]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - kernel/sched.c
Merge branch 'flock' of git://git.kernel.org/pub/scm/linux/kernel/git/arnd/bkl
[mirror_ubuntu-artful-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
b52bfee4 1977 cpu = smp_processor_id();
d267f87f 1978 now = sched_clock_cpu(cpu);
b52bfee4
VP
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}
b7dadc38 1994EXPORT_SYMBOL_GPL(account_system_vtime);
b52bfee4 1995
aa483808
VP
1996static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time)
1997{
1998 if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) {
1999 u64 delta_irq = curr_irq_time - rq->prev_irq_time;
2000 rq->prev_irq_time = curr_irq_time;
2001 sched_rt_avg_update(rq, delta_irq);
2002 }
2003}
2004
305e6835
VP
2005#else
2006
2007static u64 irq_time_cpu(int cpu)
2008{
2009 return 0;
2010}
2011
aa483808
VP
2012static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { }
2013
b52bfee4
VP
2014#endif
2015
1e3c88bd
PZ
2016#include "sched_idletask.c"
2017#include "sched_fair.c"
2018#include "sched_rt.c"
34f971f6 2019#include "sched_stoptask.c"
1e3c88bd
PZ
2020#ifdef CONFIG_SCHED_DEBUG
2021# include "sched_debug.c"
2022#endif
2023
34f971f6
PZ
2024void sched_set_stop_task(int cpu, struct task_struct *stop)
2025{
2026 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
2027 struct task_struct *old_stop = cpu_rq(cpu)->stop;
2028
2029 if (stop) {
2030 /*
2031 * Make it appear like a SCHED_FIFO task, its something
2032 * userspace knows about and won't get confused about.
2033 *
2034 * Also, it will make PI more or less work without too
2035 * much confusion -- but then, stop work should not
2036 * rely on PI working anyway.
2037 */
2038 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
2039
2040 stop->sched_class = &stop_sched_class;
2041 }
2042
2043 cpu_rq(cpu)->stop = stop;
2044
2045 if (old_stop) {
2046 /*
2047 * Reset it back to a normal scheduling class so that
2048 * it can die in pieces.
2049 */
2050 old_stop->sched_class = &rt_sched_class;
2051 }
2052}
2053
14531189 2054/*
dd41f596 2055 * __normal_prio - return the priority that is based on the static prio
14531189 2056 */
14531189
IM
2057static inline int __normal_prio(struct task_struct *p)
2058{
dd41f596 2059 return p->static_prio;
14531189
IM
2060}
2061
b29739f9
IM
2062/*
2063 * Calculate the expected normal priority: i.e. priority
2064 * without taking RT-inheritance into account. Might be
2065 * boosted by interactivity modifiers. Changes upon fork,
2066 * setprio syscalls, and whenever the interactivity
2067 * estimator recalculates.
2068 */
36c8b586 2069static inline int normal_prio(struct task_struct *p)
b29739f9
IM
2070{
2071 int prio;
2072
e05606d3 2073 if (task_has_rt_policy(p))
b29739f9
IM
2074 prio = MAX_RT_PRIO-1 - p->rt_priority;
2075 else
2076 prio = __normal_prio(p);
2077 return prio;
2078}
2079
2080/*
2081 * Calculate the current priority, i.e. the priority
2082 * taken into account by the scheduler. This value might
2083 * be boosted by RT tasks, or might be boosted by
2084 * interactivity modifiers. Will be RT if the task got
2085 * RT-boosted. If not then it returns p->normal_prio.
2086 */
36c8b586 2087static int effective_prio(struct task_struct *p)
b29739f9
IM
2088{
2089 p->normal_prio = normal_prio(p);
2090 /*
2091 * If we are RT tasks or we were boosted to RT priority,
2092 * keep the priority unchanged. Otherwise, update priority
2093 * to the normal priority:
2094 */
2095 if (!rt_prio(p->prio))
2096 return p->normal_prio;
2097 return p->prio;
2098}
2099
1da177e4
LT
2100/**
2101 * task_curr - is this task currently executing on a CPU?
2102 * @p: the task in question.
2103 */
36c8b586 2104inline int task_curr(const struct task_struct *p)
1da177e4
LT
2105{
2106 return cpu_curr(task_cpu(p)) == p;
2107}
2108
cb469845
SR
2109static inline void check_class_changed(struct rq *rq, struct task_struct *p,
2110 const struct sched_class *prev_class,
2111 int oldprio, int running)
2112{
2113 if (prev_class != p->sched_class) {
2114 if (prev_class->switched_from)
2115 prev_class->switched_from(rq, p, running);
2116 p->sched_class->switched_to(rq, p, running);
2117 } else
2118 p->sched_class->prio_changed(rq, p, oldprio, running);
2119}
2120
1da177e4 2121#ifdef CONFIG_SMP
cc367732
IM
2122/*
2123 * Is this task likely cache-hot:
2124 */
e7693a36 2125static int
cc367732
IM
2126task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2127{
2128 s64 delta;
2129
e6c8fba7
PZ
2130 if (p->sched_class != &fair_sched_class)
2131 return 0;
2132
ef8002f6
NR
2133 if (unlikely(p->policy == SCHED_IDLE))
2134 return 0;
2135
f540a608
IM
2136 /*
2137 * Buddy candidates are cache hot:
2138 */
f685ceac 2139 if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
4793241b
PZ
2140 (&p->se == cfs_rq_of(&p->se)->next ||
2141 &p->se == cfs_rq_of(&p->se)->last))
f540a608
IM
2142 return 1;
2143
6bc1665b
IM
2144 if (sysctl_sched_migration_cost == -1)
2145 return 1;
2146 if (sysctl_sched_migration_cost == 0)
2147 return 0;
2148
cc367732
IM
2149 delta = now - p->se.exec_start;
2150
2151 return delta < (s64)sysctl_sched_migration_cost;
2152}
2153
dd41f596 2154void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
c65cc870 2155{
e2912009
PZ
2156#ifdef CONFIG_SCHED_DEBUG
2157 /*
2158 * We should never call set_task_cpu() on a blocked task,
2159 * ttwu() will sort out the placement.
2160 */
077614ee
PZ
2161 WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
2162 !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
e2912009
PZ
2163#endif
2164
de1d7286 2165 trace_sched_migrate_task(p, new_cpu);
cbc34ed1 2166
0c69774e
PZ
2167 if (task_cpu(p) != new_cpu) {
2168 p->se.nr_migrations++;
2169 perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0);
2170 }
dd41f596
IM
2171
2172 __set_task_cpu(p, new_cpu);
c65cc870
IM
2173}
2174
969c7921 2175struct migration_arg {
36c8b586 2176 struct task_struct *task;
1da177e4 2177 int dest_cpu;
70b97a7f 2178};
1da177e4 2179
969c7921
TH
2180static int migration_cpu_stop(void *data);
2181
1da177e4
LT
2182/*
2183 * The task's runqueue lock must be held.
2184 * Returns true if you have to wait for migration thread.
2185 */
969c7921 2186static bool migrate_task(struct task_struct *p, int dest_cpu)
1da177e4 2187{
70b97a7f 2188 struct rq *rq = task_rq(p);
1da177e4
LT
2189
2190 /*
2191 * If the task is not on a runqueue (and not running), then
e2912009 2192 * the next wake-up will properly place the task.
1da177e4 2193 */
969c7921 2194 return p->se.on_rq || task_running(rq, p);
1da177e4
LT
2195}
2196
2197/*
2198 * wait_task_inactive - wait for a thread to unschedule.
2199 *
85ba2d86
RM
2200 * If @match_state is nonzero, it's the @p->state value just checked and
2201 * not expected to change. If it changes, i.e. @p might have woken up,
2202 * then return zero. When we succeed in waiting for @p to be off its CPU,
2203 * we return a positive number (its total switch count). If a second call
2204 * a short while later returns the same number, the caller can be sure that
2205 * @p has remained unscheduled the whole time.
2206 *
1da177e4
LT
2207 * The caller must ensure that the task *will* unschedule sometime soon,
2208 * else this function might spin for a *long* time. This function can't
2209 * be called with interrupts off, or it may introduce deadlock with
2210 * smp_call_function() if an IPI is sent by the same process we are
2211 * waiting to become inactive.
2212 */
85ba2d86 2213unsigned long wait_task_inactive(struct task_struct *p, long match_state)
1da177e4
LT
2214{
2215 unsigned long flags;
dd41f596 2216 int running, on_rq;
85ba2d86 2217 unsigned long ncsw;
70b97a7f 2218 struct rq *rq;
1da177e4 2219
3a5c359a
AK
2220 for (;;) {
2221 /*
2222 * We do the initial early heuristics without holding
2223 * any task-queue locks at all. We'll only try to get
2224 * the runqueue lock when things look like they will
2225 * work out!
2226 */
2227 rq = task_rq(p);
fa490cfd 2228
3a5c359a
AK
2229 /*
2230 * If the task is actively running on another CPU
2231 * still, just relax and busy-wait without holding
2232 * any locks.
2233 *
2234 * NOTE! Since we don't hold any locks, it's not
2235 * even sure that "rq" stays as the right runqueue!
2236 * But we don't care, since "task_running()" will
2237 * return false if the runqueue has changed and p
2238 * is actually now running somewhere else!
2239 */
85ba2d86
RM
2240 while (task_running(rq, p)) {
2241 if (match_state && unlikely(p->state != match_state))
2242 return 0;
3a5c359a 2243 cpu_relax();
85ba2d86 2244 }
fa490cfd 2245
3a5c359a
AK
2246 /*
2247 * Ok, time to look more closely! We need the rq
2248 * lock now, to be *sure*. If we're wrong, we'll
2249 * just go back and repeat.
2250 */
2251 rq = task_rq_lock(p, &flags);
27a9da65 2252 trace_sched_wait_task(p);
3a5c359a
AK
2253 running = task_running(rq, p);
2254 on_rq = p->se.on_rq;
85ba2d86 2255 ncsw = 0;
f31e11d8 2256 if (!match_state || p->state == match_state)
93dcf55f 2257 ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
3a5c359a 2258 task_rq_unlock(rq, &flags);
fa490cfd 2259
85ba2d86
RM
2260 /*
2261 * If it changed from the expected state, bail out now.
2262 */
2263 if (unlikely(!ncsw))
2264 break;
2265
3a5c359a
AK
2266 /*
2267 * Was it really running after all now that we
2268 * checked with the proper locks actually held?
2269 *
2270 * Oops. Go back and try again..
2271 */
2272 if (unlikely(running)) {
2273 cpu_relax();
2274 continue;
2275 }
fa490cfd 2276
3a5c359a
AK
2277 /*
2278 * It's not enough that it's not actively running,
2279 * it must be off the runqueue _entirely_, and not
2280 * preempted!
2281 *
80dd99b3 2282 * So if it was still runnable (but just not actively
3a5c359a
AK
2283 * running right now), it's preempted, and we should
2284 * yield - it could be a while.
2285 */
2286 if (unlikely(on_rq)) {
2287 schedule_timeout_uninterruptible(1);
2288 continue;
2289 }
fa490cfd 2290
3a5c359a
AK
2291 /*
2292 * Ahh, all good. It wasn't running, and it wasn't
2293 * runnable, which means that it will never become
2294 * running in the future either. We're all done!
2295 */
2296 break;
2297 }
85ba2d86
RM
2298
2299 return ncsw;
1da177e4
LT
2300}
2301
2302/***
2303 * kick_process - kick a running thread to enter/exit the kernel
2304 * @p: the to-be-kicked thread
2305 *
2306 * Cause a process which is running on another CPU to enter
2307 * kernel-mode, without any delay. (to get signals handled.)
2308 *
2309 * NOTE: this function doesnt have to take the runqueue lock,
2310 * because all it wants to ensure is that the remote task enters
2311 * the kernel. If the IPI races and the task has been migrated
2312 * to another CPU then no harm is done and the purpose has been
2313 * achieved as well.
2314 */
36c8b586 2315void kick_process(struct task_struct *p)
1da177e4
LT
2316{
2317 int cpu;
2318
2319 preempt_disable();
2320 cpu = task_cpu(p);
2321 if ((cpu != smp_processor_id()) && task_curr(p))
2322 smp_send_reschedule(cpu);
2323 preempt_enable();
2324}
b43e3521 2325EXPORT_SYMBOL_GPL(kick_process);
476d139c 2326#endif /* CONFIG_SMP */
1da177e4 2327
0793a61d
TG
2328/**
2329 * task_oncpu_function_call - call a function on the cpu on which a task runs
2330 * @p: the task to evaluate
2331 * @func: the function to be called
2332 * @info: the function call argument
2333 *
2334 * Calls the function @func when the task is currently running. This might
2335 * be on the current CPU, which just calls the function directly
2336 */
2337void task_oncpu_function_call(struct task_struct *p,
2338 void (*func) (void *info), void *info)
2339{
2340 int cpu;
2341
2342 preempt_disable();
2343 cpu = task_cpu(p);
2344 if (task_curr(p))
2345 smp_call_function_single(cpu, func, info, 1);
2346 preempt_enable();
2347}
2348
970b13ba 2349#ifdef CONFIG_SMP
30da688e
ON
2350/*
2351 * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.
2352 */
5da9a0fb
PZ
2353static int select_fallback_rq(int cpu, struct task_struct *p)
2354{
2355 int dest_cpu;
2356 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu));
2357
2358 /* Look for allowed, online CPU in same node. */
2359 for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask)
2360 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
2361 return dest_cpu;
2362
2363 /* Any allowed, online CPU? */
2364 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask);
2365 if (dest_cpu < nr_cpu_ids)
2366 return dest_cpu;
2367
2368 /* No more Mr. Nice Guy. */
897f0b3c 2369 if (unlikely(dest_cpu >= nr_cpu_ids)) {
9084bb82 2370 dest_cpu = cpuset_cpus_allowed_fallback(p);
5da9a0fb
PZ
2371 /*
2372 * Don't tell them about moving exiting tasks or
2373 * kernel threads (both mm NULL), since they never
2374 * leave kernel.
2375 */
2376 if (p->mm && printk_ratelimit()) {
2377 printk(KERN_INFO "process %d (%s) no "
2378 "longer affine to cpu%d\n",
2379 task_pid_nr(p), p->comm, cpu);
2380 }
2381 }
2382
2383 return dest_cpu;
2384}
2385
e2912009 2386/*
30da688e 2387 * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable.
e2912009 2388 */
970b13ba 2389static inline
0017d735 2390int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags)
970b13ba 2391{
0017d735 2392 int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags);
e2912009
PZ
2393
2394 /*
2395 * In order not to call set_task_cpu() on a blocking task we need
2396 * to rely on ttwu() to place the task on a valid ->cpus_allowed
2397 * cpu.
2398 *
2399 * Since this is common to all placement strategies, this lives here.
2400 *
2401 * [ this allows ->select_task() to simply return task_cpu(p) and
2402 * not worry about this generic constraint ]
2403 */
2404 if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
70f11205 2405 !cpu_online(cpu)))
5da9a0fb 2406 cpu = select_fallback_rq(task_cpu(p), p);
e2912009
PZ
2407
2408 return cpu;
970b13ba 2409}
09a40af5
MG
2410
2411static void update_avg(u64 *avg, u64 sample)
2412{
2413 s64 diff = sample - *avg;
2414 *avg += diff >> 3;
2415}
970b13ba
PZ
2416#endif
2417
9ed3811a
TH
2418static inline void ttwu_activate(struct task_struct *p, struct rq *rq,
2419 bool is_sync, bool is_migrate, bool is_local,
2420 unsigned long en_flags)
2421{
2422 schedstat_inc(p, se.statistics.nr_wakeups);
2423 if (is_sync)
2424 schedstat_inc(p, se.statistics.nr_wakeups_sync);
2425 if (is_migrate)
2426 schedstat_inc(p, se.statistics.nr_wakeups_migrate);
2427 if (is_local)
2428 schedstat_inc(p, se.statistics.nr_wakeups_local);
2429 else
2430 schedstat_inc(p, se.statistics.nr_wakeups_remote);
2431
2432 activate_task(rq, p, en_flags);
2433}
2434
2435static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq,
2436 int wake_flags, bool success)
2437{
2438 trace_sched_wakeup(p, success);
2439 check_preempt_curr(rq, p, wake_flags);
2440
2441 p->state = TASK_RUNNING;
2442#ifdef CONFIG_SMP
2443 if (p->sched_class->task_woken)
2444 p->sched_class->task_woken(rq, p);
2445
2446 if (unlikely(rq->idle_stamp)) {
2447 u64 delta = rq->clock - rq->idle_stamp;
2448 u64 max = 2*sysctl_sched_migration_cost;
2449
2450 if (delta > max)
2451 rq->avg_idle = max;
2452 else
2453 update_avg(&rq->avg_idle, delta);
2454 rq->idle_stamp = 0;
2455 }
2456#endif
21aa9af0
TH
2457 /* if a worker is waking up, notify workqueue */
2458 if ((p->flags & PF_WQ_WORKER) && success)
2459 wq_worker_waking_up(p, cpu_of(rq));
9ed3811a
TH
2460}
2461
2462/**
1da177e4 2463 * try_to_wake_up - wake up a thread
9ed3811a 2464 * @p: the thread to be awakened
1da177e4 2465 * @state: the mask of task states that can be woken
9ed3811a 2466 * @wake_flags: wake modifier flags (WF_*)
1da177e4
LT
2467 *
2468 * Put it on the run-queue if it's not already there. The "current"
2469 * thread is always on the run-queue (except when the actual
2470 * re-schedule is in progress), and as such you're allowed to do
2471 * the simpler "current->state = TASK_RUNNING" to mark yourself
2472 * runnable without the overhead of this.
2473 *
9ed3811a
TH
2474 * Returns %true if @p was woken up, %false if it was already running
2475 * or @state didn't match @p's state.
1da177e4 2476 */
7d478721
PZ
2477static int try_to_wake_up(struct task_struct *p, unsigned int state,
2478 int wake_flags)
1da177e4 2479{
cc367732 2480 int cpu, orig_cpu, this_cpu, success = 0;
1da177e4 2481 unsigned long flags;
371fd7e7 2482 unsigned long en_flags = ENQUEUE_WAKEUP;
ab3b3aa5 2483 struct rq *rq;
1da177e4 2484
e9c84311 2485 this_cpu = get_cpu();
2398f2c6 2486
04e2f174 2487 smp_wmb();
ab3b3aa5 2488 rq = task_rq_lock(p, &flags);
e9c84311 2489 if (!(p->state & state))
1da177e4
LT
2490 goto out;
2491
dd41f596 2492 if (p->se.on_rq)
1da177e4
LT
2493 goto out_running;
2494
2495 cpu = task_cpu(p);
cc367732 2496 orig_cpu = cpu;
1da177e4
LT
2497
2498#ifdef CONFIG_SMP
2499 if (unlikely(task_running(rq, p)))
2500 goto out_activate;
2501
e9c84311
PZ
2502 /*
2503 * In order to handle concurrent wakeups and release the rq->lock
2504 * we put the task in TASK_WAKING state.
eb24073b
IM
2505 *
2506 * First fix up the nr_uninterruptible count:
e9c84311 2507 */
cc87f76a
PZ
2508 if (task_contributes_to_load(p)) {
2509 if (likely(cpu_online(orig_cpu)))
2510 rq->nr_uninterruptible--;
2511 else
2512 this_rq()->nr_uninterruptible--;
2513 }
e9c84311 2514 p->state = TASK_WAKING;
efbbd05a 2515
371fd7e7 2516 if (p->sched_class->task_waking) {
efbbd05a 2517 p->sched_class->task_waking(rq, p);
371fd7e7
PZ
2518 en_flags |= ENQUEUE_WAKING;
2519 }
efbbd05a 2520
0017d735
PZ
2521 cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags);
2522 if (cpu != orig_cpu)
5d2f5a61 2523 set_task_cpu(p, cpu);
0017d735 2524 __task_rq_unlock(rq);
ab19cb23 2525
0970d299
PZ
2526 rq = cpu_rq(cpu);
2527 raw_spin_lock(&rq->lock);
f5dc3753 2528
0970d299
PZ
2529 /*
2530 * We migrated the task without holding either rq->lock, however
2531 * since the task is not on the task list itself, nobody else
2532 * will try and migrate the task, hence the rq should match the
2533 * cpu we just moved it to.
2534 */
2535 WARN_ON(task_cpu(p) != cpu);
e9c84311 2536 WARN_ON(p->state != TASK_WAKING);
1da177e4 2537
e7693a36
GH
2538#ifdef CONFIG_SCHEDSTATS
2539 schedstat_inc(rq, ttwu_count);
2540 if (cpu == this_cpu)
2541 schedstat_inc(rq, ttwu_local);
2542 else {
2543 struct sched_domain *sd;
2544 for_each_domain(this_cpu, sd) {
758b2cdc 2545 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
e7693a36
GH
2546 schedstat_inc(sd, ttwu_wake_remote);
2547 break;
2548 }
2549 }
2550 }
6d6bc0ad 2551#endif /* CONFIG_SCHEDSTATS */
e7693a36 2552
1da177e4
LT
2553out_activate:
2554#endif /* CONFIG_SMP */
9ed3811a
TH
2555 ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu,
2556 cpu == this_cpu, en_flags);
1da177e4 2557 success = 1;
1da177e4 2558out_running:
9ed3811a 2559 ttwu_post_activation(p, rq, wake_flags, success);
1da177e4
LT
2560out:
2561 task_rq_unlock(rq, &flags);
e9c84311 2562 put_cpu();
1da177e4
LT
2563
2564 return success;
2565}
2566
21aa9af0
TH
2567/**
2568 * try_to_wake_up_local - try to wake up a local task with rq lock held
2569 * @p: the thread to be awakened
2570 *
2571 * Put @p on the run-queue if it's not alredy there. The caller must
2572 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2573 * the current task. this_rq() stays locked over invocation.
2574 */
2575static void try_to_wake_up_local(struct task_struct *p)
2576{
2577 struct rq *rq = task_rq(p);
2578 bool success = false;
2579
2580 BUG_ON(rq != this_rq());
2581 BUG_ON(p == current);
2582 lockdep_assert_held(&rq->lock);
2583
2584 if (!(p->state & TASK_NORMAL))
2585 return;
2586
2587 if (!p->se.on_rq) {
2588 if (likely(!task_running(rq, p))) {
2589 schedstat_inc(rq, ttwu_count);
2590 schedstat_inc(rq, ttwu_local);
2591 }
2592 ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP);
2593 success = true;
2594 }
2595 ttwu_post_activation(p, rq, 0, success);
2596}
2597
50fa610a
DH
2598/**
2599 * wake_up_process - Wake up a specific process
2600 * @p: The process to be woken up.
2601 *
2602 * Attempt to wake up the nominated process and move it to the set of runnable
2603 * processes. Returns 1 if the process was woken up, 0 if it was already
2604 * running.
2605 *
2606 * It may be assumed that this function implies a write memory barrier before
2607 * changing the task state if and only if any tasks are woken up.
2608 */
7ad5b3a5 2609int wake_up_process(struct task_struct *p)
1da177e4 2610{
d9514f6c 2611 return try_to_wake_up(p, TASK_ALL, 0);
1da177e4 2612}
1da177e4
LT
2613EXPORT_SYMBOL(wake_up_process);
2614
7ad5b3a5 2615int wake_up_state(struct task_struct *p, unsigned int state)
1da177e4
LT
2616{
2617 return try_to_wake_up(p, state, 0);
2618}
2619
1da177e4
LT
2620/*
2621 * Perform scheduler related setup for a newly forked process p.
2622 * p is forked by current.
dd41f596
IM
2623 *
2624 * __sched_fork() is basic setup used by init_idle() too:
2625 */
2626static void __sched_fork(struct task_struct *p)
2627{
dd41f596
IM
2628 p->se.exec_start = 0;
2629 p->se.sum_exec_runtime = 0;
f6cf891c 2630 p->se.prev_sum_exec_runtime = 0;
6c594c21 2631 p->se.nr_migrations = 0;
6cfb0d5d
IM
2632
2633#ifdef CONFIG_SCHEDSTATS
41acab88 2634 memset(&p->se.statistics, 0, sizeof(p->se.statistics));
6cfb0d5d 2635#endif
476d139c 2636
fa717060 2637 INIT_LIST_HEAD(&p->rt.run_list);
dd41f596 2638 p->se.on_rq = 0;
4a55bd5e 2639 INIT_LIST_HEAD(&p->se.group_node);
476d139c 2640
e107be36
AK
2641#ifdef CONFIG_PREEMPT_NOTIFIERS
2642 INIT_HLIST_HEAD(&p->preempt_notifiers);
2643#endif
dd41f596
IM
2644}
2645
2646/*
2647 * fork()/clone()-time setup:
2648 */
2649void sched_fork(struct task_struct *p, int clone_flags)
2650{
2651 int cpu = get_cpu();
2652
2653 __sched_fork(p);
06b83b5f 2654 /*
0017d735 2655 * We mark the process as running here. This guarantees that
06b83b5f
PZ
2656 * nobody will actually run it, and a signal or other external
2657 * event cannot wake it up and insert it on the runqueue either.
2658 */
0017d735 2659 p->state = TASK_RUNNING;
dd41f596 2660
b9dc29e7
MG
2661 /*
2662 * Revert to default priority/policy on fork if requested.
2663 */
2664 if (unlikely(p->sched_reset_on_fork)) {
f83f9ac2 2665 if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) {
b9dc29e7 2666 p->policy = SCHED_NORMAL;
f83f9ac2
PW
2667 p->normal_prio = p->static_prio;
2668 }
b9dc29e7 2669
6c697bdf
MG
2670 if (PRIO_TO_NICE(p->static_prio) < 0) {
2671 p->static_prio = NICE_TO_PRIO(0);
f83f9ac2 2672 p->normal_prio = p->static_prio;
6c697bdf
MG
2673 set_load_weight(p);
2674 }
2675
b9dc29e7
MG
2676 /*
2677 * We don't need the reset flag anymore after the fork. It has
2678 * fulfilled its duty:
2679 */
2680 p->sched_reset_on_fork = 0;
2681 }
ca94c442 2682
f83f9ac2
PW
2683 /*
2684 * Make sure we do not leak PI boosting priority to the child.
2685 */
2686 p->prio = current->normal_prio;
2687
2ddbf952
HS
2688 if (!rt_prio(p->prio))
2689 p->sched_class = &fair_sched_class;
b29739f9 2690
cd29fe6f
PZ
2691 if (p->sched_class->task_fork)
2692 p->sched_class->task_fork(p);
2693
86951599
PZ
2694 /*
2695 * The child is not yet in the pid-hash so no cgroup attach races,
2696 * and the cgroup is pinned to this child due to cgroup_fork()
2697 * is ran before sched_fork().
2698 *
2699 * Silence PROVE_RCU.
2700 */
2701 rcu_read_lock();
5f3edc1b 2702 set_task_cpu(p, cpu);
86951599 2703 rcu_read_unlock();
5f3edc1b 2704
52f17b6c 2705#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
dd41f596 2706 if (likely(sched_info_on()))
52f17b6c 2707 memset(&p->sched_info, 0, sizeof(p->sched_info));
1da177e4 2708#endif
d6077cb8 2709#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
4866cde0
NP
2710 p->oncpu = 0;
2711#endif
1da177e4 2712#ifdef CONFIG_PREEMPT
4866cde0 2713 /* Want to start with kernel preemption disabled. */
a1261f54 2714 task_thread_info(p)->preempt_count = 1;
1da177e4 2715#endif
917b627d
GH
2716 plist_node_init(&p->pushable_tasks, MAX_PRIO);
2717
476d139c 2718 put_cpu();
1da177e4
LT
2719}
2720
2721/*
2722 * wake_up_new_task - wake up a newly created task for the first time.
2723 *
2724 * This function will do some initial scheduler statistics housekeeping
2725 * that must be done for every newly created context, then puts the task
2726 * on the runqueue and wakes it.
2727 */
7ad5b3a5 2728void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
1da177e4
LT
2729{
2730 unsigned long flags;
dd41f596 2731 struct rq *rq;
c890692b 2732 int cpu __maybe_unused = get_cpu();
fabf318e
PZ
2733
2734#ifdef CONFIG_SMP
0017d735
PZ
2735 rq = task_rq_lock(p, &flags);
2736 p->state = TASK_WAKING;
2737
fabf318e
PZ
2738 /*
2739 * Fork balancing, do it here and not earlier because:
2740 * - cpus_allowed can change in the fork path
2741 * - any previously selected cpu might disappear through hotplug
2742 *
0017d735
PZ
2743 * We set TASK_WAKING so that select_task_rq() can drop rq->lock
2744 * without people poking at ->cpus_allowed.
fabf318e 2745 */
0017d735 2746 cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0);
fabf318e 2747 set_task_cpu(p, cpu);
1da177e4 2748
06b83b5f 2749 p->state = TASK_RUNNING;
0017d735
PZ
2750 task_rq_unlock(rq, &flags);
2751#endif
2752
2753 rq = task_rq_lock(p, &flags);
cd29fe6f 2754 activate_task(rq, p, 0);
27a9da65 2755 trace_sched_wakeup_new(p, 1);
a7558e01 2756 check_preempt_curr(rq, p, WF_FORK);
9a897c5a 2757#ifdef CONFIG_SMP
efbbd05a
PZ
2758 if (p->sched_class->task_woken)
2759 p->sched_class->task_woken(rq, p);
9a897c5a 2760#endif
dd41f596 2761 task_rq_unlock(rq, &flags);
fabf318e 2762 put_cpu();
1da177e4
LT
2763}
2764
e107be36
AK
2765#ifdef CONFIG_PREEMPT_NOTIFIERS
2766
2767/**
80dd99b3 2768 * preempt_notifier_register - tell me when current is being preempted & rescheduled
421cee29 2769 * @notifier: notifier struct to register
e107be36
AK
2770 */
2771void preempt_notifier_register(struct preempt_notifier *notifier)
2772{
2773 hlist_add_head(&notifier->link, &current->preempt_notifiers);
2774}
2775EXPORT_SYMBOL_GPL(preempt_notifier_register);
2776
2777/**
2778 * preempt_notifier_unregister - no longer interested in preemption notifications
421cee29 2779 * @notifier: notifier struct to unregister
e107be36
AK
2780 *
2781 * This is safe to call from within a preemption notifier.
2782 */
2783void preempt_notifier_unregister(struct preempt_notifier *notifier)
2784{
2785 hlist_del(&notifier->link);
2786}
2787EXPORT_SYMBOL_GPL(preempt_notifier_unregister);
2788
2789static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2790{
2791 struct preempt_notifier *notifier;
2792 struct hlist_node *node;
2793
2794 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2795 notifier->ops->sched_in(notifier, raw_smp_processor_id());
2796}
2797
2798static void
2799fire_sched_out_preempt_notifiers(struct task_struct *curr,
2800 struct task_struct *next)
2801{
2802 struct preempt_notifier *notifier;
2803 struct hlist_node *node;
2804
2805 hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
2806 notifier->ops->sched_out(notifier, next);
2807}
2808
6d6bc0ad 2809#else /* !CONFIG_PREEMPT_NOTIFIERS */
e107be36
AK
2810
2811static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2812{
2813}
2814
2815static void
2816fire_sched_out_preempt_notifiers(struct task_struct *curr,
2817 struct task_struct *next)
2818{
2819}
2820
6d6bc0ad 2821#endif /* CONFIG_PREEMPT_NOTIFIERS */
e107be36 2822
4866cde0
NP
2823/**
2824 * prepare_task_switch - prepare to switch tasks
2825 * @rq: the runqueue preparing to switch
421cee29 2826 * @prev: the current task that is being switched out
4866cde0
NP
2827 * @next: the task we are going to switch to.
2828 *
2829 * This is called with the rq lock held and interrupts off. It must
2830 * be paired with a subsequent finish_task_switch after the context
2831 * switch.
2832 *
2833 * prepare_task_switch sets up locking and calls architecture specific
2834 * hooks.
2835 */
e107be36
AK
2836static inline void
2837prepare_task_switch(struct rq *rq, struct task_struct *prev,
2838 struct task_struct *next)
4866cde0 2839{
e107be36 2840 fire_sched_out_preempt_notifiers(prev, next);
4866cde0
NP
2841 prepare_lock_switch(rq, next);
2842 prepare_arch_switch(next);
2843}
2844
1da177e4
LT
2845/**
2846 * finish_task_switch - clean up after a task-switch
344babaa 2847 * @rq: runqueue associated with task-switch
1da177e4
LT
2848 * @prev: the thread we just switched away from.
2849 *
4866cde0
NP
2850 * finish_task_switch must be called after the context switch, paired
2851 * with a prepare_task_switch call before the context switch.
2852 * finish_task_switch will reconcile locking set up by prepare_task_switch,
2853 * and do any other architecture-specific cleanup actions.
1da177e4
LT
2854 *
2855 * Note that we may have delayed dropping an mm in context_switch(). If
41a2d6cf 2856 * so, we finish that here outside of the runqueue lock. (Doing it
1da177e4
LT
2857 * with the lock held can cause deadlocks; see schedule() for
2858 * details.)
2859 */
a9957449 2860static void finish_task_switch(struct rq *rq, struct task_struct *prev)
1da177e4
LT
2861 __releases(rq->lock)
2862{
1da177e4 2863 struct mm_struct *mm = rq->prev_mm;
55a101f8 2864 long prev_state;
1da177e4
LT
2865
2866 rq->prev_mm = NULL;
2867
2868 /*
2869 * A task struct has one reference for the use as "current".
c394cc9f 2870 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
55a101f8
ON
2871 * schedule one last time. The schedule call will never return, and
2872 * the scheduled task must drop that reference.
c394cc9f 2873 * The test for TASK_DEAD must occur while the runqueue locks are
1da177e4
LT
2874 * still held, otherwise prev could be scheduled on another cpu, die
2875 * there before we look at prev->state, and then the reference would
2876 * be dropped twice.
2877 * Manfred Spraul <manfred@colorfullife.com>
2878 */
55a101f8 2879 prev_state = prev->state;
4866cde0 2880 finish_arch_switch(prev);
8381f65d
JI
2881#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2882 local_irq_disable();
2883#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
49f47433 2884 perf_event_task_sched_in(current);
8381f65d
JI
2885#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
2886 local_irq_enable();
2887#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
4866cde0 2888 finish_lock_switch(rq, prev);
e8fa1362 2889
e107be36 2890 fire_sched_in_preempt_notifiers(current);
1da177e4
LT
2891 if (mm)
2892 mmdrop(mm);
c394cc9f 2893 if (unlikely(prev_state == TASK_DEAD)) {
c6fd91f0 2894 /*
2895 * Remove function-return probe instances associated with this
2896 * task and put them back on the free list.
9761eea8 2897 */
c6fd91f0 2898 kprobe_flush_task(prev);
1da177e4 2899 put_task_struct(prev);
c6fd91f0 2900 }
1da177e4
LT
2901}
2902
3f029d3c
GH
2903#ifdef CONFIG_SMP
2904
2905/* assumes rq->lock is held */
2906static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
2907{
2908 if (prev->sched_class->pre_schedule)
2909 prev->sched_class->pre_schedule(rq, prev);
2910}
2911
2912/* rq->lock is NOT held, but preemption is disabled */
2913static inline void post_schedule(struct rq *rq)
2914{
2915 if (rq->post_schedule) {
2916 unsigned long flags;
2917
05fa785c 2918 raw_spin_lock_irqsave(&rq->lock, flags);
3f029d3c
GH
2919 if (rq->curr->sched_class->post_schedule)
2920 rq->curr->sched_class->post_schedule(rq);
05fa785c 2921 raw_spin_unlock_irqrestore(&rq->lock, flags);
3f029d3c
GH
2922
2923 rq->post_schedule = 0;
2924 }
2925}
2926
2927#else
da19ab51 2928
3f029d3c
GH
2929static inline void pre_schedule(struct rq *rq, struct task_struct *p)
2930{
2931}
2932
2933static inline void post_schedule(struct rq *rq)
2934{
1da177e4
LT
2935}
2936
3f029d3c
GH
2937#endif
2938
1da177e4
LT
2939/**
2940 * schedule_tail - first thing a freshly forked thread must call.
2941 * @prev: the thread we just switched away from.
2942 */
36c8b586 2943asmlinkage void schedule_tail(struct task_struct *prev)
1da177e4
LT
2944 __releases(rq->lock)
2945{
70b97a7f
IM
2946 struct rq *rq = this_rq();
2947
4866cde0 2948 finish_task_switch(rq, prev);
da19ab51 2949
3f029d3c
GH
2950 /*
2951 * FIXME: do we need to worry about rq being invalidated by the
2952 * task_switch?
2953 */
2954 post_schedule(rq);
70b97a7f 2955
4866cde0
NP
2956#ifdef __ARCH_WANT_UNLOCKED_CTXSW
2957 /* In this case, finish_task_switch does not reenable preemption */
2958 preempt_enable();
2959#endif
1da177e4 2960 if (current->set_child_tid)
b488893a 2961 put_user(task_pid_vnr(current), current->set_child_tid);
1da177e4
LT
2962}
2963
2964/*
2965 * context_switch - switch to the new MM and the new
2966 * thread's register state.
2967 */
dd41f596 2968static inline void
70b97a7f 2969context_switch(struct rq *rq, struct task_struct *prev,
36c8b586 2970 struct task_struct *next)
1da177e4 2971{
dd41f596 2972 struct mm_struct *mm, *oldmm;
1da177e4 2973
e107be36 2974 prepare_task_switch(rq, prev, next);
27a9da65 2975 trace_sched_switch(prev, next);
dd41f596
IM
2976 mm = next->mm;
2977 oldmm = prev->active_mm;
9226d125
ZA
2978 /*
2979 * For paravirt, this is coupled with an exit in switch_to to
2980 * combine the page table reload and the switch backend into
2981 * one hypercall.
2982 */
224101ed 2983 arch_start_context_switch(prev);
9226d125 2984
31915ab4 2985 if (!mm) {
1da177e4
LT
2986 next->active_mm = oldmm;
2987 atomic_inc(&oldmm->mm_count);
2988 enter_lazy_tlb(oldmm, next);
2989 } else
2990 switch_mm(oldmm, mm, next);
2991
31915ab4 2992 if (!prev->mm) {
1da177e4 2993 prev->active_mm = NULL;
1da177e4
LT
2994 rq->prev_mm = oldmm;
2995 }
3a5f5e48
IM
2996 /*
2997 * Since the runqueue lock will be released by the next
2998 * task (which is an invalid locking op but in the case
2999 * of the scheduler it's an obvious special-case), so we
3000 * do an early lockdep release here:
3001 */
3002#ifndef __ARCH_WANT_UNLOCKED_CTXSW
8a25d5de 3003 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
3a5f5e48 3004#endif
1da177e4
LT
3005
3006 /* Here we just switch the register state and the stack. */
3007 switch_to(prev, next, prev);
3008
dd41f596
IM
3009 barrier();
3010 /*
3011 * this_rq must be evaluated again because prev may have moved
3012 * CPUs since it called schedule(), thus the 'rq' on its stack
3013 * frame will be invalid.
3014 */
3015 finish_task_switch(this_rq(), prev);
1da177e4
LT
3016}
3017
3018/*
3019 * nr_running, nr_uninterruptible and nr_context_switches:
3020 *
3021 * externally visible scheduler statistics: current number of runnable
3022 * threads, current number of uninterruptible-sleeping threads, total
3023 * number of context switches performed since bootup.
3024 */
3025unsigned long nr_running(void)
3026{
3027 unsigned long i, sum = 0;
3028
3029 for_each_online_cpu(i)
3030 sum += cpu_rq(i)->nr_running;
3031
3032 return sum;
f711f609 3033}
1da177e4
LT
3034
3035unsigned long nr_uninterruptible(void)
f711f609 3036{
1da177e4 3037 unsigned long i, sum = 0;
f711f609 3038
0a945022 3039 for_each_possible_cpu(i)
1da177e4 3040 sum += cpu_rq(i)->nr_uninterruptible;
f711f609
GS
3041
3042 /*
1da177e4
LT
3043 * Since we read the counters lockless, it might be slightly
3044 * inaccurate. Do not allow it to go below zero though:
f711f609 3045 */
1da177e4
LT
3046 if (unlikely((long)sum < 0))
3047 sum = 0;
f711f609 3048
1da177e4 3049 return sum;
f711f609 3050}
f711f609 3051
1da177e4 3052unsigned long long nr_context_switches(void)
46cb4b7c 3053{
cc94abfc
SR
3054 int i;
3055 unsigned long long sum = 0;
46cb4b7c 3056
0a945022 3057 for_each_possible_cpu(i)
1da177e4 3058 sum += cpu_rq(i)->nr_switches;
46cb4b7c 3059
1da177e4
LT
3060 return sum;
3061}
483b4ee6 3062
1da177e4
LT
3063unsigned long nr_iowait(void)
3064{
3065 unsigned long i, sum = 0;
483b4ee6 3066
0a945022 3067 for_each_possible_cpu(i)
1da177e4 3068 sum += atomic_read(&cpu_rq(i)->nr_iowait);
46cb4b7c 3069
1da177e4
LT
3070 return sum;
3071}
483b4ee6 3072
8c215bd3 3073unsigned long nr_iowait_cpu(int cpu)
69d25870 3074{
8c215bd3 3075 struct rq *this = cpu_rq(cpu);
69d25870
AV
3076 return atomic_read(&this->nr_iowait);
3077}
46cb4b7c 3078
69d25870
AV
3079unsigned long this_cpu_load(void)
3080{
3081 struct rq *this = this_rq();
3082 return this->cpu_load[0];
3083}
e790fb0b 3084
46cb4b7c 3085
dce48a84
TG
3086/* Variables and functions for calc_load */
3087static atomic_long_t calc_load_tasks;
3088static unsigned long calc_load_update;
3089unsigned long avenrun[3];
3090EXPORT_SYMBOL(avenrun);
46cb4b7c 3091
74f5187a
PZ
3092static long calc_load_fold_active(struct rq *this_rq)
3093{
3094 long nr_active, delta = 0;
3095
3096 nr_active = this_rq->nr_running;
3097 nr_active += (long) this_rq->nr_uninterruptible;
3098
3099 if (nr_active != this_rq->calc_load_active) {
3100 delta = nr_active - this_rq->calc_load_active;
3101 this_rq->calc_load_active = nr_active;
3102 }
3103
3104 return delta;
3105}
3106
3107#ifdef CONFIG_NO_HZ
3108/*
3109 * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
3110 *
3111 * When making the ILB scale, we should try to pull this in as well.
3112 */
3113static atomic_long_t calc_load_tasks_idle;
3114
3115static void calc_load_account_idle(struct rq *this_rq)
3116{
3117 long delta;
3118
3119 delta = calc_load_fold_active(this_rq);
3120 if (delta)
3121 atomic_long_add(delta, &calc_load_tasks_idle);
3122}
3123
3124static long calc_load_fold_idle(void)
3125{
3126 long delta = 0;
3127
3128 /*
3129 * Its got a race, we don't care...
3130 */
3131 if (atomic_long_read(&calc_load_tasks_idle))
3132 delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
3133
3134 return delta;
3135}
3136#else
3137static void calc_load_account_idle(struct rq *this_rq)
3138{
3139}
3140
3141static inline long calc_load_fold_idle(void)
3142{
3143 return 0;
3144}
3145#endif
3146
2d02494f
TG
3147/**
3148 * get_avenrun - get the load average array
3149 * @loads: pointer to dest load array
3150 * @offset: offset to add
3151 * @shift: shift count to shift the result left
3152 *
3153 * These values are estimates at best, so no need for locking.
3154 */
3155void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
3156{
3157 loads[0] = (avenrun[0] + offset) << shift;
3158 loads[1] = (avenrun[1] + offset) << shift;
3159 loads[2] = (avenrun[2] + offset) << shift;
46cb4b7c 3160}
46cb4b7c 3161
dce48a84
TG
3162static unsigned long
3163calc_load(unsigned long load, unsigned long exp, unsigned long active)
db1b1fef 3164{
dce48a84
TG
3165 load *= exp;
3166 load += active * (FIXED_1 - exp);
3167 return load >> FSHIFT;
3168}
46cb4b7c
SS
3169
3170/*
dce48a84
TG
3171 * calc_load - update the avenrun load estimates 10 ticks after the
3172 * CPUs have updated calc_load_tasks.
7835b98b 3173 */
dce48a84 3174void calc_global_load(void)
7835b98b 3175{
dce48a84
TG
3176 unsigned long upd = calc_load_update + 10;
3177 long active;
1da177e4 3178
dce48a84
TG
3179 if (time_before(jiffies, upd))
3180 return;
1da177e4 3181
dce48a84
TG
3182 active = atomic_long_read(&calc_load_tasks);
3183 active = active > 0 ? active * FIXED_1 : 0;
1da177e4 3184
dce48a84
TG
3185 avenrun[0] = calc_load(avenrun[0], EXP_1, active);
3186 avenrun[1] = calc_load(avenrun[1], EXP_5, active);
3187 avenrun[2] = calc_load(avenrun[2], EXP_15, active);
dd41f596 3188
dce48a84
TG
3189 calc_load_update += LOAD_FREQ;
3190}
1da177e4 3191
dce48a84 3192/*
74f5187a
PZ
3193 * Called from update_cpu_load() to periodically update this CPU's
3194 * active count.
dce48a84
TG
3195 */
3196static void calc_load_account_active(struct rq *this_rq)
3197{
74f5187a 3198 long delta;
08c183f3 3199
74f5187a
PZ
3200 if (time_before(jiffies, this_rq->calc_load_update))
3201 return;
783609c6 3202
74f5187a
PZ
3203 delta = calc_load_fold_active(this_rq);
3204 delta += calc_load_fold_idle();
3205 if (delta)
dce48a84 3206 atomic_long_add(delta, &calc_load_tasks);
74f5187a
PZ
3207
3208 this_rq->calc_load_update += LOAD_FREQ;
46cb4b7c
SS
3209}
3210
fdf3e95d
VP
3211/*
3212 * The exact cpuload at various idx values, calculated at every tick would be
3213 * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
3214 *
3215 * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called
3216 * on nth tick when cpu may be busy, then we have:
3217 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3218 * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load
3219 *
3220 * decay_load_missed() below does efficient calculation of
3221 * load = ((2^idx - 1) / 2^idx)^(n-1) * load
3222 * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load
3223 *
3224 * The calculation is approximated on a 128 point scale.
3225 * degrade_zero_ticks is the number of ticks after which load at any
3226 * particular idx is approximated to be zero.
3227 * degrade_factor is a precomputed table, a row for each load idx.
3228 * Each column corresponds to degradation factor for a power of two ticks,
3229 * based on 128 point scale.
3230 * Example:
3231 * row 2, col 3 (=12) says that the degradation at load idx 2 after
3232 * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8).
3233 *
3234 * With this power of 2 load factors, we can degrade the load n times
3235 * by looking at 1 bits in n and doing as many mult/shift instead of
3236 * n mult/shifts needed by the exact degradation.
3237 */
3238#define DEGRADE_SHIFT 7
3239static const unsigned char
3240 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
3241static const unsigned char
3242 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
3243 {0, 0, 0, 0, 0, 0, 0, 0},
3244 {64, 32, 8, 0, 0, 0, 0, 0},
3245 {96, 72, 40, 12, 1, 0, 0},
3246 {112, 98, 75, 43, 15, 1, 0},
3247 {120, 112, 98, 76, 45, 16, 2} };
3248
3249/*
3250 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
3251 * would be when CPU is idle and so we just decay the old load without
3252 * adding any new load.
3253 */
3254static unsigned long
3255decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
3256{
3257 int j = 0;
3258
3259 if (!missed_updates)
3260 return load;
3261
3262 if (missed_updates >= degrade_zero_ticks[idx])
3263 return 0;
3264
3265 if (idx == 1)
3266 return load >> missed_updates;
3267
3268 while (missed_updates) {
3269 if (missed_updates % 2)
3270 load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;
3271
3272 missed_updates >>= 1;
3273 j++;
3274 }
3275 return load;
3276}
3277
46cb4b7c 3278/*
dd41f596 3279 * Update rq->cpu_load[] statistics. This function is usually called every
fdf3e95d
VP
3280 * scheduler tick (TICK_NSEC). With tickless idle this will not be called
3281 * every tick. We fix it up based on jiffies.
46cb4b7c 3282 */
dd41f596 3283static void update_cpu_load(struct rq *this_rq)
46cb4b7c 3284{
495eca49 3285 unsigned long this_load = this_rq->load.weight;
fdf3e95d
VP
3286 unsigned long curr_jiffies = jiffies;
3287 unsigned long pending_updates;
dd41f596 3288 int i, scale;
46cb4b7c 3289
dd41f596 3290 this_rq->nr_load_updates++;
46cb4b7c 3291
fdf3e95d
VP
3292 /* Avoid repeated calls on same jiffy, when moving in and out of idle */
3293 if (curr_jiffies == this_rq->last_load_update_tick)
3294 return;
3295
3296 pending_updates = curr_jiffies - this_rq->last_load_update_tick;
3297 this_rq->last_load_update_tick = curr_jiffies;
3298
dd41f596 3299 /* Update our load: */
fdf3e95d
VP
3300 this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
3301 for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
dd41f596 3302 unsigned long old_load, new_load;
7d1e6a9b 3303
dd41f596 3304 /* scale is effectively 1 << i now, and >> i divides by scale */
46cb4b7c 3305
dd41f596 3306 old_load = this_rq->cpu_load[i];
fdf3e95d 3307 old_load = decay_load_missed(old_load, pending_updates - 1, i);
dd41f596 3308 new_load = this_load;
a25707f3
IM
3309 /*
3310 * Round up the averaging division if load is increasing. This
3311 * prevents us from getting stuck on 9 if the load is 10, for
3312 * example.
3313 */
3314 if (new_load > old_load)
fdf3e95d
VP
3315 new_load += scale - 1;
3316
3317 this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
dd41f596 3318 }
da2b71ed
SS
3319
3320 sched_avg_update(this_rq);
fdf3e95d
VP
3321}
3322
3323static void update_cpu_load_active(struct rq *this_rq)
3324{
3325 update_cpu_load(this_rq);
46cb4b7c 3326
74f5187a 3327 calc_load_account_active(this_rq);
46cb4b7c
SS
3328}
3329
dd41f596 3330#ifdef CONFIG_SMP
8a0be9ef 3331
46cb4b7c 3332/*
38022906
PZ
3333 * sched_exec - execve() is a valuable balancing opportunity, because at
3334 * this point the task has the smallest effective memory and cache footprint.
46cb4b7c 3335 */
38022906 3336void sched_exec(void)
46cb4b7c 3337{
38022906 3338 struct task_struct *p = current;
1da177e4 3339 unsigned long flags;
70b97a7f 3340 struct rq *rq;
0017d735 3341 int dest_cpu;
46cb4b7c 3342
1da177e4 3343 rq = task_rq_lock(p, &flags);
0017d735
PZ
3344 dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0);
3345 if (dest_cpu == smp_processor_id())
3346 goto unlock;
38022906 3347
46cb4b7c 3348 /*
38022906 3349 * select_task_rq() can race against ->cpus_allowed
46cb4b7c 3350 */
30da688e 3351 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) &&
969c7921
TH
3352 likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) {
3353 struct migration_arg arg = { p, dest_cpu };
46cb4b7c 3354
1da177e4 3355 task_rq_unlock(rq, &flags);
969c7921 3356 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1da177e4
LT
3357 return;
3358 }
0017d735 3359unlock:
1da177e4 3360 task_rq_unlock(rq, &flags);
1da177e4 3361}
dd41f596 3362
1da177e4
LT
3363#endif
3364
1da177e4
LT
3365DEFINE_PER_CPU(struct kernel_stat, kstat);
3366
3367EXPORT_PER_CPU_SYMBOL(kstat);
3368
3369/*
c5f8d995 3370 * Return any ns on the sched_clock that have not yet been accounted in
f06febc9 3371 * @p in case that task is currently running.
c5f8d995
HS
3372 *
3373 * Called with task_rq_lock() held on @rq.
1da177e4 3374 */
c5f8d995
HS
3375static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3376{
3377 u64 ns = 0;
3378
3379 if (task_current(rq, p)) {
3380 update_rq_clock(rq);
305e6835 3381 ns = rq->clock_task - p->se.exec_start;
c5f8d995
HS
3382 if ((s64)ns < 0)
3383 ns = 0;
3384 }
3385
3386 return ns;
3387}
3388
bb34d92f 3389unsigned long long task_delta_exec(struct task_struct *p)
1da177e4 3390{
1da177e4 3391 unsigned long flags;
41b86e9c 3392 struct rq *rq;
bb34d92f 3393 u64 ns = 0;
48f24c4d 3394
41b86e9c 3395 rq = task_rq_lock(p, &flags);
c5f8d995
HS
3396 ns = do_task_delta_exec(p, rq);
3397 task_rq_unlock(rq, &flags);
1508487e 3398
c5f8d995
HS
3399 return ns;
3400}
f06febc9 3401
c5f8d995
HS
3402/*
3403 * Return accounted runtime for the task.
3404 * In case the task is currently running, return the runtime plus current's
3405 * pending runtime that have not been accounted yet.
3406 */
3407unsigned long long task_sched_runtime(struct task_struct *p)
3408{
3409 unsigned long flags;
3410 struct rq *rq;
3411 u64 ns = 0;
3412
3413 rq = task_rq_lock(p, &flags);
3414 ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
3415 task_rq_unlock(rq, &flags);
3416
3417 return ns;
3418}
48f24c4d 3419
c5f8d995
HS
3420/*
3421 * Return sum_exec_runtime for the thread group.
3422 * In case the task is currently running, return the sum plus current's
3423 * pending runtime that have not been accounted yet.
3424 *
3425 * Note that the thread group might have other running tasks as well,
3426 * so the return value not includes other pending runtime that other
3427 * running tasks might have.
3428 */
3429unsigned long long thread_group_sched_runtime(struct task_struct *p)
3430{
3431 struct task_cputime totals;
3432 unsigned long flags;
3433 struct rq *rq;
3434 u64 ns;
3435
3436 rq = task_rq_lock(p, &flags);
3437 thread_group_cputime(p, &totals);
3438 ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq);
41b86e9c 3439 task_rq_unlock(rq, &flags);
48f24c4d 3440
1da177e4
LT
3441 return ns;
3442}
3443
1da177e4
LT
3444/*
3445 * Account user cpu time to a process.
3446 * @p: the process that the cpu time gets accounted to
1da177e4 3447 * @cputime: the cpu time spent in user space since the last update
457533a7 3448 * @cputime_scaled: cputime scaled by cpu frequency
1da177e4 3449 */
457533a7
MS
3450void account_user_time(struct task_struct *p, cputime_t cputime,
3451 cputime_t cputime_scaled)
1da177e4
LT
3452{
3453 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3454 cputime64_t tmp;
3455
457533a7 3456 /* Add user time to process. */
1da177e4 3457 p->utime = cputime_add(p->utime, cputime);
457533a7 3458 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
f06febc9 3459 account_group_user_time(p, cputime);
1da177e4
LT
3460
3461 /* Add user time to cpustat. */
3462 tmp = cputime_to_cputime64(cputime);
3463 if (TASK_NICE(p) > 0)
3464 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3465 else
3466 cpustat->user = cputime64_add(cpustat->user, tmp);
ef12fefa
BR
3467
3468 cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime);
49b5cf34
JL
3469 /* Account for user time used */
3470 acct_update_integrals(p);
1da177e4
LT
3471}
3472
94886b84
LV
3473/*
3474 * Account guest cpu time to a process.
3475 * @p: the process that the cpu time gets accounted to
3476 * @cputime: the cpu time spent in virtual machine since the last update
457533a7 3477 * @cputime_scaled: cputime scaled by cpu frequency
94886b84 3478 */
457533a7
MS
3479static void account_guest_time(struct task_struct *p, cputime_t cputime,
3480 cputime_t cputime_scaled)
94886b84
LV
3481{
3482 cputime64_t tmp;
3483 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3484
3485 tmp = cputime_to_cputime64(cputime);
3486
457533a7 3487 /* Add guest time to process. */
94886b84 3488 p->utime = cputime_add(p->utime, cputime);
457533a7 3489 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
f06febc9 3490 account_group_user_time(p, cputime);
94886b84
LV
3491 p->gtime = cputime_add(p->gtime, cputime);
3492
457533a7 3493 /* Add guest time to cpustat. */
ce0e7b28
RO
3494 if (TASK_NICE(p) > 0) {
3495 cpustat->nice = cputime64_add(cpustat->nice, tmp);
3496 cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp);
3497 } else {
3498 cpustat->user = cputime64_add(cpustat->user, tmp);
3499 cpustat->guest = cputime64_add(cpustat->guest, tmp);
3500 }
94886b84
LV
3501}
3502
1da177e4
LT
3503/*
3504 * Account system cpu time to a process.
3505 * @p: the process that the cpu time gets accounted to
3506 * @hardirq_offset: the offset to subtract from hardirq_count()
3507 * @cputime: the cpu time spent in kernel space since the last update
457533a7 3508 * @cputime_scaled: cputime scaled by cpu frequency
1da177e4
LT
3509 */
3510void account_system_time(struct task_struct *p, int hardirq_offset,
457533a7 3511 cputime_t cputime, cputime_t cputime_scaled)
1da177e4
LT
3512{
3513 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
1da177e4
LT
3514 cputime64_t tmp;
3515
983ed7a6 3516 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
457533a7 3517 account_guest_time(p, cputime, cputime_scaled);
983ed7a6
HH
3518 return;
3519 }
94886b84 3520
457533a7 3521 /* Add system time to process. */
1da177e4 3522 p->stime = cputime_add(p->stime, cputime);
457533a7 3523 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
f06febc9 3524 account_group_system_time(p, cputime);
1da177e4
LT
3525
3526 /* Add system time to cpustat. */
3527 tmp = cputime_to_cputime64(cputime);
3528 if (hardirq_count() - hardirq_offset)
3529 cpustat->irq = cputime64_add(cpustat->irq, tmp);
75e1056f 3530 else if (in_serving_softirq())
1da177e4 3531 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
1da177e4 3532 else
79741dd3
MS
3533 cpustat->system = cputime64_add(cpustat->system, tmp);
3534
ef12fefa
BR
3535 cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime);
3536
1da177e4
LT
3537 /* Account for system time used */
3538 acct_update_integrals(p);
1da177e4
LT
3539}
3540
c66f08be 3541/*
1da177e4 3542 * Account for involuntary wait time.
1da177e4 3543 * @steal: the cpu time spent in involuntary wait
c66f08be 3544 */
79741dd3 3545void account_steal_time(cputime_t cputime)
c66f08be 3546{
79741dd3
MS
3547 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
3548 cputime64_t cputime64 = cputime_to_cputime64(cputime);
3549
3550 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
c66f08be
MN
3551}
3552
1da177e4 3553/*
79741dd3
MS
3554 * Account for idle time.
3555 * @cputime: the cpu time spent in idle wait
1da177e4 3556 */
79741dd3 3557void account_idle_time(cputime_t cputime)
1da177e4
LT
3558{
3559 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
79741dd3 3560 cputime64_t cputime64 = cputime_to_cputime64(cputime);
70b97a7f 3561 struct rq *rq = this_rq();
1da177e4 3562
79741dd3
MS
3563 if (atomic_read(&rq->nr_iowait) > 0)
3564 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
3565 else
3566 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
1da177e4
LT
3567}
3568
79741dd3
MS
3569#ifndef CONFIG_VIRT_CPU_ACCOUNTING
3570
3571/*
3572 * Account a single tick of cpu time.
3573 * @p: the process that the cpu time gets accounted to
3574 * @user_tick: indicates if the tick is a user or a system tick
3575 */
3576void account_process_tick(struct task_struct *p, int user_tick)
3577{
a42548a1 3578 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
79741dd3
MS
3579 struct rq *rq = this_rq();
3580
3581 if (user_tick)
a42548a1 3582 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
f5f293a4 3583 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
a42548a1 3584 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
79741dd3
MS
3585 one_jiffy_scaled);
3586 else
a42548a1 3587 account_idle_time(cputime_one_jiffy);
79741dd3
MS
3588}
3589
3590/*
3591 * Account multiple ticks of steal time.
3592 * @p: the process from which the cpu time has been stolen
3593 * @ticks: number of stolen ticks
3594 */
3595void account_steal_ticks(unsigned long ticks)
3596{
3597 account_steal_time(jiffies_to_cputime(ticks));
3598}
3599
3600/*
3601 * Account multiple ticks of idle time.
3602 * @ticks: number of stolen ticks
3603 */
3604void account_idle_ticks(unsigned long ticks)
3605{
3606 account_idle_time(jiffies_to_cputime(ticks));
1da177e4
LT
3607}
3608
79741dd3
MS
3609#endif
3610
49048622
BS
3611/*
3612 * Use precise platform statistics if available:
3613 */
3614#ifdef CONFIG_VIRT_CPU_ACCOUNTING
d180c5bc 3615void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3616{
d99ca3b9
HS
3617 *ut = p->utime;
3618 *st = p->stime;
49048622
BS
3619}
3620
0cf55e1e 3621void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3622{
0cf55e1e
HS
3623 struct task_cputime cputime;
3624
3625 thread_group_cputime(p, &cputime);
3626
3627 *ut = cputime.utime;
3628 *st = cputime.stime;
49048622
BS
3629}
3630#else
761b1d26
HS
3631
3632#ifndef nsecs_to_cputime
b7b20df9 3633# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
761b1d26
HS
3634#endif
3635
d180c5bc 3636void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3637{
d99ca3b9 3638 cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime);
49048622
BS
3639
3640 /*
3641 * Use CFS's precise accounting:
3642 */
d180c5bc 3643 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
49048622
BS
3644
3645 if (total) {
e75e863d 3646 u64 temp = rtime;
d180c5bc 3647
e75e863d 3648 temp *= utime;
49048622 3649 do_div(temp, total);
d180c5bc
HS
3650 utime = (cputime_t)temp;
3651 } else
3652 utime = rtime;
49048622 3653
d180c5bc
HS
3654 /*
3655 * Compare with previous values, to keep monotonicity:
3656 */
761b1d26 3657 p->prev_utime = max(p->prev_utime, utime);
d99ca3b9 3658 p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime));
49048622 3659
d99ca3b9
HS
3660 *ut = p->prev_utime;
3661 *st = p->prev_stime;
49048622
BS
3662}
3663
0cf55e1e
HS
3664/*
3665 * Must be called with siglock held.
3666 */
3667void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
49048622 3668{
0cf55e1e
HS
3669 struct signal_struct *sig = p->signal;
3670 struct task_cputime cputime;
3671 cputime_t rtime, utime, total;
49048622 3672
0cf55e1e 3673 thread_group_cputime(p, &cputime);
49048622 3674
0cf55e1e
HS
3675 total = cputime_add(cputime.utime, cputime.stime);
3676 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
49048622 3677
0cf55e1e 3678 if (total) {
e75e863d 3679 u64 temp = rtime;
49048622 3680
e75e863d 3681 temp *= cputime.utime;
0cf55e1e
HS
3682 do_div(temp, total);
3683 utime = (cputime_t)temp;
3684 } else
3685 utime = rtime;
3686
3687 sig->prev_utime = max(sig->prev_utime, utime);
3688 sig->prev_stime = max(sig->prev_stime,
3689 cputime_sub(rtime, sig->prev_utime));
3690
3691 *ut = sig->prev_utime;
3692 *st = sig->prev_stime;
49048622 3693}
49048622 3694#endif
49048622 3695
7835b98b
CL
3696/*
3697 * This function gets called by the timer code, with HZ frequency.
3698 * We call it with interrupts disabled.
3699 *
3700 * It also gets called by the fork code, when changing the parent's
3701 * timeslices.
3702 */
3703void scheduler_tick(void)
3704{
7835b98b
CL
3705 int cpu = smp_processor_id();
3706 struct rq *rq = cpu_rq(cpu);
dd41f596 3707 struct task_struct *curr = rq->curr;
3e51f33f
PZ
3708
3709 sched_clock_tick();
dd41f596 3710
05fa785c 3711 raw_spin_lock(&rq->lock);
3e51f33f 3712 update_rq_clock(rq);
fdf3e95d 3713 update_cpu_load_active(rq);
fa85ae24 3714 curr->sched_class->task_tick(rq, curr, 0);
05fa785c 3715 raw_spin_unlock(&rq->lock);
7835b98b 3716
e9d2b064 3717 perf_event_task_tick();
e220d2dc 3718
e418e1c2 3719#ifdef CONFIG_SMP
dd41f596
IM
3720 rq->idle_at_tick = idle_cpu(cpu);
3721 trigger_load_balance(rq, cpu);
e418e1c2 3722#endif
1da177e4
LT
3723}
3724
132380a0 3725notrace unsigned long get_parent_ip(unsigned long addr)
6cd8a4bb
SR
3726{
3727 if (in_lock_functions(addr)) {
3728 addr = CALLER_ADDR2;
3729 if (in_lock_functions(addr))
3730 addr = CALLER_ADDR3;
3731 }
3732 return addr;
3733}
1da177e4 3734
7e49fcce
SR
3735#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3736 defined(CONFIG_PREEMPT_TRACER))
3737
43627582 3738void __kprobes add_preempt_count(int val)
1da177e4 3739{
6cd8a4bb 3740#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3741 /*
3742 * Underflow?
3743 */
9a11b49a
IM
3744 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
3745 return;
6cd8a4bb 3746#endif
1da177e4 3747 preempt_count() += val;
6cd8a4bb 3748#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3749 /*
3750 * Spinlock count overflowing soon?
3751 */
33859f7f
MOS
3752 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
3753 PREEMPT_MASK - 10);
6cd8a4bb
SR
3754#endif
3755 if (preempt_count() == val)
3756 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
1da177e4
LT
3757}
3758EXPORT_SYMBOL(add_preempt_count);
3759
43627582 3760void __kprobes sub_preempt_count(int val)
1da177e4 3761{
6cd8a4bb 3762#ifdef CONFIG_DEBUG_PREEMPT
1da177e4
LT
3763 /*
3764 * Underflow?
3765 */
01e3eb82 3766 if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
9a11b49a 3767 return;
1da177e4
LT
3768 /*
3769 * Is the spinlock portion underflowing?
3770 */
9a11b49a
IM
3771 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
3772 !(preempt_count() & PREEMPT_MASK)))
3773 return;
6cd8a4bb 3774#endif
9a11b49a 3775
6cd8a4bb
SR
3776 if (preempt_count() == val)
3777 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
1da177e4
LT
3778 preempt_count() -= val;
3779}
3780EXPORT_SYMBOL(sub_preempt_count);
3781
3782#endif
3783
3784/*
dd41f596 3785 * Print scheduling while atomic bug:
1da177e4 3786 */
dd41f596 3787static noinline void __schedule_bug(struct task_struct *prev)
1da177e4 3788{
838225b4
SS
3789 struct pt_regs *regs = get_irq_regs();
3790
3df0fc5b
PZ
3791 printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
3792 prev->comm, prev->pid, preempt_count());
838225b4 3793
dd41f596 3794 debug_show_held_locks(prev);
e21f5b15 3795 print_modules();
dd41f596
IM
3796 if (irqs_disabled())
3797 print_irqtrace_events(prev);
838225b4
SS
3798
3799 if (regs)
3800 show_regs(regs);
3801 else
3802 dump_stack();
dd41f596 3803}
1da177e4 3804
dd41f596
IM
3805/*
3806 * Various schedule()-time debugging checks and statistics:
3807 */
3808static inline void schedule_debug(struct task_struct *prev)
3809{
1da177e4 3810 /*
41a2d6cf 3811 * Test if we are atomic. Since do_exit() needs to call into
1da177e4
LT
3812 * schedule() atomically, we ignore that path for now.
3813 * Otherwise, whine if we are scheduling when we should not be.
3814 */
3f33a7ce 3815 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
dd41f596
IM
3816 __schedule_bug(prev);
3817
1da177e4
LT
3818 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
3819
2d72376b 3820 schedstat_inc(this_rq(), sched_count);
b8efb561
IM
3821#ifdef CONFIG_SCHEDSTATS
3822 if (unlikely(prev->lock_depth >= 0)) {
2d72376b
IM
3823 schedstat_inc(this_rq(), bkl_count);
3824 schedstat_inc(prev, sched_info.bkl_count);
b8efb561
IM
3825 }
3826#endif
dd41f596
IM
3827}
3828
6cecd084 3829static void put_prev_task(struct rq *rq, struct task_struct *prev)
df1c99d4 3830{
a64692a3
MG
3831 if (prev->se.on_rq)
3832 update_rq_clock(rq);
3833 rq->skip_clock_update = 0;
6cecd084 3834 prev->sched_class->put_prev_task(rq, prev);
df1c99d4
MG
3835}
3836
dd41f596
IM
3837/*
3838 * Pick up the highest-prio task:
3839 */
3840static inline struct task_struct *
b67802ea 3841pick_next_task(struct rq *rq)
dd41f596 3842{
5522d5d5 3843 const struct sched_class *class;
dd41f596 3844 struct task_struct *p;
1da177e4
LT
3845
3846 /*
dd41f596
IM
3847 * Optimization: we know that if all tasks are in
3848 * the fair class we can call that function directly:
1da177e4 3849 */
dd41f596 3850 if (likely(rq->nr_running == rq->cfs.nr_running)) {
fb8d4724 3851 p = fair_sched_class.pick_next_task(rq);
dd41f596
IM
3852 if (likely(p))
3853 return p;
1da177e4
LT
3854 }
3855
34f971f6 3856 for_each_class(class) {
fb8d4724 3857 p = class->pick_next_task(rq);
dd41f596
IM
3858 if (p)
3859 return p;
dd41f596 3860 }
34f971f6
PZ
3861
3862 BUG(); /* the idle class will always have a runnable task */
dd41f596 3863}
1da177e4 3864
dd41f596
IM
3865/*
3866 * schedule() is the main scheduler function.
3867 */
ff743345 3868asmlinkage void __sched schedule(void)
dd41f596
IM
3869{
3870 struct task_struct *prev, *next;
67ca7bde 3871 unsigned long *switch_count;
dd41f596 3872 struct rq *rq;
31656519 3873 int cpu;
dd41f596 3874
ff743345
PZ
3875need_resched:
3876 preempt_disable();
dd41f596
IM
3877 cpu = smp_processor_id();
3878 rq = cpu_rq(cpu);
25502a6c 3879 rcu_note_context_switch(cpu);
dd41f596 3880 prev = rq->curr;
dd41f596
IM
3881
3882 release_kernel_lock(prev);
3883need_resched_nonpreemptible:
3884
3885 schedule_debug(prev);
1da177e4 3886
31656519 3887 if (sched_feat(HRTICK))
f333fdc9 3888 hrtick_clear(rq);
8f4d37ec 3889
05fa785c 3890 raw_spin_lock_irq(&rq->lock);
1e819950 3891 clear_tsk_need_resched(prev);
1da177e4 3892
246d86b5 3893 switch_count = &prev->nivcsw;
1da177e4 3894 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
21aa9af0 3895 if (unlikely(signal_pending_state(prev->state, prev))) {
1da177e4 3896 prev->state = TASK_RUNNING;
21aa9af0
TH
3897 } else {
3898 /*
3899 * If a worker is going to sleep, notify and
3900 * ask workqueue whether it wants to wake up a
3901 * task to maintain concurrency. If so, wake
3902 * up the task.
3903 */
3904 if (prev->flags & PF_WQ_WORKER) {
3905 struct task_struct *to_wakeup;
3906
3907 to_wakeup = wq_worker_sleeping(prev, cpu);
3908 if (to_wakeup)
3909 try_to_wake_up_local(to_wakeup);
3910 }
371fd7e7 3911 deactivate_task(rq, prev, DEQUEUE_SLEEP);
21aa9af0 3912 }
dd41f596 3913 switch_count = &prev->nvcsw;
1da177e4
LT
3914 }
3915
3f029d3c 3916 pre_schedule(rq, prev);
f65eda4f 3917
dd41f596 3918 if (unlikely(!rq->nr_running))
1da177e4 3919 idle_balance(cpu, rq);
1da177e4 3920
df1c99d4 3921 put_prev_task(rq, prev);
b67802ea 3922 next = pick_next_task(rq);
1da177e4 3923
1da177e4 3924 if (likely(prev != next)) {
673a90a1 3925 sched_info_switch(prev, next);
49f47433 3926 perf_event_task_sched_out(prev, next);
673a90a1 3927
1da177e4
LT
3928 rq->nr_switches++;
3929 rq->curr = next;
3930 ++*switch_count;
3931
dd41f596 3932 context_switch(rq, prev, next); /* unlocks the rq */
8f4d37ec 3933 /*
246d86b5
ON
3934 * The context switch have flipped the stack from under us
3935 * and restored the local variables which were saved when
3936 * this task called schedule() in the past. prev == current
3937 * is still correct, but it can be moved to another cpu/rq.
8f4d37ec
PZ
3938 */
3939 cpu = smp_processor_id();
3940 rq = cpu_rq(cpu);
1da177e4 3941 } else
05fa785c 3942 raw_spin_unlock_irq(&rq->lock);
1da177e4 3943
3f029d3c 3944 post_schedule(rq);
1da177e4 3945
246d86b5 3946 if (unlikely(reacquire_kernel_lock(prev)))
1da177e4 3947 goto need_resched_nonpreemptible;
8f4d37ec 3948
1da177e4 3949 preempt_enable_no_resched();
ff743345 3950 if (need_resched())
1da177e4
LT
3951 goto need_resched;
3952}
1da177e4
LT
3953EXPORT_SYMBOL(schedule);
3954
c08f7829 3955#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
0d66bf6d
PZ
3956/*
3957 * Look out! "owner" is an entirely speculative pointer
3958 * access and not reliable.
3959 */
3960int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
3961{
3962 unsigned int cpu;
3963 struct rq *rq;
3964
3965 if (!sched_feat(OWNER_SPIN))
3966 return 0;
3967
3968#ifdef CONFIG_DEBUG_PAGEALLOC
3969 /*
3970 * Need to access the cpu field knowing that
3971 * DEBUG_PAGEALLOC could have unmapped it if
3972 * the mutex owner just released it and exited.
3973 */
3974 if (probe_kernel_address(&owner->cpu, cpu))
4b402210 3975 return 0;
0d66bf6d
PZ
3976#else
3977 cpu = owner->cpu;
3978#endif
3979
3980 /*
3981 * Even if the access succeeded (likely case),
3982 * the cpu field may no longer be valid.
3983 */
3984 if (cpu >= nr_cpumask_bits)
4b402210 3985 return 0;
0d66bf6d
PZ
3986
3987 /*
3988 * We need to validate that we can do a
3989 * get_cpu() and that we have the percpu area.
3990 */
3991 if (!cpu_online(cpu))
4b402210 3992 return 0;
0d66bf6d
PZ
3993
3994 rq = cpu_rq(cpu);
3995
3996 for (;;) {
3997 /*
3998 * Owner changed, break to re-assess state.
3999 */
9d0f4dcc
TC
4000 if (lock->owner != owner) {
4001 /*
4002 * If the lock has switched to a different owner,
4003 * we likely have heavy contention. Return 0 to quit
4004 * optimistic spinning and not contend further:
4005 */
4006 if (lock->owner)
4007 return 0;
0d66bf6d 4008 break;
9d0f4dcc 4009 }
0d66bf6d
PZ
4010
4011 /*
4012 * Is that owner really running on that cpu?
4013 */
4014 if (task_thread_info(rq->curr) != owner || need_resched())
4015 return 0;
4016
4017 cpu_relax();
4018 }
4b402210 4019
0d66bf6d
PZ
4020 return 1;
4021}
4022#endif
4023
1da177e4
LT
4024#ifdef CONFIG_PREEMPT
4025/*
2ed6e34f 4026 * this is the entry point to schedule() from in-kernel preemption
41a2d6cf 4027 * off of preempt_enable. Kernel preemptions off return from interrupt
1da177e4
LT
4028 * occur there and call schedule directly.
4029 */
d1f74e20 4030asmlinkage void __sched notrace preempt_schedule(void)
1da177e4
LT
4031{
4032 struct thread_info *ti = current_thread_info();
6478d880 4033
1da177e4
LT
4034 /*
4035 * If there is a non-zero preempt_count or interrupts are disabled,
41a2d6cf 4036 * we do not want to preempt the current task. Just return..
1da177e4 4037 */
beed33a8 4038 if (likely(ti->preempt_count || irqs_disabled()))
1da177e4
LT
4039 return;
4040
3a5c359a 4041 do {
d1f74e20 4042 add_preempt_count_notrace(PREEMPT_ACTIVE);
3a5c359a 4043 schedule();
d1f74e20 4044 sub_preempt_count_notrace(PREEMPT_ACTIVE);
1da177e4 4045
3a5c359a
AK
4046 /*
4047 * Check again in case we missed a preemption opportunity
4048 * between schedule and now.
4049 */
4050 barrier();
5ed0cec0 4051 } while (need_resched());
1da177e4 4052}
1da177e4
LT
4053EXPORT_SYMBOL(preempt_schedule);
4054
4055/*
2ed6e34f 4056 * this is the entry point to schedule() from kernel preemption
1da177e4
LT
4057 * off of irq context.
4058 * Note, that this is called and return with irqs disabled. This will
4059 * protect us against recursive calling from irq.
4060 */
4061asmlinkage void __sched preempt_schedule_irq(void)
4062{
4063 struct thread_info *ti = current_thread_info();
6478d880 4064
2ed6e34f 4065 /* Catch callers which need to be fixed */
1da177e4
LT
4066 BUG_ON(ti->preempt_count || !irqs_disabled());
4067
3a5c359a
AK
4068 do {
4069 add_preempt_count(PREEMPT_ACTIVE);
3a5c359a
AK
4070 local_irq_enable();
4071 schedule();
4072 local_irq_disable();
3a5c359a 4073 sub_preempt_count(PREEMPT_ACTIVE);
1da177e4 4074
3a5c359a
AK
4075 /*
4076 * Check again in case we missed a preemption opportunity
4077 * between schedule and now.
4078 */
4079 barrier();
5ed0cec0 4080 } while (need_resched());
1da177e4
LT
4081}
4082
4083#endif /* CONFIG_PREEMPT */
4084
63859d4f 4085int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
95cdf3b7 4086 void *key)
1da177e4 4087{
63859d4f 4088 return try_to_wake_up(curr->private, mode, wake_flags);
1da177e4 4089}
1da177e4
LT
4090EXPORT_SYMBOL(default_wake_function);
4091
4092/*
41a2d6cf
IM
4093 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
4094 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
1da177e4
LT
4095 * number) then we wake all the non-exclusive tasks and one exclusive task.
4096 *
4097 * There are circumstances in which we can try to wake a task which has already
41a2d6cf 4098 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
1da177e4
LT
4099 * zero in this (rare) case, and we handle it by continuing to scan the queue.
4100 */
78ddb08f 4101static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
63859d4f 4102 int nr_exclusive, int wake_flags, void *key)
1da177e4 4103{
2e45874c 4104 wait_queue_t *curr, *next;
1da177e4 4105
2e45874c 4106 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
48f24c4d
IM
4107 unsigned flags = curr->flags;
4108
63859d4f 4109 if (curr->func(curr, mode, wake_flags, key) &&
48f24c4d 4110 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
1da177e4
LT
4111 break;
4112 }
4113}
4114
4115/**
4116 * __wake_up - wake up threads blocked on a waitqueue.
4117 * @q: the waitqueue
4118 * @mode: which threads
4119 * @nr_exclusive: how many wake-one or wake-many threads to wake up
67be2dd1 4120 * @key: is directly passed to the wakeup function
50fa610a
DH
4121 *
4122 * It may be assumed that this function implies a write memory barrier before
4123 * changing the task state if and only if any tasks are woken up.
1da177e4 4124 */
7ad5b3a5 4125void __wake_up(wait_queue_head_t *q, unsigned int mode,
95cdf3b7 4126 int nr_exclusive, void *key)
1da177e4
LT
4127{
4128 unsigned long flags;
4129
4130 spin_lock_irqsave(&q->lock, flags);
4131 __wake_up_common(q, mode, nr_exclusive, 0, key);
4132 spin_unlock_irqrestore(&q->lock, flags);
4133}
1da177e4
LT
4134EXPORT_SYMBOL(__wake_up);
4135
4136/*
4137 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
4138 */
7ad5b3a5 4139void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
1da177e4
LT
4140{
4141 __wake_up_common(q, mode, 1, 0, NULL);
4142}
22c43c81 4143EXPORT_SYMBOL_GPL(__wake_up_locked);
1da177e4 4144
4ede816a
DL
4145void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
4146{
4147 __wake_up_common(q, mode, 1, 0, key);
4148}
4149
1da177e4 4150/**
4ede816a 4151 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
1da177e4
LT
4152 * @q: the waitqueue
4153 * @mode: which threads
4154 * @nr_exclusive: how many wake-one or wake-many threads to wake up
4ede816a 4155 * @key: opaque value to be passed to wakeup targets
1da177e4
LT
4156 *
4157 * The sync wakeup differs that the waker knows that it will schedule
4158 * away soon, so while the target thread will be woken up, it will not
4159 * be migrated to another CPU - ie. the two threads are 'synchronized'
4160 * with each other. This can prevent needless bouncing between CPUs.
4161 *
4162 * On UP it can prevent extra preemption.
50fa610a
DH
4163 *
4164 * It may be assumed that this function implies a write memory barrier before
4165 * changing the task state if and only if any tasks are woken up.
1da177e4 4166 */
4ede816a
DL
4167void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4168 int nr_exclusive, void *key)
1da177e4
LT
4169{
4170 unsigned long flags;
7d478721 4171 int wake_flags = WF_SYNC;
1da177e4
LT
4172
4173 if (unlikely(!q))
4174 return;
4175
4176 if (unlikely(!nr_exclusive))
7d478721 4177 wake_flags = 0;
1da177e4
LT
4178
4179 spin_lock_irqsave(&q->lock, flags);
7d478721 4180 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
1da177e4
LT
4181 spin_unlock_irqrestore(&q->lock, flags);
4182}
4ede816a
DL
4183EXPORT_SYMBOL_GPL(__wake_up_sync_key);
4184
4185/*
4186 * __wake_up_sync - see __wake_up_sync_key()
4187 */
4188void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4189{
4190 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
4191}
1da177e4
LT
4192EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4193
65eb3dc6
KD
4194/**
4195 * complete: - signals a single thread waiting on this completion
4196 * @x: holds the state of this particular completion
4197 *
4198 * This will wake up a single thread waiting on this completion. Threads will be
4199 * awakened in the same order in which they were queued.
4200 *
4201 * See also complete_all(), wait_for_completion() and related routines.
50fa610a
DH
4202 *
4203 * It may be assumed that this function implies a write memory barrier before
4204 * changing the task state if and only if any tasks are woken up.
65eb3dc6 4205 */
b15136e9 4206void complete(struct completion *x)
1da177e4
LT
4207{
4208 unsigned long flags;
4209
4210 spin_lock_irqsave(&x->wait.lock, flags);
4211 x->done++;
d9514f6c 4212 __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
1da177e4
LT
4213 spin_unlock_irqrestore(&x->wait.lock, flags);
4214}
4215EXPORT_SYMBOL(complete);
4216
65eb3dc6
KD
4217/**
4218 * complete_all: - signals all threads waiting on this completion
4219 * @x: holds the state of this particular completion
4220 *
4221 * This will wake up all threads waiting on this particular completion event.
50fa610a
DH
4222 *
4223 * It may be assumed that this function implies a write memory barrier before
4224 * changing the task state if and only if any tasks are woken up.
65eb3dc6 4225 */
b15136e9 4226void complete_all(struct completion *x)
1da177e4
LT
4227{
4228 unsigned long flags;
4229
4230 spin_lock_irqsave(&x->wait.lock, flags);
4231 x->done += UINT_MAX/2;
d9514f6c 4232 __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
1da177e4
LT
4233 spin_unlock_irqrestore(&x->wait.lock, flags);
4234}
4235EXPORT_SYMBOL(complete_all);
4236
8cbbe86d
AK
4237static inline long __sched
4238do_wait_for_common(struct completion *x, long timeout, int state)
1da177e4 4239{
1da177e4
LT
4240 if (!x->done) {
4241 DECLARE_WAITQUEUE(wait, current);
4242
a93d2f17 4243 __add_wait_queue_tail_exclusive(&x->wait, &wait);
1da177e4 4244 do {
94d3d824 4245 if (signal_pending_state(state, current)) {
ea71a546
ON
4246 timeout = -ERESTARTSYS;
4247 break;
8cbbe86d
AK
4248 }
4249 __set_current_state(state);
1da177e4
LT
4250 spin_unlock_irq(&x->wait.lock);
4251 timeout = schedule_timeout(timeout);
4252 spin_lock_irq(&x->wait.lock);
ea71a546 4253 } while (!x->done && timeout);
1da177e4 4254 __remove_wait_queue(&x->wait, &wait);
ea71a546
ON
4255 if (!x->done)
4256 return timeout;
1da177e4
LT
4257 }
4258 x->done--;
ea71a546 4259 return timeout ?: 1;
1da177e4 4260}
1da177e4 4261
8cbbe86d
AK
4262static long __sched
4263wait_for_common(struct completion *x, long timeout, int state)
1da177e4 4264{
1da177e4
LT
4265 might_sleep();
4266
4267 spin_lock_irq(&x->wait.lock);
8cbbe86d 4268 timeout = do_wait_for_common(x, timeout, state);
1da177e4 4269 spin_unlock_irq(&x->wait.lock);
8cbbe86d
AK
4270 return timeout;
4271}
1da177e4 4272
65eb3dc6
KD
4273/**
4274 * wait_for_completion: - waits for completion of a task
4275 * @x: holds the state of this particular completion
4276 *
4277 * This waits to be signaled for completion of a specific task. It is NOT
4278 * interruptible and there is no timeout.
4279 *
4280 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
4281 * and interrupt capability. Also see complete().
4282 */
b15136e9 4283void __sched wait_for_completion(struct completion *x)
8cbbe86d
AK
4284{
4285 wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
1da177e4 4286}
8cbbe86d 4287EXPORT_SYMBOL(wait_for_completion);
1da177e4 4288
65eb3dc6
KD
4289/**
4290 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
4291 * @x: holds the state of this particular completion
4292 * @timeout: timeout value in jiffies
4293 *
4294 * This waits for either a completion of a specific task to be signaled or for a
4295 * specified timeout to expire. The timeout is in jiffies. It is not
4296 * interruptible.
4297 */
b15136e9 4298unsigned long __sched
8cbbe86d 4299wait_for_completion_timeout(struct completion *x, unsigned long timeout)
1da177e4 4300{
8cbbe86d 4301 return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
1da177e4 4302}
8cbbe86d 4303EXPORT_SYMBOL(wait_for_completion_timeout);
1da177e4 4304
65eb3dc6
KD
4305/**
4306 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
4307 * @x: holds the state of this particular completion
4308 *
4309 * This waits for completion of a specific task to be signaled. It is
4310 * interruptible.
4311 */
8cbbe86d 4312int __sched wait_for_completion_interruptible(struct completion *x)
0fec171c 4313{
51e97990
AK
4314 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
4315 if (t == -ERESTARTSYS)
4316 return t;
4317 return 0;
0fec171c 4318}
8cbbe86d 4319EXPORT_SYMBOL(wait_for_completion_interruptible);
1da177e4 4320
65eb3dc6
KD
4321/**
4322 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
4323 * @x: holds the state of this particular completion
4324 * @timeout: timeout value in jiffies
4325 *
4326 * This waits for either a completion of a specific task to be signaled or for a
4327 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
4328 */
b15136e9 4329unsigned long __sched
8cbbe86d
AK
4330wait_for_completion_interruptible_timeout(struct completion *x,
4331 unsigned long timeout)
0fec171c 4332{
8cbbe86d 4333 return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
0fec171c 4334}
8cbbe86d 4335EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
1da177e4 4336
65eb3dc6
KD
4337/**
4338 * wait_for_completion_killable: - waits for completion of a task (killable)
4339 * @x: holds the state of this particular completion
4340 *
4341 * This waits to be signaled for completion of a specific task. It can be
4342 * interrupted by a kill signal.
4343 */
009e577e
MW
4344int __sched wait_for_completion_killable(struct completion *x)
4345{
4346 long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
4347 if (t == -ERESTARTSYS)
4348 return t;
4349 return 0;
4350}
4351EXPORT_SYMBOL(wait_for_completion_killable);
4352
0aa12fb4
SW
4353/**
4354 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
4355 * @x: holds the state of this particular completion
4356 * @timeout: timeout value in jiffies
4357 *
4358 * This waits for either a completion of a specific task to be
4359 * signaled or for a specified timeout to expire. It can be
4360 * interrupted by a kill signal. The timeout is in jiffies.
4361 */
4362unsigned long __sched
4363wait_for_completion_killable_timeout(struct completion *x,
4364 unsigned long timeout)
4365{
4366 return wait_for_common(x, timeout, TASK_KILLABLE);
4367}
4368EXPORT_SYMBOL(wait_for_completion_killable_timeout);
4369
be4de352
DC
4370/**
4371 * try_wait_for_completion - try to decrement a completion without blocking
4372 * @x: completion structure
4373 *
4374 * Returns: 0 if a decrement cannot be done without blocking
4375 * 1 if a decrement succeeded.
4376 *
4377 * If a completion is being used as a counting completion,
4378 * attempt to decrement the counter without blocking. This
4379 * enables us to avoid waiting if the resource the completion
4380 * is protecting is not available.
4381 */
4382bool try_wait_for_completion(struct completion *x)
4383{
7539a3b3 4384 unsigned long flags;
be4de352
DC
4385 int ret = 1;
4386
7539a3b3 4387 spin_lock_irqsave(&x->wait.lock, flags);
be4de352
DC
4388 if (!x->done)
4389 ret = 0;
4390 else
4391 x->done--;
7539a3b3 4392 spin_unlock_irqrestore(&x->wait.lock, flags);
be4de352
DC
4393 return ret;
4394}
4395EXPORT_SYMBOL(try_wait_for_completion);
4396
4397/**
4398 * completion_done - Test to see if a completion has any waiters
4399 * @x: completion structure
4400 *
4401 * Returns: 0 if there are waiters (wait_for_completion() in progress)
4402 * 1 if there are no waiters.
4403 *
4404 */
4405bool completion_done(struct completion *x)
4406{
7539a3b3 4407 unsigned long flags;
be4de352
DC
4408 int ret = 1;
4409
7539a3b3 4410 spin_lock_irqsave(&x->wait.lock, flags);
be4de352
DC
4411 if (!x->done)
4412 ret = 0;
7539a3b3 4413 spin_unlock_irqrestore(&x->wait.lock, flags);
be4de352
DC
4414 return ret;
4415}
4416EXPORT_SYMBOL(completion_done);
4417
8cbbe86d
AK
4418static long __sched
4419sleep_on_common(wait_queue_head_t *q, int state, long timeout)
1da177e4 4420{
0fec171c
IM
4421 unsigned long flags;
4422 wait_queue_t wait;
4423
4424 init_waitqueue_entry(&wait, current);
1da177e4 4425
8cbbe86d 4426 __set_current_state(state);
1da177e4 4427
8cbbe86d
AK
4428 spin_lock_irqsave(&q->lock, flags);
4429 __add_wait_queue(q, &wait);
4430 spin_unlock(&q->lock);
4431 timeout = schedule_timeout(timeout);
4432 spin_lock_irq(&q->lock);
4433 __remove_wait_queue(q, &wait);
4434 spin_unlock_irqrestore(&q->lock, flags);
4435
4436 return timeout;
4437}
4438
4439void __sched interruptible_sleep_on(wait_queue_head_t *q)
4440{
4441 sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
1da177e4 4442}
1da177e4
LT
4443EXPORT_SYMBOL(interruptible_sleep_on);
4444
0fec171c 4445long __sched
95cdf3b7 4446interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
1da177e4 4447{
8cbbe86d 4448 return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
1da177e4 4449}
1da177e4
LT
4450EXPORT_SYMBOL(interruptible_sleep_on_timeout);
4451
0fec171c 4452void __sched sleep_on(wait_queue_head_t *q)
1da177e4 4453{
8cbbe86d 4454 sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
1da177e4 4455}
1da177e4
LT
4456EXPORT_SYMBOL(sleep_on);
4457
0fec171c 4458long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
1da177e4 4459{
8cbbe86d 4460 return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
1da177e4 4461}
1da177e4
LT
4462EXPORT_SYMBOL(sleep_on_timeout);
4463
b29739f9
IM
4464#ifdef CONFIG_RT_MUTEXES
4465
4466/*
4467 * rt_mutex_setprio - set the current priority of a task
4468 * @p: task
4469 * @prio: prio value (kernel-internal form)
4470 *
4471 * This function changes the 'effective' priority of a task. It does
4472 * not touch ->normal_prio like __setscheduler().
4473 *
4474 * Used by the rt_mutex code to implement priority inheritance logic.
4475 */
36c8b586 4476void rt_mutex_setprio(struct task_struct *p, int prio)
b29739f9
IM
4477{
4478 unsigned long flags;
83b699ed 4479 int oldprio, on_rq, running;
70b97a7f 4480 struct rq *rq;
83ab0aa0 4481 const struct sched_class *prev_class;
b29739f9
IM
4482
4483 BUG_ON(prio < 0 || prio > MAX_PRIO);
4484
4485 rq = task_rq_lock(p, &flags);
4486
a8027073 4487 trace_sched_pi_setprio(p, prio);
d5f9f942 4488 oldprio = p->prio;
83ab0aa0 4489 prev_class = p->sched_class;
dd41f596 4490 on_rq = p->se.on_rq;
051a1d1a 4491 running = task_current(rq, p);
0e1f3483 4492 if (on_rq)
69be72c1 4493 dequeue_task(rq, p, 0);
0e1f3483
HS
4494 if (running)
4495 p->sched_class->put_prev_task(rq, p);
dd41f596
IM
4496
4497 if (rt_prio(prio))
4498 p->sched_class = &rt_sched_class;
4499 else
4500 p->sched_class = &fair_sched_class;
4501
b29739f9
IM
4502 p->prio = prio;
4503
0e1f3483
HS
4504 if (running)
4505 p->sched_class->set_curr_task(rq);
dd41f596 4506 if (on_rq) {
371fd7e7 4507 enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
cb469845
SR
4508
4509 check_class_changed(rq, p, prev_class, oldprio, running);
b29739f9
IM
4510 }
4511 task_rq_unlock(rq, &flags);
4512}
4513
4514#endif
4515
36c8b586 4516void set_user_nice(struct task_struct *p, long nice)
1da177e4 4517{
dd41f596 4518 int old_prio, delta, on_rq;
1da177e4 4519 unsigned long flags;
70b97a7f 4520 struct rq *rq;
1da177e4
LT
4521
4522 if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
4523 return;
4524 /*
4525 * We have to be careful, if called from sys_setpriority(),
4526 * the task might be in the middle of scheduling on another CPU.
4527 */
4528 rq = task_rq_lock(p, &flags);
4529 /*
4530 * The RT priorities are set via sched_setscheduler(), but we still
4531 * allow the 'normal' nice value to be set - but as expected
4532 * it wont have any effect on scheduling until the task is
dd41f596 4533 * SCHED_FIFO/SCHED_RR:
1da177e4 4534 */
e05606d3 4535 if (task_has_rt_policy(p)) {
1da177e4
LT
4536 p->static_prio = NICE_TO_PRIO(nice);
4537 goto out_unlock;
4538 }
dd41f596 4539 on_rq = p->se.on_rq;
c09595f6 4540 if (on_rq)
69be72c1 4541 dequeue_task(rq, p, 0);
1da177e4 4542
1da177e4 4543 p->static_prio = NICE_TO_PRIO(nice);
2dd73a4f 4544 set_load_weight(p);
b29739f9
IM
4545 old_prio = p->prio;
4546 p->prio = effective_prio(p);
4547 delta = p->prio - old_prio;
1da177e4 4548
dd41f596 4549 if (on_rq) {
371fd7e7 4550 enqueue_task(rq, p, 0);
1da177e4 4551 /*
d5f9f942
AM
4552 * If the task increased its priority or is running and
4553 * lowered its priority, then reschedule its CPU:
1da177e4 4554 */
d5f9f942 4555 if (delta < 0 || (delta > 0 && task_running(rq, p)))
1da177e4
LT
4556 resched_task(rq->curr);
4557 }
4558out_unlock:
4559 task_rq_unlock(rq, &flags);
4560}
1da177e4
LT
4561EXPORT_SYMBOL(set_user_nice);
4562
e43379f1
MM
4563/*
4564 * can_nice - check if a task can reduce its nice value
4565 * @p: task
4566 * @nice: nice value
4567 */
36c8b586 4568int can_nice(const struct task_struct *p, const int nice)
e43379f1 4569{
024f4747
MM
4570 /* convert nice value [19,-20] to rlimit style value [1,40] */
4571 int nice_rlim = 20 - nice;
48f24c4d 4572
78d7d407 4573 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
e43379f1
MM
4574 capable(CAP_SYS_NICE));
4575}
4576
1da177e4
LT
4577#ifdef __ARCH_WANT_SYS_NICE
4578
4579/*
4580 * sys_nice - change the priority of the current process.
4581 * @increment: priority increment
4582 *
4583 * sys_setpriority is a more generic, but much slower function that
4584 * does similar things.
4585 */
5add95d4 4586SYSCALL_DEFINE1(nice, int, increment)
1da177e4 4587{
48f24c4d 4588 long nice, retval;
1da177e4
LT
4589
4590 /*
4591 * Setpriority might change our priority at the same moment.
4592 * We don't have to worry. Conceptually one call occurs first
4593 * and we have a single winner.
4594 */
e43379f1
MM
4595 if (increment < -40)
4596 increment = -40;
1da177e4
LT
4597 if (increment > 40)
4598 increment = 40;
4599
2b8f836f 4600 nice = TASK_NICE(current) + increment;
1da177e4
LT
4601 if (nice < -20)
4602 nice = -20;
4603 if (nice > 19)
4604 nice = 19;
4605
e43379f1
MM
4606 if (increment < 0 && !can_nice(current, nice))
4607 return -EPERM;
4608
1da177e4
LT
4609 retval = security_task_setnice(current, nice);
4610 if (retval)
4611 return retval;
4612
4613 set_user_nice(current, nice);
4614 return 0;
4615}
4616
4617#endif
4618
4619/**
4620 * task_prio - return the priority value of a given task.
4621 * @p: the task in question.
4622 *
4623 * This is the priority value as seen by users in /proc.
4624 * RT tasks are offset by -200. Normal tasks are centered
4625 * around 0, value goes from -16 to +15.
4626 */
36c8b586 4627int task_prio(const struct task_struct *p)
1da177e4
LT
4628{
4629 return p->prio - MAX_RT_PRIO;
4630}
4631
4632/**
4633 * task_nice - return the nice value of a given task.
4634 * @p: the task in question.
4635 */
36c8b586 4636int task_nice(const struct task_struct *p)
1da177e4
LT
4637{
4638 return TASK_NICE(p);
4639}
150d8bed 4640EXPORT_SYMBOL(task_nice);
1da177e4
LT
4641
4642/**
4643 * idle_cpu - is a given cpu idle currently?
4644 * @cpu: the processor in question.
4645 */
4646int idle_cpu(int cpu)
4647{
4648 return cpu_curr(cpu) == cpu_rq(cpu)->idle;
4649}
4650
1da177e4
LT
4651/**
4652 * idle_task - return the idle task for a given cpu.
4653 * @cpu: the processor in question.
4654 */
36c8b586 4655struct task_struct *idle_task(int cpu)
1da177e4
LT
4656{
4657 return cpu_rq(cpu)->idle;
4658}
4659
4660/**
4661 * find_process_by_pid - find a process with a matching PID value.
4662 * @pid: the pid in question.
4663 */
a9957449 4664static struct task_struct *find_process_by_pid(pid_t pid)
1da177e4 4665{
228ebcbe 4666 return pid ? find_task_by_vpid(pid) : current;
1da177e4
LT
4667}
4668
4669/* Actually do priority change: must hold rq lock. */
dd41f596
IM
4670static void
4671__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
1da177e4 4672{
dd41f596 4673 BUG_ON(p->se.on_rq);
48f24c4d 4674
1da177e4
LT
4675 p->policy = policy;
4676 p->rt_priority = prio;
b29739f9
IM
4677 p->normal_prio = normal_prio(p);
4678 /* we are holding p->pi_lock already */
4679 p->prio = rt_mutex_getprio(p);
ffd44db5
PZ
4680 if (rt_prio(p->prio))
4681 p->sched_class = &rt_sched_class;
4682 else
4683 p->sched_class = &fair_sched_class;
2dd73a4f 4684 set_load_weight(p);
1da177e4
LT
4685}
4686
c69e8d9c
DH
4687/*
4688 * check the target process has a UID that matches the current process's
4689 */
4690static bool check_same_owner(struct task_struct *p)
4691{
4692 const struct cred *cred = current_cred(), *pcred;
4693 bool match;
4694
4695 rcu_read_lock();
4696 pcred = __task_cred(p);
4697 match = (cred->euid == pcred->euid ||
4698 cred->euid == pcred->uid);
4699 rcu_read_unlock();
4700 return match;
4701}
4702
961ccddd
RR
4703static int __sched_setscheduler(struct task_struct *p, int policy,
4704 struct sched_param *param, bool user)
1da177e4 4705{
83b699ed 4706 int retval, oldprio, oldpolicy = -1, on_rq, running;
1da177e4 4707 unsigned long flags;
83ab0aa0 4708 const struct sched_class *prev_class;
70b97a7f 4709 struct rq *rq;
ca94c442 4710 int reset_on_fork;
1da177e4 4711
66e5393a
SR
4712 /* may grab non-irq protected spin_locks */
4713 BUG_ON(in_interrupt());
1da177e4
LT
4714recheck:
4715 /* double check policy once rq lock held */
ca94c442
LP
4716 if (policy < 0) {
4717 reset_on_fork = p->sched_reset_on_fork;
1da177e4 4718 policy = oldpolicy = p->policy;
ca94c442
LP
4719 } else {
4720 reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
4721 policy &= ~SCHED_RESET_ON_FORK;
4722
4723 if (policy != SCHED_FIFO && policy != SCHED_RR &&
4724 policy != SCHED_NORMAL && policy != SCHED_BATCH &&
4725 policy != SCHED_IDLE)
4726 return -EINVAL;
4727 }
4728
1da177e4
LT
4729 /*
4730 * Valid priorities for SCHED_FIFO and SCHED_RR are
dd41f596
IM
4731 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
4732 * SCHED_BATCH and SCHED_IDLE is 0.
1da177e4
LT
4733 */
4734 if (param->sched_priority < 0 ||
95cdf3b7 4735 (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
d46523ea 4736 (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
1da177e4 4737 return -EINVAL;
e05606d3 4738 if (rt_policy(policy) != (param->sched_priority != 0))
1da177e4
LT
4739 return -EINVAL;
4740
37e4ab3f
OC
4741 /*
4742 * Allow unprivileged RT tasks to decrease priority:
4743 */
961ccddd 4744 if (user && !capable(CAP_SYS_NICE)) {
e05606d3 4745 if (rt_policy(policy)) {
a44702e8
ON
4746 unsigned long rlim_rtprio =
4747 task_rlimit(p, RLIMIT_RTPRIO);
8dc3e909
ON
4748
4749 /* can't set/change the rt policy */
4750 if (policy != p->policy && !rlim_rtprio)
4751 return -EPERM;
4752
4753 /* can't increase priority */
4754 if (param->sched_priority > p->rt_priority &&
4755 param->sched_priority > rlim_rtprio)
4756 return -EPERM;
4757 }
dd41f596
IM
4758 /*
4759 * Like positive nice levels, dont allow tasks to
4760 * move out of SCHED_IDLE either:
4761 */
4762 if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
4763 return -EPERM;
5fe1d75f 4764
37e4ab3f 4765 /* can't change other user's priorities */
c69e8d9c 4766 if (!check_same_owner(p))
37e4ab3f 4767 return -EPERM;
ca94c442
LP
4768
4769 /* Normal users shall not reset the sched_reset_on_fork flag */
4770 if (p->sched_reset_on_fork && !reset_on_fork)
4771 return -EPERM;
37e4ab3f 4772 }
1da177e4 4773
725aad24 4774 if (user) {
b0ae1981 4775 retval = security_task_setscheduler(p);
725aad24
JF
4776 if (retval)
4777 return retval;
4778 }
4779
b29739f9
IM
4780 /*
4781 * make sure no PI-waiters arrive (or leave) while we are
4782 * changing the priority of the task:
4783 */
1d615482 4784 raw_spin_lock_irqsave(&p->pi_lock, flags);
1da177e4
LT
4785 /*
4786 * To be able to change p->policy safely, the apropriate
4787 * runqueue lock must be held.
4788 */
b29739f9 4789 rq = __task_rq_lock(p);
dc61b1d6 4790
34f971f6
PZ
4791 /*
4792 * Changing the policy of the stop threads its a very bad idea
4793 */
4794 if (p == rq->stop) {
4795 __task_rq_unlock(rq);
4796 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4797 return -EINVAL;
4798 }
4799
dc61b1d6
PZ
4800#ifdef CONFIG_RT_GROUP_SCHED
4801 if (user) {
4802 /*
4803 * Do not allow realtime tasks into groups that have no runtime
4804 * assigned.
4805 */
4806 if (rt_bandwidth_enabled() && rt_policy(policy) &&
4807 task_group(p)->rt_bandwidth.rt_runtime == 0) {
4808 __task_rq_unlock(rq);
4809 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4810 return -EPERM;
4811 }
4812 }
4813#endif
4814
1da177e4
LT
4815 /* recheck policy now with rq lock held */
4816 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
4817 policy = oldpolicy = -1;
b29739f9 4818 __task_rq_unlock(rq);
1d615482 4819 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
1da177e4
LT
4820 goto recheck;
4821 }
dd41f596 4822 on_rq = p->se.on_rq;
051a1d1a 4823 running = task_current(rq, p);
0e1f3483 4824 if (on_rq)
2e1cb74a 4825 deactivate_task(rq, p, 0);
0e1f3483
HS
4826 if (running)
4827 p->sched_class->put_prev_task(rq, p);
f6b53205 4828
ca94c442
LP
4829 p->sched_reset_on_fork = reset_on_fork;
4830
1da177e4 4831 oldprio = p->prio;
83ab0aa0 4832 prev_class = p->sched_class;
dd41f596 4833 __setscheduler(rq, p, policy, param->sched_priority);
f6b53205 4834
0e1f3483
HS
4835 if (running)
4836 p->sched_class->set_curr_task(rq);
dd41f596
IM
4837 if (on_rq) {
4838 activate_task(rq, p, 0);
cb469845
SR
4839
4840 check_class_changed(rq, p, prev_class, oldprio, running);
1da177e4 4841 }
b29739f9 4842 __task_rq_unlock(rq);
1d615482 4843 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
b29739f9 4844
95e02ca9
TG
4845 rt_mutex_adjust_pi(p);
4846
1da177e4
LT
4847 return 0;
4848}
961ccddd
RR
4849
4850/**
4851 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
4852 * @p: the task in question.
4853 * @policy: new policy.
4854 * @param: structure containing the new RT priority.
4855 *
4856 * NOTE that the task may be already dead.
4857 */
4858int sched_setscheduler(struct task_struct *p, int policy,
4859 struct sched_param *param)
4860{
4861 return __sched_setscheduler(p, policy, param, true);
4862}
1da177e4
LT
4863EXPORT_SYMBOL_GPL(sched_setscheduler);
4864
961ccddd
RR
4865/**
4866 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
4867 * @p: the task in question.
4868 * @policy: new policy.
4869 * @param: structure containing the new RT priority.
4870 *
4871 * Just like sched_setscheduler, only don't bother checking if the
4872 * current context has permission. For example, this is needed in
4873 * stop_machine(): we create temporary high priority worker threads,
4874 * but our caller might not have that capability.
4875 */
4876int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4877 struct sched_param *param)
4878{
4879 return __sched_setscheduler(p, policy, param, false);
4880}
4881
95cdf3b7
IM
4882static int
4883do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
1da177e4 4884{
1da177e4
LT
4885 struct sched_param lparam;
4886 struct task_struct *p;
36c8b586 4887 int retval;
1da177e4
LT
4888
4889 if (!param || pid < 0)
4890 return -EINVAL;
4891 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
4892 return -EFAULT;
5fe1d75f
ON
4893
4894 rcu_read_lock();
4895 retval = -ESRCH;
1da177e4 4896 p = find_process_by_pid(pid);
5fe1d75f
ON
4897 if (p != NULL)
4898 retval = sched_setscheduler(p, policy, &lparam);
4899 rcu_read_unlock();
36c8b586 4900
1da177e4
LT
4901 return retval;
4902}
4903
4904/**
4905 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
4906 * @pid: the pid in question.
4907 * @policy: new policy.
4908 * @param: structure containing the new RT priority.
4909 */
5add95d4
HC
4910SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
4911 struct sched_param __user *, param)
1da177e4 4912{
c21761f1
JB
4913 /* negative values for policy are not valid */
4914 if (policy < 0)
4915 return -EINVAL;
4916
1da177e4
LT
4917 return do_sched_setscheduler(pid, policy, param);
4918}
4919
4920/**
4921 * sys_sched_setparam - set/change the RT priority of a thread
4922 * @pid: the pid in question.
4923 * @param: structure containing the new RT priority.
4924 */
5add95d4 4925SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
1da177e4
LT
4926{
4927 return do_sched_setscheduler(pid, -1, param);
4928}
4929
4930/**
4931 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
4932 * @pid: the pid in question.
4933 */
5add95d4 4934SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
1da177e4 4935{
36c8b586 4936 struct task_struct *p;
3a5c359a 4937 int retval;
1da177e4
LT
4938
4939 if (pid < 0)
3a5c359a 4940 return -EINVAL;
1da177e4
LT
4941
4942 retval = -ESRCH;
5fe85be0 4943 rcu_read_lock();
1da177e4
LT
4944 p = find_process_by_pid(pid);
4945 if (p) {
4946 retval = security_task_getscheduler(p);
4947 if (!retval)
ca94c442
LP
4948 retval = p->policy
4949 | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
1da177e4 4950 }
5fe85be0 4951 rcu_read_unlock();
1da177e4
LT
4952 return retval;
4953}
4954
4955/**
ca94c442 4956 * sys_sched_getparam - get the RT priority of a thread
1da177e4
LT
4957 * @pid: the pid in question.
4958 * @param: structure containing the RT priority.
4959 */
5add95d4 4960SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
1da177e4
LT
4961{
4962 struct sched_param lp;
36c8b586 4963 struct task_struct *p;
3a5c359a 4964 int retval;
1da177e4
LT
4965
4966 if (!param || pid < 0)
3a5c359a 4967 return -EINVAL;
1da177e4 4968
5fe85be0 4969 rcu_read_lock();
1da177e4
LT
4970 p = find_process_by_pid(pid);
4971 retval = -ESRCH;
4972 if (!p)
4973 goto out_unlock;
4974
4975 retval = security_task_getscheduler(p);
4976 if (retval)
4977 goto out_unlock;
4978
4979 lp.sched_priority = p->rt_priority;
5fe85be0 4980 rcu_read_unlock();
1da177e4
LT
4981
4982 /*
4983 * This one might sleep, we cannot do it with a spinlock held ...
4984 */
4985 retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
4986
1da177e4
LT
4987 return retval;
4988
4989out_unlock:
5fe85be0 4990 rcu_read_unlock();
1da177e4
LT
4991 return retval;
4992}
4993
96f874e2 4994long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
1da177e4 4995{
5a16f3d3 4996 cpumask_var_t cpus_allowed, new_mask;
36c8b586
IM
4997 struct task_struct *p;
4998 int retval;
1da177e4 4999
95402b38 5000 get_online_cpus();
23f5d142 5001 rcu_read_lock();
1da177e4
LT
5002
5003 p = find_process_by_pid(pid);
5004 if (!p) {
23f5d142 5005 rcu_read_unlock();
95402b38 5006 put_online_cpus();
1da177e4
LT
5007 return -ESRCH;
5008 }
5009
23f5d142 5010 /* Prevent p going away */
1da177e4 5011 get_task_struct(p);
23f5d142 5012 rcu_read_unlock();
1da177e4 5013
5a16f3d3
RR
5014 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5015 retval = -ENOMEM;
5016 goto out_put_task;
5017 }
5018 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5019 retval = -ENOMEM;
5020 goto out_free_cpus_allowed;
5021 }
1da177e4 5022 retval = -EPERM;
c69e8d9c 5023 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
1da177e4
LT
5024 goto out_unlock;
5025
b0ae1981 5026 retval = security_task_setscheduler(p);
e7834f8f
DQ
5027 if (retval)
5028 goto out_unlock;
5029
5a16f3d3
RR
5030 cpuset_cpus_allowed(p, cpus_allowed);
5031 cpumask_and(new_mask, in_mask, cpus_allowed);
49246274 5032again:
5a16f3d3 5033 retval = set_cpus_allowed_ptr(p, new_mask);
1da177e4 5034
8707d8b8 5035 if (!retval) {
5a16f3d3
RR
5036 cpuset_cpus_allowed(p, cpus_allowed);
5037 if (!cpumask_subset(new_mask, cpus_allowed)) {
8707d8b8
PM
5038 /*
5039 * We must have raced with a concurrent cpuset
5040 * update. Just reset the cpus_allowed to the
5041 * cpuset's cpus_allowed
5042 */
5a16f3d3 5043 cpumask_copy(new_mask, cpus_allowed);
8707d8b8
PM
5044 goto again;
5045 }
5046 }
1da177e4 5047out_unlock:
5a16f3d3
RR
5048 free_cpumask_var(new_mask);
5049out_free_cpus_allowed:
5050 free_cpumask_var(cpus_allowed);
5051out_put_task:
1da177e4 5052 put_task_struct(p);
95402b38 5053 put_online_cpus();
1da177e4
LT
5054 return retval;
5055}
5056
5057static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
96f874e2 5058 struct cpumask *new_mask)
1da177e4 5059{
96f874e2
RR
5060 if (len < cpumask_size())
5061 cpumask_clear(new_mask);
5062 else if (len > cpumask_size())
5063 len = cpumask_size();
5064
1da177e4
LT
5065 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5066}
5067
5068/**
5069 * sys_sched_setaffinity - set the cpu affinity of a process
5070 * @pid: pid of the process
5071 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5072 * @user_mask_ptr: user-space pointer to the new cpu mask
5073 */
5add95d4
HC
5074SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
5075 unsigned long __user *, user_mask_ptr)
1da177e4 5076{
5a16f3d3 5077 cpumask_var_t new_mask;
1da177e4
LT
5078 int retval;
5079
5a16f3d3
RR
5080 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5081 return -ENOMEM;
1da177e4 5082
5a16f3d3
RR
5083 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5084 if (retval == 0)
5085 retval = sched_setaffinity(pid, new_mask);
5086 free_cpumask_var(new_mask);
5087 return retval;
1da177e4
LT
5088}
5089
96f874e2 5090long sched_getaffinity(pid_t pid, struct cpumask *mask)
1da177e4 5091{
36c8b586 5092 struct task_struct *p;
31605683
TG
5093 unsigned long flags;
5094 struct rq *rq;
1da177e4 5095 int retval;
1da177e4 5096
95402b38 5097 get_online_cpus();
23f5d142 5098 rcu_read_lock();
1da177e4
LT
5099
5100 retval = -ESRCH;
5101 p = find_process_by_pid(pid);
5102 if (!p)
5103 goto out_unlock;
5104
e7834f8f
DQ
5105 retval = security_task_getscheduler(p);
5106 if (retval)
5107 goto out_unlock;
5108
31605683 5109 rq = task_rq_lock(p, &flags);
96f874e2 5110 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
31605683 5111 task_rq_unlock(rq, &flags);
1da177e4
LT
5112
5113out_unlock:
23f5d142 5114 rcu_read_unlock();
95402b38 5115 put_online_cpus();
1da177e4 5116
9531b62f 5117 return retval;
1da177e4
LT
5118}
5119
5120/**
5121 * sys_sched_getaffinity - get the cpu affinity of a process
5122 * @pid: pid of the process
5123 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
5124 * @user_mask_ptr: user-space pointer to hold the current cpu mask
5125 */
5add95d4
HC
5126SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
5127 unsigned long __user *, user_mask_ptr)
1da177e4
LT
5128{
5129 int ret;
f17c8607 5130 cpumask_var_t mask;
1da177e4 5131
84fba5ec 5132 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
cd3d8031
KM
5133 return -EINVAL;
5134 if (len & (sizeof(unsigned long)-1))
1da177e4
LT
5135 return -EINVAL;
5136
f17c8607
RR
5137 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5138 return -ENOMEM;
1da177e4 5139
f17c8607
RR
5140 ret = sched_getaffinity(pid, mask);
5141 if (ret == 0) {
8bc037fb 5142 size_t retlen = min_t(size_t, len, cpumask_size());
cd3d8031
KM
5143
5144 if (copy_to_user(user_mask_ptr, mask, retlen))
f17c8607
RR
5145 ret = -EFAULT;
5146 else
cd3d8031 5147 ret = retlen;
f17c8607
RR
5148 }
5149 free_cpumask_var(mask);
1da177e4 5150
f17c8607 5151 return ret;
1da177e4
LT
5152}
5153
5154/**
5155 * sys_sched_yield - yield the current processor to other threads.
5156 *
dd41f596
IM
5157 * This function yields the current CPU to other tasks. If there are no
5158 * other threads running on this CPU then this function will return.
1da177e4 5159 */
5add95d4 5160SYSCALL_DEFINE0(sched_yield)
1da177e4 5161{
70b97a7f 5162 struct rq *rq = this_rq_lock();
1da177e4 5163
2d72376b 5164 schedstat_inc(rq, yld_count);
4530d7ab 5165 current->sched_class->yield_task(rq);
1da177e4
LT
5166
5167 /*
5168 * Since we are going to call schedule() anyway, there's
5169 * no need to preempt or enable interrupts:
5170 */
5171 __release(rq->lock);
8a25d5de 5172 spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
9828ea9d 5173 do_raw_spin_unlock(&rq->lock);
1da177e4
LT
5174 preempt_enable_no_resched();
5175
5176 schedule();
5177
5178 return 0;
5179}
5180
d86ee480
PZ
5181static inline int should_resched(void)
5182{
5183 return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
5184}
5185
e7b38404 5186static void __cond_resched(void)
1da177e4 5187{
e7aaaa69
FW
5188 add_preempt_count(PREEMPT_ACTIVE);
5189 schedule();
5190 sub_preempt_count(PREEMPT_ACTIVE);
1da177e4
LT
5191}
5192
02b67cc3 5193int __sched _cond_resched(void)
1da177e4 5194{
d86ee480 5195 if (should_resched()) {
1da177e4
LT
5196 __cond_resched();
5197 return 1;
5198 }
5199 return 0;
5200}
02b67cc3 5201EXPORT_SYMBOL(_cond_resched);
1da177e4
LT
5202
5203/*
613afbf8 5204 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
1da177e4
LT
5205 * call schedule, and on return reacquire the lock.
5206 *
41a2d6cf 5207 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
1da177e4
LT
5208 * operations here to prevent schedule() from being called twice (once via
5209 * spin_unlock(), once by hand).
5210 */
613afbf8 5211int __cond_resched_lock(spinlock_t *lock)
1da177e4 5212{
d86ee480 5213 int resched = should_resched();
6df3cecb
JK
5214 int ret = 0;
5215
f607c668
PZ
5216 lockdep_assert_held(lock);
5217
95c354fe 5218 if (spin_needbreak(lock) || resched) {
1da177e4 5219 spin_unlock(lock);
d86ee480 5220 if (resched)
95c354fe
NP
5221 __cond_resched();
5222 else
5223 cpu_relax();
6df3cecb 5224 ret = 1;
1da177e4 5225 spin_lock(lock);
1da177e4 5226 }
6df3cecb 5227 return ret;
1da177e4 5228}
613afbf8 5229EXPORT_SYMBOL(__cond_resched_lock);
1da177e4 5230
613afbf8 5231int __sched __cond_resched_softirq(void)
1da177e4
LT
5232{
5233 BUG_ON(!in_softirq());
5234
d86ee480 5235 if (should_resched()) {
98d82567 5236 local_bh_enable();
1da177e4
LT
5237 __cond_resched();
5238 local_bh_disable();
5239 return 1;
5240 }
5241 return 0;
5242}
613afbf8 5243EXPORT_SYMBOL(__cond_resched_softirq);
1da177e4 5244
1da177e4
LT
5245/**
5246 * yield - yield the current processor to other threads.
5247 *
72fd4a35 5248 * This is a shortcut for kernel-space yielding - it marks the
1da177e4
LT
5249 * thread runnable and calls sys_sched_yield().
5250 */
5251void __sched yield(void)
5252{
5253 set_current_state(TASK_RUNNING);
5254 sys_sched_yield();
5255}
1da177e4
LT
5256EXPORT_SYMBOL(yield);
5257
5258/*
41a2d6cf 5259 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
1da177e4 5260 * that process accounting knows that this is a task in IO wait state.
1da177e4
LT
5261 */
5262void __sched io_schedule(void)
5263{
54d35f29 5264 struct rq *rq = raw_rq();
1da177e4 5265
0ff92245 5266 delayacct_blkio_start();
1da177e4 5267 atomic_inc(&rq->nr_iowait);
8f0dfc34 5268 current->in_iowait = 1;
1da177e4 5269 schedule();
8f0dfc34 5270 current->in_iowait = 0;
1da177e4 5271 atomic_dec(&rq->nr_iowait);
0ff92245 5272 delayacct_blkio_end();
1da177e4 5273}
1da177e4
LT
5274EXPORT_SYMBOL(io_schedule);
5275
5276long __sched io_schedule_timeout(long timeout)
5277{
54d35f29 5278 struct rq *rq = raw_rq();
1da177e4
LT
5279 long ret;
5280
0ff92245 5281 delayacct_blkio_start();
1da177e4 5282 atomic_inc(&rq->nr_iowait);
8f0dfc34 5283 current->in_iowait = 1;
1da177e4 5284 ret = schedule_timeout(timeout);
8f0dfc34 5285 current->in_iowait = 0;
1da177e4 5286 atomic_dec(&rq->nr_iowait);
0ff92245 5287 delayacct_blkio_end();
1da177e4
LT
5288 return ret;
5289}
5290
5291/**
5292 * sys_sched_get_priority_max - return maximum RT priority.
5293 * @policy: scheduling class.
5294 *
5295 * this syscall returns the maximum rt_priority that can be used
5296 * by a given scheduling class.
5297 */
5add95d4 5298SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
1da177e4
LT
5299{
5300 int ret = -EINVAL;
5301
5302 switch (policy) {
5303 case SCHED_FIFO:
5304 case SCHED_RR:
5305 ret = MAX_USER_RT_PRIO-1;
5306 break;
5307 case SCHED_NORMAL:
b0a9499c 5308 case SCHED_BATCH:
dd41f596 5309 case SCHED_IDLE:
1da177e4
LT
5310 ret = 0;
5311 break;
5312 }
5313 return ret;
5314}
5315
5316/**
5317 * sys_sched_get_priority_min - return minimum RT priority.
5318 * @policy: scheduling class.
5319 *
5320 * this syscall returns the minimum rt_priority that can be used
5321 * by a given scheduling class.
5322 */
5add95d4 5323SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
1da177e4
LT
5324{
5325 int ret = -EINVAL;
5326
5327 switch (policy) {
5328 case SCHED_FIFO:
5329 case SCHED_RR:
5330 ret = 1;
5331 break;
5332 case SCHED_NORMAL:
b0a9499c 5333 case SCHED_BATCH:
dd41f596 5334 case SCHED_IDLE:
1da177e4
LT
5335 ret = 0;
5336 }
5337 return ret;
5338}
5339
5340/**
5341 * sys_sched_rr_get_interval - return the default timeslice of a process.
5342 * @pid: pid of the process.
5343 * @interval: userspace pointer to the timeslice value.
5344 *
5345 * this syscall writes the default timeslice value of a given process
5346 * into the user-space timespec buffer. A value of '0' means infinity.
5347 */
17da2bd9 5348SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
754fe8d2 5349 struct timespec __user *, interval)
1da177e4 5350{
36c8b586 5351 struct task_struct *p;
a4ec24b4 5352 unsigned int time_slice;
dba091b9
TG
5353 unsigned long flags;
5354 struct rq *rq;
3a5c359a 5355 int retval;
1da177e4 5356 struct timespec t;
1da177e4
LT
5357
5358 if (pid < 0)
3a5c359a 5359 return -EINVAL;
1da177e4
LT
5360
5361 retval = -ESRCH;
1a551ae7 5362 rcu_read_lock();
1da177e4
LT
5363 p = find_process_by_pid(pid);
5364 if (!p)
5365 goto out_unlock;
5366
5367 retval = security_task_getscheduler(p);
5368 if (retval)
5369 goto out_unlock;
5370
dba091b9
TG
5371 rq = task_rq_lock(p, &flags);
5372 time_slice = p->sched_class->get_rr_interval(rq, p);
5373 task_rq_unlock(rq, &flags);
a4ec24b4 5374
1a551ae7 5375 rcu_read_unlock();
a4ec24b4 5376 jiffies_to_timespec(time_slice, &t);
1da177e4 5377 retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
1da177e4 5378 return retval;
3a5c359a 5379
1da177e4 5380out_unlock:
1a551ae7 5381 rcu_read_unlock();
1da177e4
LT
5382 return retval;
5383}
5384
7c731e0a 5385static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
36c8b586 5386
82a1fcb9 5387void sched_show_task(struct task_struct *p)
1da177e4 5388{
1da177e4 5389 unsigned long free = 0;
36c8b586 5390 unsigned state;
1da177e4 5391
1da177e4 5392 state = p->state ? __ffs(p->state) + 1 : 0;
3df0fc5b 5393 printk(KERN_INFO "%-13.13s %c", p->comm,
2ed6e34f 5394 state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
4bd77321 5395#if BITS_PER_LONG == 32
1da177e4 5396 if (state == TASK_RUNNING)
3df0fc5b 5397 printk(KERN_CONT " running ");
1da177e4 5398 else
3df0fc5b 5399 printk(KERN_CONT " %08lx ", thread_saved_pc(p));
1da177e4
LT
5400#else
5401 if (state == TASK_RUNNING)
3df0fc5b 5402 printk(KERN_CONT " running task ");
1da177e4 5403 else
3df0fc5b 5404 printk(KERN_CONT " %016lx ", thread_saved_pc(p));
1da177e4
LT
5405#endif
5406#ifdef CONFIG_DEBUG_STACK_USAGE
7c9f8861 5407 free = stack_not_used(p);
1da177e4 5408#endif
3df0fc5b 5409 printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
aa47b7e0
DR
5410 task_pid_nr(p), task_pid_nr(p->real_parent),
5411 (unsigned long)task_thread_info(p)->flags);
1da177e4 5412
5fb5e6de 5413 show_stack(p, NULL);
1da177e4
LT
5414}
5415
e59e2ae2 5416void show_state_filter(unsigned long state_filter)
1da177e4 5417{
36c8b586 5418 struct task_struct *g, *p;
1da177e4 5419
4bd77321 5420#if BITS_PER_LONG == 32
3df0fc5b
PZ
5421 printk(KERN_INFO
5422 " task PC stack pid father\n");
1da177e4 5423#else
3df0fc5b
PZ
5424 printk(KERN_INFO
5425 " task PC stack pid father\n");
1da177e4
LT
5426#endif
5427 read_lock(&tasklist_lock);
5428 do_each_thread(g, p) {
5429 /*
5430 * reset the NMI-timeout, listing all files on a slow
5431 * console might take alot of time:
5432 */
5433 touch_nmi_watchdog();
39bc89fd 5434 if (!state_filter || (p->state & state_filter))
82a1fcb9 5435 sched_show_task(p);
1da177e4
LT
5436 } while_each_thread(g, p);
5437
04c9167f
JF
5438 touch_all_softlockup_watchdogs();
5439
dd41f596
IM
5440#ifdef CONFIG_SCHED_DEBUG
5441 sysrq_sched_debug_show();
5442#endif
1da177e4 5443 read_unlock(&tasklist_lock);
e59e2ae2
IM
5444 /*
5445 * Only show locks if all tasks are dumped:
5446 */
93335a21 5447 if (!state_filter)
e59e2ae2 5448 debug_show_all_locks();
1da177e4
LT
5449}
5450
1df21055
IM
5451void __cpuinit init_idle_bootup_task(struct task_struct *idle)
5452{
dd41f596 5453 idle->sched_class = &idle_sched_class;
1df21055
IM
5454}
5455
f340c0d1
IM
5456/**
5457 * init_idle - set up an idle thread for a given CPU
5458 * @idle: task in question
5459 * @cpu: cpu the idle task belongs to
5460 *
5461 * NOTE: this function does not set the idle thread's NEED_RESCHED
5462 * flag, to make booting more robust.
5463 */
5c1e1767 5464void __cpuinit init_idle(struct task_struct *idle, int cpu)
1da177e4 5465{
70b97a7f 5466 struct rq *rq = cpu_rq(cpu);
1da177e4
LT
5467 unsigned long flags;
5468
05fa785c 5469 raw_spin_lock_irqsave(&rq->lock, flags);
5cbd54ef 5470
dd41f596 5471 __sched_fork(idle);
06b83b5f 5472 idle->state = TASK_RUNNING;
dd41f596
IM
5473 idle->se.exec_start = sched_clock();
5474
96f874e2 5475 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
6506cf6c
PZ
5476 /*
5477 * We're having a chicken and egg problem, even though we are
5478 * holding rq->lock, the cpu isn't yet set to this cpu so the
5479 * lockdep check in task_group() will fail.
5480 *
5481 * Similar case to sched_fork(). / Alternatively we could
5482 * use task_rq_lock() here and obtain the other rq->lock.
5483 *
5484 * Silence PROVE_RCU
5485 */
5486 rcu_read_lock();
dd41f596 5487 __set_task_cpu(idle, cpu);
6506cf6c 5488 rcu_read_unlock();
1da177e4 5489
1da177e4 5490 rq->curr = rq->idle = idle;
4866cde0
NP
5491#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
5492 idle->oncpu = 1;
5493#endif
05fa785c 5494 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4
LT
5495
5496 /* Set the preempt count _outside_ the spinlocks! */
8e3e076c
LT
5497#if defined(CONFIG_PREEMPT)
5498 task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
5499#else
a1261f54 5500 task_thread_info(idle)->preempt_count = 0;
8e3e076c 5501#endif
dd41f596
IM
5502 /*
5503 * The idle tasks have their own, simple scheduling class:
5504 */
5505 idle->sched_class = &idle_sched_class;
fb52607a 5506 ftrace_graph_init_task(idle);
1da177e4
LT
5507}
5508
5509/*
5510 * In a system that switches off the HZ timer nohz_cpu_mask
5511 * indicates which cpus entered this state. This is used
5512 * in the rcu update to wait only for active cpus. For system
5513 * which do not switch off the HZ timer nohz_cpu_mask should
6a7b3dc3 5514 * always be CPU_BITS_NONE.
1da177e4 5515 */
6a7b3dc3 5516cpumask_var_t nohz_cpu_mask;
1da177e4 5517
19978ca6
IM
5518/*
5519 * Increase the granularity value when there are more CPUs,
5520 * because with more CPUs the 'effective latency' as visible
5521 * to users decreases. But the relationship is not linear,
5522 * so pick a second-best guess by going with the log2 of the
5523 * number of CPUs.
5524 *
5525 * This idea comes from the SD scheduler of Con Kolivas:
5526 */
acb4a848 5527static int get_update_sysctl_factor(void)
19978ca6 5528{
4ca3ef71 5529 unsigned int cpus = min_t(int, num_online_cpus(), 8);
1983a922
CE
5530 unsigned int factor;
5531
5532 switch (sysctl_sched_tunable_scaling) {
5533 case SCHED_TUNABLESCALING_NONE:
5534 factor = 1;
5535 break;
5536 case SCHED_TUNABLESCALING_LINEAR:
5537 factor = cpus;
5538 break;
5539 case SCHED_TUNABLESCALING_LOG:
5540 default:
5541 factor = 1 + ilog2(cpus);
5542 break;
5543 }
19978ca6 5544
acb4a848
CE
5545 return factor;
5546}
19978ca6 5547
acb4a848
CE
5548static void update_sysctl(void)
5549{
5550 unsigned int factor = get_update_sysctl_factor();
19978ca6 5551
0bcdcf28
CE
5552#define SET_SYSCTL(name) \
5553 (sysctl_##name = (factor) * normalized_sysctl_##name)
5554 SET_SYSCTL(sched_min_granularity);
5555 SET_SYSCTL(sched_latency);
5556 SET_SYSCTL(sched_wakeup_granularity);
5557 SET_SYSCTL(sched_shares_ratelimit);
5558#undef SET_SYSCTL
5559}
55cd5340 5560
0bcdcf28
CE
5561static inline void sched_init_granularity(void)
5562{
5563 update_sysctl();
19978ca6
IM
5564}
5565
1da177e4
LT
5566#ifdef CONFIG_SMP
5567/*
5568 * This is how migration works:
5569 *
969c7921
TH
5570 * 1) we invoke migration_cpu_stop() on the target CPU using
5571 * stop_one_cpu().
5572 * 2) stopper starts to run (implicitly forcing the migrated thread
5573 * off the CPU)
5574 * 3) it checks whether the migrated task is still in the wrong runqueue.
5575 * 4) if it's in the wrong runqueue then the migration thread removes
1da177e4 5576 * it and puts it into the right queue.
969c7921
TH
5577 * 5) stopper completes and stop_one_cpu() returns and the migration
5578 * is done.
1da177e4
LT
5579 */
5580
5581/*
5582 * Change a given task's CPU affinity. Migrate the thread to a
5583 * proper CPU and schedule it away if the CPU it's executing on
5584 * is removed from the allowed bitmask.
5585 *
5586 * NOTE: the caller must have a valid reference to the task, the
41a2d6cf 5587 * task must not exit() & deallocate itself prematurely. The
1da177e4
LT
5588 * call is not atomic; no spinlocks may be held.
5589 */
96f874e2 5590int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1da177e4
LT
5591{
5592 unsigned long flags;
70b97a7f 5593 struct rq *rq;
969c7921 5594 unsigned int dest_cpu;
48f24c4d 5595 int ret = 0;
1da177e4 5596
65cc8e48
PZ
5597 /*
5598 * Serialize against TASK_WAKING so that ttwu() and wunt() can
5599 * drop the rq->lock and still rely on ->cpus_allowed.
5600 */
5601again:
5602 while (task_is_waking(p))
5603 cpu_relax();
1da177e4 5604 rq = task_rq_lock(p, &flags);
65cc8e48
PZ
5605 if (task_is_waking(p)) {
5606 task_rq_unlock(rq, &flags);
5607 goto again;
5608 }
e2912009 5609
6ad4c188 5610 if (!cpumask_intersects(new_mask, cpu_active_mask)) {
1da177e4
LT
5611 ret = -EINVAL;
5612 goto out;
5613 }
5614
9985b0ba 5615 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
96f874e2 5616 !cpumask_equal(&p->cpus_allowed, new_mask))) {
9985b0ba
DR
5617 ret = -EINVAL;
5618 goto out;
5619 }
5620
73fe6aae 5621 if (p->sched_class->set_cpus_allowed)
cd8ba7cd 5622 p->sched_class->set_cpus_allowed(p, new_mask);
73fe6aae 5623 else {
96f874e2
RR
5624 cpumask_copy(&p->cpus_allowed, new_mask);
5625 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
73fe6aae
GH
5626 }
5627
1da177e4 5628 /* Can the task run on the task's current CPU? If so, we're done */
96f874e2 5629 if (cpumask_test_cpu(task_cpu(p), new_mask))
1da177e4
LT
5630 goto out;
5631
969c7921
TH
5632 dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
5633 if (migrate_task(p, dest_cpu)) {
5634 struct migration_arg arg = { p, dest_cpu };
1da177e4
LT
5635 /* Need help from migration thread: drop lock and wait. */
5636 task_rq_unlock(rq, &flags);
969c7921 5637 stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
1da177e4
LT
5638 tlb_migrate_finish(p->mm);
5639 return 0;
5640 }
5641out:
5642 task_rq_unlock(rq, &flags);
48f24c4d 5643
1da177e4
LT
5644 return ret;
5645}
cd8ba7cd 5646EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
1da177e4
LT
5647
5648/*
41a2d6cf 5649 * Move (not current) task off this cpu, onto dest cpu. We're doing
1da177e4
LT
5650 * this because either it can't run here any more (set_cpus_allowed()
5651 * away from this CPU, or CPU going down), or because we're
5652 * attempting to rebalance this task on exec (sched_exec).
5653 *
5654 * So we race with normal scheduler movements, but that's OK, as long
5655 * as the task is no longer on this CPU.
efc30814
KK
5656 *
5657 * Returns non-zero if task was successfully migrated.
1da177e4 5658 */
efc30814 5659static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
1da177e4 5660{
70b97a7f 5661 struct rq *rq_dest, *rq_src;
e2912009 5662 int ret = 0;
1da177e4 5663
e761b772 5664 if (unlikely(!cpu_active(dest_cpu)))
efc30814 5665 return ret;
1da177e4
LT
5666
5667 rq_src = cpu_rq(src_cpu);
5668 rq_dest = cpu_rq(dest_cpu);
5669
5670 double_rq_lock(rq_src, rq_dest);
5671 /* Already moved. */
5672 if (task_cpu(p) != src_cpu)
b1e38734 5673 goto done;
1da177e4 5674 /* Affinity changed (again). */
96f874e2 5675 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
b1e38734 5676 goto fail;
1da177e4 5677
e2912009
PZ
5678 /*
5679 * If we're not on a rq, the next wake-up will ensure we're
5680 * placed properly.
5681 */
5682 if (p->se.on_rq) {
2e1cb74a 5683 deactivate_task(rq_src, p, 0);
e2912009 5684 set_task_cpu(p, dest_cpu);
dd41f596 5685 activate_task(rq_dest, p, 0);
15afe09b 5686 check_preempt_curr(rq_dest, p, 0);
1da177e4 5687 }
b1e38734 5688done:
efc30814 5689 ret = 1;
b1e38734 5690fail:
1da177e4 5691 double_rq_unlock(rq_src, rq_dest);
efc30814 5692 return ret;
1da177e4
LT
5693}
5694
5695/*
969c7921
TH
5696 * migration_cpu_stop - this will be executed by a highprio stopper thread
5697 * and performs thread migration by bumping thread off CPU then
5698 * 'pushing' onto another runqueue.
1da177e4 5699 */
969c7921 5700static int migration_cpu_stop(void *data)
1da177e4 5701{
969c7921 5702 struct migration_arg *arg = data;
f7b4cddc 5703
969c7921
TH
5704 /*
5705 * The original target cpu might have gone down and we might
5706 * be on another cpu but it doesn't matter.
5707 */
f7b4cddc 5708 local_irq_disable();
969c7921 5709 __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
f7b4cddc 5710 local_irq_enable();
1da177e4 5711 return 0;
f7b4cddc
ON
5712}
5713
1da177e4 5714#ifdef CONFIG_HOTPLUG_CPU
054b9108 5715/*
3a4fa0a2 5716 * Figure out where task on dead CPU should go, use force if necessary.
054b9108 5717 */
6a1bdc1b 5718void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
1da177e4 5719{
1445c08d
ON
5720 struct rq *rq = cpu_rq(dead_cpu);
5721 int needs_cpu, uninitialized_var(dest_cpu);
5722 unsigned long flags;
e76bd8d9 5723
1445c08d 5724 local_irq_save(flags);
e76bd8d9 5725
1445c08d
ON
5726 raw_spin_lock(&rq->lock);
5727 needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING);
5728 if (needs_cpu)
5729 dest_cpu = select_fallback_rq(dead_cpu, p);
5730 raw_spin_unlock(&rq->lock);
c1804d54
ON
5731 /*
5732 * It can only fail if we race with set_cpus_allowed(),
5733 * in the racer should migrate the task anyway.
5734 */
1445c08d 5735 if (needs_cpu)
c1804d54 5736 __migrate_task(p, dead_cpu, dest_cpu);
1445c08d 5737 local_irq_restore(flags);
1da177e4
LT
5738}
5739
5740/*
5741 * While a dead CPU has no uninterruptible tasks queued at this point,
5742 * it might still have a nonzero ->nr_uninterruptible counter, because
5743 * for performance reasons the counter is not stricly tracking tasks to
5744 * their home CPUs. So we just add the counter to another CPU's counter,
5745 * to keep the global sum constant after CPU-down:
5746 */
70b97a7f 5747static void migrate_nr_uninterruptible(struct rq *rq_src)
1da177e4 5748{
6ad4c188 5749 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask));
1da177e4
LT
5750 unsigned long flags;
5751
5752 local_irq_save(flags);
5753 double_rq_lock(rq_src, rq_dest);
5754 rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
5755 rq_src->nr_uninterruptible = 0;
5756 double_rq_unlock(rq_src, rq_dest);
5757 local_irq_restore(flags);
5758}
5759
5760/* Run through task list and migrate tasks from the dead cpu. */
5761static void migrate_live_tasks(int src_cpu)
5762{
48f24c4d 5763 struct task_struct *p, *t;
1da177e4 5764
f7b4cddc 5765 read_lock(&tasklist_lock);
1da177e4 5766
48f24c4d
IM
5767 do_each_thread(t, p) {
5768 if (p == current)
1da177e4
LT
5769 continue;
5770
48f24c4d
IM
5771 if (task_cpu(p) == src_cpu)
5772 move_task_off_dead_cpu(src_cpu, p);
5773 } while_each_thread(t, p);
1da177e4 5774
f7b4cddc 5775 read_unlock(&tasklist_lock);
1da177e4
LT
5776}
5777
dd41f596
IM
5778/*
5779 * Schedules idle task to be the next runnable task on current CPU.
94bc9a7b
DA
5780 * It does so by boosting its priority to highest possible.
5781 * Used by CPU offline code.
1da177e4
LT
5782 */
5783void sched_idle_next(void)
5784{
48f24c4d 5785 int this_cpu = smp_processor_id();
70b97a7f 5786 struct rq *rq = cpu_rq(this_cpu);
1da177e4
LT
5787 struct task_struct *p = rq->idle;
5788 unsigned long flags;
5789
5790 /* cpu has to be offline */
48f24c4d 5791 BUG_ON(cpu_online(this_cpu));
1da177e4 5792
48f24c4d
IM
5793 /*
5794 * Strictly not necessary since rest of the CPUs are stopped by now
5795 * and interrupts disabled on the current cpu.
1da177e4 5796 */
05fa785c 5797 raw_spin_lock_irqsave(&rq->lock, flags);
1da177e4 5798
dd41f596 5799 __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
48f24c4d 5800
94bc9a7b 5801 activate_task(rq, p, 0);
1da177e4 5802
05fa785c 5803 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4
LT
5804}
5805
48f24c4d
IM
5806/*
5807 * Ensures that the idle task is using init_mm right before its cpu goes
1da177e4
LT
5808 * offline.
5809 */
5810void idle_task_exit(void)
5811{
5812 struct mm_struct *mm = current->active_mm;
5813
5814 BUG_ON(cpu_online(smp_processor_id()));
5815
5816 if (mm != &init_mm)
5817 switch_mm(mm, &init_mm, current);
5818 mmdrop(mm);
5819}
5820
054b9108 5821/* called under rq->lock with disabled interrupts */
36c8b586 5822static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
1da177e4 5823{
70b97a7f 5824 struct rq *rq = cpu_rq(dead_cpu);
1da177e4
LT
5825
5826 /* Must be exiting, otherwise would be on tasklist. */
270f722d 5827 BUG_ON(!p->exit_state);
1da177e4
LT
5828
5829 /* Cannot have done final schedule yet: would have vanished. */
c394cc9f 5830 BUG_ON(p->state == TASK_DEAD);
1da177e4 5831
48f24c4d 5832 get_task_struct(p);
1da177e4
LT
5833
5834 /*
5835 * Drop lock around migration; if someone else moves it,
41a2d6cf 5836 * that's OK. No task can be added to this CPU, so iteration is
1da177e4
LT
5837 * fine.
5838 */
05fa785c 5839 raw_spin_unlock_irq(&rq->lock);
48f24c4d 5840 move_task_off_dead_cpu(dead_cpu, p);
05fa785c 5841 raw_spin_lock_irq(&rq->lock);
1da177e4 5842
48f24c4d 5843 put_task_struct(p);
1da177e4
LT
5844}
5845
5846/* release_task() removes task from tasklist, so we won't find dead tasks. */
5847static void migrate_dead_tasks(unsigned int dead_cpu)
5848{
70b97a7f 5849 struct rq *rq = cpu_rq(dead_cpu);
dd41f596 5850 struct task_struct *next;
48f24c4d 5851
dd41f596
IM
5852 for ( ; ; ) {
5853 if (!rq->nr_running)
5854 break;
b67802ea 5855 next = pick_next_task(rq);
dd41f596
IM
5856 if (!next)
5857 break;
79c53799 5858 next->sched_class->put_prev_task(rq, next);
dd41f596 5859 migrate_dead(dead_cpu, next);
e692ab53 5860
1da177e4
LT
5861 }
5862}
dce48a84
TG
5863
5864/*
5865 * remove the tasks which were accounted by rq from calc_load_tasks.
5866 */
5867static void calc_global_load_remove(struct rq *rq)
5868{
5869 atomic_long_sub(rq->calc_load_active, &calc_load_tasks);
a468d389 5870 rq->calc_load_active = 0;
dce48a84 5871}
1da177e4
LT
5872#endif /* CONFIG_HOTPLUG_CPU */
5873
e692ab53
NP
5874#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)
5875
5876static struct ctl_table sd_ctl_dir[] = {
e0361851
AD
5877 {
5878 .procname = "sched_domain",
c57baf1e 5879 .mode = 0555,
e0361851 5880 },
56992309 5881 {}
e692ab53
NP
5882};
5883
5884static struct ctl_table sd_ctl_root[] = {
e0361851
AD
5885 {
5886 .procname = "kernel",
c57baf1e 5887 .mode = 0555,
e0361851
AD
5888 .child = sd_ctl_dir,
5889 },
56992309 5890 {}
e692ab53
NP
5891};
5892
5893static struct ctl_table *sd_alloc_ctl_entry(int n)
5894{
5895 struct ctl_table *entry =
5cf9f062 5896 kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
e692ab53 5897
e692ab53
NP
5898 return entry;
5899}
5900
6382bc90
MM
5901static void sd_free_ctl_entry(struct ctl_table **tablep)
5902{
cd790076 5903 struct ctl_table *entry;
6382bc90 5904
cd790076
MM
5905 /*
5906 * In the intermediate directories, both the child directory and
5907 * procname are dynamically allocated and could fail but the mode
41a2d6cf 5908 * will always be set. In the lowest directory the names are
cd790076
MM
5909 * static strings and all have proc handlers.
5910 */
5911 for (entry = *tablep; entry->mode; entry++) {
6382bc90
MM
5912 if (entry->child)
5913 sd_free_ctl_entry(&entry->child);
cd790076
MM
5914 if (entry->proc_handler == NULL)
5915 kfree(entry->procname);
5916 }
6382bc90
MM
5917
5918 kfree(*tablep);
5919 *tablep = NULL;
5920}
5921
e692ab53 5922static void
e0361851 5923set_table_entry(struct ctl_table *entry,
e692ab53
NP
5924 const char *procname, void *data, int maxlen,
5925 mode_t mode, proc_handler *proc_handler)
5926{
e692ab53
NP
5927 entry->procname = procname;
5928 entry->data = data;
5929 entry->maxlen = maxlen;
5930 entry->mode = mode;
5931 entry->proc_handler = proc_handler;
5932}
5933
5934static struct ctl_table *
5935sd_alloc_ctl_domain_table(struct sched_domain *sd)
5936{
a5d8c348 5937 struct ctl_table *table = sd_alloc_ctl_entry(13);
e692ab53 5938
ad1cdc1d
MM
5939 if (table == NULL)
5940 return NULL;
5941
e0361851 5942 set_table_entry(&table[0], "min_interval", &sd->min_interval,
e692ab53 5943 sizeof(long), 0644, proc_doulongvec_minmax);
e0361851 5944 set_table_entry(&table[1], "max_interval", &sd->max_interval,
e692ab53 5945 sizeof(long), 0644, proc_doulongvec_minmax);
e0361851 5946 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
e692ab53 5947 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5948 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
e692ab53 5949 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5950 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
e692ab53 5951 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5952 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
e692ab53 5953 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5954 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
e692ab53 5955 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5956 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
e692ab53 5957 sizeof(int), 0644, proc_dointvec_minmax);
e0361851 5958 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
e692ab53 5959 sizeof(int), 0644, proc_dointvec_minmax);
ace8b3d6 5960 set_table_entry(&table[9], "cache_nice_tries",
e692ab53
NP
5961 &sd->cache_nice_tries,
5962 sizeof(int), 0644, proc_dointvec_minmax);
ace8b3d6 5963 set_table_entry(&table[10], "flags", &sd->flags,
e692ab53 5964 sizeof(int), 0644, proc_dointvec_minmax);
a5d8c348
IM
5965 set_table_entry(&table[11], "name", sd->name,
5966 CORENAME_MAX_SIZE, 0444, proc_dostring);
5967 /* &table[12] is terminator */
e692ab53
NP
5968
5969 return table;
5970}
5971
9a4e7159 5972static ctl_table *sd_alloc_ctl_cpu_table(int cpu)
e692ab53
NP
5973{
5974 struct ctl_table *entry, *table;
5975 struct sched_domain *sd;
5976 int domain_num = 0, i;
5977 char buf[32];
5978
5979 for_each_domain(cpu, sd)
5980 domain_num++;
5981 entry = table = sd_alloc_ctl_entry(domain_num + 1);
ad1cdc1d
MM
5982 if (table == NULL)
5983 return NULL;
e692ab53
NP
5984
5985 i = 0;
5986 for_each_domain(cpu, sd) {
5987 snprintf(buf, 32, "domain%d", i);
e692ab53 5988 entry->procname = kstrdup(buf, GFP_KERNEL);
c57baf1e 5989 entry->mode = 0555;
e692ab53
NP
5990 entry->child = sd_alloc_ctl_domain_table(sd);
5991 entry++;
5992 i++;
5993 }
5994 return table;
5995}
5996
5997static struct ctl_table_header *sd_sysctl_header;
6382bc90 5998static void register_sched_domain_sysctl(void)
e692ab53 5999{
6ad4c188 6000 int i, cpu_num = num_possible_cpus();
e692ab53
NP
6001 struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
6002 char buf[32];
6003
7378547f
MM
6004 WARN_ON(sd_ctl_dir[0].child);
6005 sd_ctl_dir[0].child = entry;
6006
ad1cdc1d
MM
6007 if (entry == NULL)
6008 return;
6009
6ad4c188 6010 for_each_possible_cpu(i) {
e692ab53 6011 snprintf(buf, 32, "cpu%d", i);
e692ab53 6012 entry->procname = kstrdup(buf, GFP_KERNEL);
c57baf1e 6013 entry->mode = 0555;
e692ab53 6014 entry->child = sd_alloc_ctl_cpu_table(i);
97b6ea7b 6015 entry++;
e692ab53 6016 }
7378547f
MM
6017
6018 WARN_ON(sd_sysctl_header);
e692ab53
NP
6019 sd_sysctl_header = register_sysctl_table(sd_ctl_root);
6020}
6382bc90 6021
7378547f 6022/* may be called multiple times per register */
6382bc90
MM
6023static void unregister_sched_domain_sysctl(void)
6024{
7378547f
MM
6025 if (sd_sysctl_header)
6026 unregister_sysctl_table(sd_sysctl_header);
6382bc90 6027 sd_sysctl_header = NULL;
7378547f
MM
6028 if (sd_ctl_dir[0].child)
6029 sd_free_ctl_entry(&sd_ctl_dir[0].child);
6382bc90 6030}
e692ab53 6031#else
6382bc90
MM
6032static void register_sched_domain_sysctl(void)
6033{
6034}
6035static void unregister_sched_domain_sysctl(void)
e692ab53
NP
6036{
6037}
6038#endif
6039
1f11eb6a
GH
6040static void set_rq_online(struct rq *rq)
6041{
6042 if (!rq->online) {
6043 const struct sched_class *class;
6044
c6c4927b 6045 cpumask_set_cpu(rq->cpu, rq->rd->online);
1f11eb6a
GH
6046 rq->online = 1;
6047
6048 for_each_class(class) {
6049 if (class->rq_online)
6050 class->rq_online(rq);
6051 }
6052 }
6053}
6054
6055static void set_rq_offline(struct rq *rq)
6056{
6057 if (rq->online) {
6058 const struct sched_class *class;
6059
6060 for_each_class(class) {
6061 if (class->rq_offline)
6062 class->rq_offline(rq);
6063 }
6064
c6c4927b 6065 cpumask_clear_cpu(rq->cpu, rq->rd->online);
1f11eb6a
GH
6066 rq->online = 0;
6067 }
6068}
6069
1da177e4
LT
6070/*
6071 * migration_call - callback that gets triggered when a CPU is added.
6072 * Here we can start up the necessary migration thread for the new CPU.
6073 */
48f24c4d
IM
6074static int __cpuinit
6075migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
1da177e4 6076{
48f24c4d 6077 int cpu = (long)hcpu;
1da177e4 6078 unsigned long flags;
969c7921 6079 struct rq *rq = cpu_rq(cpu);
1da177e4
LT
6080
6081 switch (action) {
5be9361c 6082
1da177e4 6083 case CPU_UP_PREPARE:
8bb78442 6084 case CPU_UP_PREPARE_FROZEN:
a468d389 6085 rq->calc_load_update = calc_load_update;
1da177e4 6086 break;
48f24c4d 6087
1da177e4 6088 case CPU_ONLINE:
8bb78442 6089 case CPU_ONLINE_FROZEN:
1f94ef59 6090 /* Update our root-domain */
05fa785c 6091 raw_spin_lock_irqsave(&rq->lock, flags);
1f94ef59 6092 if (rq->rd) {
c6c4927b 6093 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
1f11eb6a
GH
6094
6095 set_rq_online(rq);
1f94ef59 6096 }
05fa785c 6097 raw_spin_unlock_irqrestore(&rq->lock, flags);
1da177e4 6098 break;
48f24c4d 6099
1da177e4 6100#ifdef CONFIG_HOTPLUG_CPU
1da177e4 6101 case CPU_DEAD:
8bb78442 6102 case CPU_DEAD_FROZEN:
1da177e4 6103 migrate_live_tasks(cpu);
1da177e4 6104 /* Idle task back to normal (off runqueue, low prio) */
05fa785c 6105 raw_spin_lock_irq(&rq->lock);
2e1cb74a 6106 deactivate_task(rq, rq->idle, 0);
dd41f596
IM
6107 __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
6108 rq->idle->sched_class = &idle_sched_class;
1da177e4 6109 migrate_dead_tasks(cpu);
05fa785c 6110 raw_spin_unlock_irq(&rq->lock);
1da177e4
LT
6111 migrate_nr_uninterruptible(rq);
6112 BUG_ON(rq->nr_running != 0);
dce48a84 6113 calc_global_load_remove(rq);
1da177e4 6114 break;
57d885fe 6115
08f503b0
GH
6116 case CPU_DYING:
6117 case CPU_DYING_FROZEN:
57d885fe 6118 /* Update our root-domain */
05fa785c 6119 raw_spin_lock_irqsave(&rq->lock, flags);
57d885fe 6120 if (rq->rd) {
c6c4927b 6121 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
1f11eb6a 6122 set_rq_offline(rq);
57d885fe 6123 }
05fa785c 6124 raw_spin_unlock_irqrestore(&rq->lock, flags);
57d885fe 6125 break;
1da177e4
LT
6126#endif
6127 }
6128 return NOTIFY_OK;
6129}
6130
f38b0820
PM
6131/*
6132 * Register at high priority so that task migration (migrate_all_tasks)
6133 * happens before everything else. This has to be lower priority than
cdd6c482 6134 * the notifier in the perf_event subsystem, though.
1da177e4 6135 */
26c2143b 6136static struct notifier_block __cpuinitdata migration_notifier = {
1da177e4 6137 .notifier_call = migration_call,
50a323b7 6138 .priority = CPU_PRI_MIGRATION,
1da177e4
LT
6139};
6140
3a101d05
TH
6141static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
6142 unsigned long action, void *hcpu)
6143{
6144 switch (action & ~CPU_TASKS_FROZEN) {
6145 case CPU_ONLINE:
6146 case CPU_DOWN_FAILED:
6147 set_cpu_active((long)hcpu, true);
6148 return NOTIFY_OK;
6149 default:
6150 return NOTIFY_DONE;
6151 }
6152}
6153
6154static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
6155 unsigned long action, void *hcpu)
6156{
6157 switch (action & ~CPU_TASKS_FROZEN) {
6158 case CPU_DOWN_PREPARE:
6159 set_cpu_active((long)hcpu, false);
6160 return NOTIFY_OK;
6161 default:
6162 return NOTIFY_DONE;
6163 }
6164}
6165
7babe8db 6166static int __init migration_init(void)
1da177e4
LT
6167{
6168 void *cpu = (void *)(long)smp_processor_id();
07dccf33 6169 int err;
48f24c4d 6170
3a101d05 6171 /* Initialize migration for the boot CPU */
07dccf33
AM
6172 err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
6173 BUG_ON(err == NOTIFY_BAD);
1da177e4
LT
6174 migration_call(&migration_notifier, CPU_ONLINE, cpu);
6175 register_cpu_notifier(&migration_notifier);
7babe8db 6176
3a101d05
TH
6177 /* Register cpu active notifiers */
6178 cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
6179 cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);
6180
a004cd42 6181 return 0;
1da177e4 6182}
7babe8db 6183early_initcall(migration_init);
1da177e4
LT
6184#endif
6185
6186#ifdef CONFIG_SMP
476f3534 6187
3e9830dc 6188#ifdef CONFIG_SCHED_DEBUG
4dcf6aff 6189
f6630114
MT
6190static __read_mostly int sched_domain_debug_enabled;
6191
6192static int __init sched_domain_debug_setup(char *str)
6193{
6194 sched_domain_debug_enabled = 1;
6195
6196 return 0;
6197}
6198early_param("sched_debug", sched_domain_debug_setup);
6199
7c16ec58 6200static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
96f874e2 6201 struct cpumask *groupmask)
1da177e4 6202{
4dcf6aff 6203 struct sched_group *group = sd->groups;
434d53b0 6204 char str[256];
1da177e4 6205
968ea6d8 6206 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
96f874e2 6207 cpumask_clear(groupmask);
4dcf6aff
IM
6208
6209 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6210
6211 if (!(sd->flags & SD_LOAD_BALANCE)) {
3df0fc5b 6212 printk("does not load-balance\n");
4dcf6aff 6213 if (sd->parent)
3df0fc5b
PZ
6214 printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
6215 " has parent");
4dcf6aff 6216 return -1;
41c7ce9a
NP
6217 }
6218
3df0fc5b 6219 printk(KERN_CONT "span %s level %s\n", str, sd->name);
4dcf6aff 6220
758b2cdc 6221 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
3df0fc5b
PZ
6222 printk(KERN_ERR "ERROR: domain->span does not contain "
6223 "CPU%d\n", cpu);
4dcf6aff 6224 }
758b2cdc 6225 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
3df0fc5b
PZ
6226 printk(KERN_ERR "ERROR: domain->groups does not contain"
6227 " CPU%d\n", cpu);
4dcf6aff 6228 }
1da177e4 6229
4dcf6aff 6230 printk(KERN_DEBUG "%*s groups:", level + 1, "");
1da177e4 6231 do {
4dcf6aff 6232 if (!group) {
3df0fc5b
PZ
6233 printk("\n");
6234 printk(KERN_ERR "ERROR: group is NULL\n");
1da177e4
LT
6235 break;
6236 }
6237
18a3885f 6238 if (!group->cpu_power) {
3df0fc5b
PZ
6239 printk(KERN_CONT "\n");
6240 printk(KERN_ERR "ERROR: domain->cpu_power not "
6241 "set\n");
4dcf6aff
IM
6242 break;
6243 }
1da177e4 6244
758b2cdc 6245 if (!cpumask_weight(sched_group_cpus(group))) {
3df0fc5b
PZ
6246 printk(KERN_CONT "\n");
6247 printk(KERN_ERR "ERROR: empty group\n");
4dcf6aff
IM
6248 break;
6249 }
1da177e4 6250
758b2cdc 6251 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
3df0fc5b
PZ
6252 printk(KERN_CONT "\n");
6253 printk(KERN_ERR "ERROR: repeated CPUs\n");
4dcf6aff
IM
6254 break;
6255 }
1da177e4 6256
758b2cdc 6257 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
1da177e4 6258
968ea6d8 6259 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
381512cf 6260
3df0fc5b 6261 printk(KERN_CONT " %s", str);
18a3885f 6262 if (group->cpu_power != SCHED_LOAD_SCALE) {
3df0fc5b
PZ
6263 printk(KERN_CONT " (cpu_power = %d)",
6264 group->cpu_power);
381512cf 6265 }
1da177e4 6266
4dcf6aff
IM
6267 group = group->next;
6268 } while (group != sd->groups);
3df0fc5b 6269 printk(KERN_CONT "\n");
1da177e4 6270
758b2cdc 6271 if (!cpumask_equal(sched_domain_span(sd), groupmask))
3df0fc5b 6272 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
1da177e4 6273
758b2cdc
RR
6274 if (sd->parent &&
6275 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
3df0fc5b
PZ
6276 printk(KERN_ERR "ERROR: parent span is not a superset "
6277 "of domain->span\n");
4dcf6aff
IM
6278 return 0;
6279}
1da177e4 6280
4dcf6aff
IM
6281static void sched_domain_debug(struct sched_domain *sd, int cpu)
6282{
d5dd3db1 6283 cpumask_var_t groupmask;
4dcf6aff 6284 int level = 0;
1da177e4 6285
f6630114
MT
6286 if (!sched_domain_debug_enabled)
6287 return;
6288
4dcf6aff
IM
6289 if (!sd) {
6290 printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
6291 return;
6292 }
1da177e4 6293
4dcf6aff
IM
6294 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6295
d5dd3db1 6296 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
7c16ec58
MT
6297 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6298 return;
6299 }
6300
4dcf6aff 6301 for (;;) {
7c16ec58 6302 if (sched_domain_debug_one(sd, cpu, level, groupmask))
4dcf6aff 6303 break;
1da177e4
LT
6304 level++;
6305 sd = sd->parent;
33859f7f 6306 if (!sd)
4dcf6aff
IM
6307 break;
6308 }
d5dd3db1 6309 free_cpumask_var(groupmask);
1da177e4 6310}
6d6bc0ad 6311#else /* !CONFIG_SCHED_DEBUG */
48f24c4d 6312# define sched_domain_debug(sd, cpu) do { } while (0)
6d6bc0ad 6313#endif /* CONFIG_SCHED_DEBUG */
1da177e4 6314
1a20ff27 6315static int sd_degenerate(struct sched_domain *sd)
245af2c7 6316{
758b2cdc 6317 if (cpumask_weight(sched_domain_span(sd)) == 1)
245af2c7
SS
6318 return 1;
6319
6320 /* Following flags need at least 2 groups */
6321 if (sd->flags & (SD_LOAD_BALANCE |
6322 SD_BALANCE_NEWIDLE |
6323 SD_BALANCE_FORK |
89c4710e
SS
6324 SD_BALANCE_EXEC |
6325 SD_SHARE_CPUPOWER |
6326 SD_SHARE_PKG_RESOURCES)) {
245af2c7
SS
6327 if (sd->groups != sd->groups->next)
6328 return 0;
6329 }
6330
6331 /* Following flags don't use groups */
c88d5910 6332 if (sd->flags & (SD_WAKE_AFFINE))
245af2c7
SS
6333 return 0;
6334
6335 return 1;
6336}
6337
48f24c4d
IM
6338static int
6339sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
245af2c7
SS
6340{
6341 unsigned long cflags = sd->flags, pflags = parent->flags;
6342
6343 if (sd_degenerate(parent))
6344 return 1;
6345
758b2cdc 6346 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
245af2c7
SS
6347 return 0;
6348
245af2c7
SS
6349 /* Flags needing groups don't count if only 1 group in parent */
6350 if (parent->groups == parent->groups->next) {
6351 pflags &= ~(SD_LOAD_BALANCE |
6352 SD_BALANCE_NEWIDLE |
6353 SD_BALANCE_FORK |
89c4710e
SS
6354 SD_BALANCE_EXEC |
6355 SD_SHARE_CPUPOWER |
6356 SD_SHARE_PKG_RESOURCES);
5436499e
KC
6357 if (nr_node_ids == 1)
6358 pflags &= ~SD_SERIALIZE;
245af2c7
SS
6359 }
6360 if (~cflags & pflags)
6361 return 0;
6362
6363 return 1;
6364}
6365
c6c4927b
RR
6366static void free_rootdomain(struct root_domain *rd)
6367{
047106ad
PZ
6368 synchronize_sched();
6369
68e74568
RR
6370 cpupri_cleanup(&rd->cpupri);
6371
c6c4927b
RR
6372 free_cpumask_var(rd->rto_mask);
6373 free_cpumask_var(rd->online);
6374 free_cpumask_var(rd->span);
6375 kfree(rd);
6376}
6377
57d885fe
GH
6378static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6379{
a0490fa3 6380 struct root_domain *old_rd = NULL;
57d885fe 6381 unsigned long flags;
57d885fe 6382
05fa785c 6383 raw_spin_lock_irqsave(&rq->lock, flags);
57d885fe
GH
6384
6385 if (rq->rd) {
a0490fa3 6386 old_rd = rq->rd;
57d885fe 6387
c6c4927b 6388 if (cpumask_test_cpu(rq->cpu, old_rd->online))
1f11eb6a 6389 set_rq_offline(rq);
57d885fe 6390
c6c4927b 6391 cpumask_clear_cpu(rq->cpu, old_rd->span);
dc938520 6392
a0490fa3
IM
6393 /*
6394 * If we dont want to free the old_rt yet then
6395 * set old_rd to NULL to skip the freeing later
6396 * in this function:
6397 */
6398 if (!atomic_dec_and_test(&old_rd->refcount))
6399 old_rd = NULL;
57d885fe
GH
6400 }
6401
6402 atomic_inc(&rd->refcount);
6403 rq->rd = rd;
6404
c6c4927b 6405 cpumask_set_cpu(rq->cpu, rd->span);
00aec93d 6406 if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
1f11eb6a 6407 set_rq_online(rq);
57d885fe 6408
05fa785c 6409 raw_spin_unlock_irqrestore(&rq->lock, flags);
a0490fa3
IM
6410
6411 if (old_rd)
6412 free_rootdomain(old_rd);
57d885fe
GH
6413}
6414
68c38fc3 6415static int init_rootdomain(struct root_domain *rd)
57d885fe
GH
6416{
6417 memset(rd, 0, sizeof(*rd));
6418
68c38fc3 6419 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
0c910d28 6420 goto out;
68c38fc3 6421 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
c6c4927b 6422 goto free_span;
68c38fc3 6423 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
c6c4927b 6424 goto free_online;
6e0534f2 6425
68c38fc3 6426 if (cpupri_init(&rd->cpupri) != 0)
68e74568 6427 goto free_rto_mask;
c6c4927b 6428 return 0;
6e0534f2 6429
68e74568
RR
6430free_rto_mask:
6431 free_cpumask_var(rd->rto_mask);
c6c4927b
RR
6432free_online:
6433 free_cpumask_var(rd->online);
6434free_span:
6435 free_cpumask_var(rd->span);
0c910d28 6436out:
c6c4927b 6437 return -ENOMEM;
57d885fe
GH
6438}
6439
6440static void init_defrootdomain(void)
6441{
68c38fc3 6442 init_rootdomain(&def_root_domain);
c6c4927b 6443
57d885fe
GH
6444 atomic_set(&def_root_domain.refcount, 1);
6445}
6446
dc938520 6447static struct root_domain *alloc_rootdomain(void)
57d885fe
GH
6448{
6449 struct root_domain *rd;
6450
6451 rd = kmalloc(sizeof(*rd), GFP_KERNEL);
6452 if (!rd)
6453 return NULL;
6454
68c38fc3 6455 if (init_rootdomain(rd) != 0) {
c6c4927b
RR
6456 kfree(rd);
6457 return NULL;
6458 }
57d885fe
GH
6459
6460 return rd;
6461}
6462
1da177e4 6463/*
0eab9146 6464 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
1da177e4
LT
6465 * hold the hotplug lock.
6466 */
0eab9146
IM
6467static void
6468cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
1da177e4 6469{
70b97a7f 6470 struct rq *rq = cpu_rq(cpu);
245af2c7
SS
6471 struct sched_domain *tmp;
6472
669c55e9
PZ
6473 for (tmp = sd; tmp; tmp = tmp->parent)
6474 tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
6475
245af2c7 6476 /* Remove the sched domains which do not contribute to scheduling. */
f29c9b1c 6477 for (tmp = sd; tmp; ) {
245af2c7
SS
6478 struct sched_domain *parent = tmp->parent;
6479 if (!parent)
6480 break;
f29c9b1c 6481
1a848870 6482 if (sd_parent_degenerate(tmp, parent)) {
245af2c7 6483 tmp->parent = parent->parent;
1a848870
SS
6484 if (parent->parent)
6485 parent->parent->child = tmp;
f29c9b1c
LZ
6486 } else
6487 tmp = tmp->parent;
245af2c7
SS
6488 }
6489
1a848870 6490 if (sd && sd_degenerate(sd)) {
245af2c7 6491 sd = sd->parent;
1a848870
SS
6492 if (sd)
6493 sd->child = NULL;
6494 }
1da177e4
LT
6495
6496 sched_domain_debug(sd, cpu);
6497
57d885fe 6498 rq_attach_root(rq, rd);
674311d5 6499 rcu_assign_pointer(rq->sd, sd);
1da177e4
LT
6500}
6501
6502/* cpus with isolated domains */
dcc30a35 6503static cpumask_var_t cpu_isolated_map;
1da177e4
LT
6504
6505/* Setup the mask of cpus configured for isolated domains */
6506static int __init isolated_cpu_setup(char *str)
6507{
bdddd296 6508 alloc_bootmem_cpumask_var(&cpu_isolated_map);
968ea6d8 6509 cpulist_parse(str, cpu_isolated_map);
1da177e4
LT
6510 return 1;
6511}
6512
8927f494 6513__setup("isolcpus=", isolated_cpu_setup);
1da177e4
LT
6514
6515/*
6711cab4
SS
6516 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6517 * to a function which identifies what group(along with sched group) a CPU
96f874e2
RR
6518 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6519 * (due to the fact that we keep track of groups covered with a struct cpumask).
1da177e4
LT
6520 *
6521 * init_sched_build_groups will build a circular linked list of the groups
6522 * covered by the given span, and will set each group's ->cpumask correctly,
6523 * and ->cpu_power to 0.
6524 */
a616058b 6525static void
96f874e2
RR
6526init_sched_build_groups(const struct cpumask *span,
6527 const struct cpumask *cpu_map,
6528 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
7c16ec58 6529 struct sched_group **sg,
96f874e2
RR
6530 struct cpumask *tmpmask),
6531 struct cpumask *covered, struct cpumask *tmpmask)
1da177e4
LT
6532{
6533 struct sched_group *first = NULL, *last = NULL;
1da177e4
LT
6534 int i;
6535
96f874e2 6536 cpumask_clear(covered);
7c16ec58 6537
abcd083a 6538 for_each_cpu(i, span) {
6711cab4 6539 struct sched_group *sg;
7c16ec58 6540 int group = group_fn(i, cpu_map, &sg, tmpmask);
1da177e4
LT
6541 int j;
6542
758b2cdc 6543 if (cpumask_test_cpu(i, covered))
1da177e4
LT
6544 continue;
6545
758b2cdc 6546 cpumask_clear(sched_group_cpus(sg));
18a3885f 6547 sg->cpu_power = 0;
1da177e4 6548
abcd083a 6549 for_each_cpu(j, span) {
7c16ec58 6550 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
1da177e4
LT
6551 continue;
6552
96f874e2 6553 cpumask_set_cpu(j, covered);
758b2cdc 6554 cpumask_set_cpu(j, sched_group_cpus(sg));
1da177e4
LT
6555 }
6556 if (!first)
6557 first = sg;
6558 if (last)
6559 last->next = sg;
6560 last = sg;
6561 }
6562 last->next = first;
6563}
6564
9c1cfda2 6565#define SD_NODES_PER_DOMAIN 16
1da177e4 6566
9c1cfda2 6567#ifdef CONFIG_NUMA
198e2f18 6568
9c1cfda2
JH
6569/**
6570 * find_next_best_node - find the next node to include in a sched_domain
6571 * @node: node whose sched_domain we're building
6572 * @used_nodes: nodes already in the sched_domain
6573 *
41a2d6cf 6574 * Find the next node to include in a given scheduling domain. Simply
9c1cfda2
JH
6575 * finds the closest node not already in the @used_nodes map.
6576 *
6577 * Should use nodemask_t.
6578 */
c5f59f08 6579static int find_next_best_node(int node, nodemask_t *used_nodes)
9c1cfda2
JH
6580{
6581 int i, n, val, min_val, best_node = 0;
6582
6583 min_val = INT_MAX;
6584
076ac2af 6585 for (i = 0; i < nr_node_ids; i++) {
9c1cfda2 6586 /* Start at @node */
076ac2af 6587 n = (node + i) % nr_node_ids;
9c1cfda2
JH
6588
6589 if (!nr_cpus_node(n))
6590 continue;
6591
6592 /* Skip already used nodes */
c5f59f08 6593 if (node_isset(n, *used_nodes))
9c1cfda2
JH
6594 continue;
6595
6596 /* Simple min distance search */
6597 val = node_distance(node, n);
6598
6599 if (val < min_val) {
6600 min_val = val;
6601 best_node = n;
6602 }
6603 }
6604
c5f59f08 6605 node_set(best_node, *used_nodes);
9c1cfda2
JH
6606 return best_node;
6607}
6608
6609/**
6610 * sched_domain_node_span - get a cpumask for a node's sched_domain
6611 * @node: node whose cpumask we're constructing
73486722 6612 * @span: resulting cpumask
9c1cfda2 6613 *
41a2d6cf 6614 * Given a node, construct a good cpumask for its sched_domain to span. It
9c1cfda2
JH
6615 * should be one that prevents unnecessary balancing, but also spreads tasks
6616 * out optimally.
6617 */
96f874e2 6618static void sched_domain_node_span(int node, struct cpumask *span)
9c1cfda2 6619{
c5f59f08 6620 nodemask_t used_nodes;
48f24c4d 6621 int i;
9c1cfda2 6622
6ca09dfc 6623 cpumask_clear(span);
c5f59f08 6624 nodes_clear(used_nodes);
9c1cfda2 6625
6ca09dfc 6626 cpumask_or(span, span, cpumask_of_node(node));
c5f59f08 6627 node_set(node, used_nodes);
9c1cfda2
JH
6628
6629 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
c5f59f08 6630 int next_node = find_next_best_node(node, &used_nodes);
48f24c4d 6631
6ca09dfc 6632 cpumask_or(span, span, cpumask_of_node(next_node));
9c1cfda2 6633 }
9c1cfda2 6634}
6d6bc0ad 6635#endif /* CONFIG_NUMA */
9c1cfda2 6636
5c45bf27 6637int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
48f24c4d 6638
6c99e9ad
RR
6639/*
6640 * The cpus mask in sched_group and sched_domain hangs off the end.
4200efd9
IM
6641 *
6642 * ( See the the comments in include/linux/sched.h:struct sched_group
6643 * and struct sched_domain. )
6c99e9ad
RR
6644 */
6645struct static_sched_group {
6646 struct sched_group sg;
6647 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
6648};
6649
6650struct static_sched_domain {
6651 struct sched_domain sd;
6652 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
6653};
6654
49a02c51
AH
6655struct s_data {
6656#ifdef CONFIG_NUMA
6657 int sd_allnodes;
6658 cpumask_var_t domainspan;
6659 cpumask_var_t covered;
6660 cpumask_var_t notcovered;
6661#endif
6662 cpumask_var_t nodemask;
6663 cpumask_var_t this_sibling_map;
6664 cpumask_var_t this_core_map;
01a08546 6665 cpumask_var_t this_book_map;
49a02c51
AH
6666 cpumask_var_t send_covered;
6667 cpumask_var_t tmpmask;
6668 struct sched_group **sched_group_nodes;
6669 struct root_domain *rd;
6670};
6671
2109b99e
AH
6672enum s_alloc {
6673 sa_sched_groups = 0,
6674 sa_rootdomain,
6675 sa_tmpmask,
6676 sa_send_covered,
01a08546 6677 sa_this_book_map,
2109b99e
AH
6678 sa_this_core_map,
6679 sa_this_sibling_map,
6680 sa_nodemask,
6681 sa_sched_group_nodes,
6682#ifdef CONFIG_NUMA
6683 sa_notcovered,
6684 sa_covered,
6685 sa_domainspan,
6686#endif
6687 sa_none,
6688};
6689
9c1cfda2 6690/*
48f24c4d 6691 * SMT sched-domains:
9c1cfda2 6692 */
1da177e4 6693#ifdef CONFIG_SCHED_SMT
6c99e9ad 6694static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
1871e52c 6695static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
48f24c4d 6696
41a2d6cf 6697static int
96f874e2
RR
6698cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6699 struct sched_group **sg, struct cpumask *unused)
1da177e4 6700{
6711cab4 6701 if (sg)
1871e52c 6702 *sg = &per_cpu(sched_groups, cpu).sg;
1da177e4
LT
6703 return cpu;
6704}
6d6bc0ad 6705#endif /* CONFIG_SCHED_SMT */
1da177e4 6706
48f24c4d
IM
6707/*
6708 * multi-core sched-domains:
6709 */
1e9f28fa 6710#ifdef CONFIG_SCHED_MC
6c99e9ad
RR
6711static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6712static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
1e9f28fa 6713
41a2d6cf 6714static int
96f874e2
RR
6715cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6716 struct sched_group **sg, struct cpumask *mask)
1e9f28fa 6717{
6711cab4 6718 int group;
f269893c 6719#ifdef CONFIG_SCHED_SMT
c69fc56d 6720 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
96f874e2 6721 group = cpumask_first(mask);
f269893c
HC
6722#else
6723 group = cpu;
6724#endif
6711cab4 6725 if (sg)
6c99e9ad 6726 *sg = &per_cpu(sched_group_core, group).sg;
6711cab4 6727 return group;
1e9f28fa 6728}
f269893c 6729#endif /* CONFIG_SCHED_MC */
1e9f28fa 6730
01a08546
HC
6731/*
6732 * book sched-domains:
6733 */
6734#ifdef CONFIG_SCHED_BOOK
6735static DEFINE_PER_CPU(struct static_sched_domain, book_domains);
6736static DEFINE_PER_CPU(struct static_sched_group, sched_group_book);
6737
41a2d6cf 6738static int
01a08546
HC
6739cpu_to_book_group(int cpu, const struct cpumask *cpu_map,
6740 struct sched_group **sg, struct cpumask *mask)
1e9f28fa 6741{
01a08546
HC
6742 int group = cpu;
6743#ifdef CONFIG_SCHED_MC
6744 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
6745 group = cpumask_first(mask);
6746#elif defined(CONFIG_SCHED_SMT)
6747 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
6748 group = cpumask_first(mask);
6749#endif
6711cab4 6750 if (sg)
01a08546
HC
6751 *sg = &per_cpu(sched_group_book, group).sg;
6752 return group;
1e9f28fa 6753}
01a08546 6754#endif /* CONFIG_SCHED_BOOK */
1e9f28fa 6755
6c99e9ad
RR
6756static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6757static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
48f24c4d 6758
41a2d6cf 6759static int
96f874e2
RR
6760cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6761 struct sched_group **sg, struct cpumask *mask)
1da177e4 6762{
6711cab4 6763 int group;
01a08546
HC
6764#ifdef CONFIG_SCHED_BOOK
6765 cpumask_and(mask, cpu_book_mask(cpu), cpu_map);
6766 group = cpumask_first(mask);
6767#elif defined(CONFIG_SCHED_MC)
6ca09dfc 6768 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
96f874e2 6769 group = cpumask_first(mask);
1e9f28fa 6770#elif defined(CONFIG_SCHED_SMT)
c69fc56d 6771 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
96f874e2 6772 group = cpumask_first(mask);
1da177e4 6773#else
6711cab4 6774 group = cpu;
1da177e4 6775#endif
6711cab4 6776 if (sg)
6c99e9ad 6777 *sg = &per_cpu(sched_group_phys, group).sg;
6711cab4 6778 return group;
1da177e4
LT
6779}
6780
6781#ifdef CONFIG_NUMA
1da177e4 6782/*
9c1cfda2
JH
6783 * The init_sched_build_groups can't handle what we want to do with node
6784 * groups, so roll our own. Now each node has its own list of groups which
6785 * gets dynamically allocated.
1da177e4 6786 */
62ea9ceb 6787static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
434d53b0 6788static struct sched_group ***sched_group_nodes_bycpu;
1da177e4 6789
62ea9ceb 6790static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
6c99e9ad 6791static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
9c1cfda2 6792
96f874e2
RR
6793static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
6794 struct sched_group **sg,
6795 struct cpumask *nodemask)
9c1cfda2 6796{
6711cab4
SS
6797 int group;
6798
6ca09dfc 6799 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
96f874e2 6800 group = cpumask_first(nodemask);
6711cab4
SS
6801
6802 if (sg)
6c99e9ad 6803 *sg = &per_cpu(sched_group_allnodes, group).sg;
6711cab4 6804 return group;
1da177e4 6805}
6711cab4 6806
08069033
SS
6807static void init_numa_sched_groups_power(struct sched_group *group_head)
6808{
6809 struct sched_group *sg = group_head;
6810 int j;
6811
6812 if (!sg)
6813 return;
3a5c359a 6814 do {
758b2cdc 6815 for_each_cpu(j, sched_group_cpus(sg)) {
3a5c359a 6816 struct sched_domain *sd;
08069033 6817
6c99e9ad 6818 sd = &per_cpu(phys_domains, j).sd;
13318a71 6819 if (j != group_first_cpu(sd->groups)) {
3a5c359a
AK
6820 /*
6821 * Only add "power" once for each
6822 * physical package.
6823 */
6824 continue;
6825 }
08069033 6826
18a3885f 6827 sg->cpu_power += sd->groups->cpu_power;
3a5c359a
AK
6828 }
6829 sg = sg->next;
6830 } while (sg != group_head);
08069033 6831}
0601a88d
AH
6832
6833static int build_numa_sched_groups(struct s_data *d,
6834 const struct cpumask *cpu_map, int num)
6835{
6836 struct sched_domain *sd;
6837 struct sched_group *sg, *prev;
6838 int n, j;
6839
6840 cpumask_clear(d->covered);
6841 cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
6842 if (cpumask_empty(d->nodemask)) {
6843 d->sched_group_nodes[num] = NULL;
6844 goto out;
6845 }
6846
6847 sched_domain_node_span(num, d->domainspan);
6848 cpumask_and(d->domainspan, d->domainspan, cpu_map);
6849
6850 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6851 GFP_KERNEL, num);
6852 if (!sg) {
3df0fc5b
PZ
6853 printk(KERN_WARNING "Can not alloc domain group for node %d\n",
6854 num);
0601a88d
AH
6855 return -ENOMEM;
6856 }
6857 d->sched_group_nodes[num] = sg;
6858
6859 for_each_cpu(j, d->nodemask) {
6860 sd = &per_cpu(node_domains, j).sd;
6861 sd->groups = sg;
6862 }
6863
18a3885f 6864 sg->cpu_power = 0;
0601a88d
AH
6865 cpumask_copy(sched_group_cpus(sg), d->nodemask);
6866 sg->next = sg;
6867 cpumask_or(d->covered, d->covered, d->nodemask);
6868
6869 prev = sg;
6870 for (j = 0; j < nr_node_ids; j++) {
6871 n = (num + j) % nr_node_ids;
6872 cpumask_complement(d->notcovered, d->covered);
6873 cpumask_and(d->tmpmask, d->notcovered, cpu_map);
6874 cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
6875 if (cpumask_empty(d->tmpmask))
6876 break;
6877 cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
6878 if (cpumask_empty(d->tmpmask))
6879 continue;
6880 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
6881 GFP_KERNEL, num);
6882 if (!sg) {
3df0fc5b
PZ
6883 printk(KERN_WARNING
6884 "Can not alloc domain group for node %d\n", j);
0601a88d
AH
6885 return -ENOMEM;
6886 }
18a3885f 6887 sg->cpu_power = 0;
0601a88d
AH
6888 cpumask_copy(sched_group_cpus(sg), d->tmpmask);
6889 sg->next = prev->next;
6890 cpumask_or(d->covered, d->covered, d->tmpmask);
6891 prev->next = sg;
6892 prev = sg;
6893 }
6894out:
6895 return 0;
6896}
6d6bc0ad 6897#endif /* CONFIG_NUMA */
1da177e4 6898
a616058b 6899#ifdef CONFIG_NUMA
51888ca2 6900/* Free memory allocated for various sched_group structures */
96f874e2
RR
6901static void free_sched_groups(const struct cpumask *cpu_map,
6902 struct cpumask *nodemask)
51888ca2 6903{
a616058b 6904 int cpu, i;
51888ca2 6905
abcd083a 6906 for_each_cpu(cpu, cpu_map) {
51888ca2
SV
6907 struct sched_group **sched_group_nodes
6908 = sched_group_nodes_bycpu[cpu];
6909
51888ca2
SV
6910 if (!sched_group_nodes)
6911 continue;
6912
076ac2af 6913 for (i = 0; i < nr_node_ids; i++) {
51888ca2
SV
6914 struct sched_group *oldsg, *sg = sched_group_nodes[i];
6915
6ca09dfc 6916 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
96f874e2 6917 if (cpumask_empty(nodemask))
51888ca2
SV
6918 continue;
6919
6920 if (sg == NULL)
6921 continue;
6922 sg = sg->next;
6923next_sg:
6924 oldsg = sg;
6925 sg = sg->next;
6926 kfree(oldsg);
6927 if (oldsg != sched_group_nodes[i])
6928 goto next_sg;
6929 }
6930 kfree(sched_group_nodes);
6931 sched_group_nodes_bycpu[cpu] = NULL;
6932 }
51888ca2 6933}
6d6bc0ad 6934#else /* !CONFIG_NUMA */
96f874e2
RR
6935static void free_sched_groups(const struct cpumask *cpu_map,
6936 struct cpumask *nodemask)
a616058b
SS
6937{
6938}
6d6bc0ad 6939#endif /* CONFIG_NUMA */
51888ca2 6940
89c4710e
SS
6941/*
6942 * Initialize sched groups cpu_power.
6943 *
6944 * cpu_power indicates the capacity of sched group, which is used while
6945 * distributing the load between different sched groups in a sched domain.
6946 * Typically cpu_power for all the groups in a sched domain will be same unless
6947 * there are asymmetries in the topology. If there are asymmetries, group
6948 * having more cpu_power will pickup more load compared to the group having
6949 * less cpu_power.
89c4710e
SS
6950 */
6951static void init_sched_groups_power(int cpu, struct sched_domain *sd)
6952{
6953 struct sched_domain *child;
6954 struct sched_group *group;
f93e65c1
PZ
6955 long power;
6956 int weight;
89c4710e
SS
6957
6958 WARN_ON(!sd || !sd->groups);
6959
13318a71 6960 if (cpu != group_first_cpu(sd->groups))
89c4710e
SS
6961 return;
6962
6963 child = sd->child;
6964
18a3885f 6965 sd->groups->cpu_power = 0;
5517d86b 6966
f93e65c1
PZ
6967 if (!child) {
6968 power = SCHED_LOAD_SCALE;
6969 weight = cpumask_weight(sched_domain_span(sd));
6970 /*
6971 * SMT siblings share the power of a single core.
a52bfd73
PZ
6972 * Usually multiple threads get a better yield out of
6973 * that one core than a single thread would have,
6974 * reflect that in sd->smt_gain.
f93e65c1 6975 */
a52bfd73
PZ
6976 if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
6977 power *= sd->smt_gain;
f93e65c1 6978 power /= weight;
a52bfd73
PZ
6979 power >>= SCHED_LOAD_SHIFT;
6980 }
18a3885f 6981 sd->groups->cpu_power += power;
89c4710e
SS
6982 return;
6983 }
6984
89c4710e 6985 /*
f93e65c1 6986 * Add cpu_power of each child group to this groups cpu_power.
89c4710e
SS
6987 */
6988 group = child->groups;
6989 do {
18a3885f 6990 sd->groups->cpu_power += group->cpu_power;
89c4710e
SS
6991 group = group->next;
6992 } while (group != child->groups);
6993}
6994
7c16ec58
MT
6995/*
6996 * Initializers for schedule domains
6997 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
6998 */
6999
a5d8c348
IM
7000#ifdef CONFIG_SCHED_DEBUG
7001# define SD_INIT_NAME(sd, type) sd->name = #type
7002#else
7003# define SD_INIT_NAME(sd, type) do { } while (0)
7004#endif
7005
7c16ec58 7006#define SD_INIT(sd, type) sd_init_##type(sd)
a5d8c348 7007
7c16ec58
MT
7008#define SD_INIT_FUNC(type) \
7009static noinline void sd_init_##type(struct sched_domain *sd) \
7010{ \
7011 memset(sd, 0, sizeof(*sd)); \
7012 *sd = SD_##type##_INIT; \
1d3504fc 7013 sd->level = SD_LV_##type; \
a5d8c348 7014 SD_INIT_NAME(sd, type); \
7c16ec58
MT
7015}
7016
7017SD_INIT_FUNC(CPU)
7018#ifdef CONFIG_NUMA
7019 SD_INIT_FUNC(ALLNODES)
7020 SD_INIT_FUNC(NODE)
7021#endif
7022#ifdef CONFIG_SCHED_SMT
7023 SD_INIT_FUNC(SIBLING)
7024#endif
7025#ifdef CONFIG_SCHED_MC
7026 SD_INIT_FUNC(MC)
7027#endif
01a08546
HC
7028#ifdef CONFIG_SCHED_BOOK
7029 SD_INIT_FUNC(BOOK)
7030#endif
7c16ec58 7031
1d3504fc
HS
7032static int default_relax_domain_level = -1;
7033
7034static int __init setup_relax_domain_level(char *str)
7035{
30e0e178
LZ
7036 unsigned long val;
7037
7038 val = simple_strtoul(str, NULL, 0);
7039 if (val < SD_LV_MAX)
7040 default_relax_domain_level = val;
7041
1d3504fc
HS
7042 return 1;
7043}
7044__setup("relax_domain_level=", setup_relax_domain_level);
7045
7046static void set_domain_attribute(struct sched_domain *sd,
7047 struct sched_domain_attr *attr)
7048{
7049 int request;
7050
7051 if (!attr || attr->relax_domain_level < 0) {
7052 if (default_relax_domain_level < 0)
7053 return;
7054 else
7055 request = default_relax_domain_level;
7056 } else
7057 request = attr->relax_domain_level;
7058 if (request < sd->level) {
7059 /* turn off idle balance on this domain */
c88d5910 7060 sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
1d3504fc
HS
7061 } else {
7062 /* turn on idle balance on this domain */
c88d5910 7063 sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
1d3504fc
HS
7064 }
7065}
7066
2109b99e
AH
7067static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
7068 const struct cpumask *cpu_map)
7069{
7070 switch (what) {
7071 case sa_sched_groups:
7072 free_sched_groups(cpu_map, d->tmpmask); /* fall through */
7073 d->sched_group_nodes = NULL;
7074 case sa_rootdomain:
7075 free_rootdomain(d->rd); /* fall through */
7076 case sa_tmpmask:
7077 free_cpumask_var(d->tmpmask); /* fall through */
7078 case sa_send_covered:
7079 free_cpumask_var(d->send_covered); /* fall through */
01a08546
HC
7080 case sa_this_book_map:
7081 free_cpumask_var(d->this_book_map); /* fall through */
2109b99e
AH
7082 case sa_this_core_map:
7083 free_cpumask_var(d->this_core_map); /* fall through */
7084 case sa_this_sibling_map:
7085 free_cpumask_var(d->this_sibling_map); /* fall through */
7086 case sa_nodemask:
7087 free_cpumask_var(d->nodemask); /* fall through */
7088 case sa_sched_group_nodes:
d1b55138 7089#ifdef CONFIG_NUMA
2109b99e
AH
7090 kfree(d->sched_group_nodes); /* fall through */
7091 case sa_notcovered:
7092 free_cpumask_var(d->notcovered); /* fall through */
7093 case sa_covered:
7094 free_cpumask_var(d->covered); /* fall through */
7095 case sa_domainspan:
7096 free_cpumask_var(d->domainspan); /* fall through */
3404c8d9 7097#endif
2109b99e
AH
7098 case sa_none:
7099 break;
7100 }
7101}
3404c8d9 7102
2109b99e
AH
7103static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
7104 const struct cpumask *cpu_map)
7105{
3404c8d9 7106#ifdef CONFIG_NUMA
2109b99e
AH
7107 if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
7108 return sa_none;
7109 if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
7110 return sa_domainspan;
7111 if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
7112 return sa_covered;
7113 /* Allocate the per-node list of sched groups */
7114 d->sched_group_nodes = kcalloc(nr_node_ids,
7115 sizeof(struct sched_group *), GFP_KERNEL);
7116 if (!d->sched_group_nodes) {
3df0fc5b 7117 printk(KERN_WARNING "Can not alloc sched group node list\n");
2109b99e 7118 return sa_notcovered;
d1b55138 7119 }
2109b99e 7120 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
d1b55138 7121#endif
2109b99e
AH
7122 if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
7123 return sa_sched_group_nodes;
7124 if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
7125 return sa_nodemask;
7126 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
7127 return sa_this_sibling_map;
01a08546 7128 if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL))
2109b99e 7129 return sa_this_core_map;
01a08546
HC
7130 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
7131 return sa_this_book_map;
2109b99e
AH
7132 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
7133 return sa_send_covered;
7134 d->rd = alloc_rootdomain();
7135 if (!d->rd) {
3df0fc5b 7136 printk(KERN_WARNING "Cannot alloc root domain\n");
2109b99e 7137 return sa_tmpmask;
57d885fe 7138 }
2109b99e
AH
7139 return sa_rootdomain;
7140}
57d885fe 7141
7f4588f3
AH
7142static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
7143 const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
7144{
7145 struct sched_domain *sd = NULL;
7c16ec58 7146#ifdef CONFIG_NUMA
7f4588f3 7147 struct sched_domain *parent;
1da177e4 7148
7f4588f3
AH
7149 d->sd_allnodes = 0;
7150 if (cpumask_weight(cpu_map) >
7151 SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
7152 sd = &per_cpu(allnodes_domains, i).sd;
7153 SD_INIT(sd, ALLNODES);
1d3504fc 7154 set_domain_attribute(sd, attr);
7f4588f3
AH
7155 cpumask_copy(sched_domain_span(sd), cpu_map);
7156 cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
7157 d->sd_allnodes = 1;
7158 }
7159 parent = sd;
7160
7161 sd = &per_cpu(node_domains, i).sd;
7162 SD_INIT(sd, NODE);
7163 set_domain_attribute(sd, attr);
7164 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
7165 sd->parent = parent;
7166 if (parent)
7167 parent->child = sd;
7168 cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
1da177e4 7169#endif
7f4588f3
AH
7170 return sd;
7171}
1da177e4 7172
87cce662
AH
7173static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
7174 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7175 struct sched_domain *parent, int i)
7176{
7177 struct sched_domain *sd;
7178 sd = &per_cpu(phys_domains, i).sd;
7179 SD_INIT(sd, CPU);
7180 set_domain_attribute(sd, attr);
7181 cpumask_copy(sched_domain_span(sd), d->nodemask);
7182 sd->parent = parent;
7183 if (parent)
7184 parent->child = sd;
7185 cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
7186 return sd;
7187}
1da177e4 7188
01a08546
HC
7189static struct sched_domain *__build_book_sched_domain(struct s_data *d,
7190 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7191 struct sched_domain *parent, int i)
7192{
7193 struct sched_domain *sd = parent;
7194#ifdef CONFIG_SCHED_BOOK
7195 sd = &per_cpu(book_domains, i).sd;
7196 SD_INIT(sd, BOOK);
7197 set_domain_attribute(sd, attr);
7198 cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i));
7199 sd->parent = parent;
7200 parent->child = sd;
7201 cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask);
7202#endif
7203 return sd;
7204}
7205
410c4081
AH
7206static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
7207 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7208 struct sched_domain *parent, int i)
7209{
7210 struct sched_domain *sd = parent;
1e9f28fa 7211#ifdef CONFIG_SCHED_MC
410c4081
AH
7212 sd = &per_cpu(core_domains, i).sd;
7213 SD_INIT(sd, MC);
7214 set_domain_attribute(sd, attr);
7215 cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
7216 sd->parent = parent;
7217 parent->child = sd;
7218 cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
1e9f28fa 7219#endif
410c4081
AH
7220 return sd;
7221}
1e9f28fa 7222
d8173535
AH
7223static struct sched_domain *__build_smt_sched_domain(struct s_data *d,
7224 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7225 struct sched_domain *parent, int i)
7226{
7227 struct sched_domain *sd = parent;
1da177e4 7228#ifdef CONFIG_SCHED_SMT
d8173535
AH
7229 sd = &per_cpu(cpu_domains, i).sd;
7230 SD_INIT(sd, SIBLING);
7231 set_domain_attribute(sd, attr);
7232 cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
7233 sd->parent = parent;
7234 parent->child = sd;
7235 cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
1da177e4 7236#endif
d8173535
AH
7237 return sd;
7238}
1da177e4 7239
0e8e85c9
AH
7240static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
7241 const struct cpumask *cpu_map, int cpu)
7242{
7243 switch (l) {
1da177e4 7244#ifdef CONFIG_SCHED_SMT
0e8e85c9
AH
7245 case SD_LV_SIBLING: /* set up CPU (sibling) groups */
7246 cpumask_and(d->this_sibling_map, cpu_map,
7247 topology_thread_cpumask(cpu));
7248 if (cpu == cpumask_first(d->this_sibling_map))
7249 init_sched_build_groups(d->this_sibling_map, cpu_map,
7250 &cpu_to_cpu_group,
7251 d->send_covered, d->tmpmask);
7252 break;
1da177e4 7253#endif
1e9f28fa 7254#ifdef CONFIG_SCHED_MC
a2af04cd
AH
7255 case SD_LV_MC: /* set up multi-core groups */
7256 cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
7257 if (cpu == cpumask_first(d->this_core_map))
7258 init_sched_build_groups(d->this_core_map, cpu_map,
7259 &cpu_to_core_group,
7260 d->send_covered, d->tmpmask);
7261 break;
01a08546
HC
7262#endif
7263#ifdef CONFIG_SCHED_BOOK
7264 case SD_LV_BOOK: /* set up book groups */
7265 cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu));
7266 if (cpu == cpumask_first(d->this_book_map))
7267 init_sched_build_groups(d->this_book_map, cpu_map,
7268 &cpu_to_book_group,
7269 d->send_covered, d->tmpmask);
7270 break;
1e9f28fa 7271#endif
86548096
AH
7272 case SD_LV_CPU: /* set up physical groups */
7273 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
7274 if (!cpumask_empty(d->nodemask))
7275 init_sched_build_groups(d->nodemask, cpu_map,
7276 &cpu_to_phys_group,
7277 d->send_covered, d->tmpmask);
7278 break;
1da177e4 7279#ifdef CONFIG_NUMA
de616e36
AH
7280 case SD_LV_ALLNODES:
7281 init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
7282 d->send_covered, d->tmpmask);
7283 break;
7284#endif
0e8e85c9
AH
7285 default:
7286 break;
7c16ec58 7287 }
0e8e85c9 7288}
9c1cfda2 7289
2109b99e
AH
7290/*
7291 * Build sched domains for a given set of cpus and attach the sched domains
7292 * to the individual cpus
7293 */
7294static int __build_sched_domains(const struct cpumask *cpu_map,
7295 struct sched_domain_attr *attr)
7296{
7297 enum s_alloc alloc_state = sa_none;
7298 struct s_data d;
294b0c96 7299 struct sched_domain *sd;
2109b99e 7300 int i;
7c16ec58 7301#ifdef CONFIG_NUMA
2109b99e 7302 d.sd_allnodes = 0;
7c16ec58 7303#endif
9c1cfda2 7304
2109b99e
AH
7305 alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
7306 if (alloc_state != sa_rootdomain)
7307 goto error;
7308 alloc_state = sa_sched_groups;
9c1cfda2 7309
1da177e4 7310 /*
1a20ff27 7311 * Set up domains for cpus specified by the cpu_map.
1da177e4 7312 */
abcd083a 7313 for_each_cpu(i, cpu_map) {
49a02c51
AH
7314 cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
7315 cpu_map);
9761eea8 7316
7f4588f3 7317 sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
87cce662 7318 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
01a08546 7319 sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
410c4081 7320 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
d8173535 7321 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
1da177e4 7322 }
9c1cfda2 7323
abcd083a 7324 for_each_cpu(i, cpu_map) {
0e8e85c9 7325 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
01a08546 7326 build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
a2af04cd 7327 build_sched_groups(&d, SD_LV_MC, cpu_map, i);
1da177e4 7328 }
9c1cfda2 7329
1da177e4 7330 /* Set up physical groups */
86548096
AH
7331 for (i = 0; i < nr_node_ids; i++)
7332 build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
9c1cfda2 7333
1da177e4
LT
7334#ifdef CONFIG_NUMA
7335 /* Set up node groups */
de616e36
AH
7336 if (d.sd_allnodes)
7337 build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
9c1cfda2 7338
0601a88d
AH
7339 for (i = 0; i < nr_node_ids; i++)
7340 if (build_numa_sched_groups(&d, cpu_map, i))
51888ca2 7341 goto error;
1da177e4
LT
7342#endif
7343
7344 /* Calculate CPU power for physical packages and nodes */
5c45bf27 7345#ifdef CONFIG_SCHED_SMT
abcd083a 7346 for_each_cpu(i, cpu_map) {
294b0c96 7347 sd = &per_cpu(cpu_domains, i).sd;
89c4710e 7348 init_sched_groups_power(i, sd);
5c45bf27 7349 }
1da177e4 7350#endif
1e9f28fa 7351#ifdef CONFIG_SCHED_MC
abcd083a 7352 for_each_cpu(i, cpu_map) {
294b0c96 7353 sd = &per_cpu(core_domains, i).sd;
89c4710e 7354 init_sched_groups_power(i, sd);
5c45bf27
SS
7355 }
7356#endif
01a08546
HC
7357#ifdef CONFIG_SCHED_BOOK
7358 for_each_cpu(i, cpu_map) {
7359 sd = &per_cpu(book_domains, i).sd;
7360 init_sched_groups_power(i, sd);
7361 }
7362#endif
1e9f28fa 7363
abcd083a 7364 for_each_cpu(i, cpu_map) {
294b0c96 7365 sd = &per_cpu(phys_domains, i).sd;
89c4710e 7366 init_sched_groups_power(i, sd);
1da177e4
LT
7367 }
7368
9c1cfda2 7369#ifdef CONFIG_NUMA
076ac2af 7370 for (i = 0; i < nr_node_ids; i++)
49a02c51 7371 init_numa_sched_groups_power(d.sched_group_nodes[i]);
9c1cfda2 7372
49a02c51 7373 if (d.sd_allnodes) {
6711cab4 7374 struct sched_group *sg;
f712c0c7 7375
96f874e2 7376 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
49a02c51 7377 d.tmpmask);
f712c0c7
SS
7378 init_numa_sched_groups_power(sg);
7379 }
9c1cfda2
JH
7380#endif
7381
1da177e4 7382 /* Attach the domains */
abcd083a 7383 for_each_cpu(i, cpu_map) {
1da177e4 7384#ifdef CONFIG_SCHED_SMT
6c99e9ad 7385 sd = &per_cpu(cpu_domains, i).sd;
1e9f28fa 7386#elif defined(CONFIG_SCHED_MC)
6c99e9ad 7387 sd = &per_cpu(core_domains, i).sd;
01a08546
HC
7388#elif defined(CONFIG_SCHED_BOOK)
7389 sd = &per_cpu(book_domains, i).sd;
1da177e4 7390#else
6c99e9ad 7391 sd = &per_cpu(phys_domains, i).sd;
1da177e4 7392#endif
49a02c51 7393 cpu_attach_domain(sd, d.rd, i);
1da177e4 7394 }
51888ca2 7395
2109b99e
AH
7396 d.sched_group_nodes = NULL; /* don't free this we still need it */
7397 __free_domain_allocs(&d, sa_tmpmask, cpu_map);
7398 return 0;
51888ca2 7399
51888ca2 7400error:
2109b99e
AH
7401 __free_domain_allocs(&d, alloc_state, cpu_map);
7402 return -ENOMEM;
1da177e4 7403}
029190c5 7404
96f874e2 7405static int build_sched_domains(const struct cpumask *cpu_map)
1d3504fc
HS
7406{
7407 return __build_sched_domains(cpu_map, NULL);
7408}
7409
acc3f5d7 7410static cpumask_var_t *doms_cur; /* current sched domains */
029190c5 7411static int ndoms_cur; /* number of sched domains in 'doms_cur' */
4285f594
IM
7412static struct sched_domain_attr *dattr_cur;
7413 /* attribues of custom domains in 'doms_cur' */
029190c5
PJ
7414
7415/*
7416 * Special case: If a kmalloc of a doms_cur partition (array of
4212823f
RR
7417 * cpumask) fails, then fallback to a single sched domain,
7418 * as determined by the single cpumask fallback_doms.
029190c5 7419 */
4212823f 7420static cpumask_var_t fallback_doms;
029190c5 7421
ee79d1bd
HC
7422/*
7423 * arch_update_cpu_topology lets virtualized architectures update the
7424 * cpu core maps. It is supposed to return 1 if the topology changed
7425 * or 0 if it stayed the same.
7426 */
7427int __attribute__((weak)) arch_update_cpu_topology(void)
22e52b07 7428{
ee79d1bd 7429 return 0;
22e52b07
HC
7430}
7431
acc3f5d7
RR
7432cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
7433{
7434 int i;
7435 cpumask_var_t *doms;
7436
7437 doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
7438 if (!doms)
7439 return NULL;
7440 for (i = 0; i < ndoms; i++) {
7441 if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
7442 free_sched_domains(doms, i);
7443 return NULL;
7444 }
7445 }
7446 return doms;
7447}
7448
7449void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
7450{
7451 unsigned int i;
7452 for (i = 0; i < ndoms; i++)
7453 free_cpumask_var(doms[i]);
7454 kfree(doms);
7455}
7456
1a20ff27 7457/*
41a2d6cf 7458 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
029190c5
PJ
7459 * For now this just excludes isolated cpus, but could be used to
7460 * exclude other special cases in the future.
1a20ff27 7461 */
96f874e2 7462static int arch_init_sched_domains(const struct cpumask *cpu_map)
1a20ff27 7463{
7378547f
MM
7464 int err;
7465
22e52b07 7466 arch_update_cpu_topology();
029190c5 7467 ndoms_cur = 1;
acc3f5d7 7468 doms_cur = alloc_sched_domains(ndoms_cur);
029190c5 7469 if (!doms_cur)
acc3f5d7
RR
7470 doms_cur = &fallback_doms;
7471 cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
1d3504fc 7472 dattr_cur = NULL;
acc3f5d7 7473 err = build_sched_domains(doms_cur[0]);
6382bc90 7474 register_sched_domain_sysctl();
7378547f
MM
7475
7476 return err;
1a20ff27
DG
7477}
7478
96f874e2
RR
7479static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7480 struct cpumask *tmpmask)
1da177e4 7481{
7c16ec58 7482 free_sched_groups(cpu_map, tmpmask);
9c1cfda2 7483}
1da177e4 7484
1a20ff27
DG
7485/*
7486 * Detach sched domains from a group of cpus specified in cpu_map
7487 * These cpus will now be attached to the NULL domain
7488 */
96f874e2 7489static void detach_destroy_domains(const struct cpumask *cpu_map)
1a20ff27 7490{
96f874e2
RR
7491 /* Save because hotplug lock held. */
7492 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
1a20ff27
DG
7493 int i;
7494
abcd083a 7495 for_each_cpu(i, cpu_map)
57d885fe 7496 cpu_attach_domain(NULL, &def_root_domain, i);
1a20ff27 7497 synchronize_sched();
96f874e2 7498 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
1a20ff27
DG
7499}
7500
1d3504fc
HS
7501/* handle null as "default" */
7502static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7503 struct sched_domain_attr *new, int idx_new)
7504{
7505 struct sched_domain_attr tmp;
7506
7507 /* fast path */
7508 if (!new && !cur)
7509 return 1;
7510
7511 tmp = SD_ATTR_INIT;
7512 return !memcmp(cur ? (cur + idx_cur) : &tmp,
7513 new ? (new + idx_new) : &tmp,
7514 sizeof(struct sched_domain_attr));
7515}
7516
029190c5
PJ
7517/*
7518 * Partition sched domains as specified by the 'ndoms_new'
41a2d6cf 7519 * cpumasks in the array doms_new[] of cpumasks. This compares
029190c5
PJ
7520 * doms_new[] to the current sched domain partitioning, doms_cur[].
7521 * It destroys each deleted domain and builds each new domain.
7522 *
acc3f5d7 7523 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
41a2d6cf
IM
7524 * The masks don't intersect (don't overlap.) We should setup one
7525 * sched domain for each mask. CPUs not in any of the cpumasks will
7526 * not be load balanced. If the same cpumask appears both in the
029190c5
PJ
7527 * current 'doms_cur' domains and in the new 'doms_new', we can leave
7528 * it as it is.
7529 *
acc3f5d7
RR
7530 * The passed in 'doms_new' should be allocated using
7531 * alloc_sched_domains. This routine takes ownership of it and will
7532 * free_sched_domains it when done with it. If the caller failed the
7533 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
7534 * and partition_sched_domains() will fallback to the single partition
7535 * 'fallback_doms', it also forces the domains to be rebuilt.
029190c5 7536 *
96f874e2 7537 * If doms_new == NULL it will be replaced with cpu_online_mask.
700018e0
LZ
7538 * ndoms_new == 0 is a special case for destroying existing domains,
7539 * and it will not create the default domain.
dfb512ec 7540 *
029190c5
PJ
7541 * Call with hotplug lock held
7542 */
acc3f5d7 7543void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
1d3504fc 7544 struct sched_domain_attr *dattr_new)
029190c5 7545{
dfb512ec 7546 int i, j, n;
d65bd5ec 7547 int new_topology;
029190c5 7548
712555ee 7549 mutex_lock(&sched_domains_mutex);
a1835615 7550
7378547f
MM
7551 /* always unregister in case we don't destroy any domains */
7552 unregister_sched_domain_sysctl();
7553
d65bd5ec
HC
7554 /* Let architecture update cpu core mappings. */
7555 new_topology = arch_update_cpu_topology();
7556
dfb512ec 7557 n = doms_new ? ndoms_new : 0;
029190c5
PJ
7558
7559 /* Destroy deleted domains */
7560 for (i = 0; i < ndoms_cur; i++) {
d65bd5ec 7561 for (j = 0; j < n && !new_topology; j++) {
acc3f5d7 7562 if (cpumask_equal(doms_cur[i], doms_new[j])
1d3504fc 7563 && dattrs_equal(dattr_cur, i, dattr_new, j))
029190c5
PJ
7564 goto match1;
7565 }
7566 /* no match - a current sched domain not in new doms_new[] */
acc3f5d7 7567 detach_destroy_domains(doms_cur[i]);
029190c5
PJ
7568match1:
7569 ;
7570 }
7571
e761b772
MK
7572 if (doms_new == NULL) {
7573 ndoms_cur = 0;
acc3f5d7 7574 doms_new = &fallback_doms;
6ad4c188 7575 cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
faa2f98f 7576 WARN_ON_ONCE(dattr_new);
e761b772
MK
7577 }
7578
029190c5
PJ
7579 /* Build new domains */
7580 for (i = 0; i < ndoms_new; i++) {
d65bd5ec 7581 for (j = 0; j < ndoms_cur && !new_topology; j++) {
acc3f5d7 7582 if (cpumask_equal(doms_new[i], doms_cur[j])
1d3504fc 7583 && dattrs_equal(dattr_new, i, dattr_cur, j))
029190c5
PJ
7584 goto match2;
7585 }
7586 /* no match - add a new doms_new */
acc3f5d7 7587 __build_sched_domains(doms_new[i],
1d3504fc 7588 dattr_new ? dattr_new + i : NULL);
029190c5
PJ
7589match2:
7590 ;
7591 }
7592
7593 /* Remember the new sched domains */
acc3f5d7
RR
7594 if (doms_cur != &fallback_doms)
7595 free_sched_domains(doms_cur, ndoms_cur);
1d3504fc 7596 kfree(dattr_cur); /* kfree(NULL) is safe */
029190c5 7597 doms_cur = doms_new;
1d3504fc 7598 dattr_cur = dattr_new;
029190c5 7599 ndoms_cur = ndoms_new;
7378547f
MM
7600
7601 register_sched_domain_sysctl();
a1835615 7602
712555ee 7603 mutex_unlock(&sched_domains_mutex);
029190c5
PJ
7604}
7605
5c45bf27 7606#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
c70f22d2 7607static void arch_reinit_sched_domains(void)
5c45bf27 7608{
95402b38 7609 get_online_cpus();
dfb512ec
MK
7610
7611 /* Destroy domains first to force the rebuild */
7612 partition_sched_domains(0, NULL, NULL);
7613
e761b772 7614 rebuild_sched_domains();
95402b38 7615 put_online_cpus();
5c45bf27
SS
7616}
7617
7618static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7619{
afb8a9b7 7620 unsigned int level = 0;
5c45bf27 7621
afb8a9b7
GS
7622 if (sscanf(buf, "%u", &level) != 1)
7623 return -EINVAL;
7624
7625 /*
7626 * level is always be positive so don't check for
7627 * level < POWERSAVINGS_BALANCE_NONE which is 0
7628 * What happens on 0 or 1 byte write,
7629 * need to check for count as well?
7630 */
7631
7632 if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS)
5c45bf27
SS
7633 return -EINVAL;
7634
7635 if (smt)
afb8a9b7 7636 sched_smt_power_savings = level;
5c45bf27 7637 else
afb8a9b7 7638 sched_mc_power_savings = level;
5c45bf27 7639
c70f22d2 7640 arch_reinit_sched_domains();
5c45bf27 7641
c70f22d2 7642 return count;
5c45bf27
SS
7643}
7644
5c45bf27 7645#ifdef CONFIG_SCHED_MC
f718cd4a 7646static ssize_t sched_mc_power_savings_show(struct sysdev_class *class,
c9be0a36 7647 struct sysdev_class_attribute *attr,
f718cd4a 7648 char *page)
5c45bf27
SS
7649{
7650 return sprintf(page, "%u\n", sched_mc_power_savings);
7651}
f718cd4a 7652static ssize_t sched_mc_power_savings_store(struct sysdev_class *class,
c9be0a36 7653 struct sysdev_class_attribute *attr,
48f24c4d 7654 const char *buf, size_t count)
5c45bf27
SS
7655{
7656 return sched_power_savings_store(buf, count, 0);
7657}
f718cd4a
AK
7658static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644,
7659 sched_mc_power_savings_show,
7660 sched_mc_power_savings_store);
5c45bf27
SS
7661#endif
7662
7663#ifdef CONFIG_SCHED_SMT
f718cd4a 7664static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev,
c9be0a36 7665 struct sysdev_class_attribute *attr,
f718cd4a 7666 char *page)
5c45bf27
SS
7667{
7668 return sprintf(page, "%u\n", sched_smt_power_savings);
7669}
f718cd4a 7670static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev,
c9be0a36 7671 struct sysdev_class_attribute *attr,
48f24c4d 7672 const char *buf, size_t count)
5c45bf27
SS
7673{
7674 return sched_power_savings_store(buf, count, 1);
7675}
f718cd4a
AK
7676static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644,
7677 sched_smt_power_savings_show,
6707de00
AB
7678 sched_smt_power_savings_store);
7679#endif
7680
39aac648 7681int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
6707de00
AB
7682{
7683 int err = 0;
7684
7685#ifdef CONFIG_SCHED_SMT
7686 if (smt_capable())
7687 err = sysfs_create_file(&cls->kset.kobj,
7688 &attr_sched_smt_power_savings.attr);
7689#endif
7690#ifdef CONFIG_SCHED_MC
7691 if (!err && mc_capable())
7692 err = sysfs_create_file(&cls->kset.kobj,
7693 &attr_sched_mc_power_savings.attr);
7694#endif
7695 return err;
7696}
6d6bc0ad 7697#endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
5c45bf27 7698
1da177e4 7699/*
3a101d05
TH
7700 * Update cpusets according to cpu_active mask. If cpusets are
7701 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
7702 * around partition_sched_domains().
1da177e4 7703 */
0b2e918a
TH
7704static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
7705 void *hcpu)
e761b772 7706{
3a101d05 7707 switch (action & ~CPU_TASKS_FROZEN) {
e761b772 7708 case CPU_ONLINE:
6ad4c188 7709 case CPU_DOWN_FAILED:
3a101d05 7710 cpuset_update_active_cpus();
e761b772 7711 return NOTIFY_OK;
3a101d05
TH
7712 default:
7713 return NOTIFY_DONE;
7714 }
7715}
e761b772 7716
0b2e918a
TH
7717static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
7718 void *hcpu)
3a101d05
TH
7719{
7720 switch (action & ~CPU_TASKS_FROZEN) {
7721 case CPU_DOWN_PREPARE:
7722 cpuset_update_active_cpus();
7723 return NOTIFY_OK;
e761b772
MK
7724 default:
7725 return NOTIFY_DONE;
7726 }
7727}
e761b772
MK
7728
7729static int update_runtime(struct notifier_block *nfb,
7730 unsigned long action, void *hcpu)
1da177e4 7731{
7def2be1
PZ
7732 int cpu = (int)(long)hcpu;
7733
1da177e4 7734 switch (action) {
1da177e4 7735 case CPU_DOWN_PREPARE:
8bb78442 7736 case CPU_DOWN_PREPARE_FROZEN:
7def2be1 7737 disable_runtime(cpu_rq(cpu));
1da177e4
LT
7738 return NOTIFY_OK;
7739
1da177e4 7740 case CPU_DOWN_FAILED:
8bb78442 7741 case CPU_DOWN_FAILED_FROZEN:
1da177e4 7742 case CPU_ONLINE:
8bb78442 7743 case CPU_ONLINE_FROZEN:
7def2be1 7744 enable_runtime(cpu_rq(cpu));
e761b772
MK
7745 return NOTIFY_OK;
7746
1da177e4
LT
7747 default:
7748 return NOTIFY_DONE;
7749 }
1da177e4 7750}
1da177e4
LT
7751
7752void __init sched_init_smp(void)
7753{
dcc30a35
RR
7754 cpumask_var_t non_isolated_cpus;
7755
7756 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
cb5fd13f 7757 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
5c1e1767 7758
434d53b0
MT
7759#if defined(CONFIG_NUMA)
7760 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
7761 GFP_KERNEL);
7762 BUG_ON(sched_group_nodes_bycpu == NULL);
7763#endif
95402b38 7764 get_online_cpus();
712555ee 7765 mutex_lock(&sched_domains_mutex);
6ad4c188 7766 arch_init_sched_domains(cpu_active_mask);
dcc30a35
RR
7767 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
7768 if (cpumask_empty(non_isolated_cpus))
7769 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
712555ee 7770 mutex_unlock(&sched_domains_mutex);
95402b38 7771 put_online_cpus();
e761b772 7772
3a101d05
TH
7773 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
7774 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
e761b772
MK
7775
7776 /* RT runtime code needs to handle some hotplug events */
7777 hotcpu_notifier(update_runtime, 0);
7778
b328ca18 7779 init_hrtick();
5c1e1767
NP
7780
7781 /* Move init over to a non-isolated CPU */
dcc30a35 7782 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
5c1e1767 7783 BUG();
19978ca6 7784 sched_init_granularity();
dcc30a35 7785 free_cpumask_var(non_isolated_cpus);
4212823f 7786
0e3900e6 7787 init_sched_rt_class();
1da177e4
LT
7788}
7789#else
7790void __init sched_init_smp(void)
7791{
19978ca6 7792 sched_init_granularity();
1da177e4
LT
7793}
7794#endif /* CONFIG_SMP */
7795
cd1bb94b
AB
7796const_debug unsigned int sysctl_timer_migration = 1;
7797
1da177e4
LT
7798int in_sched_functions(unsigned long addr)
7799{
1da177e4
LT
7800 return in_lock_functions(addr) ||
7801 (addr >= (unsigned long)__sched_text_start
7802 && addr < (unsigned long)__sched_text_end);
7803}
7804
a9957449 7805static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
dd41f596
IM
7806{
7807 cfs_rq->tasks_timeline = RB_ROOT;
4a55bd5e 7808 INIT_LIST_HEAD(&cfs_rq->tasks);
dd41f596
IM
7809#ifdef CONFIG_FAIR_GROUP_SCHED
7810 cfs_rq->rq = rq;
7811#endif
67e9fb2a 7812 cfs_rq->min_vruntime = (u64)(-(1LL << 20));
dd41f596
IM
7813}
7814
fa85ae24
PZ
7815static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
7816{
7817 struct rt_prio_array *array;
7818 int i;
7819
7820 array = &rt_rq->active;
7821 for (i = 0; i < MAX_RT_PRIO; i++) {
7822 INIT_LIST_HEAD(array->queue + i);
7823 __clear_bit(i, array->bitmap);
7824 }
7825 /* delimiter for bitsearch: */
7826 __set_bit(MAX_RT_PRIO, array->bitmap);
7827
052f1dc7 7828#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
e864c499 7829 rt_rq->highest_prio.curr = MAX_RT_PRIO;
398a153b 7830#ifdef CONFIG_SMP
e864c499 7831 rt_rq->highest_prio.next = MAX_RT_PRIO;
48d5e258 7832#endif
48d5e258 7833#endif
fa85ae24
PZ
7834#ifdef CONFIG_SMP
7835 rt_rq->rt_nr_migratory = 0;
fa85ae24 7836 rt_rq->overloaded = 0;
05fa785c 7837 plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock);
fa85ae24
PZ
7838#endif
7839
7840 rt_rq->rt_time = 0;
7841 rt_rq->rt_throttled = 0;
ac086bc2 7842 rt_rq->rt_runtime = 0;
0986b11b 7843 raw_spin_lock_init(&rt_rq->rt_runtime_lock);
6f505b16 7844
052f1dc7 7845#ifdef CONFIG_RT_GROUP_SCHED
23b0fdfc 7846 rt_rq->rt_nr_boosted = 0;
6f505b16
PZ
7847 rt_rq->rq = rq;
7848#endif
fa85ae24
PZ
7849}
7850
6f505b16 7851#ifdef CONFIG_FAIR_GROUP_SCHED
ec7dc8ac
DG
7852static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
7853 struct sched_entity *se, int cpu, int add,
7854 struct sched_entity *parent)
6f505b16 7855{
ec7dc8ac 7856 struct rq *rq = cpu_rq(cpu);
6f505b16
PZ
7857 tg->cfs_rq[cpu] = cfs_rq;
7858 init_cfs_rq(cfs_rq, rq);
7859 cfs_rq->tg = tg;
7860 if (add)
7861 list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
7862
7863 tg->se[cpu] = se;
354d60c2
DG
7864 /* se could be NULL for init_task_group */
7865 if (!se)
7866 return;
7867
ec7dc8ac
DG
7868 if (!parent)
7869 se->cfs_rq = &rq->cfs;
7870 else
7871 se->cfs_rq = parent->my_q;
7872
6f505b16
PZ
7873 se->my_q = cfs_rq;
7874 se->load.weight = tg->shares;
e05510d0 7875 se->load.inv_weight = 0;
ec7dc8ac 7876 se->parent = parent;
6f505b16 7877}
052f1dc7 7878#endif
6f505b16 7879
052f1dc7 7880#ifdef CONFIG_RT_GROUP_SCHED
ec7dc8ac
DG
7881static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq,
7882 struct sched_rt_entity *rt_se, int cpu, int add,
7883 struct sched_rt_entity *parent)
6f505b16 7884{
ec7dc8ac
DG
7885 struct rq *rq = cpu_rq(cpu);
7886
6f505b16
PZ
7887 tg->rt_rq[cpu] = rt_rq;
7888 init_rt_rq(rt_rq, rq);
7889 rt_rq->tg = tg;
ac086bc2 7890 rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime;
6f505b16
PZ
7891 if (add)
7892 list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
7893
7894 tg->rt_se[cpu] = rt_se;
354d60c2
DG
7895 if (!rt_se)
7896 return;
7897
ec7dc8ac
DG
7898 if (!parent)
7899 rt_se->rt_rq = &rq->rt;
7900 else
7901 rt_se->rt_rq = parent->my_q;
7902
6f505b16 7903 rt_se->my_q = rt_rq;
ec7dc8ac 7904 rt_se->parent = parent;
6f505b16
PZ
7905 INIT_LIST_HEAD(&rt_se->run_list);
7906}
7907#endif
7908
1da177e4
LT
7909void __init sched_init(void)
7910{
dd41f596 7911 int i, j;
434d53b0
MT
7912 unsigned long alloc_size = 0, ptr;
7913
7914#ifdef CONFIG_FAIR_GROUP_SCHED
7915 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
7916#endif
7917#ifdef CONFIG_RT_GROUP_SCHED
7918 alloc_size += 2 * nr_cpu_ids * sizeof(void **);
eff766a6 7919#endif
df7c8e84 7920#ifdef CONFIG_CPUMASK_OFFSTACK
8c083f08 7921 alloc_size += num_possible_cpus() * cpumask_size();
434d53b0 7922#endif
434d53b0 7923 if (alloc_size) {
36b7b6d4 7924 ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
434d53b0
MT
7925
7926#ifdef CONFIG_FAIR_GROUP_SCHED
7927 init_task_group.se = (struct sched_entity **)ptr;
7928 ptr += nr_cpu_ids * sizeof(void **);
7929
7930 init_task_group.cfs_rq = (struct cfs_rq **)ptr;
7931 ptr += nr_cpu_ids * sizeof(void **);
eff766a6 7932
6d6bc0ad 7933#endif /* CONFIG_FAIR_GROUP_SCHED */
434d53b0
MT
7934#ifdef CONFIG_RT_GROUP_SCHED
7935 init_task_group.rt_se = (struct sched_rt_entity **)ptr;
7936 ptr += nr_cpu_ids * sizeof(void **);
7937
7938 init_task_group.rt_rq = (struct rt_rq **)ptr;
eff766a6
PZ
7939 ptr += nr_cpu_ids * sizeof(void **);
7940
6d6bc0ad 7941#endif /* CONFIG_RT_GROUP_SCHED */
df7c8e84
RR
7942#ifdef CONFIG_CPUMASK_OFFSTACK
7943 for_each_possible_cpu(i) {
7944 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
7945 ptr += cpumask_size();
7946 }
7947#endif /* CONFIG_CPUMASK_OFFSTACK */
434d53b0 7948 }
dd41f596 7949
57d885fe
GH
7950#ifdef CONFIG_SMP
7951 init_defrootdomain();
7952#endif
7953
d0b27fa7
PZ
7954 init_rt_bandwidth(&def_rt_bandwidth,
7955 global_rt_period(), global_rt_runtime());
7956
7957#ifdef CONFIG_RT_GROUP_SCHED
7958 init_rt_bandwidth(&init_task_group.rt_bandwidth,
7959 global_rt_period(), global_rt_runtime());
6d6bc0ad 7960#endif /* CONFIG_RT_GROUP_SCHED */
d0b27fa7 7961
7c941438 7962#ifdef CONFIG_CGROUP_SCHED
6f505b16 7963 list_add(&init_task_group.list, &task_groups);
f473aa5e
PZ
7964 INIT_LIST_HEAD(&init_task_group.children);
7965
7c941438 7966#endif /* CONFIG_CGROUP_SCHED */
6f505b16 7967
4a6cc4bd
JK
7968#if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP
7969 update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long),
7970 __alignof__(unsigned long));
7971#endif
0a945022 7972 for_each_possible_cpu(i) {
70b97a7f 7973 struct rq *rq;
1da177e4
LT
7974
7975 rq = cpu_rq(i);
05fa785c 7976 raw_spin_lock_init(&rq->lock);
7897986b 7977 rq->nr_running = 0;
dce48a84
TG
7978 rq->calc_load_active = 0;
7979 rq->calc_load_update = jiffies + LOAD_FREQ;
dd41f596 7980 init_cfs_rq(&rq->cfs, rq);
6f505b16 7981 init_rt_rq(&rq->rt, rq);
dd41f596 7982#ifdef CONFIG_FAIR_GROUP_SCHED
4cf86d77 7983 init_task_group.shares = init_task_group_load;
6f505b16 7984 INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
354d60c2
DG
7985#ifdef CONFIG_CGROUP_SCHED
7986 /*
7987 * How much cpu bandwidth does init_task_group get?
7988 *
7989 * In case of task-groups formed thr' the cgroup filesystem, it
7990 * gets 100% of the cpu resources in the system. This overall
7991 * system cpu resource is divided among the tasks of
7992 * init_task_group and its child task-groups in a fair manner,
7993 * based on each entity's (task or task-group's) weight
7994 * (se->load.weight).
7995 *
7996 * In other words, if init_task_group has 10 tasks of weight
7997 * 1024) and two child groups A0 and A1 (of weight 1024 each),
7998 * then A0's share of the cpu resource is:
7999 *
0d905bca 8000 * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
354d60c2
DG
8001 *
8002 * We achieve this by letting init_task_group's tasks sit
8003 * directly in rq->cfs (i.e init_task_group->se[] = NULL).
8004 */
ec7dc8ac 8005 init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL);
052f1dc7 8006#endif
354d60c2
DG
8007#endif /* CONFIG_FAIR_GROUP_SCHED */
8008
8009 rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
052f1dc7 8010#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 8011 INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
354d60c2 8012#ifdef CONFIG_CGROUP_SCHED
ec7dc8ac 8013 init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL);
354d60c2 8014#endif
dd41f596 8015#endif
1da177e4 8016
dd41f596
IM
8017 for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
8018 rq->cpu_load[j] = 0;
fdf3e95d
VP
8019
8020 rq->last_load_update_tick = jiffies;
8021
1da177e4 8022#ifdef CONFIG_SMP
41c7ce9a 8023 rq->sd = NULL;
57d885fe 8024 rq->rd = NULL;
e51fd5e2 8025 rq->cpu_power = SCHED_LOAD_SCALE;
3f029d3c 8026 rq->post_schedule = 0;
1da177e4 8027 rq->active_balance = 0;
dd41f596 8028 rq->next_balance = jiffies;
1da177e4 8029 rq->push_cpu = 0;
0a2966b4 8030 rq->cpu = i;
1f11eb6a 8031 rq->online = 0;
eae0c9df
MG
8032 rq->idle_stamp = 0;
8033 rq->avg_idle = 2*sysctl_sched_migration_cost;
dc938520 8034 rq_attach_root(rq, &def_root_domain);
83cd4fe2
VP
8035#ifdef CONFIG_NO_HZ
8036 rq->nohz_balance_kick = 0;
8037 init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i));
8038#endif
1da177e4 8039#endif
8f4d37ec 8040 init_rq_hrtick(rq);
1da177e4 8041 atomic_set(&rq->nr_iowait, 0);
1da177e4
LT
8042 }
8043
2dd73a4f 8044 set_load_weight(&init_task);
b50f60ce 8045
e107be36
AK
8046#ifdef CONFIG_PREEMPT_NOTIFIERS
8047 INIT_HLIST_HEAD(&init_task.preempt_notifiers);
8048#endif
8049
c9819f45 8050#ifdef CONFIG_SMP
962cf36c 8051 open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);
c9819f45
CL
8052#endif
8053
b50f60ce 8054#ifdef CONFIG_RT_MUTEXES
1d615482 8055 plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock);
b50f60ce
HC
8056#endif
8057
1da177e4
LT
8058 /*
8059 * The boot idle thread does lazy MMU switching as well:
8060 */
8061 atomic_inc(&init_mm.mm_count);
8062 enter_lazy_tlb(&init_mm, current);
8063
8064 /*
8065 * Make us the idle thread. Technically, schedule() should not be
8066 * called from this thread, however somewhere below it might be,
8067 * but because we are the idle thread, we just pick up running again
8068 * when this runqueue becomes "idle".
8069 */
8070 init_idle(current, smp_processor_id());
dce48a84
TG
8071
8072 calc_load_update = jiffies + LOAD_FREQ;
8073
dd41f596
IM
8074 /*
8075 * During early bootup we pretend to be a normal task:
8076 */
8077 current->sched_class = &fair_sched_class;
6892b75e 8078
6a7b3dc3 8079 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
49557e62 8080 zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT);
bf4d83f6 8081#ifdef CONFIG_SMP
7d1e6a9b 8082#ifdef CONFIG_NO_HZ
83cd4fe2
VP
8083 zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
8084 alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT);
8085 atomic_set(&nohz.load_balancer, nr_cpu_ids);
8086 atomic_set(&nohz.first_pick_cpu, nr_cpu_ids);
8087 atomic_set(&nohz.second_pick_cpu, nr_cpu_ids);
7d1e6a9b 8088#endif
bdddd296
RR
8089 /* May be allocated at isolcpus cmdline parse time */
8090 if (cpu_isolated_map == NULL)
8091 zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
bf4d83f6 8092#endif /* SMP */
6a7b3dc3 8093
cdd6c482 8094 perf_event_init();
0d905bca 8095
6892b75e 8096 scheduler_running = 1;
1da177e4
LT
8097}
8098
8099#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
e4aafea2
FW
8100static inline int preempt_count_equals(int preempt_offset)
8101{
234da7bc 8102 int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
e4aafea2
FW
8103
8104 return (nested == PREEMPT_INATOMIC_BASE + preempt_offset);
8105}
8106
d894837f 8107void __might_sleep(const char *file, int line, int preempt_offset)
1da177e4 8108{
48f24c4d 8109#ifdef in_atomic
1da177e4
LT
8110 static unsigned long prev_jiffy; /* ratelimiting */
8111
e4aafea2
FW
8112 if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
8113 system_state != SYSTEM_RUNNING || oops_in_progress)
aef745fc
IM
8114 return;
8115 if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
8116 return;
8117 prev_jiffy = jiffies;
8118
3df0fc5b
PZ
8119 printk(KERN_ERR
8120 "BUG: sleeping function called from invalid context at %s:%d\n",
8121 file, line);
8122 printk(KERN_ERR
8123 "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
8124 in_atomic(), irqs_disabled(),
8125 current->pid, current->comm);
aef745fc
IM
8126
8127 debug_show_held_locks(current);
8128 if (irqs_disabled())
8129 print_irqtrace_events(current);
8130 dump_stack();
1da177e4
LT
8131#endif
8132}
8133EXPORT_SYMBOL(__might_sleep);
8134#endif
8135
8136#ifdef CONFIG_MAGIC_SYSRQ
3a5e4dc1
AK
8137static void normalize_task(struct rq *rq, struct task_struct *p)
8138{
8139 int on_rq;
3e51f33f 8140
3a5e4dc1
AK
8141 on_rq = p->se.on_rq;
8142 if (on_rq)
8143 deactivate_task(rq, p, 0);
8144 __setscheduler(rq, p, SCHED_NORMAL, 0);
8145 if (on_rq) {
8146 activate_task(rq, p, 0);
8147 resched_task(rq->curr);
8148 }
8149}
8150
1da177e4
LT
8151void normalize_rt_tasks(void)
8152{
a0f98a1c 8153 struct task_struct *g, *p;
1da177e4 8154 unsigned long flags;
70b97a7f 8155 struct rq *rq;
1da177e4 8156
4cf5d77a 8157 read_lock_irqsave(&tasklist_lock, flags);
a0f98a1c 8158 do_each_thread(g, p) {
178be793
IM
8159 /*
8160 * Only normalize user tasks:
8161 */
8162 if (!p->mm)
8163 continue;
8164
6cfb0d5d 8165 p->se.exec_start = 0;
6cfb0d5d 8166#ifdef CONFIG_SCHEDSTATS
41acab88
LDM
8167 p->se.statistics.wait_start = 0;
8168 p->se.statistics.sleep_start = 0;
8169 p->se.statistics.block_start = 0;
6cfb0d5d 8170#endif
dd41f596
IM
8171
8172 if (!rt_task(p)) {
8173 /*
8174 * Renice negative nice level userspace
8175 * tasks back to 0:
8176 */
8177 if (TASK_NICE(p) < 0 && p->mm)
8178 set_user_nice(p, 0);
1da177e4 8179 continue;
dd41f596 8180 }
1da177e4 8181
1d615482 8182 raw_spin_lock(&p->pi_lock);
b29739f9 8183 rq = __task_rq_lock(p);
1da177e4 8184
178be793 8185 normalize_task(rq, p);
3a5e4dc1 8186
b29739f9 8187 __task_rq_unlock(rq);
1d615482 8188 raw_spin_unlock(&p->pi_lock);
a0f98a1c
IM
8189 } while_each_thread(g, p);
8190
4cf5d77a 8191 read_unlock_irqrestore(&tasklist_lock, flags);
1da177e4
LT
8192}
8193
8194#endif /* CONFIG_MAGIC_SYSRQ */
1df5c10a 8195
67fc4e0c 8196#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
1df5c10a 8197/*
67fc4e0c 8198 * These functions are only useful for the IA64 MCA handling, or kdb.
1df5c10a
LT
8199 *
8200 * They can only be called when the whole system has been
8201 * stopped - every CPU needs to be quiescent, and no scheduling
8202 * activity can take place. Using them for anything else would
8203 * be a serious bug, and as a result, they aren't even visible
8204 * under any other configuration.
8205 */
8206
8207/**
8208 * curr_task - return the current task for a given cpu.
8209 * @cpu: the processor in question.
8210 *
8211 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8212 */
36c8b586 8213struct task_struct *curr_task(int cpu)
1df5c10a
LT
8214{
8215 return cpu_curr(cpu);
8216}
8217
67fc4e0c
JW
8218#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */
8219
8220#ifdef CONFIG_IA64
1df5c10a
LT
8221/**
8222 * set_curr_task - set the current task for a given cpu.
8223 * @cpu: the processor in question.
8224 * @p: the task pointer to set.
8225 *
8226 * Description: This function must only be used when non-maskable interrupts
41a2d6cf
IM
8227 * are serviced on a separate stack. It allows the architecture to switch the
8228 * notion of the current task on a cpu in a non-blocking manner. This function
1df5c10a
LT
8229 * must be called with all CPU's synchronized, and interrupts disabled, the
8230 * and caller must save the original value of the current task (see
8231 * curr_task() above) and restore that value before reenabling interrupts and
8232 * re-starting the system.
8233 *
8234 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
8235 */
36c8b586 8236void set_curr_task(int cpu, struct task_struct *p)
1df5c10a
LT
8237{
8238 cpu_curr(cpu) = p;
8239}
8240
8241#endif
29f59db3 8242
bccbe08a
PZ
8243#ifdef CONFIG_FAIR_GROUP_SCHED
8244static void free_fair_sched_group(struct task_group *tg)
6f505b16
PZ
8245{
8246 int i;
8247
8248 for_each_possible_cpu(i) {
8249 if (tg->cfs_rq)
8250 kfree(tg->cfs_rq[i]);
8251 if (tg->se)
8252 kfree(tg->se[i]);
6f505b16
PZ
8253 }
8254
8255 kfree(tg->cfs_rq);
8256 kfree(tg->se);
6f505b16
PZ
8257}
8258
ec7dc8ac
DG
8259static
8260int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
29f59db3 8261{
29f59db3 8262 struct cfs_rq *cfs_rq;
eab17229 8263 struct sched_entity *se;
9b5b7751 8264 struct rq *rq;
29f59db3
SV
8265 int i;
8266
434d53b0 8267 tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
29f59db3
SV
8268 if (!tg->cfs_rq)
8269 goto err;
434d53b0 8270 tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
29f59db3
SV
8271 if (!tg->se)
8272 goto err;
052f1dc7
PZ
8273
8274 tg->shares = NICE_0_LOAD;
29f59db3
SV
8275
8276 for_each_possible_cpu(i) {
9b5b7751 8277 rq = cpu_rq(i);
29f59db3 8278
eab17229
LZ
8279 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8280 GFP_KERNEL, cpu_to_node(i));
29f59db3
SV
8281 if (!cfs_rq)
8282 goto err;
8283
eab17229
LZ
8284 se = kzalloc_node(sizeof(struct sched_entity),
8285 GFP_KERNEL, cpu_to_node(i));
29f59db3 8286 if (!se)
dfc12eb2 8287 goto err_free_rq;
29f59db3 8288
eab17229 8289 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
bccbe08a
PZ
8290 }
8291
8292 return 1;
8293
49246274 8294err_free_rq:
dfc12eb2 8295 kfree(cfs_rq);
49246274 8296err:
bccbe08a
PZ
8297 return 0;
8298}
8299
8300static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8301{
8302 list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list,
8303 &cpu_rq(cpu)->leaf_cfs_rq_list);
8304}
8305
8306static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8307{
8308 list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list);
8309}
6d6bc0ad 8310#else /* !CONFG_FAIR_GROUP_SCHED */
bccbe08a
PZ
8311static inline void free_fair_sched_group(struct task_group *tg)
8312{
8313}
8314
ec7dc8ac
DG
8315static inline
8316int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8317{
8318 return 1;
8319}
8320
8321static inline void register_fair_sched_group(struct task_group *tg, int cpu)
8322{
8323}
8324
8325static inline void unregister_fair_sched_group(struct task_group *tg, int cpu)
8326{
8327}
6d6bc0ad 8328#endif /* CONFIG_FAIR_GROUP_SCHED */
052f1dc7
PZ
8329
8330#ifdef CONFIG_RT_GROUP_SCHED
bccbe08a
PZ
8331static void free_rt_sched_group(struct task_group *tg)
8332{
8333 int i;
8334
d0b27fa7
PZ
8335 destroy_rt_bandwidth(&tg->rt_bandwidth);
8336
bccbe08a
PZ
8337 for_each_possible_cpu(i) {
8338 if (tg->rt_rq)
8339 kfree(tg->rt_rq[i]);
8340 if (tg->rt_se)
8341 kfree(tg->rt_se[i]);
8342 }
8343
8344 kfree(tg->rt_rq);
8345 kfree(tg->rt_se);
8346}
8347
ec7dc8ac
DG
8348static
8349int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8350{
8351 struct rt_rq *rt_rq;
eab17229 8352 struct sched_rt_entity *rt_se;
bccbe08a
PZ
8353 struct rq *rq;
8354 int i;
8355
434d53b0 8356 tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL);
bccbe08a
PZ
8357 if (!tg->rt_rq)
8358 goto err;
434d53b0 8359 tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL);
bccbe08a
PZ
8360 if (!tg->rt_se)
8361 goto err;
8362
d0b27fa7
PZ
8363 init_rt_bandwidth(&tg->rt_bandwidth,
8364 ktime_to_ns(def_rt_bandwidth.rt_period), 0);
bccbe08a
PZ
8365
8366 for_each_possible_cpu(i) {
8367 rq = cpu_rq(i);
8368
eab17229
LZ
8369 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8370 GFP_KERNEL, cpu_to_node(i));
6f505b16
PZ
8371 if (!rt_rq)
8372 goto err;
29f59db3 8373
eab17229
LZ
8374 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8375 GFP_KERNEL, cpu_to_node(i));
6f505b16 8376 if (!rt_se)
dfc12eb2 8377 goto err_free_rq;
29f59db3 8378
eab17229 8379 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
29f59db3
SV
8380 }
8381
bccbe08a
PZ
8382 return 1;
8383
49246274 8384err_free_rq:
dfc12eb2 8385 kfree(rt_rq);
49246274 8386err:
bccbe08a
PZ
8387 return 0;
8388}
8389
8390static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8391{
8392 list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list,
8393 &cpu_rq(cpu)->leaf_rt_rq_list);
8394}
8395
8396static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8397{
8398 list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list);
8399}
6d6bc0ad 8400#else /* !CONFIG_RT_GROUP_SCHED */
bccbe08a
PZ
8401static inline void free_rt_sched_group(struct task_group *tg)
8402{
8403}
8404
ec7dc8ac
DG
8405static inline
8406int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
bccbe08a
PZ
8407{
8408 return 1;
8409}
8410
8411static inline void register_rt_sched_group(struct task_group *tg, int cpu)
8412{
8413}
8414
8415static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
8416{
8417}
6d6bc0ad 8418#endif /* CONFIG_RT_GROUP_SCHED */
bccbe08a 8419
7c941438 8420#ifdef CONFIG_CGROUP_SCHED
bccbe08a
PZ
8421static void free_sched_group(struct task_group *tg)
8422{
8423 free_fair_sched_group(tg);
8424 free_rt_sched_group(tg);
8425 kfree(tg);
8426}
8427
8428/* allocate runqueue etc for a new task group */
ec7dc8ac 8429struct task_group *sched_create_group(struct task_group *parent)
bccbe08a
PZ
8430{
8431 struct task_group *tg;
8432 unsigned long flags;
8433 int i;
8434
8435 tg = kzalloc(sizeof(*tg), GFP_KERNEL);
8436 if (!tg)
8437 return ERR_PTR(-ENOMEM);
8438
ec7dc8ac 8439 if (!alloc_fair_sched_group(tg, parent))
bccbe08a
PZ
8440 goto err;
8441
ec7dc8ac 8442 if (!alloc_rt_sched_group(tg, parent))
bccbe08a
PZ
8443 goto err;
8444
8ed36996 8445 spin_lock_irqsave(&task_group_lock, flags);
9b5b7751 8446 for_each_possible_cpu(i) {
bccbe08a
PZ
8447 register_fair_sched_group(tg, i);
8448 register_rt_sched_group(tg, i);
9b5b7751 8449 }
6f505b16 8450 list_add_rcu(&tg->list, &task_groups);
f473aa5e
PZ
8451
8452 WARN_ON(!parent); /* root should already exist */
8453
8454 tg->parent = parent;
f473aa5e 8455 INIT_LIST_HEAD(&tg->children);
09f2724a 8456 list_add_rcu(&tg->siblings, &parent->children);
8ed36996 8457 spin_unlock_irqrestore(&task_group_lock, flags);
29f59db3 8458
9b5b7751 8459 return tg;
29f59db3
SV
8460
8461err:
6f505b16 8462 free_sched_group(tg);
29f59db3
SV
8463 return ERR_PTR(-ENOMEM);
8464}
8465
9b5b7751 8466/* rcu callback to free various structures associated with a task group */
6f505b16 8467static void free_sched_group_rcu(struct rcu_head *rhp)
29f59db3 8468{
29f59db3 8469 /* now it should be safe to free those cfs_rqs */
6f505b16 8470 free_sched_group(container_of(rhp, struct task_group, rcu));
29f59db3
SV
8471}
8472
9b5b7751 8473/* Destroy runqueue etc associated with a task group */
4cf86d77 8474void sched_destroy_group(struct task_group *tg)
29f59db3 8475{
8ed36996 8476 unsigned long flags;
9b5b7751 8477 int i;
29f59db3 8478
8ed36996 8479 spin_lock_irqsave(&task_group_lock, flags);
9b5b7751 8480 for_each_possible_cpu(i) {
bccbe08a
PZ
8481 unregister_fair_sched_group(tg, i);
8482 unregister_rt_sched_group(tg, i);
9b5b7751 8483 }
6f505b16 8484 list_del_rcu(&tg->list);
f473aa5e 8485 list_del_rcu(&tg->siblings);
8ed36996 8486 spin_unlock_irqrestore(&task_group_lock, flags);
9b5b7751 8487
9b5b7751 8488 /* wait for possible concurrent references to cfs_rqs complete */
6f505b16 8489 call_rcu(&tg->rcu, free_sched_group_rcu);
29f59db3
SV
8490}
8491
9b5b7751 8492/* change task's runqueue when it moves between groups.
3a252015
IM
8493 * The caller of this function should have put the task in its new group
8494 * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
8495 * reflect its new group.
9b5b7751
SV
8496 */
8497void sched_move_task(struct task_struct *tsk)
29f59db3
SV
8498{
8499 int on_rq, running;
8500 unsigned long flags;
8501 struct rq *rq;
8502
8503 rq = task_rq_lock(tsk, &flags);
8504
051a1d1a 8505 running = task_current(rq, tsk);
29f59db3
SV
8506 on_rq = tsk->se.on_rq;
8507
0e1f3483 8508 if (on_rq)
29f59db3 8509 dequeue_task(rq, tsk, 0);
0e1f3483
HS
8510 if (unlikely(running))
8511 tsk->sched_class->put_prev_task(rq, tsk);
29f59db3 8512
6f505b16 8513 set_task_rq(tsk, task_cpu(tsk));
29f59db3 8514
810b3817
PZ
8515#ifdef CONFIG_FAIR_GROUP_SCHED
8516 if (tsk->sched_class->moved_group)
88ec22d3 8517 tsk->sched_class->moved_group(tsk, on_rq);
810b3817
PZ
8518#endif
8519
0e1f3483
HS
8520 if (unlikely(running))
8521 tsk->sched_class->set_curr_task(rq);
8522 if (on_rq)
371fd7e7 8523 enqueue_task(rq, tsk, 0);
29f59db3 8524
29f59db3
SV
8525 task_rq_unlock(rq, &flags);
8526}
7c941438 8527#endif /* CONFIG_CGROUP_SCHED */
29f59db3 8528
052f1dc7 8529#ifdef CONFIG_FAIR_GROUP_SCHED
c09595f6 8530static void __set_se_shares(struct sched_entity *se, unsigned long shares)
29f59db3
SV
8531{
8532 struct cfs_rq *cfs_rq = se->cfs_rq;
29f59db3
SV
8533 int on_rq;
8534
29f59db3 8535 on_rq = se->on_rq;
62fb1851 8536 if (on_rq)
29f59db3
SV
8537 dequeue_entity(cfs_rq, se, 0);
8538
8539 se->load.weight = shares;
e05510d0 8540 se->load.inv_weight = 0;
29f59db3 8541
62fb1851 8542 if (on_rq)
29f59db3 8543 enqueue_entity(cfs_rq, se, 0);
c09595f6 8544}
62fb1851 8545
c09595f6
PZ
8546static void set_se_shares(struct sched_entity *se, unsigned long shares)
8547{
8548 struct cfs_rq *cfs_rq = se->cfs_rq;
8549 struct rq *rq = cfs_rq->rq;
8550 unsigned long flags;
8551
05fa785c 8552 raw_spin_lock_irqsave(&rq->lock, flags);
c09595f6 8553 __set_se_shares(se, shares);
05fa785c 8554 raw_spin_unlock_irqrestore(&rq->lock, flags);
29f59db3
SV
8555}
8556
8ed36996
PZ
8557static DEFINE_MUTEX(shares_mutex);
8558
4cf86d77 8559int sched_group_set_shares(struct task_group *tg, unsigned long shares)
29f59db3
SV
8560{
8561 int i;
8ed36996 8562 unsigned long flags;
c61935fd 8563
ec7dc8ac
DG
8564 /*
8565 * We can't change the weight of the root cgroup.
8566 */
8567 if (!tg->se[0])
8568 return -EINVAL;
8569
18d95a28
PZ
8570 if (shares < MIN_SHARES)
8571 shares = MIN_SHARES;
cb4ad1ff
MX
8572 else if (shares > MAX_SHARES)
8573 shares = MAX_SHARES;
62fb1851 8574
8ed36996 8575 mutex_lock(&shares_mutex);
9b5b7751 8576 if (tg->shares == shares)
5cb350ba 8577 goto done;
29f59db3 8578
8ed36996 8579 spin_lock_irqsave(&task_group_lock, flags);
bccbe08a
PZ
8580 for_each_possible_cpu(i)
8581 unregister_fair_sched_group(tg, i);
f473aa5e 8582 list_del_rcu(&tg->siblings);
8ed36996 8583 spin_unlock_irqrestore(&task_group_lock, flags);
6b2d7700
SV
8584
8585 /* wait for any ongoing reference to this group to finish */
8586 synchronize_sched();
8587
8588 /*
8589 * Now we are free to modify the group's share on each cpu
8590 * w/o tripping rebalance_share or load_balance_fair.
8591 */
9b5b7751 8592 tg->shares = shares;
c09595f6
PZ
8593 for_each_possible_cpu(i) {
8594 /*
8595 * force a rebalance
8596 */
8597 cfs_rq_set_shares(tg->cfs_rq[i], 0);
cb4ad1ff 8598 set_se_shares(tg->se[i], shares);
c09595f6 8599 }
29f59db3 8600
6b2d7700
SV
8601 /*
8602 * Enable load balance activity on this group, by inserting it back on
8603 * each cpu's rq->leaf_cfs_rq_list.
8604 */
8ed36996 8605 spin_lock_irqsave(&task_group_lock, flags);
bccbe08a
PZ
8606 for_each_possible_cpu(i)
8607 register_fair_sched_group(tg, i);
f473aa5e 8608 list_add_rcu(&tg->siblings, &tg->parent->children);
8ed36996 8609 spin_unlock_irqrestore(&task_group_lock, flags);
5cb350ba 8610done:
8ed36996 8611 mutex_unlock(&shares_mutex);
9b5b7751 8612 return 0;
29f59db3
SV
8613}
8614
5cb350ba
DG
8615unsigned long sched_group_shares(struct task_group *tg)
8616{
8617 return tg->shares;
8618}
052f1dc7 8619#endif
5cb350ba 8620
052f1dc7 8621#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 8622/*
9f0c1e56 8623 * Ensure that the real time constraints are schedulable.
6f505b16 8624 */
9f0c1e56
PZ
8625static DEFINE_MUTEX(rt_constraints_mutex);
8626
8627static unsigned long to_ratio(u64 period, u64 runtime)
8628{
8629 if (runtime == RUNTIME_INF)
9a7e0b18 8630 return 1ULL << 20;
9f0c1e56 8631
9a7e0b18 8632 return div64_u64(runtime << 20, period);
9f0c1e56
PZ
8633}
8634
9a7e0b18
PZ
8635/* Must be called with tasklist_lock held */
8636static inline int tg_has_rt_tasks(struct task_group *tg)
b40b2e8e 8637{
9a7e0b18 8638 struct task_struct *g, *p;
b40b2e8e 8639
9a7e0b18
PZ
8640 do_each_thread(g, p) {
8641 if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg)
8642 return 1;
8643 } while_each_thread(g, p);
b40b2e8e 8644
9a7e0b18
PZ
8645 return 0;
8646}
b40b2e8e 8647
9a7e0b18
PZ
8648struct rt_schedulable_data {
8649 struct task_group *tg;
8650 u64 rt_period;
8651 u64 rt_runtime;
8652};
b40b2e8e 8653
9a7e0b18
PZ
8654static int tg_schedulable(struct task_group *tg, void *data)
8655{
8656 struct rt_schedulable_data *d = data;
8657 struct task_group *child;
8658 unsigned long total, sum = 0;
8659 u64 period, runtime;
b40b2e8e 8660
9a7e0b18
PZ
8661 period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8662 runtime = tg->rt_bandwidth.rt_runtime;
b40b2e8e 8663
9a7e0b18
PZ
8664 if (tg == d->tg) {
8665 period = d->rt_period;
8666 runtime = d->rt_runtime;
b40b2e8e 8667 }
b40b2e8e 8668
4653f803
PZ
8669 /*
8670 * Cannot have more runtime than the period.
8671 */
8672 if (runtime > period && runtime != RUNTIME_INF)
8673 return -EINVAL;
6f505b16 8674
4653f803
PZ
8675 /*
8676 * Ensure we don't starve existing RT tasks.
8677 */
9a7e0b18
PZ
8678 if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
8679 return -EBUSY;
6f505b16 8680
9a7e0b18 8681 total = to_ratio(period, runtime);
6f505b16 8682
4653f803
PZ
8683 /*
8684 * Nobody can have more than the global setting allows.
8685 */
8686 if (total > to_ratio(global_rt_period(), global_rt_runtime()))
8687 return -EINVAL;
6f505b16 8688
4653f803
PZ
8689 /*
8690 * The sum of our children's runtime should not exceed our own.
8691 */
9a7e0b18
PZ
8692 list_for_each_entry_rcu(child, &tg->children, siblings) {
8693 period = ktime_to_ns(child->rt_bandwidth.rt_period);
8694 runtime = child->rt_bandwidth.rt_runtime;
6f505b16 8695
9a7e0b18
PZ
8696 if (child == d->tg) {
8697 period = d->rt_period;
8698 runtime = d->rt_runtime;
8699 }
6f505b16 8700
9a7e0b18 8701 sum += to_ratio(period, runtime);
9f0c1e56 8702 }
6f505b16 8703
9a7e0b18
PZ
8704 if (sum > total)
8705 return -EINVAL;
8706
8707 return 0;
6f505b16
PZ
8708}
8709
9a7e0b18 8710static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
521f1a24 8711{
9a7e0b18
PZ
8712 struct rt_schedulable_data data = {
8713 .tg = tg,
8714 .rt_period = period,
8715 .rt_runtime = runtime,
8716 };
8717
8718 return walk_tg_tree(tg_schedulable, tg_nop, &data);
521f1a24
DG
8719}
8720
d0b27fa7
PZ
8721static int tg_set_bandwidth(struct task_group *tg,
8722 u64 rt_period, u64 rt_runtime)
6f505b16 8723{
ac086bc2 8724 int i, err = 0;
9f0c1e56 8725
9f0c1e56 8726 mutex_lock(&rt_constraints_mutex);
521f1a24 8727 read_lock(&tasklist_lock);
9a7e0b18
PZ
8728 err = __rt_schedulable(tg, rt_period, rt_runtime);
8729 if (err)
9f0c1e56 8730 goto unlock;
ac086bc2 8731
0986b11b 8732 raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
d0b27fa7
PZ
8733 tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
8734 tg->rt_bandwidth.rt_runtime = rt_runtime;
ac086bc2
PZ
8735
8736 for_each_possible_cpu(i) {
8737 struct rt_rq *rt_rq = tg->rt_rq[i];
8738
0986b11b 8739 raw_spin_lock(&rt_rq->rt_runtime_lock);
ac086bc2 8740 rt_rq->rt_runtime = rt_runtime;
0986b11b 8741 raw_spin_unlock(&rt_rq->rt_runtime_lock);
ac086bc2 8742 }
0986b11b 8743 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
49246274 8744unlock:
521f1a24 8745 read_unlock(&tasklist_lock);
9f0c1e56
PZ
8746 mutex_unlock(&rt_constraints_mutex);
8747
8748 return err;
6f505b16
PZ
8749}
8750
d0b27fa7
PZ
8751int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8752{
8753 u64 rt_runtime, rt_period;
8754
8755 rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
8756 rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
8757 if (rt_runtime_us < 0)
8758 rt_runtime = RUNTIME_INF;
8759
8760 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8761}
8762
9f0c1e56
PZ
8763long sched_group_rt_runtime(struct task_group *tg)
8764{
8765 u64 rt_runtime_us;
8766
d0b27fa7 8767 if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
9f0c1e56
PZ
8768 return -1;
8769
d0b27fa7 8770 rt_runtime_us = tg->rt_bandwidth.rt_runtime;
9f0c1e56
PZ
8771 do_div(rt_runtime_us, NSEC_PER_USEC);
8772 return rt_runtime_us;
8773}
d0b27fa7
PZ
8774
8775int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
8776{
8777 u64 rt_runtime, rt_period;
8778
8779 rt_period = (u64)rt_period_us * NSEC_PER_USEC;
8780 rt_runtime = tg->rt_bandwidth.rt_runtime;
8781
619b0488
R
8782 if (rt_period == 0)
8783 return -EINVAL;
8784
d0b27fa7
PZ
8785 return tg_set_bandwidth(tg, rt_period, rt_runtime);
8786}
8787
8788long sched_group_rt_period(struct task_group *tg)
8789{
8790 u64 rt_period_us;
8791
8792 rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
8793 do_div(rt_period_us, NSEC_PER_USEC);
8794 return rt_period_us;
8795}
8796
8797static int sched_rt_global_constraints(void)
8798{
4653f803 8799 u64 runtime, period;
d0b27fa7
PZ
8800 int ret = 0;
8801
ec5d4989
HS
8802 if (sysctl_sched_rt_period <= 0)
8803 return -EINVAL;
8804
4653f803
PZ
8805 runtime = global_rt_runtime();
8806 period = global_rt_period();
8807
8808 /*
8809 * Sanity check on the sysctl variables.
8810 */
8811 if (runtime > period && runtime != RUNTIME_INF)
8812 return -EINVAL;
10b612f4 8813
d0b27fa7 8814 mutex_lock(&rt_constraints_mutex);
9a7e0b18 8815 read_lock(&tasklist_lock);
4653f803 8816 ret = __rt_schedulable(NULL, 0, 0);
9a7e0b18 8817 read_unlock(&tasklist_lock);
d0b27fa7
PZ
8818 mutex_unlock(&rt_constraints_mutex);
8819
8820 return ret;
8821}
54e99124
DG
8822
8823int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8824{
8825 /* Don't accept realtime tasks when there is no way for them to run */
8826 if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
8827 return 0;
8828
8829 return 1;
8830}
8831
6d6bc0ad 8832#else /* !CONFIG_RT_GROUP_SCHED */
d0b27fa7
PZ
8833static int sched_rt_global_constraints(void)
8834{
ac086bc2
PZ
8835 unsigned long flags;
8836 int i;
8837
ec5d4989
HS
8838 if (sysctl_sched_rt_period <= 0)
8839 return -EINVAL;
8840
60aa605d
PZ
8841 /*
8842 * There's always some RT tasks in the root group
8843 * -- migration, kstopmachine etc..
8844 */
8845 if (sysctl_sched_rt_runtime == 0)
8846 return -EBUSY;
8847
0986b11b 8848 raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
ac086bc2
PZ
8849 for_each_possible_cpu(i) {
8850 struct rt_rq *rt_rq = &cpu_rq(i)->rt;
8851
0986b11b 8852 raw_spin_lock(&rt_rq->rt_runtime_lock);
ac086bc2 8853 rt_rq->rt_runtime = global_rt_runtime();
0986b11b 8854 raw_spin_unlock(&rt_rq->rt_runtime_lock);
ac086bc2 8855 }
0986b11b 8856 raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
ac086bc2 8857
d0b27fa7
PZ
8858 return 0;
8859}
6d6bc0ad 8860#endif /* CONFIG_RT_GROUP_SCHED */
d0b27fa7
PZ
8861
8862int sched_rt_handler(struct ctl_table *table, int write,
8d65af78 8863 void __user *buffer, size_t *lenp,
d0b27fa7
PZ
8864 loff_t *ppos)
8865{
8866 int ret;
8867 int old_period, old_runtime;
8868 static DEFINE_MUTEX(mutex);
8869
8870 mutex_lock(&mutex);
8871 old_period = sysctl_sched_rt_period;
8872 old_runtime = sysctl_sched_rt_runtime;
8873
8d65af78 8874 ret = proc_dointvec(table, write, buffer, lenp, ppos);
d0b27fa7
PZ
8875
8876 if (!ret && write) {
8877 ret = sched_rt_global_constraints();
8878 if (ret) {
8879 sysctl_sched_rt_period = old_period;
8880 sysctl_sched_rt_runtime = old_runtime;
8881 } else {
8882 def_rt_bandwidth.rt_runtime = global_rt_runtime();
8883 def_rt_bandwidth.rt_period =
8884 ns_to_ktime(global_rt_period());
8885 }
8886 }
8887 mutex_unlock(&mutex);
8888
8889 return ret;
8890}
68318b8e 8891
052f1dc7 8892#ifdef CONFIG_CGROUP_SCHED
68318b8e
SV
8893
8894/* return corresponding task_group object of a cgroup */
2b01dfe3 8895static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
68318b8e 8896{
2b01dfe3
PM
8897 return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id),
8898 struct task_group, css);
68318b8e
SV
8899}
8900
8901static struct cgroup_subsys_state *
2b01dfe3 8902cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
68318b8e 8903{
ec7dc8ac 8904 struct task_group *tg, *parent;
68318b8e 8905
2b01dfe3 8906 if (!cgrp->parent) {
68318b8e 8907 /* This is early initialization for the top cgroup */
68318b8e
SV
8908 return &init_task_group.css;
8909 }
8910
ec7dc8ac
DG
8911 parent = cgroup_tg(cgrp->parent);
8912 tg = sched_create_group(parent);
68318b8e
SV
8913 if (IS_ERR(tg))
8914 return ERR_PTR(-ENOMEM);
8915
68318b8e
SV
8916 return &tg->css;
8917}
8918
41a2d6cf
IM
8919static void
8920cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
68318b8e 8921{
2b01dfe3 8922 struct task_group *tg = cgroup_tg(cgrp);
68318b8e
SV
8923
8924 sched_destroy_group(tg);
8925}
8926
41a2d6cf 8927static int
be367d09 8928cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
68318b8e 8929{
b68aa230 8930#ifdef CONFIG_RT_GROUP_SCHED
54e99124 8931 if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk))
b68aa230
PZ
8932 return -EINVAL;
8933#else
68318b8e
SV
8934 /* We don't support RT-tasks being in separate groups */
8935 if (tsk->sched_class != &fair_sched_class)
8936 return -EINVAL;
b68aa230 8937#endif
be367d09
BB
8938 return 0;
8939}
68318b8e 8940
be367d09
BB
8941static int
8942cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
8943 struct task_struct *tsk, bool threadgroup)
8944{
8945 int retval = cpu_cgroup_can_attach_task(cgrp, tsk);
8946 if (retval)
8947 return retval;
8948 if (threadgroup) {
8949 struct task_struct *c;
8950 rcu_read_lock();
8951 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8952 retval = cpu_cgroup_can_attach_task(cgrp, c);
8953 if (retval) {
8954 rcu_read_unlock();
8955 return retval;
8956 }
8957 }
8958 rcu_read_unlock();
8959 }
68318b8e
SV
8960 return 0;
8961}
8962
8963static void
2b01dfe3 8964cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
be367d09
BB
8965 struct cgroup *old_cont, struct task_struct *tsk,
8966 bool threadgroup)
68318b8e
SV
8967{
8968 sched_move_task(tsk);
be367d09
BB
8969 if (threadgroup) {
8970 struct task_struct *c;
8971 rcu_read_lock();
8972 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
8973 sched_move_task(c);
8974 }
8975 rcu_read_unlock();
8976 }
68318b8e
SV
8977}
8978
052f1dc7 8979#ifdef CONFIG_FAIR_GROUP_SCHED
f4c753b7 8980static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
2b01dfe3 8981 u64 shareval)
68318b8e 8982{
2b01dfe3 8983 return sched_group_set_shares(cgroup_tg(cgrp), shareval);
68318b8e
SV
8984}
8985
f4c753b7 8986static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
68318b8e 8987{
2b01dfe3 8988 struct task_group *tg = cgroup_tg(cgrp);
68318b8e
SV
8989
8990 return (u64) tg->shares;
8991}
6d6bc0ad 8992#endif /* CONFIG_FAIR_GROUP_SCHED */
68318b8e 8993
052f1dc7 8994#ifdef CONFIG_RT_GROUP_SCHED
0c70814c 8995static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
06ecb27c 8996 s64 val)
6f505b16 8997{
06ecb27c 8998 return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
6f505b16
PZ
8999}
9000
06ecb27c 9001static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
6f505b16 9002{
06ecb27c 9003 return sched_group_rt_runtime(cgroup_tg(cgrp));
6f505b16 9004}
d0b27fa7
PZ
9005
9006static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
9007 u64 rt_period_us)
9008{
9009 return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
9010}
9011
9012static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
9013{
9014 return sched_group_rt_period(cgroup_tg(cgrp));
9015}
6d6bc0ad 9016#endif /* CONFIG_RT_GROUP_SCHED */
6f505b16 9017
fe5c7cc2 9018static struct cftype cpu_files[] = {
052f1dc7 9019#ifdef CONFIG_FAIR_GROUP_SCHED
fe5c7cc2
PM
9020 {
9021 .name = "shares",
f4c753b7
PM
9022 .read_u64 = cpu_shares_read_u64,
9023 .write_u64 = cpu_shares_write_u64,
fe5c7cc2 9024 },
052f1dc7
PZ
9025#endif
9026#ifdef CONFIG_RT_GROUP_SCHED
6f505b16 9027 {
9f0c1e56 9028 .name = "rt_runtime_us",
06ecb27c
PM
9029 .read_s64 = cpu_rt_runtime_read,
9030 .write_s64 = cpu_rt_runtime_write,
6f505b16 9031 },
d0b27fa7
PZ
9032 {
9033 .name = "rt_period_us",
f4c753b7
PM
9034 .read_u64 = cpu_rt_period_read_uint,
9035 .write_u64 = cpu_rt_period_write_uint,
d0b27fa7 9036 },
052f1dc7 9037#endif
68318b8e
SV
9038};
9039
9040static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
9041{
fe5c7cc2 9042 return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files));
68318b8e
SV
9043}
9044
9045struct cgroup_subsys cpu_cgroup_subsys = {
38605cae
IM
9046 .name = "cpu",
9047 .create = cpu_cgroup_create,
9048 .destroy = cpu_cgroup_destroy,
9049 .can_attach = cpu_cgroup_can_attach,
9050 .attach = cpu_cgroup_attach,
9051 .populate = cpu_cgroup_populate,
9052 .subsys_id = cpu_cgroup_subsys_id,
68318b8e
SV
9053 .early_init = 1,
9054};
9055
052f1dc7 9056#endif /* CONFIG_CGROUP_SCHED */
d842de87
SV
9057
9058#ifdef CONFIG_CGROUP_CPUACCT
9059
9060/*
9061 * CPU accounting code for task groups.
9062 *
9063 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
9064 * (balbir@in.ibm.com).
9065 */
9066
934352f2 9067/* track cpu usage of a group of tasks and its child groups */
d842de87
SV
9068struct cpuacct {
9069 struct cgroup_subsys_state css;
9070 /* cpuusage holds pointer to a u64-type object on every cpu */
43cf38eb 9071 u64 __percpu *cpuusage;
ef12fefa 9072 struct percpu_counter cpustat[CPUACCT_STAT_NSTATS];
934352f2 9073 struct cpuacct *parent;
d842de87
SV
9074};
9075
9076struct cgroup_subsys cpuacct_subsys;
9077
9078/* return cpu accounting group corresponding to this container */
32cd756a 9079static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
d842de87 9080{
32cd756a 9081 return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id),
d842de87
SV
9082 struct cpuacct, css);
9083}
9084
9085/* return cpu accounting group to which this task belongs */
9086static inline struct cpuacct *task_ca(struct task_struct *tsk)
9087{
9088 return container_of(task_subsys_state(tsk, cpuacct_subsys_id),
9089 struct cpuacct, css);
9090}
9091
9092/* create a new cpu accounting group */
9093static struct cgroup_subsys_state *cpuacct_create(
32cd756a 9094 struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87
SV
9095{
9096 struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL);
ef12fefa 9097 int i;
d842de87
SV
9098
9099 if (!ca)
ef12fefa 9100 goto out;
d842de87
SV
9101
9102 ca->cpuusage = alloc_percpu(u64);
ef12fefa
BR
9103 if (!ca->cpuusage)
9104 goto out_free_ca;
9105
9106 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
9107 if (percpu_counter_init(&ca->cpustat[i], 0))
9108 goto out_free_counters;
d842de87 9109
934352f2
BR
9110 if (cgrp->parent)
9111 ca->parent = cgroup_ca(cgrp->parent);
9112
d842de87 9113 return &ca->css;
ef12fefa
BR
9114
9115out_free_counters:
9116 while (--i >= 0)
9117 percpu_counter_destroy(&ca->cpustat[i]);
9118 free_percpu(ca->cpuusage);
9119out_free_ca:
9120 kfree(ca);
9121out:
9122 return ERR_PTR(-ENOMEM);
d842de87
SV
9123}
9124
9125/* destroy an existing cpu accounting group */
41a2d6cf 9126static void
32cd756a 9127cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87 9128{
32cd756a 9129 struct cpuacct *ca = cgroup_ca(cgrp);
ef12fefa 9130 int i;
d842de87 9131
ef12fefa
BR
9132 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
9133 percpu_counter_destroy(&ca->cpustat[i]);
d842de87
SV
9134 free_percpu(ca->cpuusage);
9135 kfree(ca);
9136}
9137
720f5498
KC
9138static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
9139{
b36128c8 9140 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
720f5498
KC
9141 u64 data;
9142
9143#ifndef CONFIG_64BIT
9144 /*
9145 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
9146 */
05fa785c 9147 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
720f5498 9148 data = *cpuusage;
05fa785c 9149 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
720f5498
KC
9150#else
9151 data = *cpuusage;
9152#endif
9153
9154 return data;
9155}
9156
9157static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
9158{
b36128c8 9159 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
720f5498
KC
9160
9161#ifndef CONFIG_64BIT
9162 /*
9163 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
9164 */
05fa785c 9165 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
720f5498 9166 *cpuusage = val;
05fa785c 9167 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
720f5498
KC
9168#else
9169 *cpuusage = val;
9170#endif
9171}
9172
d842de87 9173/* return total cpu usage (in nanoseconds) of a group */
32cd756a 9174static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
d842de87 9175{
32cd756a 9176 struct cpuacct *ca = cgroup_ca(cgrp);
d842de87
SV
9177 u64 totalcpuusage = 0;
9178 int i;
9179
720f5498
KC
9180 for_each_present_cpu(i)
9181 totalcpuusage += cpuacct_cpuusage_read(ca, i);
d842de87
SV
9182
9183 return totalcpuusage;
9184}
9185
0297b803
DG
9186static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
9187 u64 reset)
9188{
9189 struct cpuacct *ca = cgroup_ca(cgrp);
9190 int err = 0;
9191 int i;
9192
9193 if (reset) {
9194 err = -EINVAL;
9195 goto out;
9196 }
9197
720f5498
KC
9198 for_each_present_cpu(i)
9199 cpuacct_cpuusage_write(ca, i, 0);
0297b803 9200
0297b803
DG
9201out:
9202 return err;
9203}
9204
e9515c3c
KC
9205static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
9206 struct seq_file *m)
9207{
9208 struct cpuacct *ca = cgroup_ca(cgroup);
9209 u64 percpu;
9210 int i;
9211
9212 for_each_present_cpu(i) {
9213 percpu = cpuacct_cpuusage_read(ca, i);
9214 seq_printf(m, "%llu ", (unsigned long long) percpu);
9215 }
9216 seq_printf(m, "\n");
9217 return 0;
9218}
9219
ef12fefa
BR
9220static const char *cpuacct_stat_desc[] = {
9221 [CPUACCT_STAT_USER] = "user",
9222 [CPUACCT_STAT_SYSTEM] = "system",
9223};
9224
9225static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft,
9226 struct cgroup_map_cb *cb)
9227{
9228 struct cpuacct *ca = cgroup_ca(cgrp);
9229 int i;
9230
9231 for (i = 0; i < CPUACCT_STAT_NSTATS; i++) {
9232 s64 val = percpu_counter_read(&ca->cpustat[i]);
9233 val = cputime64_to_clock_t(val);
9234 cb->fill(cb, cpuacct_stat_desc[i], val);
9235 }
9236 return 0;
9237}
9238
d842de87
SV
9239static struct cftype files[] = {
9240 {
9241 .name = "usage",
f4c753b7
PM
9242 .read_u64 = cpuusage_read,
9243 .write_u64 = cpuusage_write,
d842de87 9244 },
e9515c3c
KC
9245 {
9246 .name = "usage_percpu",
9247 .read_seq_string = cpuacct_percpu_seq_read,
9248 },
ef12fefa
BR
9249 {
9250 .name = "stat",
9251 .read_map = cpuacct_stats_show,
9252 },
d842de87
SV
9253};
9254
32cd756a 9255static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
d842de87 9256{
32cd756a 9257 return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files));
d842de87
SV
9258}
9259
9260/*
9261 * charge this task's execution time to its accounting group.
9262 *
9263 * called with rq->lock held.
9264 */
9265static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9266{
9267 struct cpuacct *ca;
934352f2 9268 int cpu;
d842de87 9269
c40c6f85 9270 if (unlikely(!cpuacct_subsys.active))
d842de87
SV
9271 return;
9272
934352f2 9273 cpu = task_cpu(tsk);
a18b83b7
BR
9274
9275 rcu_read_lock();
9276
d842de87 9277 ca = task_ca(tsk);
d842de87 9278
934352f2 9279 for (; ca; ca = ca->parent) {
b36128c8 9280 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
d842de87
SV
9281 *cpuusage += cputime;
9282 }
a18b83b7
BR
9283
9284 rcu_read_unlock();
d842de87
SV
9285}
9286
fa535a77
AB
9287/*
9288 * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large
9289 * in cputime_t units. As a result, cpuacct_update_stats calls
9290 * percpu_counter_add with values large enough to always overflow the
9291 * per cpu batch limit causing bad SMP scalability.
9292 *
9293 * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we
9294 * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled
9295 * and enabled. We cap it at INT_MAX which is the largest allowed batch value.
9296 */
9297#ifdef CONFIG_SMP
9298#define CPUACCT_BATCH \
9299 min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX)
9300#else
9301#define CPUACCT_BATCH 0
9302#endif
9303
ef12fefa
BR
9304/*
9305 * Charge the system/user time to the task's accounting group.
9306 */
9307static void cpuacct_update_stats(struct task_struct *tsk,
9308 enum cpuacct_stat_index idx, cputime_t val)
9309{
9310 struct cpuacct *ca;
fa535a77 9311 int batch = CPUACCT_BATCH;
ef12fefa
BR
9312
9313 if (unlikely(!cpuacct_subsys.active))
9314 return;
9315
9316 rcu_read_lock();
9317 ca = task_ca(tsk);
9318
9319 do {
fa535a77 9320 __percpu_counter_add(&ca->cpustat[idx], val, batch);
ef12fefa
BR
9321 ca = ca->parent;
9322 } while (ca);
9323 rcu_read_unlock();
9324}
9325
d842de87
SV
9326struct cgroup_subsys cpuacct_subsys = {
9327 .name = "cpuacct",
9328 .create = cpuacct_create,
9329 .destroy = cpuacct_destroy,
9330 .populate = cpuacct_populate,
9331 .subsys_id = cpuacct_subsys_id,
9332};
9333#endif /* CONFIG_CGROUP_CPUACCT */
03b042bf
PM
9334
9335#ifndef CONFIG_SMP
9336
03b042bf
PM
9337void synchronize_sched_expedited(void)
9338{
fc390cde 9339 barrier();
03b042bf
PM
9340}
9341EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9342
9343#else /* #ifndef CONFIG_SMP */
9344
cc631fb7 9345static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0);
03b042bf 9346
cc631fb7 9347static int synchronize_sched_expedited_cpu_stop(void *data)
03b042bf 9348{
969c7921
TH
9349 /*
9350 * There must be a full memory barrier on each affected CPU
9351 * between the time that try_stop_cpus() is called and the
9352 * time that it returns.
9353 *
9354 * In the current initial implementation of cpu_stop, the
9355 * above condition is already met when the control reaches
9356 * this point and the following smp_mb() is not strictly
9357 * necessary. Do smp_mb() anyway for documentation and
9358 * robustness against future implementation changes.
9359 */
cc631fb7 9360 smp_mb(); /* See above comment block. */
969c7921 9361 return 0;
03b042bf 9362}
03b042bf
PM
9363
9364/*
9365 * Wait for an rcu-sched grace period to elapse, but use "big hammer"
9366 * approach to force grace period to end quickly. This consumes
9367 * significant time on all CPUs, and is thus not recommended for
9368 * any sort of common-case code.
9369 *
9370 * Note that it is illegal to call this function while holding any
9371 * lock that is acquired by a CPU-hotplug notifier. Failing to
9372 * observe this restriction will result in deadlock.
9373 */
9374void synchronize_sched_expedited(void)
9375{
969c7921 9376 int snap, trycount = 0;
03b042bf
PM
9377
9378 smp_mb(); /* ensure prior mod happens before capturing snap. */
969c7921 9379 snap = atomic_read(&synchronize_sched_expedited_count) + 1;
03b042bf 9380 get_online_cpus();
969c7921
TH
9381 while (try_stop_cpus(cpu_online_mask,
9382 synchronize_sched_expedited_cpu_stop,
94458d5e 9383 NULL) == -EAGAIN) {
03b042bf
PM
9384 put_online_cpus();
9385 if (trycount++ < 10)
9386 udelay(trycount * num_online_cpus());
9387 else {
9388 synchronize_sched();
9389 return;
9390 }
969c7921 9391 if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) {
03b042bf
PM
9392 smp_mb(); /* ensure test happens before caller kfree */
9393 return;
9394 }
9395 get_online_cpus();
9396 }
969c7921 9397 atomic_inc(&synchronize_sched_expedited_count);
cc631fb7 9398 smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */
03b042bf 9399 put_online_cpus();
03b042bf
PM
9400}
9401EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
9402
9403#endif /* #else #ifndef CONFIG_SMP */