]>
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> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
1da177e4 | 76 | |
96f951ed | 77 | #include <asm/switch_to.h> |
5517d86b | 78 | #include <asm/tlb.h> |
838225b4 | 79 | #include <asm/irq_regs.h> |
db7e527d | 80 | #include <asm/mutex.h> |
e6e6685a GC |
81 | #ifdef CONFIG_PARAVIRT |
82 | #include <asm/paravirt.h> | |
83 | #endif | |
1da177e4 | 84 | |
029632fb | 85 | #include "sched.h" |
ea138446 | 86 | #include "../workqueue_internal.h" |
29d5e047 | 87 | #include "../smpboot.h" |
6e0534f2 | 88 | |
a8d154b0 | 89 | #define CREATE_TRACE_POINTS |
ad8d75ff | 90 | #include <trace/events/sched.h> |
a8d154b0 | 91 | |
029632fb | 92 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 93 | { |
58088ad0 PT |
94 | unsigned long delta; |
95 | ktime_t soft, hard, now; | |
d0b27fa7 | 96 | |
58088ad0 PT |
97 | for (;;) { |
98 | if (hrtimer_active(period_timer)) | |
99 | break; | |
100 | ||
101 | now = hrtimer_cb_get_time(period_timer); | |
102 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 103 | |
58088ad0 PT |
104 | soft = hrtimer_get_softexpires(period_timer); |
105 | hard = hrtimer_get_expires(period_timer); | |
106 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
107 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
108 | HRTIMER_MODE_ABS_PINNED, 0); | |
109 | } | |
110 | } | |
111 | ||
029632fb PZ |
112 | DEFINE_MUTEX(sched_domains_mutex); |
113 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 114 | |
fe44d621 | 115 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 116 | |
029632fb | 117 | void update_rq_clock(struct rq *rq) |
3e51f33f | 118 | { |
fe44d621 | 119 | s64 delta; |
305e6835 | 120 | |
61eadef6 | 121 | if (rq->skip_clock_update > 0) |
f26f9aff | 122 | return; |
aa483808 | 123 | |
fe44d621 PZ |
124 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
125 | rq->clock += delta; | |
126 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
127 | } |
128 | ||
bf5c91ba IM |
129 | /* |
130 | * Debugging: various feature bits | |
131 | */ | |
f00b45c1 | 132 | |
f00b45c1 PZ |
133 | #define SCHED_FEAT(name, enabled) \ |
134 | (1UL << __SCHED_FEAT_##name) * enabled | | |
135 | ||
bf5c91ba | 136 | const_debug unsigned int sysctl_sched_features = |
391e43da | 137 | #include "features.h" |
f00b45c1 PZ |
138 | 0; |
139 | ||
140 | #undef SCHED_FEAT | |
141 | ||
142 | #ifdef CONFIG_SCHED_DEBUG | |
143 | #define SCHED_FEAT(name, enabled) \ | |
144 | #name , | |
145 | ||
1292531f | 146 | static const char * const sched_feat_names[] = { |
391e43da | 147 | #include "features.h" |
f00b45c1 PZ |
148 | }; |
149 | ||
150 | #undef SCHED_FEAT | |
151 | ||
34f3a814 | 152 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 153 | { |
f00b45c1 PZ |
154 | int i; |
155 | ||
f8b6d1cc | 156 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
157 | if (!(sysctl_sched_features & (1UL << i))) |
158 | seq_puts(m, "NO_"); | |
159 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 160 | } |
34f3a814 | 161 | seq_puts(m, "\n"); |
f00b45c1 | 162 | |
34f3a814 | 163 | return 0; |
f00b45c1 PZ |
164 | } |
165 | ||
f8b6d1cc PZ |
166 | #ifdef HAVE_JUMP_LABEL |
167 | ||
c5905afb IM |
168 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
169 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
170 | |
171 | #define SCHED_FEAT(name, enabled) \ | |
172 | jump_label_key__##enabled , | |
173 | ||
c5905afb | 174 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
175 | #include "features.h" |
176 | }; | |
177 | ||
178 | #undef SCHED_FEAT | |
179 | ||
180 | static void sched_feat_disable(int i) | |
181 | { | |
c5905afb IM |
182 | if (static_key_enabled(&sched_feat_keys[i])) |
183 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
184 | } |
185 | ||
186 | static void sched_feat_enable(int i) | |
187 | { | |
c5905afb IM |
188 | if (!static_key_enabled(&sched_feat_keys[i])) |
189 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
190 | } |
191 | #else | |
192 | static void sched_feat_disable(int i) { }; | |
193 | static void sched_feat_enable(int i) { }; | |
194 | #endif /* HAVE_JUMP_LABEL */ | |
195 | ||
1a687c2e | 196 | static int sched_feat_set(char *cmp) |
f00b45c1 | 197 | { |
f00b45c1 | 198 | int i; |
1a687c2e | 199 | int neg = 0; |
f00b45c1 | 200 | |
524429c3 | 201 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
202 | neg = 1; |
203 | cmp += 3; | |
204 | } | |
205 | ||
f8b6d1cc | 206 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 207 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 208 | if (neg) { |
f00b45c1 | 209 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
210 | sched_feat_disable(i); |
211 | } else { | |
f00b45c1 | 212 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
213 | sched_feat_enable(i); |
214 | } | |
f00b45c1 PZ |
215 | break; |
216 | } | |
217 | } | |
218 | ||
1a687c2e MG |
219 | return i; |
220 | } | |
221 | ||
222 | static ssize_t | |
223 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
224 | size_t cnt, loff_t *ppos) | |
225 | { | |
226 | char buf[64]; | |
227 | char *cmp; | |
228 | int i; | |
229 | ||
230 | if (cnt > 63) | |
231 | cnt = 63; | |
232 | ||
233 | if (copy_from_user(&buf, ubuf, cnt)) | |
234 | return -EFAULT; | |
235 | ||
236 | buf[cnt] = 0; | |
237 | cmp = strstrip(buf); | |
238 | ||
239 | i = sched_feat_set(cmp); | |
f8b6d1cc | 240 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
241 | return -EINVAL; |
242 | ||
42994724 | 243 | *ppos += cnt; |
f00b45c1 PZ |
244 | |
245 | return cnt; | |
246 | } | |
247 | ||
34f3a814 LZ |
248 | static int sched_feat_open(struct inode *inode, struct file *filp) |
249 | { | |
250 | return single_open(filp, sched_feat_show, NULL); | |
251 | } | |
252 | ||
828c0950 | 253 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
254 | .open = sched_feat_open, |
255 | .write = sched_feat_write, | |
256 | .read = seq_read, | |
257 | .llseek = seq_lseek, | |
258 | .release = single_release, | |
f00b45c1 PZ |
259 | }; |
260 | ||
261 | static __init int sched_init_debug(void) | |
262 | { | |
f00b45c1 PZ |
263 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
264 | &sched_feat_fops); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | late_initcall(sched_init_debug); | |
f8b6d1cc | 269 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 270 | |
b82d9fdd PZ |
271 | /* |
272 | * Number of tasks to iterate in a single balance run. | |
273 | * Limited because this is done with IRQs disabled. | |
274 | */ | |
275 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
276 | ||
e9e9250b PZ |
277 | /* |
278 | * period over which we average the RT time consumption, measured | |
279 | * in ms. | |
280 | * | |
281 | * default: 1s | |
282 | */ | |
283 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
284 | ||
fa85ae24 | 285 | /* |
9f0c1e56 | 286 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
287 | * default: 1s |
288 | */ | |
9f0c1e56 | 289 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 290 | |
029632fb | 291 | __read_mostly int scheduler_running; |
6892b75e | 292 | |
9f0c1e56 PZ |
293 | /* |
294 | * part of the period that we allow rt tasks to run in us. | |
295 | * default: 0.95s | |
296 | */ | |
297 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 298 | |
fa85ae24 | 299 | |
1da177e4 | 300 | |
0970d299 | 301 | /* |
0122ec5b | 302 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 303 | */ |
70b97a7f | 304 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
305 | __acquires(rq->lock) |
306 | { | |
0970d299 PZ |
307 | struct rq *rq; |
308 | ||
0122ec5b PZ |
309 | lockdep_assert_held(&p->pi_lock); |
310 | ||
3a5c359a | 311 | for (;;) { |
0970d299 | 312 | rq = task_rq(p); |
05fa785c | 313 | raw_spin_lock(&rq->lock); |
65cc8e48 | 314 | if (likely(rq == task_rq(p))) |
3a5c359a | 315 | return rq; |
05fa785c | 316 | raw_spin_unlock(&rq->lock); |
b29739f9 | 317 | } |
b29739f9 IM |
318 | } |
319 | ||
1da177e4 | 320 | /* |
0122ec5b | 321 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 322 | */ |
70b97a7f | 323 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 324 | __acquires(p->pi_lock) |
1da177e4 LT |
325 | __acquires(rq->lock) |
326 | { | |
70b97a7f | 327 | struct rq *rq; |
1da177e4 | 328 | |
3a5c359a | 329 | for (;;) { |
0122ec5b | 330 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 331 | rq = task_rq(p); |
05fa785c | 332 | raw_spin_lock(&rq->lock); |
65cc8e48 | 333 | if (likely(rq == task_rq(p))) |
3a5c359a | 334 | return rq; |
0122ec5b PZ |
335 | raw_spin_unlock(&rq->lock); |
336 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 337 | } |
1da177e4 LT |
338 | } |
339 | ||
a9957449 | 340 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
341 | __releases(rq->lock) |
342 | { | |
05fa785c | 343 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
344 | } |
345 | ||
0122ec5b PZ |
346 | static inline void |
347 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 348 | __releases(rq->lock) |
0122ec5b | 349 | __releases(p->pi_lock) |
1da177e4 | 350 | { |
0122ec5b PZ |
351 | raw_spin_unlock(&rq->lock); |
352 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
353 | } |
354 | ||
1da177e4 | 355 | /* |
cc2a73b5 | 356 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 357 | */ |
a9957449 | 358 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
359 | __acquires(rq->lock) |
360 | { | |
70b97a7f | 361 | struct rq *rq; |
1da177e4 LT |
362 | |
363 | local_irq_disable(); | |
364 | rq = this_rq(); | |
05fa785c | 365 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
366 | |
367 | return rq; | |
368 | } | |
369 | ||
8f4d37ec PZ |
370 | #ifdef CONFIG_SCHED_HRTICK |
371 | /* | |
372 | * Use HR-timers to deliver accurate preemption points. | |
373 | * | |
374 | * Its all a bit involved since we cannot program an hrt while holding the | |
375 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
376 | * reschedule event. | |
377 | * | |
378 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
379 | * rq->lock. | |
380 | */ | |
8f4d37ec | 381 | |
8f4d37ec PZ |
382 | static void hrtick_clear(struct rq *rq) |
383 | { | |
384 | if (hrtimer_active(&rq->hrtick_timer)) | |
385 | hrtimer_cancel(&rq->hrtick_timer); | |
386 | } | |
387 | ||
8f4d37ec PZ |
388 | /* |
389 | * High-resolution timer tick. | |
390 | * Runs from hardirq context with interrupts disabled. | |
391 | */ | |
392 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
393 | { | |
394 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
395 | ||
396 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
397 | ||
05fa785c | 398 | raw_spin_lock(&rq->lock); |
3e51f33f | 399 | update_rq_clock(rq); |
8f4d37ec | 400 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 401 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
402 | |
403 | return HRTIMER_NORESTART; | |
404 | } | |
405 | ||
95e904c7 | 406 | #ifdef CONFIG_SMP |
31656519 PZ |
407 | /* |
408 | * called from hardirq (IPI) context | |
409 | */ | |
410 | static void __hrtick_start(void *arg) | |
b328ca18 | 411 | { |
31656519 | 412 | struct rq *rq = arg; |
b328ca18 | 413 | |
05fa785c | 414 | raw_spin_lock(&rq->lock); |
31656519 PZ |
415 | hrtimer_restart(&rq->hrtick_timer); |
416 | rq->hrtick_csd_pending = 0; | |
05fa785c | 417 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
418 | } |
419 | ||
31656519 PZ |
420 | /* |
421 | * Called to set the hrtick timer state. | |
422 | * | |
423 | * called with rq->lock held and irqs disabled | |
424 | */ | |
029632fb | 425 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 426 | { |
31656519 PZ |
427 | struct hrtimer *timer = &rq->hrtick_timer; |
428 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 429 | |
cc584b21 | 430 | hrtimer_set_expires(timer, time); |
31656519 PZ |
431 | |
432 | if (rq == this_rq()) { | |
433 | hrtimer_restart(timer); | |
434 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 435 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
436 | rq->hrtick_csd_pending = 1; |
437 | } | |
b328ca18 PZ |
438 | } |
439 | ||
440 | static int | |
441 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
442 | { | |
443 | int cpu = (int)(long)hcpu; | |
444 | ||
445 | switch (action) { | |
446 | case CPU_UP_CANCELED: | |
447 | case CPU_UP_CANCELED_FROZEN: | |
448 | case CPU_DOWN_PREPARE: | |
449 | case CPU_DOWN_PREPARE_FROZEN: | |
450 | case CPU_DEAD: | |
451 | case CPU_DEAD_FROZEN: | |
31656519 | 452 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
453 | return NOTIFY_OK; |
454 | } | |
455 | ||
456 | return NOTIFY_DONE; | |
457 | } | |
458 | ||
fa748203 | 459 | static __init void init_hrtick(void) |
b328ca18 PZ |
460 | { |
461 | hotcpu_notifier(hotplug_hrtick, 0); | |
462 | } | |
31656519 PZ |
463 | #else |
464 | /* | |
465 | * Called to set the hrtick timer state. | |
466 | * | |
467 | * called with rq->lock held and irqs disabled | |
468 | */ | |
029632fb | 469 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 470 | { |
7f1e2ca9 | 471 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 472 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 473 | } |
b328ca18 | 474 | |
006c75f1 | 475 | static inline void init_hrtick(void) |
8f4d37ec | 476 | { |
8f4d37ec | 477 | } |
31656519 | 478 | #endif /* CONFIG_SMP */ |
8f4d37ec | 479 | |
31656519 | 480 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 481 | { |
31656519 PZ |
482 | #ifdef CONFIG_SMP |
483 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 484 | |
31656519 PZ |
485 | rq->hrtick_csd.flags = 0; |
486 | rq->hrtick_csd.func = __hrtick_start; | |
487 | rq->hrtick_csd.info = rq; | |
488 | #endif | |
8f4d37ec | 489 | |
31656519 PZ |
490 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
491 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 492 | } |
006c75f1 | 493 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
494 | static inline void hrtick_clear(struct rq *rq) |
495 | { | |
496 | } | |
497 | ||
8f4d37ec PZ |
498 | static inline void init_rq_hrtick(struct rq *rq) |
499 | { | |
500 | } | |
501 | ||
b328ca18 PZ |
502 | static inline void init_hrtick(void) |
503 | { | |
504 | } | |
006c75f1 | 505 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 506 | |
c24d20db IM |
507 | /* |
508 | * resched_task - mark a task 'to be rescheduled now'. | |
509 | * | |
510 | * On UP this means the setting of the need_resched flag, on SMP it | |
511 | * might also involve a cross-CPU call to trigger the scheduler on | |
512 | * the target CPU. | |
513 | */ | |
514 | #ifdef CONFIG_SMP | |
515 | ||
516 | #ifndef tsk_is_polling | |
16a80163 | 517 | #define tsk_is_polling(t) 0 |
c24d20db IM |
518 | #endif |
519 | ||
029632fb | 520 | void resched_task(struct task_struct *p) |
c24d20db IM |
521 | { |
522 | int cpu; | |
523 | ||
05fa785c | 524 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 525 | |
5ed0cec0 | 526 | if (test_tsk_need_resched(p)) |
c24d20db IM |
527 | return; |
528 | ||
5ed0cec0 | 529 | set_tsk_need_resched(p); |
c24d20db IM |
530 | |
531 | cpu = task_cpu(p); | |
532 | if (cpu == smp_processor_id()) | |
533 | return; | |
534 | ||
535 | /* NEED_RESCHED must be visible before we test polling */ | |
536 | smp_mb(); | |
537 | if (!tsk_is_polling(p)) | |
538 | smp_send_reschedule(cpu); | |
539 | } | |
540 | ||
029632fb | 541 | void resched_cpu(int cpu) |
c24d20db IM |
542 | { |
543 | struct rq *rq = cpu_rq(cpu); | |
544 | unsigned long flags; | |
545 | ||
05fa785c | 546 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
547 | return; |
548 | resched_task(cpu_curr(cpu)); | |
05fa785c | 549 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 550 | } |
06d8308c | 551 | |
3451d024 | 552 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
553 | /* |
554 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
555 | * from an idle cpu. This is good for power-savings. | |
556 | * | |
557 | * We don't do similar optimization for completely idle system, as | |
558 | * selecting an idle cpu will add more delays to the timers than intended | |
559 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
560 | */ | |
561 | int get_nohz_timer_target(void) | |
562 | { | |
563 | int cpu = smp_processor_id(); | |
564 | int i; | |
565 | struct sched_domain *sd; | |
566 | ||
057f3fad | 567 | rcu_read_lock(); |
83cd4fe2 | 568 | for_each_domain(cpu, sd) { |
057f3fad PZ |
569 | for_each_cpu(i, sched_domain_span(sd)) { |
570 | if (!idle_cpu(i)) { | |
571 | cpu = i; | |
572 | goto unlock; | |
573 | } | |
574 | } | |
83cd4fe2 | 575 | } |
057f3fad PZ |
576 | unlock: |
577 | rcu_read_unlock(); | |
83cd4fe2 VP |
578 | return cpu; |
579 | } | |
06d8308c TG |
580 | /* |
581 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
582 | * idle CPU then this timer might expire before the next timer event | |
583 | * which is scheduled to wake up that CPU. In case of a completely | |
584 | * idle system the next event might even be infinite time into the | |
585 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
586 | * leaves the inner idle loop so the newly added timer is taken into | |
587 | * account when the CPU goes back to idle and evaluates the timer | |
588 | * wheel for the next timer event. | |
589 | */ | |
1c20091e | 590 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
591 | { |
592 | struct rq *rq = cpu_rq(cpu); | |
593 | ||
594 | if (cpu == smp_processor_id()) | |
595 | return; | |
596 | ||
597 | /* | |
598 | * This is safe, as this function is called with the timer | |
599 | * wheel base lock of (cpu) held. When the CPU is on the way | |
600 | * to idle and has not yet set rq->curr to idle then it will | |
601 | * be serialized on the timer wheel base lock and take the new | |
602 | * timer into account automatically. | |
603 | */ | |
604 | if (rq->curr != rq->idle) | |
605 | return; | |
45bf76df | 606 | |
45bf76df | 607 | /* |
06d8308c TG |
608 | * We can set TIF_RESCHED on the idle task of the other CPU |
609 | * lockless. The worst case is that the other CPU runs the | |
610 | * idle task through an additional NOOP schedule() | |
45bf76df | 611 | */ |
5ed0cec0 | 612 | set_tsk_need_resched(rq->idle); |
45bf76df | 613 | |
06d8308c TG |
614 | /* NEED_RESCHED must be visible before we test polling */ |
615 | smp_mb(); | |
616 | if (!tsk_is_polling(rq->idle)) | |
617 | smp_send_reschedule(cpu); | |
45bf76df IM |
618 | } |
619 | ||
1c20091e FW |
620 | static bool wake_up_extended_nohz_cpu(int cpu) |
621 | { | |
622 | if (tick_nohz_extended_cpu(cpu)) { | |
623 | if (cpu != smp_processor_id() || | |
624 | tick_nohz_tick_stopped()) | |
625 | smp_send_reschedule(cpu); | |
626 | return true; | |
627 | } | |
628 | ||
629 | return false; | |
630 | } | |
631 | ||
632 | void wake_up_nohz_cpu(int cpu) | |
633 | { | |
634 | if (!wake_up_extended_nohz_cpu(cpu)) | |
635 | wake_up_idle_cpu(cpu); | |
636 | } | |
637 | ||
ca38062e | 638 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 639 | { |
1c792db7 SS |
640 | int cpu = smp_processor_id(); |
641 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
642 | } |
643 | ||
3451d024 | 644 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 645 | |
ca38062e | 646 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 647 | { |
ca38062e | 648 | return false; |
2069dd75 PZ |
649 | } |
650 | ||
3451d024 | 651 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 652 | |
029632fb | 653 | void sched_avg_update(struct rq *rq) |
18d95a28 | 654 | { |
e9e9250b PZ |
655 | s64 period = sched_avg_period(); |
656 | ||
657 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
658 | /* |
659 | * Inline assembly required to prevent the compiler | |
660 | * optimising this loop into a divmod call. | |
661 | * See __iter_div_u64_rem() for another example of this. | |
662 | */ | |
663 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
664 | rq->age_stamp += period; |
665 | rq->rt_avg /= 2; | |
666 | } | |
18d95a28 PZ |
667 | } |
668 | ||
6d6bc0ad | 669 | #else /* !CONFIG_SMP */ |
029632fb | 670 | void resched_task(struct task_struct *p) |
18d95a28 | 671 | { |
05fa785c | 672 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 673 | set_tsk_need_resched(p); |
18d95a28 | 674 | } |
6d6bc0ad | 675 | #endif /* CONFIG_SMP */ |
18d95a28 | 676 | |
a790de99 PT |
677 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
678 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 679 | /* |
8277434e PT |
680 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
681 | * node and @up when leaving it for the final time. | |
682 | * | |
683 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 684 | */ |
029632fb | 685 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 686 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
687 | { |
688 | struct task_group *parent, *child; | |
eb755805 | 689 | int ret; |
c09595f6 | 690 | |
8277434e PT |
691 | parent = from; |
692 | ||
c09595f6 | 693 | down: |
eb755805 PZ |
694 | ret = (*down)(parent, data); |
695 | if (ret) | |
8277434e | 696 | goto out; |
c09595f6 PZ |
697 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
698 | parent = child; | |
699 | goto down; | |
700 | ||
701 | up: | |
702 | continue; | |
703 | } | |
eb755805 | 704 | ret = (*up)(parent, data); |
8277434e PT |
705 | if (ret || parent == from) |
706 | goto out; | |
c09595f6 PZ |
707 | |
708 | child = parent; | |
709 | parent = parent->parent; | |
710 | if (parent) | |
711 | goto up; | |
8277434e | 712 | out: |
eb755805 | 713 | return ret; |
c09595f6 PZ |
714 | } |
715 | ||
029632fb | 716 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 717 | { |
e2b245f8 | 718 | return 0; |
eb755805 | 719 | } |
18d95a28 PZ |
720 | #endif |
721 | ||
45bf76df IM |
722 | static void set_load_weight(struct task_struct *p) |
723 | { | |
f05998d4 NR |
724 | int prio = p->static_prio - MAX_RT_PRIO; |
725 | struct load_weight *load = &p->se.load; | |
726 | ||
dd41f596 IM |
727 | /* |
728 | * SCHED_IDLE tasks get minimal weight: | |
729 | */ | |
730 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 731 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 732 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
733 | return; |
734 | } | |
71f8bd46 | 735 | |
c8b28116 | 736 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 737 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
738 | } |
739 | ||
371fd7e7 | 740 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 741 | { |
a64692a3 | 742 | update_rq_clock(rq); |
dd41f596 | 743 | sched_info_queued(p); |
371fd7e7 | 744 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
745 | } |
746 | ||
371fd7e7 | 747 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 748 | { |
a64692a3 | 749 | update_rq_clock(rq); |
46ac22ba | 750 | sched_info_dequeued(p); |
371fd7e7 | 751 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
752 | } |
753 | ||
029632fb | 754 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
755 | { |
756 | if (task_contributes_to_load(p)) | |
757 | rq->nr_uninterruptible--; | |
758 | ||
371fd7e7 | 759 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
760 | } |
761 | ||
029632fb | 762 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
763 | { |
764 | if (task_contributes_to_load(p)) | |
765 | rq->nr_uninterruptible++; | |
766 | ||
371fd7e7 | 767 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
768 | } |
769 | ||
fe44d621 | 770 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 771 | { |
095c0aa8 GC |
772 | /* |
773 | * In theory, the compile should just see 0 here, and optimize out the call | |
774 | * to sched_rt_avg_update. But I don't trust it... | |
775 | */ | |
776 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
777 | s64 steal = 0, irq_delta = 0; | |
778 | #endif | |
779 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 780 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
781 | |
782 | /* | |
783 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
784 | * this case when a previous update_rq_clock() happened inside a | |
785 | * {soft,}irq region. | |
786 | * | |
787 | * When this happens, we stop ->clock_task and only update the | |
788 | * prev_irq_time stamp to account for the part that fit, so that a next | |
789 | * update will consume the rest. This ensures ->clock_task is | |
790 | * monotonic. | |
791 | * | |
792 | * It does however cause some slight miss-attribution of {soft,}irq | |
793 | * time, a more accurate solution would be to update the irq_time using | |
794 | * the current rq->clock timestamp, except that would require using | |
795 | * atomic ops. | |
796 | */ | |
797 | if (irq_delta > delta) | |
798 | irq_delta = delta; | |
799 | ||
800 | rq->prev_irq_time += irq_delta; | |
801 | delta -= irq_delta; | |
095c0aa8 GC |
802 | #endif |
803 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 804 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
805 | u64 st; |
806 | ||
807 | steal = paravirt_steal_clock(cpu_of(rq)); | |
808 | steal -= rq->prev_steal_time_rq; | |
809 | ||
810 | if (unlikely(steal > delta)) | |
811 | steal = delta; | |
812 | ||
813 | st = steal_ticks(steal); | |
814 | steal = st * TICK_NSEC; | |
815 | ||
816 | rq->prev_steal_time_rq += steal; | |
817 | ||
818 | delta -= steal; | |
819 | } | |
820 | #endif | |
821 | ||
fe44d621 PZ |
822 | rq->clock_task += delta; |
823 | ||
095c0aa8 GC |
824 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
825 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
826 | sched_rt_avg_update(rq, irq_delta + steal); | |
827 | #endif | |
aa483808 VP |
828 | } |
829 | ||
34f971f6 PZ |
830 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
831 | { | |
832 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
833 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
834 | ||
835 | if (stop) { | |
836 | /* | |
837 | * Make it appear like a SCHED_FIFO task, its something | |
838 | * userspace knows about and won't get confused about. | |
839 | * | |
840 | * Also, it will make PI more or less work without too | |
841 | * much confusion -- but then, stop work should not | |
842 | * rely on PI working anyway. | |
843 | */ | |
844 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
845 | ||
846 | stop->sched_class = &stop_sched_class; | |
847 | } | |
848 | ||
849 | cpu_rq(cpu)->stop = stop; | |
850 | ||
851 | if (old_stop) { | |
852 | /* | |
853 | * Reset it back to a normal scheduling class so that | |
854 | * it can die in pieces. | |
855 | */ | |
856 | old_stop->sched_class = &rt_sched_class; | |
857 | } | |
858 | } | |
859 | ||
14531189 | 860 | /* |
dd41f596 | 861 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 862 | */ |
14531189 IM |
863 | static inline int __normal_prio(struct task_struct *p) |
864 | { | |
dd41f596 | 865 | return p->static_prio; |
14531189 IM |
866 | } |
867 | ||
b29739f9 IM |
868 | /* |
869 | * Calculate the expected normal priority: i.e. priority | |
870 | * without taking RT-inheritance into account. Might be | |
871 | * boosted by interactivity modifiers. Changes upon fork, | |
872 | * setprio syscalls, and whenever the interactivity | |
873 | * estimator recalculates. | |
874 | */ | |
36c8b586 | 875 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
876 | { |
877 | int prio; | |
878 | ||
e05606d3 | 879 | if (task_has_rt_policy(p)) |
b29739f9 IM |
880 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
881 | else | |
882 | prio = __normal_prio(p); | |
883 | return prio; | |
884 | } | |
885 | ||
886 | /* | |
887 | * Calculate the current priority, i.e. the priority | |
888 | * taken into account by the scheduler. This value might | |
889 | * be boosted by RT tasks, or might be boosted by | |
890 | * interactivity modifiers. Will be RT if the task got | |
891 | * RT-boosted. If not then it returns p->normal_prio. | |
892 | */ | |
36c8b586 | 893 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
894 | { |
895 | p->normal_prio = normal_prio(p); | |
896 | /* | |
897 | * If we are RT tasks or we were boosted to RT priority, | |
898 | * keep the priority unchanged. Otherwise, update priority | |
899 | * to the normal priority: | |
900 | */ | |
901 | if (!rt_prio(p->prio)) | |
902 | return p->normal_prio; | |
903 | return p->prio; | |
904 | } | |
905 | ||
1da177e4 LT |
906 | /** |
907 | * task_curr - is this task currently executing on a CPU? | |
908 | * @p: the task in question. | |
909 | */ | |
36c8b586 | 910 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
911 | { |
912 | return cpu_curr(task_cpu(p)) == p; | |
913 | } | |
914 | ||
cb469845 SR |
915 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
916 | const struct sched_class *prev_class, | |
da7a735e | 917 | int oldprio) |
cb469845 SR |
918 | { |
919 | if (prev_class != p->sched_class) { | |
920 | if (prev_class->switched_from) | |
da7a735e PZ |
921 | prev_class->switched_from(rq, p); |
922 | p->sched_class->switched_to(rq, p); | |
923 | } else if (oldprio != p->prio) | |
924 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
925 | } |
926 | ||
029632fb | 927 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
928 | { |
929 | const struct sched_class *class; | |
930 | ||
931 | if (p->sched_class == rq->curr->sched_class) { | |
932 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
933 | } else { | |
934 | for_each_class(class) { | |
935 | if (class == rq->curr->sched_class) | |
936 | break; | |
937 | if (class == p->sched_class) { | |
938 | resched_task(rq->curr); | |
939 | break; | |
940 | } | |
941 | } | |
942 | } | |
943 | ||
944 | /* | |
945 | * A queue event has occurred, and we're going to schedule. In | |
946 | * this case, we can save a useless back to back clock update. | |
947 | */ | |
fd2f4419 | 948 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
949 | rq->skip_clock_update = 1; |
950 | } | |
951 | ||
582b336e MT |
952 | static ATOMIC_NOTIFIER_HEAD(task_migration_notifier); |
953 | ||
954 | void register_task_migration_notifier(struct notifier_block *n) | |
955 | { | |
956 | atomic_notifier_chain_register(&task_migration_notifier, n); | |
957 | } | |
958 | ||
1da177e4 | 959 | #ifdef CONFIG_SMP |
dd41f596 | 960 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 961 | { |
e2912009 PZ |
962 | #ifdef CONFIG_SCHED_DEBUG |
963 | /* | |
964 | * We should never call set_task_cpu() on a blocked task, | |
965 | * ttwu() will sort out the placement. | |
966 | */ | |
077614ee PZ |
967 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
968 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
969 | |
970 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
971 | /* |
972 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
973 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
974 | * | |
975 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 976 | * see task_group(). |
6c6c54e1 PZ |
977 | * |
978 | * Furthermore, all task_rq users should acquire both locks, see | |
979 | * task_rq_lock(). | |
980 | */ | |
0122ec5b PZ |
981 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
982 | lockdep_is_held(&task_rq(p)->lock))); | |
983 | #endif | |
e2912009 PZ |
984 | #endif |
985 | ||
de1d7286 | 986 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 987 | |
0c69774e | 988 | if (task_cpu(p) != new_cpu) { |
582b336e MT |
989 | struct task_migration_notifier tmn; |
990 | ||
0a74bef8 PT |
991 | if (p->sched_class->migrate_task_rq) |
992 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 993 | p->se.nr_migrations++; |
a8b0ca17 | 994 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
582b336e MT |
995 | |
996 | tmn.task = p; | |
997 | tmn.from_cpu = task_cpu(p); | |
998 | tmn.to_cpu = new_cpu; | |
999 | ||
1000 | atomic_notifier_call_chain(&task_migration_notifier, 0, &tmn); | |
0c69774e | 1001 | } |
dd41f596 IM |
1002 | |
1003 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1004 | } |
1005 | ||
969c7921 | 1006 | struct migration_arg { |
36c8b586 | 1007 | struct task_struct *task; |
1da177e4 | 1008 | int dest_cpu; |
70b97a7f | 1009 | }; |
1da177e4 | 1010 | |
969c7921 TH |
1011 | static int migration_cpu_stop(void *data); |
1012 | ||
1da177e4 LT |
1013 | /* |
1014 | * wait_task_inactive - wait for a thread to unschedule. | |
1015 | * | |
85ba2d86 RM |
1016 | * If @match_state is nonzero, it's the @p->state value just checked and |
1017 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1018 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1019 | * we return a positive number (its total switch count). If a second call | |
1020 | * a short while later returns the same number, the caller can be sure that | |
1021 | * @p has remained unscheduled the whole time. | |
1022 | * | |
1da177e4 LT |
1023 | * The caller must ensure that the task *will* unschedule sometime soon, |
1024 | * else this function might spin for a *long* time. This function can't | |
1025 | * be called with interrupts off, or it may introduce deadlock with | |
1026 | * smp_call_function() if an IPI is sent by the same process we are | |
1027 | * waiting to become inactive. | |
1028 | */ | |
85ba2d86 | 1029 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1030 | { |
1031 | unsigned long flags; | |
dd41f596 | 1032 | int running, on_rq; |
85ba2d86 | 1033 | unsigned long ncsw; |
70b97a7f | 1034 | struct rq *rq; |
1da177e4 | 1035 | |
3a5c359a AK |
1036 | for (;;) { |
1037 | /* | |
1038 | * We do the initial early heuristics without holding | |
1039 | * any task-queue locks at all. We'll only try to get | |
1040 | * the runqueue lock when things look like they will | |
1041 | * work out! | |
1042 | */ | |
1043 | rq = task_rq(p); | |
fa490cfd | 1044 | |
3a5c359a AK |
1045 | /* |
1046 | * If the task is actively running on another CPU | |
1047 | * still, just relax and busy-wait without holding | |
1048 | * any locks. | |
1049 | * | |
1050 | * NOTE! Since we don't hold any locks, it's not | |
1051 | * even sure that "rq" stays as the right runqueue! | |
1052 | * But we don't care, since "task_running()" will | |
1053 | * return false if the runqueue has changed and p | |
1054 | * is actually now running somewhere else! | |
1055 | */ | |
85ba2d86 RM |
1056 | while (task_running(rq, p)) { |
1057 | if (match_state && unlikely(p->state != match_state)) | |
1058 | return 0; | |
3a5c359a | 1059 | cpu_relax(); |
85ba2d86 | 1060 | } |
fa490cfd | 1061 | |
3a5c359a AK |
1062 | /* |
1063 | * Ok, time to look more closely! We need the rq | |
1064 | * lock now, to be *sure*. If we're wrong, we'll | |
1065 | * just go back and repeat. | |
1066 | */ | |
1067 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1068 | trace_sched_wait_task(p); |
3a5c359a | 1069 | running = task_running(rq, p); |
fd2f4419 | 1070 | on_rq = p->on_rq; |
85ba2d86 | 1071 | ncsw = 0; |
f31e11d8 | 1072 | if (!match_state || p->state == match_state) |
93dcf55f | 1073 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1074 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1075 | |
85ba2d86 RM |
1076 | /* |
1077 | * If it changed from the expected state, bail out now. | |
1078 | */ | |
1079 | if (unlikely(!ncsw)) | |
1080 | break; | |
1081 | ||
3a5c359a AK |
1082 | /* |
1083 | * Was it really running after all now that we | |
1084 | * checked with the proper locks actually held? | |
1085 | * | |
1086 | * Oops. Go back and try again.. | |
1087 | */ | |
1088 | if (unlikely(running)) { | |
1089 | cpu_relax(); | |
1090 | continue; | |
1091 | } | |
fa490cfd | 1092 | |
3a5c359a AK |
1093 | /* |
1094 | * It's not enough that it's not actively running, | |
1095 | * it must be off the runqueue _entirely_, and not | |
1096 | * preempted! | |
1097 | * | |
80dd99b3 | 1098 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1099 | * running right now), it's preempted, and we should |
1100 | * yield - it could be a while. | |
1101 | */ | |
1102 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1103 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1104 | ||
1105 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1106 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1107 | continue; |
1108 | } | |
fa490cfd | 1109 | |
3a5c359a AK |
1110 | /* |
1111 | * Ahh, all good. It wasn't running, and it wasn't | |
1112 | * runnable, which means that it will never become | |
1113 | * running in the future either. We're all done! | |
1114 | */ | |
1115 | break; | |
1116 | } | |
85ba2d86 RM |
1117 | |
1118 | return ncsw; | |
1da177e4 LT |
1119 | } |
1120 | ||
1121 | /*** | |
1122 | * kick_process - kick a running thread to enter/exit the kernel | |
1123 | * @p: the to-be-kicked thread | |
1124 | * | |
1125 | * Cause a process which is running on another CPU to enter | |
1126 | * kernel-mode, without any delay. (to get signals handled.) | |
1127 | * | |
25985edc | 1128 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1129 | * because all it wants to ensure is that the remote task enters |
1130 | * the kernel. If the IPI races and the task has been migrated | |
1131 | * to another CPU then no harm is done and the purpose has been | |
1132 | * achieved as well. | |
1133 | */ | |
36c8b586 | 1134 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1135 | { |
1136 | int cpu; | |
1137 | ||
1138 | preempt_disable(); | |
1139 | cpu = task_cpu(p); | |
1140 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1141 | smp_send_reschedule(cpu); | |
1142 | preempt_enable(); | |
1143 | } | |
b43e3521 | 1144 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1145 | #endif /* CONFIG_SMP */ |
1da177e4 | 1146 | |
970b13ba | 1147 | #ifdef CONFIG_SMP |
30da688e | 1148 | /* |
013fdb80 | 1149 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1150 | */ |
5da9a0fb PZ |
1151 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1152 | { | |
aa00d89c TC |
1153 | int nid = cpu_to_node(cpu); |
1154 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1155 | enum { cpuset, possible, fail } state = cpuset; |
1156 | int dest_cpu; | |
5da9a0fb | 1157 | |
aa00d89c TC |
1158 | /* |
1159 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1160 | * will return -1. There is no cpu on the node, and we should | |
1161 | * select the cpu on the other node. | |
1162 | */ | |
1163 | if (nid != -1) { | |
1164 | nodemask = cpumask_of_node(nid); | |
1165 | ||
1166 | /* Look for allowed, online CPU in same node. */ | |
1167 | for_each_cpu(dest_cpu, nodemask) { | |
1168 | if (!cpu_online(dest_cpu)) | |
1169 | continue; | |
1170 | if (!cpu_active(dest_cpu)) | |
1171 | continue; | |
1172 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1173 | return dest_cpu; | |
1174 | } | |
2baab4e9 | 1175 | } |
5da9a0fb | 1176 | |
2baab4e9 PZ |
1177 | for (;;) { |
1178 | /* Any allowed, online CPU? */ | |
e3831edd | 1179 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1180 | if (!cpu_online(dest_cpu)) |
1181 | continue; | |
1182 | if (!cpu_active(dest_cpu)) | |
1183 | continue; | |
1184 | goto out; | |
1185 | } | |
5da9a0fb | 1186 | |
2baab4e9 PZ |
1187 | switch (state) { |
1188 | case cpuset: | |
1189 | /* No more Mr. Nice Guy. */ | |
1190 | cpuset_cpus_allowed_fallback(p); | |
1191 | state = possible; | |
1192 | break; | |
1193 | ||
1194 | case possible: | |
1195 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1196 | state = fail; | |
1197 | break; | |
1198 | ||
1199 | case fail: | |
1200 | BUG(); | |
1201 | break; | |
1202 | } | |
1203 | } | |
1204 | ||
1205 | out: | |
1206 | if (state != cpuset) { | |
1207 | /* | |
1208 | * Don't tell them about moving exiting tasks or | |
1209 | * kernel threads (both mm NULL), since they never | |
1210 | * leave kernel. | |
1211 | */ | |
1212 | if (p->mm && printk_ratelimit()) { | |
1213 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1214 | task_pid_nr(p), p->comm, cpu); | |
1215 | } | |
5da9a0fb PZ |
1216 | } |
1217 | ||
1218 | return dest_cpu; | |
1219 | } | |
1220 | ||
e2912009 | 1221 | /* |
013fdb80 | 1222 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1223 | */ |
970b13ba | 1224 | static inline |
7608dec2 | 1225 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1226 | { |
7608dec2 | 1227 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1228 | |
1229 | /* | |
1230 | * In order not to call set_task_cpu() on a blocking task we need | |
1231 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1232 | * cpu. | |
1233 | * | |
1234 | * Since this is common to all placement strategies, this lives here. | |
1235 | * | |
1236 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1237 | * not worry about this generic constraint ] | |
1238 | */ | |
fa17b507 | 1239 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1240 | !cpu_online(cpu))) |
5da9a0fb | 1241 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1242 | |
1243 | return cpu; | |
970b13ba | 1244 | } |
09a40af5 MG |
1245 | |
1246 | static void update_avg(u64 *avg, u64 sample) | |
1247 | { | |
1248 | s64 diff = sample - *avg; | |
1249 | *avg += diff >> 3; | |
1250 | } | |
970b13ba PZ |
1251 | #endif |
1252 | ||
d7c01d27 | 1253 | static void |
b84cb5df | 1254 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1255 | { |
d7c01d27 | 1256 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1257 | struct rq *rq = this_rq(); |
1258 | ||
d7c01d27 PZ |
1259 | #ifdef CONFIG_SMP |
1260 | int this_cpu = smp_processor_id(); | |
1261 | ||
1262 | if (cpu == this_cpu) { | |
1263 | schedstat_inc(rq, ttwu_local); | |
1264 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1265 | } else { | |
1266 | struct sched_domain *sd; | |
1267 | ||
1268 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1269 | rcu_read_lock(); |
d7c01d27 PZ |
1270 | for_each_domain(this_cpu, sd) { |
1271 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1272 | schedstat_inc(sd, ttwu_wake_remote); | |
1273 | break; | |
1274 | } | |
1275 | } | |
057f3fad | 1276 | rcu_read_unlock(); |
d7c01d27 | 1277 | } |
f339b9dc PZ |
1278 | |
1279 | if (wake_flags & WF_MIGRATED) | |
1280 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1281 | ||
d7c01d27 PZ |
1282 | #endif /* CONFIG_SMP */ |
1283 | ||
1284 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1285 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1286 | |
1287 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1288 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1289 | |
d7c01d27 PZ |
1290 | #endif /* CONFIG_SCHEDSTATS */ |
1291 | } | |
1292 | ||
1293 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1294 | { | |
9ed3811a | 1295 | activate_task(rq, p, en_flags); |
fd2f4419 | 1296 | p->on_rq = 1; |
c2f7115e PZ |
1297 | |
1298 | /* if a worker is waking up, notify workqueue */ | |
1299 | if (p->flags & PF_WQ_WORKER) | |
1300 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1301 | } |
1302 | ||
23f41eeb PZ |
1303 | /* |
1304 | * Mark the task runnable and perform wakeup-preemption. | |
1305 | */ | |
89363381 | 1306 | static void |
23f41eeb | 1307 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1308 | { |
9ed3811a | 1309 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1310 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1311 | |
1312 | p->state = TASK_RUNNING; | |
1313 | #ifdef CONFIG_SMP | |
1314 | if (p->sched_class->task_woken) | |
1315 | p->sched_class->task_woken(rq, p); | |
1316 | ||
e69c6341 | 1317 | if (rq->idle_stamp) { |
9ed3811a TH |
1318 | u64 delta = rq->clock - rq->idle_stamp; |
1319 | u64 max = 2*sysctl_sched_migration_cost; | |
1320 | ||
1321 | if (delta > max) | |
1322 | rq->avg_idle = max; | |
1323 | else | |
1324 | update_avg(&rq->avg_idle, delta); | |
1325 | rq->idle_stamp = 0; | |
1326 | } | |
1327 | #endif | |
1328 | } | |
1329 | ||
c05fbafb PZ |
1330 | static void |
1331 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1332 | { | |
1333 | #ifdef CONFIG_SMP | |
1334 | if (p->sched_contributes_to_load) | |
1335 | rq->nr_uninterruptible--; | |
1336 | #endif | |
1337 | ||
1338 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1339 | ttwu_do_wakeup(rq, p, wake_flags); | |
1340 | } | |
1341 | ||
1342 | /* | |
1343 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1344 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1345 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1346 | * the task is still ->on_rq. | |
1347 | */ | |
1348 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1349 | { | |
1350 | struct rq *rq; | |
1351 | int ret = 0; | |
1352 | ||
1353 | rq = __task_rq_lock(p); | |
1354 | if (p->on_rq) { | |
1355 | ttwu_do_wakeup(rq, p, wake_flags); | |
1356 | ret = 1; | |
1357 | } | |
1358 | __task_rq_unlock(rq); | |
1359 | ||
1360 | return ret; | |
1361 | } | |
1362 | ||
317f3941 | 1363 | #ifdef CONFIG_SMP |
fa14ff4a | 1364 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1365 | { |
1366 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1367 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1368 | struct task_struct *p; | |
317f3941 PZ |
1369 | |
1370 | raw_spin_lock(&rq->lock); | |
1371 | ||
fa14ff4a PZ |
1372 | while (llist) { |
1373 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1374 | llist = llist_next(llist); | |
317f3941 PZ |
1375 | ttwu_do_activate(rq, p, 0); |
1376 | } | |
1377 | ||
1378 | raw_spin_unlock(&rq->lock); | |
1379 | } | |
1380 | ||
1381 | void scheduler_ipi(void) | |
1382 | { | |
ca38062e | 1383 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1384 | return; |
1385 | ||
1386 | /* | |
1387 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1388 | * traditionally all their work was done from the interrupt return | |
1389 | * path. Now that we actually do some work, we need to make sure | |
1390 | * we do call them. | |
1391 | * | |
1392 | * Some archs already do call them, luckily irq_enter/exit nest | |
1393 | * properly. | |
1394 | * | |
1395 | * Arguably we should visit all archs and update all handlers, | |
1396 | * however a fair share of IPIs are still resched only so this would | |
1397 | * somewhat pessimize the simple resched case. | |
1398 | */ | |
1399 | irq_enter(); | |
fa14ff4a | 1400 | sched_ttwu_pending(); |
ca38062e SS |
1401 | |
1402 | /* | |
1403 | * Check if someone kicked us for doing the nohz idle load balance. | |
1404 | */ | |
6eb57e0d SS |
1405 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1406 | this_rq()->idle_balance = 1; | |
ca38062e | 1407 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1408 | } |
c5d753a5 | 1409 | irq_exit(); |
317f3941 PZ |
1410 | } |
1411 | ||
1412 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1413 | { | |
fa14ff4a | 1414 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1415 | smp_send_reschedule(cpu); |
1416 | } | |
d6aa8f85 | 1417 | |
39be3501 | 1418 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1419 | { |
1420 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1421 | } | |
d6aa8f85 | 1422 | #endif /* CONFIG_SMP */ |
317f3941 | 1423 | |
c05fbafb PZ |
1424 | static void ttwu_queue(struct task_struct *p, int cpu) |
1425 | { | |
1426 | struct rq *rq = cpu_rq(cpu); | |
1427 | ||
17d9f311 | 1428 | #if defined(CONFIG_SMP) |
39be3501 | 1429 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1430 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1431 | ttwu_queue_remote(p, cpu); |
1432 | return; | |
1433 | } | |
1434 | #endif | |
1435 | ||
c05fbafb PZ |
1436 | raw_spin_lock(&rq->lock); |
1437 | ttwu_do_activate(rq, p, 0); | |
1438 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1439 | } |
1440 | ||
1441 | /** | |
1da177e4 | 1442 | * try_to_wake_up - wake up a thread |
9ed3811a | 1443 | * @p: the thread to be awakened |
1da177e4 | 1444 | * @state: the mask of task states that can be woken |
9ed3811a | 1445 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1446 | * |
1447 | * Put it on the run-queue if it's not already there. The "current" | |
1448 | * thread is always on the run-queue (except when the actual | |
1449 | * re-schedule is in progress), and as such you're allowed to do | |
1450 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1451 | * runnable without the overhead of this. | |
1452 | * | |
9ed3811a TH |
1453 | * Returns %true if @p was woken up, %false if it was already running |
1454 | * or @state didn't match @p's state. | |
1da177e4 | 1455 | */ |
e4a52bcb PZ |
1456 | static int |
1457 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1458 | { |
1da177e4 | 1459 | unsigned long flags; |
c05fbafb | 1460 | int cpu, success = 0; |
2398f2c6 | 1461 | |
04e2f174 | 1462 | smp_wmb(); |
013fdb80 | 1463 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1464 | if (!(p->state & state)) |
1da177e4 LT |
1465 | goto out; |
1466 | ||
c05fbafb | 1467 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1468 | cpu = task_cpu(p); |
1da177e4 | 1469 | |
c05fbafb PZ |
1470 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1471 | goto stat; | |
1da177e4 | 1472 | |
1da177e4 | 1473 | #ifdef CONFIG_SMP |
e9c84311 | 1474 | /* |
c05fbafb PZ |
1475 | * If the owning (remote) cpu is still in the middle of schedule() with |
1476 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1477 | */ |
f3e94786 | 1478 | while (p->on_cpu) |
e4a52bcb | 1479 | cpu_relax(); |
0970d299 | 1480 | /* |
e4a52bcb | 1481 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1482 | */ |
e4a52bcb | 1483 | smp_rmb(); |
1da177e4 | 1484 | |
a8e4f2ea | 1485 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1486 | p->state = TASK_WAKING; |
e7693a36 | 1487 | |
e4a52bcb | 1488 | if (p->sched_class->task_waking) |
74f8e4b2 | 1489 | p->sched_class->task_waking(p); |
efbbd05a | 1490 | |
7608dec2 | 1491 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1492 | if (task_cpu(p) != cpu) { |
1493 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1494 | set_task_cpu(p, cpu); |
f339b9dc | 1495 | } |
1da177e4 | 1496 | #endif /* CONFIG_SMP */ |
1da177e4 | 1497 | |
c05fbafb PZ |
1498 | ttwu_queue(p, cpu); |
1499 | stat: | |
b84cb5df | 1500 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1501 | out: |
013fdb80 | 1502 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1503 | |
1504 | return success; | |
1505 | } | |
1506 | ||
21aa9af0 TH |
1507 | /** |
1508 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1509 | * @p: the thread to be awakened | |
1510 | * | |
2acca55e | 1511 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1512 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1513 | * the current task. |
21aa9af0 TH |
1514 | */ |
1515 | static void try_to_wake_up_local(struct task_struct *p) | |
1516 | { | |
1517 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1518 | |
1519 | BUG_ON(rq != this_rq()); | |
1520 | BUG_ON(p == current); | |
1521 | lockdep_assert_held(&rq->lock); | |
1522 | ||
2acca55e PZ |
1523 | if (!raw_spin_trylock(&p->pi_lock)) { |
1524 | raw_spin_unlock(&rq->lock); | |
1525 | raw_spin_lock(&p->pi_lock); | |
1526 | raw_spin_lock(&rq->lock); | |
1527 | } | |
1528 | ||
21aa9af0 | 1529 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1530 | goto out; |
21aa9af0 | 1531 | |
fd2f4419 | 1532 | if (!p->on_rq) |
d7c01d27 PZ |
1533 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1534 | ||
23f41eeb | 1535 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1536 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1537 | out: |
1538 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1539 | } |
1540 | ||
50fa610a DH |
1541 | /** |
1542 | * wake_up_process - Wake up a specific process | |
1543 | * @p: The process to be woken up. | |
1544 | * | |
1545 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1546 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1547 | * running. | |
1548 | * | |
1549 | * It may be assumed that this function implies a write memory barrier before | |
1550 | * changing the task state if and only if any tasks are woken up. | |
1551 | */ | |
7ad5b3a5 | 1552 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1553 | { |
9067ac85 ON |
1554 | WARN_ON(task_is_stopped_or_traced(p)); |
1555 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1556 | } |
1da177e4 LT |
1557 | EXPORT_SYMBOL(wake_up_process); |
1558 | ||
7ad5b3a5 | 1559 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1560 | { |
1561 | return try_to_wake_up(p, state, 0); | |
1562 | } | |
1563 | ||
1da177e4 LT |
1564 | /* |
1565 | * Perform scheduler related setup for a newly forked process p. | |
1566 | * p is forked by current. | |
dd41f596 IM |
1567 | * |
1568 | * __sched_fork() is basic setup used by init_idle() too: | |
1569 | */ | |
1570 | static void __sched_fork(struct task_struct *p) | |
1571 | { | |
fd2f4419 PZ |
1572 | p->on_rq = 0; |
1573 | ||
1574 | p->se.on_rq = 0; | |
dd41f596 IM |
1575 | p->se.exec_start = 0; |
1576 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1577 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1578 | p->se.nr_migrations = 0; |
da7a735e | 1579 | p->se.vruntime = 0; |
fd2f4419 | 1580 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 1581 | |
f4e26b12 PT |
1582 | /* |
1583 | * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be | |
1584 | * removed when useful for applications beyond shares distribution (e.g. | |
1585 | * load-balance). | |
1586 | */ | |
1587 | #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED) | |
9d85f21c PT |
1588 | p->se.avg.runnable_avg_period = 0; |
1589 | p->se.avg.runnable_avg_sum = 0; | |
1590 | #endif | |
6cfb0d5d | 1591 | #ifdef CONFIG_SCHEDSTATS |
41acab88 | 1592 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1593 | #endif |
476d139c | 1594 | |
fa717060 | 1595 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1596 | |
e107be36 AK |
1597 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1598 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1599 | #endif | |
cbee9f88 PZ |
1600 | |
1601 | #ifdef CONFIG_NUMA_BALANCING | |
1602 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
1603 | p->mm->numa_next_scan = jiffies; | |
b8593bfd | 1604 | p->mm->numa_next_reset = jiffies; |
cbee9f88 PZ |
1605 | p->mm->numa_scan_seq = 0; |
1606 | } | |
1607 | ||
1608 | p->node_stamp = 0ULL; | |
1609 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
1610 | p->numa_migrate_seq = p->mm ? p->mm->numa_scan_seq - 1 : 0; | |
4b96a29b | 1611 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 PZ |
1612 | p->numa_work.next = &p->numa_work; |
1613 | #endif /* CONFIG_NUMA_BALANCING */ | |
dd41f596 IM |
1614 | } |
1615 | ||
1a687c2e | 1616 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1617 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1618 | void set_numabalancing_state(bool enabled) |
1619 | { | |
1620 | if (enabled) | |
1621 | sched_feat_set("NUMA"); | |
1622 | else | |
1623 | sched_feat_set("NO_NUMA"); | |
1624 | } | |
3105b86a MG |
1625 | #else |
1626 | __read_mostly bool numabalancing_enabled; | |
1627 | ||
1628 | void set_numabalancing_state(bool enabled) | |
1629 | { | |
1630 | numabalancing_enabled = enabled; | |
dd41f596 | 1631 | } |
3105b86a | 1632 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1633 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1634 | |
1635 | /* | |
1636 | * fork()/clone()-time setup: | |
1637 | */ | |
3e51e3ed | 1638 | void sched_fork(struct task_struct *p) |
dd41f596 | 1639 | { |
0122ec5b | 1640 | unsigned long flags; |
dd41f596 IM |
1641 | int cpu = get_cpu(); |
1642 | ||
1643 | __sched_fork(p); | |
06b83b5f | 1644 | /* |
0017d735 | 1645 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1646 | * nobody will actually run it, and a signal or other external |
1647 | * event cannot wake it up and insert it on the runqueue either. | |
1648 | */ | |
0017d735 | 1649 | p->state = TASK_RUNNING; |
dd41f596 | 1650 | |
c350a04e MG |
1651 | /* |
1652 | * Make sure we do not leak PI boosting priority to the child. | |
1653 | */ | |
1654 | p->prio = current->normal_prio; | |
1655 | ||
b9dc29e7 MG |
1656 | /* |
1657 | * Revert to default priority/policy on fork if requested. | |
1658 | */ | |
1659 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1660 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1661 | p->policy = SCHED_NORMAL; |
6c697bdf | 1662 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1663 | p->rt_priority = 0; |
1664 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1665 | p->static_prio = NICE_TO_PRIO(0); | |
1666 | ||
1667 | p->prio = p->normal_prio = __normal_prio(p); | |
1668 | set_load_weight(p); | |
6c697bdf | 1669 | |
b9dc29e7 MG |
1670 | /* |
1671 | * We don't need the reset flag anymore after the fork. It has | |
1672 | * fulfilled its duty: | |
1673 | */ | |
1674 | p->sched_reset_on_fork = 0; | |
1675 | } | |
ca94c442 | 1676 | |
2ddbf952 HS |
1677 | if (!rt_prio(p->prio)) |
1678 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1679 | |
cd29fe6f PZ |
1680 | if (p->sched_class->task_fork) |
1681 | p->sched_class->task_fork(p); | |
1682 | ||
86951599 PZ |
1683 | /* |
1684 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1685 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1686 | * is ran before sched_fork(). | |
1687 | * | |
1688 | * Silence PROVE_RCU. | |
1689 | */ | |
0122ec5b | 1690 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1691 | set_task_cpu(p, cpu); |
0122ec5b | 1692 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1693 | |
52f17b6c | 1694 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1695 | if (likely(sched_info_on())) |
52f17b6c | 1696 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1697 | #endif |
3ca7a440 PZ |
1698 | #if defined(CONFIG_SMP) |
1699 | p->on_cpu = 0; | |
4866cde0 | 1700 | #endif |
bdd4e85d | 1701 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1702 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1703 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1704 | #endif |
806c09a7 | 1705 | #ifdef CONFIG_SMP |
917b627d | 1706 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1707 | #endif |
917b627d | 1708 | |
476d139c | 1709 | put_cpu(); |
1da177e4 LT |
1710 | } |
1711 | ||
1712 | /* | |
1713 | * wake_up_new_task - wake up a newly created task for the first time. | |
1714 | * | |
1715 | * This function will do some initial scheduler statistics housekeeping | |
1716 | * that must be done for every newly created context, then puts the task | |
1717 | * on the runqueue and wakes it. | |
1718 | */ | |
3e51e3ed | 1719 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1720 | { |
1721 | unsigned long flags; | |
dd41f596 | 1722 | struct rq *rq; |
fabf318e | 1723 | |
ab2515c4 | 1724 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1725 | #ifdef CONFIG_SMP |
1726 | /* | |
1727 | * Fork balancing, do it here and not earlier because: | |
1728 | * - cpus_allowed can change in the fork path | |
1729 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1730 | */ |
ab2515c4 | 1731 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1732 | #endif |
1733 | ||
ab2515c4 | 1734 | rq = __task_rq_lock(p); |
cd29fe6f | 1735 | activate_task(rq, p, 0); |
fd2f4419 | 1736 | p->on_rq = 1; |
89363381 | 1737 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1738 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1739 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1740 | if (p->sched_class->task_woken) |
1741 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1742 | #endif |
0122ec5b | 1743 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1744 | } |
1745 | ||
e107be36 AK |
1746 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1747 | ||
1748 | /** | |
80dd99b3 | 1749 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1750 | * @notifier: notifier struct to register |
e107be36 AK |
1751 | */ |
1752 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1753 | { | |
1754 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1755 | } | |
1756 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1757 | ||
1758 | /** | |
1759 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1760 | * @notifier: notifier struct to unregister |
e107be36 AK |
1761 | * |
1762 | * This is safe to call from within a preemption notifier. | |
1763 | */ | |
1764 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1765 | { | |
1766 | hlist_del(¬ifier->link); | |
1767 | } | |
1768 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1769 | ||
1770 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1771 | { | |
1772 | struct preempt_notifier *notifier; | |
e107be36 | 1773 | |
b67bfe0d | 1774 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1775 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
1776 | } | |
1777 | ||
1778 | static void | |
1779 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1780 | struct task_struct *next) | |
1781 | { | |
1782 | struct preempt_notifier *notifier; | |
e107be36 | 1783 | |
b67bfe0d | 1784 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1785 | notifier->ops->sched_out(notifier, next); |
1786 | } | |
1787 | ||
6d6bc0ad | 1788 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1789 | |
1790 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1791 | { | |
1792 | } | |
1793 | ||
1794 | static void | |
1795 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1796 | struct task_struct *next) | |
1797 | { | |
1798 | } | |
1799 | ||
6d6bc0ad | 1800 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1801 | |
4866cde0 NP |
1802 | /** |
1803 | * prepare_task_switch - prepare to switch tasks | |
1804 | * @rq: the runqueue preparing to switch | |
421cee29 | 1805 | * @prev: the current task that is being switched out |
4866cde0 NP |
1806 | * @next: the task we are going to switch to. |
1807 | * | |
1808 | * This is called with the rq lock held and interrupts off. It must | |
1809 | * be paired with a subsequent finish_task_switch after the context | |
1810 | * switch. | |
1811 | * | |
1812 | * prepare_task_switch sets up locking and calls architecture specific | |
1813 | * hooks. | |
1814 | */ | |
e107be36 AK |
1815 | static inline void |
1816 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1817 | struct task_struct *next) | |
4866cde0 | 1818 | { |
895dd92c | 1819 | trace_sched_switch(prev, next); |
fe4b04fa PZ |
1820 | sched_info_switch(prev, next); |
1821 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1822 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1823 | prepare_lock_switch(rq, next); |
1824 | prepare_arch_switch(next); | |
1825 | } | |
1826 | ||
1da177e4 LT |
1827 | /** |
1828 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1829 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1830 | * @prev: the thread we just switched away from. |
1831 | * | |
4866cde0 NP |
1832 | * finish_task_switch must be called after the context switch, paired |
1833 | * with a prepare_task_switch call before the context switch. | |
1834 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1835 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1836 | * |
1837 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1838 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1839 | * with the lock held can cause deadlocks; see schedule() for |
1840 | * details.) | |
1841 | */ | |
a9957449 | 1842 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1843 | __releases(rq->lock) |
1844 | { | |
1da177e4 | 1845 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1846 | long prev_state; |
1da177e4 LT |
1847 | |
1848 | rq->prev_mm = NULL; | |
1849 | ||
1850 | /* | |
1851 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1852 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1853 | * schedule one last time. The schedule call will never return, and |
1854 | * the scheduled task must drop that reference. | |
c394cc9f | 1855 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1856 | * still held, otherwise prev could be scheduled on another cpu, die |
1857 | * there before we look at prev->state, and then the reference would | |
1858 | * be dropped twice. | |
1859 | * Manfred Spraul <manfred@colorfullife.com> | |
1860 | */ | |
55a101f8 | 1861 | prev_state = prev->state; |
bf9fae9f | 1862 | vtime_task_switch(prev); |
4866cde0 | 1863 | finish_arch_switch(prev); |
a8d757ef | 1864 | perf_event_task_sched_in(prev, current); |
4866cde0 | 1865 | finish_lock_switch(rq, prev); |
01f23e16 | 1866 | finish_arch_post_lock_switch(); |
e8fa1362 | 1867 | |
e107be36 | 1868 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1869 | if (mm) |
1870 | mmdrop(mm); | |
c394cc9f | 1871 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1872 | /* |
1873 | * Remove function-return probe instances associated with this | |
1874 | * task and put them back on the free list. | |
9761eea8 | 1875 | */ |
c6fd91f0 | 1876 | kprobe_flush_task(prev); |
1da177e4 | 1877 | put_task_struct(prev); |
c6fd91f0 | 1878 | } |
1da177e4 LT |
1879 | } |
1880 | ||
3f029d3c GH |
1881 | #ifdef CONFIG_SMP |
1882 | ||
1883 | /* assumes rq->lock is held */ | |
1884 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1885 | { | |
1886 | if (prev->sched_class->pre_schedule) | |
1887 | prev->sched_class->pre_schedule(rq, prev); | |
1888 | } | |
1889 | ||
1890 | /* rq->lock is NOT held, but preemption is disabled */ | |
1891 | static inline void post_schedule(struct rq *rq) | |
1892 | { | |
1893 | if (rq->post_schedule) { | |
1894 | unsigned long flags; | |
1895 | ||
05fa785c | 1896 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1897 | if (rq->curr->sched_class->post_schedule) |
1898 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1899 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1900 | |
1901 | rq->post_schedule = 0; | |
1902 | } | |
1903 | } | |
1904 | ||
1905 | #else | |
da19ab51 | 1906 | |
3f029d3c GH |
1907 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1908 | { | |
1909 | } | |
1910 | ||
1911 | static inline void post_schedule(struct rq *rq) | |
1912 | { | |
1da177e4 LT |
1913 | } |
1914 | ||
3f029d3c GH |
1915 | #endif |
1916 | ||
1da177e4 LT |
1917 | /** |
1918 | * schedule_tail - first thing a freshly forked thread must call. | |
1919 | * @prev: the thread we just switched away from. | |
1920 | */ | |
36c8b586 | 1921 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1922 | __releases(rq->lock) |
1923 | { | |
70b97a7f IM |
1924 | struct rq *rq = this_rq(); |
1925 | ||
4866cde0 | 1926 | finish_task_switch(rq, prev); |
da19ab51 | 1927 | |
3f029d3c GH |
1928 | /* |
1929 | * FIXME: do we need to worry about rq being invalidated by the | |
1930 | * task_switch? | |
1931 | */ | |
1932 | post_schedule(rq); | |
70b97a7f | 1933 | |
4866cde0 NP |
1934 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
1935 | /* In this case, finish_task_switch does not reenable preemption */ | |
1936 | preempt_enable(); | |
1937 | #endif | |
1da177e4 | 1938 | if (current->set_child_tid) |
b488893a | 1939 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1940 | } |
1941 | ||
1942 | /* | |
1943 | * context_switch - switch to the new MM and the new | |
1944 | * thread's register state. | |
1945 | */ | |
dd41f596 | 1946 | static inline void |
70b97a7f | 1947 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1948 | struct task_struct *next) |
1da177e4 | 1949 | { |
dd41f596 | 1950 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1951 | |
e107be36 | 1952 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 1953 | |
dd41f596 IM |
1954 | mm = next->mm; |
1955 | oldmm = prev->active_mm; | |
9226d125 ZA |
1956 | /* |
1957 | * For paravirt, this is coupled with an exit in switch_to to | |
1958 | * combine the page table reload and the switch backend into | |
1959 | * one hypercall. | |
1960 | */ | |
224101ed | 1961 | arch_start_context_switch(prev); |
9226d125 | 1962 | |
31915ab4 | 1963 | if (!mm) { |
1da177e4 LT |
1964 | next->active_mm = oldmm; |
1965 | atomic_inc(&oldmm->mm_count); | |
1966 | enter_lazy_tlb(oldmm, next); | |
1967 | } else | |
1968 | switch_mm(oldmm, mm, next); | |
1969 | ||
31915ab4 | 1970 | if (!prev->mm) { |
1da177e4 | 1971 | prev->active_mm = NULL; |
1da177e4 LT |
1972 | rq->prev_mm = oldmm; |
1973 | } | |
3a5f5e48 IM |
1974 | /* |
1975 | * Since the runqueue lock will be released by the next | |
1976 | * task (which is an invalid locking op but in the case | |
1977 | * of the scheduler it's an obvious special-case), so we | |
1978 | * do an early lockdep release here: | |
1979 | */ | |
1980 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1981 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1982 | #endif |
1da177e4 | 1983 | |
91d1aa43 | 1984 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
1985 | /* Here we just switch the register state and the stack. */ |
1986 | switch_to(prev, next, prev); | |
1987 | ||
dd41f596 IM |
1988 | barrier(); |
1989 | /* | |
1990 | * this_rq must be evaluated again because prev may have moved | |
1991 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1992 | * frame will be invalid. | |
1993 | */ | |
1994 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1995 | } |
1996 | ||
1997 | /* | |
1c3e8264 | 1998 | * nr_running and nr_context_switches: |
1da177e4 LT |
1999 | * |
2000 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2001 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2002 | */ |
2003 | unsigned long nr_running(void) | |
2004 | { | |
2005 | unsigned long i, sum = 0; | |
2006 | ||
2007 | for_each_online_cpu(i) | |
2008 | sum += cpu_rq(i)->nr_running; | |
2009 | ||
2010 | return sum; | |
f711f609 | 2011 | } |
1da177e4 | 2012 | |
1da177e4 | 2013 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2014 | { |
cc94abfc SR |
2015 | int i; |
2016 | unsigned long long sum = 0; | |
46cb4b7c | 2017 | |
0a945022 | 2018 | for_each_possible_cpu(i) |
1da177e4 | 2019 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2020 | |
1da177e4 LT |
2021 | return sum; |
2022 | } | |
483b4ee6 | 2023 | |
1da177e4 LT |
2024 | unsigned long nr_iowait(void) |
2025 | { | |
2026 | unsigned long i, sum = 0; | |
483b4ee6 | 2027 | |
0a945022 | 2028 | for_each_possible_cpu(i) |
1da177e4 | 2029 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2030 | |
1da177e4 LT |
2031 | return sum; |
2032 | } | |
483b4ee6 | 2033 | |
8c215bd3 | 2034 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2035 | { |
8c215bd3 | 2036 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2037 | return atomic_read(&this->nr_iowait); |
2038 | } | |
46cb4b7c | 2039 | |
69d25870 AV |
2040 | unsigned long this_cpu_load(void) |
2041 | { | |
2042 | struct rq *this = this_rq(); | |
2043 | return this->cpu_load[0]; | |
2044 | } | |
e790fb0b | 2045 | |
46cb4b7c | 2046 | |
5167e8d5 PZ |
2047 | /* |
2048 | * Global load-average calculations | |
2049 | * | |
2050 | * We take a distributed and async approach to calculating the global load-avg | |
2051 | * in order to minimize overhead. | |
2052 | * | |
2053 | * The global load average is an exponentially decaying average of nr_running + | |
2054 | * nr_uninterruptible. | |
2055 | * | |
2056 | * Once every LOAD_FREQ: | |
2057 | * | |
2058 | * nr_active = 0; | |
2059 | * for_each_possible_cpu(cpu) | |
2060 | * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; | |
2061 | * | |
2062 | * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) | |
2063 | * | |
2064 | * Due to a number of reasons the above turns in the mess below: | |
2065 | * | |
2066 | * - for_each_possible_cpu() is prohibitively expensive on machines with | |
2067 | * serious number of cpus, therefore we need to take a distributed approach | |
2068 | * to calculating nr_active. | |
2069 | * | |
2070 | * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 | |
2071 | * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } | |
2072 | * | |
2073 | * So assuming nr_active := 0 when we start out -- true per definition, we | |
2074 | * can simply take per-cpu deltas and fold those into a global accumulate | |
2075 | * to obtain the same result. See calc_load_fold_active(). | |
2076 | * | |
2077 | * Furthermore, in order to avoid synchronizing all per-cpu delta folding | |
2078 | * across the machine, we assume 10 ticks is sufficient time for every | |
2079 | * cpu to have completed this task. | |
2080 | * | |
2081 | * This places an upper-bound on the IRQ-off latency of the machine. Then | |
2082 | * again, being late doesn't loose the delta, just wrecks the sample. | |
2083 | * | |
2084 | * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because | |
2085 | * this would add another cross-cpu cacheline miss and atomic operation | |
2086 | * to the wakeup path. Instead we increment on whatever cpu the task ran | |
2087 | * when it went into uninterruptible state and decrement on whatever cpu | |
2088 | * did the wakeup. This means that only the sum of nr_uninterruptible over | |
2089 | * all cpus yields the correct result. | |
2090 | * | |
2091 | * This covers the NO_HZ=n code, for extra head-aches, see the comment below. | |
2092 | */ | |
2093 | ||
dce48a84 TG |
2094 | /* Variables and functions for calc_load */ |
2095 | static atomic_long_t calc_load_tasks; | |
2096 | static unsigned long calc_load_update; | |
2097 | unsigned long avenrun[3]; | |
5167e8d5 PZ |
2098 | EXPORT_SYMBOL(avenrun); /* should be removed */ |
2099 | ||
2100 | /** | |
2101 | * get_avenrun - get the load average array | |
2102 | * @loads: pointer to dest load array | |
2103 | * @offset: offset to add | |
2104 | * @shift: shift count to shift the result left | |
2105 | * | |
2106 | * These values are estimates at best, so no need for locking. | |
2107 | */ | |
2108 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2109 | { | |
2110 | loads[0] = (avenrun[0] + offset) << shift; | |
2111 | loads[1] = (avenrun[1] + offset) << shift; | |
2112 | loads[2] = (avenrun[2] + offset) << shift; | |
2113 | } | |
46cb4b7c | 2114 | |
74f5187a PZ |
2115 | static long calc_load_fold_active(struct rq *this_rq) |
2116 | { | |
2117 | long nr_active, delta = 0; | |
2118 | ||
2119 | nr_active = this_rq->nr_running; | |
2120 | nr_active += (long) this_rq->nr_uninterruptible; | |
2121 | ||
2122 | if (nr_active != this_rq->calc_load_active) { | |
2123 | delta = nr_active - this_rq->calc_load_active; | |
2124 | this_rq->calc_load_active = nr_active; | |
2125 | } | |
2126 | ||
2127 | return delta; | |
2128 | } | |
2129 | ||
5167e8d5 PZ |
2130 | /* |
2131 | * a1 = a0 * e + a * (1 - e) | |
2132 | */ | |
0f004f5a PZ |
2133 | static unsigned long |
2134 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2135 | { | |
2136 | load *= exp; | |
2137 | load += active * (FIXED_1 - exp); | |
2138 | load += 1UL << (FSHIFT - 1); | |
2139 | return load >> FSHIFT; | |
2140 | } | |
2141 | ||
3451d024 | 2142 | #ifdef CONFIG_NO_HZ_COMMON |
74f5187a | 2143 | /* |
5167e8d5 PZ |
2144 | * Handle NO_HZ for the global load-average. |
2145 | * | |
2146 | * Since the above described distributed algorithm to compute the global | |
2147 | * load-average relies on per-cpu sampling from the tick, it is affected by | |
2148 | * NO_HZ. | |
2149 | * | |
2150 | * The basic idea is to fold the nr_active delta into a global idle-delta upon | |
2151 | * entering NO_HZ state such that we can include this as an 'extra' cpu delta | |
2152 | * when we read the global state. | |
2153 | * | |
2154 | * Obviously reality has to ruin such a delightfully simple scheme: | |
2155 | * | |
2156 | * - When we go NO_HZ idle during the window, we can negate our sample | |
2157 | * contribution, causing under-accounting. | |
2158 | * | |
2159 | * We avoid this by keeping two idle-delta counters and flipping them | |
2160 | * when the window starts, thus separating old and new NO_HZ load. | |
2161 | * | |
2162 | * The only trick is the slight shift in index flip for read vs write. | |
2163 | * | |
2164 | * 0s 5s 10s 15s | |
2165 | * +10 +10 +10 +10 | |
2166 | * |-|-----------|-|-----------|-|-----------|-| | |
2167 | * r:0 0 1 1 0 0 1 1 0 | |
2168 | * w:0 1 1 0 0 1 1 0 0 | |
2169 | * | |
2170 | * This ensures we'll fold the old idle contribution in this window while | |
2171 | * accumlating the new one. | |
2172 | * | |
2173 | * - When we wake up from NO_HZ idle during the window, we push up our | |
2174 | * contribution, since we effectively move our sample point to a known | |
2175 | * busy state. | |
2176 | * | |
2177 | * This is solved by pushing the window forward, and thus skipping the | |
2178 | * sample, for this cpu (effectively using the idle-delta for this cpu which | |
2179 | * was in effect at the time the window opened). This also solves the issue | |
2180 | * of having to deal with a cpu having been in NOHZ idle for multiple | |
2181 | * LOAD_FREQ intervals. | |
74f5187a PZ |
2182 | * |
2183 | * When making the ILB scale, we should try to pull this in as well. | |
2184 | */ | |
5167e8d5 PZ |
2185 | static atomic_long_t calc_load_idle[2]; |
2186 | static int calc_load_idx; | |
74f5187a | 2187 | |
5167e8d5 | 2188 | static inline int calc_load_write_idx(void) |
74f5187a | 2189 | { |
5167e8d5 PZ |
2190 | int idx = calc_load_idx; |
2191 | ||
2192 | /* | |
2193 | * See calc_global_nohz(), if we observe the new index, we also | |
2194 | * need to observe the new update time. | |
2195 | */ | |
2196 | smp_rmb(); | |
2197 | ||
2198 | /* | |
2199 | * If the folding window started, make sure we start writing in the | |
2200 | * next idle-delta. | |
2201 | */ | |
2202 | if (!time_before(jiffies, calc_load_update)) | |
2203 | idx++; | |
2204 | ||
2205 | return idx & 1; | |
2206 | } | |
2207 | ||
2208 | static inline int calc_load_read_idx(void) | |
2209 | { | |
2210 | return calc_load_idx & 1; | |
2211 | } | |
2212 | ||
2213 | void calc_load_enter_idle(void) | |
2214 | { | |
2215 | struct rq *this_rq = this_rq(); | |
74f5187a PZ |
2216 | long delta; |
2217 | ||
5167e8d5 PZ |
2218 | /* |
2219 | * We're going into NOHZ mode, if there's any pending delta, fold it | |
2220 | * into the pending idle delta. | |
2221 | */ | |
74f5187a | 2222 | delta = calc_load_fold_active(this_rq); |
5167e8d5 PZ |
2223 | if (delta) { |
2224 | int idx = calc_load_write_idx(); | |
2225 | atomic_long_add(delta, &calc_load_idle[idx]); | |
2226 | } | |
74f5187a PZ |
2227 | } |
2228 | ||
5167e8d5 | 2229 | void calc_load_exit_idle(void) |
74f5187a | 2230 | { |
5167e8d5 PZ |
2231 | struct rq *this_rq = this_rq(); |
2232 | ||
2233 | /* | |
2234 | * If we're still before the sample window, we're done. | |
2235 | */ | |
2236 | if (time_before(jiffies, this_rq->calc_load_update)) | |
2237 | return; | |
74f5187a PZ |
2238 | |
2239 | /* | |
5167e8d5 PZ |
2240 | * We woke inside or after the sample window, this means we're already |
2241 | * accounted through the nohz accounting, so skip the entire deal and | |
2242 | * sync up for the next window. | |
74f5187a | 2243 | */ |
5167e8d5 PZ |
2244 | this_rq->calc_load_update = calc_load_update; |
2245 | if (time_before(jiffies, this_rq->calc_load_update + 10)) | |
2246 | this_rq->calc_load_update += LOAD_FREQ; | |
2247 | } | |
2248 | ||
2249 | static long calc_load_fold_idle(void) | |
2250 | { | |
2251 | int idx = calc_load_read_idx(); | |
2252 | long delta = 0; | |
2253 | ||
2254 | if (atomic_long_read(&calc_load_idle[idx])) | |
2255 | delta = atomic_long_xchg(&calc_load_idle[idx], 0); | |
74f5187a PZ |
2256 | |
2257 | return delta; | |
2258 | } | |
0f004f5a PZ |
2259 | |
2260 | /** | |
2261 | * fixed_power_int - compute: x^n, in O(log n) time | |
2262 | * | |
2263 | * @x: base of the power | |
2264 | * @frac_bits: fractional bits of @x | |
2265 | * @n: power to raise @x to. | |
2266 | * | |
2267 | * By exploiting the relation between the definition of the natural power | |
2268 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2269 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2270 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2271 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2272 | * of course trivially computable in O(log_2 n), the length of our binary | |
2273 | * vector. | |
2274 | */ | |
2275 | static unsigned long | |
2276 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2277 | { | |
2278 | unsigned long result = 1UL << frac_bits; | |
2279 | ||
2280 | if (n) for (;;) { | |
2281 | if (n & 1) { | |
2282 | result *= x; | |
2283 | result += 1UL << (frac_bits - 1); | |
2284 | result >>= frac_bits; | |
2285 | } | |
2286 | n >>= 1; | |
2287 | if (!n) | |
2288 | break; | |
2289 | x *= x; | |
2290 | x += 1UL << (frac_bits - 1); | |
2291 | x >>= frac_bits; | |
2292 | } | |
2293 | ||
2294 | return result; | |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * a1 = a0 * e + a * (1 - e) | |
2299 | * | |
2300 | * a2 = a1 * e + a * (1 - e) | |
2301 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2302 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2303 | * | |
2304 | * a3 = a2 * e + a * (1 - e) | |
2305 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2306 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2307 | * | |
2308 | * ... | |
2309 | * | |
2310 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2311 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2312 | * = a0 * e^n + a * (1 - e^n) | |
2313 | * | |
2314 | * [1] application of the geometric series: | |
2315 | * | |
2316 | * n 1 - x^(n+1) | |
2317 | * S_n := \Sum x^i = ------------- | |
2318 | * i=0 1 - x | |
2319 | */ | |
2320 | static unsigned long | |
2321 | calc_load_n(unsigned long load, unsigned long exp, | |
2322 | unsigned long active, unsigned int n) | |
2323 | { | |
2324 | ||
2325 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2326 | } | |
2327 | ||
2328 | /* | |
2329 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2330 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2331 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2332 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2333 | * | |
2334 | * Once we've updated the global active value, we need to apply the exponential | |
2335 | * weights adjusted to the number of cycles missed. | |
2336 | */ | |
c308b56b | 2337 | static void calc_global_nohz(void) |
0f004f5a PZ |
2338 | { |
2339 | long delta, active, n; | |
2340 | ||
5167e8d5 PZ |
2341 | if (!time_before(jiffies, calc_load_update + 10)) { |
2342 | /* | |
2343 | * Catch-up, fold however many we are behind still | |
2344 | */ | |
2345 | delta = jiffies - calc_load_update - 10; | |
2346 | n = 1 + (delta / LOAD_FREQ); | |
0f004f5a | 2347 | |
5167e8d5 PZ |
2348 | active = atomic_long_read(&calc_load_tasks); |
2349 | active = active > 0 ? active * FIXED_1 : 0; | |
0f004f5a | 2350 | |
5167e8d5 PZ |
2351 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
2352 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2353 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
0f004f5a | 2354 | |
5167e8d5 PZ |
2355 | calc_load_update += n * LOAD_FREQ; |
2356 | } | |
74f5187a | 2357 | |
5167e8d5 PZ |
2358 | /* |
2359 | * Flip the idle index... | |
2360 | * | |
2361 | * Make sure we first write the new time then flip the index, so that | |
2362 | * calc_load_write_idx() will see the new time when it reads the new | |
2363 | * index, this avoids a double flip messing things up. | |
2364 | */ | |
2365 | smp_wmb(); | |
2366 | calc_load_idx++; | |
74f5187a | 2367 | } |
3451d024 | 2368 | #else /* !CONFIG_NO_HZ_COMMON */ |
0f004f5a | 2369 | |
5167e8d5 PZ |
2370 | static inline long calc_load_fold_idle(void) { return 0; } |
2371 | static inline void calc_global_nohz(void) { } | |
74f5187a | 2372 | |
3451d024 | 2373 | #endif /* CONFIG_NO_HZ_COMMON */ |
46cb4b7c | 2374 | |
46cb4b7c | 2375 | /* |
dce48a84 TG |
2376 | * calc_load - update the avenrun load estimates 10 ticks after the |
2377 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2378 | */ |
0f004f5a | 2379 | void calc_global_load(unsigned long ticks) |
7835b98b | 2380 | { |
5167e8d5 | 2381 | long active, delta; |
1da177e4 | 2382 | |
0f004f5a | 2383 | if (time_before(jiffies, calc_load_update + 10)) |
dce48a84 | 2384 | return; |
1da177e4 | 2385 | |
5167e8d5 PZ |
2386 | /* |
2387 | * Fold the 'old' idle-delta to include all NO_HZ cpus. | |
2388 | */ | |
2389 | delta = calc_load_fold_idle(); | |
2390 | if (delta) | |
2391 | atomic_long_add(delta, &calc_load_tasks); | |
2392 | ||
dce48a84 TG |
2393 | active = atomic_long_read(&calc_load_tasks); |
2394 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2395 | |
dce48a84 TG |
2396 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2397 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2398 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2399 | |
dce48a84 | 2400 | calc_load_update += LOAD_FREQ; |
c308b56b PZ |
2401 | |
2402 | /* | |
5167e8d5 | 2403 | * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. |
c308b56b PZ |
2404 | */ |
2405 | calc_global_nohz(); | |
dce48a84 | 2406 | } |
1da177e4 | 2407 | |
dce48a84 | 2408 | /* |
74f5187a PZ |
2409 | * Called from update_cpu_load() to periodically update this CPU's |
2410 | * active count. | |
dce48a84 TG |
2411 | */ |
2412 | static void calc_load_account_active(struct rq *this_rq) | |
2413 | { | |
74f5187a | 2414 | long delta; |
08c183f3 | 2415 | |
74f5187a PZ |
2416 | if (time_before(jiffies, this_rq->calc_load_update)) |
2417 | return; | |
783609c6 | 2418 | |
74f5187a | 2419 | delta = calc_load_fold_active(this_rq); |
74f5187a | 2420 | if (delta) |
dce48a84 | 2421 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2422 | |
2423 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2424 | } |
2425 | ||
5167e8d5 PZ |
2426 | /* |
2427 | * End of global load-average stuff | |
2428 | */ | |
2429 | ||
fdf3e95d VP |
2430 | /* |
2431 | * The exact cpuload at various idx values, calculated at every tick would be | |
2432 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2433 | * | |
2434 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2435 | * on nth tick when cpu may be busy, then we have: | |
2436 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2437 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2438 | * | |
2439 | * decay_load_missed() below does efficient calculation of | |
2440 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2441 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2442 | * | |
2443 | * The calculation is approximated on a 128 point scale. | |
2444 | * degrade_zero_ticks is the number of ticks after which load at any | |
2445 | * particular idx is approximated to be zero. | |
2446 | * degrade_factor is a precomputed table, a row for each load idx. | |
2447 | * Each column corresponds to degradation factor for a power of two ticks, | |
2448 | * based on 128 point scale. | |
2449 | * Example: | |
2450 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2451 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2452 | * | |
2453 | * With this power of 2 load factors, we can degrade the load n times | |
2454 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2455 | * n mult/shifts needed by the exact degradation. | |
2456 | */ | |
2457 | #define DEGRADE_SHIFT 7 | |
2458 | static const unsigned char | |
2459 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2460 | static const unsigned char | |
2461 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2462 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2463 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2464 | {96, 72, 40, 12, 1, 0, 0}, | |
2465 | {112, 98, 75, 43, 15, 1, 0}, | |
2466 | {120, 112, 98, 76, 45, 16, 2} }; | |
2467 | ||
2468 | /* | |
2469 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2470 | * would be when CPU is idle and so we just decay the old load without | |
2471 | * adding any new load. | |
2472 | */ | |
2473 | static unsigned long | |
2474 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2475 | { | |
2476 | int j = 0; | |
2477 | ||
2478 | if (!missed_updates) | |
2479 | return load; | |
2480 | ||
2481 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2482 | return 0; | |
2483 | ||
2484 | if (idx == 1) | |
2485 | return load >> missed_updates; | |
2486 | ||
2487 | while (missed_updates) { | |
2488 | if (missed_updates % 2) | |
2489 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2490 | ||
2491 | missed_updates >>= 1; | |
2492 | j++; | |
2493 | } | |
2494 | return load; | |
2495 | } | |
2496 | ||
46cb4b7c | 2497 | /* |
dd41f596 | 2498 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2499 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2500 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2501 | */ |
556061b0 PZ |
2502 | static void __update_cpu_load(struct rq *this_rq, unsigned long this_load, |
2503 | unsigned long pending_updates) | |
46cb4b7c | 2504 | { |
dd41f596 | 2505 | int i, scale; |
46cb4b7c | 2506 | |
dd41f596 | 2507 | this_rq->nr_load_updates++; |
46cb4b7c | 2508 | |
dd41f596 | 2509 | /* Update our load: */ |
fdf3e95d VP |
2510 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2511 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2512 | unsigned long old_load, new_load; |
7d1e6a9b | 2513 | |
dd41f596 | 2514 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2515 | |
dd41f596 | 2516 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2517 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2518 | new_load = this_load; |
a25707f3 IM |
2519 | /* |
2520 | * Round up the averaging division if load is increasing. This | |
2521 | * prevents us from getting stuck on 9 if the load is 10, for | |
2522 | * example. | |
2523 | */ | |
2524 | if (new_load > old_load) | |
fdf3e95d VP |
2525 | new_load += scale - 1; |
2526 | ||
2527 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2528 | } |
da2b71ed SS |
2529 | |
2530 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2531 | } |
2532 | ||
3451d024 | 2533 | #ifdef CONFIG_NO_HZ_COMMON |
5aaa0b7a PZ |
2534 | /* |
2535 | * There is no sane way to deal with nohz on smp when using jiffies because the | |
2536 | * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading | |
2537 | * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}. | |
2538 | * | |
2539 | * Therefore we cannot use the delta approach from the regular tick since that | |
2540 | * would seriously skew the load calculation. However we'll make do for those | |
2541 | * updates happening while idle (nohz_idle_balance) or coming out of idle | |
2542 | * (tick_nohz_idle_exit). | |
2543 | * | |
2544 | * This means we might still be one tick off for nohz periods. | |
2545 | */ | |
2546 | ||
556061b0 PZ |
2547 | /* |
2548 | * Called from nohz_idle_balance() to update the load ratings before doing the | |
2549 | * idle balance. | |
2550 | */ | |
2551 | void update_idle_cpu_load(struct rq *this_rq) | |
2552 | { | |
5aaa0b7a | 2553 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
556061b0 PZ |
2554 | unsigned long load = this_rq->load.weight; |
2555 | unsigned long pending_updates; | |
2556 | ||
2557 | /* | |
5aaa0b7a | 2558 | * bail if there's load or we're actually up-to-date. |
556061b0 PZ |
2559 | */ |
2560 | if (load || curr_jiffies == this_rq->last_load_update_tick) | |
2561 | return; | |
2562 | ||
2563 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2564 | this_rq->last_load_update_tick = curr_jiffies; | |
2565 | ||
2566 | __update_cpu_load(this_rq, load, pending_updates); | |
2567 | } | |
2568 | ||
5aaa0b7a PZ |
2569 | /* |
2570 | * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed. | |
2571 | */ | |
2572 | void update_cpu_load_nohz(void) | |
2573 | { | |
2574 | struct rq *this_rq = this_rq(); | |
2575 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); | |
2576 | unsigned long pending_updates; | |
2577 | ||
2578 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2579 | return; | |
2580 | ||
2581 | raw_spin_lock(&this_rq->lock); | |
2582 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2583 | if (pending_updates) { | |
2584 | this_rq->last_load_update_tick = curr_jiffies; | |
2585 | /* | |
2586 | * We were idle, this means load 0, the current load might be | |
2587 | * !0 due to remote wakeups and the sort. | |
2588 | */ | |
2589 | __update_cpu_load(this_rq, 0, pending_updates); | |
2590 | } | |
2591 | raw_spin_unlock(&this_rq->lock); | |
2592 | } | |
3451d024 | 2593 | #endif /* CONFIG_NO_HZ_COMMON */ |
5aaa0b7a | 2594 | |
556061b0 PZ |
2595 | /* |
2596 | * Called from scheduler_tick() | |
2597 | */ | |
fdf3e95d VP |
2598 | static void update_cpu_load_active(struct rq *this_rq) |
2599 | { | |
556061b0 | 2600 | /* |
5aaa0b7a | 2601 | * See the mess around update_idle_cpu_load() / update_cpu_load_nohz(). |
556061b0 PZ |
2602 | */ |
2603 | this_rq->last_load_update_tick = jiffies; | |
2604 | __update_cpu_load(this_rq, this_rq->load.weight, 1); | |
46cb4b7c | 2605 | |
74f5187a | 2606 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2607 | } |
2608 | ||
dd41f596 | 2609 | #ifdef CONFIG_SMP |
8a0be9ef | 2610 | |
46cb4b7c | 2611 | /* |
38022906 PZ |
2612 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2613 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2614 | */ |
38022906 | 2615 | void sched_exec(void) |
46cb4b7c | 2616 | { |
38022906 | 2617 | struct task_struct *p = current; |
1da177e4 | 2618 | unsigned long flags; |
0017d735 | 2619 | int dest_cpu; |
46cb4b7c | 2620 | |
8f42ced9 | 2621 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2622 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2623 | if (dest_cpu == smp_processor_id()) |
2624 | goto unlock; | |
38022906 | 2625 | |
8f42ced9 | 2626 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2627 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2628 | |
8f42ced9 PZ |
2629 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2630 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2631 | return; |
2632 | } | |
0017d735 | 2633 | unlock: |
8f42ced9 | 2634 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2635 | } |
dd41f596 | 2636 | |
1da177e4 LT |
2637 | #endif |
2638 | ||
1da177e4 | 2639 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2640 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2641 | |
2642 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2643 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2644 | |
2645 | /* | |
c5f8d995 | 2646 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2647 | * @p in case that task is currently running. |
c5f8d995 HS |
2648 | * |
2649 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2650 | */ |
c5f8d995 HS |
2651 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2652 | { | |
2653 | u64 ns = 0; | |
2654 | ||
2655 | if (task_current(rq, p)) { | |
2656 | update_rq_clock(rq); | |
305e6835 | 2657 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2658 | if ((s64)ns < 0) |
2659 | ns = 0; | |
2660 | } | |
2661 | ||
2662 | return ns; | |
2663 | } | |
2664 | ||
bb34d92f | 2665 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2666 | { |
1da177e4 | 2667 | unsigned long flags; |
41b86e9c | 2668 | struct rq *rq; |
bb34d92f | 2669 | u64 ns = 0; |
48f24c4d | 2670 | |
41b86e9c | 2671 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2672 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2673 | task_rq_unlock(rq, p, &flags); |
1508487e | 2674 | |
c5f8d995 HS |
2675 | return ns; |
2676 | } | |
f06febc9 | 2677 | |
c5f8d995 HS |
2678 | /* |
2679 | * Return accounted runtime for the task. | |
2680 | * In case the task is currently running, return the runtime plus current's | |
2681 | * pending runtime that have not been accounted yet. | |
2682 | */ | |
2683 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2684 | { | |
2685 | unsigned long flags; | |
2686 | struct rq *rq; | |
2687 | u64 ns = 0; | |
2688 | ||
2689 | rq = task_rq_lock(p, &flags); | |
2690 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2691 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2692 | |
2693 | return ns; | |
2694 | } | |
48f24c4d | 2695 | |
7835b98b CL |
2696 | /* |
2697 | * This function gets called by the timer code, with HZ frequency. | |
2698 | * We call it with interrupts disabled. | |
7835b98b CL |
2699 | */ |
2700 | void scheduler_tick(void) | |
2701 | { | |
7835b98b CL |
2702 | int cpu = smp_processor_id(); |
2703 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2704 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2705 | |
2706 | sched_clock_tick(); | |
dd41f596 | 2707 | |
05fa785c | 2708 | raw_spin_lock(&rq->lock); |
3e51f33f | 2709 | update_rq_clock(rq); |
fdf3e95d | 2710 | update_cpu_load_active(rq); |
fa85ae24 | 2711 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2712 | raw_spin_unlock(&rq->lock); |
7835b98b | 2713 | |
e9d2b064 | 2714 | perf_event_task_tick(); |
e220d2dc | 2715 | |
e418e1c2 | 2716 | #ifdef CONFIG_SMP |
6eb57e0d | 2717 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2718 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2719 | #endif |
1da177e4 LT |
2720 | } |
2721 | ||
132380a0 | 2722 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2723 | { |
2724 | if (in_lock_functions(addr)) { | |
2725 | addr = CALLER_ADDR2; | |
2726 | if (in_lock_functions(addr)) | |
2727 | addr = CALLER_ADDR3; | |
2728 | } | |
2729 | return addr; | |
2730 | } | |
1da177e4 | 2731 | |
7e49fcce SR |
2732 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2733 | defined(CONFIG_PREEMPT_TRACER)) | |
2734 | ||
43627582 | 2735 | void __kprobes add_preempt_count(int val) |
1da177e4 | 2736 | { |
6cd8a4bb | 2737 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2738 | /* |
2739 | * Underflow? | |
2740 | */ | |
9a11b49a IM |
2741 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2742 | return; | |
6cd8a4bb | 2743 | #endif |
1da177e4 | 2744 | preempt_count() += val; |
6cd8a4bb | 2745 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2746 | /* |
2747 | * Spinlock count overflowing soon? | |
2748 | */ | |
33859f7f MOS |
2749 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2750 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2751 | #endif |
2752 | if (preempt_count() == val) | |
2753 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2754 | } |
2755 | EXPORT_SYMBOL(add_preempt_count); | |
2756 | ||
43627582 | 2757 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 2758 | { |
6cd8a4bb | 2759 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2760 | /* |
2761 | * Underflow? | |
2762 | */ | |
01e3eb82 | 2763 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2764 | return; |
1da177e4 LT |
2765 | /* |
2766 | * Is the spinlock portion underflowing? | |
2767 | */ | |
9a11b49a IM |
2768 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2769 | !(preempt_count() & PREEMPT_MASK))) | |
2770 | return; | |
6cd8a4bb | 2771 | #endif |
9a11b49a | 2772 | |
6cd8a4bb SR |
2773 | if (preempt_count() == val) |
2774 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2775 | preempt_count() -= val; |
2776 | } | |
2777 | EXPORT_SYMBOL(sub_preempt_count); | |
2778 | ||
2779 | #endif | |
2780 | ||
2781 | /* | |
dd41f596 | 2782 | * Print scheduling while atomic bug: |
1da177e4 | 2783 | */ |
dd41f596 | 2784 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2785 | { |
664dfa65 DJ |
2786 | if (oops_in_progress) |
2787 | return; | |
2788 | ||
3df0fc5b PZ |
2789 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2790 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2791 | |
dd41f596 | 2792 | debug_show_held_locks(prev); |
e21f5b15 | 2793 | print_modules(); |
dd41f596 IM |
2794 | if (irqs_disabled()) |
2795 | print_irqtrace_events(prev); | |
6135fc1e | 2796 | dump_stack(); |
373d4d09 | 2797 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2798 | } |
1da177e4 | 2799 | |
dd41f596 IM |
2800 | /* |
2801 | * Various schedule()-time debugging checks and statistics: | |
2802 | */ | |
2803 | static inline void schedule_debug(struct task_struct *prev) | |
2804 | { | |
1da177e4 | 2805 | /* |
41a2d6cf | 2806 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
2807 | * schedule() atomically, we ignore that path for now. |
2808 | * Otherwise, whine if we are scheduling when we should not be. | |
2809 | */ | |
3f33a7ce | 2810 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 2811 | __schedule_bug(prev); |
b3fbab05 | 2812 | rcu_sleep_check(); |
dd41f596 | 2813 | |
1da177e4 LT |
2814 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2815 | ||
2d72376b | 2816 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2817 | } |
2818 | ||
6cecd084 | 2819 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2820 | { |
61eadef6 | 2821 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2822 | update_rq_clock(rq); |
6cecd084 | 2823 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2824 | } |
2825 | ||
dd41f596 IM |
2826 | /* |
2827 | * Pick up the highest-prio task: | |
2828 | */ | |
2829 | static inline struct task_struct * | |
b67802ea | 2830 | pick_next_task(struct rq *rq) |
dd41f596 | 2831 | { |
5522d5d5 | 2832 | const struct sched_class *class; |
dd41f596 | 2833 | struct task_struct *p; |
1da177e4 LT |
2834 | |
2835 | /* | |
dd41f596 IM |
2836 | * Optimization: we know that if all tasks are in |
2837 | * the fair class we can call that function directly: | |
1da177e4 | 2838 | */ |
953bfcd1 | 2839 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2840 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2841 | if (likely(p)) |
2842 | return p; | |
1da177e4 LT |
2843 | } |
2844 | ||
34f971f6 | 2845 | for_each_class(class) { |
fb8d4724 | 2846 | p = class->pick_next_task(rq); |
dd41f596 IM |
2847 | if (p) |
2848 | return p; | |
dd41f596 | 2849 | } |
34f971f6 PZ |
2850 | |
2851 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2852 | } |
1da177e4 | 2853 | |
dd41f596 | 2854 | /* |
c259e01a | 2855 | * __schedule() is the main scheduler function. |
edde96ea PE |
2856 | * |
2857 | * The main means of driving the scheduler and thus entering this function are: | |
2858 | * | |
2859 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2860 | * | |
2861 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2862 | * paths. For example, see arch/x86/entry_64.S. | |
2863 | * | |
2864 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2865 | * interrupt handler scheduler_tick(). | |
2866 | * | |
2867 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2868 | * task to the run-queue and that's it. | |
2869 | * | |
2870 | * Now, if the new task added to the run-queue preempts the current | |
2871 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2872 | * called on the nearest possible occasion: | |
2873 | * | |
2874 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2875 | * | |
2876 | * - in syscall or exception context, at the next outmost | |
2877 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2878 | * spin_unlock()!) | |
2879 | * | |
2880 | * - in IRQ context, return from interrupt-handler to | |
2881 | * preemptible context | |
2882 | * | |
2883 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2884 | * then at the next: | |
2885 | * | |
2886 | * - cond_resched() call | |
2887 | * - explicit schedule() call | |
2888 | * - return from syscall or exception to user-space | |
2889 | * - return from interrupt-handler to user-space | |
dd41f596 | 2890 | */ |
c259e01a | 2891 | static void __sched __schedule(void) |
dd41f596 IM |
2892 | { |
2893 | struct task_struct *prev, *next; | |
67ca7bde | 2894 | unsigned long *switch_count; |
dd41f596 | 2895 | struct rq *rq; |
31656519 | 2896 | int cpu; |
dd41f596 | 2897 | |
ff743345 PZ |
2898 | need_resched: |
2899 | preempt_disable(); | |
dd41f596 IM |
2900 | cpu = smp_processor_id(); |
2901 | rq = cpu_rq(cpu); | |
25502a6c | 2902 | rcu_note_context_switch(cpu); |
dd41f596 | 2903 | prev = rq->curr; |
dd41f596 | 2904 | |
dd41f596 | 2905 | schedule_debug(prev); |
1da177e4 | 2906 | |
31656519 | 2907 | if (sched_feat(HRTICK)) |
f333fdc9 | 2908 | hrtick_clear(rq); |
8f4d37ec | 2909 | |
05fa785c | 2910 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2911 | |
246d86b5 | 2912 | switch_count = &prev->nivcsw; |
1da177e4 | 2913 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2914 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2915 | prev->state = TASK_RUNNING; |
21aa9af0 | 2916 | } else { |
2acca55e PZ |
2917 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2918 | prev->on_rq = 0; | |
2919 | ||
21aa9af0 | 2920 | /* |
2acca55e PZ |
2921 | * If a worker went to sleep, notify and ask workqueue |
2922 | * whether it wants to wake up a task to maintain | |
2923 | * concurrency. | |
21aa9af0 TH |
2924 | */ |
2925 | if (prev->flags & PF_WQ_WORKER) { | |
2926 | struct task_struct *to_wakeup; | |
2927 | ||
2928 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2929 | if (to_wakeup) | |
2930 | try_to_wake_up_local(to_wakeup); | |
2931 | } | |
21aa9af0 | 2932 | } |
dd41f596 | 2933 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2934 | } |
2935 | ||
3f029d3c | 2936 | pre_schedule(rq, prev); |
f65eda4f | 2937 | |
dd41f596 | 2938 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2939 | idle_balance(cpu, rq); |
1da177e4 | 2940 | |
df1c99d4 | 2941 | put_prev_task(rq, prev); |
b67802ea | 2942 | next = pick_next_task(rq); |
f26f9aff MG |
2943 | clear_tsk_need_resched(prev); |
2944 | rq->skip_clock_update = 0; | |
1da177e4 | 2945 | |
1da177e4 | 2946 | if (likely(prev != next)) { |
1da177e4 LT |
2947 | rq->nr_switches++; |
2948 | rq->curr = next; | |
2949 | ++*switch_count; | |
2950 | ||
dd41f596 | 2951 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2952 | /* |
246d86b5 ON |
2953 | * The context switch have flipped the stack from under us |
2954 | * and restored the local variables which were saved when | |
2955 | * this task called schedule() in the past. prev == current | |
2956 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2957 | */ |
2958 | cpu = smp_processor_id(); | |
2959 | rq = cpu_rq(cpu); | |
1da177e4 | 2960 | } else |
05fa785c | 2961 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2962 | |
3f029d3c | 2963 | post_schedule(rq); |
1da177e4 | 2964 | |
ba74c144 | 2965 | sched_preempt_enable_no_resched(); |
ff743345 | 2966 | if (need_resched()) |
1da177e4 LT |
2967 | goto need_resched; |
2968 | } | |
c259e01a | 2969 | |
9c40cef2 TG |
2970 | static inline void sched_submit_work(struct task_struct *tsk) |
2971 | { | |
3c7d5184 | 2972 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2973 | return; |
2974 | /* | |
2975 | * If we are going to sleep and we have plugged IO queued, | |
2976 | * make sure to submit it to avoid deadlocks. | |
2977 | */ | |
2978 | if (blk_needs_flush_plug(tsk)) | |
2979 | blk_schedule_flush_plug(tsk); | |
2980 | } | |
2981 | ||
6ebbe7a0 | 2982 | asmlinkage void __sched schedule(void) |
c259e01a | 2983 | { |
9c40cef2 TG |
2984 | struct task_struct *tsk = current; |
2985 | ||
2986 | sched_submit_work(tsk); | |
c259e01a TG |
2987 | __schedule(); |
2988 | } | |
1da177e4 LT |
2989 | EXPORT_SYMBOL(schedule); |
2990 | ||
91d1aa43 | 2991 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2992 | asmlinkage void __sched schedule_user(void) |
2993 | { | |
2994 | /* | |
2995 | * If we come here after a random call to set_need_resched(), | |
2996 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2997 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2998 | * we find a better solution. | |
2999 | */ | |
91d1aa43 | 3000 | user_exit(); |
20ab65e3 | 3001 | schedule(); |
91d1aa43 | 3002 | user_enter(); |
20ab65e3 FW |
3003 | } |
3004 | #endif | |
3005 | ||
c5491ea7 TG |
3006 | /** |
3007 | * schedule_preempt_disabled - called with preemption disabled | |
3008 | * | |
3009 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3010 | */ | |
3011 | void __sched schedule_preempt_disabled(void) | |
3012 | { | |
ba74c144 | 3013 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3014 | schedule(); |
3015 | preempt_disable(); | |
3016 | } | |
3017 | ||
c08f7829 | 3018 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3019 | |
c6eb3dda PZ |
3020 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3021 | { | |
c6eb3dda | 3022 | if (lock->owner != owner) |
307bf980 | 3023 | return false; |
0d66bf6d PZ |
3024 | |
3025 | /* | |
c6eb3dda PZ |
3026 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3027 | * lock->owner still matches owner, if that fails, owner might | |
3028 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3029 | * ensures the memory stays valid. | |
0d66bf6d | 3030 | */ |
c6eb3dda | 3031 | barrier(); |
0d66bf6d | 3032 | |
307bf980 | 3033 | return owner->on_cpu; |
c6eb3dda | 3034 | } |
0d66bf6d | 3035 | |
c6eb3dda PZ |
3036 | /* |
3037 | * Look out! "owner" is an entirely speculative pointer | |
3038 | * access and not reliable. | |
3039 | */ | |
3040 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3041 | { | |
3042 | if (!sched_feat(OWNER_SPIN)) | |
3043 | return 0; | |
0d66bf6d | 3044 | |
307bf980 | 3045 | rcu_read_lock(); |
c6eb3dda PZ |
3046 | while (owner_running(lock, owner)) { |
3047 | if (need_resched()) | |
307bf980 | 3048 | break; |
0d66bf6d | 3049 | |
335d7afb | 3050 | arch_mutex_cpu_relax(); |
0d66bf6d | 3051 | } |
307bf980 | 3052 | rcu_read_unlock(); |
4b402210 | 3053 | |
c6eb3dda | 3054 | /* |
307bf980 TG |
3055 | * We break out the loop above on need_resched() and when the |
3056 | * owner changed, which is a sign for heavy contention. Return | |
3057 | * success only when lock->owner is NULL. | |
c6eb3dda | 3058 | */ |
307bf980 | 3059 | return lock->owner == NULL; |
0d66bf6d PZ |
3060 | } |
3061 | #endif | |
3062 | ||
1da177e4 LT |
3063 | #ifdef CONFIG_PREEMPT |
3064 | /* | |
2ed6e34f | 3065 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3066 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3067 | * occur there and call schedule directly. |
3068 | */ | |
d1f74e20 | 3069 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3070 | { |
3071 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3072 | |
1da177e4 LT |
3073 | /* |
3074 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3075 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3076 | */ |
beed33a8 | 3077 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3078 | return; |
3079 | ||
3a5c359a | 3080 | do { |
d1f74e20 | 3081 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3082 | __schedule(); |
d1f74e20 | 3083 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3084 | |
3a5c359a AK |
3085 | /* |
3086 | * Check again in case we missed a preemption opportunity | |
3087 | * between schedule and now. | |
3088 | */ | |
3089 | barrier(); | |
5ed0cec0 | 3090 | } while (need_resched()); |
1da177e4 | 3091 | } |
1da177e4 LT |
3092 | EXPORT_SYMBOL(preempt_schedule); |
3093 | ||
3094 | /* | |
2ed6e34f | 3095 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3096 | * off of irq context. |
3097 | * Note, that this is called and return with irqs disabled. This will | |
3098 | * protect us against recursive calling from irq. | |
3099 | */ | |
3100 | asmlinkage void __sched preempt_schedule_irq(void) | |
3101 | { | |
3102 | struct thread_info *ti = current_thread_info(); | |
b22366cd | 3103 | enum ctx_state prev_state; |
6478d880 | 3104 | |
2ed6e34f | 3105 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3106 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3107 | ||
b22366cd FW |
3108 | prev_state = exception_enter(); |
3109 | ||
3a5c359a AK |
3110 | do { |
3111 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3112 | local_irq_enable(); |
c259e01a | 3113 | __schedule(); |
3a5c359a | 3114 | local_irq_disable(); |
3a5c359a | 3115 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3116 | |
3a5c359a AK |
3117 | /* |
3118 | * Check again in case we missed a preemption opportunity | |
3119 | * between schedule and now. | |
3120 | */ | |
3121 | barrier(); | |
5ed0cec0 | 3122 | } while (need_resched()); |
b22366cd FW |
3123 | |
3124 | exception_exit(prev_state); | |
1da177e4 LT |
3125 | } |
3126 | ||
3127 | #endif /* CONFIG_PREEMPT */ | |
3128 | ||
63859d4f | 3129 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3130 | void *key) |
1da177e4 | 3131 | { |
63859d4f | 3132 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3133 | } |
1da177e4 LT |
3134 | EXPORT_SYMBOL(default_wake_function); |
3135 | ||
3136 | /* | |
41a2d6cf IM |
3137 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3138 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3139 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3140 | * | |
3141 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3142 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3143 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3144 | */ | |
78ddb08f | 3145 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3146 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3147 | { |
2e45874c | 3148 | wait_queue_t *curr, *next; |
1da177e4 | 3149 | |
2e45874c | 3150 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3151 | unsigned flags = curr->flags; |
3152 | ||
63859d4f | 3153 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3154 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3155 | break; |
3156 | } | |
3157 | } | |
3158 | ||
3159 | /** | |
3160 | * __wake_up - wake up threads blocked on a waitqueue. | |
3161 | * @q: the waitqueue | |
3162 | * @mode: which threads | |
3163 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3164 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3165 | * |
3166 | * It may be assumed that this function implies a write memory barrier before | |
3167 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3168 | */ |
7ad5b3a5 | 3169 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3170 | int nr_exclusive, void *key) |
1da177e4 LT |
3171 | { |
3172 | unsigned long flags; | |
3173 | ||
3174 | spin_lock_irqsave(&q->lock, flags); | |
3175 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3176 | spin_unlock_irqrestore(&q->lock, flags); | |
3177 | } | |
1da177e4 LT |
3178 | EXPORT_SYMBOL(__wake_up); |
3179 | ||
3180 | /* | |
3181 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3182 | */ | |
63b20011 | 3183 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3184 | { |
63b20011 | 3185 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3186 | } |
22c43c81 | 3187 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3188 | |
4ede816a DL |
3189 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3190 | { | |
3191 | __wake_up_common(q, mode, 1, 0, key); | |
3192 | } | |
bf294b41 | 3193 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3194 | |
1da177e4 | 3195 | /** |
4ede816a | 3196 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3197 | * @q: the waitqueue |
3198 | * @mode: which threads | |
3199 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3200 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3201 | * |
3202 | * The sync wakeup differs that the waker knows that it will schedule | |
3203 | * away soon, so while the target thread will be woken up, it will not | |
3204 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3205 | * with each other. This can prevent needless bouncing between CPUs. | |
3206 | * | |
3207 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3208 | * |
3209 | * It may be assumed that this function implies a write memory barrier before | |
3210 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3211 | */ |
4ede816a DL |
3212 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3213 | int nr_exclusive, void *key) | |
1da177e4 LT |
3214 | { |
3215 | unsigned long flags; | |
7d478721 | 3216 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3217 | |
3218 | if (unlikely(!q)) | |
3219 | return; | |
3220 | ||
3221 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3222 | wake_flags = 0; |
1da177e4 LT |
3223 | |
3224 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3225 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3226 | spin_unlock_irqrestore(&q->lock, flags); |
3227 | } | |
4ede816a DL |
3228 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3229 | ||
3230 | /* | |
3231 | * __wake_up_sync - see __wake_up_sync_key() | |
3232 | */ | |
3233 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3234 | { | |
3235 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3236 | } | |
1da177e4 LT |
3237 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3238 | ||
65eb3dc6 KD |
3239 | /** |
3240 | * complete: - signals a single thread waiting on this completion | |
3241 | * @x: holds the state of this particular completion | |
3242 | * | |
3243 | * This will wake up a single thread waiting on this completion. Threads will be | |
3244 | * awakened in the same order in which they were queued. | |
3245 | * | |
3246 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3247 | * |
3248 | * It may be assumed that this function implies a write memory barrier before | |
3249 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3250 | */ |
b15136e9 | 3251 | void complete(struct completion *x) |
1da177e4 LT |
3252 | { |
3253 | unsigned long flags; | |
3254 | ||
3255 | spin_lock_irqsave(&x->wait.lock, flags); | |
3256 | x->done++; | |
d9514f6c | 3257 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3258 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3259 | } | |
3260 | EXPORT_SYMBOL(complete); | |
3261 | ||
65eb3dc6 KD |
3262 | /** |
3263 | * complete_all: - signals all threads waiting on this completion | |
3264 | * @x: holds the state of this particular completion | |
3265 | * | |
3266 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3267 | * |
3268 | * It may be assumed that this function implies a write memory barrier before | |
3269 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3270 | */ |
b15136e9 | 3271 | void complete_all(struct completion *x) |
1da177e4 LT |
3272 | { |
3273 | unsigned long flags; | |
3274 | ||
3275 | spin_lock_irqsave(&x->wait.lock, flags); | |
3276 | x->done += UINT_MAX/2; | |
d9514f6c | 3277 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3278 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3279 | } | |
3280 | EXPORT_SYMBOL(complete_all); | |
3281 | ||
8cbbe86d | 3282 | static inline long __sched |
686855f5 VD |
3283 | do_wait_for_common(struct completion *x, |
3284 | long (*action)(long), long timeout, int state) | |
1da177e4 | 3285 | { |
1da177e4 LT |
3286 | if (!x->done) { |
3287 | DECLARE_WAITQUEUE(wait, current); | |
3288 | ||
a93d2f17 | 3289 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3290 | do { |
94d3d824 | 3291 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3292 | timeout = -ERESTARTSYS; |
3293 | break; | |
8cbbe86d AK |
3294 | } |
3295 | __set_current_state(state); | |
1da177e4 | 3296 | spin_unlock_irq(&x->wait.lock); |
686855f5 | 3297 | timeout = action(timeout); |
1da177e4 | 3298 | spin_lock_irq(&x->wait.lock); |
ea71a546 | 3299 | } while (!x->done && timeout); |
1da177e4 | 3300 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3301 | if (!x->done) |
3302 | return timeout; | |
1da177e4 LT |
3303 | } |
3304 | x->done--; | |
ea71a546 | 3305 | return timeout ?: 1; |
1da177e4 | 3306 | } |
1da177e4 | 3307 | |
686855f5 VD |
3308 | static inline long __sched |
3309 | __wait_for_common(struct completion *x, | |
3310 | long (*action)(long), long timeout, int state) | |
1da177e4 | 3311 | { |
1da177e4 LT |
3312 | might_sleep(); |
3313 | ||
3314 | spin_lock_irq(&x->wait.lock); | |
686855f5 | 3315 | timeout = do_wait_for_common(x, action, timeout, state); |
1da177e4 | 3316 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3317 | return timeout; |
3318 | } | |
1da177e4 | 3319 | |
686855f5 VD |
3320 | static long __sched |
3321 | wait_for_common(struct completion *x, long timeout, int state) | |
3322 | { | |
3323 | return __wait_for_common(x, schedule_timeout, timeout, state); | |
3324 | } | |
3325 | ||
3326 | static long __sched | |
3327 | wait_for_common_io(struct completion *x, long timeout, int state) | |
3328 | { | |
3329 | return __wait_for_common(x, io_schedule_timeout, timeout, state); | |
3330 | } | |
3331 | ||
65eb3dc6 KD |
3332 | /** |
3333 | * wait_for_completion: - waits for completion of a task | |
3334 | * @x: holds the state of this particular completion | |
3335 | * | |
3336 | * This waits to be signaled for completion of a specific task. It is NOT | |
3337 | * interruptible and there is no timeout. | |
3338 | * | |
3339 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3340 | * and interrupt capability. Also see complete(). | |
3341 | */ | |
b15136e9 | 3342 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3343 | { |
3344 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3345 | } |
8cbbe86d | 3346 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3347 | |
65eb3dc6 KD |
3348 | /** |
3349 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3350 | * @x: holds the state of this particular completion | |
3351 | * @timeout: timeout value in jiffies | |
3352 | * | |
3353 | * This waits for either a completion of a specific task to be signaled or for a | |
3354 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3355 | * interruptible. | |
c6dc7f05 BF |
3356 | * |
3357 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3358 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3359 | */ |
b15136e9 | 3360 | unsigned long __sched |
8cbbe86d | 3361 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3362 | { |
8cbbe86d | 3363 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3364 | } |
8cbbe86d | 3365 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3366 | |
686855f5 VD |
3367 | /** |
3368 | * wait_for_completion_io: - waits for completion of a task | |
3369 | * @x: holds the state of this particular completion | |
3370 | * | |
3371 | * This waits to be signaled for completion of a specific task. It is NOT | |
3372 | * interruptible and there is no timeout. The caller is accounted as waiting | |
3373 | * for IO. | |
3374 | */ | |
3375 | void __sched wait_for_completion_io(struct completion *x) | |
3376 | { | |
3377 | wait_for_common_io(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
3378 | } | |
3379 | EXPORT_SYMBOL(wait_for_completion_io); | |
3380 | ||
3381 | /** | |
3382 | * wait_for_completion_io_timeout: - waits for completion of a task (w/timeout) | |
3383 | * @x: holds the state of this particular completion | |
3384 | * @timeout: timeout value in jiffies | |
3385 | * | |
3386 | * This waits for either a completion of a specific task to be signaled or for a | |
3387 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3388 | * interruptible. The caller is accounted as waiting for IO. | |
3389 | * | |
3390 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3391 | * jiffies left till timeout) if completed. | |
3392 | */ | |
3393 | unsigned long __sched | |
3394 | wait_for_completion_io_timeout(struct completion *x, unsigned long timeout) | |
3395 | { | |
3396 | return wait_for_common_io(x, timeout, TASK_UNINTERRUPTIBLE); | |
3397 | } | |
3398 | EXPORT_SYMBOL(wait_for_completion_io_timeout); | |
3399 | ||
65eb3dc6 KD |
3400 | /** |
3401 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3402 | * @x: holds the state of this particular completion | |
3403 | * | |
3404 | * This waits for completion of a specific task to be signaled. It is | |
3405 | * interruptible. | |
c6dc7f05 BF |
3406 | * |
3407 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3408 | */ |
8cbbe86d | 3409 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3410 | { |
51e97990 AK |
3411 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3412 | if (t == -ERESTARTSYS) | |
3413 | return t; | |
3414 | return 0; | |
0fec171c | 3415 | } |
8cbbe86d | 3416 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3417 | |
65eb3dc6 KD |
3418 | /** |
3419 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3420 | * @x: holds the state of this particular completion | |
3421 | * @timeout: timeout value in jiffies | |
3422 | * | |
3423 | * This waits for either a completion of a specific task to be signaled or for a | |
3424 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3425 | * |
3426 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3427 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3428 | */ |
6bf41237 | 3429 | long __sched |
8cbbe86d AK |
3430 | wait_for_completion_interruptible_timeout(struct completion *x, |
3431 | unsigned long timeout) | |
0fec171c | 3432 | { |
8cbbe86d | 3433 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3434 | } |
8cbbe86d | 3435 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3436 | |
65eb3dc6 KD |
3437 | /** |
3438 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3439 | * @x: holds the state of this particular completion | |
3440 | * | |
3441 | * This waits to be signaled for completion of a specific task. It can be | |
3442 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3443 | * |
3444 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3445 | */ |
009e577e MW |
3446 | int __sched wait_for_completion_killable(struct completion *x) |
3447 | { | |
3448 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3449 | if (t == -ERESTARTSYS) | |
3450 | return t; | |
3451 | return 0; | |
3452 | } | |
3453 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3454 | ||
0aa12fb4 SW |
3455 | /** |
3456 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3457 | * @x: holds the state of this particular completion | |
3458 | * @timeout: timeout value in jiffies | |
3459 | * | |
3460 | * This waits for either a completion of a specific task to be | |
3461 | * signaled or for a specified timeout to expire. It can be | |
3462 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3463 | * |
3464 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3465 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3466 | */ |
6bf41237 | 3467 | long __sched |
0aa12fb4 SW |
3468 | wait_for_completion_killable_timeout(struct completion *x, |
3469 | unsigned long timeout) | |
3470 | { | |
3471 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3472 | } | |
3473 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3474 | ||
be4de352 DC |
3475 | /** |
3476 | * try_wait_for_completion - try to decrement a completion without blocking | |
3477 | * @x: completion structure | |
3478 | * | |
3479 | * Returns: 0 if a decrement cannot be done without blocking | |
3480 | * 1 if a decrement succeeded. | |
3481 | * | |
3482 | * If a completion is being used as a counting completion, | |
3483 | * attempt to decrement the counter without blocking. This | |
3484 | * enables us to avoid waiting if the resource the completion | |
3485 | * is protecting is not available. | |
3486 | */ | |
3487 | bool try_wait_for_completion(struct completion *x) | |
3488 | { | |
7539a3b3 | 3489 | unsigned long flags; |
be4de352 DC |
3490 | int ret = 1; |
3491 | ||
7539a3b3 | 3492 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3493 | if (!x->done) |
3494 | ret = 0; | |
3495 | else | |
3496 | x->done--; | |
7539a3b3 | 3497 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3498 | return ret; |
3499 | } | |
3500 | EXPORT_SYMBOL(try_wait_for_completion); | |
3501 | ||
3502 | /** | |
3503 | * completion_done - Test to see if a completion has any waiters | |
3504 | * @x: completion structure | |
3505 | * | |
3506 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3507 | * 1 if there are no waiters. | |
3508 | * | |
3509 | */ | |
3510 | bool completion_done(struct completion *x) | |
3511 | { | |
7539a3b3 | 3512 | unsigned long flags; |
be4de352 DC |
3513 | int ret = 1; |
3514 | ||
7539a3b3 | 3515 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3516 | if (!x->done) |
3517 | ret = 0; | |
7539a3b3 | 3518 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3519 | return ret; |
3520 | } | |
3521 | EXPORT_SYMBOL(completion_done); | |
3522 | ||
8cbbe86d AK |
3523 | static long __sched |
3524 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3525 | { |
0fec171c IM |
3526 | unsigned long flags; |
3527 | wait_queue_t wait; | |
3528 | ||
3529 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3530 | |
8cbbe86d | 3531 | __set_current_state(state); |
1da177e4 | 3532 | |
8cbbe86d AK |
3533 | spin_lock_irqsave(&q->lock, flags); |
3534 | __add_wait_queue(q, &wait); | |
3535 | spin_unlock(&q->lock); | |
3536 | timeout = schedule_timeout(timeout); | |
3537 | spin_lock_irq(&q->lock); | |
3538 | __remove_wait_queue(q, &wait); | |
3539 | spin_unlock_irqrestore(&q->lock, flags); | |
3540 | ||
3541 | return timeout; | |
3542 | } | |
3543 | ||
3544 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3545 | { | |
3546 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3547 | } |
1da177e4 LT |
3548 | EXPORT_SYMBOL(interruptible_sleep_on); |
3549 | ||
0fec171c | 3550 | long __sched |
95cdf3b7 | 3551 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3552 | { |
8cbbe86d | 3553 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3554 | } |
1da177e4 LT |
3555 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3556 | ||
0fec171c | 3557 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3558 | { |
8cbbe86d | 3559 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3560 | } |
1da177e4 LT |
3561 | EXPORT_SYMBOL(sleep_on); |
3562 | ||
0fec171c | 3563 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3564 | { |
8cbbe86d | 3565 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3566 | } |
1da177e4 LT |
3567 | EXPORT_SYMBOL(sleep_on_timeout); |
3568 | ||
b29739f9 IM |
3569 | #ifdef CONFIG_RT_MUTEXES |
3570 | ||
3571 | /* | |
3572 | * rt_mutex_setprio - set the current priority of a task | |
3573 | * @p: task | |
3574 | * @prio: prio value (kernel-internal form) | |
3575 | * | |
3576 | * This function changes the 'effective' priority of a task. It does | |
3577 | * not touch ->normal_prio like __setscheduler(). | |
3578 | * | |
3579 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3580 | */ | |
36c8b586 | 3581 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3582 | { |
83b699ed | 3583 | int oldprio, on_rq, running; |
70b97a7f | 3584 | struct rq *rq; |
83ab0aa0 | 3585 | const struct sched_class *prev_class; |
b29739f9 IM |
3586 | |
3587 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3588 | ||
0122ec5b | 3589 | rq = __task_rq_lock(p); |
b29739f9 | 3590 | |
1c4dd99b TG |
3591 | /* |
3592 | * Idle task boosting is a nono in general. There is one | |
3593 | * exception, when PREEMPT_RT and NOHZ is active: | |
3594 | * | |
3595 | * The idle task calls get_next_timer_interrupt() and holds | |
3596 | * the timer wheel base->lock on the CPU and another CPU wants | |
3597 | * to access the timer (probably to cancel it). We can safely | |
3598 | * ignore the boosting request, as the idle CPU runs this code | |
3599 | * with interrupts disabled and will complete the lock | |
3600 | * protected section without being interrupted. So there is no | |
3601 | * real need to boost. | |
3602 | */ | |
3603 | if (unlikely(p == rq->idle)) { | |
3604 | WARN_ON(p != rq->curr); | |
3605 | WARN_ON(p->pi_blocked_on); | |
3606 | goto out_unlock; | |
3607 | } | |
3608 | ||
a8027073 | 3609 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3610 | oldprio = p->prio; |
83ab0aa0 | 3611 | prev_class = p->sched_class; |
fd2f4419 | 3612 | on_rq = p->on_rq; |
051a1d1a | 3613 | running = task_current(rq, p); |
0e1f3483 | 3614 | if (on_rq) |
69be72c1 | 3615 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3616 | if (running) |
3617 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3618 | |
3619 | if (rt_prio(prio)) | |
3620 | p->sched_class = &rt_sched_class; | |
3621 | else | |
3622 | p->sched_class = &fair_sched_class; | |
3623 | ||
b29739f9 IM |
3624 | p->prio = prio; |
3625 | ||
0e1f3483 HS |
3626 | if (running) |
3627 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3628 | if (on_rq) |
371fd7e7 | 3629 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3630 | |
da7a735e | 3631 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3632 | out_unlock: |
0122ec5b | 3633 | __task_rq_unlock(rq); |
b29739f9 | 3634 | } |
b29739f9 | 3635 | #endif |
36c8b586 | 3636 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3637 | { |
dd41f596 | 3638 | int old_prio, delta, on_rq; |
1da177e4 | 3639 | unsigned long flags; |
70b97a7f | 3640 | struct rq *rq; |
1da177e4 LT |
3641 | |
3642 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3643 | return; | |
3644 | /* | |
3645 | * We have to be careful, if called from sys_setpriority(), | |
3646 | * the task might be in the middle of scheduling on another CPU. | |
3647 | */ | |
3648 | rq = task_rq_lock(p, &flags); | |
3649 | /* | |
3650 | * The RT priorities are set via sched_setscheduler(), but we still | |
3651 | * allow the 'normal' nice value to be set - but as expected | |
3652 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3653 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3654 | */ |
e05606d3 | 3655 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3656 | p->static_prio = NICE_TO_PRIO(nice); |
3657 | goto out_unlock; | |
3658 | } | |
fd2f4419 | 3659 | on_rq = p->on_rq; |
c09595f6 | 3660 | if (on_rq) |
69be72c1 | 3661 | dequeue_task(rq, p, 0); |
1da177e4 | 3662 | |
1da177e4 | 3663 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3664 | set_load_weight(p); |
b29739f9 IM |
3665 | old_prio = p->prio; |
3666 | p->prio = effective_prio(p); | |
3667 | delta = p->prio - old_prio; | |
1da177e4 | 3668 | |
dd41f596 | 3669 | if (on_rq) { |
371fd7e7 | 3670 | enqueue_task(rq, p, 0); |
1da177e4 | 3671 | /* |
d5f9f942 AM |
3672 | * If the task increased its priority or is running and |
3673 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3674 | */ |
d5f9f942 | 3675 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3676 | resched_task(rq->curr); |
3677 | } | |
3678 | out_unlock: | |
0122ec5b | 3679 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3680 | } |
1da177e4 LT |
3681 | EXPORT_SYMBOL(set_user_nice); |
3682 | ||
e43379f1 MM |
3683 | /* |
3684 | * can_nice - check if a task can reduce its nice value | |
3685 | * @p: task | |
3686 | * @nice: nice value | |
3687 | */ | |
36c8b586 | 3688 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3689 | { |
024f4747 MM |
3690 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3691 | int nice_rlim = 20 - nice; | |
48f24c4d | 3692 | |
78d7d407 | 3693 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3694 | capable(CAP_SYS_NICE)); |
3695 | } | |
3696 | ||
1da177e4 LT |
3697 | #ifdef __ARCH_WANT_SYS_NICE |
3698 | ||
3699 | /* | |
3700 | * sys_nice - change the priority of the current process. | |
3701 | * @increment: priority increment | |
3702 | * | |
3703 | * sys_setpriority is a more generic, but much slower function that | |
3704 | * does similar things. | |
3705 | */ | |
5add95d4 | 3706 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3707 | { |
48f24c4d | 3708 | long nice, retval; |
1da177e4 LT |
3709 | |
3710 | /* | |
3711 | * Setpriority might change our priority at the same moment. | |
3712 | * We don't have to worry. Conceptually one call occurs first | |
3713 | * and we have a single winner. | |
3714 | */ | |
e43379f1 MM |
3715 | if (increment < -40) |
3716 | increment = -40; | |
1da177e4 LT |
3717 | if (increment > 40) |
3718 | increment = 40; | |
3719 | ||
2b8f836f | 3720 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3721 | if (nice < -20) |
3722 | nice = -20; | |
3723 | if (nice > 19) | |
3724 | nice = 19; | |
3725 | ||
e43379f1 MM |
3726 | if (increment < 0 && !can_nice(current, nice)) |
3727 | return -EPERM; | |
3728 | ||
1da177e4 LT |
3729 | retval = security_task_setnice(current, nice); |
3730 | if (retval) | |
3731 | return retval; | |
3732 | ||
3733 | set_user_nice(current, nice); | |
3734 | return 0; | |
3735 | } | |
3736 | ||
3737 | #endif | |
3738 | ||
3739 | /** | |
3740 | * task_prio - return the priority value of a given task. | |
3741 | * @p: the task in question. | |
3742 | * | |
3743 | * This is the priority value as seen by users in /proc. | |
3744 | * RT tasks are offset by -200. Normal tasks are centered | |
3745 | * around 0, value goes from -16 to +15. | |
3746 | */ | |
36c8b586 | 3747 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3748 | { |
3749 | return p->prio - MAX_RT_PRIO; | |
3750 | } | |
3751 | ||
3752 | /** | |
3753 | * task_nice - return the nice value of a given task. | |
3754 | * @p: the task in question. | |
3755 | */ | |
36c8b586 | 3756 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3757 | { |
3758 | return TASK_NICE(p); | |
3759 | } | |
150d8bed | 3760 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3761 | |
3762 | /** | |
3763 | * idle_cpu - is a given cpu idle currently? | |
3764 | * @cpu: the processor in question. | |
3765 | */ | |
3766 | int idle_cpu(int cpu) | |
3767 | { | |
908a3283 TG |
3768 | struct rq *rq = cpu_rq(cpu); |
3769 | ||
3770 | if (rq->curr != rq->idle) | |
3771 | return 0; | |
3772 | ||
3773 | if (rq->nr_running) | |
3774 | return 0; | |
3775 | ||
3776 | #ifdef CONFIG_SMP | |
3777 | if (!llist_empty(&rq->wake_list)) | |
3778 | return 0; | |
3779 | #endif | |
3780 | ||
3781 | return 1; | |
1da177e4 LT |
3782 | } |
3783 | ||
1da177e4 LT |
3784 | /** |
3785 | * idle_task - return the idle task for a given cpu. | |
3786 | * @cpu: the processor in question. | |
3787 | */ | |
36c8b586 | 3788 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3789 | { |
3790 | return cpu_rq(cpu)->idle; | |
3791 | } | |
3792 | ||
3793 | /** | |
3794 | * find_process_by_pid - find a process with a matching PID value. | |
3795 | * @pid: the pid in question. | |
3796 | */ | |
a9957449 | 3797 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3798 | { |
228ebcbe | 3799 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3800 | } |
3801 | ||
3802 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3803 | static void |
3804 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3805 | { |
1da177e4 LT |
3806 | p->policy = policy; |
3807 | p->rt_priority = prio; | |
b29739f9 IM |
3808 | p->normal_prio = normal_prio(p); |
3809 | /* we are holding p->pi_lock already */ | |
3810 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3811 | if (rt_prio(p->prio)) |
3812 | p->sched_class = &rt_sched_class; | |
3813 | else | |
3814 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3815 | set_load_weight(p); |
1da177e4 LT |
3816 | } |
3817 | ||
c69e8d9c DH |
3818 | /* |
3819 | * check the target process has a UID that matches the current process's | |
3820 | */ | |
3821 | static bool check_same_owner(struct task_struct *p) | |
3822 | { | |
3823 | const struct cred *cred = current_cred(), *pcred; | |
3824 | bool match; | |
3825 | ||
3826 | rcu_read_lock(); | |
3827 | pcred = __task_cred(p); | |
9c806aa0 EB |
3828 | match = (uid_eq(cred->euid, pcred->euid) || |
3829 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3830 | rcu_read_unlock(); |
3831 | return match; | |
3832 | } | |
3833 | ||
961ccddd | 3834 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3835 | const struct sched_param *param, bool user) |
1da177e4 | 3836 | { |
83b699ed | 3837 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3838 | unsigned long flags; |
83ab0aa0 | 3839 | const struct sched_class *prev_class; |
70b97a7f | 3840 | struct rq *rq; |
ca94c442 | 3841 | int reset_on_fork; |
1da177e4 | 3842 | |
66e5393a SR |
3843 | /* may grab non-irq protected spin_locks */ |
3844 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3845 | recheck: |
3846 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3847 | if (policy < 0) { |
3848 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3849 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3850 | } else { |
3851 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3852 | policy &= ~SCHED_RESET_ON_FORK; | |
3853 | ||
3854 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3855 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3856 | policy != SCHED_IDLE) | |
3857 | return -EINVAL; | |
3858 | } | |
3859 | ||
1da177e4 LT |
3860 | /* |
3861 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3862 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3863 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3864 | */ |
3865 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3866 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3867 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3868 | return -EINVAL; |
e05606d3 | 3869 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3870 | return -EINVAL; |
3871 | ||
37e4ab3f OC |
3872 | /* |
3873 | * Allow unprivileged RT tasks to decrease priority: | |
3874 | */ | |
961ccddd | 3875 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3876 | if (rt_policy(policy)) { |
a44702e8 ON |
3877 | unsigned long rlim_rtprio = |
3878 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3879 | |
3880 | /* can't set/change the rt policy */ | |
3881 | if (policy != p->policy && !rlim_rtprio) | |
3882 | return -EPERM; | |
3883 | ||
3884 | /* can't increase priority */ | |
3885 | if (param->sched_priority > p->rt_priority && | |
3886 | param->sched_priority > rlim_rtprio) | |
3887 | return -EPERM; | |
3888 | } | |
c02aa73b | 3889 | |
dd41f596 | 3890 | /* |
c02aa73b DH |
3891 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3892 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3893 | */ |
c02aa73b DH |
3894 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3895 | if (!can_nice(p, TASK_NICE(p))) | |
3896 | return -EPERM; | |
3897 | } | |
5fe1d75f | 3898 | |
37e4ab3f | 3899 | /* can't change other user's priorities */ |
c69e8d9c | 3900 | if (!check_same_owner(p)) |
37e4ab3f | 3901 | return -EPERM; |
ca94c442 LP |
3902 | |
3903 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3904 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3905 | return -EPERM; | |
37e4ab3f | 3906 | } |
1da177e4 | 3907 | |
725aad24 | 3908 | if (user) { |
b0ae1981 | 3909 | retval = security_task_setscheduler(p); |
725aad24 JF |
3910 | if (retval) |
3911 | return retval; | |
3912 | } | |
3913 | ||
b29739f9 IM |
3914 | /* |
3915 | * make sure no PI-waiters arrive (or leave) while we are | |
3916 | * changing the priority of the task: | |
0122ec5b | 3917 | * |
25985edc | 3918 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3919 | * runqueue lock must be held. |
3920 | */ | |
0122ec5b | 3921 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3922 | |
34f971f6 PZ |
3923 | /* |
3924 | * Changing the policy of the stop threads its a very bad idea | |
3925 | */ | |
3926 | if (p == rq->stop) { | |
0122ec5b | 3927 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3928 | return -EINVAL; |
3929 | } | |
3930 | ||
a51e9198 DF |
3931 | /* |
3932 | * If not changing anything there's no need to proceed further: | |
3933 | */ | |
3934 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
3935 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 3936 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3937 | return 0; |
3938 | } | |
3939 | ||
dc61b1d6 PZ |
3940 | #ifdef CONFIG_RT_GROUP_SCHED |
3941 | if (user) { | |
3942 | /* | |
3943 | * Do not allow realtime tasks into groups that have no runtime | |
3944 | * assigned. | |
3945 | */ | |
3946 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3947 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3948 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3949 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3950 | return -EPERM; |
3951 | } | |
3952 | } | |
3953 | #endif | |
3954 | ||
1da177e4 LT |
3955 | /* recheck policy now with rq lock held */ |
3956 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3957 | policy = oldpolicy = -1; | |
0122ec5b | 3958 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3959 | goto recheck; |
3960 | } | |
fd2f4419 | 3961 | on_rq = p->on_rq; |
051a1d1a | 3962 | running = task_current(rq, p); |
0e1f3483 | 3963 | if (on_rq) |
4ca9b72b | 3964 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3965 | if (running) |
3966 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3967 | |
ca94c442 LP |
3968 | p->sched_reset_on_fork = reset_on_fork; |
3969 | ||
1da177e4 | 3970 | oldprio = p->prio; |
83ab0aa0 | 3971 | prev_class = p->sched_class; |
dd41f596 | 3972 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 3973 | |
0e1f3483 HS |
3974 | if (running) |
3975 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3976 | if (on_rq) |
4ca9b72b | 3977 | enqueue_task(rq, p, 0); |
cb469845 | 3978 | |
da7a735e | 3979 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3980 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3981 | |
95e02ca9 TG |
3982 | rt_mutex_adjust_pi(p); |
3983 | ||
1da177e4 LT |
3984 | return 0; |
3985 | } | |
961ccddd RR |
3986 | |
3987 | /** | |
3988 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3989 | * @p: the task in question. | |
3990 | * @policy: new policy. | |
3991 | * @param: structure containing the new RT priority. | |
3992 | * | |
3993 | * NOTE that the task may be already dead. | |
3994 | */ | |
3995 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3996 | const struct sched_param *param) |
961ccddd RR |
3997 | { |
3998 | return __sched_setscheduler(p, policy, param, true); | |
3999 | } | |
1da177e4 LT |
4000 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4001 | ||
961ccddd RR |
4002 | /** |
4003 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4004 | * @p: the task in question. | |
4005 | * @policy: new policy. | |
4006 | * @param: structure containing the new RT priority. | |
4007 | * | |
4008 | * Just like sched_setscheduler, only don't bother checking if the | |
4009 | * current context has permission. For example, this is needed in | |
4010 | * stop_machine(): we create temporary high priority worker threads, | |
4011 | * but our caller might not have that capability. | |
4012 | */ | |
4013 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4014 | const struct sched_param *param) |
961ccddd RR |
4015 | { |
4016 | return __sched_setscheduler(p, policy, param, false); | |
4017 | } | |
4018 | ||
95cdf3b7 IM |
4019 | static int |
4020 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4021 | { |
1da177e4 LT |
4022 | struct sched_param lparam; |
4023 | struct task_struct *p; | |
36c8b586 | 4024 | int retval; |
1da177e4 LT |
4025 | |
4026 | if (!param || pid < 0) | |
4027 | return -EINVAL; | |
4028 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4029 | return -EFAULT; | |
5fe1d75f ON |
4030 | |
4031 | rcu_read_lock(); | |
4032 | retval = -ESRCH; | |
1da177e4 | 4033 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4034 | if (p != NULL) |
4035 | retval = sched_setscheduler(p, policy, &lparam); | |
4036 | rcu_read_unlock(); | |
36c8b586 | 4037 | |
1da177e4 LT |
4038 | return retval; |
4039 | } | |
4040 | ||
4041 | /** | |
4042 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4043 | * @pid: the pid in question. | |
4044 | * @policy: new policy. | |
4045 | * @param: structure containing the new RT priority. | |
4046 | */ | |
5add95d4 HC |
4047 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4048 | struct sched_param __user *, param) | |
1da177e4 | 4049 | { |
c21761f1 JB |
4050 | /* negative values for policy are not valid */ |
4051 | if (policy < 0) | |
4052 | return -EINVAL; | |
4053 | ||
1da177e4 LT |
4054 | return do_sched_setscheduler(pid, policy, param); |
4055 | } | |
4056 | ||
4057 | /** | |
4058 | * sys_sched_setparam - set/change the RT priority of a thread | |
4059 | * @pid: the pid in question. | |
4060 | * @param: structure containing the new RT priority. | |
4061 | */ | |
5add95d4 | 4062 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4063 | { |
4064 | return do_sched_setscheduler(pid, -1, param); | |
4065 | } | |
4066 | ||
4067 | /** | |
4068 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4069 | * @pid: the pid in question. | |
4070 | */ | |
5add95d4 | 4071 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4072 | { |
36c8b586 | 4073 | struct task_struct *p; |
3a5c359a | 4074 | int retval; |
1da177e4 LT |
4075 | |
4076 | if (pid < 0) | |
3a5c359a | 4077 | return -EINVAL; |
1da177e4 LT |
4078 | |
4079 | retval = -ESRCH; | |
5fe85be0 | 4080 | rcu_read_lock(); |
1da177e4 LT |
4081 | p = find_process_by_pid(pid); |
4082 | if (p) { | |
4083 | retval = security_task_getscheduler(p); | |
4084 | if (!retval) | |
ca94c442 LP |
4085 | retval = p->policy |
4086 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4087 | } |
5fe85be0 | 4088 | rcu_read_unlock(); |
1da177e4 LT |
4089 | return retval; |
4090 | } | |
4091 | ||
4092 | /** | |
ca94c442 | 4093 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4094 | * @pid: the pid in question. |
4095 | * @param: structure containing the RT priority. | |
4096 | */ | |
5add95d4 | 4097 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4098 | { |
4099 | struct sched_param lp; | |
36c8b586 | 4100 | struct task_struct *p; |
3a5c359a | 4101 | int retval; |
1da177e4 LT |
4102 | |
4103 | if (!param || pid < 0) | |
3a5c359a | 4104 | return -EINVAL; |
1da177e4 | 4105 | |
5fe85be0 | 4106 | rcu_read_lock(); |
1da177e4 LT |
4107 | p = find_process_by_pid(pid); |
4108 | retval = -ESRCH; | |
4109 | if (!p) | |
4110 | goto out_unlock; | |
4111 | ||
4112 | retval = security_task_getscheduler(p); | |
4113 | if (retval) | |
4114 | goto out_unlock; | |
4115 | ||
4116 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4117 | rcu_read_unlock(); |
1da177e4 LT |
4118 | |
4119 | /* | |
4120 | * This one might sleep, we cannot do it with a spinlock held ... | |
4121 | */ | |
4122 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4123 | ||
1da177e4 LT |
4124 | return retval; |
4125 | ||
4126 | out_unlock: | |
5fe85be0 | 4127 | rcu_read_unlock(); |
1da177e4 LT |
4128 | return retval; |
4129 | } | |
4130 | ||
96f874e2 | 4131 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4132 | { |
5a16f3d3 | 4133 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4134 | struct task_struct *p; |
4135 | int retval; | |
1da177e4 | 4136 | |
95402b38 | 4137 | get_online_cpus(); |
23f5d142 | 4138 | rcu_read_lock(); |
1da177e4 LT |
4139 | |
4140 | p = find_process_by_pid(pid); | |
4141 | if (!p) { | |
23f5d142 | 4142 | rcu_read_unlock(); |
95402b38 | 4143 | put_online_cpus(); |
1da177e4 LT |
4144 | return -ESRCH; |
4145 | } | |
4146 | ||
23f5d142 | 4147 | /* Prevent p going away */ |
1da177e4 | 4148 | get_task_struct(p); |
23f5d142 | 4149 | rcu_read_unlock(); |
1da177e4 | 4150 | |
5a16f3d3 RR |
4151 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4152 | retval = -ENOMEM; | |
4153 | goto out_put_task; | |
4154 | } | |
4155 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4156 | retval = -ENOMEM; | |
4157 | goto out_free_cpus_allowed; | |
4158 | } | |
1da177e4 | 4159 | retval = -EPERM; |
4c44aaaf EB |
4160 | if (!check_same_owner(p)) { |
4161 | rcu_read_lock(); | |
4162 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4163 | rcu_read_unlock(); | |
4164 | goto out_unlock; | |
4165 | } | |
4166 | rcu_read_unlock(); | |
4167 | } | |
1da177e4 | 4168 | |
b0ae1981 | 4169 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4170 | if (retval) |
4171 | goto out_unlock; | |
4172 | ||
5a16f3d3 RR |
4173 | cpuset_cpus_allowed(p, cpus_allowed); |
4174 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4175 | again: |
5a16f3d3 | 4176 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4177 | |
8707d8b8 | 4178 | if (!retval) { |
5a16f3d3 RR |
4179 | cpuset_cpus_allowed(p, cpus_allowed); |
4180 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4181 | /* |
4182 | * We must have raced with a concurrent cpuset | |
4183 | * update. Just reset the cpus_allowed to the | |
4184 | * cpuset's cpus_allowed | |
4185 | */ | |
5a16f3d3 | 4186 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4187 | goto again; |
4188 | } | |
4189 | } | |
1da177e4 | 4190 | out_unlock: |
5a16f3d3 RR |
4191 | free_cpumask_var(new_mask); |
4192 | out_free_cpus_allowed: | |
4193 | free_cpumask_var(cpus_allowed); | |
4194 | out_put_task: | |
1da177e4 | 4195 | put_task_struct(p); |
95402b38 | 4196 | put_online_cpus(); |
1da177e4 LT |
4197 | return retval; |
4198 | } | |
4199 | ||
4200 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4201 | struct cpumask *new_mask) |
1da177e4 | 4202 | { |
96f874e2 RR |
4203 | if (len < cpumask_size()) |
4204 | cpumask_clear(new_mask); | |
4205 | else if (len > cpumask_size()) | |
4206 | len = cpumask_size(); | |
4207 | ||
1da177e4 LT |
4208 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4209 | } | |
4210 | ||
4211 | /** | |
4212 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4213 | * @pid: pid of the process | |
4214 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4215 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4216 | */ | |
5add95d4 HC |
4217 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4218 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4219 | { |
5a16f3d3 | 4220 | cpumask_var_t new_mask; |
1da177e4 LT |
4221 | int retval; |
4222 | ||
5a16f3d3 RR |
4223 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4224 | return -ENOMEM; | |
1da177e4 | 4225 | |
5a16f3d3 RR |
4226 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4227 | if (retval == 0) | |
4228 | retval = sched_setaffinity(pid, new_mask); | |
4229 | free_cpumask_var(new_mask); | |
4230 | return retval; | |
1da177e4 LT |
4231 | } |
4232 | ||
96f874e2 | 4233 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4234 | { |
36c8b586 | 4235 | struct task_struct *p; |
31605683 | 4236 | unsigned long flags; |
1da177e4 | 4237 | int retval; |
1da177e4 | 4238 | |
95402b38 | 4239 | get_online_cpus(); |
23f5d142 | 4240 | rcu_read_lock(); |
1da177e4 LT |
4241 | |
4242 | retval = -ESRCH; | |
4243 | p = find_process_by_pid(pid); | |
4244 | if (!p) | |
4245 | goto out_unlock; | |
4246 | ||
e7834f8f DQ |
4247 | retval = security_task_getscheduler(p); |
4248 | if (retval) | |
4249 | goto out_unlock; | |
4250 | ||
013fdb80 | 4251 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4252 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4253 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4254 | |
4255 | out_unlock: | |
23f5d142 | 4256 | rcu_read_unlock(); |
95402b38 | 4257 | put_online_cpus(); |
1da177e4 | 4258 | |
9531b62f | 4259 | return retval; |
1da177e4 LT |
4260 | } |
4261 | ||
4262 | /** | |
4263 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4264 | * @pid: pid of the process | |
4265 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4266 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4267 | */ | |
5add95d4 HC |
4268 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4269 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4270 | { |
4271 | int ret; | |
f17c8607 | 4272 | cpumask_var_t mask; |
1da177e4 | 4273 | |
84fba5ec | 4274 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4275 | return -EINVAL; |
4276 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4277 | return -EINVAL; |
4278 | ||
f17c8607 RR |
4279 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4280 | return -ENOMEM; | |
1da177e4 | 4281 | |
f17c8607 RR |
4282 | ret = sched_getaffinity(pid, mask); |
4283 | if (ret == 0) { | |
8bc037fb | 4284 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4285 | |
4286 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4287 | ret = -EFAULT; |
4288 | else | |
cd3d8031 | 4289 | ret = retlen; |
f17c8607 RR |
4290 | } |
4291 | free_cpumask_var(mask); | |
1da177e4 | 4292 | |
f17c8607 | 4293 | return ret; |
1da177e4 LT |
4294 | } |
4295 | ||
4296 | /** | |
4297 | * sys_sched_yield - yield the current processor to other threads. | |
4298 | * | |
dd41f596 IM |
4299 | * This function yields the current CPU to other tasks. If there are no |
4300 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4301 | */ |
5add95d4 | 4302 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4303 | { |
70b97a7f | 4304 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4305 | |
2d72376b | 4306 | schedstat_inc(rq, yld_count); |
4530d7ab | 4307 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4308 | |
4309 | /* | |
4310 | * Since we are going to call schedule() anyway, there's | |
4311 | * no need to preempt or enable interrupts: | |
4312 | */ | |
4313 | __release(rq->lock); | |
8a25d5de | 4314 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4315 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4316 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4317 | |
4318 | schedule(); | |
4319 | ||
4320 | return 0; | |
4321 | } | |
4322 | ||
d86ee480 PZ |
4323 | static inline int should_resched(void) |
4324 | { | |
4325 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4326 | } | |
4327 | ||
e7b38404 | 4328 | static void __cond_resched(void) |
1da177e4 | 4329 | { |
e7aaaa69 | 4330 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4331 | __schedule(); |
e7aaaa69 | 4332 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4333 | } |
4334 | ||
02b67cc3 | 4335 | int __sched _cond_resched(void) |
1da177e4 | 4336 | { |
d86ee480 | 4337 | if (should_resched()) { |
1da177e4 LT |
4338 | __cond_resched(); |
4339 | return 1; | |
4340 | } | |
4341 | return 0; | |
4342 | } | |
02b67cc3 | 4343 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4344 | |
4345 | /* | |
613afbf8 | 4346 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4347 | * call schedule, and on return reacquire the lock. |
4348 | * | |
41a2d6cf | 4349 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4350 | * operations here to prevent schedule() from being called twice (once via |
4351 | * spin_unlock(), once by hand). | |
4352 | */ | |
613afbf8 | 4353 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4354 | { |
d86ee480 | 4355 | int resched = should_resched(); |
6df3cecb JK |
4356 | int ret = 0; |
4357 | ||
f607c668 PZ |
4358 | lockdep_assert_held(lock); |
4359 | ||
95c354fe | 4360 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4361 | spin_unlock(lock); |
d86ee480 | 4362 | if (resched) |
95c354fe NP |
4363 | __cond_resched(); |
4364 | else | |
4365 | cpu_relax(); | |
6df3cecb | 4366 | ret = 1; |
1da177e4 | 4367 | spin_lock(lock); |
1da177e4 | 4368 | } |
6df3cecb | 4369 | return ret; |
1da177e4 | 4370 | } |
613afbf8 | 4371 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4372 | |
613afbf8 | 4373 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4374 | { |
4375 | BUG_ON(!in_softirq()); | |
4376 | ||
d86ee480 | 4377 | if (should_resched()) { |
98d82567 | 4378 | local_bh_enable(); |
1da177e4 LT |
4379 | __cond_resched(); |
4380 | local_bh_disable(); | |
4381 | return 1; | |
4382 | } | |
4383 | return 0; | |
4384 | } | |
613afbf8 | 4385 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4386 | |
1da177e4 LT |
4387 | /** |
4388 | * yield - yield the current processor to other threads. | |
4389 | * | |
8e3fabfd PZ |
4390 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4391 | * | |
4392 | * The scheduler is at all times free to pick the calling task as the most | |
4393 | * eligible task to run, if removing the yield() call from your code breaks | |
4394 | * it, its already broken. | |
4395 | * | |
4396 | * Typical broken usage is: | |
4397 | * | |
4398 | * while (!event) | |
4399 | * yield(); | |
4400 | * | |
4401 | * where one assumes that yield() will let 'the other' process run that will | |
4402 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4403 | * happen. Never use yield() as a progress guarantee!! | |
4404 | * | |
4405 | * If you want to use yield() to wait for something, use wait_event(). | |
4406 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4407 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4408 | */ |
4409 | void __sched yield(void) | |
4410 | { | |
4411 | set_current_state(TASK_RUNNING); | |
4412 | sys_sched_yield(); | |
4413 | } | |
1da177e4 LT |
4414 | EXPORT_SYMBOL(yield); |
4415 | ||
d95f4122 MG |
4416 | /** |
4417 | * yield_to - yield the current processor to another thread in | |
4418 | * your thread group, or accelerate that thread toward the | |
4419 | * processor it's on. | |
16addf95 RD |
4420 | * @p: target task |
4421 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4422 | * |
4423 | * It's the caller's job to ensure that the target task struct | |
4424 | * can't go away on us before we can do any checks. | |
4425 | * | |
7b270f60 PZ |
4426 | * Returns: |
4427 | * true (>0) if we indeed boosted the target task. | |
4428 | * false (0) if we failed to boost the target. | |
4429 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4430 | */ |
4431 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4432 | { | |
4433 | struct task_struct *curr = current; | |
4434 | struct rq *rq, *p_rq; | |
4435 | unsigned long flags; | |
c3c18640 | 4436 | int yielded = 0; |
d95f4122 MG |
4437 | |
4438 | local_irq_save(flags); | |
4439 | rq = this_rq(); | |
4440 | ||
4441 | again: | |
4442 | p_rq = task_rq(p); | |
7b270f60 PZ |
4443 | /* |
4444 | * If we're the only runnable task on the rq and target rq also | |
4445 | * has only one task, there's absolutely no point in yielding. | |
4446 | */ | |
4447 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4448 | yielded = -ESRCH; | |
4449 | goto out_irq; | |
4450 | } | |
4451 | ||
d95f4122 MG |
4452 | double_rq_lock(rq, p_rq); |
4453 | while (task_rq(p) != p_rq) { | |
4454 | double_rq_unlock(rq, p_rq); | |
4455 | goto again; | |
4456 | } | |
4457 | ||
4458 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4459 | goto out_unlock; |
d95f4122 MG |
4460 | |
4461 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4462 | goto out_unlock; |
d95f4122 MG |
4463 | |
4464 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4465 | goto out_unlock; |
d95f4122 MG |
4466 | |
4467 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4468 | if (yielded) { |
d95f4122 | 4469 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4470 | /* |
4471 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4472 | * fairness. | |
4473 | */ | |
4474 | if (preempt && rq != p_rq) | |
4475 | resched_task(p_rq->curr); | |
4476 | } | |
d95f4122 | 4477 | |
7b270f60 | 4478 | out_unlock: |
d95f4122 | 4479 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4480 | out_irq: |
d95f4122 MG |
4481 | local_irq_restore(flags); |
4482 | ||
7b270f60 | 4483 | if (yielded > 0) |
d95f4122 MG |
4484 | schedule(); |
4485 | ||
4486 | return yielded; | |
4487 | } | |
4488 | EXPORT_SYMBOL_GPL(yield_to); | |
4489 | ||
1da177e4 | 4490 | /* |
41a2d6cf | 4491 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4492 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4493 | */ |
4494 | void __sched io_schedule(void) | |
4495 | { | |
54d35f29 | 4496 | struct rq *rq = raw_rq(); |
1da177e4 | 4497 | |
0ff92245 | 4498 | delayacct_blkio_start(); |
1da177e4 | 4499 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4500 | blk_flush_plug(current); |
8f0dfc34 | 4501 | current->in_iowait = 1; |
1da177e4 | 4502 | schedule(); |
8f0dfc34 | 4503 | current->in_iowait = 0; |
1da177e4 | 4504 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4505 | delayacct_blkio_end(); |
1da177e4 | 4506 | } |
1da177e4 LT |
4507 | EXPORT_SYMBOL(io_schedule); |
4508 | ||
4509 | long __sched io_schedule_timeout(long timeout) | |
4510 | { | |
54d35f29 | 4511 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4512 | long ret; |
4513 | ||
0ff92245 | 4514 | delayacct_blkio_start(); |
1da177e4 | 4515 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4516 | blk_flush_plug(current); |
8f0dfc34 | 4517 | current->in_iowait = 1; |
1da177e4 | 4518 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4519 | current->in_iowait = 0; |
1da177e4 | 4520 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4521 | delayacct_blkio_end(); |
1da177e4 LT |
4522 | return ret; |
4523 | } | |
4524 | ||
4525 | /** | |
4526 | * sys_sched_get_priority_max - return maximum RT priority. | |
4527 | * @policy: scheduling class. | |
4528 | * | |
4529 | * this syscall returns the maximum rt_priority that can be used | |
4530 | * by a given scheduling class. | |
4531 | */ | |
5add95d4 | 4532 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4533 | { |
4534 | int ret = -EINVAL; | |
4535 | ||
4536 | switch (policy) { | |
4537 | case SCHED_FIFO: | |
4538 | case SCHED_RR: | |
4539 | ret = MAX_USER_RT_PRIO-1; | |
4540 | break; | |
4541 | case SCHED_NORMAL: | |
b0a9499c | 4542 | case SCHED_BATCH: |
dd41f596 | 4543 | case SCHED_IDLE: |
1da177e4 LT |
4544 | ret = 0; |
4545 | break; | |
4546 | } | |
4547 | return ret; | |
4548 | } | |
4549 | ||
4550 | /** | |
4551 | * sys_sched_get_priority_min - return minimum RT priority. | |
4552 | * @policy: scheduling class. | |
4553 | * | |
4554 | * this syscall returns the minimum rt_priority that can be used | |
4555 | * by a given scheduling class. | |
4556 | */ | |
5add95d4 | 4557 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4558 | { |
4559 | int ret = -EINVAL; | |
4560 | ||
4561 | switch (policy) { | |
4562 | case SCHED_FIFO: | |
4563 | case SCHED_RR: | |
4564 | ret = 1; | |
4565 | break; | |
4566 | case SCHED_NORMAL: | |
b0a9499c | 4567 | case SCHED_BATCH: |
dd41f596 | 4568 | case SCHED_IDLE: |
1da177e4 LT |
4569 | ret = 0; |
4570 | } | |
4571 | return ret; | |
4572 | } | |
4573 | ||
4574 | /** | |
4575 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4576 | * @pid: pid of the process. | |
4577 | * @interval: userspace pointer to the timeslice value. | |
4578 | * | |
4579 | * this syscall writes the default timeslice value of a given process | |
4580 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4581 | */ | |
17da2bd9 | 4582 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4583 | struct timespec __user *, interval) |
1da177e4 | 4584 | { |
36c8b586 | 4585 | struct task_struct *p; |
a4ec24b4 | 4586 | unsigned int time_slice; |
dba091b9 TG |
4587 | unsigned long flags; |
4588 | struct rq *rq; | |
3a5c359a | 4589 | int retval; |
1da177e4 | 4590 | struct timespec t; |
1da177e4 LT |
4591 | |
4592 | if (pid < 0) | |
3a5c359a | 4593 | return -EINVAL; |
1da177e4 LT |
4594 | |
4595 | retval = -ESRCH; | |
1a551ae7 | 4596 | rcu_read_lock(); |
1da177e4 LT |
4597 | p = find_process_by_pid(pid); |
4598 | if (!p) | |
4599 | goto out_unlock; | |
4600 | ||
4601 | retval = security_task_getscheduler(p); | |
4602 | if (retval) | |
4603 | goto out_unlock; | |
4604 | ||
dba091b9 TG |
4605 | rq = task_rq_lock(p, &flags); |
4606 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4607 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4608 | |
1a551ae7 | 4609 | rcu_read_unlock(); |
a4ec24b4 | 4610 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4611 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4612 | return retval; |
3a5c359a | 4613 | |
1da177e4 | 4614 | out_unlock: |
1a551ae7 | 4615 | rcu_read_unlock(); |
1da177e4 LT |
4616 | return retval; |
4617 | } | |
4618 | ||
7c731e0a | 4619 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4620 | |
82a1fcb9 | 4621 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4622 | { |
1da177e4 | 4623 | unsigned long free = 0; |
4e79752c | 4624 | int ppid; |
36c8b586 | 4625 | unsigned state; |
1da177e4 | 4626 | |
1da177e4 | 4627 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4628 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4629 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4630 | #if BITS_PER_LONG == 32 |
1da177e4 | 4631 | if (state == TASK_RUNNING) |
3df0fc5b | 4632 | printk(KERN_CONT " running "); |
1da177e4 | 4633 | else |
3df0fc5b | 4634 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4635 | #else |
4636 | if (state == TASK_RUNNING) | |
3df0fc5b | 4637 | printk(KERN_CONT " running task "); |
1da177e4 | 4638 | else |
3df0fc5b | 4639 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4640 | #endif |
4641 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4642 | free = stack_not_used(p); |
1da177e4 | 4643 | #endif |
4e79752c PM |
4644 | rcu_read_lock(); |
4645 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4646 | rcu_read_unlock(); | |
3df0fc5b | 4647 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4648 | task_pid_nr(p), ppid, |
aa47b7e0 | 4649 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4650 | |
5fb5e6de | 4651 | show_stack(p, NULL); |
1da177e4 LT |
4652 | } |
4653 | ||
e59e2ae2 | 4654 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4655 | { |
36c8b586 | 4656 | struct task_struct *g, *p; |
1da177e4 | 4657 | |
4bd77321 | 4658 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4659 | printk(KERN_INFO |
4660 | " task PC stack pid father\n"); | |
1da177e4 | 4661 | #else |
3df0fc5b PZ |
4662 | printk(KERN_INFO |
4663 | " task PC stack pid father\n"); | |
1da177e4 | 4664 | #endif |
510f5acc | 4665 | rcu_read_lock(); |
1da177e4 LT |
4666 | do_each_thread(g, p) { |
4667 | /* | |
4668 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4669 | * console might take a lot of time: |
1da177e4 LT |
4670 | */ |
4671 | touch_nmi_watchdog(); | |
39bc89fd | 4672 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4673 | sched_show_task(p); |
1da177e4 LT |
4674 | } while_each_thread(g, p); |
4675 | ||
04c9167f JF |
4676 | touch_all_softlockup_watchdogs(); |
4677 | ||
dd41f596 IM |
4678 | #ifdef CONFIG_SCHED_DEBUG |
4679 | sysrq_sched_debug_show(); | |
4680 | #endif | |
510f5acc | 4681 | rcu_read_unlock(); |
e59e2ae2 IM |
4682 | /* |
4683 | * Only show locks if all tasks are dumped: | |
4684 | */ | |
93335a21 | 4685 | if (!state_filter) |
e59e2ae2 | 4686 | debug_show_all_locks(); |
1da177e4 LT |
4687 | } |
4688 | ||
1df21055 IM |
4689 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4690 | { | |
dd41f596 | 4691 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4692 | } |
4693 | ||
f340c0d1 IM |
4694 | /** |
4695 | * init_idle - set up an idle thread for a given CPU | |
4696 | * @idle: task in question | |
4697 | * @cpu: cpu the idle task belongs to | |
4698 | * | |
4699 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4700 | * flag, to make booting more robust. | |
4701 | */ | |
5c1e1767 | 4702 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4703 | { |
70b97a7f | 4704 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4705 | unsigned long flags; |
4706 | ||
05fa785c | 4707 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4708 | |
dd41f596 | 4709 | __sched_fork(idle); |
06b83b5f | 4710 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4711 | idle->se.exec_start = sched_clock(); |
4712 | ||
1e1b6c51 | 4713 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4714 | /* |
4715 | * We're having a chicken and egg problem, even though we are | |
4716 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4717 | * lockdep check in task_group() will fail. | |
4718 | * | |
4719 | * Similar case to sched_fork(). / Alternatively we could | |
4720 | * use task_rq_lock() here and obtain the other rq->lock. | |
4721 | * | |
4722 | * Silence PROVE_RCU | |
4723 | */ | |
4724 | rcu_read_lock(); | |
dd41f596 | 4725 | __set_task_cpu(idle, cpu); |
6506cf6c | 4726 | rcu_read_unlock(); |
1da177e4 | 4727 | |
1da177e4 | 4728 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4729 | #if defined(CONFIG_SMP) |
4730 | idle->on_cpu = 1; | |
4866cde0 | 4731 | #endif |
05fa785c | 4732 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4733 | |
4734 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4735 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4736 | |
dd41f596 IM |
4737 | /* |
4738 | * The idle tasks have their own, simple scheduling class: | |
4739 | */ | |
4740 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4741 | ftrace_graph_init_idle_task(idle, cpu); |
6a61671b | 4742 | vtime_init_idle(idle); |
f1c6f1a7 CE |
4743 | #if defined(CONFIG_SMP) |
4744 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4745 | #endif | |
19978ca6 IM |
4746 | } |
4747 | ||
1da177e4 | 4748 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4749 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4750 | { | |
4751 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4752 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4753 | |
4754 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4755 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4756 | } |
4757 | ||
1da177e4 LT |
4758 | /* |
4759 | * This is how migration works: | |
4760 | * | |
969c7921 TH |
4761 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4762 | * stop_one_cpu(). | |
4763 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4764 | * off the CPU) | |
4765 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4766 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4767 | * it and puts it into the right queue. |
969c7921 TH |
4768 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4769 | * is done. | |
1da177e4 LT |
4770 | */ |
4771 | ||
4772 | /* | |
4773 | * Change a given task's CPU affinity. Migrate the thread to a | |
4774 | * proper CPU and schedule it away if the CPU it's executing on | |
4775 | * is removed from the allowed bitmask. | |
4776 | * | |
4777 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4778 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4779 | * call is not atomic; no spinlocks may be held. |
4780 | */ | |
96f874e2 | 4781 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4782 | { |
4783 | unsigned long flags; | |
70b97a7f | 4784 | struct rq *rq; |
969c7921 | 4785 | unsigned int dest_cpu; |
48f24c4d | 4786 | int ret = 0; |
1da177e4 LT |
4787 | |
4788 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4789 | |
db44fc01 YZ |
4790 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4791 | goto out; | |
4792 | ||
6ad4c188 | 4793 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4794 | ret = -EINVAL; |
4795 | goto out; | |
4796 | } | |
4797 | ||
db44fc01 | 4798 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4799 | ret = -EINVAL; |
4800 | goto out; | |
4801 | } | |
4802 | ||
1e1b6c51 | 4803 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4804 | |
1da177e4 | 4805 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4806 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4807 | goto out; |
4808 | ||
969c7921 | 4809 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4810 | if (p->on_rq) { |
969c7921 | 4811 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4812 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4813 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4814 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4815 | tlb_migrate_finish(p->mm); |
4816 | return 0; | |
4817 | } | |
4818 | out: | |
0122ec5b | 4819 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4820 | |
1da177e4 LT |
4821 | return ret; |
4822 | } | |
cd8ba7cd | 4823 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4824 | |
4825 | /* | |
41a2d6cf | 4826 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4827 | * this because either it can't run here any more (set_cpus_allowed() |
4828 | * away from this CPU, or CPU going down), or because we're | |
4829 | * attempting to rebalance this task on exec (sched_exec). | |
4830 | * | |
4831 | * So we race with normal scheduler movements, but that's OK, as long | |
4832 | * as the task is no longer on this CPU. | |
efc30814 KK |
4833 | * |
4834 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4835 | */ |
efc30814 | 4836 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4837 | { |
70b97a7f | 4838 | struct rq *rq_dest, *rq_src; |
e2912009 | 4839 | int ret = 0; |
1da177e4 | 4840 | |
e761b772 | 4841 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4842 | return ret; |
1da177e4 LT |
4843 | |
4844 | rq_src = cpu_rq(src_cpu); | |
4845 | rq_dest = cpu_rq(dest_cpu); | |
4846 | ||
0122ec5b | 4847 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4848 | double_rq_lock(rq_src, rq_dest); |
4849 | /* Already moved. */ | |
4850 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4851 | goto done; |
1da177e4 | 4852 | /* Affinity changed (again). */ |
fa17b507 | 4853 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4854 | goto fail; |
1da177e4 | 4855 | |
e2912009 PZ |
4856 | /* |
4857 | * If we're not on a rq, the next wake-up will ensure we're | |
4858 | * placed properly. | |
4859 | */ | |
fd2f4419 | 4860 | if (p->on_rq) { |
4ca9b72b | 4861 | dequeue_task(rq_src, p, 0); |
e2912009 | 4862 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4863 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4864 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4865 | } |
b1e38734 | 4866 | done: |
efc30814 | 4867 | ret = 1; |
b1e38734 | 4868 | fail: |
1da177e4 | 4869 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4870 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4871 | return ret; |
1da177e4 LT |
4872 | } |
4873 | ||
4874 | /* | |
969c7921 TH |
4875 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4876 | * and performs thread migration by bumping thread off CPU then | |
4877 | * 'pushing' onto another runqueue. | |
1da177e4 | 4878 | */ |
969c7921 | 4879 | static int migration_cpu_stop(void *data) |
1da177e4 | 4880 | { |
969c7921 | 4881 | struct migration_arg *arg = data; |
f7b4cddc | 4882 | |
969c7921 TH |
4883 | /* |
4884 | * The original target cpu might have gone down and we might | |
4885 | * be on another cpu but it doesn't matter. | |
4886 | */ | |
f7b4cddc | 4887 | local_irq_disable(); |
969c7921 | 4888 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4889 | local_irq_enable(); |
1da177e4 | 4890 | return 0; |
f7b4cddc ON |
4891 | } |
4892 | ||
1da177e4 | 4893 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4894 | |
054b9108 | 4895 | /* |
48c5ccae PZ |
4896 | * Ensures that the idle task is using init_mm right before its cpu goes |
4897 | * offline. | |
054b9108 | 4898 | */ |
48c5ccae | 4899 | void idle_task_exit(void) |
1da177e4 | 4900 | { |
48c5ccae | 4901 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4902 | |
48c5ccae | 4903 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4904 | |
48c5ccae PZ |
4905 | if (mm != &init_mm) |
4906 | switch_mm(mm, &init_mm, current); | |
4907 | mmdrop(mm); | |
1da177e4 LT |
4908 | } |
4909 | ||
4910 | /* | |
5d180232 PZ |
4911 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4912 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4913 | * nr_active count is stable. | |
4914 | * | |
4915 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4916 | */ |
5d180232 | 4917 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4918 | { |
5d180232 PZ |
4919 | long delta = calc_load_fold_active(rq); |
4920 | if (delta) | |
4921 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4922 | } |
4923 | ||
48f24c4d | 4924 | /* |
48c5ccae PZ |
4925 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4926 | * try_to_wake_up()->select_task_rq(). | |
4927 | * | |
4928 | * Called with rq->lock held even though we'er in stop_machine() and | |
4929 | * there's no concurrency possible, we hold the required locks anyway | |
4930 | * because of lock validation efforts. | |
1da177e4 | 4931 | */ |
48c5ccae | 4932 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4933 | { |
70b97a7f | 4934 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4935 | struct task_struct *next, *stop = rq->stop; |
4936 | int dest_cpu; | |
1da177e4 LT |
4937 | |
4938 | /* | |
48c5ccae PZ |
4939 | * Fudge the rq selection such that the below task selection loop |
4940 | * doesn't get stuck on the currently eligible stop task. | |
4941 | * | |
4942 | * We're currently inside stop_machine() and the rq is either stuck | |
4943 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4944 | * either way we should never end up calling schedule() until we're | |
4945 | * done here. | |
1da177e4 | 4946 | */ |
48c5ccae | 4947 | rq->stop = NULL; |
48f24c4d | 4948 | |
dd41f596 | 4949 | for ( ; ; ) { |
48c5ccae PZ |
4950 | /* |
4951 | * There's this thread running, bail when that's the only | |
4952 | * remaining thread. | |
4953 | */ | |
4954 | if (rq->nr_running == 1) | |
dd41f596 | 4955 | break; |
48c5ccae | 4956 | |
b67802ea | 4957 | next = pick_next_task(rq); |
48c5ccae | 4958 | BUG_ON(!next); |
79c53799 | 4959 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4960 | |
48c5ccae PZ |
4961 | /* Find suitable destination for @next, with force if needed. */ |
4962 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4963 | raw_spin_unlock(&rq->lock); | |
4964 | ||
4965 | __migrate_task(next, dead_cpu, dest_cpu); | |
4966 | ||
4967 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4968 | } |
dce48a84 | 4969 | |
48c5ccae | 4970 | rq->stop = stop; |
dce48a84 | 4971 | } |
48c5ccae | 4972 | |
1da177e4 LT |
4973 | #endif /* CONFIG_HOTPLUG_CPU */ |
4974 | ||
e692ab53 NP |
4975 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4976 | ||
4977 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4978 | { |
4979 | .procname = "sched_domain", | |
c57baf1e | 4980 | .mode = 0555, |
e0361851 | 4981 | }, |
56992309 | 4982 | {} |
e692ab53 NP |
4983 | }; |
4984 | ||
4985 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4986 | { |
4987 | .procname = "kernel", | |
c57baf1e | 4988 | .mode = 0555, |
e0361851 AD |
4989 | .child = sd_ctl_dir, |
4990 | }, | |
56992309 | 4991 | {} |
e692ab53 NP |
4992 | }; |
4993 | ||
4994 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4995 | { | |
4996 | struct ctl_table *entry = | |
5cf9f062 | 4997 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4998 | |
e692ab53 NP |
4999 | return entry; |
5000 | } | |
5001 | ||
6382bc90 MM |
5002 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5003 | { | |
cd790076 | 5004 | struct ctl_table *entry; |
6382bc90 | 5005 | |
cd790076 MM |
5006 | /* |
5007 | * In the intermediate directories, both the child directory and | |
5008 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5009 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5010 | * static strings and all have proc handlers. |
5011 | */ | |
5012 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5013 | if (entry->child) |
5014 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5015 | if (entry->proc_handler == NULL) |
5016 | kfree(entry->procname); | |
5017 | } | |
6382bc90 MM |
5018 | |
5019 | kfree(*tablep); | |
5020 | *tablep = NULL; | |
5021 | } | |
5022 | ||
201c373e NK |
5023 | static int min_load_idx = 0; |
5024 | static int max_load_idx = CPU_LOAD_IDX_MAX; | |
5025 | ||
e692ab53 | 5026 | static void |
e0361851 | 5027 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5028 | const char *procname, void *data, int maxlen, |
201c373e NK |
5029 | umode_t mode, proc_handler *proc_handler, |
5030 | bool load_idx) | |
e692ab53 | 5031 | { |
e692ab53 NP |
5032 | entry->procname = procname; |
5033 | entry->data = data; | |
5034 | entry->maxlen = maxlen; | |
5035 | entry->mode = mode; | |
5036 | entry->proc_handler = proc_handler; | |
201c373e NK |
5037 | |
5038 | if (load_idx) { | |
5039 | entry->extra1 = &min_load_idx; | |
5040 | entry->extra2 = &max_load_idx; | |
5041 | } | |
e692ab53 NP |
5042 | } |
5043 | ||
5044 | static struct ctl_table * | |
5045 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5046 | { | |
a5d8c348 | 5047 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5048 | |
ad1cdc1d MM |
5049 | if (table == NULL) |
5050 | return NULL; | |
5051 | ||
e0361851 | 5052 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5053 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5054 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5055 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5056 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5057 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5058 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5059 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5060 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5061 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5062 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5063 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5064 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5065 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5066 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5067 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5068 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5069 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5070 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5071 | &sd->cache_nice_tries, |
201c373e | 5072 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5073 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5074 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 5075 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 5076 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 5077 | /* &table[12] is terminator */ |
e692ab53 NP |
5078 | |
5079 | return table; | |
5080 | } | |
5081 | ||
9a4e7159 | 5082 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5083 | { |
5084 | struct ctl_table *entry, *table; | |
5085 | struct sched_domain *sd; | |
5086 | int domain_num = 0, i; | |
5087 | char buf[32]; | |
5088 | ||
5089 | for_each_domain(cpu, sd) | |
5090 | domain_num++; | |
5091 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5092 | if (table == NULL) |
5093 | return NULL; | |
e692ab53 NP |
5094 | |
5095 | i = 0; | |
5096 | for_each_domain(cpu, sd) { | |
5097 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5098 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5099 | entry->mode = 0555; |
e692ab53 NP |
5100 | entry->child = sd_alloc_ctl_domain_table(sd); |
5101 | entry++; | |
5102 | i++; | |
5103 | } | |
5104 | return table; | |
5105 | } | |
5106 | ||
5107 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5108 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5109 | { |
6ad4c188 | 5110 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5111 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5112 | char buf[32]; | |
5113 | ||
7378547f MM |
5114 | WARN_ON(sd_ctl_dir[0].child); |
5115 | sd_ctl_dir[0].child = entry; | |
5116 | ||
ad1cdc1d MM |
5117 | if (entry == NULL) |
5118 | return; | |
5119 | ||
6ad4c188 | 5120 | for_each_possible_cpu(i) { |
e692ab53 | 5121 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5122 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5123 | entry->mode = 0555; |
e692ab53 | 5124 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5125 | entry++; |
e692ab53 | 5126 | } |
7378547f MM |
5127 | |
5128 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5129 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5130 | } | |
6382bc90 | 5131 | |
7378547f | 5132 | /* may be called multiple times per register */ |
6382bc90 MM |
5133 | static void unregister_sched_domain_sysctl(void) |
5134 | { | |
7378547f MM |
5135 | if (sd_sysctl_header) |
5136 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5137 | sd_sysctl_header = NULL; |
7378547f MM |
5138 | if (sd_ctl_dir[0].child) |
5139 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5140 | } |
e692ab53 | 5141 | #else |
6382bc90 MM |
5142 | static void register_sched_domain_sysctl(void) |
5143 | { | |
5144 | } | |
5145 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5146 | { |
5147 | } | |
5148 | #endif | |
5149 | ||
1f11eb6a GH |
5150 | static void set_rq_online(struct rq *rq) |
5151 | { | |
5152 | if (!rq->online) { | |
5153 | const struct sched_class *class; | |
5154 | ||
c6c4927b | 5155 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5156 | rq->online = 1; |
5157 | ||
5158 | for_each_class(class) { | |
5159 | if (class->rq_online) | |
5160 | class->rq_online(rq); | |
5161 | } | |
5162 | } | |
5163 | } | |
5164 | ||
5165 | static void set_rq_offline(struct rq *rq) | |
5166 | { | |
5167 | if (rq->online) { | |
5168 | const struct sched_class *class; | |
5169 | ||
5170 | for_each_class(class) { | |
5171 | if (class->rq_offline) | |
5172 | class->rq_offline(rq); | |
5173 | } | |
5174 | ||
c6c4927b | 5175 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5176 | rq->online = 0; |
5177 | } | |
5178 | } | |
5179 | ||
1da177e4 LT |
5180 | /* |
5181 | * migration_call - callback that gets triggered when a CPU is added. | |
5182 | * Here we can start up the necessary migration thread for the new CPU. | |
5183 | */ | |
48f24c4d IM |
5184 | static int __cpuinit |
5185 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5186 | { |
48f24c4d | 5187 | int cpu = (long)hcpu; |
1da177e4 | 5188 | unsigned long flags; |
969c7921 | 5189 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5190 | |
48c5ccae | 5191 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5192 | |
1da177e4 | 5193 | case CPU_UP_PREPARE: |
a468d389 | 5194 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5195 | break; |
48f24c4d | 5196 | |
1da177e4 | 5197 | case CPU_ONLINE: |
1f94ef59 | 5198 | /* Update our root-domain */ |
05fa785c | 5199 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5200 | if (rq->rd) { |
c6c4927b | 5201 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5202 | |
5203 | set_rq_online(rq); | |
1f94ef59 | 5204 | } |
05fa785c | 5205 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5206 | break; |
48f24c4d | 5207 | |
1da177e4 | 5208 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5209 | case CPU_DYING: |
317f3941 | 5210 | sched_ttwu_pending(); |
57d885fe | 5211 | /* Update our root-domain */ |
05fa785c | 5212 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5213 | if (rq->rd) { |
c6c4927b | 5214 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5215 | set_rq_offline(rq); |
57d885fe | 5216 | } |
48c5ccae PZ |
5217 | migrate_tasks(cpu); |
5218 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5219 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5220 | break; |
48c5ccae | 5221 | |
5d180232 | 5222 | case CPU_DEAD: |
f319da0c | 5223 | calc_load_migrate(rq); |
57d885fe | 5224 | break; |
1da177e4 LT |
5225 | #endif |
5226 | } | |
49c022e6 PZ |
5227 | |
5228 | update_max_interval(); | |
5229 | ||
1da177e4 LT |
5230 | return NOTIFY_OK; |
5231 | } | |
5232 | ||
f38b0820 PM |
5233 | /* |
5234 | * Register at high priority so that task migration (migrate_all_tasks) | |
5235 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5236 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5237 | */ |
26c2143b | 5238 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5239 | .notifier_call = migration_call, |
50a323b7 | 5240 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5241 | }; |
5242 | ||
3a101d05 TH |
5243 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5244 | unsigned long action, void *hcpu) | |
5245 | { | |
5246 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5247 | case CPU_STARTING: |
3a101d05 TH |
5248 | case CPU_DOWN_FAILED: |
5249 | set_cpu_active((long)hcpu, true); | |
5250 | return NOTIFY_OK; | |
5251 | default: | |
5252 | return NOTIFY_DONE; | |
5253 | } | |
5254 | } | |
5255 | ||
5256 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5257 | unsigned long action, void *hcpu) | |
5258 | { | |
5259 | switch (action & ~CPU_TASKS_FROZEN) { | |
5260 | case CPU_DOWN_PREPARE: | |
5261 | set_cpu_active((long)hcpu, false); | |
5262 | return NOTIFY_OK; | |
5263 | default: | |
5264 | return NOTIFY_DONE; | |
5265 | } | |
5266 | } | |
5267 | ||
7babe8db | 5268 | static int __init migration_init(void) |
1da177e4 LT |
5269 | { |
5270 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5271 | int err; |
48f24c4d | 5272 | |
3a101d05 | 5273 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5274 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5275 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5276 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5277 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5278 | |
3a101d05 TH |
5279 | /* Register cpu active notifiers */ |
5280 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5281 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5282 | ||
a004cd42 | 5283 | return 0; |
1da177e4 | 5284 | } |
7babe8db | 5285 | early_initcall(migration_init); |
1da177e4 LT |
5286 | #endif |
5287 | ||
5288 | #ifdef CONFIG_SMP | |
476f3534 | 5289 | |
4cb98839 PZ |
5290 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5291 | ||
3e9830dc | 5292 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5293 | |
d039ac60 | 5294 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5295 | |
d039ac60 | 5296 | static int __init sched_debug_setup(char *str) |
f6630114 | 5297 | { |
d039ac60 | 5298 | sched_debug_enabled = 1; |
f6630114 MT |
5299 | |
5300 | return 0; | |
5301 | } | |
d039ac60 PZ |
5302 | early_param("sched_debug", sched_debug_setup); |
5303 | ||
5304 | static inline bool sched_debug(void) | |
5305 | { | |
5306 | return sched_debug_enabled; | |
5307 | } | |
f6630114 | 5308 | |
7c16ec58 | 5309 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5310 | struct cpumask *groupmask) |
1da177e4 | 5311 | { |
4dcf6aff | 5312 | struct sched_group *group = sd->groups; |
434d53b0 | 5313 | char str[256]; |
1da177e4 | 5314 | |
968ea6d8 | 5315 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5316 | cpumask_clear(groupmask); |
4dcf6aff IM |
5317 | |
5318 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5319 | ||
5320 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5321 | printk("does not load-balance\n"); |
4dcf6aff | 5322 | if (sd->parent) |
3df0fc5b PZ |
5323 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5324 | " has parent"); | |
4dcf6aff | 5325 | return -1; |
41c7ce9a NP |
5326 | } |
5327 | ||
3df0fc5b | 5328 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5329 | |
758b2cdc | 5330 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5331 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5332 | "CPU%d\n", cpu); | |
4dcf6aff | 5333 | } |
758b2cdc | 5334 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5335 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5336 | " CPU%d\n", cpu); | |
4dcf6aff | 5337 | } |
1da177e4 | 5338 | |
4dcf6aff | 5339 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5340 | do { |
4dcf6aff | 5341 | if (!group) { |
3df0fc5b PZ |
5342 | printk("\n"); |
5343 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5344 | break; |
5345 | } | |
5346 | ||
c3decf0d PZ |
5347 | /* |
5348 | * Even though we initialize ->power to something semi-sane, | |
5349 | * we leave power_orig unset. This allows us to detect if | |
5350 | * domain iteration is still funny without causing /0 traps. | |
5351 | */ | |
5352 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5353 | printk(KERN_CONT "\n"); |
5354 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5355 | "set\n"); | |
4dcf6aff IM |
5356 | break; |
5357 | } | |
1da177e4 | 5358 | |
758b2cdc | 5359 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5360 | printk(KERN_CONT "\n"); |
5361 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5362 | break; |
5363 | } | |
1da177e4 | 5364 | |
cb83b629 PZ |
5365 | if (!(sd->flags & SD_OVERLAP) && |
5366 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5367 | printk(KERN_CONT "\n"); |
5368 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5369 | break; |
5370 | } | |
1da177e4 | 5371 | |
758b2cdc | 5372 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5373 | |
968ea6d8 | 5374 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5375 | |
3df0fc5b | 5376 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5377 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5378 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5379 | group->sgp->power); |
381512cf | 5380 | } |
1da177e4 | 5381 | |
4dcf6aff IM |
5382 | group = group->next; |
5383 | } while (group != sd->groups); | |
3df0fc5b | 5384 | printk(KERN_CONT "\n"); |
1da177e4 | 5385 | |
758b2cdc | 5386 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5387 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5388 | |
758b2cdc RR |
5389 | if (sd->parent && |
5390 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5391 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5392 | "of domain->span\n"); | |
4dcf6aff IM |
5393 | return 0; |
5394 | } | |
1da177e4 | 5395 | |
4dcf6aff IM |
5396 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5397 | { | |
5398 | int level = 0; | |
1da177e4 | 5399 | |
d039ac60 | 5400 | if (!sched_debug_enabled) |
f6630114 MT |
5401 | return; |
5402 | ||
4dcf6aff IM |
5403 | if (!sd) { |
5404 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5405 | return; | |
5406 | } | |
1da177e4 | 5407 | |
4dcf6aff IM |
5408 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5409 | ||
5410 | for (;;) { | |
4cb98839 | 5411 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5412 | break; |
1da177e4 LT |
5413 | level++; |
5414 | sd = sd->parent; | |
33859f7f | 5415 | if (!sd) |
4dcf6aff IM |
5416 | break; |
5417 | } | |
1da177e4 | 5418 | } |
6d6bc0ad | 5419 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5420 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5421 | static inline bool sched_debug(void) |
5422 | { | |
5423 | return false; | |
5424 | } | |
6d6bc0ad | 5425 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5426 | |
1a20ff27 | 5427 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5428 | { |
758b2cdc | 5429 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5430 | return 1; |
5431 | ||
5432 | /* Following flags need at least 2 groups */ | |
5433 | if (sd->flags & (SD_LOAD_BALANCE | | |
5434 | SD_BALANCE_NEWIDLE | | |
5435 | SD_BALANCE_FORK | | |
89c4710e SS |
5436 | SD_BALANCE_EXEC | |
5437 | SD_SHARE_CPUPOWER | | |
5438 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5439 | if (sd->groups != sd->groups->next) |
5440 | return 0; | |
5441 | } | |
5442 | ||
5443 | /* Following flags don't use groups */ | |
c88d5910 | 5444 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5445 | return 0; |
5446 | ||
5447 | return 1; | |
5448 | } | |
5449 | ||
48f24c4d IM |
5450 | static int |
5451 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5452 | { |
5453 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5454 | ||
5455 | if (sd_degenerate(parent)) | |
5456 | return 1; | |
5457 | ||
758b2cdc | 5458 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5459 | return 0; |
5460 | ||
245af2c7 SS |
5461 | /* Flags needing groups don't count if only 1 group in parent */ |
5462 | if (parent->groups == parent->groups->next) { | |
5463 | pflags &= ~(SD_LOAD_BALANCE | | |
5464 | SD_BALANCE_NEWIDLE | | |
5465 | SD_BALANCE_FORK | | |
89c4710e SS |
5466 | SD_BALANCE_EXEC | |
5467 | SD_SHARE_CPUPOWER | | |
5468 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5469 | if (nr_node_ids == 1) |
5470 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5471 | } |
5472 | if (~cflags & pflags) | |
5473 | return 0; | |
5474 | ||
5475 | return 1; | |
5476 | } | |
5477 | ||
dce840a0 | 5478 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5479 | { |
dce840a0 | 5480 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5481 | |
68e74568 | 5482 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5483 | free_cpumask_var(rd->rto_mask); |
5484 | free_cpumask_var(rd->online); | |
5485 | free_cpumask_var(rd->span); | |
5486 | kfree(rd); | |
5487 | } | |
5488 | ||
57d885fe GH |
5489 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5490 | { | |
a0490fa3 | 5491 | struct root_domain *old_rd = NULL; |
57d885fe | 5492 | unsigned long flags; |
57d885fe | 5493 | |
05fa785c | 5494 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5495 | |
5496 | if (rq->rd) { | |
a0490fa3 | 5497 | old_rd = rq->rd; |
57d885fe | 5498 | |
c6c4927b | 5499 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5500 | set_rq_offline(rq); |
57d885fe | 5501 | |
c6c4927b | 5502 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5503 | |
a0490fa3 IM |
5504 | /* |
5505 | * If we dont want to free the old_rt yet then | |
5506 | * set old_rd to NULL to skip the freeing later | |
5507 | * in this function: | |
5508 | */ | |
5509 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5510 | old_rd = NULL; | |
57d885fe GH |
5511 | } |
5512 | ||
5513 | atomic_inc(&rd->refcount); | |
5514 | rq->rd = rd; | |
5515 | ||
c6c4927b | 5516 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5517 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5518 | set_rq_online(rq); |
57d885fe | 5519 | |
05fa785c | 5520 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5521 | |
5522 | if (old_rd) | |
dce840a0 | 5523 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5524 | } |
5525 | ||
68c38fc3 | 5526 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5527 | { |
5528 | memset(rd, 0, sizeof(*rd)); | |
5529 | ||
68c38fc3 | 5530 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5531 | goto out; |
68c38fc3 | 5532 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5533 | goto free_span; |
68c38fc3 | 5534 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5535 | goto free_online; |
6e0534f2 | 5536 | |
68c38fc3 | 5537 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5538 | goto free_rto_mask; |
c6c4927b | 5539 | return 0; |
6e0534f2 | 5540 | |
68e74568 RR |
5541 | free_rto_mask: |
5542 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5543 | free_online: |
5544 | free_cpumask_var(rd->online); | |
5545 | free_span: | |
5546 | free_cpumask_var(rd->span); | |
0c910d28 | 5547 | out: |
c6c4927b | 5548 | return -ENOMEM; |
57d885fe GH |
5549 | } |
5550 | ||
029632fb PZ |
5551 | /* |
5552 | * By default the system creates a single root-domain with all cpus as | |
5553 | * members (mimicking the global state we have today). | |
5554 | */ | |
5555 | struct root_domain def_root_domain; | |
5556 | ||
57d885fe GH |
5557 | static void init_defrootdomain(void) |
5558 | { | |
68c38fc3 | 5559 | init_rootdomain(&def_root_domain); |
c6c4927b | 5560 | |
57d885fe GH |
5561 | atomic_set(&def_root_domain.refcount, 1); |
5562 | } | |
5563 | ||
dc938520 | 5564 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5565 | { |
5566 | struct root_domain *rd; | |
5567 | ||
5568 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5569 | if (!rd) | |
5570 | return NULL; | |
5571 | ||
68c38fc3 | 5572 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5573 | kfree(rd); |
5574 | return NULL; | |
5575 | } | |
57d885fe GH |
5576 | |
5577 | return rd; | |
5578 | } | |
5579 | ||
e3589f6c PZ |
5580 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5581 | { | |
5582 | struct sched_group *tmp, *first; | |
5583 | ||
5584 | if (!sg) | |
5585 | return; | |
5586 | ||
5587 | first = sg; | |
5588 | do { | |
5589 | tmp = sg->next; | |
5590 | ||
5591 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5592 | kfree(sg->sgp); | |
5593 | ||
5594 | kfree(sg); | |
5595 | sg = tmp; | |
5596 | } while (sg != first); | |
5597 | } | |
5598 | ||
dce840a0 PZ |
5599 | static void free_sched_domain(struct rcu_head *rcu) |
5600 | { | |
5601 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5602 | |
5603 | /* | |
5604 | * If its an overlapping domain it has private groups, iterate and | |
5605 | * nuke them all. | |
5606 | */ | |
5607 | if (sd->flags & SD_OVERLAP) { | |
5608 | free_sched_groups(sd->groups, 1); | |
5609 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5610 | kfree(sd->groups->sgp); |
dce840a0 | 5611 | kfree(sd->groups); |
9c3f75cb | 5612 | } |
dce840a0 PZ |
5613 | kfree(sd); |
5614 | } | |
5615 | ||
5616 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5617 | { | |
5618 | call_rcu(&sd->rcu, free_sched_domain); | |
5619 | } | |
5620 | ||
5621 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5622 | { | |
5623 | for (; sd; sd = sd->parent) | |
5624 | destroy_sched_domain(sd, cpu); | |
5625 | } | |
5626 | ||
518cd623 PZ |
5627 | /* |
5628 | * Keep a special pointer to the highest sched_domain that has | |
5629 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5630 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5631 | * | |
5632 | * Also keep a unique ID per domain (we use the first cpu number in | |
5633 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5634 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5635 | */ |
5636 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5637 | DEFINE_PER_CPU(int, sd_llc_id); | |
5638 | ||
5639 | static void update_top_cache_domain(int cpu) | |
5640 | { | |
5641 | struct sched_domain *sd; | |
5642 | int id = cpu; | |
5643 | ||
5644 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
37407ea7 | 5645 | if (sd) |
518cd623 PZ |
5646 | id = cpumask_first(sched_domain_span(sd)); |
5647 | ||
5648 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5649 | per_cpu(sd_llc_id, cpu) = id; | |
5650 | } | |
5651 | ||
1da177e4 | 5652 | /* |
0eab9146 | 5653 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5654 | * hold the hotplug lock. |
5655 | */ | |
0eab9146 IM |
5656 | static void |
5657 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5658 | { |
70b97a7f | 5659 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5660 | struct sched_domain *tmp; |
5661 | ||
5662 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5663 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5664 | struct sched_domain *parent = tmp->parent; |
5665 | if (!parent) | |
5666 | break; | |
f29c9b1c | 5667 | |
1a848870 | 5668 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5669 | tmp->parent = parent->parent; |
1a848870 SS |
5670 | if (parent->parent) |
5671 | parent->parent->child = tmp; | |
dce840a0 | 5672 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5673 | } else |
5674 | tmp = tmp->parent; | |
245af2c7 SS |
5675 | } |
5676 | ||
1a848870 | 5677 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5678 | tmp = sd; |
245af2c7 | 5679 | sd = sd->parent; |
dce840a0 | 5680 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5681 | if (sd) |
5682 | sd->child = NULL; | |
5683 | } | |
1da177e4 | 5684 | |
4cb98839 | 5685 | sched_domain_debug(sd, cpu); |
1da177e4 | 5686 | |
57d885fe | 5687 | rq_attach_root(rq, rd); |
dce840a0 | 5688 | tmp = rq->sd; |
674311d5 | 5689 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5690 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5691 | |
5692 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5693 | } |
5694 | ||
5695 | /* cpus with isolated domains */ | |
dcc30a35 | 5696 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5697 | |
5698 | /* Setup the mask of cpus configured for isolated domains */ | |
5699 | static int __init isolated_cpu_setup(char *str) | |
5700 | { | |
bdddd296 | 5701 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5702 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5703 | return 1; |
5704 | } | |
5705 | ||
8927f494 | 5706 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5707 | |
d3081f52 PZ |
5708 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5709 | { | |
5710 | return cpumask_of_node(cpu_to_node(cpu)); | |
5711 | } | |
5712 | ||
dce840a0 PZ |
5713 | struct sd_data { |
5714 | struct sched_domain **__percpu sd; | |
5715 | struct sched_group **__percpu sg; | |
9c3f75cb | 5716 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5717 | }; |
5718 | ||
49a02c51 | 5719 | struct s_data { |
21d42ccf | 5720 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5721 | struct root_domain *rd; |
5722 | }; | |
5723 | ||
2109b99e | 5724 | enum s_alloc { |
2109b99e | 5725 | sa_rootdomain, |
21d42ccf | 5726 | sa_sd, |
dce840a0 | 5727 | sa_sd_storage, |
2109b99e AH |
5728 | sa_none, |
5729 | }; | |
5730 | ||
54ab4ff4 PZ |
5731 | struct sched_domain_topology_level; |
5732 | ||
5733 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5734 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5735 | ||
e3589f6c PZ |
5736 | #define SDTL_OVERLAP 0x01 |
5737 | ||
eb7a74e6 | 5738 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5739 | sched_domain_init_f init; |
5740 | sched_domain_mask_f mask; | |
e3589f6c | 5741 | int flags; |
cb83b629 | 5742 | int numa_level; |
54ab4ff4 | 5743 | struct sd_data data; |
eb7a74e6 PZ |
5744 | }; |
5745 | ||
c1174876 PZ |
5746 | /* |
5747 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5748 | * domain traversal. | |
5749 | * | |
5750 | * Asymmetric node setups can result in situations where the domain tree is of | |
5751 | * unequal depth, make sure to skip domains that already cover the entire | |
5752 | * range. | |
5753 | * | |
5754 | * In that case build_sched_domains() will have terminated the iteration early | |
5755 | * and our sibling sd spans will be empty. Domains should always include the | |
5756 | * cpu they're built on, so check that. | |
5757 | * | |
5758 | */ | |
5759 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5760 | { | |
5761 | const struct cpumask *span = sched_domain_span(sd); | |
5762 | struct sd_data *sdd = sd->private; | |
5763 | struct sched_domain *sibling; | |
5764 | int i; | |
5765 | ||
5766 | for_each_cpu(i, span) { | |
5767 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5768 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5769 | continue; | |
5770 | ||
5771 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5772 | } | |
5773 | } | |
5774 | ||
5775 | /* | |
5776 | * Return the canonical balance cpu for this group, this is the first cpu | |
5777 | * of this group that's also in the iteration mask. | |
5778 | */ | |
5779 | int group_balance_cpu(struct sched_group *sg) | |
5780 | { | |
5781 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5782 | } | |
5783 | ||
e3589f6c PZ |
5784 | static int |
5785 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5786 | { | |
5787 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5788 | const struct cpumask *span = sched_domain_span(sd); | |
5789 | struct cpumask *covered = sched_domains_tmpmask; | |
5790 | struct sd_data *sdd = sd->private; | |
5791 | struct sched_domain *child; | |
5792 | int i; | |
5793 | ||
5794 | cpumask_clear(covered); | |
5795 | ||
5796 | for_each_cpu(i, span) { | |
5797 | struct cpumask *sg_span; | |
5798 | ||
5799 | if (cpumask_test_cpu(i, covered)) | |
5800 | continue; | |
5801 | ||
c1174876 PZ |
5802 | child = *per_cpu_ptr(sdd->sd, i); |
5803 | ||
5804 | /* See the comment near build_group_mask(). */ | |
5805 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5806 | continue; | |
5807 | ||
e3589f6c | 5808 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5809 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5810 | |
5811 | if (!sg) | |
5812 | goto fail; | |
5813 | ||
5814 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5815 | if (child->child) { |
5816 | child = child->child; | |
5817 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5818 | } else | |
5819 | cpumask_set_cpu(i, sg_span); | |
5820 | ||
5821 | cpumask_or(covered, covered, sg_span); | |
5822 | ||
74a5ce20 | 5823 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5824 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5825 | build_group_mask(sd, sg); | |
5826 | ||
c3decf0d PZ |
5827 | /* |
5828 | * Initialize sgp->power such that even if we mess up the | |
5829 | * domains and no possible iteration will get us here, we won't | |
5830 | * die on a /0 trap. | |
5831 | */ | |
5832 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 5833 | |
c1174876 PZ |
5834 | /* |
5835 | * Make sure the first group of this domain contains the | |
5836 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5837 | * breaks. See update_sg_lb_stats(). | |
5838 | */ | |
74a5ce20 | 5839 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5840 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5841 | groups = sg; |
5842 | ||
5843 | if (!first) | |
5844 | first = sg; | |
5845 | if (last) | |
5846 | last->next = sg; | |
5847 | last = sg; | |
5848 | last->next = first; | |
5849 | } | |
5850 | sd->groups = groups; | |
5851 | ||
5852 | return 0; | |
5853 | ||
5854 | fail: | |
5855 | free_sched_groups(first, 0); | |
5856 | ||
5857 | return -ENOMEM; | |
5858 | } | |
5859 | ||
dce840a0 | 5860 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5861 | { |
dce840a0 PZ |
5862 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5863 | struct sched_domain *child = sd->child; | |
1da177e4 | 5864 | |
dce840a0 PZ |
5865 | if (child) |
5866 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5867 | |
9c3f75cb | 5868 | if (sg) { |
dce840a0 | 5869 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5870 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5871 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5872 | } |
dce840a0 PZ |
5873 | |
5874 | return cpu; | |
1e9f28fa | 5875 | } |
1e9f28fa | 5876 | |
01a08546 | 5877 | /* |
dce840a0 PZ |
5878 | * build_sched_groups will build a circular linked list of the groups |
5879 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5880 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5881 | * |
5882 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5883 | */ |
e3589f6c PZ |
5884 | static int |
5885 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5886 | { |
dce840a0 PZ |
5887 | struct sched_group *first = NULL, *last = NULL; |
5888 | struct sd_data *sdd = sd->private; | |
5889 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5890 | struct cpumask *covered; |
dce840a0 | 5891 | int i; |
9c1cfda2 | 5892 | |
e3589f6c PZ |
5893 | get_group(cpu, sdd, &sd->groups); |
5894 | atomic_inc(&sd->groups->ref); | |
5895 | ||
5896 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
5897 | return 0; | |
5898 | ||
f96225fd PZ |
5899 | lockdep_assert_held(&sched_domains_mutex); |
5900 | covered = sched_domains_tmpmask; | |
5901 | ||
dce840a0 | 5902 | cpumask_clear(covered); |
6711cab4 | 5903 | |
dce840a0 PZ |
5904 | for_each_cpu(i, span) { |
5905 | struct sched_group *sg; | |
5906 | int group = get_group(i, sdd, &sg); | |
5907 | int j; | |
6711cab4 | 5908 | |
dce840a0 PZ |
5909 | if (cpumask_test_cpu(i, covered)) |
5910 | continue; | |
6711cab4 | 5911 | |
dce840a0 | 5912 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5913 | sg->sgp->power = 0; |
c1174876 | 5914 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5915 | |
dce840a0 PZ |
5916 | for_each_cpu(j, span) { |
5917 | if (get_group(j, sdd, NULL) != group) | |
5918 | continue; | |
0601a88d | 5919 | |
dce840a0 PZ |
5920 | cpumask_set_cpu(j, covered); |
5921 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5922 | } | |
0601a88d | 5923 | |
dce840a0 PZ |
5924 | if (!first) |
5925 | first = sg; | |
5926 | if (last) | |
5927 | last->next = sg; | |
5928 | last = sg; | |
5929 | } | |
5930 | last->next = first; | |
e3589f6c PZ |
5931 | |
5932 | return 0; | |
0601a88d | 5933 | } |
51888ca2 | 5934 | |
89c4710e SS |
5935 | /* |
5936 | * Initialize sched groups cpu_power. | |
5937 | * | |
5938 | * cpu_power indicates the capacity of sched group, which is used while | |
5939 | * distributing the load between different sched groups in a sched domain. | |
5940 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5941 | * there are asymmetries in the topology. If there are asymmetries, group | |
5942 | * having more cpu_power will pickup more load compared to the group having | |
5943 | * less cpu_power. | |
89c4710e SS |
5944 | */ |
5945 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5946 | { | |
e3589f6c | 5947 | struct sched_group *sg = sd->groups; |
89c4710e | 5948 | |
e3589f6c PZ |
5949 | WARN_ON(!sd || !sg); |
5950 | ||
5951 | do { | |
5952 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5953 | sg = sg->next; | |
5954 | } while (sg != sd->groups); | |
89c4710e | 5955 | |
c1174876 | 5956 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5957 | return; |
aae6d3dd | 5958 | |
d274cb30 | 5959 | update_group_power(sd, cpu); |
69e1e811 | 5960 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5961 | } |
5962 | ||
029632fb PZ |
5963 | int __weak arch_sd_sibling_asym_packing(void) |
5964 | { | |
5965 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5966 | } |
5967 | ||
7c16ec58 MT |
5968 | /* |
5969 | * Initializers for schedule domains | |
5970 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5971 | */ | |
5972 | ||
a5d8c348 IM |
5973 | #ifdef CONFIG_SCHED_DEBUG |
5974 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5975 | #else | |
5976 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5977 | #endif | |
5978 | ||
54ab4ff4 PZ |
5979 | #define SD_INIT_FUNC(type) \ |
5980 | static noinline struct sched_domain * \ | |
5981 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5982 | { \ | |
5983 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5984 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5985 | SD_INIT_NAME(sd, type); \ |
5986 | sd->private = &tl->data; \ | |
5987 | return sd; \ | |
7c16ec58 MT |
5988 | } |
5989 | ||
5990 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5991 | #ifdef CONFIG_SCHED_SMT |
5992 | SD_INIT_FUNC(SIBLING) | |
5993 | #endif | |
5994 | #ifdef CONFIG_SCHED_MC | |
5995 | SD_INIT_FUNC(MC) | |
5996 | #endif | |
01a08546 HC |
5997 | #ifdef CONFIG_SCHED_BOOK |
5998 | SD_INIT_FUNC(BOOK) | |
5999 | #endif | |
7c16ec58 | 6000 | |
1d3504fc | 6001 | static int default_relax_domain_level = -1; |
60495e77 | 6002 | int sched_domain_level_max; |
1d3504fc HS |
6003 | |
6004 | static int __init setup_relax_domain_level(char *str) | |
6005 | { | |
a841f8ce DS |
6006 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6007 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6008 | |
1d3504fc HS |
6009 | return 1; |
6010 | } | |
6011 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6012 | ||
6013 | static void set_domain_attribute(struct sched_domain *sd, | |
6014 | struct sched_domain_attr *attr) | |
6015 | { | |
6016 | int request; | |
6017 | ||
6018 | if (!attr || attr->relax_domain_level < 0) { | |
6019 | if (default_relax_domain_level < 0) | |
6020 | return; | |
6021 | else | |
6022 | request = default_relax_domain_level; | |
6023 | } else | |
6024 | request = attr->relax_domain_level; | |
6025 | if (request < sd->level) { | |
6026 | /* turn off idle balance on this domain */ | |
c88d5910 | 6027 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6028 | } else { |
6029 | /* turn on idle balance on this domain */ | |
c88d5910 | 6030 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6031 | } |
6032 | } | |
6033 | ||
54ab4ff4 PZ |
6034 | static void __sdt_free(const struct cpumask *cpu_map); |
6035 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6036 | ||
2109b99e AH |
6037 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6038 | const struct cpumask *cpu_map) | |
6039 | { | |
6040 | switch (what) { | |
2109b99e | 6041 | case sa_rootdomain: |
822ff793 PZ |
6042 | if (!atomic_read(&d->rd->refcount)) |
6043 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6044 | case sa_sd: |
6045 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6046 | case sa_sd_storage: |
54ab4ff4 | 6047 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6048 | case sa_none: |
6049 | break; | |
6050 | } | |
6051 | } | |
3404c8d9 | 6052 | |
2109b99e AH |
6053 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6054 | const struct cpumask *cpu_map) | |
6055 | { | |
dce840a0 PZ |
6056 | memset(d, 0, sizeof(*d)); |
6057 | ||
54ab4ff4 PZ |
6058 | if (__sdt_alloc(cpu_map)) |
6059 | return sa_sd_storage; | |
dce840a0 PZ |
6060 | d->sd = alloc_percpu(struct sched_domain *); |
6061 | if (!d->sd) | |
6062 | return sa_sd_storage; | |
2109b99e | 6063 | d->rd = alloc_rootdomain(); |
dce840a0 | 6064 | if (!d->rd) |
21d42ccf | 6065 | return sa_sd; |
2109b99e AH |
6066 | return sa_rootdomain; |
6067 | } | |
57d885fe | 6068 | |
dce840a0 PZ |
6069 | /* |
6070 | * NULL the sd_data elements we've used to build the sched_domain and | |
6071 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6072 | * will not free the data we're using. | |
6073 | */ | |
6074 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6075 | { | |
6076 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6077 | |
6078 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6079 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6080 | ||
e3589f6c | 6081 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6082 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6083 | |
6084 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6085 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6086 | } |
6087 | ||
2c402dc3 PZ |
6088 | #ifdef CONFIG_SCHED_SMT |
6089 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6090 | { |
2c402dc3 | 6091 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6092 | } |
2c402dc3 | 6093 | #endif |
7f4588f3 | 6094 | |
d069b916 PZ |
6095 | /* |
6096 | * Topology list, bottom-up. | |
6097 | */ | |
2c402dc3 | 6098 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6099 | #ifdef CONFIG_SCHED_SMT |
6100 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6101 | #endif |
1e9f28fa | 6102 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6103 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6104 | #endif |
d069b916 PZ |
6105 | #ifdef CONFIG_SCHED_BOOK |
6106 | { sd_init_BOOK, cpu_book_mask, }, | |
6107 | #endif | |
6108 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
6109 | { NULL, }, |
6110 | }; | |
6111 | ||
6112 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6113 | ||
cb83b629 PZ |
6114 | #ifdef CONFIG_NUMA |
6115 | ||
6116 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
6117 | static int *sched_domains_numa_distance; |
6118 | static struct cpumask ***sched_domains_numa_masks; | |
6119 | static int sched_domains_curr_level; | |
6120 | ||
cb83b629 PZ |
6121 | static inline int sd_local_flags(int level) |
6122 | { | |
10717dcd | 6123 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
6124 | return 0; |
6125 | ||
6126 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6127 | } | |
6128 | ||
6129 | static struct sched_domain * | |
6130 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6131 | { | |
6132 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6133 | int level = tl->numa_level; | |
6134 | int sd_weight = cpumask_weight( | |
6135 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6136 | ||
6137 | *sd = (struct sched_domain){ | |
6138 | .min_interval = sd_weight, | |
6139 | .max_interval = 2*sd_weight, | |
6140 | .busy_factor = 32, | |
870a0bb5 | 6141 | .imbalance_pct = 125, |
cb83b629 PZ |
6142 | .cache_nice_tries = 2, |
6143 | .busy_idx = 3, | |
6144 | .idle_idx = 2, | |
6145 | .newidle_idx = 0, | |
6146 | .wake_idx = 0, | |
6147 | .forkexec_idx = 0, | |
6148 | ||
6149 | .flags = 1*SD_LOAD_BALANCE | |
6150 | | 1*SD_BALANCE_NEWIDLE | |
6151 | | 0*SD_BALANCE_EXEC | |
6152 | | 0*SD_BALANCE_FORK | |
6153 | | 0*SD_BALANCE_WAKE | |
6154 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 6155 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
6156 | | 0*SD_SHARE_PKG_RESOURCES |
6157 | | 1*SD_SERIALIZE | |
6158 | | 0*SD_PREFER_SIBLING | |
6159 | | sd_local_flags(level) | |
6160 | , | |
6161 | .last_balance = jiffies, | |
6162 | .balance_interval = sd_weight, | |
6163 | }; | |
6164 | SD_INIT_NAME(sd, NUMA); | |
6165 | sd->private = &tl->data; | |
6166 | ||
6167 | /* | |
6168 | * Ugly hack to pass state to sd_numa_mask()... | |
6169 | */ | |
6170 | sched_domains_curr_level = tl->numa_level; | |
6171 | ||
6172 | return sd; | |
6173 | } | |
6174 | ||
6175 | static const struct cpumask *sd_numa_mask(int cpu) | |
6176 | { | |
6177 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6178 | } | |
6179 | ||
d039ac60 PZ |
6180 | static void sched_numa_warn(const char *str) |
6181 | { | |
6182 | static int done = false; | |
6183 | int i,j; | |
6184 | ||
6185 | if (done) | |
6186 | return; | |
6187 | ||
6188 | done = true; | |
6189 | ||
6190 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6191 | ||
6192 | for (i = 0; i < nr_node_ids; i++) { | |
6193 | printk(KERN_WARNING " "); | |
6194 | for (j = 0; j < nr_node_ids; j++) | |
6195 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6196 | printk(KERN_CONT "\n"); | |
6197 | } | |
6198 | printk(KERN_WARNING "\n"); | |
6199 | } | |
6200 | ||
6201 | static bool find_numa_distance(int distance) | |
6202 | { | |
6203 | int i; | |
6204 | ||
6205 | if (distance == node_distance(0, 0)) | |
6206 | return true; | |
6207 | ||
6208 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6209 | if (sched_domains_numa_distance[i] == distance) | |
6210 | return true; | |
6211 | } | |
6212 | ||
6213 | return false; | |
6214 | } | |
6215 | ||
cb83b629 PZ |
6216 | static void sched_init_numa(void) |
6217 | { | |
6218 | int next_distance, curr_distance = node_distance(0, 0); | |
6219 | struct sched_domain_topology_level *tl; | |
6220 | int level = 0; | |
6221 | int i, j, k; | |
6222 | ||
cb83b629 PZ |
6223 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6224 | if (!sched_domains_numa_distance) | |
6225 | return; | |
6226 | ||
6227 | /* | |
6228 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6229 | * unique distances in the node_distance() table. | |
6230 | * | |
6231 | * Assumes node_distance(0,j) includes all distances in | |
6232 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6233 | */ |
6234 | next_distance = curr_distance; | |
6235 | for (i = 0; i < nr_node_ids; i++) { | |
6236 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6237 | for (k = 0; k < nr_node_ids; k++) { |
6238 | int distance = node_distance(i, k); | |
6239 | ||
6240 | if (distance > curr_distance && | |
6241 | (distance < next_distance || | |
6242 | next_distance == curr_distance)) | |
6243 | next_distance = distance; | |
6244 | ||
6245 | /* | |
6246 | * While not a strong assumption it would be nice to know | |
6247 | * about cases where if node A is connected to B, B is not | |
6248 | * equally connected to A. | |
6249 | */ | |
6250 | if (sched_debug() && node_distance(k, i) != distance) | |
6251 | sched_numa_warn("Node-distance not symmetric"); | |
6252 | ||
6253 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6254 | sched_numa_warn("Node-0 not representative"); | |
6255 | } | |
6256 | if (next_distance != curr_distance) { | |
6257 | sched_domains_numa_distance[level++] = next_distance; | |
6258 | sched_domains_numa_levels = level; | |
6259 | curr_distance = next_distance; | |
6260 | } else break; | |
cb83b629 | 6261 | } |
d039ac60 PZ |
6262 | |
6263 | /* | |
6264 | * In case of sched_debug() we verify the above assumption. | |
6265 | */ | |
6266 | if (!sched_debug()) | |
6267 | break; | |
cb83b629 PZ |
6268 | } |
6269 | /* | |
6270 | * 'level' contains the number of unique distances, excluding the | |
6271 | * identity distance node_distance(i,i). | |
6272 | * | |
6273 | * The sched_domains_nume_distance[] array includes the actual distance | |
6274 | * numbers. | |
6275 | */ | |
6276 | ||
5f7865f3 TC |
6277 | /* |
6278 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6279 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6280 | * the array will contain less then 'level' members. This could be | |
6281 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6282 | * in other functions. | |
6283 | * | |
6284 | * We reset it to 'level' at the end of this function. | |
6285 | */ | |
6286 | sched_domains_numa_levels = 0; | |
6287 | ||
cb83b629 PZ |
6288 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6289 | if (!sched_domains_numa_masks) | |
6290 | return; | |
6291 | ||
6292 | /* | |
6293 | * Now for each level, construct a mask per node which contains all | |
6294 | * cpus of nodes that are that many hops away from us. | |
6295 | */ | |
6296 | for (i = 0; i < level; i++) { | |
6297 | sched_domains_numa_masks[i] = | |
6298 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6299 | if (!sched_domains_numa_masks[i]) | |
6300 | return; | |
6301 | ||
6302 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6303 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6304 | if (!mask) |
6305 | return; | |
6306 | ||
6307 | sched_domains_numa_masks[i][j] = mask; | |
6308 | ||
6309 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6310 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6311 | continue; |
6312 | ||
6313 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6314 | } | |
6315 | } | |
6316 | } | |
6317 | ||
6318 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6319 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6320 | if (!tl) | |
6321 | return; | |
6322 | ||
6323 | /* | |
6324 | * Copy the default topology bits.. | |
6325 | */ | |
6326 | for (i = 0; default_topology[i].init; i++) | |
6327 | tl[i] = default_topology[i]; | |
6328 | ||
6329 | /* | |
6330 | * .. and append 'j' levels of NUMA goodness. | |
6331 | */ | |
6332 | for (j = 0; j < level; i++, j++) { | |
6333 | tl[i] = (struct sched_domain_topology_level){ | |
6334 | .init = sd_numa_init, | |
6335 | .mask = sd_numa_mask, | |
6336 | .flags = SDTL_OVERLAP, | |
6337 | .numa_level = j, | |
6338 | }; | |
6339 | } | |
6340 | ||
6341 | sched_domain_topology = tl; | |
5f7865f3 TC |
6342 | |
6343 | sched_domains_numa_levels = level; | |
cb83b629 | 6344 | } |
301a5cba TC |
6345 | |
6346 | static void sched_domains_numa_masks_set(int cpu) | |
6347 | { | |
6348 | int i, j; | |
6349 | int node = cpu_to_node(cpu); | |
6350 | ||
6351 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6352 | for (j = 0; j < nr_node_ids; j++) { | |
6353 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6354 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6355 | } | |
6356 | } | |
6357 | } | |
6358 | ||
6359 | static void sched_domains_numa_masks_clear(int cpu) | |
6360 | { | |
6361 | int i, j; | |
6362 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6363 | for (j = 0; j < nr_node_ids; j++) | |
6364 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6365 | } | |
6366 | } | |
6367 | ||
6368 | /* | |
6369 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6370 | * are onlined. | |
6371 | */ | |
6372 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6373 | unsigned long action, | |
6374 | void *hcpu) | |
6375 | { | |
6376 | int cpu = (long)hcpu; | |
6377 | ||
6378 | switch (action & ~CPU_TASKS_FROZEN) { | |
6379 | case CPU_ONLINE: | |
6380 | sched_domains_numa_masks_set(cpu); | |
6381 | break; | |
6382 | ||
6383 | case CPU_DEAD: | |
6384 | sched_domains_numa_masks_clear(cpu); | |
6385 | break; | |
6386 | ||
6387 | default: | |
6388 | return NOTIFY_DONE; | |
6389 | } | |
6390 | ||
6391 | return NOTIFY_OK; | |
cb83b629 PZ |
6392 | } |
6393 | #else | |
6394 | static inline void sched_init_numa(void) | |
6395 | { | |
6396 | } | |
301a5cba TC |
6397 | |
6398 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6399 | unsigned long action, | |
6400 | void *hcpu) | |
6401 | { | |
6402 | return 0; | |
6403 | } | |
cb83b629 PZ |
6404 | #endif /* CONFIG_NUMA */ |
6405 | ||
54ab4ff4 PZ |
6406 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6407 | { | |
6408 | struct sched_domain_topology_level *tl; | |
6409 | int j; | |
6410 | ||
6411 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6412 | struct sd_data *sdd = &tl->data; | |
6413 | ||
6414 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6415 | if (!sdd->sd) | |
6416 | return -ENOMEM; | |
6417 | ||
6418 | sdd->sg = alloc_percpu(struct sched_group *); | |
6419 | if (!sdd->sg) | |
6420 | return -ENOMEM; | |
6421 | ||
9c3f75cb PZ |
6422 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6423 | if (!sdd->sgp) | |
6424 | return -ENOMEM; | |
6425 | ||
54ab4ff4 PZ |
6426 | for_each_cpu(j, cpu_map) { |
6427 | struct sched_domain *sd; | |
6428 | struct sched_group *sg; | |
9c3f75cb | 6429 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6430 | |
6431 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6432 | GFP_KERNEL, cpu_to_node(j)); | |
6433 | if (!sd) | |
6434 | return -ENOMEM; | |
6435 | ||
6436 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6437 | ||
6438 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6439 | GFP_KERNEL, cpu_to_node(j)); | |
6440 | if (!sg) | |
6441 | return -ENOMEM; | |
6442 | ||
30b4e9eb IM |
6443 | sg->next = sg; |
6444 | ||
54ab4ff4 | 6445 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6446 | |
c1174876 | 6447 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6448 | GFP_KERNEL, cpu_to_node(j)); |
6449 | if (!sgp) | |
6450 | return -ENOMEM; | |
6451 | ||
6452 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6453 | } |
6454 | } | |
6455 | ||
6456 | return 0; | |
6457 | } | |
6458 | ||
6459 | static void __sdt_free(const struct cpumask *cpu_map) | |
6460 | { | |
6461 | struct sched_domain_topology_level *tl; | |
6462 | int j; | |
6463 | ||
6464 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6465 | struct sd_data *sdd = &tl->data; | |
6466 | ||
6467 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6468 | struct sched_domain *sd; |
6469 | ||
6470 | if (sdd->sd) { | |
6471 | sd = *per_cpu_ptr(sdd->sd, j); | |
6472 | if (sd && (sd->flags & SD_OVERLAP)) | |
6473 | free_sched_groups(sd->groups, 0); | |
6474 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6475 | } | |
6476 | ||
6477 | if (sdd->sg) | |
6478 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6479 | if (sdd->sgp) | |
6480 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6481 | } |
6482 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6483 | sdd->sd = NULL; |
54ab4ff4 | 6484 | free_percpu(sdd->sg); |
fb2cf2c6 | 6485 | sdd->sg = NULL; |
9c3f75cb | 6486 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6487 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6488 | } |
6489 | } | |
6490 | ||
2c402dc3 PZ |
6491 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6492 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6493 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6494 | int cpu) |
6495 | { | |
54ab4ff4 | 6496 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6497 | if (!sd) |
d069b916 | 6498 | return child; |
2c402dc3 | 6499 | |
2c402dc3 | 6500 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6501 | if (child) { |
6502 | sd->level = child->level + 1; | |
6503 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6504 | child->parent = sd; |
60495e77 | 6505 | } |
d069b916 | 6506 | sd->child = child; |
a841f8ce | 6507 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6508 | |
6509 | return sd; | |
6510 | } | |
6511 | ||
2109b99e AH |
6512 | /* |
6513 | * Build sched domains for a given set of cpus and attach the sched domains | |
6514 | * to the individual cpus | |
6515 | */ | |
dce840a0 PZ |
6516 | static int build_sched_domains(const struct cpumask *cpu_map, |
6517 | struct sched_domain_attr *attr) | |
2109b99e AH |
6518 | { |
6519 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6520 | struct sched_domain *sd; |
2109b99e | 6521 | struct s_data d; |
822ff793 | 6522 | int i, ret = -ENOMEM; |
9c1cfda2 | 6523 | |
2109b99e AH |
6524 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6525 | if (alloc_state != sa_rootdomain) | |
6526 | goto error; | |
9c1cfda2 | 6527 | |
dce840a0 | 6528 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6529 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6530 | struct sched_domain_topology_level *tl; |
6531 | ||
3bd65a80 | 6532 | sd = NULL; |
e3589f6c | 6533 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6534 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6535 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6536 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6537 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6538 | break; | |
e3589f6c | 6539 | } |
d274cb30 | 6540 | |
d069b916 PZ |
6541 | while (sd->child) |
6542 | sd = sd->child; | |
6543 | ||
21d42ccf | 6544 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6545 | } |
6546 | ||
6547 | /* Build the groups for the domains */ | |
6548 | for_each_cpu(i, cpu_map) { | |
6549 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6550 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6551 | if (sd->flags & SD_OVERLAP) { |
6552 | if (build_overlap_sched_groups(sd, i)) | |
6553 | goto error; | |
6554 | } else { | |
6555 | if (build_sched_groups(sd, i)) | |
6556 | goto error; | |
6557 | } | |
1cf51902 | 6558 | } |
a06dadbe | 6559 | } |
9c1cfda2 | 6560 | |
1da177e4 | 6561 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6562 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6563 | if (!cpumask_test_cpu(i, cpu_map)) | |
6564 | continue; | |
9c1cfda2 | 6565 | |
dce840a0 PZ |
6566 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6567 | claim_allocations(i, sd); | |
cd4ea6ae | 6568 | init_sched_groups_power(i, sd); |
dce840a0 | 6569 | } |
f712c0c7 | 6570 | } |
9c1cfda2 | 6571 | |
1da177e4 | 6572 | /* Attach the domains */ |
dce840a0 | 6573 | rcu_read_lock(); |
abcd083a | 6574 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6575 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6576 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6577 | } |
dce840a0 | 6578 | rcu_read_unlock(); |
51888ca2 | 6579 | |
822ff793 | 6580 | ret = 0; |
51888ca2 | 6581 | error: |
2109b99e | 6582 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6583 | return ret; |
1da177e4 | 6584 | } |
029190c5 | 6585 | |
acc3f5d7 | 6586 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6587 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6588 | static struct sched_domain_attr *dattr_cur; |
6589 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6590 | |
6591 | /* | |
6592 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6593 | * cpumask) fails, then fallback to a single sched domain, |
6594 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6595 | */ |
4212823f | 6596 | static cpumask_var_t fallback_doms; |
029190c5 | 6597 | |
ee79d1bd HC |
6598 | /* |
6599 | * arch_update_cpu_topology lets virtualized architectures update the | |
6600 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6601 | * or 0 if it stayed the same. | |
6602 | */ | |
6603 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6604 | { |
ee79d1bd | 6605 | return 0; |
22e52b07 HC |
6606 | } |
6607 | ||
acc3f5d7 RR |
6608 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6609 | { | |
6610 | int i; | |
6611 | cpumask_var_t *doms; | |
6612 | ||
6613 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6614 | if (!doms) | |
6615 | return NULL; | |
6616 | for (i = 0; i < ndoms; i++) { | |
6617 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6618 | free_sched_domains(doms, i); | |
6619 | return NULL; | |
6620 | } | |
6621 | } | |
6622 | return doms; | |
6623 | } | |
6624 | ||
6625 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6626 | { | |
6627 | unsigned int i; | |
6628 | for (i = 0; i < ndoms; i++) | |
6629 | free_cpumask_var(doms[i]); | |
6630 | kfree(doms); | |
6631 | } | |
6632 | ||
1a20ff27 | 6633 | /* |
41a2d6cf | 6634 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6635 | * For now this just excludes isolated cpus, but could be used to |
6636 | * exclude other special cases in the future. | |
1a20ff27 | 6637 | */ |
c4a8849a | 6638 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6639 | { |
7378547f MM |
6640 | int err; |
6641 | ||
22e52b07 | 6642 | arch_update_cpu_topology(); |
029190c5 | 6643 | ndoms_cur = 1; |
acc3f5d7 | 6644 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6645 | if (!doms_cur) |
acc3f5d7 RR |
6646 | doms_cur = &fallback_doms; |
6647 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6648 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6649 | register_sched_domain_sysctl(); |
7378547f MM |
6650 | |
6651 | return err; | |
1a20ff27 DG |
6652 | } |
6653 | ||
1a20ff27 DG |
6654 | /* |
6655 | * Detach sched domains from a group of cpus specified in cpu_map | |
6656 | * These cpus will now be attached to the NULL domain | |
6657 | */ | |
96f874e2 | 6658 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6659 | { |
6660 | int i; | |
6661 | ||
dce840a0 | 6662 | rcu_read_lock(); |
abcd083a | 6663 | for_each_cpu(i, cpu_map) |
57d885fe | 6664 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6665 | rcu_read_unlock(); |
1a20ff27 DG |
6666 | } |
6667 | ||
1d3504fc HS |
6668 | /* handle null as "default" */ |
6669 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6670 | struct sched_domain_attr *new, int idx_new) | |
6671 | { | |
6672 | struct sched_domain_attr tmp; | |
6673 | ||
6674 | /* fast path */ | |
6675 | if (!new && !cur) | |
6676 | return 1; | |
6677 | ||
6678 | tmp = SD_ATTR_INIT; | |
6679 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6680 | new ? (new + idx_new) : &tmp, | |
6681 | sizeof(struct sched_domain_attr)); | |
6682 | } | |
6683 | ||
029190c5 PJ |
6684 | /* |
6685 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6686 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6687 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6688 | * It destroys each deleted domain and builds each new domain. | |
6689 | * | |
acc3f5d7 | 6690 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6691 | * The masks don't intersect (don't overlap.) We should setup one |
6692 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6693 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6694 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6695 | * it as it is. | |
6696 | * | |
acc3f5d7 RR |
6697 | * The passed in 'doms_new' should be allocated using |
6698 | * alloc_sched_domains. This routine takes ownership of it and will | |
6699 | * free_sched_domains it when done with it. If the caller failed the | |
6700 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6701 | * and partition_sched_domains() will fallback to the single partition | |
6702 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6703 | * |
96f874e2 | 6704 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6705 | * ndoms_new == 0 is a special case for destroying existing domains, |
6706 | * and it will not create the default domain. | |
dfb512ec | 6707 | * |
029190c5 PJ |
6708 | * Call with hotplug lock held |
6709 | */ | |
acc3f5d7 | 6710 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6711 | struct sched_domain_attr *dattr_new) |
029190c5 | 6712 | { |
dfb512ec | 6713 | int i, j, n; |
d65bd5ec | 6714 | int new_topology; |
029190c5 | 6715 | |
712555ee | 6716 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6717 | |
7378547f MM |
6718 | /* always unregister in case we don't destroy any domains */ |
6719 | unregister_sched_domain_sysctl(); | |
6720 | ||
d65bd5ec HC |
6721 | /* Let architecture update cpu core mappings. */ |
6722 | new_topology = arch_update_cpu_topology(); | |
6723 | ||
dfb512ec | 6724 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6725 | |
6726 | /* Destroy deleted domains */ | |
6727 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6728 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6729 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6730 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6731 | goto match1; |
6732 | } | |
6733 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6734 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6735 | match1: |
6736 | ; | |
6737 | } | |
6738 | ||
e761b772 MK |
6739 | if (doms_new == NULL) { |
6740 | ndoms_cur = 0; | |
acc3f5d7 | 6741 | doms_new = &fallback_doms; |
6ad4c188 | 6742 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6743 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6744 | } |
6745 | ||
029190c5 PJ |
6746 | /* Build new domains */ |
6747 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6748 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6749 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6750 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6751 | goto match2; |
6752 | } | |
6753 | /* no match - add a new doms_new */ | |
dce840a0 | 6754 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6755 | match2: |
6756 | ; | |
6757 | } | |
6758 | ||
6759 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6760 | if (doms_cur != &fallback_doms) |
6761 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6762 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6763 | doms_cur = doms_new; |
1d3504fc | 6764 | dattr_cur = dattr_new; |
029190c5 | 6765 | ndoms_cur = ndoms_new; |
7378547f MM |
6766 | |
6767 | register_sched_domain_sysctl(); | |
a1835615 | 6768 | |
712555ee | 6769 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6770 | } |
6771 | ||
d35be8ba SB |
6772 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6773 | ||
1da177e4 | 6774 | /* |
3a101d05 TH |
6775 | * Update cpusets according to cpu_active mask. If cpusets are |
6776 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6777 | * around partition_sched_domains(). | |
d35be8ba SB |
6778 | * |
6779 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6780 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6781 | */ |
0b2e918a TH |
6782 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6783 | void *hcpu) | |
e761b772 | 6784 | { |
d35be8ba SB |
6785 | switch (action) { |
6786 | case CPU_ONLINE_FROZEN: | |
6787 | case CPU_DOWN_FAILED_FROZEN: | |
6788 | ||
6789 | /* | |
6790 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6791 | * resume sequence. As long as this is not the last online | |
6792 | * operation in the resume sequence, just build a single sched | |
6793 | * domain, ignoring cpusets. | |
6794 | */ | |
6795 | num_cpus_frozen--; | |
6796 | if (likely(num_cpus_frozen)) { | |
6797 | partition_sched_domains(1, NULL, NULL); | |
6798 | break; | |
6799 | } | |
6800 | ||
6801 | /* | |
6802 | * This is the last CPU online operation. So fall through and | |
6803 | * restore the original sched domains by considering the | |
6804 | * cpuset configurations. | |
6805 | */ | |
6806 | ||
e761b772 | 6807 | case CPU_ONLINE: |
6ad4c188 | 6808 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6809 | cpuset_update_active_cpus(true); |
d35be8ba | 6810 | break; |
3a101d05 TH |
6811 | default: |
6812 | return NOTIFY_DONE; | |
6813 | } | |
d35be8ba | 6814 | return NOTIFY_OK; |
3a101d05 | 6815 | } |
e761b772 | 6816 | |
0b2e918a TH |
6817 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6818 | void *hcpu) | |
3a101d05 | 6819 | { |
d35be8ba | 6820 | switch (action) { |
3a101d05 | 6821 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6822 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6823 | break; |
6824 | case CPU_DOWN_PREPARE_FROZEN: | |
6825 | num_cpus_frozen++; | |
6826 | partition_sched_domains(1, NULL, NULL); | |
6827 | break; | |
e761b772 MK |
6828 | default: |
6829 | return NOTIFY_DONE; | |
6830 | } | |
d35be8ba | 6831 | return NOTIFY_OK; |
e761b772 | 6832 | } |
e761b772 | 6833 | |
1da177e4 LT |
6834 | void __init sched_init_smp(void) |
6835 | { | |
dcc30a35 RR |
6836 | cpumask_var_t non_isolated_cpus; |
6837 | ||
6838 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6839 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6840 | |
cb83b629 PZ |
6841 | sched_init_numa(); |
6842 | ||
95402b38 | 6843 | get_online_cpus(); |
712555ee | 6844 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6845 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6846 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6847 | if (cpumask_empty(non_isolated_cpus)) | |
6848 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6849 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6850 | put_online_cpus(); |
e761b772 | 6851 | |
301a5cba | 6852 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6853 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6854 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6855 | |
6856 | /* RT runtime code needs to handle some hotplug events */ | |
6857 | hotcpu_notifier(update_runtime, 0); | |
6858 | ||
b328ca18 | 6859 | init_hrtick(); |
5c1e1767 NP |
6860 | |
6861 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6862 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6863 | BUG(); |
19978ca6 | 6864 | sched_init_granularity(); |
dcc30a35 | 6865 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6866 | |
0e3900e6 | 6867 | init_sched_rt_class(); |
1da177e4 LT |
6868 | } |
6869 | #else | |
6870 | void __init sched_init_smp(void) | |
6871 | { | |
19978ca6 | 6872 | sched_init_granularity(); |
1da177e4 LT |
6873 | } |
6874 | #endif /* CONFIG_SMP */ | |
6875 | ||
cd1bb94b AB |
6876 | const_debug unsigned int sysctl_timer_migration = 1; |
6877 | ||
1da177e4 LT |
6878 | int in_sched_functions(unsigned long addr) |
6879 | { | |
1da177e4 LT |
6880 | return in_lock_functions(addr) || |
6881 | (addr >= (unsigned long)__sched_text_start | |
6882 | && addr < (unsigned long)__sched_text_end); | |
6883 | } | |
6884 | ||
029632fb | 6885 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6886 | /* |
6887 | * Default task group. | |
6888 | * Every task in system belongs to this group at bootup. | |
6889 | */ | |
029632fb | 6890 | struct task_group root_task_group; |
35cf4e50 | 6891 | LIST_HEAD(task_groups); |
052f1dc7 | 6892 | #endif |
6f505b16 | 6893 | |
029632fb | 6894 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6895 | |
1da177e4 LT |
6896 | void __init sched_init(void) |
6897 | { | |
dd41f596 | 6898 | int i, j; |
434d53b0 MT |
6899 | unsigned long alloc_size = 0, ptr; |
6900 | ||
6901 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6902 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6903 | #endif | |
6904 | #ifdef CONFIG_RT_GROUP_SCHED | |
6905 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6906 | #endif |
df7c8e84 | 6907 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6908 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6909 | #endif |
434d53b0 | 6910 | if (alloc_size) { |
36b7b6d4 | 6911 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6912 | |
6913 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6914 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6915 | ptr += nr_cpu_ids * sizeof(void **); |
6916 | ||
07e06b01 | 6917 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6918 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6919 | |
6d6bc0ad | 6920 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6921 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6922 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6923 | ptr += nr_cpu_ids * sizeof(void **); |
6924 | ||
07e06b01 | 6925 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6926 | ptr += nr_cpu_ids * sizeof(void **); |
6927 | ||
6d6bc0ad | 6928 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6929 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6930 | for_each_possible_cpu(i) { | |
6931 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6932 | ptr += cpumask_size(); | |
6933 | } | |
6934 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6935 | } |
dd41f596 | 6936 | |
57d885fe GH |
6937 | #ifdef CONFIG_SMP |
6938 | init_defrootdomain(); | |
6939 | #endif | |
6940 | ||
d0b27fa7 PZ |
6941 | init_rt_bandwidth(&def_rt_bandwidth, |
6942 | global_rt_period(), global_rt_runtime()); | |
6943 | ||
6944 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6945 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6946 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6947 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6948 | |
7c941438 | 6949 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6950 | list_add(&root_task_group.list, &task_groups); |
6951 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6952 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6953 | autogroup_init(&init_task); |
54c707e9 | 6954 | |
7c941438 | 6955 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6956 | |
54c707e9 GC |
6957 | #ifdef CONFIG_CGROUP_CPUACCT |
6958 | root_cpuacct.cpustat = &kernel_cpustat; | |
6959 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
6960 | /* Too early, not expected to fail */ | |
6961 | BUG_ON(!root_cpuacct.cpuusage); | |
6962 | #endif | |
0a945022 | 6963 | for_each_possible_cpu(i) { |
70b97a7f | 6964 | struct rq *rq; |
1da177e4 LT |
6965 | |
6966 | rq = cpu_rq(i); | |
05fa785c | 6967 | raw_spin_lock_init(&rq->lock); |
7897986b | 6968 | rq->nr_running = 0; |
dce48a84 TG |
6969 | rq->calc_load_active = 0; |
6970 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6971 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6972 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6973 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6974 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6975 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6976 | /* |
07e06b01 | 6977 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6978 | * |
6979 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6980 | * gets 100% of the cpu resources in the system. This overall | |
6981 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6982 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6983 | * based on each entity's (task or task-group's) weight |
6984 | * (se->load.weight). | |
6985 | * | |
07e06b01 | 6986 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6987 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6988 | * then A0's share of the cpu resource is: | |
6989 | * | |
0d905bca | 6990 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6991 | * |
07e06b01 YZ |
6992 | * We achieve this by letting root_task_group's tasks sit |
6993 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6994 | */ |
ab84d31e | 6995 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6996 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6997 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6998 | ||
6999 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7000 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7001 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7002 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7003 | #endif |
1da177e4 | 7004 | |
dd41f596 IM |
7005 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7006 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7007 | |
7008 | rq->last_load_update_tick = jiffies; | |
7009 | ||
1da177e4 | 7010 | #ifdef CONFIG_SMP |
41c7ce9a | 7011 | rq->sd = NULL; |
57d885fe | 7012 | rq->rd = NULL; |
1399fa78 | 7013 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 7014 | rq->post_schedule = 0; |
1da177e4 | 7015 | rq->active_balance = 0; |
dd41f596 | 7016 | rq->next_balance = jiffies; |
1da177e4 | 7017 | rq->push_cpu = 0; |
0a2966b4 | 7018 | rq->cpu = i; |
1f11eb6a | 7019 | rq->online = 0; |
eae0c9df MG |
7020 | rq->idle_stamp = 0; |
7021 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
7022 | |
7023 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7024 | ||
dc938520 | 7025 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7026 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7027 | rq->nohz_flags = 0; |
83cd4fe2 | 7028 | #endif |
1da177e4 | 7029 | #endif |
8f4d37ec | 7030 | init_rq_hrtick(rq); |
1da177e4 | 7031 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7032 | } |
7033 | ||
2dd73a4f | 7034 | set_load_weight(&init_task); |
b50f60ce | 7035 | |
e107be36 AK |
7036 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7037 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7038 | #endif | |
7039 | ||
b50f60ce | 7040 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 7041 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
7042 | #endif |
7043 | ||
1da177e4 LT |
7044 | /* |
7045 | * The boot idle thread does lazy MMU switching as well: | |
7046 | */ | |
7047 | atomic_inc(&init_mm.mm_count); | |
7048 | enter_lazy_tlb(&init_mm, current); | |
7049 | ||
7050 | /* | |
7051 | * Make us the idle thread. Technically, schedule() should not be | |
7052 | * called from this thread, however somewhere below it might be, | |
7053 | * but because we are the idle thread, we just pick up running again | |
7054 | * when this runqueue becomes "idle". | |
7055 | */ | |
7056 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7057 | |
7058 | calc_load_update = jiffies + LOAD_FREQ; | |
7059 | ||
dd41f596 IM |
7060 | /* |
7061 | * During early bootup we pretend to be a normal task: | |
7062 | */ | |
7063 | current->sched_class = &fair_sched_class; | |
6892b75e | 7064 | |
bf4d83f6 | 7065 | #ifdef CONFIG_SMP |
4cb98839 | 7066 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7067 | /* May be allocated at isolcpus cmdline parse time */ |
7068 | if (cpu_isolated_map == NULL) | |
7069 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7070 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7071 | #endif |
7072 | init_sched_fair_class(); | |
6a7b3dc3 | 7073 | |
6892b75e | 7074 | scheduler_running = 1; |
1da177e4 LT |
7075 | } |
7076 | ||
d902db1e | 7077 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7078 | static inline int preempt_count_equals(int preempt_offset) |
7079 | { | |
234da7bc | 7080 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7081 | |
4ba8216c | 7082 | return (nested == preempt_offset); |
e4aafea2 FW |
7083 | } |
7084 | ||
d894837f | 7085 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7086 | { |
1da177e4 LT |
7087 | static unsigned long prev_jiffy; /* ratelimiting */ |
7088 | ||
b3fbab05 | 7089 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7090 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7091 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7092 | return; |
7093 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7094 | return; | |
7095 | prev_jiffy = jiffies; | |
7096 | ||
3df0fc5b PZ |
7097 | printk(KERN_ERR |
7098 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7099 | file, line); | |
7100 | printk(KERN_ERR | |
7101 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7102 | in_atomic(), irqs_disabled(), | |
7103 | current->pid, current->comm); | |
aef745fc IM |
7104 | |
7105 | debug_show_held_locks(current); | |
7106 | if (irqs_disabled()) | |
7107 | print_irqtrace_events(current); | |
7108 | dump_stack(); | |
1da177e4 LT |
7109 | } |
7110 | EXPORT_SYMBOL(__might_sleep); | |
7111 | #endif | |
7112 | ||
7113 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7114 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7115 | { | |
da7a735e PZ |
7116 | const struct sched_class *prev_class = p->sched_class; |
7117 | int old_prio = p->prio; | |
3a5e4dc1 | 7118 | int on_rq; |
3e51f33f | 7119 | |
fd2f4419 | 7120 | on_rq = p->on_rq; |
3a5e4dc1 | 7121 | if (on_rq) |
4ca9b72b | 7122 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7123 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7124 | if (on_rq) { | |
4ca9b72b | 7125 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7126 | resched_task(rq->curr); |
7127 | } | |
da7a735e PZ |
7128 | |
7129 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7130 | } |
7131 | ||
1da177e4 LT |
7132 | void normalize_rt_tasks(void) |
7133 | { | |
a0f98a1c | 7134 | struct task_struct *g, *p; |
1da177e4 | 7135 | unsigned long flags; |
70b97a7f | 7136 | struct rq *rq; |
1da177e4 | 7137 | |
4cf5d77a | 7138 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7139 | do_each_thread(g, p) { |
178be793 IM |
7140 | /* |
7141 | * Only normalize user tasks: | |
7142 | */ | |
7143 | if (!p->mm) | |
7144 | continue; | |
7145 | ||
6cfb0d5d | 7146 | p->se.exec_start = 0; |
6cfb0d5d | 7147 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7148 | p->se.statistics.wait_start = 0; |
7149 | p->se.statistics.sleep_start = 0; | |
7150 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7151 | #endif |
dd41f596 IM |
7152 | |
7153 | if (!rt_task(p)) { | |
7154 | /* | |
7155 | * Renice negative nice level userspace | |
7156 | * tasks back to 0: | |
7157 | */ | |
7158 | if (TASK_NICE(p) < 0 && p->mm) | |
7159 | set_user_nice(p, 0); | |
1da177e4 | 7160 | continue; |
dd41f596 | 7161 | } |
1da177e4 | 7162 | |
1d615482 | 7163 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7164 | rq = __task_rq_lock(p); |
1da177e4 | 7165 | |
178be793 | 7166 | normalize_task(rq, p); |
3a5e4dc1 | 7167 | |
b29739f9 | 7168 | __task_rq_unlock(rq); |
1d615482 | 7169 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7170 | } while_each_thread(g, p); |
7171 | ||
4cf5d77a | 7172 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7173 | } |
7174 | ||
7175 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7176 | |
67fc4e0c | 7177 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7178 | /* |
67fc4e0c | 7179 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7180 | * |
7181 | * They can only be called when the whole system has been | |
7182 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7183 | * activity can take place. Using them for anything else would | |
7184 | * be a serious bug, and as a result, they aren't even visible | |
7185 | * under any other configuration. | |
7186 | */ | |
7187 | ||
7188 | /** | |
7189 | * curr_task - return the current task for a given cpu. | |
7190 | * @cpu: the processor in question. | |
7191 | * | |
7192 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7193 | */ | |
36c8b586 | 7194 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7195 | { |
7196 | return cpu_curr(cpu); | |
7197 | } | |
7198 | ||
67fc4e0c JW |
7199 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7200 | ||
7201 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7202 | /** |
7203 | * set_curr_task - set the current task for a given cpu. | |
7204 | * @cpu: the processor in question. | |
7205 | * @p: the task pointer to set. | |
7206 | * | |
7207 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7208 | * are serviced on a separate stack. It allows the architecture to switch the |
7209 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7210 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7211 | * and caller must save the original value of the current task (see | |
7212 | * curr_task() above) and restore that value before reenabling interrupts and | |
7213 | * re-starting the system. | |
7214 | * | |
7215 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7216 | */ | |
36c8b586 | 7217 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7218 | { |
7219 | cpu_curr(cpu) = p; | |
7220 | } | |
7221 | ||
7222 | #endif | |
29f59db3 | 7223 | |
7c941438 | 7224 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7225 | /* task_group_lock serializes the addition/removal of task groups */ |
7226 | static DEFINE_SPINLOCK(task_group_lock); | |
7227 | ||
bccbe08a PZ |
7228 | static void free_sched_group(struct task_group *tg) |
7229 | { | |
7230 | free_fair_sched_group(tg); | |
7231 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7232 | autogroup_free(tg); |
bccbe08a PZ |
7233 | kfree(tg); |
7234 | } | |
7235 | ||
7236 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7237 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7238 | { |
7239 | struct task_group *tg; | |
bccbe08a PZ |
7240 | |
7241 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7242 | if (!tg) | |
7243 | return ERR_PTR(-ENOMEM); | |
7244 | ||
ec7dc8ac | 7245 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7246 | goto err; |
7247 | ||
ec7dc8ac | 7248 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7249 | goto err; |
7250 | ||
ace783b9 LZ |
7251 | return tg; |
7252 | ||
7253 | err: | |
7254 | free_sched_group(tg); | |
7255 | return ERR_PTR(-ENOMEM); | |
7256 | } | |
7257 | ||
7258 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7259 | { | |
7260 | unsigned long flags; | |
7261 | ||
8ed36996 | 7262 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7263 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7264 | |
7265 | WARN_ON(!parent); /* root should already exist */ | |
7266 | ||
7267 | tg->parent = parent; | |
f473aa5e | 7268 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7269 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7270 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7271 | } |
7272 | ||
9b5b7751 | 7273 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7274 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7275 | { |
29f59db3 | 7276 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7277 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7278 | } |
7279 | ||
9b5b7751 | 7280 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7281 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7282 | { |
7283 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7284 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7285 | } | |
7286 | ||
7287 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7288 | { |
8ed36996 | 7289 | unsigned long flags; |
9b5b7751 | 7290 | int i; |
29f59db3 | 7291 | |
3d4b47b4 PZ |
7292 | /* end participation in shares distribution */ |
7293 | for_each_possible_cpu(i) | |
bccbe08a | 7294 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7295 | |
7296 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7297 | list_del_rcu(&tg->list); |
f473aa5e | 7298 | list_del_rcu(&tg->siblings); |
8ed36996 | 7299 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7300 | } |
7301 | ||
9b5b7751 | 7302 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7303 | * The caller of this function should have put the task in its new group |
7304 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7305 | * reflect its new group. | |
9b5b7751 SV |
7306 | */ |
7307 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7308 | { |
8323f26c | 7309 | struct task_group *tg; |
29f59db3 SV |
7310 | int on_rq, running; |
7311 | unsigned long flags; | |
7312 | struct rq *rq; | |
7313 | ||
7314 | rq = task_rq_lock(tsk, &flags); | |
7315 | ||
051a1d1a | 7316 | running = task_current(rq, tsk); |
fd2f4419 | 7317 | on_rq = tsk->on_rq; |
29f59db3 | 7318 | |
0e1f3483 | 7319 | if (on_rq) |
29f59db3 | 7320 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7321 | if (unlikely(running)) |
7322 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7323 | |
8323f26c PZ |
7324 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
7325 | lockdep_is_held(&tsk->sighand->siglock)), | |
7326 | struct task_group, css); | |
7327 | tg = autogroup_task_group(tsk, tg); | |
7328 | tsk->sched_task_group = tg; | |
7329 | ||
810b3817 | 7330 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7331 | if (tsk->sched_class->task_move_group) |
7332 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7333 | else | |
810b3817 | 7334 | #endif |
b2b5ce02 | 7335 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7336 | |
0e1f3483 HS |
7337 | if (unlikely(running)) |
7338 | tsk->sched_class->set_curr_task(rq); | |
7339 | if (on_rq) | |
371fd7e7 | 7340 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7341 | |
0122ec5b | 7342 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7343 | } |
7c941438 | 7344 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7345 | |
a790de99 | 7346 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7347 | static unsigned long to_ratio(u64 period, u64 runtime) |
7348 | { | |
7349 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7350 | return 1ULL << 20; |
9f0c1e56 | 7351 | |
9a7e0b18 | 7352 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7353 | } |
a790de99 PT |
7354 | #endif |
7355 | ||
7356 | #ifdef CONFIG_RT_GROUP_SCHED | |
7357 | /* | |
7358 | * Ensure that the real time constraints are schedulable. | |
7359 | */ | |
7360 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7361 | |
9a7e0b18 PZ |
7362 | /* Must be called with tasklist_lock held */ |
7363 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7364 | { |
9a7e0b18 | 7365 | struct task_struct *g, *p; |
b40b2e8e | 7366 | |
9a7e0b18 | 7367 | do_each_thread(g, p) { |
029632fb | 7368 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7369 | return 1; |
7370 | } while_each_thread(g, p); | |
b40b2e8e | 7371 | |
9a7e0b18 PZ |
7372 | return 0; |
7373 | } | |
b40b2e8e | 7374 | |
9a7e0b18 PZ |
7375 | struct rt_schedulable_data { |
7376 | struct task_group *tg; | |
7377 | u64 rt_period; | |
7378 | u64 rt_runtime; | |
7379 | }; | |
b40b2e8e | 7380 | |
a790de99 | 7381 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7382 | { |
7383 | struct rt_schedulable_data *d = data; | |
7384 | struct task_group *child; | |
7385 | unsigned long total, sum = 0; | |
7386 | u64 period, runtime; | |
b40b2e8e | 7387 | |
9a7e0b18 PZ |
7388 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7389 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7390 | |
9a7e0b18 PZ |
7391 | if (tg == d->tg) { |
7392 | period = d->rt_period; | |
7393 | runtime = d->rt_runtime; | |
b40b2e8e | 7394 | } |
b40b2e8e | 7395 | |
4653f803 PZ |
7396 | /* |
7397 | * Cannot have more runtime than the period. | |
7398 | */ | |
7399 | if (runtime > period && runtime != RUNTIME_INF) | |
7400 | return -EINVAL; | |
6f505b16 | 7401 | |
4653f803 PZ |
7402 | /* |
7403 | * Ensure we don't starve existing RT tasks. | |
7404 | */ | |
9a7e0b18 PZ |
7405 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7406 | return -EBUSY; | |
6f505b16 | 7407 | |
9a7e0b18 | 7408 | total = to_ratio(period, runtime); |
6f505b16 | 7409 | |
4653f803 PZ |
7410 | /* |
7411 | * Nobody can have more than the global setting allows. | |
7412 | */ | |
7413 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7414 | return -EINVAL; | |
6f505b16 | 7415 | |
4653f803 PZ |
7416 | /* |
7417 | * The sum of our children's runtime should not exceed our own. | |
7418 | */ | |
9a7e0b18 PZ |
7419 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7420 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7421 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7422 | |
9a7e0b18 PZ |
7423 | if (child == d->tg) { |
7424 | period = d->rt_period; | |
7425 | runtime = d->rt_runtime; | |
7426 | } | |
6f505b16 | 7427 | |
9a7e0b18 | 7428 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7429 | } |
6f505b16 | 7430 | |
9a7e0b18 PZ |
7431 | if (sum > total) |
7432 | return -EINVAL; | |
7433 | ||
7434 | return 0; | |
6f505b16 PZ |
7435 | } |
7436 | ||
9a7e0b18 | 7437 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7438 | { |
8277434e PT |
7439 | int ret; |
7440 | ||
9a7e0b18 PZ |
7441 | struct rt_schedulable_data data = { |
7442 | .tg = tg, | |
7443 | .rt_period = period, | |
7444 | .rt_runtime = runtime, | |
7445 | }; | |
7446 | ||
8277434e PT |
7447 | rcu_read_lock(); |
7448 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7449 | rcu_read_unlock(); | |
7450 | ||
7451 | return ret; | |
521f1a24 DG |
7452 | } |
7453 | ||
ab84d31e | 7454 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7455 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7456 | { |
ac086bc2 | 7457 | int i, err = 0; |
9f0c1e56 | 7458 | |
9f0c1e56 | 7459 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7460 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7461 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7462 | if (err) | |
9f0c1e56 | 7463 | goto unlock; |
ac086bc2 | 7464 | |
0986b11b | 7465 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7466 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7467 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7468 | |
7469 | for_each_possible_cpu(i) { | |
7470 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7471 | ||
0986b11b | 7472 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7473 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7474 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7475 | } |
0986b11b | 7476 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7477 | unlock: |
521f1a24 | 7478 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7479 | mutex_unlock(&rt_constraints_mutex); |
7480 | ||
7481 | return err; | |
6f505b16 PZ |
7482 | } |
7483 | ||
25cc7da7 | 7484 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7485 | { |
7486 | u64 rt_runtime, rt_period; | |
7487 | ||
7488 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7489 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7490 | if (rt_runtime_us < 0) | |
7491 | rt_runtime = RUNTIME_INF; | |
7492 | ||
ab84d31e | 7493 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7494 | } |
7495 | ||
25cc7da7 | 7496 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7497 | { |
7498 | u64 rt_runtime_us; | |
7499 | ||
d0b27fa7 | 7500 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7501 | return -1; |
7502 | ||
d0b27fa7 | 7503 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7504 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7505 | return rt_runtime_us; | |
7506 | } | |
d0b27fa7 | 7507 | |
25cc7da7 | 7508 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7509 | { |
7510 | u64 rt_runtime, rt_period; | |
7511 | ||
7512 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7513 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7514 | ||
619b0488 R |
7515 | if (rt_period == 0) |
7516 | return -EINVAL; | |
7517 | ||
ab84d31e | 7518 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7519 | } |
7520 | ||
25cc7da7 | 7521 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7522 | { |
7523 | u64 rt_period_us; | |
7524 | ||
7525 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7526 | do_div(rt_period_us, NSEC_PER_USEC); | |
7527 | return rt_period_us; | |
7528 | } | |
7529 | ||
7530 | static int sched_rt_global_constraints(void) | |
7531 | { | |
4653f803 | 7532 | u64 runtime, period; |
d0b27fa7 PZ |
7533 | int ret = 0; |
7534 | ||
ec5d4989 HS |
7535 | if (sysctl_sched_rt_period <= 0) |
7536 | return -EINVAL; | |
7537 | ||
4653f803 PZ |
7538 | runtime = global_rt_runtime(); |
7539 | period = global_rt_period(); | |
7540 | ||
7541 | /* | |
7542 | * Sanity check on the sysctl variables. | |
7543 | */ | |
7544 | if (runtime > period && runtime != RUNTIME_INF) | |
7545 | return -EINVAL; | |
10b612f4 | 7546 | |
d0b27fa7 | 7547 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7548 | read_lock(&tasklist_lock); |
4653f803 | 7549 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7550 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7551 | mutex_unlock(&rt_constraints_mutex); |
7552 | ||
7553 | return ret; | |
7554 | } | |
54e99124 | 7555 | |
25cc7da7 | 7556 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7557 | { |
7558 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7559 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7560 | return 0; | |
7561 | ||
7562 | return 1; | |
7563 | } | |
7564 | ||
6d6bc0ad | 7565 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7566 | static int sched_rt_global_constraints(void) |
7567 | { | |
ac086bc2 PZ |
7568 | unsigned long flags; |
7569 | int i; | |
7570 | ||
ec5d4989 HS |
7571 | if (sysctl_sched_rt_period <= 0) |
7572 | return -EINVAL; | |
7573 | ||
60aa605d PZ |
7574 | /* |
7575 | * There's always some RT tasks in the root group | |
7576 | * -- migration, kstopmachine etc.. | |
7577 | */ | |
7578 | if (sysctl_sched_rt_runtime == 0) | |
7579 | return -EBUSY; | |
7580 | ||
0986b11b | 7581 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7582 | for_each_possible_cpu(i) { |
7583 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7584 | ||
0986b11b | 7585 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7586 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7587 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7588 | } |
0986b11b | 7589 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7590 | |
d0b27fa7 PZ |
7591 | return 0; |
7592 | } | |
6d6bc0ad | 7593 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7594 | |
ce0dbbbb CW |
7595 | int sched_rr_handler(struct ctl_table *table, int write, |
7596 | void __user *buffer, size_t *lenp, | |
7597 | loff_t *ppos) | |
7598 | { | |
7599 | int ret; | |
7600 | static DEFINE_MUTEX(mutex); | |
7601 | ||
7602 | mutex_lock(&mutex); | |
7603 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7604 | /* make sure that internally we keep jiffies */ | |
7605 | /* also, writing zero resets timeslice to default */ | |
7606 | if (!ret && write) { | |
7607 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7608 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7609 | } | |
7610 | mutex_unlock(&mutex); | |
7611 | return ret; | |
7612 | } | |
7613 | ||
d0b27fa7 | 7614 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7615 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7616 | loff_t *ppos) |
7617 | { | |
7618 | int ret; | |
7619 | int old_period, old_runtime; | |
7620 | static DEFINE_MUTEX(mutex); | |
7621 | ||
7622 | mutex_lock(&mutex); | |
7623 | old_period = sysctl_sched_rt_period; | |
7624 | old_runtime = sysctl_sched_rt_runtime; | |
7625 | ||
8d65af78 | 7626 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7627 | |
7628 | if (!ret && write) { | |
7629 | ret = sched_rt_global_constraints(); | |
7630 | if (ret) { | |
7631 | sysctl_sched_rt_period = old_period; | |
7632 | sysctl_sched_rt_runtime = old_runtime; | |
7633 | } else { | |
7634 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7635 | def_rt_bandwidth.rt_period = | |
7636 | ns_to_ktime(global_rt_period()); | |
7637 | } | |
7638 | } | |
7639 | mutex_unlock(&mutex); | |
7640 | ||
7641 | return ret; | |
7642 | } | |
68318b8e | 7643 | |
052f1dc7 | 7644 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7645 | |
7646 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7647 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7648 | { |
2b01dfe3 PM |
7649 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7650 | struct task_group, css); | |
68318b8e SV |
7651 | } |
7652 | ||
92fb9748 | 7653 | static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp) |
68318b8e | 7654 | { |
ec7dc8ac | 7655 | struct task_group *tg, *parent; |
68318b8e | 7656 | |
2b01dfe3 | 7657 | if (!cgrp->parent) { |
68318b8e | 7658 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7659 | return &root_task_group.css; |
68318b8e SV |
7660 | } |
7661 | ||
ec7dc8ac DG |
7662 | parent = cgroup_tg(cgrp->parent); |
7663 | tg = sched_create_group(parent); | |
68318b8e SV |
7664 | if (IS_ERR(tg)) |
7665 | return ERR_PTR(-ENOMEM); | |
7666 | ||
68318b8e SV |
7667 | return &tg->css; |
7668 | } | |
7669 | ||
ace783b9 LZ |
7670 | static int cpu_cgroup_css_online(struct cgroup *cgrp) |
7671 | { | |
7672 | struct task_group *tg = cgroup_tg(cgrp); | |
7673 | struct task_group *parent; | |
7674 | ||
7675 | if (!cgrp->parent) | |
7676 | return 0; | |
7677 | ||
7678 | parent = cgroup_tg(cgrp->parent); | |
7679 | sched_online_group(tg, parent); | |
7680 | return 0; | |
7681 | } | |
7682 | ||
92fb9748 | 7683 | static void cpu_cgroup_css_free(struct cgroup *cgrp) |
68318b8e | 7684 | { |
2b01dfe3 | 7685 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7686 | |
7687 | sched_destroy_group(tg); | |
7688 | } | |
7689 | ||
ace783b9 LZ |
7690 | static void cpu_cgroup_css_offline(struct cgroup *cgrp) |
7691 | { | |
7692 | struct task_group *tg = cgroup_tg(cgrp); | |
7693 | ||
7694 | sched_offline_group(tg); | |
7695 | } | |
7696 | ||
761b3ef5 | 7697 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 7698 | struct cgroup_taskset *tset) |
68318b8e | 7699 | { |
bb9d97b6 TH |
7700 | struct task_struct *task; |
7701 | ||
7702 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7703 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7704 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7705 | return -EINVAL; | |
b68aa230 | 7706 | #else |
bb9d97b6 TH |
7707 | /* We don't support RT-tasks being in separate groups */ |
7708 | if (task->sched_class != &fair_sched_class) | |
7709 | return -EINVAL; | |
b68aa230 | 7710 | #endif |
bb9d97b6 | 7711 | } |
be367d09 BB |
7712 | return 0; |
7713 | } | |
68318b8e | 7714 | |
761b3ef5 | 7715 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 7716 | struct cgroup_taskset *tset) |
68318b8e | 7717 | { |
bb9d97b6 TH |
7718 | struct task_struct *task; |
7719 | ||
7720 | cgroup_taskset_for_each(task, cgrp, tset) | |
7721 | sched_move_task(task); | |
68318b8e SV |
7722 | } |
7723 | ||
068c5cc5 | 7724 | static void |
761b3ef5 LZ |
7725 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
7726 | struct task_struct *task) | |
068c5cc5 PZ |
7727 | { |
7728 | /* | |
7729 | * cgroup_exit() is called in the copy_process() failure path. | |
7730 | * Ignore this case since the task hasn't ran yet, this avoids | |
7731 | * trying to poke a half freed task state from generic code. | |
7732 | */ | |
7733 | if (!(task->flags & PF_EXITING)) | |
7734 | return; | |
7735 | ||
7736 | sched_move_task(task); | |
7737 | } | |
7738 | ||
052f1dc7 | 7739 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7740 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7741 | u64 shareval) |
68318b8e | 7742 | { |
c8b28116 | 7743 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7744 | } |
7745 | ||
f4c753b7 | 7746 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7747 | { |
2b01dfe3 | 7748 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7749 | |
c8b28116 | 7750 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7751 | } |
ab84d31e PT |
7752 | |
7753 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7754 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7755 | ||
ab84d31e PT |
7756 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7757 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7758 | ||
a790de99 PT |
7759 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7760 | ||
ab84d31e PT |
7761 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7762 | { | |
56f570e5 | 7763 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7764 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7765 | |
7766 | if (tg == &root_task_group) | |
7767 | return -EINVAL; | |
7768 | ||
7769 | /* | |
7770 | * Ensure we have at some amount of bandwidth every period. This is | |
7771 | * to prevent reaching a state of large arrears when throttled via | |
7772 | * entity_tick() resulting in prolonged exit starvation. | |
7773 | */ | |
7774 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7775 | return -EINVAL; | |
7776 | ||
7777 | /* | |
7778 | * Likewise, bound things on the otherside by preventing insane quota | |
7779 | * periods. This also allows us to normalize in computing quota | |
7780 | * feasibility. | |
7781 | */ | |
7782 | if (period > max_cfs_quota_period) | |
7783 | return -EINVAL; | |
7784 | ||
a790de99 PT |
7785 | mutex_lock(&cfs_constraints_mutex); |
7786 | ret = __cfs_schedulable(tg, period, quota); | |
7787 | if (ret) | |
7788 | goto out_unlock; | |
7789 | ||
58088ad0 | 7790 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7791 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7792 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7793 | raw_spin_lock_irq(&cfs_b->lock); |
7794 | cfs_b->period = ns_to_ktime(period); | |
7795 | cfs_b->quota = quota; | |
58088ad0 | 7796 | |
a9cf55b2 | 7797 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7798 | /* restart the period timer (if active) to handle new period expiry */ |
7799 | if (runtime_enabled && cfs_b->timer_active) { | |
7800 | /* force a reprogram */ | |
7801 | cfs_b->timer_active = 0; | |
7802 | __start_cfs_bandwidth(cfs_b); | |
7803 | } | |
ab84d31e PT |
7804 | raw_spin_unlock_irq(&cfs_b->lock); |
7805 | ||
7806 | for_each_possible_cpu(i) { | |
7807 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7808 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7809 | |
7810 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7811 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7812 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7813 | |
029632fb | 7814 | if (cfs_rq->throttled) |
671fd9da | 7815 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7816 | raw_spin_unlock_irq(&rq->lock); |
7817 | } | |
a790de99 PT |
7818 | out_unlock: |
7819 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7820 | |
a790de99 | 7821 | return ret; |
ab84d31e PT |
7822 | } |
7823 | ||
7824 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7825 | { | |
7826 | u64 quota, period; | |
7827 | ||
029632fb | 7828 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7829 | if (cfs_quota_us < 0) |
7830 | quota = RUNTIME_INF; | |
7831 | else | |
7832 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7833 | ||
7834 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7835 | } | |
7836 | ||
7837 | long tg_get_cfs_quota(struct task_group *tg) | |
7838 | { | |
7839 | u64 quota_us; | |
7840 | ||
029632fb | 7841 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7842 | return -1; |
7843 | ||
029632fb | 7844 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7845 | do_div(quota_us, NSEC_PER_USEC); |
7846 | ||
7847 | return quota_us; | |
7848 | } | |
7849 | ||
7850 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7851 | { | |
7852 | u64 quota, period; | |
7853 | ||
7854 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7855 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7856 | |
ab84d31e PT |
7857 | return tg_set_cfs_bandwidth(tg, period, quota); |
7858 | } | |
7859 | ||
7860 | long tg_get_cfs_period(struct task_group *tg) | |
7861 | { | |
7862 | u64 cfs_period_us; | |
7863 | ||
029632fb | 7864 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7865 | do_div(cfs_period_us, NSEC_PER_USEC); |
7866 | ||
7867 | return cfs_period_us; | |
7868 | } | |
7869 | ||
7870 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7871 | { | |
7872 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7873 | } | |
7874 | ||
7875 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7876 | s64 cfs_quota_us) | |
7877 | { | |
7878 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7879 | } | |
7880 | ||
7881 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7882 | { | |
7883 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7884 | } | |
7885 | ||
7886 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7887 | u64 cfs_period_us) | |
7888 | { | |
7889 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7890 | } | |
7891 | ||
a790de99 PT |
7892 | struct cfs_schedulable_data { |
7893 | struct task_group *tg; | |
7894 | u64 period, quota; | |
7895 | }; | |
7896 | ||
7897 | /* | |
7898 | * normalize group quota/period to be quota/max_period | |
7899 | * note: units are usecs | |
7900 | */ | |
7901 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7902 | struct cfs_schedulable_data *d) | |
7903 | { | |
7904 | u64 quota, period; | |
7905 | ||
7906 | if (tg == d->tg) { | |
7907 | period = d->period; | |
7908 | quota = d->quota; | |
7909 | } else { | |
7910 | period = tg_get_cfs_period(tg); | |
7911 | quota = tg_get_cfs_quota(tg); | |
7912 | } | |
7913 | ||
7914 | /* note: these should typically be equivalent */ | |
7915 | if (quota == RUNTIME_INF || quota == -1) | |
7916 | return RUNTIME_INF; | |
7917 | ||
7918 | return to_ratio(period, quota); | |
7919 | } | |
7920 | ||
7921 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7922 | { | |
7923 | struct cfs_schedulable_data *d = data; | |
029632fb | 7924 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7925 | s64 quota = 0, parent_quota = -1; |
7926 | ||
7927 | if (!tg->parent) { | |
7928 | quota = RUNTIME_INF; | |
7929 | } else { | |
029632fb | 7930 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7931 | |
7932 | quota = normalize_cfs_quota(tg, d); | |
7933 | parent_quota = parent_b->hierarchal_quota; | |
7934 | ||
7935 | /* | |
7936 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7937 | * limit is set | |
7938 | */ | |
7939 | if (quota == RUNTIME_INF) | |
7940 | quota = parent_quota; | |
7941 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7942 | return -EINVAL; | |
7943 | } | |
7944 | cfs_b->hierarchal_quota = quota; | |
7945 | ||
7946 | return 0; | |
7947 | } | |
7948 | ||
7949 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7950 | { | |
8277434e | 7951 | int ret; |
a790de99 PT |
7952 | struct cfs_schedulable_data data = { |
7953 | .tg = tg, | |
7954 | .period = period, | |
7955 | .quota = quota, | |
7956 | }; | |
7957 | ||
7958 | if (quota != RUNTIME_INF) { | |
7959 | do_div(data.period, NSEC_PER_USEC); | |
7960 | do_div(data.quota, NSEC_PER_USEC); | |
7961 | } | |
7962 | ||
8277434e PT |
7963 | rcu_read_lock(); |
7964 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7965 | rcu_read_unlock(); | |
7966 | ||
7967 | return ret; | |
a790de99 | 7968 | } |
e8da1b18 NR |
7969 | |
7970 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7971 | struct cgroup_map_cb *cb) | |
7972 | { | |
7973 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7974 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7975 | |
7976 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7977 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7978 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7979 | ||
7980 | return 0; | |
7981 | } | |
ab84d31e | 7982 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7983 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7984 | |
052f1dc7 | 7985 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7986 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7987 | s64 val) |
6f505b16 | 7988 | { |
06ecb27c | 7989 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7990 | } |
7991 | ||
06ecb27c | 7992 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7993 | { |
06ecb27c | 7994 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7995 | } |
d0b27fa7 PZ |
7996 | |
7997 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7998 | u64 rt_period_us) | |
7999 | { | |
8000 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8001 | } | |
8002 | ||
8003 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8004 | { | |
8005 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8006 | } | |
6d6bc0ad | 8007 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8008 | |
fe5c7cc2 | 8009 | static struct cftype cpu_files[] = { |
052f1dc7 | 8010 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8011 | { |
8012 | .name = "shares", | |
f4c753b7 PM |
8013 | .read_u64 = cpu_shares_read_u64, |
8014 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8015 | }, |
052f1dc7 | 8016 | #endif |
ab84d31e PT |
8017 | #ifdef CONFIG_CFS_BANDWIDTH |
8018 | { | |
8019 | .name = "cfs_quota_us", | |
8020 | .read_s64 = cpu_cfs_quota_read_s64, | |
8021 | .write_s64 = cpu_cfs_quota_write_s64, | |
8022 | }, | |
8023 | { | |
8024 | .name = "cfs_period_us", | |
8025 | .read_u64 = cpu_cfs_period_read_u64, | |
8026 | .write_u64 = cpu_cfs_period_write_u64, | |
8027 | }, | |
e8da1b18 NR |
8028 | { |
8029 | .name = "stat", | |
8030 | .read_map = cpu_stats_show, | |
8031 | }, | |
ab84d31e | 8032 | #endif |
052f1dc7 | 8033 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8034 | { |
9f0c1e56 | 8035 | .name = "rt_runtime_us", |
06ecb27c PM |
8036 | .read_s64 = cpu_rt_runtime_read, |
8037 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8038 | }, |
d0b27fa7 PZ |
8039 | { |
8040 | .name = "rt_period_us", | |
f4c753b7 PM |
8041 | .read_u64 = cpu_rt_period_read_uint, |
8042 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8043 | }, |
052f1dc7 | 8044 | #endif |
4baf6e33 | 8045 | { } /* terminate */ |
68318b8e SV |
8046 | }; |
8047 | ||
68318b8e | 8048 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 8049 | .name = "cpu", |
92fb9748 TH |
8050 | .css_alloc = cpu_cgroup_css_alloc, |
8051 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8052 | .css_online = cpu_cgroup_css_online, |
8053 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8054 | .can_attach = cpu_cgroup_can_attach, |
8055 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8056 | .exit = cpu_cgroup_exit, |
38605cae | 8057 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 8058 | .base_cftypes = cpu_files, |
68318b8e SV |
8059 | .early_init = 1, |
8060 | }; | |
8061 | ||
052f1dc7 | 8062 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8063 | |
8064 | #ifdef CONFIG_CGROUP_CPUACCT | |
8065 | ||
8066 | /* | |
8067 | * CPU accounting code for task groups. | |
8068 | * | |
8069 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8070 | * (balbir@in.ibm.com). | |
8071 | */ | |
8072 | ||
73fbec60 FW |
8073 | struct cpuacct root_cpuacct; |
8074 | ||
d842de87 | 8075 | /* create a new cpu accounting group */ |
92fb9748 | 8076 | static struct cgroup_subsys_state *cpuacct_css_alloc(struct cgroup *cgrp) |
d842de87 | 8077 | { |
54c707e9 | 8078 | struct cpuacct *ca; |
d842de87 | 8079 | |
54c707e9 GC |
8080 | if (!cgrp->parent) |
8081 | return &root_cpuacct.css; | |
8082 | ||
8083 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 8084 | if (!ca) |
ef12fefa | 8085 | goto out; |
d842de87 SV |
8086 | |
8087 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8088 | if (!ca->cpuusage) |
8089 | goto out_free_ca; | |
8090 | ||
54c707e9 GC |
8091 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
8092 | if (!ca->cpustat) | |
8093 | goto out_free_cpuusage; | |
934352f2 | 8094 | |
d842de87 | 8095 | return &ca->css; |
ef12fefa | 8096 | |
54c707e9 | 8097 | out_free_cpuusage: |
ef12fefa BR |
8098 | free_percpu(ca->cpuusage); |
8099 | out_free_ca: | |
8100 | kfree(ca); | |
8101 | out: | |
8102 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8103 | } |
8104 | ||
8105 | /* destroy an existing cpu accounting group */ | |
92fb9748 | 8106 | static void cpuacct_css_free(struct cgroup *cgrp) |
d842de87 | 8107 | { |
32cd756a | 8108 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 8109 | |
54c707e9 | 8110 | free_percpu(ca->cpustat); |
d842de87 SV |
8111 | free_percpu(ca->cpuusage); |
8112 | kfree(ca); | |
8113 | } | |
8114 | ||
720f5498 KC |
8115 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8116 | { | |
b36128c8 | 8117 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8118 | u64 data; |
8119 | ||
8120 | #ifndef CONFIG_64BIT | |
8121 | /* | |
8122 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8123 | */ | |
05fa785c | 8124 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8125 | data = *cpuusage; |
05fa785c | 8126 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8127 | #else |
8128 | data = *cpuusage; | |
8129 | #endif | |
8130 | ||
8131 | return data; | |
8132 | } | |
8133 | ||
8134 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8135 | { | |
b36128c8 | 8136 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8137 | |
8138 | #ifndef CONFIG_64BIT | |
8139 | /* | |
8140 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8141 | */ | |
05fa785c | 8142 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8143 | *cpuusage = val; |
05fa785c | 8144 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8145 | #else |
8146 | *cpuusage = val; | |
8147 | #endif | |
8148 | } | |
8149 | ||
d842de87 | 8150 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8151 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8152 | { |
32cd756a | 8153 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8154 | u64 totalcpuusage = 0; |
8155 | int i; | |
8156 | ||
720f5498 KC |
8157 | for_each_present_cpu(i) |
8158 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8159 | |
8160 | return totalcpuusage; | |
8161 | } | |
8162 | ||
0297b803 DG |
8163 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8164 | u64 reset) | |
8165 | { | |
8166 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8167 | int err = 0; | |
8168 | int i; | |
8169 | ||
8170 | if (reset) { | |
8171 | err = -EINVAL; | |
8172 | goto out; | |
8173 | } | |
8174 | ||
720f5498 KC |
8175 | for_each_present_cpu(i) |
8176 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8177 | |
0297b803 DG |
8178 | out: |
8179 | return err; | |
8180 | } | |
8181 | ||
e9515c3c KC |
8182 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8183 | struct seq_file *m) | |
8184 | { | |
8185 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8186 | u64 percpu; | |
8187 | int i; | |
8188 | ||
8189 | for_each_present_cpu(i) { | |
8190 | percpu = cpuacct_cpuusage_read(ca, i); | |
8191 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8192 | } | |
8193 | seq_printf(m, "\n"); | |
8194 | return 0; | |
8195 | } | |
8196 | ||
ef12fefa BR |
8197 | static const char *cpuacct_stat_desc[] = { |
8198 | [CPUACCT_STAT_USER] = "user", | |
8199 | [CPUACCT_STAT_SYSTEM] = "system", | |
8200 | }; | |
8201 | ||
8202 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8203 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8204 | { |
8205 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8206 | int cpu; |
8207 | s64 val = 0; | |
ef12fefa | 8208 | |
54c707e9 GC |
8209 | for_each_online_cpu(cpu) { |
8210 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8211 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8212 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8213 | } |
54c707e9 GC |
8214 | val = cputime64_to_clock_t(val); |
8215 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8216 | |
54c707e9 GC |
8217 | val = 0; |
8218 | for_each_online_cpu(cpu) { | |
8219 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8220 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8221 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8222 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8223 | } |
54c707e9 GC |
8224 | |
8225 | val = cputime64_to_clock_t(val); | |
8226 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8227 | ||
ef12fefa BR |
8228 | return 0; |
8229 | } | |
8230 | ||
d842de87 SV |
8231 | static struct cftype files[] = { |
8232 | { | |
8233 | .name = "usage", | |
f4c753b7 PM |
8234 | .read_u64 = cpuusage_read, |
8235 | .write_u64 = cpuusage_write, | |
d842de87 | 8236 | }, |
e9515c3c KC |
8237 | { |
8238 | .name = "usage_percpu", | |
8239 | .read_seq_string = cpuacct_percpu_seq_read, | |
8240 | }, | |
ef12fefa BR |
8241 | { |
8242 | .name = "stat", | |
8243 | .read_map = cpuacct_stats_show, | |
8244 | }, | |
4baf6e33 | 8245 | { } /* terminate */ |
d842de87 SV |
8246 | }; |
8247 | ||
d842de87 SV |
8248 | /* |
8249 | * charge this task's execution time to its accounting group. | |
8250 | * | |
8251 | * called with rq->lock held. | |
8252 | */ | |
029632fb | 8253 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8254 | { |
8255 | struct cpuacct *ca; | |
934352f2 | 8256 | int cpu; |
d842de87 | 8257 | |
c40c6f85 | 8258 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8259 | return; |
8260 | ||
934352f2 | 8261 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8262 | |
8263 | rcu_read_lock(); | |
8264 | ||
d842de87 | 8265 | ca = task_ca(tsk); |
d842de87 | 8266 | |
44252e42 | 8267 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8268 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8269 | *cpuusage += cputime; |
8270 | } | |
a18b83b7 BR |
8271 | |
8272 | rcu_read_unlock(); | |
d842de87 SV |
8273 | } |
8274 | ||
8275 | struct cgroup_subsys cpuacct_subsys = { | |
8276 | .name = "cpuacct", | |
92fb9748 TH |
8277 | .css_alloc = cpuacct_css_alloc, |
8278 | .css_free = cpuacct_css_free, | |
d842de87 | 8279 | .subsys_id = cpuacct_subsys_id, |
4baf6e33 | 8280 | .base_cftypes = files, |
d842de87 SV |
8281 | }; |
8282 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
b637a328 PM |
8283 | |
8284 | void dump_cpu_task(int cpu) | |
8285 | { | |
8286 | pr_info("Task dump for CPU %d:\n", cpu); | |
8287 | sched_show_task(cpu_curr(cpu)); | |
8288 | } |