifdef CONFIG_CPUMASK_OFFSTACK
8c083f08 7908 alloc_size += num_possible_cpus() * cpumask_size();
434d53b0 7909#endif
434d53b0 7910 if (alloc_size) {
36b7b6d4 7911 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
434d53b0
MT
7912
7913#ifdef CONFIG_FAIR_GROUP_SCHED
7914 init_task_group.se = (struct sched_entity **)ptr;
7915 ptr += nr_cpu_ids * sizeof(void **);
7916
7917 init_task_group.cfs_rq = (struct cfs_rq **)ptr;
7918 ptr += nr_cpu_ids * sizeof(void **);
eff766a6 7919
6d6bc0ad 7920#endif /* CONFIG_FAIR_GROUP_SCHED */
434d53b0
MT
7921#ifdef CONFIG_RT_GROUP_SCHED
7922 init_task_group.rt_se = (struct sched_rt_entity **)ptr;
7923 ptr += nr_cpu_ids * sizeof(void **);
7924
7925 init_task_group.rt_rq = (struct rt_rq **)ptr;
eff766a6
PZ
7926 ptr += nr_cpu_ids * sizeof(void **);
7927
6d6bc0ad 7928#endif /* CONFIG_RT_GROUP_SCHED */
df7c8e84
RR
7929#ifdef CONFIG_CPUMASK_OFFSTACK
7930 for_each_possible_cpu(i) {
7931 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
7932 ptr += cpumask_size();
7933 }
7934#endif /* CONFIG_CPUMASK_OFFSTACK */
434d53b0 7935 }
dd41f596 7936
57d885fe
GH
7937#ifdef CONFIG_SMP
7938 init_defrootdomain();
7939#endif
7940
d0b27fa7
PZ
7941 init_rt_bandwidth(&def_rt_bandwidth,
7942 global_rt_period(), global_rt_runtime());
7943
7944#ifdef CONFIG_RT_GROUP_SCHED
7945 init_rt_bandwidth(&init_task_group.rt_bandwidth,
7946 global_rt_period(), global_rt_runtime());
6d6bc0ad 7947#endif /* CONFIG_RT_GROUP_SCHED */
d0b27fa7 7948
7c941438 7949#ifdef CONFIG_CGROUP_SCHED
6f505b16 7950 list_add(&init_task_group.list, &task_groups);
f473aa5e
PZ
7951 INIT_LIST_HEAD(&init_task_group.children);
7952
7c941438 7953#endif /* CONFIG_CGROUP_SCHED */
6f505b16 7954
4a6cc4bd
JK
7955#if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP
7956 update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long),
7957 __alignof__(unsigned long));
7958#endif
0a945022 7959 for_each_possible_cpu(i) {
70b97a7f 7960 struct rq *rq;
1da177e4
LT
7961
7962 rq = cpu_rq(i);
05fa785c 7963 raw_spin_lock_init(&rq->lock);
7897986b 7964 rq->nr_running = 0;
dce48a84
TG
7965 rq->calc_load_active = 0;
7966 rq->calc_load_update = jiffies + LOAD_FREQ;
dd41f596 7967 init_cfs_rq(&rq->cfs, rq);
6f505b16 7968 init_rt_rq(&rq->rt, rq);
dd41f596 7969#ifdef CONFIG_FAIR_GROUP_SCHED
4cf86d77 7970 init_task_group.shares = init_task_group_load;
6f505b16 7971 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
354d60c2
DG
7972#ifdef CONFIG_CGROUP_SCHED
7973 /*
7974 * How much cpu bandwidth does init_task_group get?
7975 *
7976 * In case of task-groups formed thr' the cgroup filesystem, it
7977 * gets 100% of the cpu resources in the system. This overall
7978 * system cpu resource is divided among the tasks of
7979 * init_task_group and its child task-groups in a fair manner,
7980 * based on each entity's (task or task-group's) weight
7981 * (se->load.weight).
7982 *
7983 * In other words, if init_task_group has 10 tasks of weight
7984 * 1024) and two child groups A0 and A1 (of weight 1024 each),
7985 * then A0's share of the cpu resource is:
7986 *
0d905bca 7987 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
354d60c2
DG
7988 *
7989 * We achieve this by letting init_task_group's tasks sit
7990 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
7991 */
ec7dc8ac 7992 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL);
052f1dc7 7993#endif
354d60c2
DG
7994#endif /* CONFIG_FAIR_GROUP_SCHED */
7995
7996 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
052f1dc7 7997#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 7998 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
354d60c2 7999#ifdef CONFIG_CGROUP_SCHED
ec7dc8ac 8000 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL);
354d60c2 8001#endif
dd41f596 8002#endif
1da177e4 8003
dd41f596
IM
8004 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
8005 rq->cpu_load[j] = 0;
fdf3e95d
VP
8006
8007 rq->last_load_update_tick = jiffies;
8008
1da177e4 8009#ifdef CONFIG_SMP
41c7ce9a 8010 rq->sd = NULL;
57d885fe 8011 rq->rd = NULL;
e51fd5e2 8012 rq->cpu_power = SCHED_LOAD_SCALE;
3f029d3c 8013 rq->post_schedule = 0;
1da177e4 8014 rq->active_balance = 0;
dd41f596 8015 rq->next_balance = jiffies;
1da177e4 8016 rq->push_cpu = 0;
0a2966b4 8017 rq->cpu = i;
1f11eb6a 8018 rq->online = 0;
eae0c9df
MG
8019 rq->idle_stamp = 0;
8020 rq->avg_idle = 2*sysctl_sched_migration_cost;
dc938520 8021 rq_attach_root(rq, &def_root_domain);
83cd4fe2
VP
8022#ifdef CONFIG_NO_HZ
8023 rq->nohz_balance_kick = 0;
8024 init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i));
8025#endif
1da177e4 8026#endif
8f4d37ec 8027 init_rq_hrtick(rq);
1da177e4 8028 atomic_set(&rq->nr_iowait, 0);
1da177e4
LT
8029 }
8030
2dd73a4f 8031 set_load_weight(&init_task);
b50f60ce 8032
e107be36
AK
8033#ifdef CONFIG_PREEMPT_NOTIFIERS
8034 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
8035#endif
8036
c9819f45 8037#ifdef CONFIG_SMP
962cf36c 8038 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
c9819f45
CL
8039#endif
8040
b50f60ce 8041#ifdef CONFIG_RT_MUTEXES
1d615482 8042 plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock);
b50f60ce
HC
8043#endif
8044
1da177e4
LT
8045 /*
8046 * The boot idle thread does lazy MMU switching as well:
8047 */
8048 atomic_inc(&init_mm.mm_count);
8049 enter_lazy_tlb(&init_mm, current);
8050
8051 /*
8052 * Make us the idle thread. Technically, schedule() should not be
8053 * called from this thread, however somewhere below it might be,
8054 * but because we are the idle thread, we just pick up running again
8055 * when this runqueue becomes "idle".
8056 */
8057 init_idle(current, smp_processor_id());
dce48a84
TG
8058
8059 calc_load_update = jiffies + LOAD_FREQ;
8060
dd41f596
IM
8061 /*
8062 * During early bootup we pretend to be a normal task:
8063 */
8064 current->sched_class = &fair_sched_class;
6892b75e 8065
6a7b3dc3 8066 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
49557e62 8067 zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT);
bf4d83f6 8068#ifdef CONFIG_SMP
7d1e6a9b 8069#ifdef CONFIG_NO_HZ
83cd4fe2
VP
8070 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
8071 alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT);
8072 atomic_set(&nohz.load_balancer, nr_cpu_ids);
8073 atomic_set(&nohz.first_pick_cpu, nr_cpu_ids);
8074 atomic_set(&nohz.second_pick_cpu, nr_cpu_ids);
7d1e6a9b 8075#endif
bdddd296
RR
8076 /* May be allocated at isolcpus cmdline parse time */
8077 if (cpu_isolated_map == NULL)
8078 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
bf4d83f6 8079#endif /* SMP */
6a7b3dc3 8080
cdd6c482 8081 perf_event_init();
0d905bca 8082
6892b75e 8083 scheduler_running = 1;
1da177e4
LT
8084}
8085
8086#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
e4aafea2
FW
8087static inline int preempt_count_equals(int preempt_offset)
8088{
234da7bc 8089 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
e4aafea2
FW
8090
8091 return (nested == PREEMPT_INATOMIC_BASE + preempt_offset);
8092}
8093
d894837f 8094void __might_sleep(const char *file, int line, int preempt_offset)
1da177e4 8095{
48f24c4d 8096#ifdef in_atomic
1da177e4
LT
8097 static unsigned long prev_jiffy; /* ratelimiting */
8098
e4aafea2
FW
8099 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
8100 system_state != SYSTEM_RUNNING || oops_in_progress)
aef745fc
IM
8101 return;
8102 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
8103 return;
8104 prev_jiffy = jiffies;
8105
3df0fc5b
PZ
8106 printk(KERN_ERR
8107 "BUG: sleeping function called from invalid context at %s:%d\n",
8108 file, line);
8109 printk(KERN_ERR
8110 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
8111 in_atomic(), irqs_disabled(),
8112 current->pid, current->comm);
aef745fc
IM
8113
8114 debug_show_held_locks(current);
8115 if (irqs_disabled())
8116 print_irqtrace_events(current);
8117 dump_stack();
1da177e4
LT
8118#endif
8119}
8120EXPORT_SYMBOL(__might_sleep);
8121#endif
8122
8123#ifdef CONFIG_MAGIC_SYSRQ
3a5e4dc1
AK
8124static void normalize_task(struct rq *rq, struct task_struct *p)
8125{
8126 int on_rq;
3e51f33f 8127
3a5e4dc1
AK
8128 on_rq = p->se.on_rq;
8129 if (on_rq)
8130 deactivate_task(rq, p, 0);
8131 __setscheduler(rq, p, SCHED_NORMAL, 0);
8132 if (on_rq) {
8133 activate_task(rq, p, 0);
8134 resched_task(rq->curr);
8135 }
8136}
8137
1da177e4
LT
8138void normalize_rt_tasks(void)
8139{
a0f98a1c 8140 struct task_struct *g, *p;
1da177e4 8141 unsigned long flags;
70b97a7f 8142 struct rq *rq;
1da177e4 8143
4cf5d77a 8144 read_lock_irqsave(&tasklist_lock, flags);
a0f98a1c 8145 do_each_thread(g, p) {
178be793
IM
8146 /*
8147 * Only normalize user tasks:
8148 */
8149 if (!p->mm)
8150 continue;
8151
6cfb0d5d 8152 p->se.exec_start = 0;
6cfb0d5d 8153#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
8154 p->se.statistics.wait_start = 0;
8155 p->se.statistics.sleep_start = 0;
8156 p->se.statistics.block_start = 0;
6cfb0d5d 8157#endif
dd41f596
IM
8158
8159 if (!rt_task(p)) {
8160 /*
8161 * Renice negative nice level userspace
8162 * tasks back to 0:
8163 */
8164 if (TASK_NICE(p) < 0 && p->mm)
8165 set_user_nice(p, 0);
1da177e4 8166 continue;
dd41f596 8167 }
1da177e4 8168
1d615482 8169 raw_spin_lock(&p->pi_lock);
b29739f9 8170 rq = __task_rq_lock(p);
1da177e4 8171
178be793 8172 normalize_task(rq, p);
3a5e4dc1 8173
b29739f9 8174 __task_rq_unlock(rq);
1d615482 8175 raw_spin_unlock(&p->pi_lock);
a0f98a1c
IM
8176 } while_each_thread(g, p);
8177
4cf5d77a 8178 read_unlock_irqrestore(&tasklist_lock, flags);
1da177e4
LT
8179}
8180
8181#endif /* CONFIG_MAGIC_SYSRQ */
1df5c10a 8182
67fc4e0c 8183#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
1df5c10a 8184/*
67fc4e0c 8185 * These functions are only useful for the IA64 MCA handling, or kdb.
1df5c10a
LT
8186 *
8187 * They can only be called when the whole system has been
8188 * stopped - every CPU needs to be quiescent, and no scheduling
8189 * activity can take place. Using them for anything else would
8190 * be a serious bug, and as a result, they aren't even visible
8191 * under any other configuration.
8192 */
8193
8194/**
8195 * curr_task - return the current task for a given cpu.
8196 * @cpu: the processor in question.
8197 *
8198 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8199 */
36c8b586 8200struct task_struct *curr_task(int cpu)
1df5c10a
LT
8201{
8202 return cpu_curr(cpu);
8203}
8204
67fc4e0c
JW
8205#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
8206
8207#ifdef CONFIG_IA64
1df5c10a
LT
8208/**
8209 * set_curr_task - set the current task for a given cpu.
8210 * @cpu: the processor in question.
8211 * @p: the task pointer to set.
8212 *
8213 * Description: This function must only be used when non-maskable interrupts
41a2d6cf
IM
8214 * are serviced on a separate stack. It allows the architecture to switch the
8215 * notion of the current task on a cpu in a non-blocking manner. This function
1df5c10a
LT
8216 * must be called with all CPU's synchronized, and interrupts disabled, the
8217 * and caller must save the original value of the current task (see
8218 * curr_task() above) and restore that value before reenabling interrupts and
8219 * re-starting the system.
8220 *
8221 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8222 */
36c8b586 8223void set_curr_task(int cpu, struct task_struct *p)
1df5c10a
LT
8224{
8225 cpu_curr(cpu) = p;
8226}
8227
8228#endif
29f59db3 8229
bccbe08a
PZ
8230#ifdef CONFIG_FAIR_GROUP_SCHED
8231static void free_fair_sched_group(struct task_group *tg)
6f505b16
PZ
8232{
8233 int i;
8234
8235 for_each_possible_cpu(i) {
8236 if (tg->cfs_rq)
8237 kfree(tg->cfs_rq[i]);
8238 if (tg->se)
8239 kfree(tg->se[i]);
6f505b16
PZ
8240 }
8241
8242 kfree(tg->cfs_rq);
8243 kfree(tg->se);
6f505b16
PZ
8244}
8245
ec7dc8ac
DG
8246static
8247int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
29f59db3 8248{
29f59db3 8249 struct cfs_rq *cfs_rq;
eab17229 8250 struct sched_entity *se;
9b5b7751 8251 struct rq *rq;
29f59db3
SV
8252 int i;
8253
434d53b0 8254 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
29f59db3
SV
8255 if (!tg->cfs_rq)
8256 goto err;
434d53b0 8257 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
29f59db3
SV
8258 if (!tg->se)
8259 goto err;
052f1dc7
PZ
8260
8261 tg->shares = NICE_0_LOAD;
29f59db3
SV
8262
8263 for_each_possible_cpu(i) {
9b5b7751 8264 rq = cpu_rq(i);
29f59db3 8265
eab17229
LZ
8266 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8267 GFP_KERNEL, cpu_to_node(i));
29f59db3
SV
8268 if (!cfs_rq)
8269 goto err;
8270
eab17229
LZ
8271 se = kzalloc_node(sizeof(struct sched_entity),
8272 GFP_KERNEL, cpu_to_node(i));
29f59db3 8273 if (!se)
dfc12eb2 8274 goto err_free_rq;
29f59db3 8275
eab17229 8276 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
bccbe08a
PZ
8277 }
8278
8279 return 1;
8280
49246274 8281err_free_rq:
dfc12eb2 8282 kfree(cfs_rq);
49246274 8283err:
bccbe08a
PZ
8284 return 0;
8285}
8286
8287static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8288{
8289 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list,
8290 &cpu_rq(cpu)->leaf_cfs_rq_list);
8291}
8292
8293static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8294{
8295 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list);
8296}
6d6bc0ad 8297#else /* !CONFG_FAIR_GROUP_SCHED */
bccbe08a
PZ
8298static inline void free_fair_sched_group(struct task_group *tg)
8299{
8300}
8301
ec7dc8ac
DG
8302static inline
8303int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8304{
8305 return 1;
8306}
8307
8308static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8309{
8310}
8311
8312static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8313{
8314}
6d6bc0ad 8315#endif /* CONFIG_FAIR_GROUP_SCHED */
052f1dc7
PZ
8316
8317#ifdef CONFIG_RT_GROUP_SCHED
bccbe08a
PZ
8318static void free_rt_sched_group(struct task_group *tg)
8319{
8320 int i;
8321
d0b27fa7
PZ
8322 destroy_rt_bandwidth(&tg->rt_bandwidth);
8323
bccbe08a
PZ
8324 for_each_possible_cpu(i) {
8325 if (tg->rt_rq)
8326 kfree(tg->rt_rq[i]);
8327 if (tg->rt_se)
8328 kfree(tg->rt_se[i]);
8329 }
8330
8331 kfree(tg->rt_rq);
8332 kfree(tg->rt_se);
8333}
8334
ec7dc8ac
DG
8335static
8336int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8337{
8338 struct rt_rq *rt_rq;
eab17229 8339 struct sched_rt_entity *rt_se;
bccbe08a
PZ
8340 struct rq *rq;
8341 int i;
8342
434d53b0 8343 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
bccbe08a
PZ
8344 if (!tg->rt_rq)
8345 goto err;
434d53b0 8346 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
bccbe08a
PZ
8347 if (!tg->rt_se)
8348 goto err;
8349
d0b27fa7
PZ
8350 init_rt_bandwidth(&tg->rt_bandwidth,
8351 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
bccbe08a
PZ
8352
8353 for_each_possible_cpu(i) {
8354 rq = cpu_rq(i);
8355
eab17229
LZ
8356 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8357 GFP_KERNEL, cpu_to_node(i));
6f505b16
PZ
8358 if (!rt_rq)
8359 goto err;
29f59db3 8360
eab17229
LZ
8361 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8362 GFP_KERNEL, cpu_to_node(i));
6f505b16 8363 if (!rt_se)
dfc12eb2 8364 goto err_free_rq;
29f59db3 8365
eab17229 8366 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
29f59db3
SV
8367 }
8368
bccbe08a
PZ
8369 return 1;
8370
49246274 8371err_free_rq:
dfc12eb2 8372 kfree(rt_rq);
49246274 8373err:
bccbe08a
PZ
8374 return 0;
8375}
8376
8377static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8378{
8379 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list,
8380 &cpu_rq(cpu)->leaf_rt_rq_list);
8381}
8382
8383static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8384{
8385 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list);
8386}
6d6bc0ad 8387#else /* !CONFIG_RT_GROUP_SCHED */
bccbe08a
PZ
8388static inline void free_rt_sched_group(struct task_group *tg)
8389{
8390}
8391
ec7dc8ac
DG
8392static inline
8393int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8394{
8395 return 1;
8396}
8397
8398static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8399{
8400}
8401
8402static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8403{
8404}
6d6bc0ad 8405#endif /* CONFIG_RT_GROUP_SCHED */
bccbe08a 8406
7c941438 8407#ifdef CONFIG_CGROUP_SCHED
bccbe08a
PZ
8408static void free_sched_group(struct task_group *tg)
8409{
8410 free_fair_sched_group(tg);
8411 free_rt_sched_group(tg);
8412 kfree(tg);
8413}
8414
8415/* allocate runqueue etc for a new task group */
ec7dc8ac 8416struct task_group *sched_create_group(struct task_group *parent)
bccbe08a
PZ
8417{
8418 struct task_group *tg;
8419 unsigned long flags;
8420 int i;
8421
8422 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8423 if (!tg)
8424 return ERR_PTR(-ENOMEM);
8425
ec7dc8ac 8426 if (!alloc_fair_sched_group(tg, parent))
bccbe08a
PZ
8427 goto err;
8428
ec7dc8ac 8429 if (!alloc_rt_sched_group(tg, parent))
bccbe08a
PZ
8430 goto err;
8431
8ed36996 8432 spin_lock_irqsave(&task_group_lock, flags);
9b5b7751 8433 for_each_possible_cpu(i) {
bccbe08a
PZ
8434 register_fair_sched_group(tg, i);
8435 register_rt_sched_group(tg, i);
9b5b7751 8436 }
6f505b16 8437 list_add_rcu(&tg->list, &task_groups);
f473aa5e
PZ
8438
8439 WARN_ON(!parent); /* root should already exist */
8440
8441 tg->parent = parent;
f473aa5e 8442 INIT_LIST_HEAD(&tg->children);
09f2724a 8443 list_add_rcu(&tg->siblings, &parent->children);
8ed36996 8444 spin_unlock_irqrestore(&task_group_lock, flags);
29f59db3 8445
9b5b7751 8446 return tg;
29f59db3
SV
8447
8448err:
6f505b16 8449 free_sched_group(tg);
29f59db3
SV
8450 return ERR_PTR(-ENOMEM);
8451}
8452
9b5b7751 8453/* rcu callback to free various structures associated with a task group */
6f505b16 8454static void free_sched_group_rcu(struct rcu_head *rhp)
29f59db3 8455{
29f59db3 8456 /* now it should be safe to free those cfs_rqs */
6f505b16 8457 free_sched_group(container_of(rhp, struct task_group, rcu));
29f59db3
SV
8458}
8459
9b5b7751 8460/* Destroy runqueue etc associated with a task group */
4cf86d77 8461void sched_destroy_group(struct task_group *tg)
29f59db3 8462{
8ed36996 8463 unsigned long flags;
9b5b7751 8464 int i;
29f59db3 8465
8ed36996 8466 spin_lock_irqsave(&task_group_lock, flags);
9b5b7751 8467 for_each_possible_cpu(i) {
bccbe08a
PZ
8468 unregister_fair_sched_group(tg, i);
8469 unregister_rt_sched_group(tg, i);
9b5b7751 8470 }
6f505b16 8471 list_del_rcu(&tg->list);
f473aa5e 8472 list_del_rcu(&tg->siblings);
8ed36996 8473 spin_unlock_irqrestore(&task_group_lock, flags);
9b5b7751 8474
9b5b7751 8475 /* wait for possible concurrent references to cfs_rqs complete */
6f505b16 8476 call_rcu(&tg->rcu, free_sched_group_rcu);
29f59db3
SV
8477}
8478
9b5b7751 8479/* change task's runqueue when it moves between groups.
3a252015
IM
8480 * The caller of this function should have put the task in its new group
8481 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8482 * reflect its new group.
9b5b7751
SV
8483 */
8484void sched_move_task(struct task_struct *tsk)
29f59db3
SV
8485{
8486 int on_rq, running;
8487 unsigned long flags;
8488 struct rq *rq;
8489
8490 rq = task_rq_lock(tsk, &flags);
8491
051a1d1a 8492 running = task_current(rq, tsk);
29f59db3
SV
8493 on_rq = tsk->se.on_rq;
8494
0e1f3483 8495 if (on_rq)
29f59db3 8496 dequeue_task(rq, tsk, 0);
0e1f3483
HS
8497 if (unlikely(running))
8498 tsk->sched_class->put_prev_task(rq, tsk);
29f59db3 8499
6f505b16 8500 set_task_rq(tsk, task_cpu(tsk));
29f59db3 8501
810b3817
PZ
8502#ifdef CONFIG_FAIR_GROUP_SCHED
8503 if (tsk->sched_class->moved_group)
88ec22d3 8504 tsk->sched_class->moved_group(tsk, on_rq);
810b3817
PZ
8505#endif
8506
0e1f3483
HS
8507 if (unlikely(running))
8508 tsk->sched_class->set_curr_task(rq);
8509 if (on_rq)
371fd7e7 8510 enqueue_task(rq, tsk, 0);
29f59db3 8511
29f59db3
SV
8512 task_rq_unlock(rq, &flags);
8513}
7c941438 8514#endif /* CONFIG_CGROUP_SCHED */
29f59db3 8515
052f1dc7 8516#ifdef CONFIG_FAIR_GROUP_SCHED
c09595f6 8517static void __set_se_shares(struct sched_entity *se, unsigned long shares)
29f59db3
SV
8518{
8519 struct cfs_rq *cfs_rq = se->cfs_rq;
29f59db3
SV
8520 int on_rq;
8521
29f59db3 8522 on_rq = se->on_rq;
62fb1851 8523 if (on_rq)
29f59db3
SV
8524 dequeue_entity(cfs_rq, se, 0);
8525
8526 se->load.weight = shares;
e05510d0 8527 se->load.inv_weight = 0;
29f59db3 8528
62fb1851 8529 if (on_rq)
29f59db3 8530 enqueue_entity(cfs_rq, se, 0);
c09595f6 8531}
62fb1851 8532
c09595f6
PZ
8533static void set_se_shares(struct sched_entity *se, unsigned long shares)
8534{
8535 struct cfs_rq *cfs_rq = se->cfs_rq;
8536 struct rq *rq = cfs_rq->rq;
8537 unsigned long flags;
8538
05fa785c 8539 raw_spin_lock_irqsave(&rq->lock, flags);
c09595f6 8540 __set_se_shares(se, shares);
05fa785c 8541 raw_spin_unlock_irqrestore(&rq->lock, flags);
29f59db3
SV
8542}
8543
8ed36996
PZ
8544static DEFINE_MUTEX(shares_mutex);
8545
4cf86d77 8546int sched_group_set_shares(struct task_group *tg, unsigned long shares)
29f59db3
SV
8547{
8548 int i;
8ed36996 8549 unsigned long flags;
c61935fd 8550
ec7dc8ac
DG
8551 /*
8552 * We can't change the weight of the root cgroup.
8553 */
8554 if (!tg->se[0])
8555 return -EINVAL;
8556
18d95a28
PZ
8557 if (shares < MIN_SHARES)
8558 shares = MIN_SHARES;
cb4ad1ff
MX
8559 else if (shares > MAX_SHARES)
8560 shares = MAX_SHARES;
62fb1851 8561
8ed36996 8562 mutex_lock(&shares_mutex);
9b5b7751 8563 if (tg->shares == shares)
5cb350ba 8564 goto done;
29f59db3 8565
8ed36996 8566 spin_lock_irqsave(&task_group_lock, flags);
bccbe08a
PZ
8567 for_each_possible_cpu(i)
8568 unregister_fair_sched_group(tg, i);
f473aa5e 8569 list_del_rcu(&tg->siblings);
8ed36996 8570 spin_unlock_irqrestore(&task_group_lock, flags);
6b2d7700
SV
8571
8572 /* wait for any ongoing reference to this group to finish */
8573 synchronize_sched();
8574
8575 /*
8576 * Now we are free to modify the group's share on each cpu
8577 * w/o tripping rebalance_share or load_balance_fair.
8578 */
9b5b7751 8579 tg->shares = shares;
c09595f6
PZ
8580 for_each_possible_cpu(i) {
8581 /*
8582 * force a rebalance
8583 */
8584 cfs_rq_set_shares(tg->cfs_rq[i], 0);
cb4ad1ff 8585 set_se_shares(tg->se[i], shares);
c09595f6 8586 }
29f59db3 8587
6b2d7700
SV
8588 /*
8589 * Enable load balance activity on this group, by inserting it back on
8590 * each cpu's rq->leaf_cfs_rq_list.
8591 */
8ed36996 8592 spin_lock_irqsave(&task_group_lock, flags);
bccbe08a
PZ
8593 for_each_possible_cpu(i)
8594 register_fair_sched_group(tg, i);
f473aa5e 8595 list_add_rcu(&tg->siblings, &tg->parent->children);
8ed36996 8596 spin_unlock_irqrestore(&task_group_lock, flags);
5cb350ba 8597done:
8ed36996 8598 mutex_unlock(&shares_mutex);
9b5b7751 8599 return 0;
29f59db3
SV
8600}
8601
5cb350ba
DG
8602unsigned long sched_group_shares(struct task_group *tg)
8603{
8604 return tg->shares;
8605}
052f1dc7 8606#endif
5cb350ba 8607
052f1dc7 8608#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 8609/*
9f0c1e56 8610 * Ensure that the real time constraints are schedulable.
6f505b16 8611 */
9f0c1e56
PZ
8612static DEFINE_MUTEX(rt_constraints_mutex);
8613
8614static unsigned long to_ratio(u64 period, u64 runtime)
8615{
8616 if (runtime == RUNTIME_INF)
9a7e0b18 8617 return 1ULL << 20;
9f0c1e56 8618
9a7e0b18 8619 return div64_u64(runtime << 20, period);
9f0c1e56
PZ
8620}
8621
9a7e0b18
PZ
8622/* Must be called with tasklist_lock held */
8623static inline int tg_has_rt_tasks(struct task_group *tg)
b40b2e8e 8624{
9a7e0b18 8625 struct task_struct *g, *p;
b40b2e8e 8626
9a7e0b18
PZ
8627 do_each_thread(g, p) {
8628 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8629 return 1;
8630 } while_each_thread(g, p);
b40b2e8e 8631
9a7e0b18
PZ
8632 return 0;
8633}
b40b2e8e 8634
9a7e0b18
PZ
8635struct rt_schedulable_data {
8636 struct task_group *tg;
8637 u64 rt_period;
8638 u64 rt_runtime;
8639};
b40b2e8e 8640
9a7e0b18
PZ
8641static int tg_schedulable(struct task_group *tg, void *data)
8642{
8643 struct rt_schedulable_data *d = data;
8644 struct task_group *child;
8645 unsigned long total, sum = 0;
8646 u64 period, runtime;
b40b2e8e 8647
9a7e0b18
PZ
8648 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8649 runtime = tg->rt_bandwidth.rt_runtime;
b40b2e8e 8650
9a7e0b18
PZ
8651 if (tg == d->tg) {
8652 period = d->rt_period;
8653 runtime = d->rt_runtime;
b40b2e8e 8654 }
b40b2e8e 8655
4653f803
PZ
8656 /*
8657 * Cannot have more runtime than the period.
8658 */
8659 if (runtime > period && runtime != RUNTIME_INF)
8660 return -EINVAL;
6f505b16 8661
4653f803
PZ
8662 /*
8663 * Ensure we don't starve existing RT tasks.
8664 */
9a7e0b18
PZ
8665 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8666 return -EBUSY;
6f505b16 8667
9a7e0b18 8668 total = to_ratio(period, runtime);
6f505b16 8669
4653f803
PZ
8670 /*
8671 * Nobody can have more than the global setting allows.
8672 */
8673 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8674 return -EINVAL;
6f505b16 8675
4653f803
PZ
8676 /*
8677 * The sum of our children's runtime should not exceed our own.
8678 */
9a7e0b18
PZ
8679 list_for_each_entry_rcu(child, &tg->children, siblings) {
8680 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8681 runtime = child->rt_bandwidth.rt_runtime;
6f505b16 8682
9a7e0b18
PZ
8683 if (child == d->tg) {
8684 period = d->rt_period;
8685 runtime = d->rt_runtime;
8686 }
6f505b16 8687
9a7e0b18 8688 sum += to_ratio(period, runtime);
9f0c1e56 8689 }
6f505b16 8690
9a7e0b18
PZ
8691 if (sum > total)
8692 return -EINVAL;
8693
8694 return 0;
6f505b16
PZ
8695}
8696
9a7e0b18 8697static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
521f1a24 8698{
9a7e0b18
PZ
8699 struct rt_schedulable_data data = {
8700 .tg = tg,
8701 .rt_period = period,
8702 .rt_runtime = runtime,
8703 };
8704
8705 return walk_tg_tree(tg_schedulable, tg_nop, &data);
521f1a24
DG
8706}
8707
d0b27fa7
PZ
8708static int tg_set_bandwidth(struct task_group *tg,
8709 u64 rt_period, u64 rt_runtime)
6f505b16 8710{
ac086bc2 8711 int i, err = 0;
9f0c1e56 8712
9f0c1e56 8713 mutex_lock(&rt_constraints_mutex);
521f1a24 8714 read_lock(&tasklist_lock);
9a7e0b18
PZ
8715 err = __rt_schedulable(tg, rt_period, rt_runtime);
8716 if (err)
9f0c1e56 8717 goto unlock;
ac086bc2 8718
0986b11b 8719 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
d0b27fa7
PZ
8720 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8721 tg->rt_bandwidth.rt_runtime = rt_runtime;
ac086bc2
PZ
8722
8723 for_each_possible_cpu(i) {
8724 struct rt_rq *rt_rq = tg->rt_rq[i];
8725
0986b11b 8726 raw_spin_lock(&rt_rq->rt_runtime_lock);
ac086bc2 8727 rt_rq->rt_runtime = rt_runtime;
0986b11b 8728 raw_spin_unlock(&rt_rq->rt_runtime_lock);
ac086bc2 8729 }
0986b11b 8730 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
49246274 8731unlock:
521f1a24 8732 read_unlock(&tasklist_lock);
9f0c1e56
PZ
8733 mutex_unlock(&rt_constraints_mutex);
8734
8735 return err;
6f505b16
PZ
8736}
8737
d0b27fa7
PZ
8738int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8739{
8740 u64 rt_runtime, rt_period;
8741
8742 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8743 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8744 if (rt_runtime_us < 0)
8745 rt_runtime = RUNTIME_INF;
8746
8747 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8748}
8749
9f0c1e56
PZ
8750long sched_group_rt_runtime(struct task_group *tg)
8751{
8752 u64 rt_runtime_us;
8753
d0b27fa7 8754 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
9f0c1e56
PZ
8755 return -1;
8756
d0b27fa7 8757 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
9f0c1e56
PZ
8758 do_div(rt_runtime_us, NSEC_PER_USEC);
8759 return rt_runtime_us;
8760}
d0b27fa7
PZ
8761
8762int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8763{
8764 u64 rt_runtime, rt_period;
8765
8766 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8767 rt_runtime = tg->rt_bandwidth.rt_runtime;
8768
619b0488
R
8769 if (rt_period == 0)
8770 return -EINVAL;
8771
d0b27fa7
PZ
8772 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8773}
8774
8775long sched_group_rt_period(struct task_group *tg)
8776{
8777 u64 rt_period_us;
8778
8779 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8780 do_div(rt_period_us, NSEC_PER_USEC);
8781 return rt_period_us;
8782}
8783
8784static int sched_rt_global_constraints(void)
8785{
4653f803 8786 u64 runtime, period;
d0b27fa7
PZ
8787 int ret = 0;
8788
ec5d4989
HS
8789 if (sysctl_sched_rt_period <= 0)
8790 return -EINVAL;
8791
4653f803
PZ
8792 runtime = global_rt_runtime();
8793 period = global_rt_period();
8794
8795 /*
8796 * Sanity check on the sysctl variables.
8797 */
8798 if (runtime > period && runtime != RUNTIME_INF)
8799 return -EINVAL;
10b612f4 8800
d0b27fa7 8801 mutex_lock(&rt_constraints_mutex);
9a7e0b18 8802 read_lock(&tasklist_lock);
4653f803 8803 ret = __rt_schedulable(NULL, 0, 0);
9a7e0b18 8804 read_unlock(&tasklist_lock);
d0b27fa7
PZ
8805 mutex_unlock(&rt_constraints_mutex);
8806
8807 return ret;
8808}
54e99124
DG
8809
8810int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8811{
8812 /* Don't accept realtime tasks when there is no way for them to run */
8813 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
8814 return 0;
8815
8816 return 1;
8817}
8818
6d6bc0ad 8819#else /* !CONFIG_RT_GROUP_SCHED */
d0b27fa7
PZ
8820static int sched_rt_global_constraints(void)
8821{
ac086bc2
PZ
8822 unsigned long flags;
8823 int i;
8824
ec5d4989
HS
8825 if (sysctl_sched_rt_period <= 0)
8826 return -EINVAL;
8827
60aa605d
PZ
8828 /*
8829 * There's always some RT tasks in the root group
8830 * -- migration, kstopmachine etc..
8831 */
8832 if (sysctl_sched_rt_runtime == 0)
8833 return -EBUSY;
8834
0986b11b 8835 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
ac086bc2
PZ
8836 for_each_possible_cpu(i) {
8837 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
8838
0986b11b 8839 raw_spin_lock(&rt_rq->rt_runtime_lock);
ac086bc2 8840 rt_rq->rt_runtime = global_rt_runtime();
0986b11b 8841 raw_spin_unlock(&rt_rq->rt_runtime_lock);
ac086bc2 8842 }
0986b11b 8843 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
ac086bc2 8844
d0b27fa7
PZ
8845 return 0;
8846}
6d6bc0ad 8847#endif /* CONFIG_RT_GROUP_SCHED */
d0b27fa7
PZ
8848
8849int sched_rt_handler(struct ctl_table *table, int write,
8d65af78 8850 void __user *buffer, size_t *lenp,
d0b27fa7
PZ
8851 loff_t *ppos)
8852{
8853 int ret;
8854 int old_period, old_runtime;
8855 static DEFINE_MUTEX(mutex);
8856
8857 mutex_lock(&mutex);
8858 old_period = sysctl_sched_rt_period;
8859 old_runtime = sysctl_sched_rt_runtime;
8860
8d65af78 8861 ret = proc_dointvec(table, write, buffer, lenp, ppos);
d0b27fa7
PZ
8862
8863 if (!ret && write) {
8864 ret = sched_rt_global_constraints();
8865 if (ret) {
8866 sysctl_sched_rt_period = old_period;
8867 sysctl_sched_rt_runtime = old_runtime;
8868 } else {
8869 def_rt_bandwidth.rt_runtime = global_rt_runtime();
8870 def_rt_bandwidth.rt_period =
8871 ns_to_ktime(global_rt_period());
8872 }
8873 }
8874 mutex_unlock(&mutex);
8875
8876 return ret;
8877}
68318b8e 8878
052f1dc7 8879#ifdef CONFIG_CGROUP_SCHED
68318b8e
SV
8880
8881/* return corresponding task_group object of a cgroup */
2b01dfe3 8882static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
68318b8e 8883{
2b01dfe3
PM
8884 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
8885 struct task_group, css);
68318b8e
SV
8886}
8887
8888static struct cgroup_subsys_state *
2b01dfe3 8889cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
68318b8e 8890{
ec7dc8ac 8891 struct task_group *tg, *parent;
68318b8e 8892
2b01dfe3 8893 if (!cgrp->parent) {
68318b8e 8894 /* This is early initialization for the top cgroup */
68318b8e
SV
8895 return &init_task_group.css;
8896 }
8897
ec7dc8ac
DG
8898 parent = cgroup_tg(cgrp->parent);
8899 tg = sched_create_group(parent);
68318b8e
SV
8900 if (IS_ERR(tg))
8901 return ERR_PTR(-ENOMEM);
8902
68318b8e
SV
8903 return &tg->css;
8904}
8905
41a2d6cf
IM
8906static void
8907cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
68318b8e 8908{
2b01dfe3 8909 struct task_group *tg = cgroup_tg(cgrp);
68318b8e
SV
8910
8911 sched_destroy_group(tg);
8912}
8913
41a2d6cf 8914static int
be367d09 8915cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
68318b8e 8916{
b68aa230 8917#ifdef CONFIG_RT_GROUP_SCHED
54e99124 8918 if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk))
b68aa230
PZ
8919 return -EINVAL;
8920#else
68318b8e
SV
8921 /* We don't support RT-tasks being in separate groups */
8922 if (tsk->sched_class != &fair_sched_class)
8923 return -EINVAL;
b68aa230 8924#endif
be367d09
BB
8925 return 0;
8926}
68318b8e 8927
be367d09
BB
8928static int
8929cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
8930 struct task_struct *tsk, bool threadgroup)
8931{
8932 int retval = cpu_cgroup_can_attach_task(cgrp, tsk);
8933 if (retval)
8934 return retval;
8935 if (threadgroup) {
8936 struct task_struct *c;
8937 rcu_read_lock();
8938 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8939 retval = cpu_cgroup_can_attach_task(cgrp, c);
8940 if (retval) {
8941 rcu_read_unlock();
8942 return retval;
8943 }
8944 }
8945 rcu_read_unlock();
8946 }
68318b8e
SV
8947 return 0;
8948}
8949
8950static void
2b01dfe3 8951cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
be367d09
BB
8952 struct cgroup *old_cont, struct task_struct *tsk,
8953 bool threadgroup)
68318b8e
SV
8954{
8955 sched_move_task(tsk);
be367d09
BB
8956 if (threadgroup) {
8957 struct task_struct *c;
8958 rcu_read_lock();
8959 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8960 sched_move_task(c);
8961 }
8962 rcu_read_unlock();
8963 }
68318b8e
SV
8964}
8965
052f1dc7 8966#ifdef CONFIG_FAIR_GROUP_SCHED
f4c753b7 8967static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
2b01dfe3 8968 u64 shareval)
68318b8e 8969{
2b01dfe3 8970 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
68318b8e
SV
8971}
8972
f4c753b7 8973static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
68318b8e 8974{
2b01dfe3 8975 struct task_group *tg = cgroup_tg(cgrp);
68318b8e
SV
8976
8977 return (u64) tg->shares;
8978}
6d6bc0ad 8979#endif /* CONFIG_FAIR_GROUP_SCHED */
68318b8e 8980
052f1dc7 8981#ifdef CONFIG_RT_GROUP_SCHED
0c70814c 8982static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
06ecb27c 8983 s64 val)
6f505b16 8984{
06ecb27c 8985 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
6f505b16
PZ
8986}
8987
06ecb27c 8988static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
6f505b16 8989{
06ecb27c 8990 return sched_group_rt_runtime(cgroup_tg(cgrp));
6f505b16 8991}
d0b27fa7
PZ
8992
8993static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
8994 u64 rt_period_us)
8995{
8996 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
8997}
8998
8999static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
9000{
9001 return sched_group_rt_period(cgroup_tg(cgrp));
9002}
6d6bc0ad 9003#endif /* CONFIG_RT_GROUP_SCHED */
6f505b16 9004
fe5c7cc2 9005static struct cftype cpu_files[] = {
052f1dc7 9006#ifdef CONFIG_FAIR_GROUP_SCHED
fe5c7cc2
PM
9007 {
9008 .name = "shares",
f4c753b7
PM
9009 .read_u64 = cpu_shares_read_u64,
9010 .write_u64 = cpu_shares_write_u64,
fe5c7cc2 9011 },
052f1dc7
PZ
9012#endif
9013#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 9014 {
9f0c1e56 9015 .name = "rt_runtime_us",
06ecb27c
PM
9016 .read_s64 = cpu_rt_runtime_read,
9017 .write_s64 = cpu_rt_runtime_write,
6f505b16 9018 },
d0b27fa7
PZ
9019 {
9020 .name = "rt_period_us",
f4c753b7
PM
9021 .read_u64 = cpu_rt_period_read_uint,
9022 .write_u64 = cpu_rt_period_write_uint,
d0b27fa7 9023 },
052f1dc7 9024#endif
68318b8e
SV
9025};
9026
9027static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
9028{
fe5c7cc2 9029 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
68318b8e
SV
9030}
9031
9032struct cgroup_subsys cpu_cgroup_subsys = {
38605cae
IM
9033 .name = "cpu",
9034 .create = cpu_cgroup_create,
9035 .destroy = cpu_cgroup_destroy,
9036 .can_attach = cpu_cgroup_can_attach,
9037 .attach = cpu_cgroup_attach,
9038 .populate = cpu_cgroup_populate,
9039 .subsys_id = cpu_cgroup_subsys_id,
68318b8e
SV
9040 .early_init = 1,
9041};
9042
052f1dc7 9043#endif /* CONFIG_CGROUP_SCHED */
d842de87
SV
9044
9045#ifdef CONFIG_CGROUP_CPUACCT
9046
9047/*
9048 * CPU accounting code for task groups.
9049 *
9050 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
9051 * (balbir@in.ibm.com).
9052 */
9053
934352f2 9054/* track cpu usage of a group of tasks and its child groups */
d842de87
SV
9055struct cpuacct {
9056 struct cgroup_subsys_state css;
9057 /* cpuusage holds pointer to a u64-type object on every cpu */
43cf38eb 9058 u64 __percpu *cpuusage;
ef12fefa 9059 struct percpu_counter cpustat[CPUACCT_STAT_NSTATS];
934352f2 9060 struct cpuacct *parent;
d842de87
SV
9061};
9062
9063struct cgroup_subsys cpuacct_subsys;
9064
9065/* return cpu accounting group corresponding to this container */
32cd756a 9066static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
d842de87 9067{
32cd756a 9068 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
d842de87
SV
9069 struct cpuacct, css);
9070}
9071
9072/* return cpu accounting group to which this task belongs */
9073static inline struct cpuacct *task_ca(struct task_struct *tsk)
9074{
9075 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
9076 struct cpuacct, css);
9077}
9078
9079/* create a new cpu accounting group */
9080static struct cgroup_subsys_state *cpuacct_create(
32cd756a 9081 struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87
SV
9082{
9083 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
ef12fefa 9084 int i;
d842de87
SV
9085
9086 if (!ca)
ef12fefa 9087 goto out;
d842de87
SV
9088
9089 ca->cpuusage = alloc_percpu(u64);
ef12fefa
BR
9090 if (!ca->cpuusage)
9091 goto out_free_ca;
9092
9093 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
9094 if (percpu_counter_init(&ca->cpustat[i], 0))
9095 goto out_free_counters;
d842de87 9096
934352f2
BR
9097 if (cgrp->parent)
9098 ca->parent = cgroup_ca(cgrp->parent);
9099
d842de87 9100 return &ca->css;
ef12fefa
BR
9101
9102out_free_counters:
9103 while (--i >= 0)
9104 percpu_counter_destroy(&ca->cpustat[i]);
9105 free_percpu(ca->cpuusage);
9106out_free_ca:
9107 kfree(ca);
9108out:
9109 return ERR_PTR(-ENOMEM);
d842de87
SV
9110}
9111
9112/* destroy an existing cpu accounting group */
41a2d6cf 9113static void
32cd756a 9114cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87 9115{
32cd756a 9116 struct cpuacct *ca = cgroup_ca(cgrp);
ef12fefa 9117 int i;
d842de87 9118
ef12fefa
BR
9119 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
9120 percpu_counter_destroy(&ca->cpustat[i]);
d842de87
SV
9121 free_percpu(ca->cpuusage);
9122 kfree(ca);
9123}
9124
720f5498
KC
9125static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
9126{
b36128c8 9127 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
720f5498
KC
9128 u64 data;
9129
9130#ifndef CONFIG_64BIT
9131 /*
9132 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
9133 */
05fa785c 9134 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
720f5498 9135 data = *cpuusage;
05fa785c 9136 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
720f5498
KC
9137#else
9138 data = *cpuusage;
9139#endif
9140
9141 return data;
9142}
9143
9144static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
9145{
b36128c8 9146 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
720f5498
KC
9147
9148#ifndef CONFIG_64BIT
9149 /*
9150 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
9151 */
05fa785c 9152 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
720f5498 9153 *cpuusage = val;
05fa785c 9154 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
720f5498
KC
9155#else
9156 *cpuusage = val;
9157#endif
9158}
9159
d842de87 9160/* return total cpu usage (in nanoseconds) of a group */
32cd756a 9161static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
d842de87 9162{
32cd756a 9163 struct cpuacct *ca = cgroup_ca(cgrp);
d842de87
SV
9164 u64 totalcpuusage = 0;
9165 int i;
9166
720f5498
KC
9167 for_each_present_cpu(i)
9168 totalcpuusage += cpuacct_cpuusage_read(ca, i);
d842de87
SV
9169
9170 return totalcpuusage;
9171}
9172
0297b803
DG
9173static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
9174 u64 reset)
9175{
9176 struct cpuacct *ca = cgroup_ca(cgrp);
9177 int err = 0;
9178 int i;
9179
9180 if (reset) {
9181 err = -EINVAL;
9182 goto out;
9183 }
9184
720f5498
KC
9185 for_each_present_cpu(i)
9186 cpuacct_cpuusage_write(ca, i, 0);
0297b803 9187
0297b803
DG
9188out:
9189 return err;
9190}
9191
e9515c3c
KC
9192static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
9193 struct seq_file *m)
9194{
9195 struct cpuacct *ca = cgroup_ca(cgroup);
9196 u64 percpu;
9197 int i;
9198
9199 for_each_present_cpu(i) {
9200 percpu = cpuacct_cpuusage_read(ca, i);
9201 seq_printf(m, "%llu ", (unsigned long long) percpu);
9202 }
9203 seq_printf(m, "\n");
9204 return 0;
9205}
9206
ef12fefa
BR
9207static const char *cpuacct_stat_desc[] = {
9208 [CPUACCT_STAT_USER] = "user",
9209 [CPUACCT_STAT_SYSTEM] = "system",
9210};
9211
9212static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
9213 struct cgroup_map_cb *cb)
9214{
9215 struct cpuacct *ca = cgroup_ca(cgrp);
9216 int i;
9217
9218 for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
9219 s64 val = percpu_counter_read(&ca->cpustat[i]);
9220 val = cputime64_to_clock_t(val);
9221 cb->fill(cb, cpuacct_stat_desc[i], val);
9222 }
9223 return 0;
9224}
9225
d842de87
SV
9226static struct cftype files[] = {
9227 {
9228 .name = "usage",
f4c753b7
PM
9229 .read_u64 = cpuusage_read,
9230 .write_u64 = cpuusage_write,
d842de87 9231 },
e9515c3c
KC
9232 {
9233 .name = "usage_percpu",
9234 .read_seq_string = cpuacct_percpu_seq_read,
9235 },
ef12fefa
BR
9236 {
9237 .name = "stat",
9238 .read_map = cpuacct_stats_show,
9239 },
d842de87
SV
9240};
9241
32cd756a 9242static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87 9243{
32cd756a 9244 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
d842de87
SV
9245}
9246
9247/*
9248 * charge this task's execution time to its accounting group.
9249 *
9250 * called with rq->lock held.
9251 */
9252static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9253{
9254 struct cpuacct *ca;
934352f2 9255 int cpu;
d842de87 9256
c40c6f85 9257 if (unlikely(!cpuacct_subsys.active))
d842de87
SV
9258 return;
9259
934352f2 9260 cpu = task_cpu(tsk);
a18b83b7
BR
9261
9262 rcu_read_lock();
9263
d842de87 9264 ca = task_ca(tsk);
d842de87 9265
934352f2 9266 for (; ca; ca = ca->parent) {
b36128c8 9267 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
d842de87
SV
9268 *cpuusage += cputime;
9269 }
a18b83b7
BR
9270
9271 rcu_read_unlock();
d842de87
SV
9272}
9273
fa535a77
AB
9274/*
9275 * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large
9276 * in cputime_t units. As a result, cpuacct_update_stats calls
9277 * percpu_counter_add with values large enough to always overflow the
9278 * per cpu batch limit causing bad SMP scalability.
9279 *
9280 * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we
9281 * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled
9282 * and enabled. We cap it at INT_MAX which is the largest allowed batch value.
9283 */
9284#ifdef CONFIG_SMP
9285#define CPUACCT_BATCH \
9286 min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX)
9287#else
9288#define CPUACCT_BATCH 0
9289#endif
9290
ef12fefa
BR
9291/*
9292 * Charge the system/user time to the task's accounting group.
9293 */
9294static void cpuacct_update_stats(struct task_struct *tsk,
9295 enum cpuacct_stat_index idx, cputime_t val)
9296{
9297 struct cpuacct *ca;
fa535a77 9298 int batch = CPUACCT_BATCH;
ef12fefa
BR
9299
9300 if (unlikely(!cpuacct_subsys.active))
9301 return;
9302
9303 rcu_read_lock();
9304 ca = task_ca(tsk);
9305
9306 do {
fa535a77 9307 __percpu_counter_add(&ca->cpustat[idx], val, batch);
ef12fefa
BR
9308 ca = ca->parent;
9309 } while (ca);
9310 rcu_read_unlock();
9311}
9312
d842de87
SV
9313struct cgroup_subsys cpuacct_subsys = {
9314 .name = "cpuacct",
9315 .create = cpuacct_create,
9316 .destroy = cpuacct_destroy,
9317 .populate = cpuacct_populate,
9318 .subsys_id = cpuacct_subsys_id,
9319};
9320#endif /* CONFIG_CGROUP_CPUACCT */
03b042bf
PM
9321
9322#ifndef CONFIG_SMP
9323
03b042bf
PM
9324void synchronize_sched_expedited(void)
9325{
fc390cde 9326 barrier();
03b042bf
PM
9327}
9328EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9329
9330#else /* #ifndef CONFIG_SMP */
9331
cc631fb7 9332static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0);
03b042bf 9333
cc631fb7 9334static int synchronize_sched_expedited_cpu_stop(void *data)
03b042bf 9335{
969c7921
TH
9336 /*
9337 * There must be a full memory barrier on each affected CPU
9338 * between the time that try_stop_cpus() is called and the
9339 * time that it returns.
9340 *
9341 * In the current initial implementation of cpu_stop, the
9342 * above condition is already met when the control reaches
9343 * this point and the following smp_mb() is not strictly
9344 * necessary. Do smp_mb() anyway for documentation and
9345 * robustness against future implementation changes.
9346 */
cc631fb7 9347 smp_mb(); /* See above comment block. */
969c7921 9348 return 0;
03b042bf 9349}
03b042bf
PM
9350
9351/*
9352 * Wait for an rcu-sched grace period to elapse, but use "big hammer"
9353 * approach to force grace period to end quickly. This consumes
9354 * significant time on all CPUs, and is thus not recommended for
9355 * any sort of common-case code.
9356 *
9357 * Note that it is illegal to call this function while holding any
9358 * lock that is acquired by a CPU-hotplug notifier. Failing to
9359 * observe this restriction will result in deadlock.
9360 */
9361void synchronize_sched_expedited(void)
9362{
969c7921 9363 int snap, trycount = 0;
03b042bf
PM
9364
9365 smp_mb(); /* ensure prior mod happens before capturing snap. */
969c7921 9366 snap = atomic_read(&synchronize_sched_expedited_count) + 1;
03b042bf 9367 get_online_cpus();
969c7921
TH
9368 while (try_stop_cpus(cpu_online_mask,
9369 synchronize_sched_expedited_cpu_stop,
94458d5e 9370 NULL) == -EAGAIN) {
03b042bf
PM
9371 put_online_cpus();
9372 if (trycount++ < 10)
9373 udelay(trycount * num_online_cpus());
9374 else {
9375 synchronize_sched();
9376 return;
9377 }
969c7921 9378 if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) {
03b042bf
PM
9379 smp_mb(); /* ensure test happens before caller kfree */
9380 return;
9381 }
9382 get_online_cpus();
9383 }
969c7921 9384 atomic_inc(&synchronize_sched_expedited_count);
cc631fb7 9385 smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */
03b042bf 9386 put_online_cpus();
03b042bf
PM
9387}
9388EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9389
9390#endif /* #else #ifndef CONFIG_SMP */