]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
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 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
db7e527d | 78 | #include <asm/mutex.h> |
e6e6685a GC |
79 | #ifdef CONFIG_PARAVIRT |
80 | #include <asm/paravirt.h> | |
81 | #endif | |
1da177e4 | 82 | |
029632fb | 83 | #include "sched.h" |
391e43da | 84 | #include "../workqueue_sched.h" |
6e0534f2 | 85 | |
a8d154b0 | 86 | #define CREATE_TRACE_POINTS |
ad8d75ff | 87 | #include <trace/events/sched.h> |
a8d154b0 | 88 | |
029632fb | 89 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 90 | { |
58088ad0 PT |
91 | unsigned long delta; |
92 | ktime_t soft, hard, now; | |
d0b27fa7 | 93 | |
58088ad0 PT |
94 | for (;;) { |
95 | if (hrtimer_active(period_timer)) | |
96 | break; | |
97 | ||
98 | now = hrtimer_cb_get_time(period_timer); | |
99 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 100 | |
58088ad0 PT |
101 | soft = hrtimer_get_softexpires(period_timer); |
102 | hard = hrtimer_get_expires(period_timer); | |
103 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
104 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
105 | HRTIMER_MODE_ABS_PINNED, 0); | |
106 | } | |
107 | } | |
108 | ||
029632fb PZ |
109 | DEFINE_MUTEX(sched_domains_mutex); |
110 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 111 | |
fe44d621 | 112 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 113 | |
029632fb | 114 | void update_rq_clock(struct rq *rq) |
3e51f33f | 115 | { |
fe44d621 | 116 | s64 delta; |
305e6835 | 117 | |
61eadef6 | 118 | if (rq->skip_clock_update > 0) |
f26f9aff | 119 | return; |
aa483808 | 120 | |
fe44d621 PZ |
121 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
122 | rq->clock += delta; | |
123 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
124 | } |
125 | ||
bf5c91ba IM |
126 | /* |
127 | * Debugging: various feature bits | |
128 | */ | |
f00b45c1 | 129 | |
f00b45c1 PZ |
130 | #define SCHED_FEAT(name, enabled) \ |
131 | (1UL << __SCHED_FEAT_##name) * enabled | | |
132 | ||
bf5c91ba | 133 | const_debug unsigned int sysctl_sched_features = |
391e43da | 134 | #include "features.h" |
f00b45c1 PZ |
135 | 0; |
136 | ||
137 | #undef SCHED_FEAT | |
138 | ||
139 | #ifdef CONFIG_SCHED_DEBUG | |
140 | #define SCHED_FEAT(name, enabled) \ | |
141 | #name , | |
142 | ||
983ed7a6 | 143 | static __read_mostly char *sched_feat_names[] = { |
391e43da | 144 | #include "features.h" |
f00b45c1 PZ |
145 | NULL |
146 | }; | |
147 | ||
148 | #undef SCHED_FEAT | |
149 | ||
34f3a814 | 150 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 151 | { |
f00b45c1 PZ |
152 | int i; |
153 | ||
f8b6d1cc | 154 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
155 | if (!(sysctl_sched_features & (1UL << i))) |
156 | seq_puts(m, "NO_"); | |
157 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 158 | } |
34f3a814 | 159 | seq_puts(m, "\n"); |
f00b45c1 | 160 | |
34f3a814 | 161 | return 0; |
f00b45c1 PZ |
162 | } |
163 | ||
f8b6d1cc PZ |
164 | #ifdef HAVE_JUMP_LABEL |
165 | ||
166 | #define jump_label_key__true jump_label_key_enabled | |
167 | #define jump_label_key__false jump_label_key_disabled | |
168 | ||
169 | #define SCHED_FEAT(name, enabled) \ | |
170 | jump_label_key__##enabled , | |
171 | ||
172 | struct jump_label_key sched_feat_keys[__SCHED_FEAT_NR] = { | |
173 | #include "features.h" | |
174 | }; | |
175 | ||
176 | #undef SCHED_FEAT | |
177 | ||
178 | static void sched_feat_disable(int i) | |
179 | { | |
180 | if (jump_label_enabled(&sched_feat_keys[i])) | |
181 | jump_label_dec(&sched_feat_keys[i]); | |
182 | } | |
183 | ||
184 | static void sched_feat_enable(int i) | |
185 | { | |
186 | if (!jump_label_enabled(&sched_feat_keys[i])) | |
187 | jump_label_inc(&sched_feat_keys[i]); | |
188 | } | |
189 | #else | |
190 | static void sched_feat_disable(int i) { }; | |
191 | static void sched_feat_enable(int i) { }; | |
192 | #endif /* HAVE_JUMP_LABEL */ | |
193 | ||
f00b45c1 PZ |
194 | static ssize_t |
195 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
196 | size_t cnt, loff_t *ppos) | |
197 | { | |
198 | char buf[64]; | |
7740191c | 199 | char *cmp; |
f00b45c1 PZ |
200 | int neg = 0; |
201 | int i; | |
202 | ||
203 | if (cnt > 63) | |
204 | cnt = 63; | |
205 | ||
206 | if (copy_from_user(&buf, ubuf, cnt)) | |
207 | return -EFAULT; | |
208 | ||
209 | buf[cnt] = 0; | |
7740191c | 210 | cmp = strstrip(buf); |
f00b45c1 | 211 | |
524429c3 | 212 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
213 | neg = 1; |
214 | cmp += 3; | |
215 | } | |
216 | ||
f8b6d1cc | 217 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 218 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 219 | if (neg) { |
f00b45c1 | 220 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
221 | sched_feat_disable(i); |
222 | } else { | |
f00b45c1 | 223 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
224 | sched_feat_enable(i); |
225 | } | |
f00b45c1 PZ |
226 | break; |
227 | } | |
228 | } | |
229 | ||
f8b6d1cc | 230 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
231 | return -EINVAL; |
232 | ||
42994724 | 233 | *ppos += cnt; |
f00b45c1 PZ |
234 | |
235 | return cnt; | |
236 | } | |
237 | ||
34f3a814 LZ |
238 | static int sched_feat_open(struct inode *inode, struct file *filp) |
239 | { | |
240 | return single_open(filp, sched_feat_show, NULL); | |
241 | } | |
242 | ||
828c0950 | 243 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
244 | .open = sched_feat_open, |
245 | .write = sched_feat_write, | |
246 | .read = seq_read, | |
247 | .llseek = seq_lseek, | |
248 | .release = single_release, | |
f00b45c1 PZ |
249 | }; |
250 | ||
251 | static __init int sched_init_debug(void) | |
252 | { | |
f00b45c1 PZ |
253 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
254 | &sched_feat_fops); | |
255 | ||
256 | return 0; | |
257 | } | |
258 | late_initcall(sched_init_debug); | |
f8b6d1cc | 259 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 260 | |
b82d9fdd PZ |
261 | /* |
262 | * Number of tasks to iterate in a single balance run. | |
263 | * Limited because this is done with IRQs disabled. | |
264 | */ | |
265 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
266 | ||
e9e9250b PZ |
267 | /* |
268 | * period over which we average the RT time consumption, measured | |
269 | * in ms. | |
270 | * | |
271 | * default: 1s | |
272 | */ | |
273 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
274 | ||
fa85ae24 | 275 | /* |
9f0c1e56 | 276 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
277 | * default: 1s |
278 | */ | |
9f0c1e56 | 279 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 280 | |
029632fb | 281 | __read_mostly int scheduler_running; |
6892b75e | 282 | |
9f0c1e56 PZ |
283 | /* |
284 | * part of the period that we allow rt tasks to run in us. | |
285 | * default: 0.95s | |
286 | */ | |
287 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 288 | |
fa85ae24 | 289 | |
1da177e4 | 290 | |
0970d299 | 291 | /* |
0122ec5b | 292 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 293 | */ |
70b97a7f | 294 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
295 | __acquires(rq->lock) |
296 | { | |
0970d299 PZ |
297 | struct rq *rq; |
298 | ||
0122ec5b PZ |
299 | lockdep_assert_held(&p->pi_lock); |
300 | ||
3a5c359a | 301 | for (;;) { |
0970d299 | 302 | rq = task_rq(p); |
05fa785c | 303 | raw_spin_lock(&rq->lock); |
65cc8e48 | 304 | if (likely(rq == task_rq(p))) |
3a5c359a | 305 | return rq; |
05fa785c | 306 | raw_spin_unlock(&rq->lock); |
b29739f9 | 307 | } |
b29739f9 IM |
308 | } |
309 | ||
1da177e4 | 310 | /* |
0122ec5b | 311 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 312 | */ |
70b97a7f | 313 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 314 | __acquires(p->pi_lock) |
1da177e4 LT |
315 | __acquires(rq->lock) |
316 | { | |
70b97a7f | 317 | struct rq *rq; |
1da177e4 | 318 | |
3a5c359a | 319 | for (;;) { |
0122ec5b | 320 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 321 | rq = task_rq(p); |
05fa785c | 322 | raw_spin_lock(&rq->lock); |
65cc8e48 | 323 | if (likely(rq == task_rq(p))) |
3a5c359a | 324 | return rq; |
0122ec5b PZ |
325 | raw_spin_unlock(&rq->lock); |
326 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 327 | } |
1da177e4 LT |
328 | } |
329 | ||
a9957449 | 330 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
331 | __releases(rq->lock) |
332 | { | |
05fa785c | 333 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
334 | } |
335 | ||
0122ec5b PZ |
336 | static inline void |
337 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 338 | __releases(rq->lock) |
0122ec5b | 339 | __releases(p->pi_lock) |
1da177e4 | 340 | { |
0122ec5b PZ |
341 | raw_spin_unlock(&rq->lock); |
342 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
343 | } |
344 | ||
1da177e4 | 345 | /* |
cc2a73b5 | 346 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 347 | */ |
a9957449 | 348 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
349 | __acquires(rq->lock) |
350 | { | |
70b97a7f | 351 | struct rq *rq; |
1da177e4 LT |
352 | |
353 | local_irq_disable(); | |
354 | rq = this_rq(); | |
05fa785c | 355 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
356 | |
357 | return rq; | |
358 | } | |
359 | ||
8f4d37ec PZ |
360 | #ifdef CONFIG_SCHED_HRTICK |
361 | /* | |
362 | * Use HR-timers to deliver accurate preemption points. | |
363 | * | |
364 | * Its all a bit involved since we cannot program an hrt while holding the | |
365 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
366 | * reschedule event. | |
367 | * | |
368 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
369 | * rq->lock. | |
370 | */ | |
8f4d37ec | 371 | |
8f4d37ec PZ |
372 | static void hrtick_clear(struct rq *rq) |
373 | { | |
374 | if (hrtimer_active(&rq->hrtick_timer)) | |
375 | hrtimer_cancel(&rq->hrtick_timer); | |
376 | } | |
377 | ||
8f4d37ec PZ |
378 | /* |
379 | * High-resolution timer tick. | |
380 | * Runs from hardirq context with interrupts disabled. | |
381 | */ | |
382 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
383 | { | |
384 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
385 | ||
386 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
387 | ||
05fa785c | 388 | raw_spin_lock(&rq->lock); |
3e51f33f | 389 | update_rq_clock(rq); |
8f4d37ec | 390 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 391 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
392 | |
393 | return HRTIMER_NORESTART; | |
394 | } | |
395 | ||
95e904c7 | 396 | #ifdef CONFIG_SMP |
31656519 PZ |
397 | /* |
398 | * called from hardirq (IPI) context | |
399 | */ | |
400 | static void __hrtick_start(void *arg) | |
b328ca18 | 401 | { |
31656519 | 402 | struct rq *rq = arg; |
b328ca18 | 403 | |
05fa785c | 404 | raw_spin_lock(&rq->lock); |
31656519 PZ |
405 | hrtimer_restart(&rq->hrtick_timer); |
406 | rq->hrtick_csd_pending = 0; | |
05fa785c | 407 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
408 | } |
409 | ||
31656519 PZ |
410 | /* |
411 | * Called to set the hrtick timer state. | |
412 | * | |
413 | * called with rq->lock held and irqs disabled | |
414 | */ | |
029632fb | 415 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 416 | { |
31656519 PZ |
417 | struct hrtimer *timer = &rq->hrtick_timer; |
418 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 419 | |
cc584b21 | 420 | hrtimer_set_expires(timer, time); |
31656519 PZ |
421 | |
422 | if (rq == this_rq()) { | |
423 | hrtimer_restart(timer); | |
424 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 425 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
426 | rq->hrtick_csd_pending = 1; |
427 | } | |
b328ca18 PZ |
428 | } |
429 | ||
430 | static int | |
431 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
432 | { | |
433 | int cpu = (int)(long)hcpu; | |
434 | ||
435 | switch (action) { | |
436 | case CPU_UP_CANCELED: | |
437 | case CPU_UP_CANCELED_FROZEN: | |
438 | case CPU_DOWN_PREPARE: | |
439 | case CPU_DOWN_PREPARE_FROZEN: | |
440 | case CPU_DEAD: | |
441 | case CPU_DEAD_FROZEN: | |
31656519 | 442 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
443 | return NOTIFY_OK; |
444 | } | |
445 | ||
446 | return NOTIFY_DONE; | |
447 | } | |
448 | ||
fa748203 | 449 | static __init void init_hrtick(void) |
b328ca18 PZ |
450 | { |
451 | hotcpu_notifier(hotplug_hrtick, 0); | |
452 | } | |
31656519 PZ |
453 | #else |
454 | /* | |
455 | * Called to set the hrtick timer state. | |
456 | * | |
457 | * called with rq->lock held and irqs disabled | |
458 | */ | |
029632fb | 459 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 460 | { |
7f1e2ca9 | 461 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 462 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 463 | } |
b328ca18 | 464 | |
006c75f1 | 465 | static inline void init_hrtick(void) |
8f4d37ec | 466 | { |
8f4d37ec | 467 | } |
31656519 | 468 | #endif /* CONFIG_SMP */ |
8f4d37ec | 469 | |
31656519 | 470 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 471 | { |
31656519 PZ |
472 | #ifdef CONFIG_SMP |
473 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 474 | |
31656519 PZ |
475 | rq->hrtick_csd.flags = 0; |
476 | rq->hrtick_csd.func = __hrtick_start; | |
477 | rq->hrtick_csd.info = rq; | |
478 | #endif | |
8f4d37ec | 479 | |
31656519 PZ |
480 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
481 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 482 | } |
006c75f1 | 483 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
484 | static inline void hrtick_clear(struct rq *rq) |
485 | { | |
486 | } | |
487 | ||
8f4d37ec PZ |
488 | static inline void init_rq_hrtick(struct rq *rq) |
489 | { | |
490 | } | |
491 | ||
b328ca18 PZ |
492 | static inline void init_hrtick(void) |
493 | { | |
494 | } | |
006c75f1 | 495 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 496 | |
c24d20db IM |
497 | /* |
498 | * resched_task - mark a task 'to be rescheduled now'. | |
499 | * | |
500 | * On UP this means the setting of the need_resched flag, on SMP it | |
501 | * might also involve a cross-CPU call to trigger the scheduler on | |
502 | * the target CPU. | |
503 | */ | |
504 | #ifdef CONFIG_SMP | |
505 | ||
506 | #ifndef tsk_is_polling | |
507 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
508 | #endif | |
509 | ||
029632fb | 510 | void resched_task(struct task_struct *p) |
c24d20db IM |
511 | { |
512 | int cpu; | |
513 | ||
05fa785c | 514 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 515 | |
5ed0cec0 | 516 | if (test_tsk_need_resched(p)) |
c24d20db IM |
517 | return; |
518 | ||
5ed0cec0 | 519 | set_tsk_need_resched(p); |
c24d20db IM |
520 | |
521 | cpu = task_cpu(p); | |
522 | if (cpu == smp_processor_id()) | |
523 | return; | |
524 | ||
525 | /* NEED_RESCHED must be visible before we test polling */ | |
526 | smp_mb(); | |
527 | if (!tsk_is_polling(p)) | |
528 | smp_send_reschedule(cpu); | |
529 | } | |
530 | ||
029632fb | 531 | void resched_cpu(int cpu) |
c24d20db IM |
532 | { |
533 | struct rq *rq = cpu_rq(cpu); | |
534 | unsigned long flags; | |
535 | ||
05fa785c | 536 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
537 | return; |
538 | resched_task(cpu_curr(cpu)); | |
05fa785c | 539 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 540 | } |
06d8308c TG |
541 | |
542 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
543 | /* |
544 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
545 | * from an idle cpu. This is good for power-savings. | |
546 | * | |
547 | * We don't do similar optimization for completely idle system, as | |
548 | * selecting an idle cpu will add more delays to the timers than intended | |
549 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
550 | */ | |
551 | int get_nohz_timer_target(void) | |
552 | { | |
553 | int cpu = smp_processor_id(); | |
554 | int i; | |
555 | struct sched_domain *sd; | |
556 | ||
057f3fad | 557 | rcu_read_lock(); |
83cd4fe2 | 558 | for_each_domain(cpu, sd) { |
057f3fad PZ |
559 | for_each_cpu(i, sched_domain_span(sd)) { |
560 | if (!idle_cpu(i)) { | |
561 | cpu = i; | |
562 | goto unlock; | |
563 | } | |
564 | } | |
83cd4fe2 | 565 | } |
057f3fad PZ |
566 | unlock: |
567 | rcu_read_unlock(); | |
83cd4fe2 VP |
568 | return cpu; |
569 | } | |
06d8308c TG |
570 | /* |
571 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
572 | * idle CPU then this timer might expire before the next timer event | |
573 | * which is scheduled to wake up that CPU. In case of a completely | |
574 | * idle system the next event might even be infinite time into the | |
575 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
576 | * leaves the inner idle loop so the newly added timer is taken into | |
577 | * account when the CPU goes back to idle and evaluates the timer | |
578 | * wheel for the next timer event. | |
579 | */ | |
580 | void wake_up_idle_cpu(int cpu) | |
581 | { | |
582 | struct rq *rq = cpu_rq(cpu); | |
583 | ||
584 | if (cpu == smp_processor_id()) | |
585 | return; | |
586 | ||
587 | /* | |
588 | * This is safe, as this function is called with the timer | |
589 | * wheel base lock of (cpu) held. When the CPU is on the way | |
590 | * to idle and has not yet set rq->curr to idle then it will | |
591 | * be serialized on the timer wheel base lock and take the new | |
592 | * timer into account automatically. | |
593 | */ | |
594 | if (rq->curr != rq->idle) | |
595 | return; | |
45bf76df | 596 | |
45bf76df | 597 | /* |
06d8308c TG |
598 | * We can set TIF_RESCHED on the idle task of the other CPU |
599 | * lockless. The worst case is that the other CPU runs the | |
600 | * idle task through an additional NOOP schedule() | |
45bf76df | 601 | */ |
5ed0cec0 | 602 | set_tsk_need_resched(rq->idle); |
45bf76df | 603 | |
06d8308c TG |
604 | /* NEED_RESCHED must be visible before we test polling */ |
605 | smp_mb(); | |
606 | if (!tsk_is_polling(rq->idle)) | |
607 | smp_send_reschedule(cpu); | |
45bf76df IM |
608 | } |
609 | ||
ca38062e | 610 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 611 | { |
1c792db7 SS |
612 | int cpu = smp_processor_id(); |
613 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
614 | } |
615 | ||
ca38062e | 616 | #else /* CONFIG_NO_HZ */ |
45bf76df | 617 | |
ca38062e | 618 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 619 | { |
ca38062e | 620 | return false; |
2069dd75 PZ |
621 | } |
622 | ||
6d6bc0ad | 623 | #endif /* CONFIG_NO_HZ */ |
d842de87 | 624 | |
029632fb | 625 | void sched_avg_update(struct rq *rq) |
18d95a28 | 626 | { |
e9e9250b PZ |
627 | s64 period = sched_avg_period(); |
628 | ||
629 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
630 | /* |
631 | * Inline assembly required to prevent the compiler | |
632 | * optimising this loop into a divmod call. | |
633 | * See __iter_div_u64_rem() for another example of this. | |
634 | */ | |
635 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
636 | rq->age_stamp += period; |
637 | rq->rt_avg /= 2; | |
638 | } | |
18d95a28 PZ |
639 | } |
640 | ||
6d6bc0ad | 641 | #else /* !CONFIG_SMP */ |
029632fb | 642 | void resched_task(struct task_struct *p) |
18d95a28 | 643 | { |
05fa785c | 644 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 645 | set_tsk_need_resched(p); |
18d95a28 | 646 | } |
6d6bc0ad | 647 | #endif /* CONFIG_SMP */ |
18d95a28 | 648 | |
a790de99 PT |
649 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
650 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 651 | /* |
8277434e PT |
652 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
653 | * node and @up when leaving it for the final time. | |
654 | * | |
655 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 656 | */ |
029632fb | 657 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 658 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
659 | { |
660 | struct task_group *parent, *child; | |
eb755805 | 661 | int ret; |
c09595f6 | 662 | |
8277434e PT |
663 | parent = from; |
664 | ||
c09595f6 | 665 | down: |
eb755805 PZ |
666 | ret = (*down)(parent, data); |
667 | if (ret) | |
8277434e | 668 | goto out; |
c09595f6 PZ |
669 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
670 | parent = child; | |
671 | goto down; | |
672 | ||
673 | up: | |
674 | continue; | |
675 | } | |
eb755805 | 676 | ret = (*up)(parent, data); |
8277434e PT |
677 | if (ret || parent == from) |
678 | goto out; | |
c09595f6 PZ |
679 | |
680 | child = parent; | |
681 | parent = parent->parent; | |
682 | if (parent) | |
683 | goto up; | |
8277434e | 684 | out: |
eb755805 | 685 | return ret; |
c09595f6 PZ |
686 | } |
687 | ||
029632fb | 688 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 689 | { |
e2b245f8 | 690 | return 0; |
eb755805 | 691 | } |
18d95a28 PZ |
692 | #endif |
693 | ||
029632fb | 694 | void update_cpu_load(struct rq *this_rq); |
9c217245 | 695 | |
45bf76df IM |
696 | static void set_load_weight(struct task_struct *p) |
697 | { | |
f05998d4 NR |
698 | int prio = p->static_prio - MAX_RT_PRIO; |
699 | struct load_weight *load = &p->se.load; | |
700 | ||
dd41f596 IM |
701 | /* |
702 | * SCHED_IDLE tasks get minimal weight: | |
703 | */ | |
704 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 705 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 706 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
707 | return; |
708 | } | |
71f8bd46 | 709 | |
c8b28116 | 710 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 711 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
712 | } |
713 | ||
371fd7e7 | 714 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 715 | { |
a64692a3 | 716 | update_rq_clock(rq); |
dd41f596 | 717 | sched_info_queued(p); |
371fd7e7 | 718 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
719 | } |
720 | ||
371fd7e7 | 721 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 722 | { |
a64692a3 | 723 | update_rq_clock(rq); |
46ac22ba | 724 | sched_info_dequeued(p); |
371fd7e7 | 725 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
726 | } |
727 | ||
029632fb | 728 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
729 | { |
730 | if (task_contributes_to_load(p)) | |
731 | rq->nr_uninterruptible--; | |
732 | ||
371fd7e7 | 733 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
734 | } |
735 | ||
029632fb | 736 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
737 | { |
738 | if (task_contributes_to_load(p)) | |
739 | rq->nr_uninterruptible++; | |
740 | ||
371fd7e7 | 741 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
742 | } |
743 | ||
b52bfee4 VP |
744 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
745 | ||
305e6835 VP |
746 | /* |
747 | * There are no locks covering percpu hardirq/softirq time. | |
748 | * They are only modified in account_system_vtime, on corresponding CPU | |
749 | * with interrupts disabled. So, writes are safe. | |
750 | * They are read and saved off onto struct rq in update_rq_clock(). | |
751 | * This may result in other CPU reading this CPU's irq time and can | |
752 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
753 | * or new value with a side effect of accounting a slice of irq time to wrong |
754 | * task when irq is in progress while we read rq->clock. That is a worthy | |
755 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 756 | */ |
b52bfee4 VP |
757 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
758 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
759 | ||
760 | static DEFINE_PER_CPU(u64, irq_start_time); | |
761 | static int sched_clock_irqtime; | |
762 | ||
763 | void enable_sched_clock_irqtime(void) | |
764 | { | |
765 | sched_clock_irqtime = 1; | |
766 | } | |
767 | ||
768 | void disable_sched_clock_irqtime(void) | |
769 | { | |
770 | sched_clock_irqtime = 0; | |
771 | } | |
772 | ||
8e92c201 PZ |
773 | #ifndef CONFIG_64BIT |
774 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
775 | ||
776 | static inline void irq_time_write_begin(void) | |
777 | { | |
778 | __this_cpu_inc(irq_time_seq.sequence); | |
779 | smp_wmb(); | |
780 | } | |
781 | ||
782 | static inline void irq_time_write_end(void) | |
783 | { | |
784 | smp_wmb(); | |
785 | __this_cpu_inc(irq_time_seq.sequence); | |
786 | } | |
787 | ||
788 | static inline u64 irq_time_read(int cpu) | |
789 | { | |
790 | u64 irq_time; | |
791 | unsigned seq; | |
792 | ||
793 | do { | |
794 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
795 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
796 | per_cpu(cpu_hardirq_time, cpu); | |
797 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
798 | ||
799 | return irq_time; | |
800 | } | |
801 | #else /* CONFIG_64BIT */ | |
802 | static inline void irq_time_write_begin(void) | |
803 | { | |
804 | } | |
805 | ||
806 | static inline void irq_time_write_end(void) | |
807 | { | |
808 | } | |
809 | ||
810 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 811 | { |
305e6835 VP |
812 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
813 | } | |
8e92c201 | 814 | #endif /* CONFIG_64BIT */ |
305e6835 | 815 | |
fe44d621 PZ |
816 | /* |
817 | * Called before incrementing preempt_count on {soft,}irq_enter | |
818 | * and before decrementing preempt_count on {soft,}irq_exit. | |
819 | */ | |
b52bfee4 VP |
820 | void account_system_vtime(struct task_struct *curr) |
821 | { | |
822 | unsigned long flags; | |
fe44d621 | 823 | s64 delta; |
b52bfee4 | 824 | int cpu; |
b52bfee4 VP |
825 | |
826 | if (!sched_clock_irqtime) | |
827 | return; | |
828 | ||
829 | local_irq_save(flags); | |
830 | ||
b52bfee4 | 831 | cpu = smp_processor_id(); |
fe44d621 PZ |
832 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
833 | __this_cpu_add(irq_start_time, delta); | |
834 | ||
8e92c201 | 835 | irq_time_write_begin(); |
b52bfee4 VP |
836 | /* |
837 | * We do not account for softirq time from ksoftirqd here. | |
838 | * We want to continue accounting softirq time to ksoftirqd thread | |
839 | * in that case, so as not to confuse scheduler with a special task | |
840 | * that do not consume any time, but still wants to run. | |
841 | */ | |
842 | if (hardirq_count()) | |
fe44d621 | 843 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 844 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 845 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 846 | |
8e92c201 | 847 | irq_time_write_end(); |
b52bfee4 VP |
848 | local_irq_restore(flags); |
849 | } | |
b7dadc38 | 850 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 851 | |
e6e6685a GC |
852 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
853 | ||
854 | #ifdef CONFIG_PARAVIRT | |
855 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 856 | { |
e6e6685a GC |
857 | if (unlikely(steal > NSEC_PER_SEC)) |
858 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 859 | |
e6e6685a GC |
860 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
861 | } | |
862 | #endif | |
863 | ||
fe44d621 | 864 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 865 | { |
095c0aa8 GC |
866 | /* |
867 | * In theory, the compile should just see 0 here, and optimize out the call | |
868 | * to sched_rt_avg_update. But I don't trust it... | |
869 | */ | |
870 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
871 | s64 steal = 0, irq_delta = 0; | |
872 | #endif | |
873 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 874 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
875 | |
876 | /* | |
877 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
878 | * this case when a previous update_rq_clock() happened inside a | |
879 | * {soft,}irq region. | |
880 | * | |
881 | * When this happens, we stop ->clock_task and only update the | |
882 | * prev_irq_time stamp to account for the part that fit, so that a next | |
883 | * update will consume the rest. This ensures ->clock_task is | |
884 | * monotonic. | |
885 | * | |
886 | * It does however cause some slight miss-attribution of {soft,}irq | |
887 | * time, a more accurate solution would be to update the irq_time using | |
888 | * the current rq->clock timestamp, except that would require using | |
889 | * atomic ops. | |
890 | */ | |
891 | if (irq_delta > delta) | |
892 | irq_delta = delta; | |
893 | ||
894 | rq->prev_irq_time += irq_delta; | |
895 | delta -= irq_delta; | |
095c0aa8 GC |
896 | #endif |
897 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
898 | if (static_branch((¶virt_steal_rq_enabled))) { | |
899 | u64 st; | |
900 | ||
901 | steal = paravirt_steal_clock(cpu_of(rq)); | |
902 | steal -= rq->prev_steal_time_rq; | |
903 | ||
904 | if (unlikely(steal > delta)) | |
905 | steal = delta; | |
906 | ||
907 | st = steal_ticks(steal); | |
908 | steal = st * TICK_NSEC; | |
909 | ||
910 | rq->prev_steal_time_rq += steal; | |
911 | ||
912 | delta -= steal; | |
913 | } | |
914 | #endif | |
915 | ||
fe44d621 PZ |
916 | rq->clock_task += delta; |
917 | ||
095c0aa8 GC |
918 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
919 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
920 | sched_rt_avg_update(rq, irq_delta + steal); | |
921 | #endif | |
aa483808 VP |
922 | } |
923 | ||
095c0aa8 | 924 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
925 | static int irqtime_account_hi_update(void) |
926 | { | |
3292beb3 | 927 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
928 | unsigned long flags; |
929 | u64 latest_ns; | |
930 | int ret = 0; | |
931 | ||
932 | local_irq_save(flags); | |
933 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
612ef28a | 934 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) |
abb74cef VP |
935 | ret = 1; |
936 | local_irq_restore(flags); | |
937 | return ret; | |
938 | } | |
939 | ||
940 | static int irqtime_account_si_update(void) | |
941 | { | |
3292beb3 | 942 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
943 | unsigned long flags; |
944 | u64 latest_ns; | |
945 | int ret = 0; | |
946 | ||
947 | local_irq_save(flags); | |
948 | latest_ns = this_cpu_read(cpu_softirq_time); | |
612ef28a | 949 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) |
abb74cef VP |
950 | ret = 1; |
951 | local_irq_restore(flags); | |
952 | return ret; | |
953 | } | |
954 | ||
fe44d621 | 955 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 956 | |
abb74cef VP |
957 | #define sched_clock_irqtime (0) |
958 | ||
095c0aa8 | 959 | #endif |
b52bfee4 | 960 | |
34f971f6 PZ |
961 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
962 | { | |
963 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
964 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
965 | ||
966 | if (stop) { | |
967 | /* | |
968 | * Make it appear like a SCHED_FIFO task, its something | |
969 | * userspace knows about and won't get confused about. | |
970 | * | |
971 | * Also, it will make PI more or less work without too | |
972 | * much confusion -- but then, stop work should not | |
973 | * rely on PI working anyway. | |
974 | */ | |
975 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
976 | ||
977 | stop->sched_class = &stop_sched_class; | |
978 | } | |
979 | ||
980 | cpu_rq(cpu)->stop = stop; | |
981 | ||
982 | if (old_stop) { | |
983 | /* | |
984 | * Reset it back to a normal scheduling class so that | |
985 | * it can die in pieces. | |
986 | */ | |
987 | old_stop->sched_class = &rt_sched_class; | |
988 | } | |
989 | } | |
990 | ||
14531189 | 991 | /* |
dd41f596 | 992 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 993 | */ |
14531189 IM |
994 | static inline int __normal_prio(struct task_struct *p) |
995 | { | |
dd41f596 | 996 | return p->static_prio; |
14531189 IM |
997 | } |
998 | ||
b29739f9 IM |
999 | /* |
1000 | * Calculate the expected normal priority: i.e. priority | |
1001 | * without taking RT-inheritance into account. Might be | |
1002 | * boosted by interactivity modifiers. Changes upon fork, | |
1003 | * setprio syscalls, and whenever the interactivity | |
1004 | * estimator recalculates. | |
1005 | */ | |
36c8b586 | 1006 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1007 | { |
1008 | int prio; | |
1009 | ||
e05606d3 | 1010 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1011 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1012 | else | |
1013 | prio = __normal_prio(p); | |
1014 | return prio; | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Calculate the current priority, i.e. the priority | |
1019 | * taken into account by the scheduler. This value might | |
1020 | * be boosted by RT tasks, or might be boosted by | |
1021 | * interactivity modifiers. Will be RT if the task got | |
1022 | * RT-boosted. If not then it returns p->normal_prio. | |
1023 | */ | |
36c8b586 | 1024 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1025 | { |
1026 | p->normal_prio = normal_prio(p); | |
1027 | /* | |
1028 | * If we are RT tasks or we were boosted to RT priority, | |
1029 | * keep the priority unchanged. Otherwise, update priority | |
1030 | * to the normal priority: | |
1031 | */ | |
1032 | if (!rt_prio(p->prio)) | |
1033 | return p->normal_prio; | |
1034 | return p->prio; | |
1035 | } | |
1036 | ||
1da177e4 LT |
1037 | /** |
1038 | * task_curr - is this task currently executing on a CPU? | |
1039 | * @p: the task in question. | |
1040 | */ | |
36c8b586 | 1041 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1042 | { |
1043 | return cpu_curr(task_cpu(p)) == p; | |
1044 | } | |
1045 | ||
cb469845 SR |
1046 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1047 | const struct sched_class *prev_class, | |
da7a735e | 1048 | int oldprio) |
cb469845 SR |
1049 | { |
1050 | if (prev_class != p->sched_class) { | |
1051 | if (prev_class->switched_from) | |
da7a735e PZ |
1052 | prev_class->switched_from(rq, p); |
1053 | p->sched_class->switched_to(rq, p); | |
1054 | } else if (oldprio != p->prio) | |
1055 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
1056 | } |
1057 | ||
029632fb | 1058 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1059 | { |
1060 | const struct sched_class *class; | |
1061 | ||
1062 | if (p->sched_class == rq->curr->sched_class) { | |
1063 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1064 | } else { | |
1065 | for_each_class(class) { | |
1066 | if (class == rq->curr->sched_class) | |
1067 | break; | |
1068 | if (class == p->sched_class) { | |
1069 | resched_task(rq->curr); | |
1070 | break; | |
1071 | } | |
1072 | } | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * A queue event has occurred, and we're going to schedule. In | |
1077 | * this case, we can save a useless back to back clock update. | |
1078 | */ | |
fd2f4419 | 1079 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1080 | rq->skip_clock_update = 1; |
1081 | } | |
1082 | ||
1da177e4 | 1083 | #ifdef CONFIG_SMP |
dd41f596 | 1084 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1085 | { |
e2912009 PZ |
1086 | #ifdef CONFIG_SCHED_DEBUG |
1087 | /* | |
1088 | * We should never call set_task_cpu() on a blocked task, | |
1089 | * ttwu() will sort out the placement. | |
1090 | */ | |
077614ee PZ |
1091 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
1092 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
1093 | |
1094 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1095 | /* |
1096 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1097 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1098 | * | |
1099 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
1100 | * see set_task_rq(). | |
1101 | * | |
1102 | * Furthermore, all task_rq users should acquire both locks, see | |
1103 | * task_rq_lock(). | |
1104 | */ | |
0122ec5b PZ |
1105 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1106 | lockdep_is_held(&task_rq(p)->lock))); | |
1107 | #endif | |
e2912009 PZ |
1108 | #endif |
1109 | ||
de1d7286 | 1110 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1111 | |
0c69774e PZ |
1112 | if (task_cpu(p) != new_cpu) { |
1113 | p->se.nr_migrations++; | |
a8b0ca17 | 1114 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1115 | } |
dd41f596 IM |
1116 | |
1117 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1118 | } |
1119 | ||
969c7921 | 1120 | struct migration_arg { |
36c8b586 | 1121 | struct task_struct *task; |
1da177e4 | 1122 | int dest_cpu; |
70b97a7f | 1123 | }; |
1da177e4 | 1124 | |
969c7921 TH |
1125 | static int migration_cpu_stop(void *data); |
1126 | ||
1da177e4 LT |
1127 | /* |
1128 | * wait_task_inactive - wait for a thread to unschedule. | |
1129 | * | |
85ba2d86 RM |
1130 | * If @match_state is nonzero, it's the @p->state value just checked and |
1131 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1132 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1133 | * we return a positive number (its total switch count). If a second call | |
1134 | * a short while later returns the same number, the caller can be sure that | |
1135 | * @p has remained unscheduled the whole time. | |
1136 | * | |
1da177e4 LT |
1137 | * The caller must ensure that the task *will* unschedule sometime soon, |
1138 | * else this function might spin for a *long* time. This function can't | |
1139 | * be called with interrupts off, or it may introduce deadlock with | |
1140 | * smp_call_function() if an IPI is sent by the same process we are | |
1141 | * waiting to become inactive. | |
1142 | */ | |
85ba2d86 | 1143 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1144 | { |
1145 | unsigned long flags; | |
dd41f596 | 1146 | int running, on_rq; |
85ba2d86 | 1147 | unsigned long ncsw; |
70b97a7f | 1148 | struct rq *rq; |
1da177e4 | 1149 | |
3a5c359a AK |
1150 | for (;;) { |
1151 | /* | |
1152 | * We do the initial early heuristics without holding | |
1153 | * any task-queue locks at all. We'll only try to get | |
1154 | * the runqueue lock when things look like they will | |
1155 | * work out! | |
1156 | */ | |
1157 | rq = task_rq(p); | |
fa490cfd | 1158 | |
3a5c359a AK |
1159 | /* |
1160 | * If the task is actively running on another CPU | |
1161 | * still, just relax and busy-wait without holding | |
1162 | * any locks. | |
1163 | * | |
1164 | * NOTE! Since we don't hold any locks, it's not | |
1165 | * even sure that "rq" stays as the right runqueue! | |
1166 | * But we don't care, since "task_running()" will | |
1167 | * return false if the runqueue has changed and p | |
1168 | * is actually now running somewhere else! | |
1169 | */ | |
85ba2d86 RM |
1170 | while (task_running(rq, p)) { |
1171 | if (match_state && unlikely(p->state != match_state)) | |
1172 | return 0; | |
3a5c359a | 1173 | cpu_relax(); |
85ba2d86 | 1174 | } |
fa490cfd | 1175 | |
3a5c359a AK |
1176 | /* |
1177 | * Ok, time to look more closely! We need the rq | |
1178 | * lock now, to be *sure*. If we're wrong, we'll | |
1179 | * just go back and repeat. | |
1180 | */ | |
1181 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1182 | trace_sched_wait_task(p); |
3a5c359a | 1183 | running = task_running(rq, p); |
fd2f4419 | 1184 | on_rq = p->on_rq; |
85ba2d86 | 1185 | ncsw = 0; |
f31e11d8 | 1186 | if (!match_state || p->state == match_state) |
93dcf55f | 1187 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1188 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1189 | |
85ba2d86 RM |
1190 | /* |
1191 | * If it changed from the expected state, bail out now. | |
1192 | */ | |
1193 | if (unlikely(!ncsw)) | |
1194 | break; | |
1195 | ||
3a5c359a AK |
1196 | /* |
1197 | * Was it really running after all now that we | |
1198 | * checked with the proper locks actually held? | |
1199 | * | |
1200 | * Oops. Go back and try again.. | |
1201 | */ | |
1202 | if (unlikely(running)) { | |
1203 | cpu_relax(); | |
1204 | continue; | |
1205 | } | |
fa490cfd | 1206 | |
3a5c359a AK |
1207 | /* |
1208 | * It's not enough that it's not actively running, | |
1209 | * it must be off the runqueue _entirely_, and not | |
1210 | * preempted! | |
1211 | * | |
80dd99b3 | 1212 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1213 | * running right now), it's preempted, and we should |
1214 | * yield - it could be a while. | |
1215 | */ | |
1216 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1217 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1218 | ||
1219 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1220 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1221 | continue; |
1222 | } | |
fa490cfd | 1223 | |
3a5c359a AK |
1224 | /* |
1225 | * Ahh, all good. It wasn't running, and it wasn't | |
1226 | * runnable, which means that it will never become | |
1227 | * running in the future either. We're all done! | |
1228 | */ | |
1229 | break; | |
1230 | } | |
85ba2d86 RM |
1231 | |
1232 | return ncsw; | |
1da177e4 LT |
1233 | } |
1234 | ||
1235 | /*** | |
1236 | * kick_process - kick a running thread to enter/exit the kernel | |
1237 | * @p: the to-be-kicked thread | |
1238 | * | |
1239 | * Cause a process which is running on another CPU to enter | |
1240 | * kernel-mode, without any delay. (to get signals handled.) | |
1241 | * | |
25985edc | 1242 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1243 | * because all it wants to ensure is that the remote task enters |
1244 | * the kernel. If the IPI races and the task has been migrated | |
1245 | * to another CPU then no harm is done and the purpose has been | |
1246 | * achieved as well. | |
1247 | */ | |
36c8b586 | 1248 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1249 | { |
1250 | int cpu; | |
1251 | ||
1252 | preempt_disable(); | |
1253 | cpu = task_cpu(p); | |
1254 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1255 | smp_send_reschedule(cpu); | |
1256 | preempt_enable(); | |
1257 | } | |
b43e3521 | 1258 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1259 | #endif /* CONFIG_SMP */ |
1da177e4 | 1260 | |
970b13ba | 1261 | #ifdef CONFIG_SMP |
30da688e | 1262 | /* |
013fdb80 | 1263 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1264 | */ |
5da9a0fb PZ |
1265 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1266 | { | |
1267 | int dest_cpu; | |
1268 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
1269 | ||
1270 | /* Look for allowed, online CPU in same node. */ | |
1271 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
fa17b507 | 1272 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb PZ |
1273 | return dest_cpu; |
1274 | ||
1275 | /* Any allowed, online CPU? */ | |
fa17b507 | 1276 | dest_cpu = cpumask_any_and(tsk_cpus_allowed(p), cpu_active_mask); |
5da9a0fb PZ |
1277 | if (dest_cpu < nr_cpu_ids) |
1278 | return dest_cpu; | |
1279 | ||
1280 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
1281 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
1282 | /* | |
1283 | * Don't tell them about moving exiting tasks or | |
1284 | * kernel threads (both mm NULL), since they never | |
1285 | * leave kernel. | |
1286 | */ | |
1287 | if (p->mm && printk_ratelimit()) { | |
1288 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
1289 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
1290 | } |
1291 | ||
1292 | return dest_cpu; | |
1293 | } | |
1294 | ||
e2912009 | 1295 | /* |
013fdb80 | 1296 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1297 | */ |
970b13ba | 1298 | static inline |
7608dec2 | 1299 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1300 | { |
7608dec2 | 1301 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1302 | |
1303 | /* | |
1304 | * In order not to call set_task_cpu() on a blocking task we need | |
1305 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1306 | * cpu. | |
1307 | * | |
1308 | * Since this is common to all placement strategies, this lives here. | |
1309 | * | |
1310 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1311 | * not worry about this generic constraint ] | |
1312 | */ | |
fa17b507 | 1313 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1314 | !cpu_online(cpu))) |
5da9a0fb | 1315 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1316 | |
1317 | return cpu; | |
970b13ba | 1318 | } |
09a40af5 MG |
1319 | |
1320 | static void update_avg(u64 *avg, u64 sample) | |
1321 | { | |
1322 | s64 diff = sample - *avg; | |
1323 | *avg += diff >> 3; | |
1324 | } | |
970b13ba PZ |
1325 | #endif |
1326 | ||
d7c01d27 | 1327 | static void |
b84cb5df | 1328 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1329 | { |
d7c01d27 | 1330 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1331 | struct rq *rq = this_rq(); |
1332 | ||
d7c01d27 PZ |
1333 | #ifdef CONFIG_SMP |
1334 | int this_cpu = smp_processor_id(); | |
1335 | ||
1336 | if (cpu == this_cpu) { | |
1337 | schedstat_inc(rq, ttwu_local); | |
1338 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1339 | } else { | |
1340 | struct sched_domain *sd; | |
1341 | ||
1342 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1343 | rcu_read_lock(); |
d7c01d27 PZ |
1344 | for_each_domain(this_cpu, sd) { |
1345 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1346 | schedstat_inc(sd, ttwu_wake_remote); | |
1347 | break; | |
1348 | } | |
1349 | } | |
057f3fad | 1350 | rcu_read_unlock(); |
d7c01d27 | 1351 | } |
f339b9dc PZ |
1352 | |
1353 | if (wake_flags & WF_MIGRATED) | |
1354 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1355 | ||
d7c01d27 PZ |
1356 | #endif /* CONFIG_SMP */ |
1357 | ||
1358 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1359 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1360 | |
1361 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1362 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1363 | |
d7c01d27 PZ |
1364 | #endif /* CONFIG_SCHEDSTATS */ |
1365 | } | |
1366 | ||
1367 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1368 | { | |
9ed3811a | 1369 | activate_task(rq, p, en_flags); |
fd2f4419 | 1370 | p->on_rq = 1; |
c2f7115e PZ |
1371 | |
1372 | /* if a worker is waking up, notify workqueue */ | |
1373 | if (p->flags & PF_WQ_WORKER) | |
1374 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1375 | } |
1376 | ||
23f41eeb PZ |
1377 | /* |
1378 | * Mark the task runnable and perform wakeup-preemption. | |
1379 | */ | |
89363381 | 1380 | static void |
23f41eeb | 1381 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1382 | { |
89363381 | 1383 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1384 | check_preempt_curr(rq, p, wake_flags); |
1385 | ||
1386 | p->state = TASK_RUNNING; | |
1387 | #ifdef CONFIG_SMP | |
1388 | if (p->sched_class->task_woken) | |
1389 | p->sched_class->task_woken(rq, p); | |
1390 | ||
e69c6341 | 1391 | if (rq->idle_stamp) { |
9ed3811a TH |
1392 | u64 delta = rq->clock - rq->idle_stamp; |
1393 | u64 max = 2*sysctl_sched_migration_cost; | |
1394 | ||
1395 | if (delta > max) | |
1396 | rq->avg_idle = max; | |
1397 | else | |
1398 | update_avg(&rq->avg_idle, delta); | |
1399 | rq->idle_stamp = 0; | |
1400 | } | |
1401 | #endif | |
1402 | } | |
1403 | ||
c05fbafb PZ |
1404 | static void |
1405 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1406 | { | |
1407 | #ifdef CONFIG_SMP | |
1408 | if (p->sched_contributes_to_load) | |
1409 | rq->nr_uninterruptible--; | |
1410 | #endif | |
1411 | ||
1412 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1413 | ttwu_do_wakeup(rq, p, wake_flags); | |
1414 | } | |
1415 | ||
1416 | /* | |
1417 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1418 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1419 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1420 | * the task is still ->on_rq. | |
1421 | */ | |
1422 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1423 | { | |
1424 | struct rq *rq; | |
1425 | int ret = 0; | |
1426 | ||
1427 | rq = __task_rq_lock(p); | |
1428 | if (p->on_rq) { | |
1429 | ttwu_do_wakeup(rq, p, wake_flags); | |
1430 | ret = 1; | |
1431 | } | |
1432 | __task_rq_unlock(rq); | |
1433 | ||
1434 | return ret; | |
1435 | } | |
1436 | ||
317f3941 | 1437 | #ifdef CONFIG_SMP |
fa14ff4a | 1438 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1439 | { |
1440 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1441 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1442 | struct task_struct *p; | |
317f3941 PZ |
1443 | |
1444 | raw_spin_lock(&rq->lock); | |
1445 | ||
fa14ff4a PZ |
1446 | while (llist) { |
1447 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1448 | llist = llist_next(llist); | |
317f3941 PZ |
1449 | ttwu_do_activate(rq, p, 0); |
1450 | } | |
1451 | ||
1452 | raw_spin_unlock(&rq->lock); | |
1453 | } | |
1454 | ||
1455 | void scheduler_ipi(void) | |
1456 | { | |
ca38062e | 1457 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1458 | return; |
1459 | ||
1460 | /* | |
1461 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1462 | * traditionally all their work was done from the interrupt return | |
1463 | * path. Now that we actually do some work, we need to make sure | |
1464 | * we do call them. | |
1465 | * | |
1466 | * Some archs already do call them, luckily irq_enter/exit nest | |
1467 | * properly. | |
1468 | * | |
1469 | * Arguably we should visit all archs and update all handlers, | |
1470 | * however a fair share of IPIs are still resched only so this would | |
1471 | * somewhat pessimize the simple resched case. | |
1472 | */ | |
1473 | irq_enter(); | |
fa14ff4a | 1474 | sched_ttwu_pending(); |
ca38062e SS |
1475 | |
1476 | /* | |
1477 | * Check if someone kicked us for doing the nohz idle load balance. | |
1478 | */ | |
6eb57e0d SS |
1479 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1480 | this_rq()->idle_balance = 1; | |
ca38062e | 1481 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1482 | } |
c5d753a5 | 1483 | irq_exit(); |
317f3941 PZ |
1484 | } |
1485 | ||
1486 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1487 | { | |
fa14ff4a | 1488 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1489 | smp_send_reschedule(cpu); |
1490 | } | |
d6aa8f85 PZ |
1491 | |
1492 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1493 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
1494 | { | |
1495 | struct rq *rq; | |
1496 | int ret = 0; | |
1497 | ||
1498 | rq = __task_rq_lock(p); | |
1499 | if (p->on_cpu) { | |
1500 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
1501 | ttwu_do_wakeup(rq, p, wake_flags); | |
1502 | ret = 1; | |
1503 | } | |
1504 | __task_rq_unlock(rq); | |
1505 | ||
1506 | return ret; | |
1507 | ||
1508 | } | |
1509 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
518cd623 PZ |
1510 | |
1511 | static inline int ttwu_share_cache(int this_cpu, int that_cpu) | |
1512 | { | |
1513 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1514 | } | |
d6aa8f85 | 1515 | #endif /* CONFIG_SMP */ |
317f3941 | 1516 | |
c05fbafb PZ |
1517 | static void ttwu_queue(struct task_struct *p, int cpu) |
1518 | { | |
1519 | struct rq *rq = cpu_rq(cpu); | |
1520 | ||
17d9f311 | 1521 | #if defined(CONFIG_SMP) |
518cd623 | 1522 | if (sched_feat(TTWU_QUEUE) && !ttwu_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1523 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1524 | ttwu_queue_remote(p, cpu); |
1525 | return; | |
1526 | } | |
1527 | #endif | |
1528 | ||
c05fbafb PZ |
1529 | raw_spin_lock(&rq->lock); |
1530 | ttwu_do_activate(rq, p, 0); | |
1531 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1532 | } |
1533 | ||
1534 | /** | |
1da177e4 | 1535 | * try_to_wake_up - wake up a thread |
9ed3811a | 1536 | * @p: the thread to be awakened |
1da177e4 | 1537 | * @state: the mask of task states that can be woken |
9ed3811a | 1538 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1539 | * |
1540 | * Put it on the run-queue if it's not already there. The "current" | |
1541 | * thread is always on the run-queue (except when the actual | |
1542 | * re-schedule is in progress), and as such you're allowed to do | |
1543 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1544 | * runnable without the overhead of this. | |
1545 | * | |
9ed3811a TH |
1546 | * Returns %true if @p was woken up, %false if it was already running |
1547 | * or @state didn't match @p's state. | |
1da177e4 | 1548 | */ |
e4a52bcb PZ |
1549 | static int |
1550 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1551 | { |
1da177e4 | 1552 | unsigned long flags; |
c05fbafb | 1553 | int cpu, success = 0; |
2398f2c6 | 1554 | |
04e2f174 | 1555 | smp_wmb(); |
013fdb80 | 1556 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1557 | if (!(p->state & state)) |
1da177e4 LT |
1558 | goto out; |
1559 | ||
c05fbafb | 1560 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1561 | cpu = task_cpu(p); |
1da177e4 | 1562 | |
c05fbafb PZ |
1563 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1564 | goto stat; | |
1da177e4 | 1565 | |
1da177e4 | 1566 | #ifdef CONFIG_SMP |
e9c84311 | 1567 | /* |
c05fbafb PZ |
1568 | * If the owning (remote) cpu is still in the middle of schedule() with |
1569 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1570 | */ |
e4a52bcb PZ |
1571 | while (p->on_cpu) { |
1572 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1573 | /* | |
d6aa8f85 PZ |
1574 | * In case the architecture enables interrupts in |
1575 | * context_switch(), we cannot busy wait, since that | |
1576 | * would lead to deadlocks when an interrupt hits and | |
1577 | * tries to wake up @prev. So bail and do a complete | |
1578 | * remote wakeup. | |
e4a52bcb | 1579 | */ |
d6aa8f85 | 1580 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 1581 | goto stat; |
d6aa8f85 | 1582 | #else |
e4a52bcb | 1583 | cpu_relax(); |
d6aa8f85 | 1584 | #endif |
371fd7e7 | 1585 | } |
0970d299 | 1586 | /* |
e4a52bcb | 1587 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1588 | */ |
e4a52bcb | 1589 | smp_rmb(); |
1da177e4 | 1590 | |
a8e4f2ea | 1591 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1592 | p->state = TASK_WAKING; |
e7693a36 | 1593 | |
e4a52bcb | 1594 | if (p->sched_class->task_waking) |
74f8e4b2 | 1595 | p->sched_class->task_waking(p); |
efbbd05a | 1596 | |
7608dec2 | 1597 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1598 | if (task_cpu(p) != cpu) { |
1599 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1600 | set_task_cpu(p, cpu); |
f339b9dc | 1601 | } |
1da177e4 | 1602 | #endif /* CONFIG_SMP */ |
1da177e4 | 1603 | |
c05fbafb PZ |
1604 | ttwu_queue(p, cpu); |
1605 | stat: | |
b84cb5df | 1606 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1607 | out: |
013fdb80 | 1608 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1609 | |
1610 | return success; | |
1611 | } | |
1612 | ||
21aa9af0 TH |
1613 | /** |
1614 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1615 | * @p: the thread to be awakened | |
1616 | * | |
2acca55e | 1617 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1618 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1619 | * the current task. |
21aa9af0 TH |
1620 | */ |
1621 | static void try_to_wake_up_local(struct task_struct *p) | |
1622 | { | |
1623 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1624 | |
1625 | BUG_ON(rq != this_rq()); | |
1626 | BUG_ON(p == current); | |
1627 | lockdep_assert_held(&rq->lock); | |
1628 | ||
2acca55e PZ |
1629 | if (!raw_spin_trylock(&p->pi_lock)) { |
1630 | raw_spin_unlock(&rq->lock); | |
1631 | raw_spin_lock(&p->pi_lock); | |
1632 | raw_spin_lock(&rq->lock); | |
1633 | } | |
1634 | ||
21aa9af0 | 1635 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1636 | goto out; |
21aa9af0 | 1637 | |
fd2f4419 | 1638 | if (!p->on_rq) |
d7c01d27 PZ |
1639 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1640 | ||
23f41eeb | 1641 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1642 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1643 | out: |
1644 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1645 | } |
1646 | ||
50fa610a DH |
1647 | /** |
1648 | * wake_up_process - Wake up a specific process | |
1649 | * @p: The process to be woken up. | |
1650 | * | |
1651 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1652 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1653 | * running. | |
1654 | * | |
1655 | * It may be assumed that this function implies a write memory barrier before | |
1656 | * changing the task state if and only if any tasks are woken up. | |
1657 | */ | |
7ad5b3a5 | 1658 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1659 | { |
d9514f6c | 1660 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1661 | } |
1da177e4 LT |
1662 | EXPORT_SYMBOL(wake_up_process); |
1663 | ||
7ad5b3a5 | 1664 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1665 | { |
1666 | return try_to_wake_up(p, state, 0); | |
1667 | } | |
1668 | ||
1da177e4 LT |
1669 | /* |
1670 | * Perform scheduler related setup for a newly forked process p. | |
1671 | * p is forked by current. | |
dd41f596 IM |
1672 | * |
1673 | * __sched_fork() is basic setup used by init_idle() too: | |
1674 | */ | |
1675 | static void __sched_fork(struct task_struct *p) | |
1676 | { | |
fd2f4419 PZ |
1677 | p->on_rq = 0; |
1678 | ||
1679 | p->se.on_rq = 0; | |
dd41f596 IM |
1680 | p->se.exec_start = 0; |
1681 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1682 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1683 | p->se.nr_migrations = 0; |
da7a735e | 1684 | p->se.vruntime = 0; |
fd2f4419 | 1685 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1686 | |
1687 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1688 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1689 | #endif |
476d139c | 1690 | |
fa717060 | 1691 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1692 | |
e107be36 AK |
1693 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1694 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1695 | #endif | |
dd41f596 IM |
1696 | } |
1697 | ||
1698 | /* | |
1699 | * fork()/clone()-time setup: | |
1700 | */ | |
3e51e3ed | 1701 | void sched_fork(struct task_struct *p) |
dd41f596 | 1702 | { |
0122ec5b | 1703 | unsigned long flags; |
dd41f596 IM |
1704 | int cpu = get_cpu(); |
1705 | ||
1706 | __sched_fork(p); | |
06b83b5f | 1707 | /* |
0017d735 | 1708 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1709 | * nobody will actually run it, and a signal or other external |
1710 | * event cannot wake it up and insert it on the runqueue either. | |
1711 | */ | |
0017d735 | 1712 | p->state = TASK_RUNNING; |
dd41f596 | 1713 | |
c350a04e MG |
1714 | /* |
1715 | * Make sure we do not leak PI boosting priority to the child. | |
1716 | */ | |
1717 | p->prio = current->normal_prio; | |
1718 | ||
b9dc29e7 MG |
1719 | /* |
1720 | * Revert to default priority/policy on fork if requested. | |
1721 | */ | |
1722 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1723 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1724 | p->policy = SCHED_NORMAL; |
6c697bdf | 1725 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1726 | p->rt_priority = 0; |
1727 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1728 | p->static_prio = NICE_TO_PRIO(0); | |
1729 | ||
1730 | p->prio = p->normal_prio = __normal_prio(p); | |
1731 | set_load_weight(p); | |
6c697bdf | 1732 | |
b9dc29e7 MG |
1733 | /* |
1734 | * We don't need the reset flag anymore after the fork. It has | |
1735 | * fulfilled its duty: | |
1736 | */ | |
1737 | p->sched_reset_on_fork = 0; | |
1738 | } | |
ca94c442 | 1739 | |
2ddbf952 HS |
1740 | if (!rt_prio(p->prio)) |
1741 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1742 | |
cd29fe6f PZ |
1743 | if (p->sched_class->task_fork) |
1744 | p->sched_class->task_fork(p); | |
1745 | ||
86951599 PZ |
1746 | /* |
1747 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1748 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1749 | * is ran before sched_fork(). | |
1750 | * | |
1751 | * Silence PROVE_RCU. | |
1752 | */ | |
0122ec5b | 1753 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1754 | set_task_cpu(p, cpu); |
0122ec5b | 1755 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1756 | |
52f17b6c | 1757 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1758 | if (likely(sched_info_on())) |
52f17b6c | 1759 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1760 | #endif |
3ca7a440 PZ |
1761 | #if defined(CONFIG_SMP) |
1762 | p->on_cpu = 0; | |
4866cde0 | 1763 | #endif |
bdd4e85d | 1764 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1765 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1766 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1767 | #endif |
806c09a7 | 1768 | #ifdef CONFIG_SMP |
917b627d | 1769 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1770 | #endif |
917b627d | 1771 | |
476d139c | 1772 | put_cpu(); |
1da177e4 LT |
1773 | } |
1774 | ||
1775 | /* | |
1776 | * wake_up_new_task - wake up a newly created task for the first time. | |
1777 | * | |
1778 | * This function will do some initial scheduler statistics housekeeping | |
1779 | * that must be done for every newly created context, then puts the task | |
1780 | * on the runqueue and wakes it. | |
1781 | */ | |
3e51e3ed | 1782 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1783 | { |
1784 | unsigned long flags; | |
dd41f596 | 1785 | struct rq *rq; |
fabf318e | 1786 | |
ab2515c4 | 1787 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1788 | #ifdef CONFIG_SMP |
1789 | /* | |
1790 | * Fork balancing, do it here and not earlier because: | |
1791 | * - cpus_allowed can change in the fork path | |
1792 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1793 | */ |
ab2515c4 | 1794 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1795 | #endif |
1796 | ||
ab2515c4 | 1797 | rq = __task_rq_lock(p); |
cd29fe6f | 1798 | activate_task(rq, p, 0); |
fd2f4419 | 1799 | p->on_rq = 1; |
89363381 | 1800 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1801 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1802 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1803 | if (p->sched_class->task_woken) |
1804 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1805 | #endif |
0122ec5b | 1806 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1807 | } |
1808 | ||
e107be36 AK |
1809 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1810 | ||
1811 | /** | |
80dd99b3 | 1812 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1813 | * @notifier: notifier struct to register |
e107be36 AK |
1814 | */ |
1815 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1816 | { | |
1817 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1818 | } | |
1819 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1820 | ||
1821 | /** | |
1822 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1823 | * @notifier: notifier struct to unregister |
e107be36 AK |
1824 | * |
1825 | * This is safe to call from within a preemption notifier. | |
1826 | */ | |
1827 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1828 | { | |
1829 | hlist_del(¬ifier->link); | |
1830 | } | |
1831 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1832 | ||
1833 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1834 | { | |
1835 | struct preempt_notifier *notifier; | |
1836 | struct hlist_node *node; | |
1837 | ||
1838 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1839 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1840 | } | |
1841 | ||
1842 | static void | |
1843 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1844 | struct task_struct *next) | |
1845 | { | |
1846 | struct preempt_notifier *notifier; | |
1847 | struct hlist_node *node; | |
1848 | ||
1849 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1850 | notifier->ops->sched_out(notifier, next); | |
1851 | } | |
1852 | ||
6d6bc0ad | 1853 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1854 | |
1855 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1856 | { | |
1857 | } | |
1858 | ||
1859 | static void | |
1860 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1861 | struct task_struct *next) | |
1862 | { | |
1863 | } | |
1864 | ||
6d6bc0ad | 1865 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1866 | |
4866cde0 NP |
1867 | /** |
1868 | * prepare_task_switch - prepare to switch tasks | |
1869 | * @rq: the runqueue preparing to switch | |
421cee29 | 1870 | * @prev: the current task that is being switched out |
4866cde0 NP |
1871 | * @next: the task we are going to switch to. |
1872 | * | |
1873 | * This is called with the rq lock held and interrupts off. It must | |
1874 | * be paired with a subsequent finish_task_switch after the context | |
1875 | * switch. | |
1876 | * | |
1877 | * prepare_task_switch sets up locking and calls architecture specific | |
1878 | * hooks. | |
1879 | */ | |
e107be36 AK |
1880 | static inline void |
1881 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1882 | struct task_struct *next) | |
4866cde0 | 1883 | { |
fe4b04fa PZ |
1884 | sched_info_switch(prev, next); |
1885 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1886 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1887 | prepare_lock_switch(rq, next); |
1888 | prepare_arch_switch(next); | |
fe4b04fa | 1889 | trace_sched_switch(prev, next); |
4866cde0 NP |
1890 | } |
1891 | ||
1da177e4 LT |
1892 | /** |
1893 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1894 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1895 | * @prev: the thread we just switched away from. |
1896 | * | |
4866cde0 NP |
1897 | * finish_task_switch must be called after the context switch, paired |
1898 | * with a prepare_task_switch call before the context switch. | |
1899 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1900 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1901 | * |
1902 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1903 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1904 | * with the lock held can cause deadlocks; see schedule() for |
1905 | * details.) | |
1906 | */ | |
a9957449 | 1907 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1908 | __releases(rq->lock) |
1909 | { | |
1da177e4 | 1910 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1911 | long prev_state; |
1da177e4 LT |
1912 | |
1913 | rq->prev_mm = NULL; | |
1914 | ||
1915 | /* | |
1916 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1917 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1918 | * schedule one last time. The schedule call will never return, and |
1919 | * the scheduled task must drop that reference. | |
c394cc9f | 1920 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1921 | * still held, otherwise prev could be scheduled on another cpu, die |
1922 | * there before we look at prev->state, and then the reference would | |
1923 | * be dropped twice. | |
1924 | * Manfred Spraul <manfred@colorfullife.com> | |
1925 | */ | |
55a101f8 | 1926 | prev_state = prev->state; |
4866cde0 | 1927 | finish_arch_switch(prev); |
8381f65d JI |
1928 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1929 | local_irq_disable(); | |
1930 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 1931 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
1932 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1933 | local_irq_enable(); | |
1934 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 1935 | finish_lock_switch(rq, prev); |
1ac9bc69 | 1936 | trace_sched_stat_sleeptime(current, rq->clock); |
e8fa1362 | 1937 | |
e107be36 | 1938 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1939 | if (mm) |
1940 | mmdrop(mm); | |
c394cc9f | 1941 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1942 | /* |
1943 | * Remove function-return probe instances associated with this | |
1944 | * task and put them back on the free list. | |
9761eea8 | 1945 | */ |
c6fd91f0 | 1946 | kprobe_flush_task(prev); |
1da177e4 | 1947 | put_task_struct(prev); |
c6fd91f0 | 1948 | } |
1da177e4 LT |
1949 | } |
1950 | ||
3f029d3c GH |
1951 | #ifdef CONFIG_SMP |
1952 | ||
1953 | /* assumes rq->lock is held */ | |
1954 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1955 | { | |
1956 | if (prev->sched_class->pre_schedule) | |
1957 | prev->sched_class->pre_schedule(rq, prev); | |
1958 | } | |
1959 | ||
1960 | /* rq->lock is NOT held, but preemption is disabled */ | |
1961 | static inline void post_schedule(struct rq *rq) | |
1962 | { | |
1963 | if (rq->post_schedule) { | |
1964 | unsigned long flags; | |
1965 | ||
05fa785c | 1966 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1967 | if (rq->curr->sched_class->post_schedule) |
1968 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1969 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1970 | |
1971 | rq->post_schedule = 0; | |
1972 | } | |
1973 | } | |
1974 | ||
1975 | #else | |
da19ab51 | 1976 | |
3f029d3c GH |
1977 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1978 | { | |
1979 | } | |
1980 | ||
1981 | static inline void post_schedule(struct rq *rq) | |
1982 | { | |
1da177e4 LT |
1983 | } |
1984 | ||
3f029d3c GH |
1985 | #endif |
1986 | ||
1da177e4 LT |
1987 | /** |
1988 | * schedule_tail - first thing a freshly forked thread must call. | |
1989 | * @prev: the thread we just switched away from. | |
1990 | */ | |
36c8b586 | 1991 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1992 | __releases(rq->lock) |
1993 | { | |
70b97a7f IM |
1994 | struct rq *rq = this_rq(); |
1995 | ||
4866cde0 | 1996 | finish_task_switch(rq, prev); |
da19ab51 | 1997 | |
3f029d3c GH |
1998 | /* |
1999 | * FIXME: do we need to worry about rq being invalidated by the | |
2000 | * task_switch? | |
2001 | */ | |
2002 | post_schedule(rq); | |
70b97a7f | 2003 | |
4866cde0 NP |
2004 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2005 | /* In this case, finish_task_switch does not reenable preemption */ | |
2006 | preempt_enable(); | |
2007 | #endif | |
1da177e4 | 2008 | if (current->set_child_tid) |
b488893a | 2009 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * context_switch - switch to the new MM and the new | |
2014 | * thread's register state. | |
2015 | */ | |
dd41f596 | 2016 | static inline void |
70b97a7f | 2017 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2018 | struct task_struct *next) |
1da177e4 | 2019 | { |
dd41f596 | 2020 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2021 | |
e107be36 | 2022 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2023 | |
dd41f596 IM |
2024 | mm = next->mm; |
2025 | oldmm = prev->active_mm; | |
9226d125 ZA |
2026 | /* |
2027 | * For paravirt, this is coupled with an exit in switch_to to | |
2028 | * combine the page table reload and the switch backend into | |
2029 | * one hypercall. | |
2030 | */ | |
224101ed | 2031 | arch_start_context_switch(prev); |
9226d125 | 2032 | |
31915ab4 | 2033 | if (!mm) { |
1da177e4 LT |
2034 | next->active_mm = oldmm; |
2035 | atomic_inc(&oldmm->mm_count); | |
2036 | enter_lazy_tlb(oldmm, next); | |
2037 | } else | |
2038 | switch_mm(oldmm, mm, next); | |
2039 | ||
31915ab4 | 2040 | if (!prev->mm) { |
1da177e4 | 2041 | prev->active_mm = NULL; |
1da177e4 LT |
2042 | rq->prev_mm = oldmm; |
2043 | } | |
3a5f5e48 IM |
2044 | /* |
2045 | * Since the runqueue lock will be released by the next | |
2046 | * task (which is an invalid locking op but in the case | |
2047 | * of the scheduler it's an obvious special-case), so we | |
2048 | * do an early lockdep release here: | |
2049 | */ | |
2050 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2051 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2052 | #endif |
1da177e4 LT |
2053 | |
2054 | /* Here we just switch the register state and the stack. */ | |
2055 | switch_to(prev, next, prev); | |
2056 | ||
dd41f596 IM |
2057 | barrier(); |
2058 | /* | |
2059 | * this_rq must be evaluated again because prev may have moved | |
2060 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2061 | * frame will be invalid. | |
2062 | */ | |
2063 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2064 | } |
2065 | ||
2066 | /* | |
2067 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2068 | * | |
2069 | * externally visible scheduler statistics: current number of runnable | |
2070 | * threads, current number of uninterruptible-sleeping threads, total | |
2071 | * number of context switches performed since bootup. | |
2072 | */ | |
2073 | unsigned long nr_running(void) | |
2074 | { | |
2075 | unsigned long i, sum = 0; | |
2076 | ||
2077 | for_each_online_cpu(i) | |
2078 | sum += cpu_rq(i)->nr_running; | |
2079 | ||
2080 | return sum; | |
f711f609 | 2081 | } |
1da177e4 LT |
2082 | |
2083 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2084 | { |
1da177e4 | 2085 | unsigned long i, sum = 0; |
f711f609 | 2086 | |
0a945022 | 2087 | for_each_possible_cpu(i) |
1da177e4 | 2088 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2089 | |
2090 | /* | |
1da177e4 LT |
2091 | * Since we read the counters lockless, it might be slightly |
2092 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2093 | */ |
1da177e4 LT |
2094 | if (unlikely((long)sum < 0)) |
2095 | sum = 0; | |
f711f609 | 2096 | |
1da177e4 | 2097 | return sum; |
f711f609 | 2098 | } |
f711f609 | 2099 | |
1da177e4 | 2100 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2101 | { |
cc94abfc SR |
2102 | int i; |
2103 | unsigned long long sum = 0; | |
46cb4b7c | 2104 | |
0a945022 | 2105 | for_each_possible_cpu(i) |
1da177e4 | 2106 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2107 | |
1da177e4 LT |
2108 | return sum; |
2109 | } | |
483b4ee6 | 2110 | |
1da177e4 LT |
2111 | unsigned long nr_iowait(void) |
2112 | { | |
2113 | unsigned long i, sum = 0; | |
483b4ee6 | 2114 | |
0a945022 | 2115 | for_each_possible_cpu(i) |
1da177e4 | 2116 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2117 | |
1da177e4 LT |
2118 | return sum; |
2119 | } | |
483b4ee6 | 2120 | |
8c215bd3 | 2121 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2122 | { |
8c215bd3 | 2123 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2124 | return atomic_read(&this->nr_iowait); |
2125 | } | |
46cb4b7c | 2126 | |
69d25870 AV |
2127 | unsigned long this_cpu_load(void) |
2128 | { | |
2129 | struct rq *this = this_rq(); | |
2130 | return this->cpu_load[0]; | |
2131 | } | |
e790fb0b | 2132 | |
46cb4b7c | 2133 | |
dce48a84 TG |
2134 | /* Variables and functions for calc_load */ |
2135 | static atomic_long_t calc_load_tasks; | |
2136 | static unsigned long calc_load_update; | |
2137 | unsigned long avenrun[3]; | |
2138 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2139 | |
74f5187a PZ |
2140 | static long calc_load_fold_active(struct rq *this_rq) |
2141 | { | |
2142 | long nr_active, delta = 0; | |
2143 | ||
2144 | nr_active = this_rq->nr_running; | |
2145 | nr_active += (long) this_rq->nr_uninterruptible; | |
2146 | ||
2147 | if (nr_active != this_rq->calc_load_active) { | |
2148 | delta = nr_active - this_rq->calc_load_active; | |
2149 | this_rq->calc_load_active = nr_active; | |
2150 | } | |
2151 | ||
2152 | return delta; | |
2153 | } | |
2154 | ||
0f004f5a PZ |
2155 | static unsigned long |
2156 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2157 | { | |
2158 | load *= exp; | |
2159 | load += active * (FIXED_1 - exp); | |
2160 | load += 1UL << (FSHIFT - 1); | |
2161 | return load >> FSHIFT; | |
2162 | } | |
2163 | ||
74f5187a PZ |
2164 | #ifdef CONFIG_NO_HZ |
2165 | /* | |
2166 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2167 | * | |
2168 | * When making the ILB scale, we should try to pull this in as well. | |
2169 | */ | |
2170 | static atomic_long_t calc_load_tasks_idle; | |
2171 | ||
029632fb | 2172 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2173 | { |
2174 | long delta; | |
2175 | ||
2176 | delta = calc_load_fold_active(this_rq); | |
2177 | if (delta) | |
2178 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2179 | } | |
2180 | ||
2181 | static long calc_load_fold_idle(void) | |
2182 | { | |
2183 | long delta = 0; | |
2184 | ||
2185 | /* | |
2186 | * Its got a race, we don't care... | |
2187 | */ | |
2188 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2189 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2190 | ||
2191 | return delta; | |
2192 | } | |
0f004f5a PZ |
2193 | |
2194 | /** | |
2195 | * fixed_power_int - compute: x^n, in O(log n) time | |
2196 | * | |
2197 | * @x: base of the power | |
2198 | * @frac_bits: fractional bits of @x | |
2199 | * @n: power to raise @x to. | |
2200 | * | |
2201 | * By exploiting the relation between the definition of the natural power | |
2202 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2203 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2204 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2205 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2206 | * of course trivially computable in O(log_2 n), the length of our binary | |
2207 | * vector. | |
2208 | */ | |
2209 | static unsigned long | |
2210 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2211 | { | |
2212 | unsigned long result = 1UL << frac_bits; | |
2213 | ||
2214 | if (n) for (;;) { | |
2215 | if (n & 1) { | |
2216 | result *= x; | |
2217 | result += 1UL << (frac_bits - 1); | |
2218 | result >>= frac_bits; | |
2219 | } | |
2220 | n >>= 1; | |
2221 | if (!n) | |
2222 | break; | |
2223 | x *= x; | |
2224 | x += 1UL << (frac_bits - 1); | |
2225 | x >>= frac_bits; | |
2226 | } | |
2227 | ||
2228 | return result; | |
2229 | } | |
2230 | ||
2231 | /* | |
2232 | * a1 = a0 * e + a * (1 - e) | |
2233 | * | |
2234 | * a2 = a1 * e + a * (1 - e) | |
2235 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2236 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2237 | * | |
2238 | * a3 = a2 * e + a * (1 - e) | |
2239 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2240 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2241 | * | |
2242 | * ... | |
2243 | * | |
2244 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2245 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2246 | * = a0 * e^n + a * (1 - e^n) | |
2247 | * | |
2248 | * [1] application of the geometric series: | |
2249 | * | |
2250 | * n 1 - x^(n+1) | |
2251 | * S_n := \Sum x^i = ------------- | |
2252 | * i=0 1 - x | |
2253 | */ | |
2254 | static unsigned long | |
2255 | calc_load_n(unsigned long load, unsigned long exp, | |
2256 | unsigned long active, unsigned int n) | |
2257 | { | |
2258 | ||
2259 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2264 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2265 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2266 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2267 | * | |
2268 | * Once we've updated the global active value, we need to apply the exponential | |
2269 | * weights adjusted to the number of cycles missed. | |
2270 | */ | |
2271 | static void calc_global_nohz(unsigned long ticks) | |
2272 | { | |
2273 | long delta, active, n; | |
2274 | ||
2275 | if (time_before(jiffies, calc_load_update)) | |
2276 | return; | |
2277 | ||
2278 | /* | |
2279 | * If we crossed a calc_load_update boundary, make sure to fold | |
2280 | * any pending idle changes, the respective CPUs might have | |
2281 | * missed the tick driven calc_load_account_active() update | |
2282 | * due to NO_HZ. | |
2283 | */ | |
2284 | delta = calc_load_fold_idle(); | |
2285 | if (delta) | |
2286 | atomic_long_add(delta, &calc_load_tasks); | |
2287 | ||
2288 | /* | |
2289 | * If we were idle for multiple load cycles, apply them. | |
2290 | */ | |
2291 | if (ticks >= LOAD_FREQ) { | |
2292 | n = ticks / LOAD_FREQ; | |
2293 | ||
2294 | active = atomic_long_read(&calc_load_tasks); | |
2295 | active = active > 0 ? active * FIXED_1 : 0; | |
2296 | ||
2297 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
2298 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2299 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
2300 | ||
2301 | calc_load_update += n * LOAD_FREQ; | |
2302 | } | |
2303 | ||
2304 | /* | |
2305 | * Its possible the remainder of the above division also crosses | |
2306 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
2307 | * which comes after this will take care of that. | |
2308 | * | |
2309 | * Consider us being 11 ticks before a cycle completion, and us | |
2310 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
2311 | * age us 4 cycles, and the test in calc_global_load() will | |
2312 | * pick up the final one. | |
2313 | */ | |
2314 | } | |
74f5187a | 2315 | #else |
029632fb | 2316 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2317 | { |
2318 | } | |
2319 | ||
2320 | static inline long calc_load_fold_idle(void) | |
2321 | { | |
2322 | return 0; | |
2323 | } | |
0f004f5a PZ |
2324 | |
2325 | static void calc_global_nohz(unsigned long ticks) | |
2326 | { | |
2327 | } | |
74f5187a PZ |
2328 | #endif |
2329 | ||
2d02494f TG |
2330 | /** |
2331 | * get_avenrun - get the load average array | |
2332 | * @loads: pointer to dest load array | |
2333 | * @offset: offset to add | |
2334 | * @shift: shift count to shift the result left | |
2335 | * | |
2336 | * These values are estimates at best, so no need for locking. | |
2337 | */ | |
2338 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2339 | { | |
2340 | loads[0] = (avenrun[0] + offset) << shift; | |
2341 | loads[1] = (avenrun[1] + offset) << shift; | |
2342 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2343 | } |
46cb4b7c | 2344 | |
46cb4b7c | 2345 | /* |
dce48a84 TG |
2346 | * calc_load - update the avenrun load estimates 10 ticks after the |
2347 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2348 | */ |
0f004f5a | 2349 | void calc_global_load(unsigned long ticks) |
7835b98b | 2350 | { |
dce48a84 | 2351 | long active; |
1da177e4 | 2352 | |
0f004f5a PZ |
2353 | calc_global_nohz(ticks); |
2354 | ||
2355 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 2356 | return; |
1da177e4 | 2357 | |
dce48a84 TG |
2358 | active = atomic_long_read(&calc_load_tasks); |
2359 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2360 | |
dce48a84 TG |
2361 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2362 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2363 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2364 | |
dce48a84 TG |
2365 | calc_load_update += LOAD_FREQ; |
2366 | } | |
1da177e4 | 2367 | |
dce48a84 | 2368 | /* |
74f5187a PZ |
2369 | * Called from update_cpu_load() to periodically update this CPU's |
2370 | * active count. | |
dce48a84 TG |
2371 | */ |
2372 | static void calc_load_account_active(struct rq *this_rq) | |
2373 | { | |
74f5187a | 2374 | long delta; |
08c183f3 | 2375 | |
74f5187a PZ |
2376 | if (time_before(jiffies, this_rq->calc_load_update)) |
2377 | return; | |
783609c6 | 2378 | |
74f5187a PZ |
2379 | delta = calc_load_fold_active(this_rq); |
2380 | delta += calc_load_fold_idle(); | |
2381 | if (delta) | |
dce48a84 | 2382 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2383 | |
2384 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2385 | } |
2386 | ||
fdf3e95d VP |
2387 | /* |
2388 | * The exact cpuload at various idx values, calculated at every tick would be | |
2389 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2390 | * | |
2391 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2392 | * on nth tick when cpu may be busy, then we have: | |
2393 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2394 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2395 | * | |
2396 | * decay_load_missed() below does efficient calculation of | |
2397 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2398 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2399 | * | |
2400 | * The calculation is approximated on a 128 point scale. | |
2401 | * degrade_zero_ticks is the number of ticks after which load at any | |
2402 | * particular idx is approximated to be zero. | |
2403 | * degrade_factor is a precomputed table, a row for each load idx. | |
2404 | * Each column corresponds to degradation factor for a power of two ticks, | |
2405 | * based on 128 point scale. | |
2406 | * Example: | |
2407 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2408 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2409 | * | |
2410 | * With this power of 2 load factors, we can degrade the load n times | |
2411 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2412 | * n mult/shifts needed by the exact degradation. | |
2413 | */ | |
2414 | #define DEGRADE_SHIFT 7 | |
2415 | static const unsigned char | |
2416 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2417 | static const unsigned char | |
2418 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2419 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2420 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2421 | {96, 72, 40, 12, 1, 0, 0}, | |
2422 | {112, 98, 75, 43, 15, 1, 0}, | |
2423 | {120, 112, 98, 76, 45, 16, 2} }; | |
2424 | ||
2425 | /* | |
2426 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2427 | * would be when CPU is idle and so we just decay the old load without | |
2428 | * adding any new load. | |
2429 | */ | |
2430 | static unsigned long | |
2431 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2432 | { | |
2433 | int j = 0; | |
2434 | ||
2435 | if (!missed_updates) | |
2436 | return load; | |
2437 | ||
2438 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2439 | return 0; | |
2440 | ||
2441 | if (idx == 1) | |
2442 | return load >> missed_updates; | |
2443 | ||
2444 | while (missed_updates) { | |
2445 | if (missed_updates % 2) | |
2446 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2447 | ||
2448 | missed_updates >>= 1; | |
2449 | j++; | |
2450 | } | |
2451 | return load; | |
2452 | } | |
2453 | ||
46cb4b7c | 2454 | /* |
dd41f596 | 2455 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2456 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2457 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2458 | */ |
029632fb | 2459 | void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 2460 | { |
495eca49 | 2461 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
2462 | unsigned long curr_jiffies = jiffies; |
2463 | unsigned long pending_updates; | |
dd41f596 | 2464 | int i, scale; |
46cb4b7c | 2465 | |
dd41f596 | 2466 | this_rq->nr_load_updates++; |
46cb4b7c | 2467 | |
fdf3e95d VP |
2468 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
2469 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2470 | return; | |
2471 | ||
2472 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2473 | this_rq->last_load_update_tick = curr_jiffies; | |
2474 | ||
dd41f596 | 2475 | /* Update our load: */ |
fdf3e95d VP |
2476 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2477 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2478 | unsigned long old_load, new_load; |
7d1e6a9b | 2479 | |
dd41f596 | 2480 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2481 | |
dd41f596 | 2482 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2483 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2484 | new_load = this_load; |
a25707f3 IM |
2485 | /* |
2486 | * Round up the averaging division if load is increasing. This | |
2487 | * prevents us from getting stuck on 9 if the load is 10, for | |
2488 | * example. | |
2489 | */ | |
2490 | if (new_load > old_load) | |
fdf3e95d VP |
2491 | new_load += scale - 1; |
2492 | ||
2493 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2494 | } |
da2b71ed SS |
2495 | |
2496 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2497 | } |
2498 | ||
2499 | static void update_cpu_load_active(struct rq *this_rq) | |
2500 | { | |
2501 | update_cpu_load(this_rq); | |
46cb4b7c | 2502 | |
74f5187a | 2503 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2504 | } |
2505 | ||
dd41f596 | 2506 | #ifdef CONFIG_SMP |
8a0be9ef | 2507 | |
46cb4b7c | 2508 | /* |
38022906 PZ |
2509 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2510 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2511 | */ |
38022906 | 2512 | void sched_exec(void) |
46cb4b7c | 2513 | { |
38022906 | 2514 | struct task_struct *p = current; |
1da177e4 | 2515 | unsigned long flags; |
0017d735 | 2516 | int dest_cpu; |
46cb4b7c | 2517 | |
8f42ced9 | 2518 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2519 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2520 | if (dest_cpu == smp_processor_id()) |
2521 | goto unlock; | |
38022906 | 2522 | |
8f42ced9 | 2523 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2524 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2525 | |
8f42ced9 PZ |
2526 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2527 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2528 | return; |
2529 | } | |
0017d735 | 2530 | unlock: |
8f42ced9 | 2531 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2532 | } |
dd41f596 | 2533 | |
1da177e4 LT |
2534 | #endif |
2535 | ||
1da177e4 | 2536 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2537 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2538 | |
2539 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2540 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2541 | |
2542 | /* | |
c5f8d995 | 2543 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2544 | * @p in case that task is currently running. |
c5f8d995 HS |
2545 | * |
2546 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2547 | */ |
c5f8d995 HS |
2548 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2549 | { | |
2550 | u64 ns = 0; | |
2551 | ||
2552 | if (task_current(rq, p)) { | |
2553 | update_rq_clock(rq); | |
305e6835 | 2554 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2555 | if ((s64)ns < 0) |
2556 | ns = 0; | |
2557 | } | |
2558 | ||
2559 | return ns; | |
2560 | } | |
2561 | ||
bb34d92f | 2562 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2563 | { |
1da177e4 | 2564 | unsigned long flags; |
41b86e9c | 2565 | struct rq *rq; |
bb34d92f | 2566 | u64 ns = 0; |
48f24c4d | 2567 | |
41b86e9c | 2568 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2569 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2570 | task_rq_unlock(rq, p, &flags); |
1508487e | 2571 | |
c5f8d995 HS |
2572 | return ns; |
2573 | } | |
f06febc9 | 2574 | |
c5f8d995 HS |
2575 | /* |
2576 | * Return accounted runtime for the task. | |
2577 | * In case the task is currently running, return the runtime plus current's | |
2578 | * pending runtime that have not been accounted yet. | |
2579 | */ | |
2580 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2581 | { | |
2582 | unsigned long flags; | |
2583 | struct rq *rq; | |
2584 | u64 ns = 0; | |
2585 | ||
2586 | rq = task_rq_lock(p, &flags); | |
2587 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2588 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2589 | |
2590 | return ns; | |
2591 | } | |
48f24c4d | 2592 | |
54c707e9 GC |
2593 | #ifdef CONFIG_CGROUP_CPUACCT |
2594 | struct cgroup_subsys cpuacct_subsys; | |
2595 | struct cpuacct root_cpuacct; | |
2596 | #endif | |
2597 | ||
be726ffd GC |
2598 | static inline void task_group_account_field(struct task_struct *p, int index, |
2599 | u64 tmp) | |
54c707e9 GC |
2600 | { |
2601 | #ifdef CONFIG_CGROUP_CPUACCT | |
2602 | struct kernel_cpustat *kcpustat; | |
2603 | struct cpuacct *ca; | |
2604 | #endif | |
2605 | /* | |
2606 | * Since all updates are sure to touch the root cgroup, we | |
2607 | * get ourselves ahead and touch it first. If the root cgroup | |
2608 | * is the only cgroup, then nothing else should be necessary. | |
2609 | * | |
2610 | */ | |
2611 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
2612 | ||
2613 | #ifdef CONFIG_CGROUP_CPUACCT | |
2614 | if (unlikely(!cpuacct_subsys.active)) | |
2615 | return; | |
2616 | ||
2617 | rcu_read_lock(); | |
2618 | ca = task_ca(p); | |
2619 | while (ca && (ca != &root_cpuacct)) { | |
2620 | kcpustat = this_cpu_ptr(ca->cpustat); | |
2621 | kcpustat->cpustat[index] += tmp; | |
2622 | ca = parent_ca(ca); | |
2623 | } | |
2624 | rcu_read_unlock(); | |
2625 | #endif | |
2626 | } | |
2627 | ||
2628 | ||
1da177e4 LT |
2629 | /* |
2630 | * Account user cpu time to a process. | |
2631 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 2632 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 2633 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 2634 | */ |
457533a7 MS |
2635 | void account_user_time(struct task_struct *p, cputime_t cputime, |
2636 | cputime_t cputime_scaled) | |
1da177e4 | 2637 | { |
3292beb3 | 2638 | int index; |
1da177e4 | 2639 | |
457533a7 | 2640 | /* Add user time to process. */ |
64861634 MS |
2641 | p->utime += cputime; |
2642 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2643 | account_group_user_time(p, cputime); |
1da177e4 | 2644 | |
3292beb3 | 2645 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
ef12fefa | 2646 | |
1da177e4 | 2647 | /* Add user time to cpustat. */ |
612ef28a | 2648 | task_group_account_field(p, index, (__force u64) cputime); |
ef12fefa | 2649 | |
49b5cf34 JL |
2650 | /* Account for user time used */ |
2651 | acct_update_integrals(p); | |
1da177e4 LT |
2652 | } |
2653 | ||
94886b84 LV |
2654 | /* |
2655 | * Account guest cpu time to a process. | |
2656 | * @p: the process that the cpu time gets accounted to | |
2657 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 2658 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 2659 | */ |
457533a7 MS |
2660 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
2661 | cputime_t cputime_scaled) | |
94886b84 | 2662 | { |
3292beb3 | 2663 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
94886b84 | 2664 | |
457533a7 | 2665 | /* Add guest time to process. */ |
64861634 MS |
2666 | p->utime += cputime; |
2667 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2668 | account_group_user_time(p, cputime); |
64861634 | 2669 | p->gtime += cputime; |
94886b84 | 2670 | |
457533a7 | 2671 | /* Add guest time to cpustat. */ |
ce0e7b28 | 2672 | if (TASK_NICE(p) > 0) { |
612ef28a MS |
2673 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
2674 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
ce0e7b28 | 2675 | } else { |
612ef28a MS |
2676 | cpustat[CPUTIME_USER] += (__force u64) cputime; |
2677 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
ce0e7b28 | 2678 | } |
94886b84 LV |
2679 | } |
2680 | ||
70a89a66 VP |
2681 | /* |
2682 | * Account system cpu time to a process and desired cpustat field | |
2683 | * @p: the process that the cpu time gets accounted to | |
2684 | * @cputime: the cpu time spent in kernel space since the last update | |
2685 | * @cputime_scaled: cputime scaled by cpu frequency | |
2686 | * @target_cputime64: pointer to cpustat field that has to be updated | |
2687 | */ | |
2688 | static inline | |
2689 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3292beb3 | 2690 | cputime_t cputime_scaled, int index) |
70a89a66 | 2691 | { |
70a89a66 | 2692 | /* Add system time to process. */ |
64861634 MS |
2693 | p->stime += cputime; |
2694 | p->stimescaled += cputime_scaled; | |
70a89a66 VP |
2695 | account_group_system_time(p, cputime); |
2696 | ||
2697 | /* Add system time to cpustat. */ | |
612ef28a | 2698 | task_group_account_field(p, index, (__force u64) cputime); |
70a89a66 VP |
2699 | |
2700 | /* Account for system time used */ | |
2701 | acct_update_integrals(p); | |
2702 | } | |
2703 | ||
1da177e4 LT |
2704 | /* |
2705 | * Account system cpu time to a process. | |
2706 | * @p: the process that the cpu time gets accounted to | |
2707 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
2708 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 2709 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
2710 | */ |
2711 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 2712 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 | 2713 | { |
3292beb3 | 2714 | int index; |
1da177e4 | 2715 | |
983ed7a6 | 2716 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 2717 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
2718 | return; |
2719 | } | |
94886b84 | 2720 | |
1da177e4 | 2721 | if (hardirq_count() - hardirq_offset) |
3292beb3 | 2722 | index = CPUTIME_IRQ; |
75e1056f | 2723 | else if (in_serving_softirq()) |
3292beb3 | 2724 | index = CPUTIME_SOFTIRQ; |
1da177e4 | 2725 | else |
3292beb3 | 2726 | index = CPUTIME_SYSTEM; |
ef12fefa | 2727 | |
3292beb3 | 2728 | __account_system_time(p, cputime, cputime_scaled, index); |
1da177e4 LT |
2729 | } |
2730 | ||
c66f08be | 2731 | /* |
1da177e4 | 2732 | * Account for involuntary wait time. |
544b4a1f | 2733 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 2734 | */ |
79741dd3 | 2735 | void account_steal_time(cputime_t cputime) |
c66f08be | 2736 | { |
3292beb3 | 2737 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
79741dd3 | 2738 | |
612ef28a | 2739 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; |
c66f08be MN |
2740 | } |
2741 | ||
1da177e4 | 2742 | /* |
79741dd3 MS |
2743 | * Account for idle time. |
2744 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 2745 | */ |
79741dd3 | 2746 | void account_idle_time(cputime_t cputime) |
1da177e4 | 2747 | { |
3292beb3 | 2748 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
70b97a7f | 2749 | struct rq *rq = this_rq(); |
1da177e4 | 2750 | |
79741dd3 | 2751 | if (atomic_read(&rq->nr_iowait) > 0) |
612ef28a | 2752 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; |
79741dd3 | 2753 | else |
612ef28a | 2754 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; |
1da177e4 LT |
2755 | } |
2756 | ||
e6e6685a GC |
2757 | static __always_inline bool steal_account_process_tick(void) |
2758 | { | |
2759 | #ifdef CONFIG_PARAVIRT | |
2760 | if (static_branch(¶virt_steal_enabled)) { | |
2761 | u64 steal, st = 0; | |
2762 | ||
2763 | steal = paravirt_steal_clock(smp_processor_id()); | |
2764 | steal -= this_rq()->prev_steal_time; | |
2765 | ||
2766 | st = steal_ticks(steal); | |
2767 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
2768 | ||
2769 | account_steal_time(st); | |
2770 | return st; | |
2771 | } | |
2772 | #endif | |
2773 | return false; | |
2774 | } | |
2775 | ||
79741dd3 MS |
2776 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
2777 | ||
abb74cef VP |
2778 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
2779 | /* | |
2780 | * Account a tick to a process and cpustat | |
2781 | * @p: the process that the cpu time gets accounted to | |
2782 | * @user_tick: is the tick from userspace | |
2783 | * @rq: the pointer to rq | |
2784 | * | |
2785 | * Tick demultiplexing follows the order | |
2786 | * - pending hardirq update | |
2787 | * - pending softirq update | |
2788 | * - user_time | |
2789 | * - idle_time | |
2790 | * - system time | |
2791 | * - check for guest_time | |
2792 | * - else account as system_time | |
2793 | * | |
2794 | * Check for hardirq is done both for system and user time as there is | |
2795 | * no timer going off while we are on hardirq and hence we may never get an | |
2796 | * opportunity to update it solely in system time. | |
2797 | * p->stime and friends are only updated on system time and not on irq | |
2798 | * softirq as those do not count in task exec_runtime any more. | |
2799 | */ | |
2800 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2801 | struct rq *rq) | |
2802 | { | |
2803 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3292beb3 | 2804 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef | 2805 | |
e6e6685a GC |
2806 | if (steal_account_process_tick()) |
2807 | return; | |
2808 | ||
abb74cef | 2809 | if (irqtime_account_hi_update()) { |
612ef28a | 2810 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; |
abb74cef | 2811 | } else if (irqtime_account_si_update()) { |
612ef28a | 2812 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; |
414bee9b VP |
2813 | } else if (this_cpu_ksoftirqd() == p) { |
2814 | /* | |
2815 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
2816 | * So, we have to handle it separately here. | |
2817 | * Also, p->stime needs to be updated for ksoftirqd. | |
2818 | */ | |
2819 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2820 | CPUTIME_SOFTIRQ); |
abb74cef VP |
2821 | } else if (user_tick) { |
2822 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2823 | } else if (p == rq->idle) { | |
2824 | account_idle_time(cputime_one_jiffy); | |
2825 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
2826 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2827 | } else { | |
2828 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2829 | CPUTIME_SYSTEM); |
abb74cef VP |
2830 | } |
2831 | } | |
2832 | ||
2833 | static void irqtime_account_idle_ticks(int ticks) | |
2834 | { | |
2835 | int i; | |
2836 | struct rq *rq = this_rq(); | |
2837 | ||
2838 | for (i = 0; i < ticks; i++) | |
2839 | irqtime_account_process_tick(current, 0, rq); | |
2840 | } | |
544b4a1f | 2841 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
2842 | static void irqtime_account_idle_ticks(int ticks) {} |
2843 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2844 | struct rq *rq) {} | |
544b4a1f | 2845 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
2846 | |
2847 | /* | |
2848 | * Account a single tick of cpu time. | |
2849 | * @p: the process that the cpu time gets accounted to | |
2850 | * @user_tick: indicates if the tick is a user or a system tick | |
2851 | */ | |
2852 | void account_process_tick(struct task_struct *p, int user_tick) | |
2853 | { | |
a42548a1 | 2854 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
2855 | struct rq *rq = this_rq(); |
2856 | ||
abb74cef VP |
2857 | if (sched_clock_irqtime) { |
2858 | irqtime_account_process_tick(p, user_tick, rq); | |
2859 | return; | |
2860 | } | |
2861 | ||
e6e6685a GC |
2862 | if (steal_account_process_tick()) |
2863 | return; | |
2864 | ||
79741dd3 | 2865 | if (user_tick) |
a42548a1 | 2866 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 2867 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 2868 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
2869 | one_jiffy_scaled); |
2870 | else | |
a42548a1 | 2871 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
2872 | } |
2873 | ||
2874 | /* | |
2875 | * Account multiple ticks of steal time. | |
2876 | * @p: the process from which the cpu time has been stolen | |
2877 | * @ticks: number of stolen ticks | |
2878 | */ | |
2879 | void account_steal_ticks(unsigned long ticks) | |
2880 | { | |
2881 | account_steal_time(jiffies_to_cputime(ticks)); | |
2882 | } | |
2883 | ||
2884 | /* | |
2885 | * Account multiple ticks of idle time. | |
2886 | * @ticks: number of stolen ticks | |
2887 | */ | |
2888 | void account_idle_ticks(unsigned long ticks) | |
2889 | { | |
abb74cef VP |
2890 | |
2891 | if (sched_clock_irqtime) { | |
2892 | irqtime_account_idle_ticks(ticks); | |
2893 | return; | |
2894 | } | |
2895 | ||
79741dd3 | 2896 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
2897 | } |
2898 | ||
79741dd3 MS |
2899 | #endif |
2900 | ||
49048622 BS |
2901 | /* |
2902 | * Use precise platform statistics if available: | |
2903 | */ | |
2904 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 2905 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2906 | { |
d99ca3b9 HS |
2907 | *ut = p->utime; |
2908 | *st = p->stime; | |
49048622 BS |
2909 | } |
2910 | ||
0cf55e1e | 2911 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2912 | { |
0cf55e1e HS |
2913 | struct task_cputime cputime; |
2914 | ||
2915 | thread_group_cputime(p, &cputime); | |
2916 | ||
2917 | *ut = cputime.utime; | |
2918 | *st = cputime.stime; | |
49048622 BS |
2919 | } |
2920 | #else | |
761b1d26 HS |
2921 | |
2922 | #ifndef nsecs_to_cputime | |
b7b20df9 | 2923 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
2924 | #endif |
2925 | ||
d180c5bc | 2926 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2927 | { |
64861634 | 2928 | cputime_t rtime, utime = p->utime, total = utime + p->stime; |
49048622 BS |
2929 | |
2930 | /* | |
2931 | * Use CFS's precise accounting: | |
2932 | */ | |
d180c5bc | 2933 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
2934 | |
2935 | if (total) { | |
64861634 | 2936 | u64 temp = (__force u64) rtime; |
d180c5bc | 2937 | |
64861634 MS |
2938 | temp *= (__force u64) utime; |
2939 | do_div(temp, (__force u32) total); | |
2940 | utime = (__force cputime_t) temp; | |
d180c5bc HS |
2941 | } else |
2942 | utime = rtime; | |
49048622 | 2943 | |
d180c5bc HS |
2944 | /* |
2945 | * Compare with previous values, to keep monotonicity: | |
2946 | */ | |
761b1d26 | 2947 | p->prev_utime = max(p->prev_utime, utime); |
64861634 | 2948 | p->prev_stime = max(p->prev_stime, rtime - p->prev_utime); |
49048622 | 2949 | |
d99ca3b9 HS |
2950 | *ut = p->prev_utime; |
2951 | *st = p->prev_stime; | |
49048622 BS |
2952 | } |
2953 | ||
0cf55e1e HS |
2954 | /* |
2955 | * Must be called with siglock held. | |
2956 | */ | |
2957 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 2958 | { |
0cf55e1e HS |
2959 | struct signal_struct *sig = p->signal; |
2960 | struct task_cputime cputime; | |
2961 | cputime_t rtime, utime, total; | |
49048622 | 2962 | |
0cf55e1e | 2963 | thread_group_cputime(p, &cputime); |
49048622 | 2964 | |
64861634 | 2965 | total = cputime.utime + cputime.stime; |
0cf55e1e | 2966 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); |
49048622 | 2967 | |
0cf55e1e | 2968 | if (total) { |
64861634 | 2969 | u64 temp = (__force u64) rtime; |
49048622 | 2970 | |
64861634 MS |
2971 | temp *= (__force u64) cputime.utime; |
2972 | do_div(temp, (__force u32) total); | |
2973 | utime = (__force cputime_t) temp; | |
0cf55e1e HS |
2974 | } else |
2975 | utime = rtime; | |
2976 | ||
2977 | sig->prev_utime = max(sig->prev_utime, utime); | |
64861634 | 2978 | sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime); |
0cf55e1e HS |
2979 | |
2980 | *ut = sig->prev_utime; | |
2981 | *st = sig->prev_stime; | |
49048622 | 2982 | } |
49048622 | 2983 | #endif |
49048622 | 2984 | |
7835b98b CL |
2985 | /* |
2986 | * This function gets called by the timer code, with HZ frequency. | |
2987 | * We call it with interrupts disabled. | |
7835b98b CL |
2988 | */ |
2989 | void scheduler_tick(void) | |
2990 | { | |
7835b98b CL |
2991 | int cpu = smp_processor_id(); |
2992 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2993 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2994 | |
2995 | sched_clock_tick(); | |
dd41f596 | 2996 | |
05fa785c | 2997 | raw_spin_lock(&rq->lock); |
3e51f33f | 2998 | update_rq_clock(rq); |
fdf3e95d | 2999 | update_cpu_load_active(rq); |
fa85ae24 | 3000 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3001 | raw_spin_unlock(&rq->lock); |
7835b98b | 3002 | |
e9d2b064 | 3003 | perf_event_task_tick(); |
e220d2dc | 3004 | |
e418e1c2 | 3005 | #ifdef CONFIG_SMP |
6eb57e0d | 3006 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 3007 | trigger_load_balance(rq, cpu); |
e418e1c2 | 3008 | #endif |
1da177e4 LT |
3009 | } |
3010 | ||
132380a0 | 3011 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3012 | { |
3013 | if (in_lock_functions(addr)) { | |
3014 | addr = CALLER_ADDR2; | |
3015 | if (in_lock_functions(addr)) | |
3016 | addr = CALLER_ADDR3; | |
3017 | } | |
3018 | return addr; | |
3019 | } | |
1da177e4 | 3020 | |
7e49fcce SR |
3021 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3022 | defined(CONFIG_PREEMPT_TRACER)) | |
3023 | ||
43627582 | 3024 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3025 | { |
6cd8a4bb | 3026 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3027 | /* |
3028 | * Underflow? | |
3029 | */ | |
9a11b49a IM |
3030 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3031 | return; | |
6cd8a4bb | 3032 | #endif |
1da177e4 | 3033 | preempt_count() += val; |
6cd8a4bb | 3034 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3035 | /* |
3036 | * Spinlock count overflowing soon? | |
3037 | */ | |
33859f7f MOS |
3038 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3039 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3040 | #endif |
3041 | if (preempt_count() == val) | |
3042 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3043 | } |
3044 | EXPORT_SYMBOL(add_preempt_count); | |
3045 | ||
43627582 | 3046 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3047 | { |
6cd8a4bb | 3048 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3049 | /* |
3050 | * Underflow? | |
3051 | */ | |
01e3eb82 | 3052 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3053 | return; |
1da177e4 LT |
3054 | /* |
3055 | * Is the spinlock portion underflowing? | |
3056 | */ | |
9a11b49a IM |
3057 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3058 | !(preempt_count() & PREEMPT_MASK))) | |
3059 | return; | |
6cd8a4bb | 3060 | #endif |
9a11b49a | 3061 | |
6cd8a4bb SR |
3062 | if (preempt_count() == val) |
3063 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3064 | preempt_count() -= val; |
3065 | } | |
3066 | EXPORT_SYMBOL(sub_preempt_count); | |
3067 | ||
3068 | #endif | |
3069 | ||
3070 | /* | |
dd41f596 | 3071 | * Print scheduling while atomic bug: |
1da177e4 | 3072 | */ |
dd41f596 | 3073 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3074 | { |
838225b4 SS |
3075 | struct pt_regs *regs = get_irq_regs(); |
3076 | ||
664dfa65 DJ |
3077 | if (oops_in_progress) |
3078 | return; | |
3079 | ||
3df0fc5b PZ |
3080 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3081 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3082 | |
dd41f596 | 3083 | debug_show_held_locks(prev); |
e21f5b15 | 3084 | print_modules(); |
dd41f596 IM |
3085 | if (irqs_disabled()) |
3086 | print_irqtrace_events(prev); | |
838225b4 SS |
3087 | |
3088 | if (regs) | |
3089 | show_regs(regs); | |
3090 | else | |
3091 | dump_stack(); | |
dd41f596 | 3092 | } |
1da177e4 | 3093 | |
dd41f596 IM |
3094 | /* |
3095 | * Various schedule()-time debugging checks and statistics: | |
3096 | */ | |
3097 | static inline void schedule_debug(struct task_struct *prev) | |
3098 | { | |
1da177e4 | 3099 | /* |
41a2d6cf | 3100 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3101 | * schedule() atomically, we ignore that path for now. |
3102 | * Otherwise, whine if we are scheduling when we should not be. | |
3103 | */ | |
3f33a7ce | 3104 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 3105 | __schedule_bug(prev); |
b3fbab05 | 3106 | rcu_sleep_check(); |
dd41f596 | 3107 | |
1da177e4 LT |
3108 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3109 | ||
2d72376b | 3110 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3111 | } |
3112 | ||
6cecd084 | 3113 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3114 | { |
61eadef6 | 3115 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 3116 | update_rq_clock(rq); |
6cecd084 | 3117 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3118 | } |
3119 | ||
dd41f596 IM |
3120 | /* |
3121 | * Pick up the highest-prio task: | |
3122 | */ | |
3123 | static inline struct task_struct * | |
b67802ea | 3124 | pick_next_task(struct rq *rq) |
dd41f596 | 3125 | { |
5522d5d5 | 3126 | const struct sched_class *class; |
dd41f596 | 3127 | struct task_struct *p; |
1da177e4 LT |
3128 | |
3129 | /* | |
dd41f596 IM |
3130 | * Optimization: we know that if all tasks are in |
3131 | * the fair class we can call that function directly: | |
1da177e4 | 3132 | */ |
953bfcd1 | 3133 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 3134 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3135 | if (likely(p)) |
3136 | return p; | |
1da177e4 LT |
3137 | } |
3138 | ||
34f971f6 | 3139 | for_each_class(class) { |
fb8d4724 | 3140 | p = class->pick_next_task(rq); |
dd41f596 IM |
3141 | if (p) |
3142 | return p; | |
dd41f596 | 3143 | } |
34f971f6 PZ |
3144 | |
3145 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3146 | } |
1da177e4 | 3147 | |
dd41f596 | 3148 | /* |
c259e01a | 3149 | * __schedule() is the main scheduler function. |
dd41f596 | 3150 | */ |
c259e01a | 3151 | static void __sched __schedule(void) |
dd41f596 IM |
3152 | { |
3153 | struct task_struct *prev, *next; | |
67ca7bde | 3154 | unsigned long *switch_count; |
dd41f596 | 3155 | struct rq *rq; |
31656519 | 3156 | int cpu; |
dd41f596 | 3157 | |
ff743345 PZ |
3158 | need_resched: |
3159 | preempt_disable(); | |
dd41f596 IM |
3160 | cpu = smp_processor_id(); |
3161 | rq = cpu_rq(cpu); | |
25502a6c | 3162 | rcu_note_context_switch(cpu); |
dd41f596 | 3163 | prev = rq->curr; |
dd41f596 | 3164 | |
dd41f596 | 3165 | schedule_debug(prev); |
1da177e4 | 3166 | |
31656519 | 3167 | if (sched_feat(HRTICK)) |
f333fdc9 | 3168 | hrtick_clear(rq); |
8f4d37ec | 3169 | |
05fa785c | 3170 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3171 | |
246d86b5 | 3172 | switch_count = &prev->nivcsw; |
1da177e4 | 3173 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3174 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3175 | prev->state = TASK_RUNNING; |
21aa9af0 | 3176 | } else { |
2acca55e PZ |
3177 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3178 | prev->on_rq = 0; | |
3179 | ||
21aa9af0 | 3180 | /* |
2acca55e PZ |
3181 | * If a worker went to sleep, notify and ask workqueue |
3182 | * whether it wants to wake up a task to maintain | |
3183 | * concurrency. | |
21aa9af0 TH |
3184 | */ |
3185 | if (prev->flags & PF_WQ_WORKER) { | |
3186 | struct task_struct *to_wakeup; | |
3187 | ||
3188 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3189 | if (to_wakeup) | |
3190 | try_to_wake_up_local(to_wakeup); | |
3191 | } | |
21aa9af0 | 3192 | } |
dd41f596 | 3193 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3194 | } |
3195 | ||
3f029d3c | 3196 | pre_schedule(rq, prev); |
f65eda4f | 3197 | |
dd41f596 | 3198 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3199 | idle_balance(cpu, rq); |
1da177e4 | 3200 | |
df1c99d4 | 3201 | put_prev_task(rq, prev); |
b67802ea | 3202 | next = pick_next_task(rq); |
f26f9aff MG |
3203 | clear_tsk_need_resched(prev); |
3204 | rq->skip_clock_update = 0; | |
1da177e4 | 3205 | |
1da177e4 | 3206 | if (likely(prev != next)) { |
1da177e4 LT |
3207 | rq->nr_switches++; |
3208 | rq->curr = next; | |
3209 | ++*switch_count; | |
3210 | ||
dd41f596 | 3211 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3212 | /* |
246d86b5 ON |
3213 | * The context switch have flipped the stack from under us |
3214 | * and restored the local variables which were saved when | |
3215 | * this task called schedule() in the past. prev == current | |
3216 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3217 | */ |
3218 | cpu = smp_processor_id(); | |
3219 | rq = cpu_rq(cpu); | |
1da177e4 | 3220 | } else |
05fa785c | 3221 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3222 | |
3f029d3c | 3223 | post_schedule(rq); |
1da177e4 | 3224 | |
1da177e4 | 3225 | preempt_enable_no_resched(); |
ff743345 | 3226 | if (need_resched()) |
1da177e4 LT |
3227 | goto need_resched; |
3228 | } | |
c259e01a | 3229 | |
9c40cef2 TG |
3230 | static inline void sched_submit_work(struct task_struct *tsk) |
3231 | { | |
3232 | if (!tsk->state) | |
3233 | return; | |
3234 | /* | |
3235 | * If we are going to sleep and we have plugged IO queued, | |
3236 | * make sure to submit it to avoid deadlocks. | |
3237 | */ | |
3238 | if (blk_needs_flush_plug(tsk)) | |
3239 | blk_schedule_flush_plug(tsk); | |
3240 | } | |
3241 | ||
6ebbe7a0 | 3242 | asmlinkage void __sched schedule(void) |
c259e01a | 3243 | { |
9c40cef2 TG |
3244 | struct task_struct *tsk = current; |
3245 | ||
3246 | sched_submit_work(tsk); | |
c259e01a TG |
3247 | __schedule(); |
3248 | } | |
1da177e4 LT |
3249 | EXPORT_SYMBOL(schedule); |
3250 | ||
c08f7829 | 3251 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3252 | |
c6eb3dda PZ |
3253 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3254 | { | |
c6eb3dda | 3255 | if (lock->owner != owner) |
307bf980 | 3256 | return false; |
0d66bf6d PZ |
3257 | |
3258 | /* | |
c6eb3dda PZ |
3259 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3260 | * lock->owner still matches owner, if that fails, owner might | |
3261 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3262 | * ensures the memory stays valid. | |
0d66bf6d | 3263 | */ |
c6eb3dda | 3264 | barrier(); |
0d66bf6d | 3265 | |
307bf980 | 3266 | return owner->on_cpu; |
c6eb3dda | 3267 | } |
0d66bf6d | 3268 | |
c6eb3dda PZ |
3269 | /* |
3270 | * Look out! "owner" is an entirely speculative pointer | |
3271 | * access and not reliable. | |
3272 | */ | |
3273 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3274 | { | |
3275 | if (!sched_feat(OWNER_SPIN)) | |
3276 | return 0; | |
0d66bf6d | 3277 | |
307bf980 | 3278 | rcu_read_lock(); |
c6eb3dda PZ |
3279 | while (owner_running(lock, owner)) { |
3280 | if (need_resched()) | |
307bf980 | 3281 | break; |
0d66bf6d | 3282 | |
335d7afb | 3283 | arch_mutex_cpu_relax(); |
0d66bf6d | 3284 | } |
307bf980 | 3285 | rcu_read_unlock(); |
4b402210 | 3286 | |
c6eb3dda | 3287 | /* |
307bf980 TG |
3288 | * We break out the loop above on need_resched() and when the |
3289 | * owner changed, which is a sign for heavy contention. Return | |
3290 | * success only when lock->owner is NULL. | |
c6eb3dda | 3291 | */ |
307bf980 | 3292 | return lock->owner == NULL; |
0d66bf6d PZ |
3293 | } |
3294 | #endif | |
3295 | ||
1da177e4 LT |
3296 | #ifdef CONFIG_PREEMPT |
3297 | /* | |
2ed6e34f | 3298 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3299 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3300 | * occur there and call schedule directly. |
3301 | */ | |
d1f74e20 | 3302 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3303 | { |
3304 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3305 | |
1da177e4 LT |
3306 | /* |
3307 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3308 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3309 | */ |
beed33a8 | 3310 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3311 | return; |
3312 | ||
3a5c359a | 3313 | do { |
d1f74e20 | 3314 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3315 | __schedule(); |
d1f74e20 | 3316 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3317 | |
3a5c359a AK |
3318 | /* |
3319 | * Check again in case we missed a preemption opportunity | |
3320 | * between schedule and now. | |
3321 | */ | |
3322 | barrier(); | |
5ed0cec0 | 3323 | } while (need_resched()); |
1da177e4 | 3324 | } |
1da177e4 LT |
3325 | EXPORT_SYMBOL(preempt_schedule); |
3326 | ||
3327 | /* | |
2ed6e34f | 3328 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3329 | * off of irq context. |
3330 | * Note, that this is called and return with irqs disabled. This will | |
3331 | * protect us against recursive calling from irq. | |
3332 | */ | |
3333 | asmlinkage void __sched preempt_schedule_irq(void) | |
3334 | { | |
3335 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3336 | |
2ed6e34f | 3337 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3338 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3339 | ||
3a5c359a AK |
3340 | do { |
3341 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3342 | local_irq_enable(); |
c259e01a | 3343 | __schedule(); |
3a5c359a | 3344 | local_irq_disable(); |
3a5c359a | 3345 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3346 | |
3a5c359a AK |
3347 | /* |
3348 | * Check again in case we missed a preemption opportunity | |
3349 | * between schedule and now. | |
3350 | */ | |
3351 | barrier(); | |
5ed0cec0 | 3352 | } while (need_resched()); |
1da177e4 LT |
3353 | } |
3354 | ||
3355 | #endif /* CONFIG_PREEMPT */ | |
3356 | ||
63859d4f | 3357 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3358 | void *key) |
1da177e4 | 3359 | { |
63859d4f | 3360 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3361 | } |
1da177e4 LT |
3362 | EXPORT_SYMBOL(default_wake_function); |
3363 | ||
3364 | /* | |
41a2d6cf IM |
3365 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3366 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3367 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3368 | * | |
3369 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3370 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3371 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3372 | */ | |
78ddb08f | 3373 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3374 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3375 | { |
2e45874c | 3376 | wait_queue_t *curr, *next; |
1da177e4 | 3377 | |
2e45874c | 3378 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3379 | unsigned flags = curr->flags; |
3380 | ||
63859d4f | 3381 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3382 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3383 | break; |
3384 | } | |
3385 | } | |
3386 | ||
3387 | /** | |
3388 | * __wake_up - wake up threads blocked on a waitqueue. | |
3389 | * @q: the waitqueue | |
3390 | * @mode: which threads | |
3391 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3392 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3393 | * |
3394 | * It may be assumed that this function implies a write memory barrier before | |
3395 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3396 | */ |
7ad5b3a5 | 3397 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3398 | int nr_exclusive, void *key) |
1da177e4 LT |
3399 | { |
3400 | unsigned long flags; | |
3401 | ||
3402 | spin_lock_irqsave(&q->lock, flags); | |
3403 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3404 | spin_unlock_irqrestore(&q->lock, flags); | |
3405 | } | |
1da177e4 LT |
3406 | EXPORT_SYMBOL(__wake_up); |
3407 | ||
3408 | /* | |
3409 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3410 | */ | |
7ad5b3a5 | 3411 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3412 | { |
3413 | __wake_up_common(q, mode, 1, 0, NULL); | |
3414 | } | |
22c43c81 | 3415 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3416 | |
4ede816a DL |
3417 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3418 | { | |
3419 | __wake_up_common(q, mode, 1, 0, key); | |
3420 | } | |
bf294b41 | 3421 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3422 | |
1da177e4 | 3423 | /** |
4ede816a | 3424 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3425 | * @q: the waitqueue |
3426 | * @mode: which threads | |
3427 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3428 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3429 | * |
3430 | * The sync wakeup differs that the waker knows that it will schedule | |
3431 | * away soon, so while the target thread will be woken up, it will not | |
3432 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3433 | * with each other. This can prevent needless bouncing between CPUs. | |
3434 | * | |
3435 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3436 | * |
3437 | * It may be assumed that this function implies a write memory barrier before | |
3438 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3439 | */ |
4ede816a DL |
3440 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3441 | int nr_exclusive, void *key) | |
1da177e4 LT |
3442 | { |
3443 | unsigned long flags; | |
7d478721 | 3444 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3445 | |
3446 | if (unlikely(!q)) | |
3447 | return; | |
3448 | ||
3449 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3450 | wake_flags = 0; |
1da177e4 LT |
3451 | |
3452 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3453 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3454 | spin_unlock_irqrestore(&q->lock, flags); |
3455 | } | |
4ede816a DL |
3456 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3457 | ||
3458 | /* | |
3459 | * __wake_up_sync - see __wake_up_sync_key() | |
3460 | */ | |
3461 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3462 | { | |
3463 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3464 | } | |
1da177e4 LT |
3465 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3466 | ||
65eb3dc6 KD |
3467 | /** |
3468 | * complete: - signals a single thread waiting on this completion | |
3469 | * @x: holds the state of this particular completion | |
3470 | * | |
3471 | * This will wake up a single thread waiting on this completion. Threads will be | |
3472 | * awakened in the same order in which they were queued. | |
3473 | * | |
3474 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3475 | * |
3476 | * It may be assumed that this function implies a write memory barrier before | |
3477 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3478 | */ |
b15136e9 | 3479 | void complete(struct completion *x) |
1da177e4 LT |
3480 | { |
3481 | unsigned long flags; | |
3482 | ||
3483 | spin_lock_irqsave(&x->wait.lock, flags); | |
3484 | x->done++; | |
d9514f6c | 3485 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3486 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3487 | } | |
3488 | EXPORT_SYMBOL(complete); | |
3489 | ||
65eb3dc6 KD |
3490 | /** |
3491 | * complete_all: - signals all threads waiting on this completion | |
3492 | * @x: holds the state of this particular completion | |
3493 | * | |
3494 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3495 | * |
3496 | * It may be assumed that this function implies a write memory barrier before | |
3497 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3498 | */ |
b15136e9 | 3499 | void complete_all(struct completion *x) |
1da177e4 LT |
3500 | { |
3501 | unsigned long flags; | |
3502 | ||
3503 | spin_lock_irqsave(&x->wait.lock, flags); | |
3504 | x->done += UINT_MAX/2; | |
d9514f6c | 3505 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3506 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3507 | } | |
3508 | EXPORT_SYMBOL(complete_all); | |
3509 | ||
8cbbe86d AK |
3510 | static inline long __sched |
3511 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3512 | { |
1da177e4 LT |
3513 | if (!x->done) { |
3514 | DECLARE_WAITQUEUE(wait, current); | |
3515 | ||
a93d2f17 | 3516 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3517 | do { |
94d3d824 | 3518 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3519 | timeout = -ERESTARTSYS; |
3520 | break; | |
8cbbe86d AK |
3521 | } |
3522 | __set_current_state(state); | |
1da177e4 LT |
3523 | spin_unlock_irq(&x->wait.lock); |
3524 | timeout = schedule_timeout(timeout); | |
3525 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3526 | } while (!x->done && timeout); |
1da177e4 | 3527 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3528 | if (!x->done) |
3529 | return timeout; | |
1da177e4 LT |
3530 | } |
3531 | x->done--; | |
ea71a546 | 3532 | return timeout ?: 1; |
1da177e4 | 3533 | } |
1da177e4 | 3534 | |
8cbbe86d AK |
3535 | static long __sched |
3536 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3537 | { |
1da177e4 LT |
3538 | might_sleep(); |
3539 | ||
3540 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3541 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3542 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3543 | return timeout; |
3544 | } | |
1da177e4 | 3545 | |
65eb3dc6 KD |
3546 | /** |
3547 | * wait_for_completion: - waits for completion of a task | |
3548 | * @x: holds the state of this particular completion | |
3549 | * | |
3550 | * This waits to be signaled for completion of a specific task. It is NOT | |
3551 | * interruptible and there is no timeout. | |
3552 | * | |
3553 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3554 | * and interrupt capability. Also see complete(). | |
3555 | */ | |
b15136e9 | 3556 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3557 | { |
3558 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3559 | } |
8cbbe86d | 3560 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3561 | |
65eb3dc6 KD |
3562 | /** |
3563 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3564 | * @x: holds the state of this particular completion | |
3565 | * @timeout: timeout value in jiffies | |
3566 | * | |
3567 | * This waits for either a completion of a specific task to be signaled or for a | |
3568 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3569 | * interruptible. | |
c6dc7f05 BF |
3570 | * |
3571 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3572 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3573 | */ |
b15136e9 | 3574 | unsigned long __sched |
8cbbe86d | 3575 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3576 | { |
8cbbe86d | 3577 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3578 | } |
8cbbe86d | 3579 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3580 | |
65eb3dc6 KD |
3581 | /** |
3582 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3583 | * @x: holds the state of this particular completion | |
3584 | * | |
3585 | * This waits for completion of a specific task to be signaled. It is | |
3586 | * interruptible. | |
c6dc7f05 BF |
3587 | * |
3588 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3589 | */ |
8cbbe86d | 3590 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3591 | { |
51e97990 AK |
3592 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3593 | if (t == -ERESTARTSYS) | |
3594 | return t; | |
3595 | return 0; | |
0fec171c | 3596 | } |
8cbbe86d | 3597 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3598 | |
65eb3dc6 KD |
3599 | /** |
3600 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3601 | * @x: holds the state of this particular completion | |
3602 | * @timeout: timeout value in jiffies | |
3603 | * | |
3604 | * This waits for either a completion of a specific task to be signaled or for a | |
3605 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3606 | * |
3607 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3608 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3609 | */ |
6bf41237 | 3610 | long __sched |
8cbbe86d AK |
3611 | wait_for_completion_interruptible_timeout(struct completion *x, |
3612 | unsigned long timeout) | |
0fec171c | 3613 | { |
8cbbe86d | 3614 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3615 | } |
8cbbe86d | 3616 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3617 | |
65eb3dc6 KD |
3618 | /** |
3619 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3620 | * @x: holds the state of this particular completion | |
3621 | * | |
3622 | * This waits to be signaled for completion of a specific task. It can be | |
3623 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3624 | * |
3625 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3626 | */ |
009e577e MW |
3627 | int __sched wait_for_completion_killable(struct completion *x) |
3628 | { | |
3629 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3630 | if (t == -ERESTARTSYS) | |
3631 | return t; | |
3632 | return 0; | |
3633 | } | |
3634 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3635 | ||
0aa12fb4 SW |
3636 | /** |
3637 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3638 | * @x: holds the state of this particular completion | |
3639 | * @timeout: timeout value in jiffies | |
3640 | * | |
3641 | * This waits for either a completion of a specific task to be | |
3642 | * signaled or for a specified timeout to expire. It can be | |
3643 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3644 | * |
3645 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3646 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3647 | */ |
6bf41237 | 3648 | long __sched |
0aa12fb4 SW |
3649 | wait_for_completion_killable_timeout(struct completion *x, |
3650 | unsigned long timeout) | |
3651 | { | |
3652 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3653 | } | |
3654 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3655 | ||
be4de352 DC |
3656 | /** |
3657 | * try_wait_for_completion - try to decrement a completion without blocking | |
3658 | * @x: completion structure | |
3659 | * | |
3660 | * Returns: 0 if a decrement cannot be done without blocking | |
3661 | * 1 if a decrement succeeded. | |
3662 | * | |
3663 | * If a completion is being used as a counting completion, | |
3664 | * attempt to decrement the counter without blocking. This | |
3665 | * enables us to avoid waiting if the resource the completion | |
3666 | * is protecting is not available. | |
3667 | */ | |
3668 | bool try_wait_for_completion(struct completion *x) | |
3669 | { | |
7539a3b3 | 3670 | unsigned long flags; |
be4de352 DC |
3671 | int ret = 1; |
3672 | ||
7539a3b3 | 3673 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3674 | if (!x->done) |
3675 | ret = 0; | |
3676 | else | |
3677 | x->done--; | |
7539a3b3 | 3678 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3679 | return ret; |
3680 | } | |
3681 | EXPORT_SYMBOL(try_wait_for_completion); | |
3682 | ||
3683 | /** | |
3684 | * completion_done - Test to see if a completion has any waiters | |
3685 | * @x: completion structure | |
3686 | * | |
3687 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3688 | * 1 if there are no waiters. | |
3689 | * | |
3690 | */ | |
3691 | bool completion_done(struct completion *x) | |
3692 | { | |
7539a3b3 | 3693 | unsigned long flags; |
be4de352 DC |
3694 | int ret = 1; |
3695 | ||
7539a3b3 | 3696 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3697 | if (!x->done) |
3698 | ret = 0; | |
7539a3b3 | 3699 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3700 | return ret; |
3701 | } | |
3702 | EXPORT_SYMBOL(completion_done); | |
3703 | ||
8cbbe86d AK |
3704 | static long __sched |
3705 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3706 | { |
0fec171c IM |
3707 | unsigned long flags; |
3708 | wait_queue_t wait; | |
3709 | ||
3710 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3711 | |
8cbbe86d | 3712 | __set_current_state(state); |
1da177e4 | 3713 | |
8cbbe86d AK |
3714 | spin_lock_irqsave(&q->lock, flags); |
3715 | __add_wait_queue(q, &wait); | |
3716 | spin_unlock(&q->lock); | |
3717 | timeout = schedule_timeout(timeout); | |
3718 | spin_lock_irq(&q->lock); | |
3719 | __remove_wait_queue(q, &wait); | |
3720 | spin_unlock_irqrestore(&q->lock, flags); | |
3721 | ||
3722 | return timeout; | |
3723 | } | |
3724 | ||
3725 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3726 | { | |
3727 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3728 | } |
1da177e4 LT |
3729 | EXPORT_SYMBOL(interruptible_sleep_on); |
3730 | ||
0fec171c | 3731 | long __sched |
95cdf3b7 | 3732 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3733 | { |
8cbbe86d | 3734 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3735 | } |
1da177e4 LT |
3736 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3737 | ||
0fec171c | 3738 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3739 | { |
8cbbe86d | 3740 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3741 | } |
1da177e4 LT |
3742 | EXPORT_SYMBOL(sleep_on); |
3743 | ||
0fec171c | 3744 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3745 | { |
8cbbe86d | 3746 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3747 | } |
1da177e4 LT |
3748 | EXPORT_SYMBOL(sleep_on_timeout); |
3749 | ||
b29739f9 IM |
3750 | #ifdef CONFIG_RT_MUTEXES |
3751 | ||
3752 | /* | |
3753 | * rt_mutex_setprio - set the current priority of a task | |
3754 | * @p: task | |
3755 | * @prio: prio value (kernel-internal form) | |
3756 | * | |
3757 | * This function changes the 'effective' priority of a task. It does | |
3758 | * not touch ->normal_prio like __setscheduler(). | |
3759 | * | |
3760 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3761 | */ | |
36c8b586 | 3762 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3763 | { |
83b699ed | 3764 | int oldprio, on_rq, running; |
70b97a7f | 3765 | struct rq *rq; |
83ab0aa0 | 3766 | const struct sched_class *prev_class; |
b29739f9 IM |
3767 | |
3768 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3769 | ||
0122ec5b | 3770 | rq = __task_rq_lock(p); |
b29739f9 | 3771 | |
a8027073 | 3772 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3773 | oldprio = p->prio; |
83ab0aa0 | 3774 | prev_class = p->sched_class; |
fd2f4419 | 3775 | on_rq = p->on_rq; |
051a1d1a | 3776 | running = task_current(rq, p); |
0e1f3483 | 3777 | if (on_rq) |
69be72c1 | 3778 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3779 | if (running) |
3780 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3781 | |
3782 | if (rt_prio(prio)) | |
3783 | p->sched_class = &rt_sched_class; | |
3784 | else | |
3785 | p->sched_class = &fair_sched_class; | |
3786 | ||
b29739f9 IM |
3787 | p->prio = prio; |
3788 | ||
0e1f3483 HS |
3789 | if (running) |
3790 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3791 | if (on_rq) |
371fd7e7 | 3792 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3793 | |
da7a735e | 3794 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3795 | __task_rq_unlock(rq); |
b29739f9 IM |
3796 | } |
3797 | ||
3798 | #endif | |
3799 | ||
36c8b586 | 3800 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3801 | { |
dd41f596 | 3802 | int old_prio, delta, on_rq; |
1da177e4 | 3803 | unsigned long flags; |
70b97a7f | 3804 | struct rq *rq; |
1da177e4 LT |
3805 | |
3806 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3807 | return; | |
3808 | /* | |
3809 | * We have to be careful, if called from sys_setpriority(), | |
3810 | * the task might be in the middle of scheduling on another CPU. | |
3811 | */ | |
3812 | rq = task_rq_lock(p, &flags); | |
3813 | /* | |
3814 | * The RT priorities are set via sched_setscheduler(), but we still | |
3815 | * allow the 'normal' nice value to be set - but as expected | |
3816 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3817 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3818 | */ |
e05606d3 | 3819 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3820 | p->static_prio = NICE_TO_PRIO(nice); |
3821 | goto out_unlock; | |
3822 | } | |
fd2f4419 | 3823 | on_rq = p->on_rq; |
c09595f6 | 3824 | if (on_rq) |
69be72c1 | 3825 | dequeue_task(rq, p, 0); |
1da177e4 | 3826 | |
1da177e4 | 3827 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3828 | set_load_weight(p); |
b29739f9 IM |
3829 | old_prio = p->prio; |
3830 | p->prio = effective_prio(p); | |
3831 | delta = p->prio - old_prio; | |
1da177e4 | 3832 | |
dd41f596 | 3833 | if (on_rq) { |
371fd7e7 | 3834 | enqueue_task(rq, p, 0); |
1da177e4 | 3835 | /* |
d5f9f942 AM |
3836 | * If the task increased its priority or is running and |
3837 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3838 | */ |
d5f9f942 | 3839 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3840 | resched_task(rq->curr); |
3841 | } | |
3842 | out_unlock: | |
0122ec5b | 3843 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3844 | } |
1da177e4 LT |
3845 | EXPORT_SYMBOL(set_user_nice); |
3846 | ||
e43379f1 MM |
3847 | /* |
3848 | * can_nice - check if a task can reduce its nice value | |
3849 | * @p: task | |
3850 | * @nice: nice value | |
3851 | */ | |
36c8b586 | 3852 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3853 | { |
024f4747 MM |
3854 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3855 | int nice_rlim = 20 - nice; | |
48f24c4d | 3856 | |
78d7d407 | 3857 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3858 | capable(CAP_SYS_NICE)); |
3859 | } | |
3860 | ||
1da177e4 LT |
3861 | #ifdef __ARCH_WANT_SYS_NICE |
3862 | ||
3863 | /* | |
3864 | * sys_nice - change the priority of the current process. | |
3865 | * @increment: priority increment | |
3866 | * | |
3867 | * sys_setpriority is a more generic, but much slower function that | |
3868 | * does similar things. | |
3869 | */ | |
5add95d4 | 3870 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3871 | { |
48f24c4d | 3872 | long nice, retval; |
1da177e4 LT |
3873 | |
3874 | /* | |
3875 | * Setpriority might change our priority at the same moment. | |
3876 | * We don't have to worry. Conceptually one call occurs first | |
3877 | * and we have a single winner. | |
3878 | */ | |
e43379f1 MM |
3879 | if (increment < -40) |
3880 | increment = -40; | |
1da177e4 LT |
3881 | if (increment > 40) |
3882 | increment = 40; | |
3883 | ||
2b8f836f | 3884 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3885 | if (nice < -20) |
3886 | nice = -20; | |
3887 | if (nice > 19) | |
3888 | nice = 19; | |
3889 | ||
e43379f1 MM |
3890 | if (increment < 0 && !can_nice(current, nice)) |
3891 | return -EPERM; | |
3892 | ||
1da177e4 LT |
3893 | retval = security_task_setnice(current, nice); |
3894 | if (retval) | |
3895 | return retval; | |
3896 | ||
3897 | set_user_nice(current, nice); | |
3898 | return 0; | |
3899 | } | |
3900 | ||
3901 | #endif | |
3902 | ||
3903 | /** | |
3904 | * task_prio - return the priority value of a given task. | |
3905 | * @p: the task in question. | |
3906 | * | |
3907 | * This is the priority value as seen by users in /proc. | |
3908 | * RT tasks are offset by -200. Normal tasks are centered | |
3909 | * around 0, value goes from -16 to +15. | |
3910 | */ | |
36c8b586 | 3911 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3912 | { |
3913 | return p->prio - MAX_RT_PRIO; | |
3914 | } | |
3915 | ||
3916 | /** | |
3917 | * task_nice - return the nice value of a given task. | |
3918 | * @p: the task in question. | |
3919 | */ | |
36c8b586 | 3920 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3921 | { |
3922 | return TASK_NICE(p); | |
3923 | } | |
150d8bed | 3924 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3925 | |
3926 | /** | |
3927 | * idle_cpu - is a given cpu idle currently? | |
3928 | * @cpu: the processor in question. | |
3929 | */ | |
3930 | int idle_cpu(int cpu) | |
3931 | { | |
908a3283 TG |
3932 | struct rq *rq = cpu_rq(cpu); |
3933 | ||
3934 | if (rq->curr != rq->idle) | |
3935 | return 0; | |
3936 | ||
3937 | if (rq->nr_running) | |
3938 | return 0; | |
3939 | ||
3940 | #ifdef CONFIG_SMP | |
3941 | if (!llist_empty(&rq->wake_list)) | |
3942 | return 0; | |
3943 | #endif | |
3944 | ||
3945 | return 1; | |
1da177e4 LT |
3946 | } |
3947 | ||
1da177e4 LT |
3948 | /** |
3949 | * idle_task - return the idle task for a given cpu. | |
3950 | * @cpu: the processor in question. | |
3951 | */ | |
36c8b586 | 3952 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3953 | { |
3954 | return cpu_rq(cpu)->idle; | |
3955 | } | |
3956 | ||
3957 | /** | |
3958 | * find_process_by_pid - find a process with a matching PID value. | |
3959 | * @pid: the pid in question. | |
3960 | */ | |
a9957449 | 3961 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3962 | { |
228ebcbe | 3963 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3964 | } |
3965 | ||
3966 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3967 | static void |
3968 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3969 | { |
1da177e4 LT |
3970 | p->policy = policy; |
3971 | p->rt_priority = prio; | |
b29739f9 IM |
3972 | p->normal_prio = normal_prio(p); |
3973 | /* we are holding p->pi_lock already */ | |
3974 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3975 | if (rt_prio(p->prio)) |
3976 | p->sched_class = &rt_sched_class; | |
3977 | else | |
3978 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3979 | set_load_weight(p); |
1da177e4 LT |
3980 | } |
3981 | ||
c69e8d9c DH |
3982 | /* |
3983 | * check the target process has a UID that matches the current process's | |
3984 | */ | |
3985 | static bool check_same_owner(struct task_struct *p) | |
3986 | { | |
3987 | const struct cred *cred = current_cred(), *pcred; | |
3988 | bool match; | |
3989 | ||
3990 | rcu_read_lock(); | |
3991 | pcred = __task_cred(p); | |
b0e77598 SH |
3992 | if (cred->user->user_ns == pcred->user->user_ns) |
3993 | match = (cred->euid == pcred->euid || | |
3994 | cred->euid == pcred->uid); | |
3995 | else | |
3996 | match = false; | |
c69e8d9c DH |
3997 | rcu_read_unlock(); |
3998 | return match; | |
3999 | } | |
4000 | ||
961ccddd | 4001 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4002 | const struct sched_param *param, bool user) |
1da177e4 | 4003 | { |
83b699ed | 4004 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4005 | unsigned long flags; |
83ab0aa0 | 4006 | const struct sched_class *prev_class; |
70b97a7f | 4007 | struct rq *rq; |
ca94c442 | 4008 | int reset_on_fork; |
1da177e4 | 4009 | |
66e5393a SR |
4010 | /* may grab non-irq protected spin_locks */ |
4011 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4012 | recheck: |
4013 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4014 | if (policy < 0) { |
4015 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4016 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4017 | } else { |
4018 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4019 | policy &= ~SCHED_RESET_ON_FORK; | |
4020 | ||
4021 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4022 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4023 | policy != SCHED_IDLE) | |
4024 | return -EINVAL; | |
4025 | } | |
4026 | ||
1da177e4 LT |
4027 | /* |
4028 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4029 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4030 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4031 | */ |
4032 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4033 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4034 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4035 | return -EINVAL; |
e05606d3 | 4036 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4037 | return -EINVAL; |
4038 | ||
37e4ab3f OC |
4039 | /* |
4040 | * Allow unprivileged RT tasks to decrease priority: | |
4041 | */ | |
961ccddd | 4042 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4043 | if (rt_policy(policy)) { |
a44702e8 ON |
4044 | unsigned long rlim_rtprio = |
4045 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4046 | |
4047 | /* can't set/change the rt policy */ | |
4048 | if (policy != p->policy && !rlim_rtprio) | |
4049 | return -EPERM; | |
4050 | ||
4051 | /* can't increase priority */ | |
4052 | if (param->sched_priority > p->rt_priority && | |
4053 | param->sched_priority > rlim_rtprio) | |
4054 | return -EPERM; | |
4055 | } | |
c02aa73b | 4056 | |
dd41f596 | 4057 | /* |
c02aa73b DH |
4058 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4059 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4060 | */ |
c02aa73b DH |
4061 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4062 | if (!can_nice(p, TASK_NICE(p))) | |
4063 | return -EPERM; | |
4064 | } | |
5fe1d75f | 4065 | |
37e4ab3f | 4066 | /* can't change other user's priorities */ |
c69e8d9c | 4067 | if (!check_same_owner(p)) |
37e4ab3f | 4068 | return -EPERM; |
ca94c442 LP |
4069 | |
4070 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4071 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4072 | return -EPERM; | |
37e4ab3f | 4073 | } |
1da177e4 | 4074 | |
725aad24 | 4075 | if (user) { |
b0ae1981 | 4076 | retval = security_task_setscheduler(p); |
725aad24 JF |
4077 | if (retval) |
4078 | return retval; | |
4079 | } | |
4080 | ||
b29739f9 IM |
4081 | /* |
4082 | * make sure no PI-waiters arrive (or leave) while we are | |
4083 | * changing the priority of the task: | |
0122ec5b | 4084 | * |
25985edc | 4085 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4086 | * runqueue lock must be held. |
4087 | */ | |
0122ec5b | 4088 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 4089 | |
34f971f6 PZ |
4090 | /* |
4091 | * Changing the policy of the stop threads its a very bad idea | |
4092 | */ | |
4093 | if (p == rq->stop) { | |
0122ec5b | 4094 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
4095 | return -EINVAL; |
4096 | } | |
4097 | ||
a51e9198 DF |
4098 | /* |
4099 | * If not changing anything there's no need to proceed further: | |
4100 | */ | |
4101 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
4102 | param->sched_priority == p->rt_priority))) { | |
4103 | ||
4104 | __task_rq_unlock(rq); | |
4105 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4106 | return 0; | |
4107 | } | |
4108 | ||
dc61b1d6 PZ |
4109 | #ifdef CONFIG_RT_GROUP_SCHED |
4110 | if (user) { | |
4111 | /* | |
4112 | * Do not allow realtime tasks into groups that have no runtime | |
4113 | * assigned. | |
4114 | */ | |
4115 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4116 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4117 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4118 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4119 | return -EPERM; |
4120 | } | |
4121 | } | |
4122 | #endif | |
4123 | ||
1da177e4 LT |
4124 | /* recheck policy now with rq lock held */ |
4125 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4126 | policy = oldpolicy = -1; | |
0122ec5b | 4127 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4128 | goto recheck; |
4129 | } | |
fd2f4419 | 4130 | on_rq = p->on_rq; |
051a1d1a | 4131 | running = task_current(rq, p); |
0e1f3483 | 4132 | if (on_rq) |
4ca9b72b | 4133 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4134 | if (running) |
4135 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4136 | |
ca94c442 LP |
4137 | p->sched_reset_on_fork = reset_on_fork; |
4138 | ||
1da177e4 | 4139 | oldprio = p->prio; |
83ab0aa0 | 4140 | prev_class = p->sched_class; |
dd41f596 | 4141 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4142 | |
0e1f3483 HS |
4143 | if (running) |
4144 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4145 | if (on_rq) |
4ca9b72b | 4146 | enqueue_task(rq, p, 0); |
cb469845 | 4147 | |
da7a735e | 4148 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4149 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4150 | |
95e02ca9 TG |
4151 | rt_mutex_adjust_pi(p); |
4152 | ||
1da177e4 LT |
4153 | return 0; |
4154 | } | |
961ccddd RR |
4155 | |
4156 | /** | |
4157 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4158 | * @p: the task in question. | |
4159 | * @policy: new policy. | |
4160 | * @param: structure containing the new RT priority. | |
4161 | * | |
4162 | * NOTE that the task may be already dead. | |
4163 | */ | |
4164 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4165 | const struct sched_param *param) |
961ccddd RR |
4166 | { |
4167 | return __sched_setscheduler(p, policy, param, true); | |
4168 | } | |
1da177e4 LT |
4169 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4170 | ||
961ccddd RR |
4171 | /** |
4172 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4173 | * @p: the task in question. | |
4174 | * @policy: new policy. | |
4175 | * @param: structure containing the new RT priority. | |
4176 | * | |
4177 | * Just like sched_setscheduler, only don't bother checking if the | |
4178 | * current context has permission. For example, this is needed in | |
4179 | * stop_machine(): we create temporary high priority worker threads, | |
4180 | * but our caller might not have that capability. | |
4181 | */ | |
4182 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4183 | const struct sched_param *param) |
961ccddd RR |
4184 | { |
4185 | return __sched_setscheduler(p, policy, param, false); | |
4186 | } | |
4187 | ||
95cdf3b7 IM |
4188 | static int |
4189 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4190 | { |
1da177e4 LT |
4191 | struct sched_param lparam; |
4192 | struct task_struct *p; | |
36c8b586 | 4193 | int retval; |
1da177e4 LT |
4194 | |
4195 | if (!param || pid < 0) | |
4196 | return -EINVAL; | |
4197 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4198 | return -EFAULT; | |
5fe1d75f ON |
4199 | |
4200 | rcu_read_lock(); | |
4201 | retval = -ESRCH; | |
1da177e4 | 4202 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4203 | if (p != NULL) |
4204 | retval = sched_setscheduler(p, policy, &lparam); | |
4205 | rcu_read_unlock(); | |
36c8b586 | 4206 | |
1da177e4 LT |
4207 | return retval; |
4208 | } | |
4209 | ||
4210 | /** | |
4211 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4212 | * @pid: the pid in question. | |
4213 | * @policy: new policy. | |
4214 | * @param: structure containing the new RT priority. | |
4215 | */ | |
5add95d4 HC |
4216 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4217 | struct sched_param __user *, param) | |
1da177e4 | 4218 | { |
c21761f1 JB |
4219 | /* negative values for policy are not valid */ |
4220 | if (policy < 0) | |
4221 | return -EINVAL; | |
4222 | ||
1da177e4 LT |
4223 | return do_sched_setscheduler(pid, policy, param); |
4224 | } | |
4225 | ||
4226 | /** | |
4227 | * sys_sched_setparam - set/change the RT priority of a thread | |
4228 | * @pid: the pid in question. | |
4229 | * @param: structure containing the new RT priority. | |
4230 | */ | |
5add95d4 | 4231 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4232 | { |
4233 | return do_sched_setscheduler(pid, -1, param); | |
4234 | } | |
4235 | ||
4236 | /** | |
4237 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4238 | * @pid: the pid in question. | |
4239 | */ | |
5add95d4 | 4240 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4241 | { |
36c8b586 | 4242 | struct task_struct *p; |
3a5c359a | 4243 | int retval; |
1da177e4 LT |
4244 | |
4245 | if (pid < 0) | |
3a5c359a | 4246 | return -EINVAL; |
1da177e4 LT |
4247 | |
4248 | retval = -ESRCH; | |
5fe85be0 | 4249 | rcu_read_lock(); |
1da177e4 LT |
4250 | p = find_process_by_pid(pid); |
4251 | if (p) { | |
4252 | retval = security_task_getscheduler(p); | |
4253 | if (!retval) | |
ca94c442 LP |
4254 | retval = p->policy |
4255 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4256 | } |
5fe85be0 | 4257 | rcu_read_unlock(); |
1da177e4 LT |
4258 | return retval; |
4259 | } | |
4260 | ||
4261 | /** | |
ca94c442 | 4262 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4263 | * @pid: the pid in question. |
4264 | * @param: structure containing the RT priority. | |
4265 | */ | |
5add95d4 | 4266 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4267 | { |
4268 | struct sched_param lp; | |
36c8b586 | 4269 | struct task_struct *p; |
3a5c359a | 4270 | int retval; |
1da177e4 LT |
4271 | |
4272 | if (!param || pid < 0) | |
3a5c359a | 4273 | return -EINVAL; |
1da177e4 | 4274 | |
5fe85be0 | 4275 | rcu_read_lock(); |
1da177e4 LT |
4276 | p = find_process_by_pid(pid); |
4277 | retval = -ESRCH; | |
4278 | if (!p) | |
4279 | goto out_unlock; | |
4280 | ||
4281 | retval = security_task_getscheduler(p); | |
4282 | if (retval) | |
4283 | goto out_unlock; | |
4284 | ||
4285 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4286 | rcu_read_unlock(); |
1da177e4 LT |
4287 | |
4288 | /* | |
4289 | * This one might sleep, we cannot do it with a spinlock held ... | |
4290 | */ | |
4291 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4292 | ||
1da177e4 LT |
4293 | return retval; |
4294 | ||
4295 | out_unlock: | |
5fe85be0 | 4296 | rcu_read_unlock(); |
1da177e4 LT |
4297 | return retval; |
4298 | } | |
4299 | ||
96f874e2 | 4300 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4301 | { |
5a16f3d3 | 4302 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4303 | struct task_struct *p; |
4304 | int retval; | |
1da177e4 | 4305 | |
95402b38 | 4306 | get_online_cpus(); |
23f5d142 | 4307 | rcu_read_lock(); |
1da177e4 LT |
4308 | |
4309 | p = find_process_by_pid(pid); | |
4310 | if (!p) { | |
23f5d142 | 4311 | rcu_read_unlock(); |
95402b38 | 4312 | put_online_cpus(); |
1da177e4 LT |
4313 | return -ESRCH; |
4314 | } | |
4315 | ||
23f5d142 | 4316 | /* Prevent p going away */ |
1da177e4 | 4317 | get_task_struct(p); |
23f5d142 | 4318 | rcu_read_unlock(); |
1da177e4 | 4319 | |
5a16f3d3 RR |
4320 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4321 | retval = -ENOMEM; | |
4322 | goto out_put_task; | |
4323 | } | |
4324 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4325 | retval = -ENOMEM; | |
4326 | goto out_free_cpus_allowed; | |
4327 | } | |
1da177e4 | 4328 | retval = -EPERM; |
f1c84dae | 4329 | if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE)) |
1da177e4 LT |
4330 | goto out_unlock; |
4331 | ||
b0ae1981 | 4332 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4333 | if (retval) |
4334 | goto out_unlock; | |
4335 | ||
5a16f3d3 RR |
4336 | cpuset_cpus_allowed(p, cpus_allowed); |
4337 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4338 | again: |
5a16f3d3 | 4339 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4340 | |
8707d8b8 | 4341 | if (!retval) { |
5a16f3d3 RR |
4342 | cpuset_cpus_allowed(p, cpus_allowed); |
4343 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4344 | /* |
4345 | * We must have raced with a concurrent cpuset | |
4346 | * update. Just reset the cpus_allowed to the | |
4347 | * cpuset's cpus_allowed | |
4348 | */ | |
5a16f3d3 | 4349 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4350 | goto again; |
4351 | } | |
4352 | } | |
1da177e4 | 4353 | out_unlock: |
5a16f3d3 RR |
4354 | free_cpumask_var(new_mask); |
4355 | out_free_cpus_allowed: | |
4356 | free_cpumask_var(cpus_allowed); | |
4357 | out_put_task: | |
1da177e4 | 4358 | put_task_struct(p); |
95402b38 | 4359 | put_online_cpus(); |
1da177e4 LT |
4360 | return retval; |
4361 | } | |
4362 | ||
4363 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4364 | struct cpumask *new_mask) |
1da177e4 | 4365 | { |
96f874e2 RR |
4366 | if (len < cpumask_size()) |
4367 | cpumask_clear(new_mask); | |
4368 | else if (len > cpumask_size()) | |
4369 | len = cpumask_size(); | |
4370 | ||
1da177e4 LT |
4371 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4372 | } | |
4373 | ||
4374 | /** | |
4375 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4376 | * @pid: pid of the process | |
4377 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4378 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4379 | */ | |
5add95d4 HC |
4380 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4381 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4382 | { |
5a16f3d3 | 4383 | cpumask_var_t new_mask; |
1da177e4 LT |
4384 | int retval; |
4385 | ||
5a16f3d3 RR |
4386 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4387 | return -ENOMEM; | |
1da177e4 | 4388 | |
5a16f3d3 RR |
4389 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4390 | if (retval == 0) | |
4391 | retval = sched_setaffinity(pid, new_mask); | |
4392 | free_cpumask_var(new_mask); | |
4393 | return retval; | |
1da177e4 LT |
4394 | } |
4395 | ||
96f874e2 | 4396 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4397 | { |
36c8b586 | 4398 | struct task_struct *p; |
31605683 | 4399 | unsigned long flags; |
1da177e4 | 4400 | int retval; |
1da177e4 | 4401 | |
95402b38 | 4402 | get_online_cpus(); |
23f5d142 | 4403 | rcu_read_lock(); |
1da177e4 LT |
4404 | |
4405 | retval = -ESRCH; | |
4406 | p = find_process_by_pid(pid); | |
4407 | if (!p) | |
4408 | goto out_unlock; | |
4409 | ||
e7834f8f DQ |
4410 | retval = security_task_getscheduler(p); |
4411 | if (retval) | |
4412 | goto out_unlock; | |
4413 | ||
013fdb80 | 4414 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4415 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4416 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4417 | |
4418 | out_unlock: | |
23f5d142 | 4419 | rcu_read_unlock(); |
95402b38 | 4420 | put_online_cpus(); |
1da177e4 | 4421 | |
9531b62f | 4422 | return retval; |
1da177e4 LT |
4423 | } |
4424 | ||
4425 | /** | |
4426 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4427 | * @pid: pid of the process | |
4428 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4429 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4430 | */ | |
5add95d4 HC |
4431 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4432 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4433 | { |
4434 | int ret; | |
f17c8607 | 4435 | cpumask_var_t mask; |
1da177e4 | 4436 | |
84fba5ec | 4437 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4438 | return -EINVAL; |
4439 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4440 | return -EINVAL; |
4441 | ||
f17c8607 RR |
4442 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4443 | return -ENOMEM; | |
1da177e4 | 4444 | |
f17c8607 RR |
4445 | ret = sched_getaffinity(pid, mask); |
4446 | if (ret == 0) { | |
8bc037fb | 4447 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4448 | |
4449 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4450 | ret = -EFAULT; |
4451 | else | |
cd3d8031 | 4452 | ret = retlen; |
f17c8607 RR |
4453 | } |
4454 | free_cpumask_var(mask); | |
1da177e4 | 4455 | |
f17c8607 | 4456 | return ret; |
1da177e4 LT |
4457 | } |
4458 | ||
4459 | /** | |
4460 | * sys_sched_yield - yield the current processor to other threads. | |
4461 | * | |
dd41f596 IM |
4462 | * This function yields the current CPU to other tasks. If there are no |
4463 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4464 | */ |
5add95d4 | 4465 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4466 | { |
70b97a7f | 4467 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4468 | |
2d72376b | 4469 | schedstat_inc(rq, yld_count); |
4530d7ab | 4470 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4471 | |
4472 | /* | |
4473 | * Since we are going to call schedule() anyway, there's | |
4474 | * no need to preempt or enable interrupts: | |
4475 | */ | |
4476 | __release(rq->lock); | |
8a25d5de | 4477 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4478 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4479 | preempt_enable_no_resched(); |
4480 | ||
4481 | schedule(); | |
4482 | ||
4483 | return 0; | |
4484 | } | |
4485 | ||
d86ee480 PZ |
4486 | static inline int should_resched(void) |
4487 | { | |
4488 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4489 | } | |
4490 | ||
e7b38404 | 4491 | static void __cond_resched(void) |
1da177e4 | 4492 | { |
e7aaaa69 | 4493 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4494 | __schedule(); |
e7aaaa69 | 4495 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4496 | } |
4497 | ||
02b67cc3 | 4498 | int __sched _cond_resched(void) |
1da177e4 | 4499 | { |
d86ee480 | 4500 | if (should_resched()) { |
1da177e4 LT |
4501 | __cond_resched(); |
4502 | return 1; | |
4503 | } | |
4504 | return 0; | |
4505 | } | |
02b67cc3 | 4506 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4507 | |
4508 | /* | |
613afbf8 | 4509 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4510 | * call schedule, and on return reacquire the lock. |
4511 | * | |
41a2d6cf | 4512 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4513 | * operations here to prevent schedule() from being called twice (once via |
4514 | * spin_unlock(), once by hand). | |
4515 | */ | |
613afbf8 | 4516 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4517 | { |
d86ee480 | 4518 | int resched = should_resched(); |
6df3cecb JK |
4519 | int ret = 0; |
4520 | ||
f607c668 PZ |
4521 | lockdep_assert_held(lock); |
4522 | ||
95c354fe | 4523 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4524 | spin_unlock(lock); |
d86ee480 | 4525 | if (resched) |
95c354fe NP |
4526 | __cond_resched(); |
4527 | else | |
4528 | cpu_relax(); | |
6df3cecb | 4529 | ret = 1; |
1da177e4 | 4530 | spin_lock(lock); |
1da177e4 | 4531 | } |
6df3cecb | 4532 | return ret; |
1da177e4 | 4533 | } |
613afbf8 | 4534 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4535 | |
613afbf8 | 4536 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4537 | { |
4538 | BUG_ON(!in_softirq()); | |
4539 | ||
d86ee480 | 4540 | if (should_resched()) { |
98d82567 | 4541 | local_bh_enable(); |
1da177e4 LT |
4542 | __cond_resched(); |
4543 | local_bh_disable(); | |
4544 | return 1; | |
4545 | } | |
4546 | return 0; | |
4547 | } | |
613afbf8 | 4548 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4549 | |
1da177e4 LT |
4550 | /** |
4551 | * yield - yield the current processor to other threads. | |
4552 | * | |
72fd4a35 | 4553 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4554 | * thread runnable and calls sys_sched_yield(). |
4555 | */ | |
4556 | void __sched yield(void) | |
4557 | { | |
4558 | set_current_state(TASK_RUNNING); | |
4559 | sys_sched_yield(); | |
4560 | } | |
1da177e4 LT |
4561 | EXPORT_SYMBOL(yield); |
4562 | ||
d95f4122 MG |
4563 | /** |
4564 | * yield_to - yield the current processor to another thread in | |
4565 | * your thread group, or accelerate that thread toward the | |
4566 | * processor it's on. | |
16addf95 RD |
4567 | * @p: target task |
4568 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4569 | * |
4570 | * It's the caller's job to ensure that the target task struct | |
4571 | * can't go away on us before we can do any checks. | |
4572 | * | |
4573 | * Returns true if we indeed boosted the target task. | |
4574 | */ | |
4575 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4576 | { | |
4577 | struct task_struct *curr = current; | |
4578 | struct rq *rq, *p_rq; | |
4579 | unsigned long flags; | |
4580 | bool yielded = 0; | |
4581 | ||
4582 | local_irq_save(flags); | |
4583 | rq = this_rq(); | |
4584 | ||
4585 | again: | |
4586 | p_rq = task_rq(p); | |
4587 | double_rq_lock(rq, p_rq); | |
4588 | while (task_rq(p) != p_rq) { | |
4589 | double_rq_unlock(rq, p_rq); | |
4590 | goto again; | |
4591 | } | |
4592 | ||
4593 | if (!curr->sched_class->yield_to_task) | |
4594 | goto out; | |
4595 | ||
4596 | if (curr->sched_class != p->sched_class) | |
4597 | goto out; | |
4598 | ||
4599 | if (task_running(p_rq, p) || p->state) | |
4600 | goto out; | |
4601 | ||
4602 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4603 | if (yielded) { |
d95f4122 | 4604 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4605 | /* |
4606 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4607 | * fairness. | |
4608 | */ | |
4609 | if (preempt && rq != p_rq) | |
4610 | resched_task(p_rq->curr); | |
916671c0 MG |
4611 | } else { |
4612 | /* | |
4613 | * We might have set it in task_yield_fair(), but are | |
4614 | * not going to schedule(), so don't want to skip | |
4615 | * the next update. | |
4616 | */ | |
4617 | rq->skip_clock_update = 0; | |
6d1cafd8 | 4618 | } |
d95f4122 MG |
4619 | |
4620 | out: | |
4621 | double_rq_unlock(rq, p_rq); | |
4622 | local_irq_restore(flags); | |
4623 | ||
4624 | if (yielded) | |
4625 | schedule(); | |
4626 | ||
4627 | return yielded; | |
4628 | } | |
4629 | EXPORT_SYMBOL_GPL(yield_to); | |
4630 | ||
1da177e4 | 4631 | /* |
41a2d6cf | 4632 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4633 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4634 | */ |
4635 | void __sched io_schedule(void) | |
4636 | { | |
54d35f29 | 4637 | struct rq *rq = raw_rq(); |
1da177e4 | 4638 | |
0ff92245 | 4639 | delayacct_blkio_start(); |
1da177e4 | 4640 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4641 | blk_flush_plug(current); |
8f0dfc34 | 4642 | current->in_iowait = 1; |
1da177e4 | 4643 | schedule(); |
8f0dfc34 | 4644 | current->in_iowait = 0; |
1da177e4 | 4645 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4646 | delayacct_blkio_end(); |
1da177e4 | 4647 | } |
1da177e4 LT |
4648 | EXPORT_SYMBOL(io_schedule); |
4649 | ||
4650 | long __sched io_schedule_timeout(long timeout) | |
4651 | { | |
54d35f29 | 4652 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4653 | long ret; |
4654 | ||
0ff92245 | 4655 | delayacct_blkio_start(); |
1da177e4 | 4656 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4657 | blk_flush_plug(current); |
8f0dfc34 | 4658 | current->in_iowait = 1; |
1da177e4 | 4659 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4660 | current->in_iowait = 0; |
1da177e4 | 4661 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4662 | delayacct_blkio_end(); |
1da177e4 LT |
4663 | return ret; |
4664 | } | |
4665 | ||
4666 | /** | |
4667 | * sys_sched_get_priority_max - return maximum RT priority. | |
4668 | * @policy: scheduling class. | |
4669 | * | |
4670 | * this syscall returns the maximum rt_priority that can be used | |
4671 | * by a given scheduling class. | |
4672 | */ | |
5add95d4 | 4673 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4674 | { |
4675 | int ret = -EINVAL; | |
4676 | ||
4677 | switch (policy) { | |
4678 | case SCHED_FIFO: | |
4679 | case SCHED_RR: | |
4680 | ret = MAX_USER_RT_PRIO-1; | |
4681 | break; | |
4682 | case SCHED_NORMAL: | |
b0a9499c | 4683 | case SCHED_BATCH: |
dd41f596 | 4684 | case SCHED_IDLE: |
1da177e4 LT |
4685 | ret = 0; |
4686 | break; | |
4687 | } | |
4688 | return ret; | |
4689 | } | |
4690 | ||
4691 | /** | |
4692 | * sys_sched_get_priority_min - return minimum RT priority. | |
4693 | * @policy: scheduling class. | |
4694 | * | |
4695 | * this syscall returns the minimum rt_priority that can be used | |
4696 | * by a given scheduling class. | |
4697 | */ | |
5add95d4 | 4698 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4699 | { |
4700 | int ret = -EINVAL; | |
4701 | ||
4702 | switch (policy) { | |
4703 | case SCHED_FIFO: | |
4704 | case SCHED_RR: | |
4705 | ret = 1; | |
4706 | break; | |
4707 | case SCHED_NORMAL: | |
b0a9499c | 4708 | case SCHED_BATCH: |
dd41f596 | 4709 | case SCHED_IDLE: |
1da177e4 LT |
4710 | ret = 0; |
4711 | } | |
4712 | return ret; | |
4713 | } | |
4714 | ||
4715 | /** | |
4716 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4717 | * @pid: pid of the process. | |
4718 | * @interval: userspace pointer to the timeslice value. | |
4719 | * | |
4720 | * this syscall writes the default timeslice value of a given process | |
4721 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4722 | */ | |
17da2bd9 | 4723 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4724 | struct timespec __user *, interval) |
1da177e4 | 4725 | { |
36c8b586 | 4726 | struct task_struct *p; |
a4ec24b4 | 4727 | unsigned int time_slice; |
dba091b9 TG |
4728 | unsigned long flags; |
4729 | struct rq *rq; | |
3a5c359a | 4730 | int retval; |
1da177e4 | 4731 | struct timespec t; |
1da177e4 LT |
4732 | |
4733 | if (pid < 0) | |
3a5c359a | 4734 | return -EINVAL; |
1da177e4 LT |
4735 | |
4736 | retval = -ESRCH; | |
1a551ae7 | 4737 | rcu_read_lock(); |
1da177e4 LT |
4738 | p = find_process_by_pid(pid); |
4739 | if (!p) | |
4740 | goto out_unlock; | |
4741 | ||
4742 | retval = security_task_getscheduler(p); | |
4743 | if (retval) | |
4744 | goto out_unlock; | |
4745 | ||
dba091b9 TG |
4746 | rq = task_rq_lock(p, &flags); |
4747 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4748 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4749 | |
1a551ae7 | 4750 | rcu_read_unlock(); |
a4ec24b4 | 4751 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4752 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4753 | return retval; |
3a5c359a | 4754 | |
1da177e4 | 4755 | out_unlock: |
1a551ae7 | 4756 | rcu_read_unlock(); |
1da177e4 LT |
4757 | return retval; |
4758 | } | |
4759 | ||
7c731e0a | 4760 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4761 | |
82a1fcb9 | 4762 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4763 | { |
1da177e4 | 4764 | unsigned long free = 0; |
36c8b586 | 4765 | unsigned state; |
1da177e4 | 4766 | |
1da177e4 | 4767 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4768 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4769 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4770 | #if BITS_PER_LONG == 32 |
1da177e4 | 4771 | if (state == TASK_RUNNING) |
3df0fc5b | 4772 | printk(KERN_CONT " running "); |
1da177e4 | 4773 | else |
3df0fc5b | 4774 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4775 | #else |
4776 | if (state == TASK_RUNNING) | |
3df0fc5b | 4777 | printk(KERN_CONT " running task "); |
1da177e4 | 4778 | else |
3df0fc5b | 4779 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4780 | #endif |
4781 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4782 | free = stack_not_used(p); |
1da177e4 | 4783 | #endif |
3df0fc5b | 4784 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
07cde260 | 4785 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), |
aa47b7e0 | 4786 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4787 | |
5fb5e6de | 4788 | show_stack(p, NULL); |
1da177e4 LT |
4789 | } |
4790 | ||
e59e2ae2 | 4791 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4792 | { |
36c8b586 | 4793 | struct task_struct *g, *p; |
1da177e4 | 4794 | |
4bd77321 | 4795 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4796 | printk(KERN_INFO |
4797 | " task PC stack pid father\n"); | |
1da177e4 | 4798 | #else |
3df0fc5b PZ |
4799 | printk(KERN_INFO |
4800 | " task PC stack pid father\n"); | |
1da177e4 | 4801 | #endif |
510f5acc | 4802 | rcu_read_lock(); |
1da177e4 LT |
4803 | do_each_thread(g, p) { |
4804 | /* | |
4805 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4806 | * console might take a lot of time: |
1da177e4 LT |
4807 | */ |
4808 | touch_nmi_watchdog(); | |
39bc89fd | 4809 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4810 | sched_show_task(p); |
1da177e4 LT |
4811 | } while_each_thread(g, p); |
4812 | ||
04c9167f JF |
4813 | touch_all_softlockup_watchdogs(); |
4814 | ||
dd41f596 IM |
4815 | #ifdef CONFIG_SCHED_DEBUG |
4816 | sysrq_sched_debug_show(); | |
4817 | #endif | |
510f5acc | 4818 | rcu_read_unlock(); |
e59e2ae2 IM |
4819 | /* |
4820 | * Only show locks if all tasks are dumped: | |
4821 | */ | |
93335a21 | 4822 | if (!state_filter) |
e59e2ae2 | 4823 | debug_show_all_locks(); |
1da177e4 LT |
4824 | } |
4825 | ||
1df21055 IM |
4826 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4827 | { | |
dd41f596 | 4828 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4829 | } |
4830 | ||
f340c0d1 IM |
4831 | /** |
4832 | * init_idle - set up an idle thread for a given CPU | |
4833 | * @idle: task in question | |
4834 | * @cpu: cpu the idle task belongs to | |
4835 | * | |
4836 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4837 | * flag, to make booting more robust. | |
4838 | */ | |
5c1e1767 | 4839 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4840 | { |
70b97a7f | 4841 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4842 | unsigned long flags; |
4843 | ||
05fa785c | 4844 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4845 | |
dd41f596 | 4846 | __sched_fork(idle); |
06b83b5f | 4847 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4848 | idle->se.exec_start = sched_clock(); |
4849 | ||
1e1b6c51 | 4850 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4851 | /* |
4852 | * We're having a chicken and egg problem, even though we are | |
4853 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4854 | * lockdep check in task_group() will fail. | |
4855 | * | |
4856 | * Similar case to sched_fork(). / Alternatively we could | |
4857 | * use task_rq_lock() here and obtain the other rq->lock. | |
4858 | * | |
4859 | * Silence PROVE_RCU | |
4860 | */ | |
4861 | rcu_read_lock(); | |
dd41f596 | 4862 | __set_task_cpu(idle, cpu); |
6506cf6c | 4863 | rcu_read_unlock(); |
1da177e4 | 4864 | |
1da177e4 | 4865 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4866 | #if defined(CONFIG_SMP) |
4867 | idle->on_cpu = 1; | |
4866cde0 | 4868 | #endif |
05fa785c | 4869 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4870 | |
4871 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4872 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4873 | |
dd41f596 IM |
4874 | /* |
4875 | * The idle tasks have their own, simple scheduling class: | |
4876 | */ | |
4877 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4878 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
4879 | #if defined(CONFIG_SMP) |
4880 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4881 | #endif | |
19978ca6 IM |
4882 | } |
4883 | ||
1da177e4 | 4884 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4885 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4886 | { | |
4887 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4888 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4889 | |
4890 | cpumask_copy(&p->cpus_allowed, new_mask); | |
4891 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
1e1b6c51 KM |
4892 | } |
4893 | ||
1da177e4 LT |
4894 | /* |
4895 | * This is how migration works: | |
4896 | * | |
969c7921 TH |
4897 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4898 | * stop_one_cpu(). | |
4899 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4900 | * off the CPU) | |
4901 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4902 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4903 | * it and puts it into the right queue. |
969c7921 TH |
4904 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4905 | * is done. | |
1da177e4 LT |
4906 | */ |
4907 | ||
4908 | /* | |
4909 | * Change a given task's CPU affinity. Migrate the thread to a | |
4910 | * proper CPU and schedule it away if the CPU it's executing on | |
4911 | * is removed from the allowed bitmask. | |
4912 | * | |
4913 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4914 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4915 | * call is not atomic; no spinlocks may be held. |
4916 | */ | |
96f874e2 | 4917 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4918 | { |
4919 | unsigned long flags; | |
70b97a7f | 4920 | struct rq *rq; |
969c7921 | 4921 | unsigned int dest_cpu; |
48f24c4d | 4922 | int ret = 0; |
1da177e4 LT |
4923 | |
4924 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4925 | |
db44fc01 YZ |
4926 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4927 | goto out; | |
4928 | ||
6ad4c188 | 4929 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4930 | ret = -EINVAL; |
4931 | goto out; | |
4932 | } | |
4933 | ||
db44fc01 | 4934 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4935 | ret = -EINVAL; |
4936 | goto out; | |
4937 | } | |
4938 | ||
1e1b6c51 | 4939 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4940 | |
1da177e4 | 4941 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4942 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4943 | goto out; |
4944 | ||
969c7921 | 4945 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4946 | if (p->on_rq) { |
969c7921 | 4947 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4948 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4949 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4950 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4951 | tlb_migrate_finish(p->mm); |
4952 | return 0; | |
4953 | } | |
4954 | out: | |
0122ec5b | 4955 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4956 | |
1da177e4 LT |
4957 | return ret; |
4958 | } | |
cd8ba7cd | 4959 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4960 | |
4961 | /* | |
41a2d6cf | 4962 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4963 | * this because either it can't run here any more (set_cpus_allowed() |
4964 | * away from this CPU, or CPU going down), or because we're | |
4965 | * attempting to rebalance this task on exec (sched_exec). | |
4966 | * | |
4967 | * So we race with normal scheduler movements, but that's OK, as long | |
4968 | * as the task is no longer on this CPU. | |
efc30814 KK |
4969 | * |
4970 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4971 | */ |
efc30814 | 4972 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4973 | { |
70b97a7f | 4974 | struct rq *rq_dest, *rq_src; |
e2912009 | 4975 | int ret = 0; |
1da177e4 | 4976 | |
e761b772 | 4977 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4978 | return ret; |
1da177e4 LT |
4979 | |
4980 | rq_src = cpu_rq(src_cpu); | |
4981 | rq_dest = cpu_rq(dest_cpu); | |
4982 | ||
0122ec5b | 4983 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4984 | double_rq_lock(rq_src, rq_dest); |
4985 | /* Already moved. */ | |
4986 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4987 | goto done; |
1da177e4 | 4988 | /* Affinity changed (again). */ |
fa17b507 | 4989 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4990 | goto fail; |
1da177e4 | 4991 | |
e2912009 PZ |
4992 | /* |
4993 | * If we're not on a rq, the next wake-up will ensure we're | |
4994 | * placed properly. | |
4995 | */ | |
fd2f4419 | 4996 | if (p->on_rq) { |
4ca9b72b | 4997 | dequeue_task(rq_src, p, 0); |
e2912009 | 4998 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4999 | enqueue_task(rq_dest, p, 0); |
15afe09b | 5000 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5001 | } |
b1e38734 | 5002 | done: |
efc30814 | 5003 | ret = 1; |
b1e38734 | 5004 | fail: |
1da177e4 | 5005 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 5006 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 5007 | return ret; |
1da177e4 LT |
5008 | } |
5009 | ||
5010 | /* | |
969c7921 TH |
5011 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5012 | * and performs thread migration by bumping thread off CPU then | |
5013 | * 'pushing' onto another runqueue. | |
1da177e4 | 5014 | */ |
969c7921 | 5015 | static int migration_cpu_stop(void *data) |
1da177e4 | 5016 | { |
969c7921 | 5017 | struct migration_arg *arg = data; |
f7b4cddc | 5018 | |
969c7921 TH |
5019 | /* |
5020 | * The original target cpu might have gone down and we might | |
5021 | * be on another cpu but it doesn't matter. | |
5022 | */ | |
f7b4cddc | 5023 | local_irq_disable(); |
969c7921 | 5024 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5025 | local_irq_enable(); |
1da177e4 | 5026 | return 0; |
f7b4cddc ON |
5027 | } |
5028 | ||
1da177e4 | 5029 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5030 | |
054b9108 | 5031 | /* |
48c5ccae PZ |
5032 | * Ensures that the idle task is using init_mm right before its cpu goes |
5033 | * offline. | |
054b9108 | 5034 | */ |
48c5ccae | 5035 | void idle_task_exit(void) |
1da177e4 | 5036 | { |
48c5ccae | 5037 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5038 | |
48c5ccae | 5039 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5040 | |
48c5ccae PZ |
5041 | if (mm != &init_mm) |
5042 | switch_mm(mm, &init_mm, current); | |
5043 | mmdrop(mm); | |
1da177e4 LT |
5044 | } |
5045 | ||
5046 | /* | |
5047 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5048 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5049 | * for performance reasons the counter is not stricly tracking tasks to | |
5050 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5051 | * to keep the global sum constant after CPU-down: | |
5052 | */ | |
70b97a7f | 5053 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5054 | { |
6ad4c188 | 5055 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5056 | |
1da177e4 LT |
5057 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5058 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5059 | } |
5060 | ||
dd41f596 | 5061 | /* |
48c5ccae | 5062 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5063 | */ |
48c5ccae | 5064 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5065 | { |
48c5ccae PZ |
5066 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5067 | rq->calc_load_active = 0; | |
1da177e4 LT |
5068 | } |
5069 | ||
48f24c4d | 5070 | /* |
48c5ccae PZ |
5071 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5072 | * try_to_wake_up()->select_task_rq(). | |
5073 | * | |
5074 | * Called with rq->lock held even though we'er in stop_machine() and | |
5075 | * there's no concurrency possible, we hold the required locks anyway | |
5076 | * because of lock validation efforts. | |
1da177e4 | 5077 | */ |
48c5ccae | 5078 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5079 | { |
70b97a7f | 5080 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5081 | struct task_struct *next, *stop = rq->stop; |
5082 | int dest_cpu; | |
1da177e4 LT |
5083 | |
5084 | /* | |
48c5ccae PZ |
5085 | * Fudge the rq selection such that the below task selection loop |
5086 | * doesn't get stuck on the currently eligible stop task. | |
5087 | * | |
5088 | * We're currently inside stop_machine() and the rq is either stuck | |
5089 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5090 | * either way we should never end up calling schedule() until we're | |
5091 | * done here. | |
1da177e4 | 5092 | */ |
48c5ccae | 5093 | rq->stop = NULL; |
48f24c4d | 5094 | |
8cb120d3 PT |
5095 | /* Ensure any throttled groups are reachable by pick_next_task */ |
5096 | unthrottle_offline_cfs_rqs(rq); | |
5097 | ||
dd41f596 | 5098 | for ( ; ; ) { |
48c5ccae PZ |
5099 | /* |
5100 | * There's this thread running, bail when that's the only | |
5101 | * remaining thread. | |
5102 | */ | |
5103 | if (rq->nr_running == 1) | |
dd41f596 | 5104 | break; |
48c5ccae | 5105 | |
b67802ea | 5106 | next = pick_next_task(rq); |
48c5ccae | 5107 | BUG_ON(!next); |
79c53799 | 5108 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5109 | |
48c5ccae PZ |
5110 | /* Find suitable destination for @next, with force if needed. */ |
5111 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5112 | raw_spin_unlock(&rq->lock); | |
5113 | ||
5114 | __migrate_task(next, dead_cpu, dest_cpu); | |
5115 | ||
5116 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5117 | } |
dce48a84 | 5118 | |
48c5ccae | 5119 | rq->stop = stop; |
dce48a84 | 5120 | } |
48c5ccae | 5121 | |
1da177e4 LT |
5122 | #endif /* CONFIG_HOTPLUG_CPU */ |
5123 | ||
e692ab53 NP |
5124 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5125 | ||
5126 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5127 | { |
5128 | .procname = "sched_domain", | |
c57baf1e | 5129 | .mode = 0555, |
e0361851 | 5130 | }, |
56992309 | 5131 | {} |
e692ab53 NP |
5132 | }; |
5133 | ||
5134 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5135 | { |
5136 | .procname = "kernel", | |
c57baf1e | 5137 | .mode = 0555, |
e0361851 AD |
5138 | .child = sd_ctl_dir, |
5139 | }, | |
56992309 | 5140 | {} |
e692ab53 NP |
5141 | }; |
5142 | ||
5143 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5144 | { | |
5145 | struct ctl_table *entry = | |
5cf9f062 | 5146 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5147 | |
e692ab53 NP |
5148 | return entry; |
5149 | } | |
5150 | ||
6382bc90 MM |
5151 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5152 | { | |
cd790076 | 5153 | struct ctl_table *entry; |
6382bc90 | 5154 | |
cd790076 MM |
5155 | /* |
5156 | * In the intermediate directories, both the child directory and | |
5157 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5158 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5159 | * static strings and all have proc handlers. |
5160 | */ | |
5161 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5162 | if (entry->child) |
5163 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5164 | if (entry->proc_handler == NULL) |
5165 | kfree(entry->procname); | |
5166 | } | |
6382bc90 MM |
5167 | |
5168 | kfree(*tablep); | |
5169 | *tablep = NULL; | |
5170 | } | |
5171 | ||
e692ab53 | 5172 | static void |
e0361851 | 5173 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5174 | const char *procname, void *data, int maxlen, |
36fcb589 | 5175 | umode_t mode, proc_handler *proc_handler) |
e692ab53 | 5176 | { |
e692ab53 NP |
5177 | entry->procname = procname; |
5178 | entry->data = data; | |
5179 | entry->maxlen = maxlen; | |
5180 | entry->mode = mode; | |
5181 | entry->proc_handler = proc_handler; | |
5182 | } | |
5183 | ||
5184 | static struct ctl_table * | |
5185 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5186 | { | |
a5d8c348 | 5187 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5188 | |
ad1cdc1d MM |
5189 | if (table == NULL) |
5190 | return NULL; | |
5191 | ||
e0361851 | 5192 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5193 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5194 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5195 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5196 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5197 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5198 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5199 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5200 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5201 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5202 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5203 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5204 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5205 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5206 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5207 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5208 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5209 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5210 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5211 | &sd->cache_nice_tries, |
5212 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5213 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5214 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5215 | set_table_entry(&table[11], "name", sd->name, |
5216 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5217 | /* &table[12] is terminator */ | |
e692ab53 NP |
5218 | |
5219 | return table; | |
5220 | } | |
5221 | ||
9a4e7159 | 5222 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5223 | { |
5224 | struct ctl_table *entry, *table; | |
5225 | struct sched_domain *sd; | |
5226 | int domain_num = 0, i; | |
5227 | char buf[32]; | |
5228 | ||
5229 | for_each_domain(cpu, sd) | |
5230 | domain_num++; | |
5231 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5232 | if (table == NULL) |
5233 | return NULL; | |
e692ab53 NP |
5234 | |
5235 | i = 0; | |
5236 | for_each_domain(cpu, sd) { | |
5237 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5238 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5239 | entry->mode = 0555; |
e692ab53 NP |
5240 | entry->child = sd_alloc_ctl_domain_table(sd); |
5241 | entry++; | |
5242 | i++; | |
5243 | } | |
5244 | return table; | |
5245 | } | |
5246 | ||
5247 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5248 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5249 | { |
6ad4c188 | 5250 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5251 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5252 | char buf[32]; | |
5253 | ||
7378547f MM |
5254 | WARN_ON(sd_ctl_dir[0].child); |
5255 | sd_ctl_dir[0].child = entry; | |
5256 | ||
ad1cdc1d MM |
5257 | if (entry == NULL) |
5258 | return; | |
5259 | ||
6ad4c188 | 5260 | for_each_possible_cpu(i) { |
e692ab53 | 5261 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5262 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5263 | entry->mode = 0555; |
e692ab53 | 5264 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5265 | entry++; |
e692ab53 | 5266 | } |
7378547f MM |
5267 | |
5268 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5269 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5270 | } | |
6382bc90 | 5271 | |
7378547f | 5272 | /* may be called multiple times per register */ |
6382bc90 MM |
5273 | static void unregister_sched_domain_sysctl(void) |
5274 | { | |
7378547f MM |
5275 | if (sd_sysctl_header) |
5276 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5277 | sd_sysctl_header = NULL; |
7378547f MM |
5278 | if (sd_ctl_dir[0].child) |
5279 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5280 | } |
e692ab53 | 5281 | #else |
6382bc90 MM |
5282 | static void register_sched_domain_sysctl(void) |
5283 | { | |
5284 | } | |
5285 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5286 | { |
5287 | } | |
5288 | #endif | |
5289 | ||
1f11eb6a GH |
5290 | static void set_rq_online(struct rq *rq) |
5291 | { | |
5292 | if (!rq->online) { | |
5293 | const struct sched_class *class; | |
5294 | ||
c6c4927b | 5295 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5296 | rq->online = 1; |
5297 | ||
5298 | for_each_class(class) { | |
5299 | if (class->rq_online) | |
5300 | class->rq_online(rq); | |
5301 | } | |
5302 | } | |
5303 | } | |
5304 | ||
5305 | static void set_rq_offline(struct rq *rq) | |
5306 | { | |
5307 | if (rq->online) { | |
5308 | const struct sched_class *class; | |
5309 | ||
5310 | for_each_class(class) { | |
5311 | if (class->rq_offline) | |
5312 | class->rq_offline(rq); | |
5313 | } | |
5314 | ||
c6c4927b | 5315 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5316 | rq->online = 0; |
5317 | } | |
5318 | } | |
5319 | ||
1da177e4 LT |
5320 | /* |
5321 | * migration_call - callback that gets triggered when a CPU is added. | |
5322 | * Here we can start up the necessary migration thread for the new CPU. | |
5323 | */ | |
48f24c4d IM |
5324 | static int __cpuinit |
5325 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5326 | { |
48f24c4d | 5327 | int cpu = (long)hcpu; |
1da177e4 | 5328 | unsigned long flags; |
969c7921 | 5329 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5330 | |
48c5ccae | 5331 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5332 | |
1da177e4 | 5333 | case CPU_UP_PREPARE: |
a468d389 | 5334 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5335 | break; |
48f24c4d | 5336 | |
1da177e4 | 5337 | case CPU_ONLINE: |
1f94ef59 | 5338 | /* Update our root-domain */ |
05fa785c | 5339 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5340 | if (rq->rd) { |
c6c4927b | 5341 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5342 | |
5343 | set_rq_online(rq); | |
1f94ef59 | 5344 | } |
05fa785c | 5345 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5346 | break; |
48f24c4d | 5347 | |
1da177e4 | 5348 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5349 | case CPU_DYING: |
317f3941 | 5350 | sched_ttwu_pending(); |
57d885fe | 5351 | /* Update our root-domain */ |
05fa785c | 5352 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5353 | if (rq->rd) { |
c6c4927b | 5354 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5355 | set_rq_offline(rq); |
57d885fe | 5356 | } |
48c5ccae PZ |
5357 | migrate_tasks(cpu); |
5358 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5359 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5360 | |
5361 | migrate_nr_uninterruptible(rq); | |
5362 | calc_global_load_remove(rq); | |
57d885fe | 5363 | break; |
1da177e4 LT |
5364 | #endif |
5365 | } | |
49c022e6 PZ |
5366 | |
5367 | update_max_interval(); | |
5368 | ||
1da177e4 LT |
5369 | return NOTIFY_OK; |
5370 | } | |
5371 | ||
f38b0820 PM |
5372 | /* |
5373 | * Register at high priority so that task migration (migrate_all_tasks) | |
5374 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5375 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5376 | */ |
26c2143b | 5377 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5378 | .notifier_call = migration_call, |
50a323b7 | 5379 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5380 | }; |
5381 | ||
3a101d05 TH |
5382 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5383 | unsigned long action, void *hcpu) | |
5384 | { | |
5385 | switch (action & ~CPU_TASKS_FROZEN) { | |
5386 | case CPU_ONLINE: | |
5387 | case CPU_DOWN_FAILED: | |
5388 | set_cpu_active((long)hcpu, true); | |
5389 | return NOTIFY_OK; | |
5390 | default: | |
5391 | return NOTIFY_DONE; | |
5392 | } | |
5393 | } | |
5394 | ||
5395 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5396 | unsigned long action, void *hcpu) | |
5397 | { | |
5398 | switch (action & ~CPU_TASKS_FROZEN) { | |
5399 | case CPU_DOWN_PREPARE: | |
5400 | set_cpu_active((long)hcpu, false); | |
5401 | return NOTIFY_OK; | |
5402 | default: | |
5403 | return NOTIFY_DONE; | |
5404 | } | |
5405 | } | |
5406 | ||
7babe8db | 5407 | static int __init migration_init(void) |
1da177e4 LT |
5408 | { |
5409 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5410 | int err; |
48f24c4d | 5411 | |
3a101d05 | 5412 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5413 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5414 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5415 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5416 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5417 | |
3a101d05 TH |
5418 | /* Register cpu active notifiers */ |
5419 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5420 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5421 | ||
a004cd42 | 5422 | return 0; |
1da177e4 | 5423 | } |
7babe8db | 5424 | early_initcall(migration_init); |
1da177e4 LT |
5425 | #endif |
5426 | ||
5427 | #ifdef CONFIG_SMP | |
476f3534 | 5428 | |
4cb98839 PZ |
5429 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5430 | ||
3e9830dc | 5431 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5432 | |
f6630114 MT |
5433 | static __read_mostly int sched_domain_debug_enabled; |
5434 | ||
5435 | static int __init sched_domain_debug_setup(char *str) | |
5436 | { | |
5437 | sched_domain_debug_enabled = 1; | |
5438 | ||
5439 | return 0; | |
5440 | } | |
5441 | early_param("sched_debug", sched_domain_debug_setup); | |
5442 | ||
7c16ec58 | 5443 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5444 | struct cpumask *groupmask) |
1da177e4 | 5445 | { |
4dcf6aff | 5446 | struct sched_group *group = sd->groups; |
434d53b0 | 5447 | char str[256]; |
1da177e4 | 5448 | |
968ea6d8 | 5449 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5450 | cpumask_clear(groupmask); |
4dcf6aff IM |
5451 | |
5452 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5453 | ||
5454 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5455 | printk("does not load-balance\n"); |
4dcf6aff | 5456 | if (sd->parent) |
3df0fc5b PZ |
5457 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5458 | " has parent"); | |
4dcf6aff | 5459 | return -1; |
41c7ce9a NP |
5460 | } |
5461 | ||
3df0fc5b | 5462 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5463 | |
758b2cdc | 5464 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5465 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5466 | "CPU%d\n", cpu); | |
4dcf6aff | 5467 | } |
758b2cdc | 5468 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5469 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5470 | " CPU%d\n", cpu); | |
4dcf6aff | 5471 | } |
1da177e4 | 5472 | |
4dcf6aff | 5473 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5474 | do { |
4dcf6aff | 5475 | if (!group) { |
3df0fc5b PZ |
5476 | printk("\n"); |
5477 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5478 | break; |
5479 | } | |
5480 | ||
9c3f75cb | 5481 | if (!group->sgp->power) { |
3df0fc5b PZ |
5482 | printk(KERN_CONT "\n"); |
5483 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5484 | "set\n"); | |
4dcf6aff IM |
5485 | break; |
5486 | } | |
1da177e4 | 5487 | |
758b2cdc | 5488 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5489 | printk(KERN_CONT "\n"); |
5490 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5491 | break; |
5492 | } | |
1da177e4 | 5493 | |
758b2cdc | 5494 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5495 | printk(KERN_CONT "\n"); |
5496 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5497 | break; |
5498 | } | |
1da177e4 | 5499 | |
758b2cdc | 5500 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5501 | |
968ea6d8 | 5502 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5503 | |
3df0fc5b | 5504 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5505 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5506 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5507 | group->sgp->power); |
381512cf | 5508 | } |
1da177e4 | 5509 | |
4dcf6aff IM |
5510 | group = group->next; |
5511 | } while (group != sd->groups); | |
3df0fc5b | 5512 | printk(KERN_CONT "\n"); |
1da177e4 | 5513 | |
758b2cdc | 5514 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5515 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5516 | |
758b2cdc RR |
5517 | if (sd->parent && |
5518 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5519 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5520 | "of domain->span\n"); | |
4dcf6aff IM |
5521 | return 0; |
5522 | } | |
1da177e4 | 5523 | |
4dcf6aff IM |
5524 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5525 | { | |
5526 | int level = 0; | |
1da177e4 | 5527 | |
f6630114 MT |
5528 | if (!sched_domain_debug_enabled) |
5529 | return; | |
5530 | ||
4dcf6aff IM |
5531 | if (!sd) { |
5532 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5533 | return; | |
5534 | } | |
1da177e4 | 5535 | |
4dcf6aff IM |
5536 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5537 | ||
5538 | for (;;) { | |
4cb98839 | 5539 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5540 | break; |
1da177e4 LT |
5541 | level++; |
5542 | sd = sd->parent; | |
33859f7f | 5543 | if (!sd) |
4dcf6aff IM |
5544 | break; |
5545 | } | |
1da177e4 | 5546 | } |
6d6bc0ad | 5547 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5548 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5549 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5550 | |
1a20ff27 | 5551 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5552 | { |
758b2cdc | 5553 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5554 | return 1; |
5555 | ||
5556 | /* Following flags need at least 2 groups */ | |
5557 | if (sd->flags & (SD_LOAD_BALANCE | | |
5558 | SD_BALANCE_NEWIDLE | | |
5559 | SD_BALANCE_FORK | | |
89c4710e SS |
5560 | SD_BALANCE_EXEC | |
5561 | SD_SHARE_CPUPOWER | | |
5562 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5563 | if (sd->groups != sd->groups->next) |
5564 | return 0; | |
5565 | } | |
5566 | ||
5567 | /* Following flags don't use groups */ | |
c88d5910 | 5568 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5569 | return 0; |
5570 | ||
5571 | return 1; | |
5572 | } | |
5573 | ||
48f24c4d IM |
5574 | static int |
5575 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5576 | { |
5577 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5578 | ||
5579 | if (sd_degenerate(parent)) | |
5580 | return 1; | |
5581 | ||
758b2cdc | 5582 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5583 | return 0; |
5584 | ||
245af2c7 SS |
5585 | /* Flags needing groups don't count if only 1 group in parent */ |
5586 | if (parent->groups == parent->groups->next) { | |
5587 | pflags &= ~(SD_LOAD_BALANCE | | |
5588 | SD_BALANCE_NEWIDLE | | |
5589 | SD_BALANCE_FORK | | |
89c4710e SS |
5590 | SD_BALANCE_EXEC | |
5591 | SD_SHARE_CPUPOWER | | |
5592 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5593 | if (nr_node_ids == 1) |
5594 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5595 | } |
5596 | if (~cflags & pflags) | |
5597 | return 0; | |
5598 | ||
5599 | return 1; | |
5600 | } | |
5601 | ||
dce840a0 | 5602 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5603 | { |
dce840a0 | 5604 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5605 | |
68e74568 | 5606 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5607 | free_cpumask_var(rd->rto_mask); |
5608 | free_cpumask_var(rd->online); | |
5609 | free_cpumask_var(rd->span); | |
5610 | kfree(rd); | |
5611 | } | |
5612 | ||
57d885fe GH |
5613 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5614 | { | |
a0490fa3 | 5615 | struct root_domain *old_rd = NULL; |
57d885fe | 5616 | unsigned long flags; |
57d885fe | 5617 | |
05fa785c | 5618 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5619 | |
5620 | if (rq->rd) { | |
a0490fa3 | 5621 | old_rd = rq->rd; |
57d885fe | 5622 | |
c6c4927b | 5623 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5624 | set_rq_offline(rq); |
57d885fe | 5625 | |
c6c4927b | 5626 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5627 | |
a0490fa3 IM |
5628 | /* |
5629 | * If we dont want to free the old_rt yet then | |
5630 | * set old_rd to NULL to skip the freeing later | |
5631 | * in this function: | |
5632 | */ | |
5633 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5634 | old_rd = NULL; | |
57d885fe GH |
5635 | } |
5636 | ||
5637 | atomic_inc(&rd->refcount); | |
5638 | rq->rd = rd; | |
5639 | ||
c6c4927b | 5640 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5641 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5642 | set_rq_online(rq); |
57d885fe | 5643 | |
05fa785c | 5644 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5645 | |
5646 | if (old_rd) | |
dce840a0 | 5647 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5648 | } |
5649 | ||
68c38fc3 | 5650 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5651 | { |
5652 | memset(rd, 0, sizeof(*rd)); | |
5653 | ||
68c38fc3 | 5654 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5655 | goto out; |
68c38fc3 | 5656 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5657 | goto free_span; |
68c38fc3 | 5658 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5659 | goto free_online; |
6e0534f2 | 5660 | |
68c38fc3 | 5661 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5662 | goto free_rto_mask; |
c6c4927b | 5663 | return 0; |
6e0534f2 | 5664 | |
68e74568 RR |
5665 | free_rto_mask: |
5666 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5667 | free_online: |
5668 | free_cpumask_var(rd->online); | |
5669 | free_span: | |
5670 | free_cpumask_var(rd->span); | |
0c910d28 | 5671 | out: |
c6c4927b | 5672 | return -ENOMEM; |
57d885fe GH |
5673 | } |
5674 | ||
029632fb PZ |
5675 | /* |
5676 | * By default the system creates a single root-domain with all cpus as | |
5677 | * members (mimicking the global state we have today). | |
5678 | */ | |
5679 | struct root_domain def_root_domain; | |
5680 | ||
57d885fe GH |
5681 | static void init_defrootdomain(void) |
5682 | { | |
68c38fc3 | 5683 | init_rootdomain(&def_root_domain); |
c6c4927b | 5684 | |
57d885fe GH |
5685 | atomic_set(&def_root_domain.refcount, 1); |
5686 | } | |
5687 | ||
dc938520 | 5688 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5689 | { |
5690 | struct root_domain *rd; | |
5691 | ||
5692 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5693 | if (!rd) | |
5694 | return NULL; | |
5695 | ||
68c38fc3 | 5696 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5697 | kfree(rd); |
5698 | return NULL; | |
5699 | } | |
57d885fe GH |
5700 | |
5701 | return rd; | |
5702 | } | |
5703 | ||
e3589f6c PZ |
5704 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5705 | { | |
5706 | struct sched_group *tmp, *first; | |
5707 | ||
5708 | if (!sg) | |
5709 | return; | |
5710 | ||
5711 | first = sg; | |
5712 | do { | |
5713 | tmp = sg->next; | |
5714 | ||
5715 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5716 | kfree(sg->sgp); | |
5717 | ||
5718 | kfree(sg); | |
5719 | sg = tmp; | |
5720 | } while (sg != first); | |
5721 | } | |
5722 | ||
dce840a0 PZ |
5723 | static void free_sched_domain(struct rcu_head *rcu) |
5724 | { | |
5725 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5726 | |
5727 | /* | |
5728 | * If its an overlapping domain it has private groups, iterate and | |
5729 | * nuke them all. | |
5730 | */ | |
5731 | if (sd->flags & SD_OVERLAP) { | |
5732 | free_sched_groups(sd->groups, 1); | |
5733 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5734 | kfree(sd->groups->sgp); |
dce840a0 | 5735 | kfree(sd->groups); |
9c3f75cb | 5736 | } |
dce840a0 PZ |
5737 | kfree(sd); |
5738 | } | |
5739 | ||
5740 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5741 | { | |
5742 | call_rcu(&sd->rcu, free_sched_domain); | |
5743 | } | |
5744 | ||
5745 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5746 | { | |
5747 | for (; sd; sd = sd->parent) | |
5748 | destroy_sched_domain(sd, cpu); | |
5749 | } | |
5750 | ||
518cd623 PZ |
5751 | /* |
5752 | * Keep a special pointer to the highest sched_domain that has | |
5753 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5754 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5755 | * | |
5756 | * Also keep a unique ID per domain (we use the first cpu number in | |
5757 | * the cpumask of the domain), this allows us to quickly tell if | |
5758 | * two cpus are in the same cache domain, see ttwu_share_cache(). | |
5759 | */ | |
5760 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5761 | DEFINE_PER_CPU(int, sd_llc_id); | |
5762 | ||
5763 | static void update_top_cache_domain(int cpu) | |
5764 | { | |
5765 | struct sched_domain *sd; | |
5766 | int id = cpu; | |
5767 | ||
5768 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
5769 | if (sd) | |
5770 | id = cpumask_first(sched_domain_span(sd)); | |
5771 | ||
5772 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5773 | per_cpu(sd_llc_id, cpu) = id; | |
5774 | } | |
5775 | ||
1da177e4 | 5776 | /* |
0eab9146 | 5777 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5778 | * hold the hotplug lock. |
5779 | */ | |
0eab9146 IM |
5780 | static void |
5781 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5782 | { |
70b97a7f | 5783 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5784 | struct sched_domain *tmp; |
5785 | ||
5786 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5787 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5788 | struct sched_domain *parent = tmp->parent; |
5789 | if (!parent) | |
5790 | break; | |
f29c9b1c | 5791 | |
1a848870 | 5792 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5793 | tmp->parent = parent->parent; |
1a848870 SS |
5794 | if (parent->parent) |
5795 | parent->parent->child = tmp; | |
dce840a0 | 5796 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5797 | } else |
5798 | tmp = tmp->parent; | |
245af2c7 SS |
5799 | } |
5800 | ||
1a848870 | 5801 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5802 | tmp = sd; |
245af2c7 | 5803 | sd = sd->parent; |
dce840a0 | 5804 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5805 | if (sd) |
5806 | sd->child = NULL; | |
5807 | } | |
1da177e4 | 5808 | |
4cb98839 | 5809 | sched_domain_debug(sd, cpu); |
1da177e4 | 5810 | |
57d885fe | 5811 | rq_attach_root(rq, rd); |
dce840a0 | 5812 | tmp = rq->sd; |
674311d5 | 5813 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5814 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5815 | |
5816 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5817 | } |
5818 | ||
5819 | /* cpus with isolated domains */ | |
dcc30a35 | 5820 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5821 | |
5822 | /* Setup the mask of cpus configured for isolated domains */ | |
5823 | static int __init isolated_cpu_setup(char *str) | |
5824 | { | |
bdddd296 | 5825 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5826 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5827 | return 1; |
5828 | } | |
5829 | ||
8927f494 | 5830 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5831 | |
9c1cfda2 | 5832 | #ifdef CONFIG_NUMA |
198e2f18 | 5833 | |
9c1cfda2 JH |
5834 | /** |
5835 | * find_next_best_node - find the next node to include in a sched_domain | |
5836 | * @node: node whose sched_domain we're building | |
5837 | * @used_nodes: nodes already in the sched_domain | |
5838 | * | |
41a2d6cf | 5839 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5840 | * finds the closest node not already in the @used_nodes map. |
5841 | * | |
5842 | * Should use nodemask_t. | |
5843 | */ | |
c5f59f08 | 5844 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 5845 | { |
7142d17e | 5846 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
5847 | |
5848 | min_val = INT_MAX; | |
5849 | ||
076ac2af | 5850 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 5851 | /* Start at @node */ |
076ac2af | 5852 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
5853 | |
5854 | if (!nr_cpus_node(n)) | |
5855 | continue; | |
5856 | ||
5857 | /* Skip already used nodes */ | |
c5f59f08 | 5858 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
5859 | continue; |
5860 | ||
5861 | /* Simple min distance search */ | |
5862 | val = node_distance(node, n); | |
5863 | ||
5864 | if (val < min_val) { | |
5865 | min_val = val; | |
5866 | best_node = n; | |
5867 | } | |
5868 | } | |
5869 | ||
7142d17e HD |
5870 | if (best_node != -1) |
5871 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
5872 | return best_node; |
5873 | } | |
5874 | ||
5875 | /** | |
5876 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5877 | * @node: node whose cpumask we're constructing | |
73486722 | 5878 | * @span: resulting cpumask |
9c1cfda2 | 5879 | * |
41a2d6cf | 5880 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
5881 | * should be one that prevents unnecessary balancing, but also spreads tasks |
5882 | * out optimally. | |
5883 | */ | |
96f874e2 | 5884 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 5885 | { |
c5f59f08 | 5886 | nodemask_t used_nodes; |
48f24c4d | 5887 | int i; |
9c1cfda2 | 5888 | |
6ca09dfc | 5889 | cpumask_clear(span); |
c5f59f08 | 5890 | nodes_clear(used_nodes); |
9c1cfda2 | 5891 | |
6ca09dfc | 5892 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 5893 | node_set(node, used_nodes); |
9c1cfda2 JH |
5894 | |
5895 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 5896 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
5897 | if (next_node < 0) |
5898 | break; | |
6ca09dfc | 5899 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 5900 | } |
9c1cfda2 | 5901 | } |
d3081f52 PZ |
5902 | |
5903 | static const struct cpumask *cpu_node_mask(int cpu) | |
5904 | { | |
5905 | lockdep_assert_held(&sched_domains_mutex); | |
5906 | ||
5907 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
5908 | ||
5909 | return sched_domains_tmpmask; | |
5910 | } | |
2c402dc3 PZ |
5911 | |
5912 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
5913 | { | |
5914 | return cpu_possible_mask; | |
5915 | } | |
6d6bc0ad | 5916 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 5917 | |
d3081f52 PZ |
5918 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5919 | { | |
5920 | return cpumask_of_node(cpu_to_node(cpu)); | |
5921 | } | |
5922 | ||
5c45bf27 | 5923 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5924 | |
dce840a0 PZ |
5925 | struct sd_data { |
5926 | struct sched_domain **__percpu sd; | |
5927 | struct sched_group **__percpu sg; | |
9c3f75cb | 5928 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5929 | }; |
5930 | ||
49a02c51 | 5931 | struct s_data { |
21d42ccf | 5932 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5933 | struct root_domain *rd; |
5934 | }; | |
5935 | ||
2109b99e | 5936 | enum s_alloc { |
2109b99e | 5937 | sa_rootdomain, |
21d42ccf | 5938 | sa_sd, |
dce840a0 | 5939 | sa_sd_storage, |
2109b99e AH |
5940 | sa_none, |
5941 | }; | |
5942 | ||
54ab4ff4 PZ |
5943 | struct sched_domain_topology_level; |
5944 | ||
5945 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5946 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5947 | ||
e3589f6c PZ |
5948 | #define SDTL_OVERLAP 0x01 |
5949 | ||
eb7a74e6 | 5950 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5951 | sched_domain_init_f init; |
5952 | sched_domain_mask_f mask; | |
e3589f6c | 5953 | int flags; |
54ab4ff4 | 5954 | struct sd_data data; |
eb7a74e6 PZ |
5955 | }; |
5956 | ||
e3589f6c PZ |
5957 | static int |
5958 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5959 | { | |
5960 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5961 | const struct cpumask *span = sched_domain_span(sd); | |
5962 | struct cpumask *covered = sched_domains_tmpmask; | |
5963 | struct sd_data *sdd = sd->private; | |
5964 | struct sched_domain *child; | |
5965 | int i; | |
5966 | ||
5967 | cpumask_clear(covered); | |
5968 | ||
5969 | for_each_cpu(i, span) { | |
5970 | struct cpumask *sg_span; | |
5971 | ||
5972 | if (cpumask_test_cpu(i, covered)) | |
5973 | continue; | |
5974 | ||
5975 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
4d78a223 | 5976 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5977 | |
5978 | if (!sg) | |
5979 | goto fail; | |
5980 | ||
5981 | sg_span = sched_group_cpus(sg); | |
5982 | ||
5983 | child = *per_cpu_ptr(sdd->sd, i); | |
5984 | if (child->child) { | |
5985 | child = child->child; | |
5986 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5987 | } else | |
5988 | cpumask_set_cpu(i, sg_span); | |
5989 | ||
5990 | cpumask_or(covered, covered, sg_span); | |
5991 | ||
5992 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
5993 | atomic_inc(&sg->sgp->ref); | |
5994 | ||
5995 | if (cpumask_test_cpu(cpu, sg_span)) | |
5996 | groups = sg; | |
5997 | ||
5998 | if (!first) | |
5999 | first = sg; | |
6000 | if (last) | |
6001 | last->next = sg; | |
6002 | last = sg; | |
6003 | last->next = first; | |
6004 | } | |
6005 | sd->groups = groups; | |
6006 | ||
6007 | return 0; | |
6008 | ||
6009 | fail: | |
6010 | free_sched_groups(first, 0); | |
6011 | ||
6012 | return -ENOMEM; | |
6013 | } | |
6014 | ||
dce840a0 | 6015 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6016 | { |
dce840a0 PZ |
6017 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6018 | struct sched_domain *child = sd->child; | |
1da177e4 | 6019 | |
dce840a0 PZ |
6020 | if (child) |
6021 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6022 | |
9c3f75cb | 6023 | if (sg) { |
dce840a0 | 6024 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 6025 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 6026 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 6027 | } |
dce840a0 PZ |
6028 | |
6029 | return cpu; | |
1e9f28fa | 6030 | } |
1e9f28fa | 6031 | |
01a08546 | 6032 | /* |
dce840a0 PZ |
6033 | * build_sched_groups will build a circular linked list of the groups |
6034 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6035 | * and ->cpu_power to 0. | |
e3589f6c PZ |
6036 | * |
6037 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6038 | */ |
e3589f6c PZ |
6039 | static int |
6040 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6041 | { |
dce840a0 PZ |
6042 | struct sched_group *first = NULL, *last = NULL; |
6043 | struct sd_data *sdd = sd->private; | |
6044 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6045 | struct cpumask *covered; |
dce840a0 | 6046 | int i; |
9c1cfda2 | 6047 | |
e3589f6c PZ |
6048 | get_group(cpu, sdd, &sd->groups); |
6049 | atomic_inc(&sd->groups->ref); | |
6050 | ||
6051 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
6052 | return 0; | |
6053 | ||
f96225fd PZ |
6054 | lockdep_assert_held(&sched_domains_mutex); |
6055 | covered = sched_domains_tmpmask; | |
6056 | ||
dce840a0 | 6057 | cpumask_clear(covered); |
6711cab4 | 6058 | |
dce840a0 PZ |
6059 | for_each_cpu(i, span) { |
6060 | struct sched_group *sg; | |
6061 | int group = get_group(i, sdd, &sg); | |
6062 | int j; | |
6711cab4 | 6063 | |
dce840a0 PZ |
6064 | if (cpumask_test_cpu(i, covered)) |
6065 | continue; | |
6711cab4 | 6066 | |
dce840a0 | 6067 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 6068 | sg->sgp->power = 0; |
0601a88d | 6069 | |
dce840a0 PZ |
6070 | for_each_cpu(j, span) { |
6071 | if (get_group(j, sdd, NULL) != group) | |
6072 | continue; | |
0601a88d | 6073 | |
dce840a0 PZ |
6074 | cpumask_set_cpu(j, covered); |
6075 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6076 | } | |
0601a88d | 6077 | |
dce840a0 PZ |
6078 | if (!first) |
6079 | first = sg; | |
6080 | if (last) | |
6081 | last->next = sg; | |
6082 | last = sg; | |
6083 | } | |
6084 | last->next = first; | |
e3589f6c PZ |
6085 | |
6086 | return 0; | |
0601a88d | 6087 | } |
51888ca2 | 6088 | |
89c4710e SS |
6089 | /* |
6090 | * Initialize sched groups cpu_power. | |
6091 | * | |
6092 | * cpu_power indicates the capacity of sched group, which is used while | |
6093 | * distributing the load between different sched groups in a sched domain. | |
6094 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6095 | * there are asymmetries in the topology. If there are asymmetries, group | |
6096 | * having more cpu_power will pickup more load compared to the group having | |
6097 | * less cpu_power. | |
89c4710e SS |
6098 | */ |
6099 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6100 | { | |
e3589f6c | 6101 | struct sched_group *sg = sd->groups; |
89c4710e | 6102 | |
e3589f6c PZ |
6103 | WARN_ON(!sd || !sg); |
6104 | ||
6105 | do { | |
6106 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6107 | sg = sg->next; | |
6108 | } while (sg != sd->groups); | |
89c4710e | 6109 | |
e3589f6c PZ |
6110 | if (cpu != group_first_cpu(sg)) |
6111 | return; | |
aae6d3dd | 6112 | |
d274cb30 | 6113 | update_group_power(sd, cpu); |
69e1e811 | 6114 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
6115 | } |
6116 | ||
029632fb PZ |
6117 | int __weak arch_sd_sibling_asym_packing(void) |
6118 | { | |
6119 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
6120 | } |
6121 | ||
7c16ec58 MT |
6122 | /* |
6123 | * Initializers for schedule domains | |
6124 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6125 | */ | |
6126 | ||
a5d8c348 IM |
6127 | #ifdef CONFIG_SCHED_DEBUG |
6128 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6129 | #else | |
6130 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6131 | #endif | |
6132 | ||
54ab4ff4 PZ |
6133 | #define SD_INIT_FUNC(type) \ |
6134 | static noinline struct sched_domain * \ | |
6135 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
6136 | { \ | |
6137 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
6138 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
6139 | SD_INIT_NAME(sd, type); \ |
6140 | sd->private = &tl->data; \ | |
6141 | return sd; \ | |
7c16ec58 MT |
6142 | } |
6143 | ||
6144 | SD_INIT_FUNC(CPU) | |
6145 | #ifdef CONFIG_NUMA | |
6146 | SD_INIT_FUNC(ALLNODES) | |
6147 | SD_INIT_FUNC(NODE) | |
6148 | #endif | |
6149 | #ifdef CONFIG_SCHED_SMT | |
6150 | SD_INIT_FUNC(SIBLING) | |
6151 | #endif | |
6152 | #ifdef CONFIG_SCHED_MC | |
6153 | SD_INIT_FUNC(MC) | |
6154 | #endif | |
01a08546 HC |
6155 | #ifdef CONFIG_SCHED_BOOK |
6156 | SD_INIT_FUNC(BOOK) | |
6157 | #endif | |
7c16ec58 | 6158 | |
1d3504fc | 6159 | static int default_relax_domain_level = -1; |
60495e77 | 6160 | int sched_domain_level_max; |
1d3504fc HS |
6161 | |
6162 | static int __init setup_relax_domain_level(char *str) | |
6163 | { | |
30e0e178 LZ |
6164 | unsigned long val; |
6165 | ||
6166 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 6167 | if (val < sched_domain_level_max) |
30e0e178 LZ |
6168 | default_relax_domain_level = val; |
6169 | ||
1d3504fc HS |
6170 | return 1; |
6171 | } | |
6172 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6173 | ||
6174 | static void set_domain_attribute(struct sched_domain *sd, | |
6175 | struct sched_domain_attr *attr) | |
6176 | { | |
6177 | int request; | |
6178 | ||
6179 | if (!attr || attr->relax_domain_level < 0) { | |
6180 | if (default_relax_domain_level < 0) | |
6181 | return; | |
6182 | else | |
6183 | request = default_relax_domain_level; | |
6184 | } else | |
6185 | request = attr->relax_domain_level; | |
6186 | if (request < sd->level) { | |
6187 | /* turn off idle balance on this domain */ | |
c88d5910 | 6188 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6189 | } else { |
6190 | /* turn on idle balance on this domain */ | |
c88d5910 | 6191 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6192 | } |
6193 | } | |
6194 | ||
54ab4ff4 PZ |
6195 | static void __sdt_free(const struct cpumask *cpu_map); |
6196 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6197 | ||
2109b99e AH |
6198 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6199 | const struct cpumask *cpu_map) | |
6200 | { | |
6201 | switch (what) { | |
2109b99e | 6202 | case sa_rootdomain: |
822ff793 PZ |
6203 | if (!atomic_read(&d->rd->refcount)) |
6204 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6205 | case sa_sd: |
6206 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6207 | case sa_sd_storage: |
54ab4ff4 | 6208 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6209 | case sa_none: |
6210 | break; | |
6211 | } | |
6212 | } | |
3404c8d9 | 6213 | |
2109b99e AH |
6214 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6215 | const struct cpumask *cpu_map) | |
6216 | { | |
dce840a0 PZ |
6217 | memset(d, 0, sizeof(*d)); |
6218 | ||
54ab4ff4 PZ |
6219 | if (__sdt_alloc(cpu_map)) |
6220 | return sa_sd_storage; | |
dce840a0 PZ |
6221 | d->sd = alloc_percpu(struct sched_domain *); |
6222 | if (!d->sd) | |
6223 | return sa_sd_storage; | |
2109b99e | 6224 | d->rd = alloc_rootdomain(); |
dce840a0 | 6225 | if (!d->rd) |
21d42ccf | 6226 | return sa_sd; |
2109b99e AH |
6227 | return sa_rootdomain; |
6228 | } | |
57d885fe | 6229 | |
dce840a0 PZ |
6230 | /* |
6231 | * NULL the sd_data elements we've used to build the sched_domain and | |
6232 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6233 | * will not free the data we're using. | |
6234 | */ | |
6235 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6236 | { | |
6237 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6238 | |
6239 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6240 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6241 | ||
e3589f6c | 6242 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6243 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6244 | |
6245 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6246 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6247 | } |
6248 | ||
2c402dc3 PZ |
6249 | #ifdef CONFIG_SCHED_SMT |
6250 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6251 | { |
2c402dc3 | 6252 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6253 | } |
2c402dc3 | 6254 | #endif |
7f4588f3 | 6255 | |
d069b916 PZ |
6256 | /* |
6257 | * Topology list, bottom-up. | |
6258 | */ | |
2c402dc3 | 6259 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6260 | #ifdef CONFIG_SCHED_SMT |
6261 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6262 | #endif |
1e9f28fa | 6263 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6264 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6265 | #endif |
d069b916 PZ |
6266 | #ifdef CONFIG_SCHED_BOOK |
6267 | { sd_init_BOOK, cpu_book_mask, }, | |
6268 | #endif | |
6269 | { sd_init_CPU, cpu_cpu_mask, }, | |
6270 | #ifdef CONFIG_NUMA | |
e3589f6c | 6271 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, |
d069b916 | 6272 | { sd_init_ALLNODES, cpu_allnodes_mask, }, |
1da177e4 | 6273 | #endif |
eb7a74e6 PZ |
6274 | { NULL, }, |
6275 | }; | |
6276 | ||
6277 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6278 | ||
54ab4ff4 PZ |
6279 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6280 | { | |
6281 | struct sched_domain_topology_level *tl; | |
6282 | int j; | |
6283 | ||
6284 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6285 | struct sd_data *sdd = &tl->data; | |
6286 | ||
6287 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6288 | if (!sdd->sd) | |
6289 | return -ENOMEM; | |
6290 | ||
6291 | sdd->sg = alloc_percpu(struct sched_group *); | |
6292 | if (!sdd->sg) | |
6293 | return -ENOMEM; | |
6294 | ||
9c3f75cb PZ |
6295 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6296 | if (!sdd->sgp) | |
6297 | return -ENOMEM; | |
6298 | ||
54ab4ff4 PZ |
6299 | for_each_cpu(j, cpu_map) { |
6300 | struct sched_domain *sd; | |
6301 | struct sched_group *sg; | |
9c3f75cb | 6302 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6303 | |
6304 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6305 | GFP_KERNEL, cpu_to_node(j)); | |
6306 | if (!sd) | |
6307 | return -ENOMEM; | |
6308 | ||
6309 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6310 | ||
6311 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6312 | GFP_KERNEL, cpu_to_node(j)); | |
6313 | if (!sg) | |
6314 | return -ENOMEM; | |
6315 | ||
6316 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
6317 | |
6318 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
6319 | GFP_KERNEL, cpu_to_node(j)); | |
6320 | if (!sgp) | |
6321 | return -ENOMEM; | |
6322 | ||
6323 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6324 | } |
6325 | } | |
6326 | ||
6327 | return 0; | |
6328 | } | |
6329 | ||
6330 | static void __sdt_free(const struct cpumask *cpu_map) | |
6331 | { | |
6332 | struct sched_domain_topology_level *tl; | |
6333 | int j; | |
6334 | ||
6335 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6336 | struct sd_data *sdd = &tl->data; | |
6337 | ||
6338 | for_each_cpu(j, cpu_map) { | |
e3589f6c PZ |
6339 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); |
6340 | if (sd && (sd->flags & SD_OVERLAP)) | |
6341 | free_sched_groups(sd->groups, 0); | |
feff8fa0 | 6342 | kfree(*per_cpu_ptr(sdd->sd, j)); |
54ab4ff4 | 6343 | kfree(*per_cpu_ptr(sdd->sg, j)); |
9c3f75cb | 6344 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
54ab4ff4 PZ |
6345 | } |
6346 | free_percpu(sdd->sd); | |
6347 | free_percpu(sdd->sg); | |
9c3f75cb | 6348 | free_percpu(sdd->sgp); |
54ab4ff4 PZ |
6349 | } |
6350 | } | |
6351 | ||
2c402dc3 PZ |
6352 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6353 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6354 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6355 | int cpu) |
6356 | { | |
54ab4ff4 | 6357 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6358 | if (!sd) |
d069b916 | 6359 | return child; |
2c402dc3 PZ |
6360 | |
6361 | set_domain_attribute(sd, attr); | |
6362 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
6363 | if (child) { |
6364 | sd->level = child->level + 1; | |
6365 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6366 | child->parent = sd; |
60495e77 | 6367 | } |
d069b916 | 6368 | sd->child = child; |
2c402dc3 PZ |
6369 | |
6370 | return sd; | |
6371 | } | |
6372 | ||
2109b99e AH |
6373 | /* |
6374 | * Build sched domains for a given set of cpus and attach the sched domains | |
6375 | * to the individual cpus | |
6376 | */ | |
dce840a0 PZ |
6377 | static int build_sched_domains(const struct cpumask *cpu_map, |
6378 | struct sched_domain_attr *attr) | |
2109b99e AH |
6379 | { |
6380 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6381 | struct sched_domain *sd; |
2109b99e | 6382 | struct s_data d; |
822ff793 | 6383 | int i, ret = -ENOMEM; |
9c1cfda2 | 6384 | |
2109b99e AH |
6385 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6386 | if (alloc_state != sa_rootdomain) | |
6387 | goto error; | |
9c1cfda2 | 6388 | |
dce840a0 | 6389 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6390 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6391 | struct sched_domain_topology_level *tl; |
6392 | ||
3bd65a80 | 6393 | sd = NULL; |
e3589f6c | 6394 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6395 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6396 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6397 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6398 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6399 | break; | |
e3589f6c | 6400 | } |
d274cb30 | 6401 | |
d069b916 PZ |
6402 | while (sd->child) |
6403 | sd = sd->child; | |
6404 | ||
21d42ccf | 6405 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6406 | } |
6407 | ||
6408 | /* Build the groups for the domains */ | |
6409 | for_each_cpu(i, cpu_map) { | |
6410 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6411 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6412 | if (sd->flags & SD_OVERLAP) { |
6413 | if (build_overlap_sched_groups(sd, i)) | |
6414 | goto error; | |
6415 | } else { | |
6416 | if (build_sched_groups(sd, i)) | |
6417 | goto error; | |
6418 | } | |
1cf51902 | 6419 | } |
a06dadbe | 6420 | } |
9c1cfda2 | 6421 | |
1da177e4 | 6422 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6423 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6424 | if (!cpumask_test_cpu(i, cpu_map)) | |
6425 | continue; | |
9c1cfda2 | 6426 | |
dce840a0 PZ |
6427 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6428 | claim_allocations(i, sd); | |
cd4ea6ae | 6429 | init_sched_groups_power(i, sd); |
dce840a0 | 6430 | } |
f712c0c7 | 6431 | } |
9c1cfda2 | 6432 | |
1da177e4 | 6433 | /* Attach the domains */ |
dce840a0 | 6434 | rcu_read_lock(); |
abcd083a | 6435 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6436 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6437 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6438 | } |
dce840a0 | 6439 | rcu_read_unlock(); |
51888ca2 | 6440 | |
822ff793 | 6441 | ret = 0; |
51888ca2 | 6442 | error: |
2109b99e | 6443 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6444 | return ret; |
1da177e4 | 6445 | } |
029190c5 | 6446 | |
acc3f5d7 | 6447 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6448 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6449 | static struct sched_domain_attr *dattr_cur; |
6450 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6451 | |
6452 | /* | |
6453 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6454 | * cpumask) fails, then fallback to a single sched domain, |
6455 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6456 | */ |
4212823f | 6457 | static cpumask_var_t fallback_doms; |
029190c5 | 6458 | |
ee79d1bd HC |
6459 | /* |
6460 | * arch_update_cpu_topology lets virtualized architectures update the | |
6461 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6462 | * or 0 if it stayed the same. | |
6463 | */ | |
6464 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6465 | { |
ee79d1bd | 6466 | return 0; |
22e52b07 HC |
6467 | } |
6468 | ||
acc3f5d7 RR |
6469 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6470 | { | |
6471 | int i; | |
6472 | cpumask_var_t *doms; | |
6473 | ||
6474 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6475 | if (!doms) | |
6476 | return NULL; | |
6477 | for (i = 0; i < ndoms; i++) { | |
6478 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6479 | free_sched_domains(doms, i); | |
6480 | return NULL; | |
6481 | } | |
6482 | } | |
6483 | return doms; | |
6484 | } | |
6485 | ||
6486 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6487 | { | |
6488 | unsigned int i; | |
6489 | for (i = 0; i < ndoms; i++) | |
6490 | free_cpumask_var(doms[i]); | |
6491 | kfree(doms); | |
6492 | } | |
6493 | ||
1a20ff27 | 6494 | /* |
41a2d6cf | 6495 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6496 | * For now this just excludes isolated cpus, but could be used to |
6497 | * exclude other special cases in the future. | |
1a20ff27 | 6498 | */ |
c4a8849a | 6499 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6500 | { |
7378547f MM |
6501 | int err; |
6502 | ||
22e52b07 | 6503 | arch_update_cpu_topology(); |
029190c5 | 6504 | ndoms_cur = 1; |
acc3f5d7 | 6505 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6506 | if (!doms_cur) |
acc3f5d7 RR |
6507 | doms_cur = &fallback_doms; |
6508 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 6509 | dattr_cur = NULL; |
dce840a0 | 6510 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6511 | register_sched_domain_sysctl(); |
7378547f MM |
6512 | |
6513 | return err; | |
1a20ff27 DG |
6514 | } |
6515 | ||
1a20ff27 DG |
6516 | /* |
6517 | * Detach sched domains from a group of cpus specified in cpu_map | |
6518 | * These cpus will now be attached to the NULL domain | |
6519 | */ | |
96f874e2 | 6520 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6521 | { |
6522 | int i; | |
6523 | ||
dce840a0 | 6524 | rcu_read_lock(); |
abcd083a | 6525 | for_each_cpu(i, cpu_map) |
57d885fe | 6526 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6527 | rcu_read_unlock(); |
1a20ff27 DG |
6528 | } |
6529 | ||
1d3504fc HS |
6530 | /* handle null as "default" */ |
6531 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6532 | struct sched_domain_attr *new, int idx_new) | |
6533 | { | |
6534 | struct sched_domain_attr tmp; | |
6535 | ||
6536 | /* fast path */ | |
6537 | if (!new && !cur) | |
6538 | return 1; | |
6539 | ||
6540 | tmp = SD_ATTR_INIT; | |
6541 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6542 | new ? (new + idx_new) : &tmp, | |
6543 | sizeof(struct sched_domain_attr)); | |
6544 | } | |
6545 | ||
029190c5 PJ |
6546 | /* |
6547 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6548 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6549 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6550 | * It destroys each deleted domain and builds each new domain. | |
6551 | * | |
acc3f5d7 | 6552 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6553 | * The masks don't intersect (don't overlap.) We should setup one |
6554 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6555 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6556 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6557 | * it as it is. | |
6558 | * | |
acc3f5d7 RR |
6559 | * The passed in 'doms_new' should be allocated using |
6560 | * alloc_sched_domains. This routine takes ownership of it and will | |
6561 | * free_sched_domains it when done with it. If the caller failed the | |
6562 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6563 | * and partition_sched_domains() will fallback to the single partition | |
6564 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6565 | * |
96f874e2 | 6566 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6567 | * ndoms_new == 0 is a special case for destroying existing domains, |
6568 | * and it will not create the default domain. | |
dfb512ec | 6569 | * |
029190c5 PJ |
6570 | * Call with hotplug lock held |
6571 | */ | |
acc3f5d7 | 6572 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6573 | struct sched_domain_attr *dattr_new) |
029190c5 | 6574 | { |
dfb512ec | 6575 | int i, j, n; |
d65bd5ec | 6576 | int new_topology; |
029190c5 | 6577 | |
712555ee | 6578 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6579 | |
7378547f MM |
6580 | /* always unregister in case we don't destroy any domains */ |
6581 | unregister_sched_domain_sysctl(); | |
6582 | ||
d65bd5ec HC |
6583 | /* Let architecture update cpu core mappings. */ |
6584 | new_topology = arch_update_cpu_topology(); | |
6585 | ||
dfb512ec | 6586 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6587 | |
6588 | /* Destroy deleted domains */ | |
6589 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6590 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6591 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6592 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6593 | goto match1; |
6594 | } | |
6595 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6596 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6597 | match1: |
6598 | ; | |
6599 | } | |
6600 | ||
e761b772 MK |
6601 | if (doms_new == NULL) { |
6602 | ndoms_cur = 0; | |
acc3f5d7 | 6603 | doms_new = &fallback_doms; |
6ad4c188 | 6604 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6605 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6606 | } |
6607 | ||
029190c5 PJ |
6608 | /* Build new domains */ |
6609 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6610 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6611 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6612 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6613 | goto match2; |
6614 | } | |
6615 | /* no match - add a new doms_new */ | |
dce840a0 | 6616 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6617 | match2: |
6618 | ; | |
6619 | } | |
6620 | ||
6621 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6622 | if (doms_cur != &fallback_doms) |
6623 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6624 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6625 | doms_cur = doms_new; |
1d3504fc | 6626 | dattr_cur = dattr_new; |
029190c5 | 6627 | ndoms_cur = ndoms_new; |
7378547f MM |
6628 | |
6629 | register_sched_domain_sysctl(); | |
a1835615 | 6630 | |
712555ee | 6631 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6632 | } |
6633 | ||
5c45bf27 | 6634 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 6635 | static void reinit_sched_domains(void) |
5c45bf27 | 6636 | { |
95402b38 | 6637 | get_online_cpus(); |
dfb512ec MK |
6638 | |
6639 | /* Destroy domains first to force the rebuild */ | |
6640 | partition_sched_domains(0, NULL, NULL); | |
6641 | ||
e761b772 | 6642 | rebuild_sched_domains(); |
95402b38 | 6643 | put_online_cpus(); |
5c45bf27 SS |
6644 | } |
6645 | ||
6646 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6647 | { | |
afb8a9b7 | 6648 | unsigned int level = 0; |
5c45bf27 | 6649 | |
afb8a9b7 GS |
6650 | if (sscanf(buf, "%u", &level) != 1) |
6651 | return -EINVAL; | |
6652 | ||
6653 | /* | |
6654 | * level is always be positive so don't check for | |
6655 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
6656 | * What happens on 0 or 1 byte write, | |
6657 | * need to check for count as well? | |
6658 | */ | |
6659 | ||
6660 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
6661 | return -EINVAL; |
6662 | ||
6663 | if (smt) | |
afb8a9b7 | 6664 | sched_smt_power_savings = level; |
5c45bf27 | 6665 | else |
afb8a9b7 | 6666 | sched_mc_power_savings = level; |
5c45bf27 | 6667 | |
c4a8849a | 6668 | reinit_sched_domains(); |
5c45bf27 | 6669 | |
c70f22d2 | 6670 | return count; |
5c45bf27 SS |
6671 | } |
6672 | ||
5c45bf27 | 6673 | #ifdef CONFIG_SCHED_MC |
8a25a2fd KS |
6674 | static ssize_t sched_mc_power_savings_show(struct device *dev, |
6675 | struct device_attribute *attr, | |
6676 | char *buf) | |
5c45bf27 | 6677 | { |
8a25a2fd | 6678 | return sprintf(buf, "%u\n", sched_mc_power_savings); |
5c45bf27 | 6679 | } |
8a25a2fd KS |
6680 | static ssize_t sched_mc_power_savings_store(struct device *dev, |
6681 | struct device_attribute *attr, | |
48f24c4d | 6682 | const char *buf, size_t count) |
5c45bf27 SS |
6683 | { |
6684 | return sched_power_savings_store(buf, count, 0); | |
6685 | } | |
8a25a2fd KS |
6686 | static DEVICE_ATTR(sched_mc_power_savings, 0644, |
6687 | sched_mc_power_savings_show, | |
6688 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6689 | #endif |
6690 | ||
6691 | #ifdef CONFIG_SCHED_SMT | |
8a25a2fd KS |
6692 | static ssize_t sched_smt_power_savings_show(struct device *dev, |
6693 | struct device_attribute *attr, | |
6694 | char *buf) | |
5c45bf27 | 6695 | { |
8a25a2fd | 6696 | return sprintf(buf, "%u\n", sched_smt_power_savings); |
5c45bf27 | 6697 | } |
8a25a2fd KS |
6698 | static ssize_t sched_smt_power_savings_store(struct device *dev, |
6699 | struct device_attribute *attr, | |
48f24c4d | 6700 | const char *buf, size_t count) |
5c45bf27 SS |
6701 | { |
6702 | return sched_power_savings_store(buf, count, 1); | |
6703 | } | |
8a25a2fd | 6704 | static DEVICE_ATTR(sched_smt_power_savings, 0644, |
f718cd4a | 6705 | sched_smt_power_savings_show, |
6707de00 AB |
6706 | sched_smt_power_savings_store); |
6707 | #endif | |
6708 | ||
8a25a2fd | 6709 | int __init sched_create_sysfs_power_savings_entries(struct device *dev) |
6707de00 AB |
6710 | { |
6711 | int err = 0; | |
6712 | ||
6713 | #ifdef CONFIG_SCHED_SMT | |
6714 | if (smt_capable()) | |
8a25a2fd | 6715 | err = device_create_file(dev, &dev_attr_sched_smt_power_savings); |
6707de00 AB |
6716 | #endif |
6717 | #ifdef CONFIG_SCHED_MC | |
6718 | if (!err && mc_capable()) | |
8a25a2fd | 6719 | err = device_create_file(dev, &dev_attr_sched_mc_power_savings); |
6707de00 AB |
6720 | #endif |
6721 | return err; | |
6722 | } | |
6d6bc0ad | 6723 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 6724 | |
1da177e4 | 6725 | /* |
3a101d05 TH |
6726 | * Update cpusets according to cpu_active mask. If cpusets are |
6727 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6728 | * around partition_sched_domains(). | |
1da177e4 | 6729 | */ |
0b2e918a TH |
6730 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6731 | void *hcpu) | |
e761b772 | 6732 | { |
3a101d05 | 6733 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 6734 | case CPU_ONLINE: |
6ad4c188 | 6735 | case CPU_DOWN_FAILED: |
3a101d05 | 6736 | cpuset_update_active_cpus(); |
e761b772 | 6737 | return NOTIFY_OK; |
3a101d05 TH |
6738 | default: |
6739 | return NOTIFY_DONE; | |
6740 | } | |
6741 | } | |
e761b772 | 6742 | |
0b2e918a TH |
6743 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6744 | void *hcpu) | |
3a101d05 TH |
6745 | { |
6746 | switch (action & ~CPU_TASKS_FROZEN) { | |
6747 | case CPU_DOWN_PREPARE: | |
6748 | cpuset_update_active_cpus(); | |
6749 | return NOTIFY_OK; | |
e761b772 MK |
6750 | default: |
6751 | return NOTIFY_DONE; | |
6752 | } | |
6753 | } | |
e761b772 | 6754 | |
1da177e4 LT |
6755 | void __init sched_init_smp(void) |
6756 | { | |
dcc30a35 RR |
6757 | cpumask_var_t non_isolated_cpus; |
6758 | ||
6759 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6760 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6761 | |
95402b38 | 6762 | get_online_cpus(); |
712555ee | 6763 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6764 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6765 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6766 | if (cpumask_empty(non_isolated_cpus)) | |
6767 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6768 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6769 | put_online_cpus(); |
e761b772 | 6770 | |
3a101d05 TH |
6771 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6772 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6773 | |
6774 | /* RT runtime code needs to handle some hotplug events */ | |
6775 | hotcpu_notifier(update_runtime, 0); | |
6776 | ||
b328ca18 | 6777 | init_hrtick(); |
5c1e1767 NP |
6778 | |
6779 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6780 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6781 | BUG(); |
19978ca6 | 6782 | sched_init_granularity(); |
dcc30a35 | 6783 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6784 | |
0e3900e6 | 6785 | init_sched_rt_class(); |
1da177e4 LT |
6786 | } |
6787 | #else | |
6788 | void __init sched_init_smp(void) | |
6789 | { | |
19978ca6 | 6790 | sched_init_granularity(); |
1da177e4 LT |
6791 | } |
6792 | #endif /* CONFIG_SMP */ | |
6793 | ||
cd1bb94b AB |
6794 | const_debug unsigned int sysctl_timer_migration = 1; |
6795 | ||
1da177e4 LT |
6796 | int in_sched_functions(unsigned long addr) |
6797 | { | |
1da177e4 LT |
6798 | return in_lock_functions(addr) || |
6799 | (addr >= (unsigned long)__sched_text_start | |
6800 | && addr < (unsigned long)__sched_text_end); | |
6801 | } | |
6802 | ||
029632fb PZ |
6803 | #ifdef CONFIG_CGROUP_SCHED |
6804 | struct task_group root_task_group; | |
052f1dc7 | 6805 | #endif |
6f505b16 | 6806 | |
029632fb | 6807 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6808 | |
1da177e4 LT |
6809 | void __init sched_init(void) |
6810 | { | |
dd41f596 | 6811 | int i, j; |
434d53b0 MT |
6812 | unsigned long alloc_size = 0, ptr; |
6813 | ||
6814 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6815 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6816 | #endif | |
6817 | #ifdef CONFIG_RT_GROUP_SCHED | |
6818 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6819 | #endif |
df7c8e84 | 6820 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6821 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6822 | #endif |
434d53b0 | 6823 | if (alloc_size) { |
36b7b6d4 | 6824 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6825 | |
6826 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6827 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6828 | ptr += nr_cpu_ids * sizeof(void **); |
6829 | ||
07e06b01 | 6830 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6831 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6832 | |
6d6bc0ad | 6833 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6834 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6835 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6836 | ptr += nr_cpu_ids * sizeof(void **); |
6837 | ||
07e06b01 | 6838 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6839 | ptr += nr_cpu_ids * sizeof(void **); |
6840 | ||
6d6bc0ad | 6841 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6842 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6843 | for_each_possible_cpu(i) { | |
6844 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6845 | ptr += cpumask_size(); | |
6846 | } | |
6847 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6848 | } |
dd41f596 | 6849 | |
57d885fe GH |
6850 | #ifdef CONFIG_SMP |
6851 | init_defrootdomain(); | |
6852 | #endif | |
6853 | ||
d0b27fa7 PZ |
6854 | init_rt_bandwidth(&def_rt_bandwidth, |
6855 | global_rt_period(), global_rt_runtime()); | |
6856 | ||
6857 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6858 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6859 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6860 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6861 | |
7c941438 | 6862 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6863 | list_add(&root_task_group.list, &task_groups); |
6864 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6865 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6866 | autogroup_init(&init_task); |
54c707e9 | 6867 | |
7c941438 | 6868 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6869 | |
54c707e9 GC |
6870 | #ifdef CONFIG_CGROUP_CPUACCT |
6871 | root_cpuacct.cpustat = &kernel_cpustat; | |
6872 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
6873 | /* Too early, not expected to fail */ | |
6874 | BUG_ON(!root_cpuacct.cpuusage); | |
6875 | #endif | |
0a945022 | 6876 | for_each_possible_cpu(i) { |
70b97a7f | 6877 | struct rq *rq; |
1da177e4 LT |
6878 | |
6879 | rq = cpu_rq(i); | |
05fa785c | 6880 | raw_spin_lock_init(&rq->lock); |
7897986b | 6881 | rq->nr_running = 0; |
dce48a84 TG |
6882 | rq->calc_load_active = 0; |
6883 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6884 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6885 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6886 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6887 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6888 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6889 | /* |
07e06b01 | 6890 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6891 | * |
6892 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6893 | * gets 100% of the cpu resources in the system. This overall | |
6894 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6895 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6896 | * based on each entity's (task or task-group's) weight |
6897 | * (se->load.weight). | |
6898 | * | |
07e06b01 | 6899 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6900 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6901 | * then A0's share of the cpu resource is: | |
6902 | * | |
0d905bca | 6903 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6904 | * |
07e06b01 YZ |
6905 | * We achieve this by letting root_task_group's tasks sit |
6906 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6907 | */ |
ab84d31e | 6908 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6909 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6910 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6911 | ||
6912 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6913 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6914 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6915 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6916 | #endif |
1da177e4 | 6917 | |
dd41f596 IM |
6918 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6919 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6920 | |
6921 | rq->last_load_update_tick = jiffies; | |
6922 | ||
1da177e4 | 6923 | #ifdef CONFIG_SMP |
41c7ce9a | 6924 | rq->sd = NULL; |
57d885fe | 6925 | rq->rd = NULL; |
1399fa78 | 6926 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6927 | rq->post_schedule = 0; |
1da177e4 | 6928 | rq->active_balance = 0; |
dd41f596 | 6929 | rq->next_balance = jiffies; |
1da177e4 | 6930 | rq->push_cpu = 0; |
0a2966b4 | 6931 | rq->cpu = i; |
1f11eb6a | 6932 | rq->online = 0; |
eae0c9df MG |
6933 | rq->idle_stamp = 0; |
6934 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 6935 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 6936 | #ifdef CONFIG_NO_HZ |
1c792db7 | 6937 | rq->nohz_flags = 0; |
83cd4fe2 | 6938 | #endif |
1da177e4 | 6939 | #endif |
8f4d37ec | 6940 | init_rq_hrtick(rq); |
1da177e4 | 6941 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6942 | } |
6943 | ||
2dd73a4f | 6944 | set_load_weight(&init_task); |
b50f60ce | 6945 | |
e107be36 AK |
6946 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6947 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6948 | #endif | |
6949 | ||
b50f60ce | 6950 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6951 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6952 | #endif |
6953 | ||
1da177e4 LT |
6954 | /* |
6955 | * The boot idle thread does lazy MMU switching as well: | |
6956 | */ | |
6957 | atomic_inc(&init_mm.mm_count); | |
6958 | enter_lazy_tlb(&init_mm, current); | |
6959 | ||
6960 | /* | |
6961 | * Make us the idle thread. Technically, schedule() should not be | |
6962 | * called from this thread, however somewhere below it might be, | |
6963 | * but because we are the idle thread, we just pick up running again | |
6964 | * when this runqueue becomes "idle". | |
6965 | */ | |
6966 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6967 | |
6968 | calc_load_update = jiffies + LOAD_FREQ; | |
6969 | ||
dd41f596 IM |
6970 | /* |
6971 | * During early bootup we pretend to be a normal task: | |
6972 | */ | |
6973 | current->sched_class = &fair_sched_class; | |
6892b75e | 6974 | |
bf4d83f6 | 6975 | #ifdef CONFIG_SMP |
4cb98839 | 6976 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6977 | /* May be allocated at isolcpus cmdline parse time */ |
6978 | if (cpu_isolated_map == NULL) | |
6979 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
029632fb PZ |
6980 | #endif |
6981 | init_sched_fair_class(); | |
6a7b3dc3 | 6982 | |
6892b75e | 6983 | scheduler_running = 1; |
1da177e4 LT |
6984 | } |
6985 | ||
d902db1e | 6986 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6987 | static inline int preempt_count_equals(int preempt_offset) |
6988 | { | |
234da7bc | 6989 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6990 | |
4ba8216c | 6991 | return (nested == preempt_offset); |
e4aafea2 FW |
6992 | } |
6993 | ||
d894837f | 6994 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6995 | { |
1da177e4 LT |
6996 | static unsigned long prev_jiffy; /* ratelimiting */ |
6997 | ||
b3fbab05 | 6998 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6999 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7000 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7001 | return; |
7002 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7003 | return; | |
7004 | prev_jiffy = jiffies; | |
7005 | ||
3df0fc5b PZ |
7006 | printk(KERN_ERR |
7007 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7008 | file, line); | |
7009 | printk(KERN_ERR | |
7010 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7011 | in_atomic(), irqs_disabled(), | |
7012 | current->pid, current->comm); | |
aef745fc IM |
7013 | |
7014 | debug_show_held_locks(current); | |
7015 | if (irqs_disabled()) | |
7016 | print_irqtrace_events(current); | |
7017 | dump_stack(); | |
1da177e4 LT |
7018 | } |
7019 | EXPORT_SYMBOL(__might_sleep); | |
7020 | #endif | |
7021 | ||
7022 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7023 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7024 | { | |
da7a735e PZ |
7025 | const struct sched_class *prev_class = p->sched_class; |
7026 | int old_prio = p->prio; | |
3a5e4dc1 | 7027 | int on_rq; |
3e51f33f | 7028 | |
fd2f4419 | 7029 | on_rq = p->on_rq; |
3a5e4dc1 | 7030 | if (on_rq) |
4ca9b72b | 7031 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7032 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7033 | if (on_rq) { | |
4ca9b72b | 7034 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7035 | resched_task(rq->curr); |
7036 | } | |
da7a735e PZ |
7037 | |
7038 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7039 | } |
7040 | ||
1da177e4 LT |
7041 | void normalize_rt_tasks(void) |
7042 | { | |
a0f98a1c | 7043 | struct task_struct *g, *p; |
1da177e4 | 7044 | unsigned long flags; |
70b97a7f | 7045 | struct rq *rq; |
1da177e4 | 7046 | |
4cf5d77a | 7047 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7048 | do_each_thread(g, p) { |
178be793 IM |
7049 | /* |
7050 | * Only normalize user tasks: | |
7051 | */ | |
7052 | if (!p->mm) | |
7053 | continue; | |
7054 | ||
6cfb0d5d | 7055 | p->se.exec_start = 0; |
6cfb0d5d | 7056 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7057 | p->se.statistics.wait_start = 0; |
7058 | p->se.statistics.sleep_start = 0; | |
7059 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7060 | #endif |
dd41f596 IM |
7061 | |
7062 | if (!rt_task(p)) { | |
7063 | /* | |
7064 | * Renice negative nice level userspace | |
7065 | * tasks back to 0: | |
7066 | */ | |
7067 | if (TASK_NICE(p) < 0 && p->mm) | |
7068 | set_user_nice(p, 0); | |
1da177e4 | 7069 | continue; |
dd41f596 | 7070 | } |
1da177e4 | 7071 | |
1d615482 | 7072 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7073 | rq = __task_rq_lock(p); |
1da177e4 | 7074 | |
178be793 | 7075 | normalize_task(rq, p); |
3a5e4dc1 | 7076 | |
b29739f9 | 7077 | __task_rq_unlock(rq); |
1d615482 | 7078 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7079 | } while_each_thread(g, p); |
7080 | ||
4cf5d77a | 7081 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7082 | } |
7083 | ||
7084 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7085 | |
67fc4e0c | 7086 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7087 | /* |
67fc4e0c | 7088 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7089 | * |
7090 | * They can only be called when the whole system has been | |
7091 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7092 | * activity can take place. Using them for anything else would | |
7093 | * be a serious bug, and as a result, they aren't even visible | |
7094 | * under any other configuration. | |
7095 | */ | |
7096 | ||
7097 | /** | |
7098 | * curr_task - return the current task for a given cpu. | |
7099 | * @cpu: the processor in question. | |
7100 | * | |
7101 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7102 | */ | |
36c8b586 | 7103 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7104 | { |
7105 | return cpu_curr(cpu); | |
7106 | } | |
7107 | ||
67fc4e0c JW |
7108 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7109 | ||
7110 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7111 | /** |
7112 | * set_curr_task - set the current task for a given cpu. | |
7113 | * @cpu: the processor in question. | |
7114 | * @p: the task pointer to set. | |
7115 | * | |
7116 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7117 | * are serviced on a separate stack. It allows the architecture to switch the |
7118 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7119 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7120 | * and caller must save the original value of the current task (see | |
7121 | * curr_task() above) and restore that value before reenabling interrupts and | |
7122 | * re-starting the system. | |
7123 | * | |
7124 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7125 | */ | |
36c8b586 | 7126 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7127 | { |
7128 | cpu_curr(cpu) = p; | |
7129 | } | |
7130 | ||
7131 | #endif | |
29f59db3 | 7132 | |
7c941438 | 7133 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7134 | /* task_group_lock serializes the addition/removal of task groups */ |
7135 | static DEFINE_SPINLOCK(task_group_lock); | |
7136 | ||
bccbe08a PZ |
7137 | static void free_sched_group(struct task_group *tg) |
7138 | { | |
7139 | free_fair_sched_group(tg); | |
7140 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7141 | autogroup_free(tg); |
bccbe08a PZ |
7142 | kfree(tg); |
7143 | } | |
7144 | ||
7145 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7146 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7147 | { |
7148 | struct task_group *tg; | |
7149 | unsigned long flags; | |
bccbe08a PZ |
7150 | |
7151 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7152 | if (!tg) | |
7153 | return ERR_PTR(-ENOMEM); | |
7154 | ||
ec7dc8ac | 7155 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7156 | goto err; |
7157 | ||
ec7dc8ac | 7158 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7159 | goto err; |
7160 | ||
8ed36996 | 7161 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7162 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7163 | |
7164 | WARN_ON(!parent); /* root should already exist */ | |
7165 | ||
7166 | tg->parent = parent; | |
f473aa5e | 7167 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7168 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7169 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7170 | |
9b5b7751 | 7171 | return tg; |
29f59db3 SV |
7172 | |
7173 | err: | |
6f505b16 | 7174 | free_sched_group(tg); |
29f59db3 SV |
7175 | return ERR_PTR(-ENOMEM); |
7176 | } | |
7177 | ||
9b5b7751 | 7178 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7179 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7180 | { |
29f59db3 | 7181 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7182 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7183 | } |
7184 | ||
9b5b7751 | 7185 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7186 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7187 | { |
8ed36996 | 7188 | unsigned long flags; |
9b5b7751 | 7189 | int i; |
29f59db3 | 7190 | |
3d4b47b4 PZ |
7191 | /* end participation in shares distribution */ |
7192 | for_each_possible_cpu(i) | |
bccbe08a | 7193 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7194 | |
7195 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7196 | list_del_rcu(&tg->list); |
f473aa5e | 7197 | list_del_rcu(&tg->siblings); |
8ed36996 | 7198 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7199 | |
9b5b7751 | 7200 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7201 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7202 | } |
7203 | ||
9b5b7751 | 7204 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7205 | * The caller of this function should have put the task in its new group |
7206 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7207 | * reflect its new group. | |
9b5b7751 SV |
7208 | */ |
7209 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7210 | { |
7211 | int on_rq, running; | |
7212 | unsigned long flags; | |
7213 | struct rq *rq; | |
7214 | ||
7215 | rq = task_rq_lock(tsk, &flags); | |
7216 | ||
051a1d1a | 7217 | running = task_current(rq, tsk); |
fd2f4419 | 7218 | on_rq = tsk->on_rq; |
29f59db3 | 7219 | |
0e1f3483 | 7220 | if (on_rq) |
29f59db3 | 7221 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7222 | if (unlikely(running)) |
7223 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7224 | |
810b3817 | 7225 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7226 | if (tsk->sched_class->task_move_group) |
7227 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7228 | else | |
810b3817 | 7229 | #endif |
b2b5ce02 | 7230 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7231 | |
0e1f3483 HS |
7232 | if (unlikely(running)) |
7233 | tsk->sched_class->set_curr_task(rq); | |
7234 | if (on_rq) | |
371fd7e7 | 7235 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7236 | |
0122ec5b | 7237 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7238 | } |
7c941438 | 7239 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7240 | |
a790de99 | 7241 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7242 | static unsigned long to_ratio(u64 period, u64 runtime) |
7243 | { | |
7244 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7245 | return 1ULL << 20; |
9f0c1e56 | 7246 | |
9a7e0b18 | 7247 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7248 | } |
a790de99 PT |
7249 | #endif |
7250 | ||
7251 | #ifdef CONFIG_RT_GROUP_SCHED | |
7252 | /* | |
7253 | * Ensure that the real time constraints are schedulable. | |
7254 | */ | |
7255 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7256 | |
9a7e0b18 PZ |
7257 | /* Must be called with tasklist_lock held */ |
7258 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7259 | { |
9a7e0b18 | 7260 | struct task_struct *g, *p; |
b40b2e8e | 7261 | |
9a7e0b18 | 7262 | do_each_thread(g, p) { |
029632fb | 7263 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7264 | return 1; |
7265 | } while_each_thread(g, p); | |
b40b2e8e | 7266 | |
9a7e0b18 PZ |
7267 | return 0; |
7268 | } | |
b40b2e8e | 7269 | |
9a7e0b18 PZ |
7270 | struct rt_schedulable_data { |
7271 | struct task_group *tg; | |
7272 | u64 rt_period; | |
7273 | u64 rt_runtime; | |
7274 | }; | |
b40b2e8e | 7275 | |
a790de99 | 7276 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7277 | { |
7278 | struct rt_schedulable_data *d = data; | |
7279 | struct task_group *child; | |
7280 | unsigned long total, sum = 0; | |
7281 | u64 period, runtime; | |
b40b2e8e | 7282 | |
9a7e0b18 PZ |
7283 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7284 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7285 | |
9a7e0b18 PZ |
7286 | if (tg == d->tg) { |
7287 | period = d->rt_period; | |
7288 | runtime = d->rt_runtime; | |
b40b2e8e | 7289 | } |
b40b2e8e | 7290 | |
4653f803 PZ |
7291 | /* |
7292 | * Cannot have more runtime than the period. | |
7293 | */ | |
7294 | if (runtime > period && runtime != RUNTIME_INF) | |
7295 | return -EINVAL; | |
6f505b16 | 7296 | |
4653f803 PZ |
7297 | /* |
7298 | * Ensure we don't starve existing RT tasks. | |
7299 | */ | |
9a7e0b18 PZ |
7300 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7301 | return -EBUSY; | |
6f505b16 | 7302 | |
9a7e0b18 | 7303 | total = to_ratio(period, runtime); |
6f505b16 | 7304 | |
4653f803 PZ |
7305 | /* |
7306 | * Nobody can have more than the global setting allows. | |
7307 | */ | |
7308 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7309 | return -EINVAL; | |
6f505b16 | 7310 | |
4653f803 PZ |
7311 | /* |
7312 | * The sum of our children's runtime should not exceed our own. | |
7313 | */ | |
9a7e0b18 PZ |
7314 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7315 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7316 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7317 | |
9a7e0b18 PZ |
7318 | if (child == d->tg) { |
7319 | period = d->rt_period; | |
7320 | runtime = d->rt_runtime; | |
7321 | } | |
6f505b16 | 7322 | |
9a7e0b18 | 7323 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7324 | } |
6f505b16 | 7325 | |
9a7e0b18 PZ |
7326 | if (sum > total) |
7327 | return -EINVAL; | |
7328 | ||
7329 | return 0; | |
6f505b16 PZ |
7330 | } |
7331 | ||
9a7e0b18 | 7332 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7333 | { |
8277434e PT |
7334 | int ret; |
7335 | ||
9a7e0b18 PZ |
7336 | struct rt_schedulable_data data = { |
7337 | .tg = tg, | |
7338 | .rt_period = period, | |
7339 | .rt_runtime = runtime, | |
7340 | }; | |
7341 | ||
8277434e PT |
7342 | rcu_read_lock(); |
7343 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7344 | rcu_read_unlock(); | |
7345 | ||
7346 | return ret; | |
521f1a24 DG |
7347 | } |
7348 | ||
ab84d31e | 7349 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7350 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7351 | { |
ac086bc2 | 7352 | int i, err = 0; |
9f0c1e56 | 7353 | |
9f0c1e56 | 7354 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7355 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7356 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7357 | if (err) | |
9f0c1e56 | 7358 | goto unlock; |
ac086bc2 | 7359 | |
0986b11b | 7360 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7361 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7362 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7363 | |
7364 | for_each_possible_cpu(i) { | |
7365 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7366 | ||
0986b11b | 7367 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7368 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7369 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7370 | } |
0986b11b | 7371 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7372 | unlock: |
521f1a24 | 7373 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7374 | mutex_unlock(&rt_constraints_mutex); |
7375 | ||
7376 | return err; | |
6f505b16 PZ |
7377 | } |
7378 | ||
d0b27fa7 PZ |
7379 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7380 | { | |
7381 | u64 rt_runtime, rt_period; | |
7382 | ||
7383 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7384 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7385 | if (rt_runtime_us < 0) | |
7386 | rt_runtime = RUNTIME_INF; | |
7387 | ||
ab84d31e | 7388 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7389 | } |
7390 | ||
9f0c1e56 PZ |
7391 | long sched_group_rt_runtime(struct task_group *tg) |
7392 | { | |
7393 | u64 rt_runtime_us; | |
7394 | ||
d0b27fa7 | 7395 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7396 | return -1; |
7397 | ||
d0b27fa7 | 7398 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7399 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7400 | return rt_runtime_us; | |
7401 | } | |
d0b27fa7 PZ |
7402 | |
7403 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7404 | { | |
7405 | u64 rt_runtime, rt_period; | |
7406 | ||
7407 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7408 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7409 | ||
619b0488 R |
7410 | if (rt_period == 0) |
7411 | return -EINVAL; | |
7412 | ||
ab84d31e | 7413 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7414 | } |
7415 | ||
7416 | long sched_group_rt_period(struct task_group *tg) | |
7417 | { | |
7418 | u64 rt_period_us; | |
7419 | ||
7420 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7421 | do_div(rt_period_us, NSEC_PER_USEC); | |
7422 | return rt_period_us; | |
7423 | } | |
7424 | ||
7425 | static int sched_rt_global_constraints(void) | |
7426 | { | |
4653f803 | 7427 | u64 runtime, period; |
d0b27fa7 PZ |
7428 | int ret = 0; |
7429 | ||
ec5d4989 HS |
7430 | if (sysctl_sched_rt_period <= 0) |
7431 | return -EINVAL; | |
7432 | ||
4653f803 PZ |
7433 | runtime = global_rt_runtime(); |
7434 | period = global_rt_period(); | |
7435 | ||
7436 | /* | |
7437 | * Sanity check on the sysctl variables. | |
7438 | */ | |
7439 | if (runtime > period && runtime != RUNTIME_INF) | |
7440 | return -EINVAL; | |
10b612f4 | 7441 | |
d0b27fa7 | 7442 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7443 | read_lock(&tasklist_lock); |
4653f803 | 7444 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7445 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7446 | mutex_unlock(&rt_constraints_mutex); |
7447 | ||
7448 | return ret; | |
7449 | } | |
54e99124 DG |
7450 | |
7451 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7452 | { | |
7453 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7454 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7455 | return 0; | |
7456 | ||
7457 | return 1; | |
7458 | } | |
7459 | ||
6d6bc0ad | 7460 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7461 | static int sched_rt_global_constraints(void) |
7462 | { | |
ac086bc2 PZ |
7463 | unsigned long flags; |
7464 | int i; | |
7465 | ||
ec5d4989 HS |
7466 | if (sysctl_sched_rt_period <= 0) |
7467 | return -EINVAL; | |
7468 | ||
60aa605d PZ |
7469 | /* |
7470 | * There's always some RT tasks in the root group | |
7471 | * -- migration, kstopmachine etc.. | |
7472 | */ | |
7473 | if (sysctl_sched_rt_runtime == 0) | |
7474 | return -EBUSY; | |
7475 | ||
0986b11b | 7476 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7477 | for_each_possible_cpu(i) { |
7478 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7479 | ||
0986b11b | 7480 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7481 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7482 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7483 | } |
0986b11b | 7484 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7485 | |
d0b27fa7 PZ |
7486 | return 0; |
7487 | } | |
6d6bc0ad | 7488 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7489 | |
7490 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7491 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7492 | loff_t *ppos) |
7493 | { | |
7494 | int ret; | |
7495 | int old_period, old_runtime; | |
7496 | static DEFINE_MUTEX(mutex); | |
7497 | ||
7498 | mutex_lock(&mutex); | |
7499 | old_period = sysctl_sched_rt_period; | |
7500 | old_runtime = sysctl_sched_rt_runtime; | |
7501 | ||
8d65af78 | 7502 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7503 | |
7504 | if (!ret && write) { | |
7505 | ret = sched_rt_global_constraints(); | |
7506 | if (ret) { | |
7507 | sysctl_sched_rt_period = old_period; | |
7508 | sysctl_sched_rt_runtime = old_runtime; | |
7509 | } else { | |
7510 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7511 | def_rt_bandwidth.rt_period = | |
7512 | ns_to_ktime(global_rt_period()); | |
7513 | } | |
7514 | } | |
7515 | mutex_unlock(&mutex); | |
7516 | ||
7517 | return ret; | |
7518 | } | |
68318b8e | 7519 | |
052f1dc7 | 7520 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7521 | |
7522 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7523 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7524 | { |
2b01dfe3 PM |
7525 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7526 | struct task_group, css); | |
68318b8e SV |
7527 | } |
7528 | ||
7529 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7530 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 7531 | { |
ec7dc8ac | 7532 | struct task_group *tg, *parent; |
68318b8e | 7533 | |
2b01dfe3 | 7534 | if (!cgrp->parent) { |
68318b8e | 7535 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7536 | return &root_task_group.css; |
68318b8e SV |
7537 | } |
7538 | ||
ec7dc8ac DG |
7539 | parent = cgroup_tg(cgrp->parent); |
7540 | tg = sched_create_group(parent); | |
68318b8e SV |
7541 | if (IS_ERR(tg)) |
7542 | return ERR_PTR(-ENOMEM); | |
7543 | ||
68318b8e SV |
7544 | return &tg->css; |
7545 | } | |
7546 | ||
41a2d6cf IM |
7547 | static void |
7548 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7549 | { |
2b01dfe3 | 7550 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7551 | |
7552 | sched_destroy_group(tg); | |
7553 | } | |
7554 | ||
bb9d97b6 TH |
7555 | static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7556 | struct cgroup_taskset *tset) | |
68318b8e | 7557 | { |
bb9d97b6 TH |
7558 | struct task_struct *task; |
7559 | ||
7560 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7561 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7562 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7563 | return -EINVAL; | |
b68aa230 | 7564 | #else |
bb9d97b6 TH |
7565 | /* We don't support RT-tasks being in separate groups */ |
7566 | if (task->sched_class != &fair_sched_class) | |
7567 | return -EINVAL; | |
b68aa230 | 7568 | #endif |
bb9d97b6 | 7569 | } |
be367d09 BB |
7570 | return 0; |
7571 | } | |
68318b8e | 7572 | |
bb9d97b6 TH |
7573 | static void cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7574 | struct cgroup_taskset *tset) | |
68318b8e | 7575 | { |
bb9d97b6 TH |
7576 | struct task_struct *task; |
7577 | ||
7578 | cgroup_taskset_for_each(task, cgrp, tset) | |
7579 | sched_move_task(task); | |
68318b8e SV |
7580 | } |
7581 | ||
068c5cc5 | 7582 | static void |
d41d5a01 PZ |
7583 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7584 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
7585 | { |
7586 | /* | |
7587 | * cgroup_exit() is called in the copy_process() failure path. | |
7588 | * Ignore this case since the task hasn't ran yet, this avoids | |
7589 | * trying to poke a half freed task state from generic code. | |
7590 | */ | |
7591 | if (!(task->flags & PF_EXITING)) | |
7592 | return; | |
7593 | ||
7594 | sched_move_task(task); | |
7595 | } | |
7596 | ||
052f1dc7 | 7597 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7598 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7599 | u64 shareval) |
68318b8e | 7600 | { |
c8b28116 | 7601 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7602 | } |
7603 | ||
f4c753b7 | 7604 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7605 | { |
2b01dfe3 | 7606 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7607 | |
c8b28116 | 7608 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7609 | } |
ab84d31e PT |
7610 | |
7611 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7612 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7613 | ||
ab84d31e PT |
7614 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7615 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7616 | ||
a790de99 PT |
7617 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7618 | ||
ab84d31e PT |
7619 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7620 | { | |
56f570e5 | 7621 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7622 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7623 | |
7624 | if (tg == &root_task_group) | |
7625 | return -EINVAL; | |
7626 | ||
7627 | /* | |
7628 | * Ensure we have at some amount of bandwidth every period. This is | |
7629 | * to prevent reaching a state of large arrears when throttled via | |
7630 | * entity_tick() resulting in prolonged exit starvation. | |
7631 | */ | |
7632 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7633 | return -EINVAL; | |
7634 | ||
7635 | /* | |
7636 | * Likewise, bound things on the otherside by preventing insane quota | |
7637 | * periods. This also allows us to normalize in computing quota | |
7638 | * feasibility. | |
7639 | */ | |
7640 | if (period > max_cfs_quota_period) | |
7641 | return -EINVAL; | |
7642 | ||
a790de99 PT |
7643 | mutex_lock(&cfs_constraints_mutex); |
7644 | ret = __cfs_schedulable(tg, period, quota); | |
7645 | if (ret) | |
7646 | goto out_unlock; | |
7647 | ||
58088ad0 | 7648 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7649 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7650 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7651 | raw_spin_lock_irq(&cfs_b->lock); |
7652 | cfs_b->period = ns_to_ktime(period); | |
7653 | cfs_b->quota = quota; | |
58088ad0 | 7654 | |
a9cf55b2 | 7655 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7656 | /* restart the period timer (if active) to handle new period expiry */ |
7657 | if (runtime_enabled && cfs_b->timer_active) { | |
7658 | /* force a reprogram */ | |
7659 | cfs_b->timer_active = 0; | |
7660 | __start_cfs_bandwidth(cfs_b); | |
7661 | } | |
ab84d31e PT |
7662 | raw_spin_unlock_irq(&cfs_b->lock); |
7663 | ||
7664 | for_each_possible_cpu(i) { | |
7665 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7666 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7667 | |
7668 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7669 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7670 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7671 | |
029632fb | 7672 | if (cfs_rq->throttled) |
671fd9da | 7673 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7674 | raw_spin_unlock_irq(&rq->lock); |
7675 | } | |
a790de99 PT |
7676 | out_unlock: |
7677 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7678 | |
a790de99 | 7679 | return ret; |
ab84d31e PT |
7680 | } |
7681 | ||
7682 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7683 | { | |
7684 | u64 quota, period; | |
7685 | ||
029632fb | 7686 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7687 | if (cfs_quota_us < 0) |
7688 | quota = RUNTIME_INF; | |
7689 | else | |
7690 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7691 | ||
7692 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7693 | } | |
7694 | ||
7695 | long tg_get_cfs_quota(struct task_group *tg) | |
7696 | { | |
7697 | u64 quota_us; | |
7698 | ||
029632fb | 7699 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7700 | return -1; |
7701 | ||
029632fb | 7702 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7703 | do_div(quota_us, NSEC_PER_USEC); |
7704 | ||
7705 | return quota_us; | |
7706 | } | |
7707 | ||
7708 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7709 | { | |
7710 | u64 quota, period; | |
7711 | ||
7712 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7713 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7714 | |
ab84d31e PT |
7715 | return tg_set_cfs_bandwidth(tg, period, quota); |
7716 | } | |
7717 | ||
7718 | long tg_get_cfs_period(struct task_group *tg) | |
7719 | { | |
7720 | u64 cfs_period_us; | |
7721 | ||
029632fb | 7722 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7723 | do_div(cfs_period_us, NSEC_PER_USEC); |
7724 | ||
7725 | return cfs_period_us; | |
7726 | } | |
7727 | ||
7728 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7729 | { | |
7730 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7731 | } | |
7732 | ||
7733 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7734 | s64 cfs_quota_us) | |
7735 | { | |
7736 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7737 | } | |
7738 | ||
7739 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7740 | { | |
7741 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7742 | } | |
7743 | ||
7744 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7745 | u64 cfs_period_us) | |
7746 | { | |
7747 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7748 | } | |
7749 | ||
a790de99 PT |
7750 | struct cfs_schedulable_data { |
7751 | struct task_group *tg; | |
7752 | u64 period, quota; | |
7753 | }; | |
7754 | ||
7755 | /* | |
7756 | * normalize group quota/period to be quota/max_period | |
7757 | * note: units are usecs | |
7758 | */ | |
7759 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7760 | struct cfs_schedulable_data *d) | |
7761 | { | |
7762 | u64 quota, period; | |
7763 | ||
7764 | if (tg == d->tg) { | |
7765 | period = d->period; | |
7766 | quota = d->quota; | |
7767 | } else { | |
7768 | period = tg_get_cfs_period(tg); | |
7769 | quota = tg_get_cfs_quota(tg); | |
7770 | } | |
7771 | ||
7772 | /* note: these should typically be equivalent */ | |
7773 | if (quota == RUNTIME_INF || quota == -1) | |
7774 | return RUNTIME_INF; | |
7775 | ||
7776 | return to_ratio(period, quota); | |
7777 | } | |
7778 | ||
7779 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7780 | { | |
7781 | struct cfs_schedulable_data *d = data; | |
029632fb | 7782 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7783 | s64 quota = 0, parent_quota = -1; |
7784 | ||
7785 | if (!tg->parent) { | |
7786 | quota = RUNTIME_INF; | |
7787 | } else { | |
029632fb | 7788 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7789 | |
7790 | quota = normalize_cfs_quota(tg, d); | |
7791 | parent_quota = parent_b->hierarchal_quota; | |
7792 | ||
7793 | /* | |
7794 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7795 | * limit is set | |
7796 | */ | |
7797 | if (quota == RUNTIME_INF) | |
7798 | quota = parent_quota; | |
7799 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7800 | return -EINVAL; | |
7801 | } | |
7802 | cfs_b->hierarchal_quota = quota; | |
7803 | ||
7804 | return 0; | |
7805 | } | |
7806 | ||
7807 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7808 | { | |
8277434e | 7809 | int ret; |
a790de99 PT |
7810 | struct cfs_schedulable_data data = { |
7811 | .tg = tg, | |
7812 | .period = period, | |
7813 | .quota = quota, | |
7814 | }; | |
7815 | ||
7816 | if (quota != RUNTIME_INF) { | |
7817 | do_div(data.period, NSEC_PER_USEC); | |
7818 | do_div(data.quota, NSEC_PER_USEC); | |
7819 | } | |
7820 | ||
8277434e PT |
7821 | rcu_read_lock(); |
7822 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7823 | rcu_read_unlock(); | |
7824 | ||
7825 | return ret; | |
a790de99 | 7826 | } |
e8da1b18 NR |
7827 | |
7828 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7829 | struct cgroup_map_cb *cb) | |
7830 | { | |
7831 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7832 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7833 | |
7834 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7835 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7836 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7837 | ||
7838 | return 0; | |
7839 | } | |
ab84d31e | 7840 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7841 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7842 | |
052f1dc7 | 7843 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7844 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7845 | s64 val) |
6f505b16 | 7846 | { |
06ecb27c | 7847 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7848 | } |
7849 | ||
06ecb27c | 7850 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7851 | { |
06ecb27c | 7852 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7853 | } |
d0b27fa7 PZ |
7854 | |
7855 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7856 | u64 rt_period_us) | |
7857 | { | |
7858 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7859 | } | |
7860 | ||
7861 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7862 | { | |
7863 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7864 | } | |
6d6bc0ad | 7865 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7866 | |
fe5c7cc2 | 7867 | static struct cftype cpu_files[] = { |
052f1dc7 | 7868 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7869 | { |
7870 | .name = "shares", | |
f4c753b7 PM |
7871 | .read_u64 = cpu_shares_read_u64, |
7872 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7873 | }, |
052f1dc7 | 7874 | #endif |
ab84d31e PT |
7875 | #ifdef CONFIG_CFS_BANDWIDTH |
7876 | { | |
7877 | .name = "cfs_quota_us", | |
7878 | .read_s64 = cpu_cfs_quota_read_s64, | |
7879 | .write_s64 = cpu_cfs_quota_write_s64, | |
7880 | }, | |
7881 | { | |
7882 | .name = "cfs_period_us", | |
7883 | .read_u64 = cpu_cfs_period_read_u64, | |
7884 | .write_u64 = cpu_cfs_period_write_u64, | |
7885 | }, | |
e8da1b18 NR |
7886 | { |
7887 | .name = "stat", | |
7888 | .read_map = cpu_stats_show, | |
7889 | }, | |
ab84d31e | 7890 | #endif |
052f1dc7 | 7891 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7892 | { |
9f0c1e56 | 7893 | .name = "rt_runtime_us", |
06ecb27c PM |
7894 | .read_s64 = cpu_rt_runtime_read, |
7895 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7896 | }, |
d0b27fa7 PZ |
7897 | { |
7898 | .name = "rt_period_us", | |
f4c753b7 PM |
7899 | .read_u64 = cpu_rt_period_read_uint, |
7900 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7901 | }, |
052f1dc7 | 7902 | #endif |
68318b8e SV |
7903 | }; |
7904 | ||
7905 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7906 | { | |
fe5c7cc2 | 7907 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7908 | } |
7909 | ||
7910 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7911 | .name = "cpu", |
7912 | .create = cpu_cgroup_create, | |
7913 | .destroy = cpu_cgroup_destroy, | |
bb9d97b6 TH |
7914 | .can_attach = cpu_cgroup_can_attach, |
7915 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7916 | .exit = cpu_cgroup_exit, |
38605cae IM |
7917 | .populate = cpu_cgroup_populate, |
7918 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7919 | .early_init = 1, |
7920 | }; | |
7921 | ||
052f1dc7 | 7922 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7923 | |
7924 | #ifdef CONFIG_CGROUP_CPUACCT | |
7925 | ||
7926 | /* | |
7927 | * CPU accounting code for task groups. | |
7928 | * | |
7929 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7930 | * (balbir@in.ibm.com). | |
7931 | */ | |
7932 | ||
d842de87 SV |
7933 | /* create a new cpu accounting group */ |
7934 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 7935 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7936 | { |
54c707e9 | 7937 | struct cpuacct *ca; |
d842de87 | 7938 | |
54c707e9 GC |
7939 | if (!cgrp->parent) |
7940 | return &root_cpuacct.css; | |
7941 | ||
7942 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 7943 | if (!ca) |
ef12fefa | 7944 | goto out; |
d842de87 SV |
7945 | |
7946 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
7947 | if (!ca->cpuusage) |
7948 | goto out_free_ca; | |
7949 | ||
54c707e9 GC |
7950 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
7951 | if (!ca->cpustat) | |
7952 | goto out_free_cpuusage; | |
934352f2 | 7953 | |
d842de87 | 7954 | return &ca->css; |
ef12fefa | 7955 | |
54c707e9 | 7956 | out_free_cpuusage: |
ef12fefa BR |
7957 | free_percpu(ca->cpuusage); |
7958 | out_free_ca: | |
7959 | kfree(ca); | |
7960 | out: | |
7961 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
7962 | } |
7963 | ||
7964 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 7965 | static void |
32cd756a | 7966 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7967 | { |
32cd756a | 7968 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 7969 | |
54c707e9 | 7970 | free_percpu(ca->cpustat); |
d842de87 SV |
7971 | free_percpu(ca->cpuusage); |
7972 | kfree(ca); | |
7973 | } | |
7974 | ||
720f5498 KC |
7975 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
7976 | { | |
b36128c8 | 7977 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7978 | u64 data; |
7979 | ||
7980 | #ifndef CONFIG_64BIT | |
7981 | /* | |
7982 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
7983 | */ | |
05fa785c | 7984 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 7985 | data = *cpuusage; |
05fa785c | 7986 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
7987 | #else |
7988 | data = *cpuusage; | |
7989 | #endif | |
7990 | ||
7991 | return data; | |
7992 | } | |
7993 | ||
7994 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
7995 | { | |
b36128c8 | 7996 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7997 | |
7998 | #ifndef CONFIG_64BIT | |
7999 | /* | |
8000 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8001 | */ | |
05fa785c | 8002 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8003 | *cpuusage = val; |
05fa785c | 8004 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8005 | #else |
8006 | *cpuusage = val; | |
8007 | #endif | |
8008 | } | |
8009 | ||
d842de87 | 8010 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8011 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8012 | { |
32cd756a | 8013 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8014 | u64 totalcpuusage = 0; |
8015 | int i; | |
8016 | ||
720f5498 KC |
8017 | for_each_present_cpu(i) |
8018 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8019 | |
8020 | return totalcpuusage; | |
8021 | } | |
8022 | ||
0297b803 DG |
8023 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8024 | u64 reset) | |
8025 | { | |
8026 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8027 | int err = 0; | |
8028 | int i; | |
8029 | ||
8030 | if (reset) { | |
8031 | err = -EINVAL; | |
8032 | goto out; | |
8033 | } | |
8034 | ||
720f5498 KC |
8035 | for_each_present_cpu(i) |
8036 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8037 | |
0297b803 DG |
8038 | out: |
8039 | return err; | |
8040 | } | |
8041 | ||
e9515c3c KC |
8042 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8043 | struct seq_file *m) | |
8044 | { | |
8045 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8046 | u64 percpu; | |
8047 | int i; | |
8048 | ||
8049 | for_each_present_cpu(i) { | |
8050 | percpu = cpuacct_cpuusage_read(ca, i); | |
8051 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8052 | } | |
8053 | seq_printf(m, "\n"); | |
8054 | return 0; | |
8055 | } | |
8056 | ||
ef12fefa BR |
8057 | static const char *cpuacct_stat_desc[] = { |
8058 | [CPUACCT_STAT_USER] = "user", | |
8059 | [CPUACCT_STAT_SYSTEM] = "system", | |
8060 | }; | |
8061 | ||
8062 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8063 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8064 | { |
8065 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8066 | int cpu; |
8067 | s64 val = 0; | |
ef12fefa | 8068 | |
54c707e9 GC |
8069 | for_each_online_cpu(cpu) { |
8070 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8071 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8072 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8073 | } |
54c707e9 GC |
8074 | val = cputime64_to_clock_t(val); |
8075 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8076 | |
54c707e9 GC |
8077 | val = 0; |
8078 | for_each_online_cpu(cpu) { | |
8079 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8080 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8081 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8082 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8083 | } |
54c707e9 GC |
8084 | |
8085 | val = cputime64_to_clock_t(val); | |
8086 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8087 | ||
ef12fefa BR |
8088 | return 0; |
8089 | } | |
8090 | ||
d842de87 SV |
8091 | static struct cftype files[] = { |
8092 | { | |
8093 | .name = "usage", | |
f4c753b7 PM |
8094 | .read_u64 = cpuusage_read, |
8095 | .write_u64 = cpuusage_write, | |
d842de87 | 8096 | }, |
e9515c3c KC |
8097 | { |
8098 | .name = "usage_percpu", | |
8099 | .read_seq_string = cpuacct_percpu_seq_read, | |
8100 | }, | |
ef12fefa BR |
8101 | { |
8102 | .name = "stat", | |
8103 | .read_map = cpuacct_stats_show, | |
8104 | }, | |
d842de87 SV |
8105 | }; |
8106 | ||
32cd756a | 8107 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8108 | { |
32cd756a | 8109 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8110 | } |
8111 | ||
8112 | /* | |
8113 | * charge this task's execution time to its accounting group. | |
8114 | * | |
8115 | * called with rq->lock held. | |
8116 | */ | |
029632fb | 8117 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8118 | { |
8119 | struct cpuacct *ca; | |
934352f2 | 8120 | int cpu; |
d842de87 | 8121 | |
c40c6f85 | 8122 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8123 | return; |
8124 | ||
934352f2 | 8125 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8126 | |
8127 | rcu_read_lock(); | |
8128 | ||
d842de87 | 8129 | ca = task_ca(tsk); |
d842de87 | 8130 | |
44252e42 | 8131 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8132 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8133 | *cpuusage += cputime; |
8134 | } | |
a18b83b7 BR |
8135 | |
8136 | rcu_read_unlock(); | |
d842de87 SV |
8137 | } |
8138 | ||
8139 | struct cgroup_subsys cpuacct_subsys = { | |
8140 | .name = "cpuacct", | |
8141 | .create = cpuacct_create, | |
8142 | .destroy = cpuacct_destroy, | |
8143 | .populate = cpuacct_populate, | |
8144 | .subsys_id = cpuacct_subsys_id, | |
8145 | }; | |
8146 | #endif /* CONFIG_CGROUP_CPUACCT */ |