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 | |
029632fb | 515 | void resched_task(struct task_struct *p) |
c24d20db IM |
516 | { |
517 | int cpu; | |
518 | ||
05fa785c | 519 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 520 | |
5ed0cec0 | 521 | if (test_tsk_need_resched(p)) |
c24d20db IM |
522 | return; |
523 | ||
5ed0cec0 | 524 | set_tsk_need_resched(p); |
c24d20db IM |
525 | |
526 | cpu = task_cpu(p); | |
527 | if (cpu == smp_processor_id()) | |
528 | return; | |
529 | ||
530 | /* NEED_RESCHED must be visible before we test polling */ | |
531 | smp_mb(); | |
532 | if (!tsk_is_polling(p)) | |
533 | smp_send_reschedule(cpu); | |
534 | } | |
535 | ||
029632fb | 536 | void resched_cpu(int cpu) |
c24d20db IM |
537 | { |
538 | struct rq *rq = cpu_rq(cpu); | |
539 | unsigned long flags; | |
540 | ||
05fa785c | 541 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
542 | return; |
543 | resched_task(cpu_curr(cpu)); | |
05fa785c | 544 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 545 | } |
06d8308c | 546 | |
3451d024 | 547 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
548 | /* |
549 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
550 | * from an idle cpu. This is good for power-savings. | |
551 | * | |
552 | * We don't do similar optimization for completely idle system, as | |
553 | * selecting an idle cpu will add more delays to the timers than intended | |
554 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
555 | */ | |
556 | int get_nohz_timer_target(void) | |
557 | { | |
558 | int cpu = smp_processor_id(); | |
559 | int i; | |
560 | struct sched_domain *sd; | |
561 | ||
057f3fad | 562 | rcu_read_lock(); |
83cd4fe2 | 563 | for_each_domain(cpu, sd) { |
057f3fad PZ |
564 | for_each_cpu(i, sched_domain_span(sd)) { |
565 | if (!idle_cpu(i)) { | |
566 | cpu = i; | |
567 | goto unlock; | |
568 | } | |
569 | } | |
83cd4fe2 | 570 | } |
057f3fad PZ |
571 | unlock: |
572 | rcu_read_unlock(); | |
83cd4fe2 VP |
573 | return cpu; |
574 | } | |
06d8308c TG |
575 | /* |
576 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
577 | * idle CPU then this timer might expire before the next timer event | |
578 | * which is scheduled to wake up that CPU. In case of a completely | |
579 | * idle system the next event might even be infinite time into the | |
580 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
581 | * leaves the inner idle loop so the newly added timer is taken into | |
582 | * account when the CPU goes back to idle and evaluates the timer | |
583 | * wheel for the next timer event. | |
584 | */ | |
1c20091e | 585 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
586 | { |
587 | struct rq *rq = cpu_rq(cpu); | |
588 | ||
589 | if (cpu == smp_processor_id()) | |
590 | return; | |
591 | ||
592 | /* | |
593 | * This is safe, as this function is called with the timer | |
594 | * wheel base lock of (cpu) held. When the CPU is on the way | |
595 | * to idle and has not yet set rq->curr to idle then it will | |
596 | * be serialized on the timer wheel base lock and take the new | |
597 | * timer into account automatically. | |
598 | */ | |
599 | if (rq->curr != rq->idle) | |
600 | return; | |
45bf76df | 601 | |
45bf76df | 602 | /* |
06d8308c TG |
603 | * We can set TIF_RESCHED on the idle task of the other CPU |
604 | * lockless. The worst case is that the other CPU runs the | |
605 | * idle task through an additional NOOP schedule() | |
45bf76df | 606 | */ |
5ed0cec0 | 607 | set_tsk_need_resched(rq->idle); |
45bf76df | 608 | |
06d8308c TG |
609 | /* NEED_RESCHED must be visible before we test polling */ |
610 | smp_mb(); | |
611 | if (!tsk_is_polling(rq->idle)) | |
612 | smp_send_reschedule(cpu); | |
45bf76df IM |
613 | } |
614 | ||
c5bfece2 | 615 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 616 | { |
c5bfece2 | 617 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
618 | if (cpu != smp_processor_id() || |
619 | tick_nohz_tick_stopped()) | |
620 | smp_send_reschedule(cpu); | |
621 | return true; | |
622 | } | |
623 | ||
624 | return false; | |
625 | } | |
626 | ||
627 | void wake_up_nohz_cpu(int cpu) | |
628 | { | |
c5bfece2 | 629 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
630 | wake_up_idle_cpu(cpu); |
631 | } | |
632 | ||
ca38062e | 633 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 634 | { |
1c792db7 SS |
635 | int cpu = smp_processor_id(); |
636 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
637 | } |
638 | ||
3451d024 | 639 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 640 | |
ca38062e | 641 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 642 | { |
ca38062e | 643 | return false; |
2069dd75 PZ |
644 | } |
645 | ||
3451d024 | 646 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 647 | |
ce831b38 FW |
648 | #ifdef CONFIG_NO_HZ_FULL |
649 | bool sched_can_stop_tick(void) | |
650 | { | |
651 | struct rq *rq; | |
652 | ||
653 | rq = this_rq(); | |
654 | ||
655 | /* Make sure rq->nr_running update is visible after the IPI */ | |
656 | smp_rmb(); | |
657 | ||
658 | /* More than one running task need preemption */ | |
659 | if (rq->nr_running > 1) | |
660 | return false; | |
661 | ||
662 | return true; | |
663 | } | |
664 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 665 | |
029632fb | 666 | void sched_avg_update(struct rq *rq) |
18d95a28 | 667 | { |
e9e9250b PZ |
668 | s64 period = sched_avg_period(); |
669 | ||
670 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
671 | /* |
672 | * Inline assembly required to prevent the compiler | |
673 | * optimising this loop into a divmod call. | |
674 | * See __iter_div_u64_rem() for another example of this. | |
675 | */ | |
676 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
677 | rq->age_stamp += period; |
678 | rq->rt_avg /= 2; | |
679 | } | |
18d95a28 PZ |
680 | } |
681 | ||
6d6bc0ad | 682 | #else /* !CONFIG_SMP */ |
029632fb | 683 | void resched_task(struct task_struct *p) |
18d95a28 | 684 | { |
05fa785c | 685 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 686 | set_tsk_need_resched(p); |
18d95a28 | 687 | } |
6d6bc0ad | 688 | #endif /* CONFIG_SMP */ |
18d95a28 | 689 | |
a790de99 PT |
690 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
691 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 692 | /* |
8277434e PT |
693 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
694 | * node and @up when leaving it for the final time. | |
695 | * | |
696 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 697 | */ |
029632fb | 698 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 699 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
700 | { |
701 | struct task_group *parent, *child; | |
eb755805 | 702 | int ret; |
c09595f6 | 703 | |
8277434e PT |
704 | parent = from; |
705 | ||
c09595f6 | 706 | down: |
eb755805 PZ |
707 | ret = (*down)(parent, data); |
708 | if (ret) | |
8277434e | 709 | goto out; |
c09595f6 PZ |
710 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
711 | parent = child; | |
712 | goto down; | |
713 | ||
714 | up: | |
715 | continue; | |
716 | } | |
eb755805 | 717 | ret = (*up)(parent, data); |
8277434e PT |
718 | if (ret || parent == from) |
719 | goto out; | |
c09595f6 PZ |
720 | |
721 | child = parent; | |
722 | parent = parent->parent; | |
723 | if (parent) | |
724 | goto up; | |
8277434e | 725 | out: |
eb755805 | 726 | return ret; |
c09595f6 PZ |
727 | } |
728 | ||
029632fb | 729 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 730 | { |
e2b245f8 | 731 | return 0; |
eb755805 | 732 | } |
18d95a28 PZ |
733 | #endif |
734 | ||
45bf76df IM |
735 | static void set_load_weight(struct task_struct *p) |
736 | { | |
f05998d4 NR |
737 | int prio = p->static_prio - MAX_RT_PRIO; |
738 | struct load_weight *load = &p->se.load; | |
739 | ||
dd41f596 IM |
740 | /* |
741 | * SCHED_IDLE tasks get minimal weight: | |
742 | */ | |
743 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 744 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 745 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
746 | return; |
747 | } | |
71f8bd46 | 748 | |
c8b28116 | 749 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 750 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
751 | } |
752 | ||
371fd7e7 | 753 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 754 | { |
a64692a3 | 755 | update_rq_clock(rq); |
dd41f596 | 756 | sched_info_queued(p); |
371fd7e7 | 757 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
758 | } |
759 | ||
371fd7e7 | 760 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 761 | { |
a64692a3 | 762 | update_rq_clock(rq); |
46ac22ba | 763 | sched_info_dequeued(p); |
371fd7e7 | 764 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
765 | } |
766 | ||
029632fb | 767 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
768 | { |
769 | if (task_contributes_to_load(p)) | |
770 | rq->nr_uninterruptible--; | |
771 | ||
371fd7e7 | 772 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
773 | } |
774 | ||
029632fb | 775 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
776 | { |
777 | if (task_contributes_to_load(p)) | |
778 | rq->nr_uninterruptible++; | |
779 | ||
371fd7e7 | 780 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
781 | } |
782 | ||
fe44d621 | 783 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 784 | { |
095c0aa8 GC |
785 | /* |
786 | * In theory, the compile should just see 0 here, and optimize out the call | |
787 | * to sched_rt_avg_update. But I don't trust it... | |
788 | */ | |
789 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
790 | s64 steal = 0, irq_delta = 0; | |
791 | #endif | |
792 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 793 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
794 | |
795 | /* | |
796 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
797 | * this case when a previous update_rq_clock() happened inside a | |
798 | * {soft,}irq region. | |
799 | * | |
800 | * When this happens, we stop ->clock_task and only update the | |
801 | * prev_irq_time stamp to account for the part that fit, so that a next | |
802 | * update will consume the rest. This ensures ->clock_task is | |
803 | * monotonic. | |
804 | * | |
805 | * It does however cause some slight miss-attribution of {soft,}irq | |
806 | * time, a more accurate solution would be to update the irq_time using | |
807 | * the current rq->clock timestamp, except that would require using | |
808 | * atomic ops. | |
809 | */ | |
810 | if (irq_delta > delta) | |
811 | irq_delta = delta; | |
812 | ||
813 | rq->prev_irq_time += irq_delta; | |
814 | delta -= irq_delta; | |
095c0aa8 GC |
815 | #endif |
816 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 817 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
818 | u64 st; |
819 | ||
820 | steal = paravirt_steal_clock(cpu_of(rq)); | |
821 | steal -= rq->prev_steal_time_rq; | |
822 | ||
823 | if (unlikely(steal > delta)) | |
824 | steal = delta; | |
825 | ||
826 | st = steal_ticks(steal); | |
827 | steal = st * TICK_NSEC; | |
828 | ||
829 | rq->prev_steal_time_rq += steal; | |
830 | ||
831 | delta -= steal; | |
832 | } | |
833 | #endif | |
834 | ||
fe44d621 PZ |
835 | rq->clock_task += delta; |
836 | ||
095c0aa8 GC |
837 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
838 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
839 | sched_rt_avg_update(rq, irq_delta + steal); | |
840 | #endif | |
aa483808 VP |
841 | } |
842 | ||
34f971f6 PZ |
843 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
844 | { | |
845 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
846 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
847 | ||
848 | if (stop) { | |
849 | /* | |
850 | * Make it appear like a SCHED_FIFO task, its something | |
851 | * userspace knows about and won't get confused about. | |
852 | * | |
853 | * Also, it will make PI more or less work without too | |
854 | * much confusion -- but then, stop work should not | |
855 | * rely on PI working anyway. | |
856 | */ | |
857 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
858 | ||
859 | stop->sched_class = &stop_sched_class; | |
860 | } | |
861 | ||
862 | cpu_rq(cpu)->stop = stop; | |
863 | ||
864 | if (old_stop) { | |
865 | /* | |
866 | * Reset it back to a normal scheduling class so that | |
867 | * it can die in pieces. | |
868 | */ | |
869 | old_stop->sched_class = &rt_sched_class; | |
870 | } | |
871 | } | |
872 | ||
14531189 | 873 | /* |
dd41f596 | 874 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 875 | */ |
14531189 IM |
876 | static inline int __normal_prio(struct task_struct *p) |
877 | { | |
dd41f596 | 878 | return p->static_prio; |
14531189 IM |
879 | } |
880 | ||
b29739f9 IM |
881 | /* |
882 | * Calculate the expected normal priority: i.e. priority | |
883 | * without taking RT-inheritance into account. Might be | |
884 | * boosted by interactivity modifiers. Changes upon fork, | |
885 | * setprio syscalls, and whenever the interactivity | |
886 | * estimator recalculates. | |
887 | */ | |
36c8b586 | 888 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
889 | { |
890 | int prio; | |
891 | ||
e05606d3 | 892 | if (task_has_rt_policy(p)) |
b29739f9 IM |
893 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
894 | else | |
895 | prio = __normal_prio(p); | |
896 | return prio; | |
897 | } | |
898 | ||
899 | /* | |
900 | * Calculate the current priority, i.e. the priority | |
901 | * taken into account by the scheduler. This value might | |
902 | * be boosted by RT tasks, or might be boosted by | |
903 | * interactivity modifiers. Will be RT if the task got | |
904 | * RT-boosted. If not then it returns p->normal_prio. | |
905 | */ | |
36c8b586 | 906 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
907 | { |
908 | p->normal_prio = normal_prio(p); | |
909 | /* | |
910 | * If we are RT tasks or we were boosted to RT priority, | |
911 | * keep the priority unchanged. Otherwise, update priority | |
912 | * to the normal priority: | |
913 | */ | |
914 | if (!rt_prio(p->prio)) | |
915 | return p->normal_prio; | |
916 | return p->prio; | |
917 | } | |
918 | ||
1da177e4 LT |
919 | /** |
920 | * task_curr - is this task currently executing on a CPU? | |
921 | * @p: the task in question. | |
922 | */ | |
36c8b586 | 923 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
924 | { |
925 | return cpu_curr(task_cpu(p)) == p; | |
926 | } | |
927 | ||
cb469845 SR |
928 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
929 | const struct sched_class *prev_class, | |
da7a735e | 930 | int oldprio) |
cb469845 SR |
931 | { |
932 | if (prev_class != p->sched_class) { | |
933 | if (prev_class->switched_from) | |
da7a735e PZ |
934 | prev_class->switched_from(rq, p); |
935 | p->sched_class->switched_to(rq, p); | |
936 | } else if (oldprio != p->prio) | |
937 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
938 | } |
939 | ||
029632fb | 940 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
941 | { |
942 | const struct sched_class *class; | |
943 | ||
944 | if (p->sched_class == rq->curr->sched_class) { | |
945 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
946 | } else { | |
947 | for_each_class(class) { | |
948 | if (class == rq->curr->sched_class) | |
949 | break; | |
950 | if (class == p->sched_class) { | |
951 | resched_task(rq->curr); | |
952 | break; | |
953 | } | |
954 | } | |
955 | } | |
956 | ||
957 | /* | |
958 | * A queue event has occurred, and we're going to schedule. In | |
959 | * this case, we can save a useless back to back clock update. | |
960 | */ | |
fd2f4419 | 961 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
962 | rq->skip_clock_update = 1; |
963 | } | |
964 | ||
582b336e MT |
965 | static ATOMIC_NOTIFIER_HEAD(task_migration_notifier); |
966 | ||
967 | void register_task_migration_notifier(struct notifier_block *n) | |
968 | { | |
969 | atomic_notifier_chain_register(&task_migration_notifier, n); | |
970 | } | |
971 | ||
1da177e4 | 972 | #ifdef CONFIG_SMP |
dd41f596 | 973 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 974 | { |
e2912009 PZ |
975 | #ifdef CONFIG_SCHED_DEBUG |
976 | /* | |
977 | * We should never call set_task_cpu() on a blocked task, | |
978 | * ttwu() will sort out the placement. | |
979 | */ | |
077614ee PZ |
980 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
981 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
982 | |
983 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
984 | /* |
985 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
986 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
987 | * | |
988 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 989 | * see task_group(). |
6c6c54e1 PZ |
990 | * |
991 | * Furthermore, all task_rq users should acquire both locks, see | |
992 | * task_rq_lock(). | |
993 | */ | |
0122ec5b PZ |
994 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
995 | lockdep_is_held(&task_rq(p)->lock))); | |
996 | #endif | |
e2912009 PZ |
997 | #endif |
998 | ||
de1d7286 | 999 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1000 | |
0c69774e | 1001 | if (task_cpu(p) != new_cpu) { |
582b336e MT |
1002 | struct task_migration_notifier tmn; |
1003 | ||
0a74bef8 PT |
1004 | if (p->sched_class->migrate_task_rq) |
1005 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1006 | p->se.nr_migrations++; |
a8b0ca17 | 1007 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
582b336e MT |
1008 | |
1009 | tmn.task = p; | |
1010 | tmn.from_cpu = task_cpu(p); | |
1011 | tmn.to_cpu = new_cpu; | |
1012 | ||
1013 | atomic_notifier_call_chain(&task_migration_notifier, 0, &tmn); | |
0c69774e | 1014 | } |
dd41f596 IM |
1015 | |
1016 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1017 | } |
1018 | ||
969c7921 | 1019 | struct migration_arg { |
36c8b586 | 1020 | struct task_struct *task; |
1da177e4 | 1021 | int dest_cpu; |
70b97a7f | 1022 | }; |
1da177e4 | 1023 | |
969c7921 TH |
1024 | static int migration_cpu_stop(void *data); |
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * wait_task_inactive - wait for a thread to unschedule. | |
1028 | * | |
85ba2d86 RM |
1029 | * If @match_state is nonzero, it's the @p->state value just checked and |
1030 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1031 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1032 | * we return a positive number (its total switch count). If a second call | |
1033 | * a short while later returns the same number, the caller can be sure that | |
1034 | * @p has remained unscheduled the whole time. | |
1035 | * | |
1da177e4 LT |
1036 | * The caller must ensure that the task *will* unschedule sometime soon, |
1037 | * else this function might spin for a *long* time. This function can't | |
1038 | * be called with interrupts off, or it may introduce deadlock with | |
1039 | * smp_call_function() if an IPI is sent by the same process we are | |
1040 | * waiting to become inactive. | |
1041 | */ | |
85ba2d86 | 1042 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1043 | { |
1044 | unsigned long flags; | |
dd41f596 | 1045 | int running, on_rq; |
85ba2d86 | 1046 | unsigned long ncsw; |
70b97a7f | 1047 | struct rq *rq; |
1da177e4 | 1048 | |
3a5c359a AK |
1049 | for (;;) { |
1050 | /* | |
1051 | * We do the initial early heuristics without holding | |
1052 | * any task-queue locks at all. We'll only try to get | |
1053 | * the runqueue lock when things look like they will | |
1054 | * work out! | |
1055 | */ | |
1056 | rq = task_rq(p); | |
fa490cfd | 1057 | |
3a5c359a AK |
1058 | /* |
1059 | * If the task is actively running on another CPU | |
1060 | * still, just relax and busy-wait without holding | |
1061 | * any locks. | |
1062 | * | |
1063 | * NOTE! Since we don't hold any locks, it's not | |
1064 | * even sure that "rq" stays as the right runqueue! | |
1065 | * But we don't care, since "task_running()" will | |
1066 | * return false if the runqueue has changed and p | |
1067 | * is actually now running somewhere else! | |
1068 | */ | |
85ba2d86 RM |
1069 | while (task_running(rq, p)) { |
1070 | if (match_state && unlikely(p->state != match_state)) | |
1071 | return 0; | |
3a5c359a | 1072 | cpu_relax(); |
85ba2d86 | 1073 | } |
fa490cfd | 1074 | |
3a5c359a AK |
1075 | /* |
1076 | * Ok, time to look more closely! We need the rq | |
1077 | * lock now, to be *sure*. If we're wrong, we'll | |
1078 | * just go back and repeat. | |
1079 | */ | |
1080 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1081 | trace_sched_wait_task(p); |
3a5c359a | 1082 | running = task_running(rq, p); |
fd2f4419 | 1083 | on_rq = p->on_rq; |
85ba2d86 | 1084 | ncsw = 0; |
f31e11d8 | 1085 | if (!match_state || p->state == match_state) |
93dcf55f | 1086 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1087 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1088 | |
85ba2d86 RM |
1089 | /* |
1090 | * If it changed from the expected state, bail out now. | |
1091 | */ | |
1092 | if (unlikely(!ncsw)) | |
1093 | break; | |
1094 | ||
3a5c359a AK |
1095 | /* |
1096 | * Was it really running after all now that we | |
1097 | * checked with the proper locks actually held? | |
1098 | * | |
1099 | * Oops. Go back and try again.. | |
1100 | */ | |
1101 | if (unlikely(running)) { | |
1102 | cpu_relax(); | |
1103 | continue; | |
1104 | } | |
fa490cfd | 1105 | |
3a5c359a AK |
1106 | /* |
1107 | * It's not enough that it's not actively running, | |
1108 | * it must be off the runqueue _entirely_, and not | |
1109 | * preempted! | |
1110 | * | |
80dd99b3 | 1111 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1112 | * running right now), it's preempted, and we should |
1113 | * yield - it could be a while. | |
1114 | */ | |
1115 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1116 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1117 | ||
1118 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1119 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1120 | continue; |
1121 | } | |
fa490cfd | 1122 | |
3a5c359a AK |
1123 | /* |
1124 | * Ahh, all good. It wasn't running, and it wasn't | |
1125 | * runnable, which means that it will never become | |
1126 | * running in the future either. We're all done! | |
1127 | */ | |
1128 | break; | |
1129 | } | |
85ba2d86 RM |
1130 | |
1131 | return ncsw; | |
1da177e4 LT |
1132 | } |
1133 | ||
1134 | /*** | |
1135 | * kick_process - kick a running thread to enter/exit the kernel | |
1136 | * @p: the to-be-kicked thread | |
1137 | * | |
1138 | * Cause a process which is running on another CPU to enter | |
1139 | * kernel-mode, without any delay. (to get signals handled.) | |
1140 | * | |
25985edc | 1141 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1142 | * because all it wants to ensure is that the remote task enters |
1143 | * the kernel. If the IPI races and the task has been migrated | |
1144 | * to another CPU then no harm is done and the purpose has been | |
1145 | * achieved as well. | |
1146 | */ | |
36c8b586 | 1147 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1148 | { |
1149 | int cpu; | |
1150 | ||
1151 | preempt_disable(); | |
1152 | cpu = task_cpu(p); | |
1153 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1154 | smp_send_reschedule(cpu); | |
1155 | preempt_enable(); | |
1156 | } | |
b43e3521 | 1157 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1158 | #endif /* CONFIG_SMP */ |
1da177e4 | 1159 | |
970b13ba | 1160 | #ifdef CONFIG_SMP |
30da688e | 1161 | /* |
013fdb80 | 1162 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1163 | */ |
5da9a0fb PZ |
1164 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1165 | { | |
aa00d89c TC |
1166 | int nid = cpu_to_node(cpu); |
1167 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1168 | enum { cpuset, possible, fail } state = cpuset; |
1169 | int dest_cpu; | |
5da9a0fb | 1170 | |
aa00d89c TC |
1171 | /* |
1172 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1173 | * will return -1. There is no cpu on the node, and we should | |
1174 | * select the cpu on the other node. | |
1175 | */ | |
1176 | if (nid != -1) { | |
1177 | nodemask = cpumask_of_node(nid); | |
1178 | ||
1179 | /* Look for allowed, online CPU in same node. */ | |
1180 | for_each_cpu(dest_cpu, nodemask) { | |
1181 | if (!cpu_online(dest_cpu)) | |
1182 | continue; | |
1183 | if (!cpu_active(dest_cpu)) | |
1184 | continue; | |
1185 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1186 | return dest_cpu; | |
1187 | } | |
2baab4e9 | 1188 | } |
5da9a0fb | 1189 | |
2baab4e9 PZ |
1190 | for (;;) { |
1191 | /* Any allowed, online CPU? */ | |
e3831edd | 1192 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1193 | if (!cpu_online(dest_cpu)) |
1194 | continue; | |
1195 | if (!cpu_active(dest_cpu)) | |
1196 | continue; | |
1197 | goto out; | |
1198 | } | |
5da9a0fb | 1199 | |
2baab4e9 PZ |
1200 | switch (state) { |
1201 | case cpuset: | |
1202 | /* No more Mr. Nice Guy. */ | |
1203 | cpuset_cpus_allowed_fallback(p); | |
1204 | state = possible; | |
1205 | break; | |
1206 | ||
1207 | case possible: | |
1208 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1209 | state = fail; | |
1210 | break; | |
1211 | ||
1212 | case fail: | |
1213 | BUG(); | |
1214 | break; | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | out: | |
1219 | if (state != cpuset) { | |
1220 | /* | |
1221 | * Don't tell them about moving exiting tasks or | |
1222 | * kernel threads (both mm NULL), since they never | |
1223 | * leave kernel. | |
1224 | */ | |
1225 | if (p->mm && printk_ratelimit()) { | |
1226 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1227 | task_pid_nr(p), p->comm, cpu); | |
1228 | } | |
5da9a0fb PZ |
1229 | } |
1230 | ||
1231 | return dest_cpu; | |
1232 | } | |
1233 | ||
e2912009 | 1234 | /* |
013fdb80 | 1235 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1236 | */ |
970b13ba | 1237 | static inline |
7608dec2 | 1238 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1239 | { |
7608dec2 | 1240 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1241 | |
1242 | /* | |
1243 | * In order not to call set_task_cpu() on a blocking task we need | |
1244 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1245 | * cpu. | |
1246 | * | |
1247 | * Since this is common to all placement strategies, this lives here. | |
1248 | * | |
1249 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1250 | * not worry about this generic constraint ] | |
1251 | */ | |
fa17b507 | 1252 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1253 | !cpu_online(cpu))) |
5da9a0fb | 1254 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1255 | |
1256 | return cpu; | |
970b13ba | 1257 | } |
09a40af5 MG |
1258 | |
1259 | static void update_avg(u64 *avg, u64 sample) | |
1260 | { | |
1261 | s64 diff = sample - *avg; | |
1262 | *avg += diff >> 3; | |
1263 | } | |
970b13ba PZ |
1264 | #endif |
1265 | ||
d7c01d27 | 1266 | static void |
b84cb5df | 1267 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1268 | { |
d7c01d27 | 1269 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1270 | struct rq *rq = this_rq(); |
1271 | ||
d7c01d27 PZ |
1272 | #ifdef CONFIG_SMP |
1273 | int this_cpu = smp_processor_id(); | |
1274 | ||
1275 | if (cpu == this_cpu) { | |
1276 | schedstat_inc(rq, ttwu_local); | |
1277 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1278 | } else { | |
1279 | struct sched_domain *sd; | |
1280 | ||
1281 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1282 | rcu_read_lock(); |
d7c01d27 PZ |
1283 | for_each_domain(this_cpu, sd) { |
1284 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1285 | schedstat_inc(sd, ttwu_wake_remote); | |
1286 | break; | |
1287 | } | |
1288 | } | |
057f3fad | 1289 | rcu_read_unlock(); |
d7c01d27 | 1290 | } |
f339b9dc PZ |
1291 | |
1292 | if (wake_flags & WF_MIGRATED) | |
1293 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1294 | ||
d7c01d27 PZ |
1295 | #endif /* CONFIG_SMP */ |
1296 | ||
1297 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1298 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1299 | |
1300 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1301 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1302 | |
d7c01d27 PZ |
1303 | #endif /* CONFIG_SCHEDSTATS */ |
1304 | } | |
1305 | ||
1306 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1307 | { | |
9ed3811a | 1308 | activate_task(rq, p, en_flags); |
fd2f4419 | 1309 | p->on_rq = 1; |
c2f7115e PZ |
1310 | |
1311 | /* if a worker is waking up, notify workqueue */ | |
1312 | if (p->flags & PF_WQ_WORKER) | |
1313 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1314 | } |
1315 | ||
23f41eeb PZ |
1316 | /* |
1317 | * Mark the task runnable and perform wakeup-preemption. | |
1318 | */ | |
89363381 | 1319 | static void |
23f41eeb | 1320 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1321 | { |
9ed3811a | 1322 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1323 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1324 | |
1325 | p->state = TASK_RUNNING; | |
1326 | #ifdef CONFIG_SMP | |
1327 | if (p->sched_class->task_woken) | |
1328 | p->sched_class->task_woken(rq, p); | |
1329 | ||
e69c6341 | 1330 | if (rq->idle_stamp) { |
9ed3811a TH |
1331 | u64 delta = rq->clock - rq->idle_stamp; |
1332 | u64 max = 2*sysctl_sched_migration_cost; | |
1333 | ||
1334 | if (delta > max) | |
1335 | rq->avg_idle = max; | |
1336 | else | |
1337 | update_avg(&rq->avg_idle, delta); | |
1338 | rq->idle_stamp = 0; | |
1339 | } | |
1340 | #endif | |
1341 | } | |
1342 | ||
c05fbafb PZ |
1343 | static void |
1344 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1345 | { | |
1346 | #ifdef CONFIG_SMP | |
1347 | if (p->sched_contributes_to_load) | |
1348 | rq->nr_uninterruptible--; | |
1349 | #endif | |
1350 | ||
1351 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1352 | ttwu_do_wakeup(rq, p, wake_flags); | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1357 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1358 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1359 | * the task is still ->on_rq. | |
1360 | */ | |
1361 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1362 | { | |
1363 | struct rq *rq; | |
1364 | int ret = 0; | |
1365 | ||
1366 | rq = __task_rq_lock(p); | |
1367 | if (p->on_rq) { | |
1368 | ttwu_do_wakeup(rq, p, wake_flags); | |
1369 | ret = 1; | |
1370 | } | |
1371 | __task_rq_unlock(rq); | |
1372 | ||
1373 | return ret; | |
1374 | } | |
1375 | ||
317f3941 | 1376 | #ifdef CONFIG_SMP |
fa14ff4a | 1377 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1378 | { |
1379 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1380 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1381 | struct task_struct *p; | |
317f3941 PZ |
1382 | |
1383 | raw_spin_lock(&rq->lock); | |
1384 | ||
fa14ff4a PZ |
1385 | while (llist) { |
1386 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1387 | llist = llist_next(llist); | |
317f3941 PZ |
1388 | ttwu_do_activate(rq, p, 0); |
1389 | } | |
1390 | ||
1391 | raw_spin_unlock(&rq->lock); | |
1392 | } | |
1393 | ||
1394 | void scheduler_ipi(void) | |
1395 | { | |
ff442c51 FW |
1396 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick() |
1397 | && !tick_nohz_full_cpu(smp_processor_id())) | |
c5d753a5 PZ |
1398 | return; |
1399 | ||
1400 | /* | |
1401 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1402 | * traditionally all their work was done from the interrupt return | |
1403 | * path. Now that we actually do some work, we need to make sure | |
1404 | * we do call them. | |
1405 | * | |
1406 | * Some archs already do call them, luckily irq_enter/exit nest | |
1407 | * properly. | |
1408 | * | |
1409 | * Arguably we should visit all archs and update all handlers, | |
1410 | * however a fair share of IPIs are still resched only so this would | |
1411 | * somewhat pessimize the simple resched case. | |
1412 | */ | |
1413 | irq_enter(); | |
ff442c51 | 1414 | tick_nohz_full_check(); |
fa14ff4a | 1415 | sched_ttwu_pending(); |
ca38062e SS |
1416 | |
1417 | /* | |
1418 | * Check if someone kicked us for doing the nohz idle load balance. | |
1419 | */ | |
6eb57e0d SS |
1420 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1421 | this_rq()->idle_balance = 1; | |
ca38062e | 1422 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1423 | } |
c5d753a5 | 1424 | irq_exit(); |
317f3941 PZ |
1425 | } |
1426 | ||
1427 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1428 | { | |
fa14ff4a | 1429 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1430 | smp_send_reschedule(cpu); |
1431 | } | |
d6aa8f85 | 1432 | |
39be3501 | 1433 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1434 | { |
1435 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1436 | } | |
d6aa8f85 | 1437 | #endif /* CONFIG_SMP */ |
317f3941 | 1438 | |
c05fbafb PZ |
1439 | static void ttwu_queue(struct task_struct *p, int cpu) |
1440 | { | |
1441 | struct rq *rq = cpu_rq(cpu); | |
1442 | ||
17d9f311 | 1443 | #if defined(CONFIG_SMP) |
39be3501 | 1444 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1445 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1446 | ttwu_queue_remote(p, cpu); |
1447 | return; | |
1448 | } | |
1449 | #endif | |
1450 | ||
c05fbafb PZ |
1451 | raw_spin_lock(&rq->lock); |
1452 | ttwu_do_activate(rq, p, 0); | |
1453 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1454 | } |
1455 | ||
1456 | /** | |
1da177e4 | 1457 | * try_to_wake_up - wake up a thread |
9ed3811a | 1458 | * @p: the thread to be awakened |
1da177e4 | 1459 | * @state: the mask of task states that can be woken |
9ed3811a | 1460 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1461 | * |
1462 | * Put it on the run-queue if it's not already there. The "current" | |
1463 | * thread is always on the run-queue (except when the actual | |
1464 | * re-schedule is in progress), and as such you're allowed to do | |
1465 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1466 | * runnable without the overhead of this. | |
1467 | * | |
9ed3811a TH |
1468 | * Returns %true if @p was woken up, %false if it was already running |
1469 | * or @state didn't match @p's state. | |
1da177e4 | 1470 | */ |
e4a52bcb PZ |
1471 | static int |
1472 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1473 | { |
1da177e4 | 1474 | unsigned long flags; |
c05fbafb | 1475 | int cpu, success = 0; |
2398f2c6 | 1476 | |
04e2f174 | 1477 | smp_wmb(); |
013fdb80 | 1478 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1479 | if (!(p->state & state)) |
1da177e4 LT |
1480 | goto out; |
1481 | ||
c05fbafb | 1482 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1483 | cpu = task_cpu(p); |
1da177e4 | 1484 | |
c05fbafb PZ |
1485 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1486 | goto stat; | |
1da177e4 | 1487 | |
1da177e4 | 1488 | #ifdef CONFIG_SMP |
e9c84311 | 1489 | /* |
c05fbafb PZ |
1490 | * If the owning (remote) cpu is still in the middle of schedule() with |
1491 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1492 | */ |
f3e94786 | 1493 | while (p->on_cpu) |
e4a52bcb | 1494 | cpu_relax(); |
0970d299 | 1495 | /* |
e4a52bcb | 1496 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1497 | */ |
e4a52bcb | 1498 | smp_rmb(); |
1da177e4 | 1499 | |
a8e4f2ea | 1500 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1501 | p->state = TASK_WAKING; |
e7693a36 | 1502 | |
e4a52bcb | 1503 | if (p->sched_class->task_waking) |
74f8e4b2 | 1504 | p->sched_class->task_waking(p); |
efbbd05a | 1505 | |
7608dec2 | 1506 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1507 | if (task_cpu(p) != cpu) { |
1508 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1509 | set_task_cpu(p, cpu); |
f339b9dc | 1510 | } |
1da177e4 | 1511 | #endif /* CONFIG_SMP */ |
1da177e4 | 1512 | |
c05fbafb PZ |
1513 | ttwu_queue(p, cpu); |
1514 | stat: | |
b84cb5df | 1515 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1516 | out: |
013fdb80 | 1517 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1518 | |
1519 | return success; | |
1520 | } | |
1521 | ||
21aa9af0 TH |
1522 | /** |
1523 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1524 | * @p: the thread to be awakened | |
1525 | * | |
2acca55e | 1526 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1527 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1528 | * the current task. |
21aa9af0 TH |
1529 | */ |
1530 | static void try_to_wake_up_local(struct task_struct *p) | |
1531 | { | |
1532 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1533 | |
383efcd0 TH |
1534 | if (WARN_ON_ONCE(rq != this_rq()) || |
1535 | WARN_ON_ONCE(p == current)) | |
1536 | return; | |
1537 | ||
21aa9af0 TH |
1538 | lockdep_assert_held(&rq->lock); |
1539 | ||
2acca55e PZ |
1540 | if (!raw_spin_trylock(&p->pi_lock)) { |
1541 | raw_spin_unlock(&rq->lock); | |
1542 | raw_spin_lock(&p->pi_lock); | |
1543 | raw_spin_lock(&rq->lock); | |
1544 | } | |
1545 | ||
21aa9af0 | 1546 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1547 | goto out; |
21aa9af0 | 1548 | |
fd2f4419 | 1549 | if (!p->on_rq) |
d7c01d27 PZ |
1550 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1551 | ||
23f41eeb | 1552 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1553 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1554 | out: |
1555 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1556 | } |
1557 | ||
50fa610a DH |
1558 | /** |
1559 | * wake_up_process - Wake up a specific process | |
1560 | * @p: The process to be woken up. | |
1561 | * | |
1562 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1563 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1564 | * running. | |
1565 | * | |
1566 | * It may be assumed that this function implies a write memory barrier before | |
1567 | * changing the task state if and only if any tasks are woken up. | |
1568 | */ | |
7ad5b3a5 | 1569 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1570 | { |
9067ac85 ON |
1571 | WARN_ON(task_is_stopped_or_traced(p)); |
1572 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1573 | } |
1da177e4 LT |
1574 | EXPORT_SYMBOL(wake_up_process); |
1575 | ||
7ad5b3a5 | 1576 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1577 | { |
1578 | return try_to_wake_up(p, state, 0); | |
1579 | } | |
1580 | ||
1da177e4 LT |
1581 | /* |
1582 | * Perform scheduler related setup for a newly forked process p. | |
1583 | * p is forked by current. | |
dd41f596 IM |
1584 | * |
1585 | * __sched_fork() is basic setup used by init_idle() too: | |
1586 | */ | |
1587 | static void __sched_fork(struct task_struct *p) | |
1588 | { | |
fd2f4419 PZ |
1589 | p->on_rq = 0; |
1590 | ||
1591 | p->se.on_rq = 0; | |
dd41f596 IM |
1592 | p->se.exec_start = 0; |
1593 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1594 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1595 | p->se.nr_migrations = 0; |
da7a735e | 1596 | p->se.vruntime = 0; |
fd2f4419 | 1597 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 1598 | |
f4e26b12 PT |
1599 | /* |
1600 | * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be | |
1601 | * removed when useful for applications beyond shares distribution (e.g. | |
1602 | * load-balance). | |
1603 | */ | |
1604 | #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED) | |
9d85f21c PT |
1605 | p->se.avg.runnable_avg_period = 0; |
1606 | p->se.avg.runnable_avg_sum = 0; | |
1607 | #endif | |
6cfb0d5d | 1608 | #ifdef CONFIG_SCHEDSTATS |
41acab88 | 1609 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1610 | #endif |
476d139c | 1611 | |
fa717060 | 1612 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1613 | |
e107be36 AK |
1614 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1615 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1616 | #endif | |
cbee9f88 PZ |
1617 | |
1618 | #ifdef CONFIG_NUMA_BALANCING | |
1619 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
1620 | p->mm->numa_next_scan = jiffies; | |
b8593bfd | 1621 | p->mm->numa_next_reset = jiffies; |
cbee9f88 PZ |
1622 | p->mm->numa_scan_seq = 0; |
1623 | } | |
1624 | ||
1625 | p->node_stamp = 0ULL; | |
1626 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
1627 | p->numa_migrate_seq = p->mm ? p->mm->numa_scan_seq - 1 : 0; | |
4b96a29b | 1628 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 PZ |
1629 | p->numa_work.next = &p->numa_work; |
1630 | #endif /* CONFIG_NUMA_BALANCING */ | |
dd41f596 IM |
1631 | } |
1632 | ||
1a687c2e | 1633 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1634 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1635 | void set_numabalancing_state(bool enabled) |
1636 | { | |
1637 | if (enabled) | |
1638 | sched_feat_set("NUMA"); | |
1639 | else | |
1640 | sched_feat_set("NO_NUMA"); | |
1641 | } | |
3105b86a MG |
1642 | #else |
1643 | __read_mostly bool numabalancing_enabled; | |
1644 | ||
1645 | void set_numabalancing_state(bool enabled) | |
1646 | { | |
1647 | numabalancing_enabled = enabled; | |
dd41f596 | 1648 | } |
3105b86a | 1649 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1650 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1651 | |
1652 | /* | |
1653 | * fork()/clone()-time setup: | |
1654 | */ | |
3e51e3ed | 1655 | void sched_fork(struct task_struct *p) |
dd41f596 | 1656 | { |
0122ec5b | 1657 | unsigned long flags; |
dd41f596 IM |
1658 | int cpu = get_cpu(); |
1659 | ||
1660 | __sched_fork(p); | |
06b83b5f | 1661 | /* |
0017d735 | 1662 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1663 | * nobody will actually run it, and a signal or other external |
1664 | * event cannot wake it up and insert it on the runqueue either. | |
1665 | */ | |
0017d735 | 1666 | p->state = TASK_RUNNING; |
dd41f596 | 1667 | |
c350a04e MG |
1668 | /* |
1669 | * Make sure we do not leak PI boosting priority to the child. | |
1670 | */ | |
1671 | p->prio = current->normal_prio; | |
1672 | ||
b9dc29e7 MG |
1673 | /* |
1674 | * Revert to default priority/policy on fork if requested. | |
1675 | */ | |
1676 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1677 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1678 | p->policy = SCHED_NORMAL; |
6c697bdf | 1679 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1680 | p->rt_priority = 0; |
1681 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1682 | p->static_prio = NICE_TO_PRIO(0); | |
1683 | ||
1684 | p->prio = p->normal_prio = __normal_prio(p); | |
1685 | set_load_weight(p); | |
6c697bdf | 1686 | |
b9dc29e7 MG |
1687 | /* |
1688 | * We don't need the reset flag anymore after the fork. It has | |
1689 | * fulfilled its duty: | |
1690 | */ | |
1691 | p->sched_reset_on_fork = 0; | |
1692 | } | |
ca94c442 | 1693 | |
2ddbf952 HS |
1694 | if (!rt_prio(p->prio)) |
1695 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1696 | |
cd29fe6f PZ |
1697 | if (p->sched_class->task_fork) |
1698 | p->sched_class->task_fork(p); | |
1699 | ||
86951599 PZ |
1700 | /* |
1701 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1702 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1703 | * is ran before sched_fork(). | |
1704 | * | |
1705 | * Silence PROVE_RCU. | |
1706 | */ | |
0122ec5b | 1707 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1708 | set_task_cpu(p, cpu); |
0122ec5b | 1709 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1710 | |
52f17b6c | 1711 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1712 | if (likely(sched_info_on())) |
52f17b6c | 1713 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1714 | #endif |
3ca7a440 PZ |
1715 | #if defined(CONFIG_SMP) |
1716 | p->on_cpu = 0; | |
4866cde0 | 1717 | #endif |
bdd4e85d | 1718 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1719 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1720 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1721 | #endif |
806c09a7 | 1722 | #ifdef CONFIG_SMP |
917b627d | 1723 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1724 | #endif |
917b627d | 1725 | |
476d139c | 1726 | put_cpu(); |
1da177e4 LT |
1727 | } |
1728 | ||
1729 | /* | |
1730 | * wake_up_new_task - wake up a newly created task for the first time. | |
1731 | * | |
1732 | * This function will do some initial scheduler statistics housekeeping | |
1733 | * that must be done for every newly created context, then puts the task | |
1734 | * on the runqueue and wakes it. | |
1735 | */ | |
3e51e3ed | 1736 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1737 | { |
1738 | unsigned long flags; | |
dd41f596 | 1739 | struct rq *rq; |
fabf318e | 1740 | |
ab2515c4 | 1741 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1742 | #ifdef CONFIG_SMP |
1743 | /* | |
1744 | * Fork balancing, do it here and not earlier because: | |
1745 | * - cpus_allowed can change in the fork path | |
1746 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1747 | */ |
ab2515c4 | 1748 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1749 | #endif |
1750 | ||
ab2515c4 | 1751 | rq = __task_rq_lock(p); |
cd29fe6f | 1752 | activate_task(rq, p, 0); |
fd2f4419 | 1753 | p->on_rq = 1; |
89363381 | 1754 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1755 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1756 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1757 | if (p->sched_class->task_woken) |
1758 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1759 | #endif |
0122ec5b | 1760 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1761 | } |
1762 | ||
e107be36 AK |
1763 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1764 | ||
1765 | /** | |
80dd99b3 | 1766 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1767 | * @notifier: notifier struct to register |
e107be36 AK |
1768 | */ |
1769 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1770 | { | |
1771 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1772 | } | |
1773 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1774 | ||
1775 | /** | |
1776 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1777 | * @notifier: notifier struct to unregister |
e107be36 AK |
1778 | * |
1779 | * This is safe to call from within a preemption notifier. | |
1780 | */ | |
1781 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1782 | { | |
1783 | hlist_del(¬ifier->link); | |
1784 | } | |
1785 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1786 | ||
1787 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1788 | { | |
1789 | struct preempt_notifier *notifier; | |
e107be36 | 1790 | |
b67bfe0d | 1791 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1792 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
1793 | } | |
1794 | ||
1795 | static void | |
1796 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1797 | struct task_struct *next) | |
1798 | { | |
1799 | struct preempt_notifier *notifier; | |
e107be36 | 1800 | |
b67bfe0d | 1801 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1802 | notifier->ops->sched_out(notifier, next); |
1803 | } | |
1804 | ||
6d6bc0ad | 1805 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1806 | |
1807 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1808 | { | |
1809 | } | |
1810 | ||
1811 | static void | |
1812 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1813 | struct task_struct *next) | |
1814 | { | |
1815 | } | |
1816 | ||
6d6bc0ad | 1817 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1818 | |
4866cde0 NP |
1819 | /** |
1820 | * prepare_task_switch - prepare to switch tasks | |
1821 | * @rq: the runqueue preparing to switch | |
421cee29 | 1822 | * @prev: the current task that is being switched out |
4866cde0 NP |
1823 | * @next: the task we are going to switch to. |
1824 | * | |
1825 | * This is called with the rq lock held and interrupts off. It must | |
1826 | * be paired with a subsequent finish_task_switch after the context | |
1827 | * switch. | |
1828 | * | |
1829 | * prepare_task_switch sets up locking and calls architecture specific | |
1830 | * hooks. | |
1831 | */ | |
e107be36 AK |
1832 | static inline void |
1833 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1834 | struct task_struct *next) | |
4866cde0 | 1835 | { |
895dd92c | 1836 | trace_sched_switch(prev, next); |
fe4b04fa PZ |
1837 | sched_info_switch(prev, next); |
1838 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1839 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1840 | prepare_lock_switch(rq, next); |
1841 | prepare_arch_switch(next); | |
1842 | } | |
1843 | ||
1da177e4 LT |
1844 | /** |
1845 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1846 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1847 | * @prev: the thread we just switched away from. |
1848 | * | |
4866cde0 NP |
1849 | * finish_task_switch must be called after the context switch, paired |
1850 | * with a prepare_task_switch call before the context switch. | |
1851 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1852 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1853 | * |
1854 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1855 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1856 | * with the lock held can cause deadlocks; see schedule() for |
1857 | * details.) | |
1858 | */ | |
a9957449 | 1859 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1860 | __releases(rq->lock) |
1861 | { | |
1da177e4 | 1862 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1863 | long prev_state; |
1da177e4 LT |
1864 | |
1865 | rq->prev_mm = NULL; | |
1866 | ||
1867 | /* | |
1868 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1869 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1870 | * schedule one last time. The schedule call will never return, and |
1871 | * the scheduled task must drop that reference. | |
c394cc9f | 1872 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1873 | * still held, otherwise prev could be scheduled on another cpu, die |
1874 | * there before we look at prev->state, and then the reference would | |
1875 | * be dropped twice. | |
1876 | * Manfred Spraul <manfred@colorfullife.com> | |
1877 | */ | |
55a101f8 | 1878 | prev_state = prev->state; |
bf9fae9f | 1879 | vtime_task_switch(prev); |
4866cde0 | 1880 | finish_arch_switch(prev); |
a8d757ef | 1881 | perf_event_task_sched_in(prev, current); |
4866cde0 | 1882 | finish_lock_switch(rq, prev); |
01f23e16 | 1883 | finish_arch_post_lock_switch(); |
e8fa1362 | 1884 | |
e107be36 | 1885 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1886 | if (mm) |
1887 | mmdrop(mm); | |
c394cc9f | 1888 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1889 | /* |
1890 | * Remove function-return probe instances associated with this | |
1891 | * task and put them back on the free list. | |
9761eea8 | 1892 | */ |
c6fd91f0 | 1893 | kprobe_flush_task(prev); |
1da177e4 | 1894 | put_task_struct(prev); |
c6fd91f0 | 1895 | } |
99e5ada9 FW |
1896 | |
1897 | tick_nohz_task_switch(current); | |
1da177e4 LT |
1898 | } |
1899 | ||
3f029d3c GH |
1900 | #ifdef CONFIG_SMP |
1901 | ||
1902 | /* assumes rq->lock is held */ | |
1903 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1904 | { | |
1905 | if (prev->sched_class->pre_schedule) | |
1906 | prev->sched_class->pre_schedule(rq, prev); | |
1907 | } | |
1908 | ||
1909 | /* rq->lock is NOT held, but preemption is disabled */ | |
1910 | static inline void post_schedule(struct rq *rq) | |
1911 | { | |
1912 | if (rq->post_schedule) { | |
1913 | unsigned long flags; | |
1914 | ||
05fa785c | 1915 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1916 | if (rq->curr->sched_class->post_schedule) |
1917 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1918 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1919 | |
1920 | rq->post_schedule = 0; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | #else | |
da19ab51 | 1925 | |
3f029d3c GH |
1926 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1927 | { | |
1928 | } | |
1929 | ||
1930 | static inline void post_schedule(struct rq *rq) | |
1931 | { | |
1da177e4 LT |
1932 | } |
1933 | ||
3f029d3c GH |
1934 | #endif |
1935 | ||
1da177e4 LT |
1936 | /** |
1937 | * schedule_tail - first thing a freshly forked thread must call. | |
1938 | * @prev: the thread we just switched away from. | |
1939 | */ | |
36c8b586 | 1940 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1941 | __releases(rq->lock) |
1942 | { | |
70b97a7f IM |
1943 | struct rq *rq = this_rq(); |
1944 | ||
4866cde0 | 1945 | finish_task_switch(rq, prev); |
da19ab51 | 1946 | |
3f029d3c GH |
1947 | /* |
1948 | * FIXME: do we need to worry about rq being invalidated by the | |
1949 | * task_switch? | |
1950 | */ | |
1951 | post_schedule(rq); | |
70b97a7f | 1952 | |
4866cde0 NP |
1953 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
1954 | /* In this case, finish_task_switch does not reenable preemption */ | |
1955 | preempt_enable(); | |
1956 | #endif | |
1da177e4 | 1957 | if (current->set_child_tid) |
b488893a | 1958 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1959 | } |
1960 | ||
1961 | /* | |
1962 | * context_switch - switch to the new MM and the new | |
1963 | * thread's register state. | |
1964 | */ | |
dd41f596 | 1965 | static inline void |
70b97a7f | 1966 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1967 | struct task_struct *next) |
1da177e4 | 1968 | { |
dd41f596 | 1969 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1970 | |
e107be36 | 1971 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 1972 | |
dd41f596 IM |
1973 | mm = next->mm; |
1974 | oldmm = prev->active_mm; | |
9226d125 ZA |
1975 | /* |
1976 | * For paravirt, this is coupled with an exit in switch_to to | |
1977 | * combine the page table reload and the switch backend into | |
1978 | * one hypercall. | |
1979 | */ | |
224101ed | 1980 | arch_start_context_switch(prev); |
9226d125 | 1981 | |
31915ab4 | 1982 | if (!mm) { |
1da177e4 LT |
1983 | next->active_mm = oldmm; |
1984 | atomic_inc(&oldmm->mm_count); | |
1985 | enter_lazy_tlb(oldmm, next); | |
1986 | } else | |
1987 | switch_mm(oldmm, mm, next); | |
1988 | ||
31915ab4 | 1989 | if (!prev->mm) { |
1da177e4 | 1990 | prev->active_mm = NULL; |
1da177e4 LT |
1991 | rq->prev_mm = oldmm; |
1992 | } | |
3a5f5e48 IM |
1993 | /* |
1994 | * Since the runqueue lock will be released by the next | |
1995 | * task (which is an invalid locking op but in the case | |
1996 | * of the scheduler it's an obvious special-case), so we | |
1997 | * do an early lockdep release here: | |
1998 | */ | |
1999 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2000 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2001 | #endif |
1da177e4 | 2002 | |
91d1aa43 | 2003 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2004 | /* Here we just switch the register state and the stack. */ |
2005 | switch_to(prev, next, prev); | |
2006 | ||
dd41f596 IM |
2007 | barrier(); |
2008 | /* | |
2009 | * this_rq must be evaluated again because prev may have moved | |
2010 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2011 | * frame will be invalid. | |
2012 | */ | |
2013 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2014 | } |
2015 | ||
2016 | /* | |
1c3e8264 | 2017 | * nr_running and nr_context_switches: |
1da177e4 LT |
2018 | * |
2019 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2020 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2021 | */ |
2022 | unsigned long nr_running(void) | |
2023 | { | |
2024 | unsigned long i, sum = 0; | |
2025 | ||
2026 | for_each_online_cpu(i) | |
2027 | sum += cpu_rq(i)->nr_running; | |
2028 | ||
2029 | return sum; | |
f711f609 | 2030 | } |
1da177e4 | 2031 | |
1da177e4 | 2032 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2033 | { |
cc94abfc SR |
2034 | int i; |
2035 | unsigned long long sum = 0; | |
46cb4b7c | 2036 | |
0a945022 | 2037 | for_each_possible_cpu(i) |
1da177e4 | 2038 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2039 | |
1da177e4 LT |
2040 | return sum; |
2041 | } | |
483b4ee6 | 2042 | |
1da177e4 LT |
2043 | unsigned long nr_iowait(void) |
2044 | { | |
2045 | unsigned long i, sum = 0; | |
483b4ee6 | 2046 | |
0a945022 | 2047 | for_each_possible_cpu(i) |
1da177e4 | 2048 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2049 | |
1da177e4 LT |
2050 | return sum; |
2051 | } | |
483b4ee6 | 2052 | |
8c215bd3 | 2053 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2054 | { |
8c215bd3 | 2055 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2056 | return atomic_read(&this->nr_iowait); |
2057 | } | |
46cb4b7c | 2058 | |
69d25870 AV |
2059 | unsigned long this_cpu_load(void) |
2060 | { | |
2061 | struct rq *this = this_rq(); | |
2062 | return this->cpu_load[0]; | |
2063 | } | |
e790fb0b | 2064 | |
46cb4b7c | 2065 | |
5167e8d5 PZ |
2066 | /* |
2067 | * Global load-average calculations | |
2068 | * | |
2069 | * We take a distributed and async approach to calculating the global load-avg | |
2070 | * in order to minimize overhead. | |
2071 | * | |
2072 | * The global load average is an exponentially decaying average of nr_running + | |
2073 | * nr_uninterruptible. | |
2074 | * | |
2075 | * Once every LOAD_FREQ: | |
2076 | * | |
2077 | * nr_active = 0; | |
2078 | * for_each_possible_cpu(cpu) | |
2079 | * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; | |
2080 | * | |
2081 | * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) | |
2082 | * | |
2083 | * Due to a number of reasons the above turns in the mess below: | |
2084 | * | |
2085 | * - for_each_possible_cpu() is prohibitively expensive on machines with | |
2086 | * serious number of cpus, therefore we need to take a distributed approach | |
2087 | * to calculating nr_active. | |
2088 | * | |
2089 | * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 | |
2090 | * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } | |
2091 | * | |
2092 | * So assuming nr_active := 0 when we start out -- true per definition, we | |
2093 | * can simply take per-cpu deltas and fold those into a global accumulate | |
2094 | * to obtain the same result. See calc_load_fold_active(). | |
2095 | * | |
2096 | * Furthermore, in order to avoid synchronizing all per-cpu delta folding | |
2097 | * across the machine, we assume 10 ticks is sufficient time for every | |
2098 | * cpu to have completed this task. | |
2099 | * | |
2100 | * This places an upper-bound on the IRQ-off latency of the machine. Then | |
2101 | * again, being late doesn't loose the delta, just wrecks the sample. | |
2102 | * | |
2103 | * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because | |
2104 | * this would add another cross-cpu cacheline miss and atomic operation | |
2105 | * to the wakeup path. Instead we increment on whatever cpu the task ran | |
2106 | * when it went into uninterruptible state and decrement on whatever cpu | |
2107 | * did the wakeup. This means that only the sum of nr_uninterruptible over | |
2108 | * all cpus yields the correct result. | |
2109 | * | |
2110 | * This covers the NO_HZ=n code, for extra head-aches, see the comment below. | |
2111 | */ | |
2112 | ||
dce48a84 TG |
2113 | /* Variables and functions for calc_load */ |
2114 | static atomic_long_t calc_load_tasks; | |
2115 | static unsigned long calc_load_update; | |
2116 | unsigned long avenrun[3]; | |
5167e8d5 PZ |
2117 | EXPORT_SYMBOL(avenrun); /* should be removed */ |
2118 | ||
2119 | /** | |
2120 | * get_avenrun - get the load average array | |
2121 | * @loads: pointer to dest load array | |
2122 | * @offset: offset to add | |
2123 | * @shift: shift count to shift the result left | |
2124 | * | |
2125 | * These values are estimates at best, so no need for locking. | |
2126 | */ | |
2127 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2128 | { | |
2129 | loads[0] = (avenrun[0] + offset) << shift; | |
2130 | loads[1] = (avenrun[1] + offset) << shift; | |
2131 | loads[2] = (avenrun[2] + offset) << shift; | |
2132 | } | |
46cb4b7c | 2133 | |
74f5187a PZ |
2134 | static long calc_load_fold_active(struct rq *this_rq) |
2135 | { | |
2136 | long nr_active, delta = 0; | |
2137 | ||
2138 | nr_active = this_rq->nr_running; | |
2139 | nr_active += (long) this_rq->nr_uninterruptible; | |
2140 | ||
2141 | if (nr_active != this_rq->calc_load_active) { | |
2142 | delta = nr_active - this_rq->calc_load_active; | |
2143 | this_rq->calc_load_active = nr_active; | |
2144 | } | |
2145 | ||
2146 | return delta; | |
2147 | } | |
2148 | ||
5167e8d5 PZ |
2149 | /* |
2150 | * a1 = a0 * e + a * (1 - e) | |
2151 | */ | |
0f004f5a PZ |
2152 | static unsigned long |
2153 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2154 | { | |
2155 | load *= exp; | |
2156 | load += active * (FIXED_1 - exp); | |
2157 | load += 1UL << (FSHIFT - 1); | |
2158 | return load >> FSHIFT; | |
2159 | } | |
2160 | ||
3451d024 | 2161 | #ifdef CONFIG_NO_HZ_COMMON |
74f5187a | 2162 | /* |
5167e8d5 PZ |
2163 | * Handle NO_HZ for the global load-average. |
2164 | * | |
2165 | * Since the above described distributed algorithm to compute the global | |
2166 | * load-average relies on per-cpu sampling from the tick, it is affected by | |
2167 | * NO_HZ. | |
2168 | * | |
2169 | * The basic idea is to fold the nr_active delta into a global idle-delta upon | |
2170 | * entering NO_HZ state such that we can include this as an 'extra' cpu delta | |
2171 | * when we read the global state. | |
2172 | * | |
2173 | * Obviously reality has to ruin such a delightfully simple scheme: | |
2174 | * | |
2175 | * - When we go NO_HZ idle during the window, we can negate our sample | |
2176 | * contribution, causing under-accounting. | |
2177 | * | |
2178 | * We avoid this by keeping two idle-delta counters and flipping them | |
2179 | * when the window starts, thus separating old and new NO_HZ load. | |
2180 | * | |
2181 | * The only trick is the slight shift in index flip for read vs write. | |
2182 | * | |
2183 | * 0s 5s 10s 15s | |
2184 | * +10 +10 +10 +10 | |
2185 | * |-|-----------|-|-----------|-|-----------|-| | |
2186 | * r:0 0 1 1 0 0 1 1 0 | |
2187 | * w:0 1 1 0 0 1 1 0 0 | |
2188 | * | |
2189 | * This ensures we'll fold the old idle contribution in this window while | |
2190 | * accumlating the new one. | |
2191 | * | |
2192 | * - When we wake up from NO_HZ idle during the window, we push up our | |
2193 | * contribution, since we effectively move our sample point to a known | |
2194 | * busy state. | |
2195 | * | |
2196 | * This is solved by pushing the window forward, and thus skipping the | |
2197 | * sample, for this cpu (effectively using the idle-delta for this cpu which | |
2198 | * was in effect at the time the window opened). This also solves the issue | |
2199 | * of having to deal with a cpu having been in NOHZ idle for multiple | |
2200 | * LOAD_FREQ intervals. | |
74f5187a PZ |
2201 | * |
2202 | * When making the ILB scale, we should try to pull this in as well. | |
2203 | */ | |
5167e8d5 PZ |
2204 | static atomic_long_t calc_load_idle[2]; |
2205 | static int calc_load_idx; | |
74f5187a | 2206 | |
5167e8d5 | 2207 | static inline int calc_load_write_idx(void) |
74f5187a | 2208 | { |
5167e8d5 PZ |
2209 | int idx = calc_load_idx; |
2210 | ||
2211 | /* | |
2212 | * See calc_global_nohz(), if we observe the new index, we also | |
2213 | * need to observe the new update time. | |
2214 | */ | |
2215 | smp_rmb(); | |
2216 | ||
2217 | /* | |
2218 | * If the folding window started, make sure we start writing in the | |
2219 | * next idle-delta. | |
2220 | */ | |
2221 | if (!time_before(jiffies, calc_load_update)) | |
2222 | idx++; | |
2223 | ||
2224 | return idx & 1; | |
2225 | } | |
2226 | ||
2227 | static inline int calc_load_read_idx(void) | |
2228 | { | |
2229 | return calc_load_idx & 1; | |
2230 | } | |
2231 | ||
2232 | void calc_load_enter_idle(void) | |
2233 | { | |
2234 | struct rq *this_rq = this_rq(); | |
74f5187a PZ |
2235 | long delta; |
2236 | ||
5167e8d5 PZ |
2237 | /* |
2238 | * We're going into NOHZ mode, if there's any pending delta, fold it | |
2239 | * into the pending idle delta. | |
2240 | */ | |
74f5187a | 2241 | delta = calc_load_fold_active(this_rq); |
5167e8d5 PZ |
2242 | if (delta) { |
2243 | int idx = calc_load_write_idx(); | |
2244 | atomic_long_add(delta, &calc_load_idle[idx]); | |
2245 | } | |
74f5187a PZ |
2246 | } |
2247 | ||
5167e8d5 | 2248 | void calc_load_exit_idle(void) |
74f5187a | 2249 | { |
5167e8d5 PZ |
2250 | struct rq *this_rq = this_rq(); |
2251 | ||
2252 | /* | |
2253 | * If we're still before the sample window, we're done. | |
2254 | */ | |
2255 | if (time_before(jiffies, this_rq->calc_load_update)) | |
2256 | return; | |
74f5187a PZ |
2257 | |
2258 | /* | |
5167e8d5 PZ |
2259 | * We woke inside or after the sample window, this means we're already |
2260 | * accounted through the nohz accounting, so skip the entire deal and | |
2261 | * sync up for the next window. | |
74f5187a | 2262 | */ |
5167e8d5 PZ |
2263 | this_rq->calc_load_update = calc_load_update; |
2264 | if (time_before(jiffies, this_rq->calc_load_update + 10)) | |
2265 | this_rq->calc_load_update += LOAD_FREQ; | |
2266 | } | |
2267 | ||
2268 | static long calc_load_fold_idle(void) | |
2269 | { | |
2270 | int idx = calc_load_read_idx(); | |
2271 | long delta = 0; | |
2272 | ||
2273 | if (atomic_long_read(&calc_load_idle[idx])) | |
2274 | delta = atomic_long_xchg(&calc_load_idle[idx], 0); | |
74f5187a PZ |
2275 | |
2276 | return delta; | |
2277 | } | |
0f004f5a PZ |
2278 | |
2279 | /** | |
2280 | * fixed_power_int - compute: x^n, in O(log n) time | |
2281 | * | |
2282 | * @x: base of the power | |
2283 | * @frac_bits: fractional bits of @x | |
2284 | * @n: power to raise @x to. | |
2285 | * | |
2286 | * By exploiting the relation between the definition of the natural power | |
2287 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2288 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2289 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2290 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2291 | * of course trivially computable in O(log_2 n), the length of our binary | |
2292 | * vector. | |
2293 | */ | |
2294 | static unsigned long | |
2295 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2296 | { | |
2297 | unsigned long result = 1UL << frac_bits; | |
2298 | ||
2299 | if (n) for (;;) { | |
2300 | if (n & 1) { | |
2301 | result *= x; | |
2302 | result += 1UL << (frac_bits - 1); | |
2303 | result >>= frac_bits; | |
2304 | } | |
2305 | n >>= 1; | |
2306 | if (!n) | |
2307 | break; | |
2308 | x *= x; | |
2309 | x += 1UL << (frac_bits - 1); | |
2310 | x >>= frac_bits; | |
2311 | } | |
2312 | ||
2313 | return result; | |
2314 | } | |
2315 | ||
2316 | /* | |
2317 | * a1 = a0 * e + a * (1 - e) | |
2318 | * | |
2319 | * a2 = a1 * e + a * (1 - e) | |
2320 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2321 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2322 | * | |
2323 | * a3 = a2 * e + a * (1 - e) | |
2324 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2325 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2326 | * | |
2327 | * ... | |
2328 | * | |
2329 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2330 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2331 | * = a0 * e^n + a * (1 - e^n) | |
2332 | * | |
2333 | * [1] application of the geometric series: | |
2334 | * | |
2335 | * n 1 - x^(n+1) | |
2336 | * S_n := \Sum x^i = ------------- | |
2337 | * i=0 1 - x | |
2338 | */ | |
2339 | static unsigned long | |
2340 | calc_load_n(unsigned long load, unsigned long exp, | |
2341 | unsigned long active, unsigned int n) | |
2342 | { | |
2343 | ||
2344 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2345 | } | |
2346 | ||
2347 | /* | |
2348 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2349 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2350 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2351 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2352 | * | |
2353 | * Once we've updated the global active value, we need to apply the exponential | |
2354 | * weights adjusted to the number of cycles missed. | |
2355 | */ | |
c308b56b | 2356 | static void calc_global_nohz(void) |
0f004f5a PZ |
2357 | { |
2358 | long delta, active, n; | |
2359 | ||
5167e8d5 PZ |
2360 | if (!time_before(jiffies, calc_load_update + 10)) { |
2361 | /* | |
2362 | * Catch-up, fold however many we are behind still | |
2363 | */ | |
2364 | delta = jiffies - calc_load_update - 10; | |
2365 | n = 1 + (delta / LOAD_FREQ); | |
0f004f5a | 2366 | |
5167e8d5 PZ |
2367 | active = atomic_long_read(&calc_load_tasks); |
2368 | active = active > 0 ? active * FIXED_1 : 0; | |
0f004f5a | 2369 | |
5167e8d5 PZ |
2370 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
2371 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2372 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
0f004f5a | 2373 | |
5167e8d5 PZ |
2374 | calc_load_update += n * LOAD_FREQ; |
2375 | } | |
74f5187a | 2376 | |
5167e8d5 PZ |
2377 | /* |
2378 | * Flip the idle index... | |
2379 | * | |
2380 | * Make sure we first write the new time then flip the index, so that | |
2381 | * calc_load_write_idx() will see the new time when it reads the new | |
2382 | * index, this avoids a double flip messing things up. | |
2383 | */ | |
2384 | smp_wmb(); | |
2385 | calc_load_idx++; | |
74f5187a | 2386 | } |
3451d024 | 2387 | #else /* !CONFIG_NO_HZ_COMMON */ |
0f004f5a | 2388 | |
5167e8d5 PZ |
2389 | static inline long calc_load_fold_idle(void) { return 0; } |
2390 | static inline void calc_global_nohz(void) { } | |
74f5187a | 2391 | |
3451d024 | 2392 | #endif /* CONFIG_NO_HZ_COMMON */ |
46cb4b7c | 2393 | |
46cb4b7c | 2394 | /* |
dce48a84 TG |
2395 | * calc_load - update the avenrun load estimates 10 ticks after the |
2396 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2397 | */ |
0f004f5a | 2398 | void calc_global_load(unsigned long ticks) |
7835b98b | 2399 | { |
5167e8d5 | 2400 | long active, delta; |
1da177e4 | 2401 | |
0f004f5a | 2402 | if (time_before(jiffies, calc_load_update + 10)) |
dce48a84 | 2403 | return; |
1da177e4 | 2404 | |
5167e8d5 PZ |
2405 | /* |
2406 | * Fold the 'old' idle-delta to include all NO_HZ cpus. | |
2407 | */ | |
2408 | delta = calc_load_fold_idle(); | |
2409 | if (delta) | |
2410 | atomic_long_add(delta, &calc_load_tasks); | |
2411 | ||
dce48a84 TG |
2412 | active = atomic_long_read(&calc_load_tasks); |
2413 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2414 | |
dce48a84 TG |
2415 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2416 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2417 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2418 | |
dce48a84 | 2419 | calc_load_update += LOAD_FREQ; |
c308b56b PZ |
2420 | |
2421 | /* | |
5167e8d5 | 2422 | * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. |
c308b56b PZ |
2423 | */ |
2424 | calc_global_nohz(); | |
dce48a84 | 2425 | } |
1da177e4 | 2426 | |
dce48a84 | 2427 | /* |
74f5187a PZ |
2428 | * Called from update_cpu_load() to periodically update this CPU's |
2429 | * active count. | |
dce48a84 TG |
2430 | */ |
2431 | static void calc_load_account_active(struct rq *this_rq) | |
2432 | { | |
74f5187a | 2433 | long delta; |
08c183f3 | 2434 | |
74f5187a PZ |
2435 | if (time_before(jiffies, this_rq->calc_load_update)) |
2436 | return; | |
783609c6 | 2437 | |
74f5187a | 2438 | delta = calc_load_fold_active(this_rq); |
74f5187a | 2439 | if (delta) |
dce48a84 | 2440 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2441 | |
2442 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2443 | } |
2444 | ||
5167e8d5 PZ |
2445 | /* |
2446 | * End of global load-average stuff | |
2447 | */ | |
2448 | ||
fdf3e95d VP |
2449 | /* |
2450 | * The exact cpuload at various idx values, calculated at every tick would be | |
2451 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2452 | * | |
2453 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2454 | * on nth tick when cpu may be busy, then we have: | |
2455 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2456 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2457 | * | |
2458 | * decay_load_missed() below does efficient calculation of | |
2459 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2460 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2461 | * | |
2462 | * The calculation is approximated on a 128 point scale. | |
2463 | * degrade_zero_ticks is the number of ticks after which load at any | |
2464 | * particular idx is approximated to be zero. | |
2465 | * degrade_factor is a precomputed table, a row for each load idx. | |
2466 | * Each column corresponds to degradation factor for a power of two ticks, | |
2467 | * based on 128 point scale. | |
2468 | * Example: | |
2469 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2470 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2471 | * | |
2472 | * With this power of 2 load factors, we can degrade the load n times | |
2473 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2474 | * n mult/shifts needed by the exact degradation. | |
2475 | */ | |
2476 | #define DEGRADE_SHIFT 7 | |
2477 | static const unsigned char | |
2478 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2479 | static const unsigned char | |
2480 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2481 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2482 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2483 | {96, 72, 40, 12, 1, 0, 0}, | |
2484 | {112, 98, 75, 43, 15, 1, 0}, | |
2485 | {120, 112, 98, 76, 45, 16, 2} }; | |
2486 | ||
2487 | /* | |
2488 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2489 | * would be when CPU is idle and so we just decay the old load without | |
2490 | * adding any new load. | |
2491 | */ | |
2492 | static unsigned long | |
2493 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2494 | { | |
2495 | int j = 0; | |
2496 | ||
2497 | if (!missed_updates) | |
2498 | return load; | |
2499 | ||
2500 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2501 | return 0; | |
2502 | ||
2503 | if (idx == 1) | |
2504 | return load >> missed_updates; | |
2505 | ||
2506 | while (missed_updates) { | |
2507 | if (missed_updates % 2) | |
2508 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2509 | ||
2510 | missed_updates >>= 1; | |
2511 | j++; | |
2512 | } | |
2513 | return load; | |
2514 | } | |
2515 | ||
46cb4b7c | 2516 | /* |
dd41f596 | 2517 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2518 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2519 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2520 | */ |
556061b0 PZ |
2521 | static void __update_cpu_load(struct rq *this_rq, unsigned long this_load, |
2522 | unsigned long pending_updates) | |
46cb4b7c | 2523 | { |
dd41f596 | 2524 | int i, scale; |
46cb4b7c | 2525 | |
dd41f596 | 2526 | this_rq->nr_load_updates++; |
46cb4b7c | 2527 | |
dd41f596 | 2528 | /* Update our load: */ |
fdf3e95d VP |
2529 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2530 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2531 | unsigned long old_load, new_load; |
7d1e6a9b | 2532 | |
dd41f596 | 2533 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2534 | |
dd41f596 | 2535 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2536 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2537 | new_load = this_load; |
a25707f3 IM |
2538 | /* |
2539 | * Round up the averaging division if load is increasing. This | |
2540 | * prevents us from getting stuck on 9 if the load is 10, for | |
2541 | * example. | |
2542 | */ | |
2543 | if (new_load > old_load) | |
fdf3e95d VP |
2544 | new_load += scale - 1; |
2545 | ||
2546 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2547 | } |
da2b71ed SS |
2548 | |
2549 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2550 | } |
2551 | ||
3451d024 | 2552 | #ifdef CONFIG_NO_HZ_COMMON |
5aaa0b7a PZ |
2553 | /* |
2554 | * There is no sane way to deal with nohz on smp when using jiffies because the | |
2555 | * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading | |
2556 | * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}. | |
2557 | * | |
2558 | * Therefore we cannot use the delta approach from the regular tick since that | |
2559 | * would seriously skew the load calculation. However we'll make do for those | |
2560 | * updates happening while idle (nohz_idle_balance) or coming out of idle | |
2561 | * (tick_nohz_idle_exit). | |
2562 | * | |
2563 | * This means we might still be one tick off for nohz periods. | |
2564 | */ | |
2565 | ||
556061b0 PZ |
2566 | /* |
2567 | * Called from nohz_idle_balance() to update the load ratings before doing the | |
2568 | * idle balance. | |
2569 | */ | |
2570 | void update_idle_cpu_load(struct rq *this_rq) | |
2571 | { | |
5aaa0b7a | 2572 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
556061b0 PZ |
2573 | unsigned long load = this_rq->load.weight; |
2574 | unsigned long pending_updates; | |
2575 | ||
2576 | /* | |
5aaa0b7a | 2577 | * bail if there's load or we're actually up-to-date. |
556061b0 PZ |
2578 | */ |
2579 | if (load || curr_jiffies == this_rq->last_load_update_tick) | |
2580 | return; | |
2581 | ||
2582 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2583 | this_rq->last_load_update_tick = curr_jiffies; | |
2584 | ||
2585 | __update_cpu_load(this_rq, load, pending_updates); | |
2586 | } | |
2587 | ||
5aaa0b7a PZ |
2588 | /* |
2589 | * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed. | |
2590 | */ | |
2591 | void update_cpu_load_nohz(void) | |
2592 | { | |
2593 | struct rq *this_rq = this_rq(); | |
2594 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); | |
2595 | unsigned long pending_updates; | |
2596 | ||
2597 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2598 | return; | |
2599 | ||
2600 | raw_spin_lock(&this_rq->lock); | |
2601 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2602 | if (pending_updates) { | |
2603 | this_rq->last_load_update_tick = curr_jiffies; | |
2604 | /* | |
2605 | * We were idle, this means load 0, the current load might be | |
2606 | * !0 due to remote wakeups and the sort. | |
2607 | */ | |
2608 | __update_cpu_load(this_rq, 0, pending_updates); | |
2609 | } | |
2610 | raw_spin_unlock(&this_rq->lock); | |
2611 | } | |
3451d024 | 2612 | #endif /* CONFIG_NO_HZ_COMMON */ |
5aaa0b7a | 2613 | |
556061b0 PZ |
2614 | /* |
2615 | * Called from scheduler_tick() | |
2616 | */ | |
fdf3e95d VP |
2617 | static void update_cpu_load_active(struct rq *this_rq) |
2618 | { | |
556061b0 | 2619 | /* |
5aaa0b7a | 2620 | * See the mess around update_idle_cpu_load() / update_cpu_load_nohz(). |
556061b0 PZ |
2621 | */ |
2622 | this_rq->last_load_update_tick = jiffies; | |
2623 | __update_cpu_load(this_rq, this_rq->load.weight, 1); | |
46cb4b7c | 2624 | |
74f5187a | 2625 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2626 | } |
2627 | ||
dd41f596 | 2628 | #ifdef CONFIG_SMP |
8a0be9ef | 2629 | |
46cb4b7c | 2630 | /* |
38022906 PZ |
2631 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2632 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2633 | */ |
38022906 | 2634 | void sched_exec(void) |
46cb4b7c | 2635 | { |
38022906 | 2636 | struct task_struct *p = current; |
1da177e4 | 2637 | unsigned long flags; |
0017d735 | 2638 | int dest_cpu; |
46cb4b7c | 2639 | |
8f42ced9 | 2640 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2641 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2642 | if (dest_cpu == smp_processor_id()) |
2643 | goto unlock; | |
38022906 | 2644 | |
8f42ced9 | 2645 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2646 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2647 | |
8f42ced9 PZ |
2648 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2649 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2650 | return; |
2651 | } | |
0017d735 | 2652 | unlock: |
8f42ced9 | 2653 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2654 | } |
dd41f596 | 2655 | |
1da177e4 LT |
2656 | #endif |
2657 | ||
1da177e4 | 2658 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2659 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2660 | |
2661 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2662 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2663 | |
2664 | /* | |
c5f8d995 | 2665 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2666 | * @p in case that task is currently running. |
c5f8d995 HS |
2667 | * |
2668 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2669 | */ |
c5f8d995 HS |
2670 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2671 | { | |
2672 | u64 ns = 0; | |
2673 | ||
2674 | if (task_current(rq, p)) { | |
2675 | update_rq_clock(rq); | |
305e6835 | 2676 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2677 | if ((s64)ns < 0) |
2678 | ns = 0; | |
2679 | } | |
2680 | ||
2681 | return ns; | |
2682 | } | |
2683 | ||
bb34d92f | 2684 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2685 | { |
1da177e4 | 2686 | unsigned long flags; |
41b86e9c | 2687 | struct rq *rq; |
bb34d92f | 2688 | u64 ns = 0; |
48f24c4d | 2689 | |
41b86e9c | 2690 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2691 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2692 | task_rq_unlock(rq, p, &flags); |
1508487e | 2693 | |
c5f8d995 HS |
2694 | return ns; |
2695 | } | |
f06febc9 | 2696 | |
c5f8d995 HS |
2697 | /* |
2698 | * Return accounted runtime for the task. | |
2699 | * In case the task is currently running, return the runtime plus current's | |
2700 | * pending runtime that have not been accounted yet. | |
2701 | */ | |
2702 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2703 | { | |
2704 | unsigned long flags; | |
2705 | struct rq *rq; | |
2706 | u64 ns = 0; | |
2707 | ||
2708 | rq = task_rq_lock(p, &flags); | |
2709 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2710 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2711 | |
2712 | return ns; | |
2713 | } | |
48f24c4d | 2714 | |
7835b98b CL |
2715 | /* |
2716 | * This function gets called by the timer code, with HZ frequency. | |
2717 | * We call it with interrupts disabled. | |
7835b98b CL |
2718 | */ |
2719 | void scheduler_tick(void) | |
2720 | { | |
7835b98b CL |
2721 | int cpu = smp_processor_id(); |
2722 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2723 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2724 | |
2725 | sched_clock_tick(); | |
dd41f596 | 2726 | |
05fa785c | 2727 | raw_spin_lock(&rq->lock); |
3e51f33f | 2728 | update_rq_clock(rq); |
fdf3e95d | 2729 | update_cpu_load_active(rq); |
fa85ae24 | 2730 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2731 | raw_spin_unlock(&rq->lock); |
7835b98b | 2732 | |
e9d2b064 | 2733 | perf_event_task_tick(); |
e220d2dc | 2734 | |
e418e1c2 | 2735 | #ifdef CONFIG_SMP |
6eb57e0d | 2736 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2737 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2738 | #endif |
265f22a9 | 2739 | rq_last_tick_reset(rq); |
1da177e4 LT |
2740 | } |
2741 | ||
265f22a9 FW |
2742 | #ifdef CONFIG_NO_HZ_FULL |
2743 | /** | |
2744 | * scheduler_tick_max_deferment | |
2745 | * | |
2746 | * Keep at least one tick per second when a single | |
2747 | * active task is running because the scheduler doesn't | |
2748 | * yet completely support full dynticks environment. | |
2749 | * | |
2750 | * This makes sure that uptime, CFS vruntime, load | |
2751 | * balancing, etc... continue to move forward, even | |
2752 | * with a very low granularity. | |
2753 | */ | |
2754 | u64 scheduler_tick_max_deferment(void) | |
2755 | { | |
2756 | struct rq *rq = this_rq(); | |
2757 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2758 | ||
2759 | next = rq->last_sched_tick + HZ; | |
2760 | ||
2761 | if (time_before_eq(next, now)) | |
2762 | return 0; | |
2763 | ||
2764 | return jiffies_to_usecs(next - now) * NSEC_PER_USEC; | |
1da177e4 | 2765 | } |
265f22a9 | 2766 | #endif |
1da177e4 | 2767 | |
132380a0 | 2768 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2769 | { |
2770 | if (in_lock_functions(addr)) { | |
2771 | addr = CALLER_ADDR2; | |
2772 | if (in_lock_functions(addr)) | |
2773 | addr = CALLER_ADDR3; | |
2774 | } | |
2775 | return addr; | |
2776 | } | |
1da177e4 | 2777 | |
7e49fcce SR |
2778 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2779 | defined(CONFIG_PREEMPT_TRACER)) | |
2780 | ||
43627582 | 2781 | void __kprobes add_preempt_count(int val) |
1da177e4 | 2782 | { |
6cd8a4bb | 2783 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2784 | /* |
2785 | * Underflow? | |
2786 | */ | |
9a11b49a IM |
2787 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2788 | return; | |
6cd8a4bb | 2789 | #endif |
1da177e4 | 2790 | preempt_count() += val; |
6cd8a4bb | 2791 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2792 | /* |
2793 | * Spinlock count overflowing soon? | |
2794 | */ | |
33859f7f MOS |
2795 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2796 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2797 | #endif |
2798 | if (preempt_count() == val) | |
2799 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2800 | } |
2801 | EXPORT_SYMBOL(add_preempt_count); | |
2802 | ||
43627582 | 2803 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 2804 | { |
6cd8a4bb | 2805 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2806 | /* |
2807 | * Underflow? | |
2808 | */ | |
01e3eb82 | 2809 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2810 | return; |
1da177e4 LT |
2811 | /* |
2812 | * Is the spinlock portion underflowing? | |
2813 | */ | |
9a11b49a IM |
2814 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2815 | !(preempt_count() & PREEMPT_MASK))) | |
2816 | return; | |
6cd8a4bb | 2817 | #endif |
9a11b49a | 2818 | |
6cd8a4bb SR |
2819 | if (preempt_count() == val) |
2820 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2821 | preempt_count() -= val; |
2822 | } | |
2823 | EXPORT_SYMBOL(sub_preempt_count); | |
2824 | ||
2825 | #endif | |
2826 | ||
2827 | /* | |
dd41f596 | 2828 | * Print scheduling while atomic bug: |
1da177e4 | 2829 | */ |
dd41f596 | 2830 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2831 | { |
664dfa65 DJ |
2832 | if (oops_in_progress) |
2833 | return; | |
2834 | ||
3df0fc5b PZ |
2835 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2836 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2837 | |
dd41f596 | 2838 | debug_show_held_locks(prev); |
e21f5b15 | 2839 | print_modules(); |
dd41f596 IM |
2840 | if (irqs_disabled()) |
2841 | print_irqtrace_events(prev); | |
6135fc1e | 2842 | dump_stack(); |
373d4d09 | 2843 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2844 | } |
1da177e4 | 2845 | |
dd41f596 IM |
2846 | /* |
2847 | * Various schedule()-time debugging checks and statistics: | |
2848 | */ | |
2849 | static inline void schedule_debug(struct task_struct *prev) | |
2850 | { | |
1da177e4 | 2851 | /* |
41a2d6cf | 2852 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
2853 | * schedule() atomically, we ignore that path for now. |
2854 | * Otherwise, whine if we are scheduling when we should not be. | |
2855 | */ | |
3f33a7ce | 2856 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 2857 | __schedule_bug(prev); |
b3fbab05 | 2858 | rcu_sleep_check(); |
dd41f596 | 2859 | |
1da177e4 LT |
2860 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2861 | ||
2d72376b | 2862 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2863 | } |
2864 | ||
6cecd084 | 2865 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2866 | { |
61eadef6 | 2867 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2868 | update_rq_clock(rq); |
6cecd084 | 2869 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2870 | } |
2871 | ||
dd41f596 IM |
2872 | /* |
2873 | * Pick up the highest-prio task: | |
2874 | */ | |
2875 | static inline struct task_struct * | |
b67802ea | 2876 | pick_next_task(struct rq *rq) |
dd41f596 | 2877 | { |
5522d5d5 | 2878 | const struct sched_class *class; |
dd41f596 | 2879 | struct task_struct *p; |
1da177e4 LT |
2880 | |
2881 | /* | |
dd41f596 IM |
2882 | * Optimization: we know that if all tasks are in |
2883 | * the fair class we can call that function directly: | |
1da177e4 | 2884 | */ |
953bfcd1 | 2885 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2886 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2887 | if (likely(p)) |
2888 | return p; | |
1da177e4 LT |
2889 | } |
2890 | ||
34f971f6 | 2891 | for_each_class(class) { |
fb8d4724 | 2892 | p = class->pick_next_task(rq); |
dd41f596 IM |
2893 | if (p) |
2894 | return p; | |
dd41f596 | 2895 | } |
34f971f6 PZ |
2896 | |
2897 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2898 | } |
1da177e4 | 2899 | |
dd41f596 | 2900 | /* |
c259e01a | 2901 | * __schedule() is the main scheduler function. |
edde96ea PE |
2902 | * |
2903 | * The main means of driving the scheduler and thus entering this function are: | |
2904 | * | |
2905 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2906 | * | |
2907 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2908 | * paths. For example, see arch/x86/entry_64.S. | |
2909 | * | |
2910 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2911 | * interrupt handler scheduler_tick(). | |
2912 | * | |
2913 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2914 | * task to the run-queue and that's it. | |
2915 | * | |
2916 | * Now, if the new task added to the run-queue preempts the current | |
2917 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2918 | * called on the nearest possible occasion: | |
2919 | * | |
2920 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2921 | * | |
2922 | * - in syscall or exception context, at the next outmost | |
2923 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2924 | * spin_unlock()!) | |
2925 | * | |
2926 | * - in IRQ context, return from interrupt-handler to | |
2927 | * preemptible context | |
2928 | * | |
2929 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2930 | * then at the next: | |
2931 | * | |
2932 | * - cond_resched() call | |
2933 | * - explicit schedule() call | |
2934 | * - return from syscall or exception to user-space | |
2935 | * - return from interrupt-handler to user-space | |
dd41f596 | 2936 | */ |
c259e01a | 2937 | static void __sched __schedule(void) |
dd41f596 IM |
2938 | { |
2939 | struct task_struct *prev, *next; | |
67ca7bde | 2940 | unsigned long *switch_count; |
dd41f596 | 2941 | struct rq *rq; |
31656519 | 2942 | int cpu; |
dd41f596 | 2943 | |
ff743345 PZ |
2944 | need_resched: |
2945 | preempt_disable(); | |
dd41f596 IM |
2946 | cpu = smp_processor_id(); |
2947 | rq = cpu_rq(cpu); | |
25502a6c | 2948 | rcu_note_context_switch(cpu); |
dd41f596 | 2949 | prev = rq->curr; |
dd41f596 | 2950 | |
dd41f596 | 2951 | schedule_debug(prev); |
1da177e4 | 2952 | |
31656519 | 2953 | if (sched_feat(HRTICK)) |
f333fdc9 | 2954 | hrtick_clear(rq); |
8f4d37ec | 2955 | |
05fa785c | 2956 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2957 | |
246d86b5 | 2958 | switch_count = &prev->nivcsw; |
1da177e4 | 2959 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2960 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2961 | prev->state = TASK_RUNNING; |
21aa9af0 | 2962 | } else { |
2acca55e PZ |
2963 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2964 | prev->on_rq = 0; | |
2965 | ||
21aa9af0 | 2966 | /* |
2acca55e PZ |
2967 | * If a worker went to sleep, notify and ask workqueue |
2968 | * whether it wants to wake up a task to maintain | |
2969 | * concurrency. | |
21aa9af0 TH |
2970 | */ |
2971 | if (prev->flags & PF_WQ_WORKER) { | |
2972 | struct task_struct *to_wakeup; | |
2973 | ||
2974 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2975 | if (to_wakeup) | |
2976 | try_to_wake_up_local(to_wakeup); | |
2977 | } | |
21aa9af0 | 2978 | } |
dd41f596 | 2979 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2980 | } |
2981 | ||
3f029d3c | 2982 | pre_schedule(rq, prev); |
f65eda4f | 2983 | |
dd41f596 | 2984 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2985 | idle_balance(cpu, rq); |
1da177e4 | 2986 | |
df1c99d4 | 2987 | put_prev_task(rq, prev); |
b67802ea | 2988 | next = pick_next_task(rq); |
f26f9aff MG |
2989 | clear_tsk_need_resched(prev); |
2990 | rq->skip_clock_update = 0; | |
1da177e4 | 2991 | |
1da177e4 | 2992 | if (likely(prev != next)) { |
1da177e4 LT |
2993 | rq->nr_switches++; |
2994 | rq->curr = next; | |
2995 | ++*switch_count; | |
2996 | ||
dd41f596 | 2997 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2998 | /* |
246d86b5 ON |
2999 | * The context switch have flipped the stack from under us |
3000 | * and restored the local variables which were saved when | |
3001 | * this task called schedule() in the past. prev == current | |
3002 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3003 | */ |
3004 | cpu = smp_processor_id(); | |
3005 | rq = cpu_rq(cpu); | |
1da177e4 | 3006 | } else |
05fa785c | 3007 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3008 | |
3f029d3c | 3009 | post_schedule(rq); |
1da177e4 | 3010 | |
ba74c144 | 3011 | sched_preempt_enable_no_resched(); |
ff743345 | 3012 | if (need_resched()) |
1da177e4 LT |
3013 | goto need_resched; |
3014 | } | |
c259e01a | 3015 | |
9c40cef2 TG |
3016 | static inline void sched_submit_work(struct task_struct *tsk) |
3017 | { | |
3c7d5184 | 3018 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3019 | return; |
3020 | /* | |
3021 | * If we are going to sleep and we have plugged IO queued, | |
3022 | * make sure to submit it to avoid deadlocks. | |
3023 | */ | |
3024 | if (blk_needs_flush_plug(tsk)) | |
3025 | blk_schedule_flush_plug(tsk); | |
3026 | } | |
3027 | ||
6ebbe7a0 | 3028 | asmlinkage void __sched schedule(void) |
c259e01a | 3029 | { |
9c40cef2 TG |
3030 | struct task_struct *tsk = current; |
3031 | ||
3032 | sched_submit_work(tsk); | |
c259e01a TG |
3033 | __schedule(); |
3034 | } | |
1da177e4 LT |
3035 | EXPORT_SYMBOL(schedule); |
3036 | ||
91d1aa43 | 3037 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
3038 | asmlinkage void __sched schedule_user(void) |
3039 | { | |
3040 | /* | |
3041 | * If we come here after a random call to set_need_resched(), | |
3042 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3043 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3044 | * we find a better solution. | |
3045 | */ | |
91d1aa43 | 3046 | user_exit(); |
20ab65e3 | 3047 | schedule(); |
91d1aa43 | 3048 | user_enter(); |
20ab65e3 FW |
3049 | } |
3050 | #endif | |
3051 | ||
c5491ea7 TG |
3052 | /** |
3053 | * schedule_preempt_disabled - called with preemption disabled | |
3054 | * | |
3055 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3056 | */ | |
3057 | void __sched schedule_preempt_disabled(void) | |
3058 | { | |
ba74c144 | 3059 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3060 | schedule(); |
3061 | preempt_disable(); | |
3062 | } | |
3063 | ||
1da177e4 LT |
3064 | #ifdef CONFIG_PREEMPT |
3065 | /* | |
2ed6e34f | 3066 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3067 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3068 | * occur there and call schedule directly. |
3069 | */ | |
d1f74e20 | 3070 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3071 | { |
3072 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3073 | |
1da177e4 LT |
3074 | /* |
3075 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3076 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3077 | */ |
beed33a8 | 3078 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3079 | return; |
3080 | ||
3a5c359a | 3081 | do { |
d1f74e20 | 3082 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3083 | __schedule(); |
d1f74e20 | 3084 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3085 | |
3a5c359a AK |
3086 | /* |
3087 | * Check again in case we missed a preemption opportunity | |
3088 | * between schedule and now. | |
3089 | */ | |
3090 | barrier(); | |
5ed0cec0 | 3091 | } while (need_resched()); |
1da177e4 | 3092 | } |
1da177e4 LT |
3093 | EXPORT_SYMBOL(preempt_schedule); |
3094 | ||
3095 | /* | |
2ed6e34f | 3096 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3097 | * off of irq context. |
3098 | * Note, that this is called and return with irqs disabled. This will | |
3099 | * protect us against recursive calling from irq. | |
3100 | */ | |
3101 | asmlinkage void __sched preempt_schedule_irq(void) | |
3102 | { | |
3103 | struct thread_info *ti = current_thread_info(); | |
b22366cd | 3104 | enum ctx_state prev_state; |
6478d880 | 3105 | |
2ed6e34f | 3106 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3107 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3108 | ||
b22366cd FW |
3109 | prev_state = exception_enter(); |
3110 | ||
3a5c359a AK |
3111 | do { |
3112 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3113 | local_irq_enable(); |
c259e01a | 3114 | __schedule(); |
3a5c359a | 3115 | local_irq_disable(); |
3a5c359a | 3116 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3117 | |
3a5c359a AK |
3118 | /* |
3119 | * Check again in case we missed a preemption opportunity | |
3120 | * between schedule and now. | |
3121 | */ | |
3122 | barrier(); | |
5ed0cec0 | 3123 | } while (need_resched()); |
b22366cd FW |
3124 | |
3125 | exception_exit(prev_state); | |
1da177e4 LT |
3126 | } |
3127 | ||
3128 | #endif /* CONFIG_PREEMPT */ | |
3129 | ||
63859d4f | 3130 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3131 | void *key) |
1da177e4 | 3132 | { |
63859d4f | 3133 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3134 | } |
1da177e4 LT |
3135 | EXPORT_SYMBOL(default_wake_function); |
3136 | ||
3137 | /* | |
41a2d6cf IM |
3138 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3139 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3140 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3141 | * | |
3142 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3143 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3144 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3145 | */ | |
78ddb08f | 3146 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3147 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3148 | { |
2e45874c | 3149 | wait_queue_t *curr, *next; |
1da177e4 | 3150 | |
2e45874c | 3151 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3152 | unsigned flags = curr->flags; |
3153 | ||
63859d4f | 3154 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3155 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3156 | break; |
3157 | } | |
3158 | } | |
3159 | ||
3160 | /** | |
3161 | * __wake_up - wake up threads blocked on a waitqueue. | |
3162 | * @q: the waitqueue | |
3163 | * @mode: which threads | |
3164 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3165 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3166 | * |
3167 | * It may be assumed that this function implies a write memory barrier before | |
3168 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3169 | */ |
7ad5b3a5 | 3170 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3171 | int nr_exclusive, void *key) |
1da177e4 LT |
3172 | { |
3173 | unsigned long flags; | |
3174 | ||
3175 | spin_lock_irqsave(&q->lock, flags); | |
3176 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3177 | spin_unlock_irqrestore(&q->lock, flags); | |
3178 | } | |
1da177e4 LT |
3179 | EXPORT_SYMBOL(__wake_up); |
3180 | ||
3181 | /* | |
3182 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3183 | */ | |
63b20011 | 3184 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3185 | { |
63b20011 | 3186 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3187 | } |
22c43c81 | 3188 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3189 | |
4ede816a DL |
3190 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3191 | { | |
3192 | __wake_up_common(q, mode, 1, 0, key); | |
3193 | } | |
bf294b41 | 3194 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3195 | |
1da177e4 | 3196 | /** |
4ede816a | 3197 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3198 | * @q: the waitqueue |
3199 | * @mode: which threads | |
3200 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3201 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3202 | * |
3203 | * The sync wakeup differs that the waker knows that it will schedule | |
3204 | * away soon, so while the target thread will be woken up, it will not | |
3205 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3206 | * with each other. This can prevent needless bouncing between CPUs. | |
3207 | * | |
3208 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3209 | * |
3210 | * It may be assumed that this function implies a write memory barrier before | |
3211 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3212 | */ |
4ede816a DL |
3213 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3214 | int nr_exclusive, void *key) | |
1da177e4 LT |
3215 | { |
3216 | unsigned long flags; | |
7d478721 | 3217 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3218 | |
3219 | if (unlikely(!q)) | |
3220 | return; | |
3221 | ||
3222 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3223 | wake_flags = 0; |
1da177e4 LT |
3224 | |
3225 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3226 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3227 | spin_unlock_irqrestore(&q->lock, flags); |
3228 | } | |
4ede816a DL |
3229 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3230 | ||
3231 | /* | |
3232 | * __wake_up_sync - see __wake_up_sync_key() | |
3233 | */ | |
3234 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3235 | { | |
3236 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3237 | } | |
1da177e4 LT |
3238 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3239 | ||
65eb3dc6 KD |
3240 | /** |
3241 | * complete: - signals a single thread waiting on this completion | |
3242 | * @x: holds the state of this particular completion | |
3243 | * | |
3244 | * This will wake up a single thread waiting on this completion. Threads will be | |
3245 | * awakened in the same order in which they were queued. | |
3246 | * | |
3247 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3248 | * |
3249 | * It may be assumed that this function implies a write memory barrier before | |
3250 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3251 | */ |
b15136e9 | 3252 | void complete(struct completion *x) |
1da177e4 LT |
3253 | { |
3254 | unsigned long flags; | |
3255 | ||
3256 | spin_lock_irqsave(&x->wait.lock, flags); | |
3257 | x->done++; | |
d9514f6c | 3258 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3259 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3260 | } | |
3261 | EXPORT_SYMBOL(complete); | |
3262 | ||
65eb3dc6 KD |
3263 | /** |
3264 | * complete_all: - signals all threads waiting on this completion | |
3265 | * @x: holds the state of this particular completion | |
3266 | * | |
3267 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3268 | * |
3269 | * It may be assumed that this function implies a write memory barrier before | |
3270 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3271 | */ |
b15136e9 | 3272 | void complete_all(struct completion *x) |
1da177e4 LT |
3273 | { |
3274 | unsigned long flags; | |
3275 | ||
3276 | spin_lock_irqsave(&x->wait.lock, flags); | |
3277 | x->done += UINT_MAX/2; | |
d9514f6c | 3278 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3279 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3280 | } | |
3281 | EXPORT_SYMBOL(complete_all); | |
3282 | ||
8cbbe86d | 3283 | static inline long __sched |
686855f5 VD |
3284 | do_wait_for_common(struct completion *x, |
3285 | long (*action)(long), long timeout, int state) | |
1da177e4 | 3286 | { |
1da177e4 LT |
3287 | if (!x->done) { |
3288 | DECLARE_WAITQUEUE(wait, current); | |
3289 | ||
a93d2f17 | 3290 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3291 | do { |
94d3d824 | 3292 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3293 | timeout = -ERESTARTSYS; |
3294 | break; | |
8cbbe86d AK |
3295 | } |
3296 | __set_current_state(state); | |
1da177e4 | 3297 | spin_unlock_irq(&x->wait.lock); |
686855f5 | 3298 | timeout = action(timeout); |
1da177e4 | 3299 | spin_lock_irq(&x->wait.lock); |
ea71a546 | 3300 | } while (!x->done && timeout); |
1da177e4 | 3301 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3302 | if (!x->done) |
3303 | return timeout; | |
1da177e4 LT |
3304 | } |
3305 | x->done--; | |
ea71a546 | 3306 | return timeout ?: 1; |
1da177e4 | 3307 | } |
1da177e4 | 3308 | |
686855f5 VD |
3309 | static inline long __sched |
3310 | __wait_for_common(struct completion *x, | |
3311 | long (*action)(long), long timeout, int state) | |
1da177e4 | 3312 | { |
1da177e4 LT |
3313 | might_sleep(); |
3314 | ||
3315 | spin_lock_irq(&x->wait.lock); | |
686855f5 | 3316 | timeout = do_wait_for_common(x, action, timeout, state); |
1da177e4 | 3317 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3318 | return timeout; |
3319 | } | |
1da177e4 | 3320 | |
686855f5 VD |
3321 | static long __sched |
3322 | wait_for_common(struct completion *x, long timeout, int state) | |
3323 | { | |
3324 | return __wait_for_common(x, schedule_timeout, timeout, state); | |
3325 | } | |
3326 | ||
3327 | static long __sched | |
3328 | wait_for_common_io(struct completion *x, long timeout, int state) | |
3329 | { | |
3330 | return __wait_for_common(x, io_schedule_timeout, timeout, state); | |
3331 | } | |
3332 | ||
65eb3dc6 KD |
3333 | /** |
3334 | * wait_for_completion: - waits for completion of a task | |
3335 | * @x: holds the state of this particular completion | |
3336 | * | |
3337 | * This waits to be signaled for completion of a specific task. It is NOT | |
3338 | * interruptible and there is no timeout. | |
3339 | * | |
3340 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3341 | * and interrupt capability. Also see complete(). | |
3342 | */ | |
b15136e9 | 3343 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3344 | { |
3345 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3346 | } |
8cbbe86d | 3347 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3348 | |
65eb3dc6 KD |
3349 | /** |
3350 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3351 | * @x: holds the state of this particular completion | |
3352 | * @timeout: timeout value in jiffies | |
3353 | * | |
3354 | * This waits for either a completion of a specific task to be signaled or for a | |
3355 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3356 | * interruptible. | |
c6dc7f05 BF |
3357 | * |
3358 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3359 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3360 | */ |
b15136e9 | 3361 | unsigned long __sched |
8cbbe86d | 3362 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3363 | { |
8cbbe86d | 3364 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3365 | } |
8cbbe86d | 3366 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3367 | |
686855f5 VD |
3368 | /** |
3369 | * wait_for_completion_io: - waits for completion of a task | |
3370 | * @x: holds the state of this particular completion | |
3371 | * | |
3372 | * This waits to be signaled for completion of a specific task. It is NOT | |
3373 | * interruptible and there is no timeout. The caller is accounted as waiting | |
3374 | * for IO. | |
3375 | */ | |
3376 | void __sched wait_for_completion_io(struct completion *x) | |
3377 | { | |
3378 | wait_for_common_io(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
3379 | } | |
3380 | EXPORT_SYMBOL(wait_for_completion_io); | |
3381 | ||
3382 | /** | |
3383 | * wait_for_completion_io_timeout: - waits for completion of a task (w/timeout) | |
3384 | * @x: holds the state of this particular completion | |
3385 | * @timeout: timeout value in jiffies | |
3386 | * | |
3387 | * This waits for either a completion of a specific task to be signaled or for a | |
3388 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3389 | * interruptible. The caller is accounted as waiting for IO. | |
3390 | * | |
3391 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3392 | * jiffies left till timeout) if completed. | |
3393 | */ | |
3394 | unsigned long __sched | |
3395 | wait_for_completion_io_timeout(struct completion *x, unsigned long timeout) | |
3396 | { | |
3397 | return wait_for_common_io(x, timeout, TASK_UNINTERRUPTIBLE); | |
3398 | } | |
3399 | EXPORT_SYMBOL(wait_for_completion_io_timeout); | |
3400 | ||
65eb3dc6 KD |
3401 | /** |
3402 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3403 | * @x: holds the state of this particular completion | |
3404 | * | |
3405 | * This waits for completion of a specific task to be signaled. It is | |
3406 | * interruptible. | |
c6dc7f05 BF |
3407 | * |
3408 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3409 | */ |
8cbbe86d | 3410 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3411 | { |
51e97990 AK |
3412 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3413 | if (t == -ERESTARTSYS) | |
3414 | return t; | |
3415 | return 0; | |
0fec171c | 3416 | } |
8cbbe86d | 3417 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3418 | |
65eb3dc6 KD |
3419 | /** |
3420 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3421 | * @x: holds the state of this particular completion | |
3422 | * @timeout: timeout value in jiffies | |
3423 | * | |
3424 | * This waits for either a completion of a specific task to be signaled or for a | |
3425 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3426 | * |
3427 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3428 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3429 | */ |
6bf41237 | 3430 | long __sched |
8cbbe86d AK |
3431 | wait_for_completion_interruptible_timeout(struct completion *x, |
3432 | unsigned long timeout) | |
0fec171c | 3433 | { |
8cbbe86d | 3434 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3435 | } |
8cbbe86d | 3436 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3437 | |
65eb3dc6 KD |
3438 | /** |
3439 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3440 | * @x: holds the state of this particular completion | |
3441 | * | |
3442 | * This waits to be signaled for completion of a specific task. It can be | |
3443 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3444 | * |
3445 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3446 | */ |
009e577e MW |
3447 | int __sched wait_for_completion_killable(struct completion *x) |
3448 | { | |
3449 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3450 | if (t == -ERESTARTSYS) | |
3451 | return t; | |
3452 | return 0; | |
3453 | } | |
3454 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3455 | ||
0aa12fb4 SW |
3456 | /** |
3457 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3458 | * @x: holds the state of this particular completion | |
3459 | * @timeout: timeout value in jiffies | |
3460 | * | |
3461 | * This waits for either a completion of a specific task to be | |
3462 | * signaled or for a specified timeout to expire. It can be | |
3463 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3464 | * |
3465 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3466 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3467 | */ |
6bf41237 | 3468 | long __sched |
0aa12fb4 SW |
3469 | wait_for_completion_killable_timeout(struct completion *x, |
3470 | unsigned long timeout) | |
3471 | { | |
3472 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3473 | } | |
3474 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3475 | ||
be4de352 DC |
3476 | /** |
3477 | * try_wait_for_completion - try to decrement a completion without blocking | |
3478 | * @x: completion structure | |
3479 | * | |
3480 | * Returns: 0 if a decrement cannot be done without blocking | |
3481 | * 1 if a decrement succeeded. | |
3482 | * | |
3483 | * If a completion is being used as a counting completion, | |
3484 | * attempt to decrement the counter without blocking. This | |
3485 | * enables us to avoid waiting if the resource the completion | |
3486 | * is protecting is not available. | |
3487 | */ | |
3488 | bool try_wait_for_completion(struct completion *x) | |
3489 | { | |
7539a3b3 | 3490 | unsigned long flags; |
be4de352 DC |
3491 | int ret = 1; |
3492 | ||
7539a3b3 | 3493 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3494 | if (!x->done) |
3495 | ret = 0; | |
3496 | else | |
3497 | x->done--; | |
7539a3b3 | 3498 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3499 | return ret; |
3500 | } | |
3501 | EXPORT_SYMBOL(try_wait_for_completion); | |
3502 | ||
3503 | /** | |
3504 | * completion_done - Test to see if a completion has any waiters | |
3505 | * @x: completion structure | |
3506 | * | |
3507 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3508 | * 1 if there are no waiters. | |
3509 | * | |
3510 | */ | |
3511 | bool completion_done(struct completion *x) | |
3512 | { | |
7539a3b3 | 3513 | unsigned long flags; |
be4de352 DC |
3514 | int ret = 1; |
3515 | ||
7539a3b3 | 3516 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3517 | if (!x->done) |
3518 | ret = 0; | |
7539a3b3 | 3519 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3520 | return ret; |
3521 | } | |
3522 | EXPORT_SYMBOL(completion_done); | |
3523 | ||
8cbbe86d AK |
3524 | static long __sched |
3525 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3526 | { |
0fec171c IM |
3527 | unsigned long flags; |
3528 | wait_queue_t wait; | |
3529 | ||
3530 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3531 | |
8cbbe86d | 3532 | __set_current_state(state); |
1da177e4 | 3533 | |
8cbbe86d AK |
3534 | spin_lock_irqsave(&q->lock, flags); |
3535 | __add_wait_queue(q, &wait); | |
3536 | spin_unlock(&q->lock); | |
3537 | timeout = schedule_timeout(timeout); | |
3538 | spin_lock_irq(&q->lock); | |
3539 | __remove_wait_queue(q, &wait); | |
3540 | spin_unlock_irqrestore(&q->lock, flags); | |
3541 | ||
3542 | return timeout; | |
3543 | } | |
3544 | ||
3545 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3546 | { | |
3547 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3548 | } |
1da177e4 LT |
3549 | EXPORT_SYMBOL(interruptible_sleep_on); |
3550 | ||
0fec171c | 3551 | long __sched |
95cdf3b7 | 3552 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3553 | { |
8cbbe86d | 3554 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3555 | } |
1da177e4 LT |
3556 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3557 | ||
0fec171c | 3558 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3559 | { |
8cbbe86d | 3560 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3561 | } |
1da177e4 LT |
3562 | EXPORT_SYMBOL(sleep_on); |
3563 | ||
0fec171c | 3564 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3565 | { |
8cbbe86d | 3566 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3567 | } |
1da177e4 LT |
3568 | EXPORT_SYMBOL(sleep_on_timeout); |
3569 | ||
b29739f9 IM |
3570 | #ifdef CONFIG_RT_MUTEXES |
3571 | ||
3572 | /* | |
3573 | * rt_mutex_setprio - set the current priority of a task | |
3574 | * @p: task | |
3575 | * @prio: prio value (kernel-internal form) | |
3576 | * | |
3577 | * This function changes the 'effective' priority of a task. It does | |
3578 | * not touch ->normal_prio like __setscheduler(). | |
3579 | * | |
3580 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3581 | */ | |
36c8b586 | 3582 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3583 | { |
83b699ed | 3584 | int oldprio, on_rq, running; |
70b97a7f | 3585 | struct rq *rq; |
83ab0aa0 | 3586 | const struct sched_class *prev_class; |
b29739f9 IM |
3587 | |
3588 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3589 | ||
0122ec5b | 3590 | rq = __task_rq_lock(p); |
b29739f9 | 3591 | |
1c4dd99b TG |
3592 | /* |
3593 | * Idle task boosting is a nono in general. There is one | |
3594 | * exception, when PREEMPT_RT and NOHZ is active: | |
3595 | * | |
3596 | * The idle task calls get_next_timer_interrupt() and holds | |
3597 | * the timer wheel base->lock on the CPU and another CPU wants | |
3598 | * to access the timer (probably to cancel it). We can safely | |
3599 | * ignore the boosting request, as the idle CPU runs this code | |
3600 | * with interrupts disabled and will complete the lock | |
3601 | * protected section without being interrupted. So there is no | |
3602 | * real need to boost. | |
3603 | */ | |
3604 | if (unlikely(p == rq->idle)) { | |
3605 | WARN_ON(p != rq->curr); | |
3606 | WARN_ON(p->pi_blocked_on); | |
3607 | goto out_unlock; | |
3608 | } | |
3609 | ||
a8027073 | 3610 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3611 | oldprio = p->prio; |
83ab0aa0 | 3612 | prev_class = p->sched_class; |
fd2f4419 | 3613 | on_rq = p->on_rq; |
051a1d1a | 3614 | running = task_current(rq, p); |
0e1f3483 | 3615 | if (on_rq) |
69be72c1 | 3616 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3617 | if (running) |
3618 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3619 | |
3620 | if (rt_prio(prio)) | |
3621 | p->sched_class = &rt_sched_class; | |
3622 | else | |
3623 | p->sched_class = &fair_sched_class; | |
3624 | ||
b29739f9 IM |
3625 | p->prio = prio; |
3626 | ||
0e1f3483 HS |
3627 | if (running) |
3628 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3629 | if (on_rq) |
371fd7e7 | 3630 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3631 | |
da7a735e | 3632 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3633 | out_unlock: |
0122ec5b | 3634 | __task_rq_unlock(rq); |
b29739f9 | 3635 | } |
b29739f9 | 3636 | #endif |
36c8b586 | 3637 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3638 | { |
dd41f596 | 3639 | int old_prio, delta, on_rq; |
1da177e4 | 3640 | unsigned long flags; |
70b97a7f | 3641 | struct rq *rq; |
1da177e4 LT |
3642 | |
3643 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3644 | return; | |
3645 | /* | |
3646 | * We have to be careful, if called from sys_setpriority(), | |
3647 | * the task might be in the middle of scheduling on another CPU. | |
3648 | */ | |
3649 | rq = task_rq_lock(p, &flags); | |
3650 | /* | |
3651 | * The RT priorities are set via sched_setscheduler(), but we still | |
3652 | * allow the 'normal' nice value to be set - but as expected | |
3653 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3654 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3655 | */ |
e05606d3 | 3656 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3657 | p->static_prio = NICE_TO_PRIO(nice); |
3658 | goto out_unlock; | |
3659 | } | |
fd2f4419 | 3660 | on_rq = p->on_rq; |
c09595f6 | 3661 | if (on_rq) |
69be72c1 | 3662 | dequeue_task(rq, p, 0); |
1da177e4 | 3663 | |
1da177e4 | 3664 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3665 | set_load_weight(p); |
b29739f9 IM |
3666 | old_prio = p->prio; |
3667 | p->prio = effective_prio(p); | |
3668 | delta = p->prio - old_prio; | |
1da177e4 | 3669 | |
dd41f596 | 3670 | if (on_rq) { |
371fd7e7 | 3671 | enqueue_task(rq, p, 0); |
1da177e4 | 3672 | /* |
d5f9f942 AM |
3673 | * If the task increased its priority or is running and |
3674 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3675 | */ |
d5f9f942 | 3676 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3677 | resched_task(rq->curr); |
3678 | } | |
3679 | out_unlock: | |
0122ec5b | 3680 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3681 | } |
1da177e4 LT |
3682 | EXPORT_SYMBOL(set_user_nice); |
3683 | ||
e43379f1 MM |
3684 | /* |
3685 | * can_nice - check if a task can reduce its nice value | |
3686 | * @p: task | |
3687 | * @nice: nice value | |
3688 | */ | |
36c8b586 | 3689 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3690 | { |
024f4747 MM |
3691 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3692 | int nice_rlim = 20 - nice; | |
48f24c4d | 3693 | |
78d7d407 | 3694 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3695 | capable(CAP_SYS_NICE)); |
3696 | } | |
3697 | ||
1da177e4 LT |
3698 | #ifdef __ARCH_WANT_SYS_NICE |
3699 | ||
3700 | /* | |
3701 | * sys_nice - change the priority of the current process. | |
3702 | * @increment: priority increment | |
3703 | * | |
3704 | * sys_setpriority is a more generic, but much slower function that | |
3705 | * does similar things. | |
3706 | */ | |
5add95d4 | 3707 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3708 | { |
48f24c4d | 3709 | long nice, retval; |
1da177e4 LT |
3710 | |
3711 | /* | |
3712 | * Setpriority might change our priority at the same moment. | |
3713 | * We don't have to worry. Conceptually one call occurs first | |
3714 | * and we have a single winner. | |
3715 | */ | |
e43379f1 MM |
3716 | if (increment < -40) |
3717 | increment = -40; | |
1da177e4 LT |
3718 | if (increment > 40) |
3719 | increment = 40; | |
3720 | ||
2b8f836f | 3721 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3722 | if (nice < -20) |
3723 | nice = -20; | |
3724 | if (nice > 19) | |
3725 | nice = 19; | |
3726 | ||
e43379f1 MM |
3727 | if (increment < 0 && !can_nice(current, nice)) |
3728 | return -EPERM; | |
3729 | ||
1da177e4 LT |
3730 | retval = security_task_setnice(current, nice); |
3731 | if (retval) | |
3732 | return retval; | |
3733 | ||
3734 | set_user_nice(current, nice); | |
3735 | return 0; | |
3736 | } | |
3737 | ||
3738 | #endif | |
3739 | ||
3740 | /** | |
3741 | * task_prio - return the priority value of a given task. | |
3742 | * @p: the task in question. | |
3743 | * | |
3744 | * This is the priority value as seen by users in /proc. | |
3745 | * RT tasks are offset by -200. Normal tasks are centered | |
3746 | * around 0, value goes from -16 to +15. | |
3747 | */ | |
36c8b586 | 3748 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3749 | { |
3750 | return p->prio - MAX_RT_PRIO; | |
3751 | } | |
3752 | ||
3753 | /** | |
3754 | * task_nice - return the nice value of a given task. | |
3755 | * @p: the task in question. | |
3756 | */ | |
36c8b586 | 3757 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3758 | { |
3759 | return TASK_NICE(p); | |
3760 | } | |
150d8bed | 3761 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3762 | |
3763 | /** | |
3764 | * idle_cpu - is a given cpu idle currently? | |
3765 | * @cpu: the processor in question. | |
3766 | */ | |
3767 | int idle_cpu(int cpu) | |
3768 | { | |
908a3283 TG |
3769 | struct rq *rq = cpu_rq(cpu); |
3770 | ||
3771 | if (rq->curr != rq->idle) | |
3772 | return 0; | |
3773 | ||
3774 | if (rq->nr_running) | |
3775 | return 0; | |
3776 | ||
3777 | #ifdef CONFIG_SMP | |
3778 | if (!llist_empty(&rq->wake_list)) | |
3779 | return 0; | |
3780 | #endif | |
3781 | ||
3782 | return 1; | |
1da177e4 LT |
3783 | } |
3784 | ||
1da177e4 LT |
3785 | /** |
3786 | * idle_task - return the idle task for a given cpu. | |
3787 | * @cpu: the processor in question. | |
3788 | */ | |
36c8b586 | 3789 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3790 | { |
3791 | return cpu_rq(cpu)->idle; | |
3792 | } | |
3793 | ||
3794 | /** | |
3795 | * find_process_by_pid - find a process with a matching PID value. | |
3796 | * @pid: the pid in question. | |
3797 | */ | |
a9957449 | 3798 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3799 | { |
228ebcbe | 3800 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3801 | } |
3802 | ||
3803 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3804 | static void |
3805 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3806 | { |
1da177e4 LT |
3807 | p->policy = policy; |
3808 | p->rt_priority = prio; | |
b29739f9 IM |
3809 | p->normal_prio = normal_prio(p); |
3810 | /* we are holding p->pi_lock already */ | |
3811 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3812 | if (rt_prio(p->prio)) |
3813 | p->sched_class = &rt_sched_class; | |
3814 | else | |
3815 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3816 | set_load_weight(p); |
1da177e4 LT |
3817 | } |
3818 | ||
c69e8d9c DH |
3819 | /* |
3820 | * check the target process has a UID that matches the current process's | |
3821 | */ | |
3822 | static bool check_same_owner(struct task_struct *p) | |
3823 | { | |
3824 | const struct cred *cred = current_cred(), *pcred; | |
3825 | bool match; | |
3826 | ||
3827 | rcu_read_lock(); | |
3828 | pcred = __task_cred(p); | |
9c806aa0 EB |
3829 | match = (uid_eq(cred->euid, pcred->euid) || |
3830 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3831 | rcu_read_unlock(); |
3832 | return match; | |
3833 | } | |
3834 | ||
961ccddd | 3835 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3836 | const struct sched_param *param, bool user) |
1da177e4 | 3837 | { |
83b699ed | 3838 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3839 | unsigned long flags; |
83ab0aa0 | 3840 | const struct sched_class *prev_class; |
70b97a7f | 3841 | struct rq *rq; |
ca94c442 | 3842 | int reset_on_fork; |
1da177e4 | 3843 | |
66e5393a SR |
3844 | /* may grab non-irq protected spin_locks */ |
3845 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3846 | recheck: |
3847 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3848 | if (policy < 0) { |
3849 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3850 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3851 | } else { |
3852 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3853 | policy &= ~SCHED_RESET_ON_FORK; | |
3854 | ||
3855 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3856 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3857 | policy != SCHED_IDLE) | |
3858 | return -EINVAL; | |
3859 | } | |
3860 | ||
1da177e4 LT |
3861 | /* |
3862 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3863 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3864 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3865 | */ |
3866 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3867 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3868 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3869 | return -EINVAL; |
e05606d3 | 3870 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3871 | return -EINVAL; |
3872 | ||
37e4ab3f OC |
3873 | /* |
3874 | * Allow unprivileged RT tasks to decrease priority: | |
3875 | */ | |
961ccddd | 3876 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3877 | if (rt_policy(policy)) { |
a44702e8 ON |
3878 | unsigned long rlim_rtprio = |
3879 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3880 | |
3881 | /* can't set/change the rt policy */ | |
3882 | if (policy != p->policy && !rlim_rtprio) | |
3883 | return -EPERM; | |
3884 | ||
3885 | /* can't increase priority */ | |
3886 | if (param->sched_priority > p->rt_priority && | |
3887 | param->sched_priority > rlim_rtprio) | |
3888 | return -EPERM; | |
3889 | } | |
c02aa73b | 3890 | |
dd41f596 | 3891 | /* |
c02aa73b DH |
3892 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3893 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3894 | */ |
c02aa73b DH |
3895 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3896 | if (!can_nice(p, TASK_NICE(p))) | |
3897 | return -EPERM; | |
3898 | } | |
5fe1d75f | 3899 | |
37e4ab3f | 3900 | /* can't change other user's priorities */ |
c69e8d9c | 3901 | if (!check_same_owner(p)) |
37e4ab3f | 3902 | return -EPERM; |
ca94c442 LP |
3903 | |
3904 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3905 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3906 | return -EPERM; | |
37e4ab3f | 3907 | } |
1da177e4 | 3908 | |
725aad24 | 3909 | if (user) { |
b0ae1981 | 3910 | retval = security_task_setscheduler(p); |
725aad24 JF |
3911 | if (retval) |
3912 | return retval; | |
3913 | } | |
3914 | ||
b29739f9 IM |
3915 | /* |
3916 | * make sure no PI-waiters arrive (or leave) while we are | |
3917 | * changing the priority of the task: | |
0122ec5b | 3918 | * |
25985edc | 3919 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3920 | * runqueue lock must be held. |
3921 | */ | |
0122ec5b | 3922 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3923 | |
34f971f6 PZ |
3924 | /* |
3925 | * Changing the policy of the stop threads its a very bad idea | |
3926 | */ | |
3927 | if (p == rq->stop) { | |
0122ec5b | 3928 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3929 | return -EINVAL; |
3930 | } | |
3931 | ||
a51e9198 DF |
3932 | /* |
3933 | * If not changing anything there's no need to proceed further: | |
3934 | */ | |
3935 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
3936 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 3937 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3938 | return 0; |
3939 | } | |
3940 | ||
dc61b1d6 PZ |
3941 | #ifdef CONFIG_RT_GROUP_SCHED |
3942 | if (user) { | |
3943 | /* | |
3944 | * Do not allow realtime tasks into groups that have no runtime | |
3945 | * assigned. | |
3946 | */ | |
3947 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3948 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3949 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3950 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3951 | return -EPERM; |
3952 | } | |
3953 | } | |
3954 | #endif | |
3955 | ||
1da177e4 LT |
3956 | /* recheck policy now with rq lock held */ |
3957 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3958 | policy = oldpolicy = -1; | |
0122ec5b | 3959 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3960 | goto recheck; |
3961 | } | |
fd2f4419 | 3962 | on_rq = p->on_rq; |
051a1d1a | 3963 | running = task_current(rq, p); |
0e1f3483 | 3964 | if (on_rq) |
4ca9b72b | 3965 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3966 | if (running) |
3967 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3968 | |
ca94c442 LP |
3969 | p->sched_reset_on_fork = reset_on_fork; |
3970 | ||
1da177e4 | 3971 | oldprio = p->prio; |
83ab0aa0 | 3972 | prev_class = p->sched_class; |
dd41f596 | 3973 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 3974 | |
0e1f3483 HS |
3975 | if (running) |
3976 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3977 | if (on_rq) |
4ca9b72b | 3978 | enqueue_task(rq, p, 0); |
cb469845 | 3979 | |
da7a735e | 3980 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3981 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3982 | |
95e02ca9 TG |
3983 | rt_mutex_adjust_pi(p); |
3984 | ||
1da177e4 LT |
3985 | return 0; |
3986 | } | |
961ccddd RR |
3987 | |
3988 | /** | |
3989 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3990 | * @p: the task in question. | |
3991 | * @policy: new policy. | |
3992 | * @param: structure containing the new RT priority. | |
3993 | * | |
3994 | * NOTE that the task may be already dead. | |
3995 | */ | |
3996 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3997 | const struct sched_param *param) |
961ccddd RR |
3998 | { |
3999 | return __sched_setscheduler(p, policy, param, true); | |
4000 | } | |
1da177e4 LT |
4001 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4002 | ||
961ccddd RR |
4003 | /** |
4004 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4005 | * @p: the task in question. | |
4006 | * @policy: new policy. | |
4007 | * @param: structure containing the new RT priority. | |
4008 | * | |
4009 | * Just like sched_setscheduler, only don't bother checking if the | |
4010 | * current context has permission. For example, this is needed in | |
4011 | * stop_machine(): we create temporary high priority worker threads, | |
4012 | * but our caller might not have that capability. | |
4013 | */ | |
4014 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4015 | const struct sched_param *param) |
961ccddd RR |
4016 | { |
4017 | return __sched_setscheduler(p, policy, param, false); | |
4018 | } | |
4019 | ||
95cdf3b7 IM |
4020 | static int |
4021 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4022 | { |
1da177e4 LT |
4023 | struct sched_param lparam; |
4024 | struct task_struct *p; | |
36c8b586 | 4025 | int retval; |
1da177e4 LT |
4026 | |
4027 | if (!param || pid < 0) | |
4028 | return -EINVAL; | |
4029 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4030 | return -EFAULT; | |
5fe1d75f ON |
4031 | |
4032 | rcu_read_lock(); | |
4033 | retval = -ESRCH; | |
1da177e4 | 4034 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4035 | if (p != NULL) |
4036 | retval = sched_setscheduler(p, policy, &lparam); | |
4037 | rcu_read_unlock(); | |
36c8b586 | 4038 | |
1da177e4 LT |
4039 | return retval; |
4040 | } | |
4041 | ||
4042 | /** | |
4043 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4044 | * @pid: the pid in question. | |
4045 | * @policy: new policy. | |
4046 | * @param: structure containing the new RT priority. | |
4047 | */ | |
5add95d4 HC |
4048 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4049 | struct sched_param __user *, param) | |
1da177e4 | 4050 | { |
c21761f1 JB |
4051 | /* negative values for policy are not valid */ |
4052 | if (policy < 0) | |
4053 | return -EINVAL; | |
4054 | ||
1da177e4 LT |
4055 | return do_sched_setscheduler(pid, policy, param); |
4056 | } | |
4057 | ||
4058 | /** | |
4059 | * sys_sched_setparam - set/change the RT priority of a thread | |
4060 | * @pid: the pid in question. | |
4061 | * @param: structure containing the new RT priority. | |
4062 | */ | |
5add95d4 | 4063 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4064 | { |
4065 | return do_sched_setscheduler(pid, -1, param); | |
4066 | } | |
4067 | ||
4068 | /** | |
4069 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4070 | * @pid: the pid in question. | |
4071 | */ | |
5add95d4 | 4072 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4073 | { |
36c8b586 | 4074 | struct task_struct *p; |
3a5c359a | 4075 | int retval; |
1da177e4 LT |
4076 | |
4077 | if (pid < 0) | |
3a5c359a | 4078 | return -EINVAL; |
1da177e4 LT |
4079 | |
4080 | retval = -ESRCH; | |
5fe85be0 | 4081 | rcu_read_lock(); |
1da177e4 LT |
4082 | p = find_process_by_pid(pid); |
4083 | if (p) { | |
4084 | retval = security_task_getscheduler(p); | |
4085 | if (!retval) | |
ca94c442 LP |
4086 | retval = p->policy |
4087 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4088 | } |
5fe85be0 | 4089 | rcu_read_unlock(); |
1da177e4 LT |
4090 | return retval; |
4091 | } | |
4092 | ||
4093 | /** | |
ca94c442 | 4094 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4095 | * @pid: the pid in question. |
4096 | * @param: structure containing the RT priority. | |
4097 | */ | |
5add95d4 | 4098 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4099 | { |
4100 | struct sched_param lp; | |
36c8b586 | 4101 | struct task_struct *p; |
3a5c359a | 4102 | int retval; |
1da177e4 LT |
4103 | |
4104 | if (!param || pid < 0) | |
3a5c359a | 4105 | return -EINVAL; |
1da177e4 | 4106 | |
5fe85be0 | 4107 | rcu_read_lock(); |
1da177e4 LT |
4108 | p = find_process_by_pid(pid); |
4109 | retval = -ESRCH; | |
4110 | if (!p) | |
4111 | goto out_unlock; | |
4112 | ||
4113 | retval = security_task_getscheduler(p); | |
4114 | if (retval) | |
4115 | goto out_unlock; | |
4116 | ||
4117 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4118 | rcu_read_unlock(); |
1da177e4 LT |
4119 | |
4120 | /* | |
4121 | * This one might sleep, we cannot do it with a spinlock held ... | |
4122 | */ | |
4123 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4124 | ||
1da177e4 LT |
4125 | return retval; |
4126 | ||
4127 | out_unlock: | |
5fe85be0 | 4128 | rcu_read_unlock(); |
1da177e4 LT |
4129 | return retval; |
4130 | } | |
4131 | ||
96f874e2 | 4132 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4133 | { |
5a16f3d3 | 4134 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4135 | struct task_struct *p; |
4136 | int retval; | |
1da177e4 | 4137 | |
95402b38 | 4138 | get_online_cpus(); |
23f5d142 | 4139 | rcu_read_lock(); |
1da177e4 LT |
4140 | |
4141 | p = find_process_by_pid(pid); | |
4142 | if (!p) { | |
23f5d142 | 4143 | rcu_read_unlock(); |
95402b38 | 4144 | put_online_cpus(); |
1da177e4 LT |
4145 | return -ESRCH; |
4146 | } | |
4147 | ||
23f5d142 | 4148 | /* Prevent p going away */ |
1da177e4 | 4149 | get_task_struct(p); |
23f5d142 | 4150 | rcu_read_unlock(); |
1da177e4 | 4151 | |
14a40ffc TH |
4152 | if (p->flags & PF_NO_SETAFFINITY) { |
4153 | retval = -EINVAL; | |
4154 | goto out_put_task; | |
4155 | } | |
5a16f3d3 RR |
4156 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4157 | retval = -ENOMEM; | |
4158 | goto out_put_task; | |
4159 | } | |
4160 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4161 | retval = -ENOMEM; | |
4162 | goto out_free_cpus_allowed; | |
4163 | } | |
1da177e4 | 4164 | retval = -EPERM; |
4c44aaaf EB |
4165 | if (!check_same_owner(p)) { |
4166 | rcu_read_lock(); | |
4167 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4168 | rcu_read_unlock(); | |
4169 | goto out_unlock; | |
4170 | } | |
4171 | rcu_read_unlock(); | |
4172 | } | |
1da177e4 | 4173 | |
b0ae1981 | 4174 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4175 | if (retval) |
4176 | goto out_unlock; | |
4177 | ||
5a16f3d3 RR |
4178 | cpuset_cpus_allowed(p, cpus_allowed); |
4179 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4180 | again: |
5a16f3d3 | 4181 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4182 | |
8707d8b8 | 4183 | if (!retval) { |
5a16f3d3 RR |
4184 | cpuset_cpus_allowed(p, cpus_allowed); |
4185 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4186 | /* |
4187 | * We must have raced with a concurrent cpuset | |
4188 | * update. Just reset the cpus_allowed to the | |
4189 | * cpuset's cpus_allowed | |
4190 | */ | |
5a16f3d3 | 4191 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4192 | goto again; |
4193 | } | |
4194 | } | |
1da177e4 | 4195 | out_unlock: |
5a16f3d3 RR |
4196 | free_cpumask_var(new_mask); |
4197 | out_free_cpus_allowed: | |
4198 | free_cpumask_var(cpus_allowed); | |
4199 | out_put_task: | |
1da177e4 | 4200 | put_task_struct(p); |
95402b38 | 4201 | put_online_cpus(); |
1da177e4 LT |
4202 | return retval; |
4203 | } | |
4204 | ||
4205 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4206 | struct cpumask *new_mask) |
1da177e4 | 4207 | { |
96f874e2 RR |
4208 | if (len < cpumask_size()) |
4209 | cpumask_clear(new_mask); | |
4210 | else if (len > cpumask_size()) | |
4211 | len = cpumask_size(); | |
4212 | ||
1da177e4 LT |
4213 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4214 | } | |
4215 | ||
4216 | /** | |
4217 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4218 | * @pid: pid of the process | |
4219 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4220 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4221 | */ | |
5add95d4 HC |
4222 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4223 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4224 | { |
5a16f3d3 | 4225 | cpumask_var_t new_mask; |
1da177e4 LT |
4226 | int retval; |
4227 | ||
5a16f3d3 RR |
4228 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4229 | return -ENOMEM; | |
1da177e4 | 4230 | |
5a16f3d3 RR |
4231 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4232 | if (retval == 0) | |
4233 | retval = sched_setaffinity(pid, new_mask); | |
4234 | free_cpumask_var(new_mask); | |
4235 | return retval; | |
1da177e4 LT |
4236 | } |
4237 | ||
96f874e2 | 4238 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4239 | { |
36c8b586 | 4240 | struct task_struct *p; |
31605683 | 4241 | unsigned long flags; |
1da177e4 | 4242 | int retval; |
1da177e4 | 4243 | |
95402b38 | 4244 | get_online_cpus(); |
23f5d142 | 4245 | rcu_read_lock(); |
1da177e4 LT |
4246 | |
4247 | retval = -ESRCH; | |
4248 | p = find_process_by_pid(pid); | |
4249 | if (!p) | |
4250 | goto out_unlock; | |
4251 | ||
e7834f8f DQ |
4252 | retval = security_task_getscheduler(p); |
4253 | if (retval) | |
4254 | goto out_unlock; | |
4255 | ||
013fdb80 | 4256 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4257 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4258 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4259 | |
4260 | out_unlock: | |
23f5d142 | 4261 | rcu_read_unlock(); |
95402b38 | 4262 | put_online_cpus(); |
1da177e4 | 4263 | |
9531b62f | 4264 | return retval; |
1da177e4 LT |
4265 | } |
4266 | ||
4267 | /** | |
4268 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4269 | * @pid: pid of the process | |
4270 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4271 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4272 | */ | |
5add95d4 HC |
4273 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4274 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4275 | { |
4276 | int ret; | |
f17c8607 | 4277 | cpumask_var_t mask; |
1da177e4 | 4278 | |
84fba5ec | 4279 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4280 | return -EINVAL; |
4281 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4282 | return -EINVAL; |
4283 | ||
f17c8607 RR |
4284 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4285 | return -ENOMEM; | |
1da177e4 | 4286 | |
f17c8607 RR |
4287 | ret = sched_getaffinity(pid, mask); |
4288 | if (ret == 0) { | |
8bc037fb | 4289 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4290 | |
4291 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4292 | ret = -EFAULT; |
4293 | else | |
cd3d8031 | 4294 | ret = retlen; |
f17c8607 RR |
4295 | } |
4296 | free_cpumask_var(mask); | |
1da177e4 | 4297 | |
f17c8607 | 4298 | return ret; |
1da177e4 LT |
4299 | } |
4300 | ||
4301 | /** | |
4302 | * sys_sched_yield - yield the current processor to other threads. | |
4303 | * | |
dd41f596 IM |
4304 | * This function yields the current CPU to other tasks. If there are no |
4305 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4306 | */ |
5add95d4 | 4307 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4308 | { |
70b97a7f | 4309 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4310 | |
2d72376b | 4311 | schedstat_inc(rq, yld_count); |
4530d7ab | 4312 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4313 | |
4314 | /* | |
4315 | * Since we are going to call schedule() anyway, there's | |
4316 | * no need to preempt or enable interrupts: | |
4317 | */ | |
4318 | __release(rq->lock); | |
8a25d5de | 4319 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4320 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4321 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4322 | |
4323 | schedule(); | |
4324 | ||
4325 | return 0; | |
4326 | } | |
4327 | ||
d86ee480 PZ |
4328 | static inline int should_resched(void) |
4329 | { | |
4330 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4331 | } | |
4332 | ||
e7b38404 | 4333 | static void __cond_resched(void) |
1da177e4 | 4334 | { |
e7aaaa69 | 4335 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4336 | __schedule(); |
e7aaaa69 | 4337 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4338 | } |
4339 | ||
02b67cc3 | 4340 | int __sched _cond_resched(void) |
1da177e4 | 4341 | { |
d86ee480 | 4342 | if (should_resched()) { |
1da177e4 LT |
4343 | __cond_resched(); |
4344 | return 1; | |
4345 | } | |
4346 | return 0; | |
4347 | } | |
02b67cc3 | 4348 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4349 | |
4350 | /* | |
613afbf8 | 4351 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4352 | * call schedule, and on return reacquire the lock. |
4353 | * | |
41a2d6cf | 4354 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4355 | * operations here to prevent schedule() from being called twice (once via |
4356 | * spin_unlock(), once by hand). | |
4357 | */ | |
613afbf8 | 4358 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4359 | { |
d86ee480 | 4360 | int resched = should_resched(); |
6df3cecb JK |
4361 | int ret = 0; |
4362 | ||
f607c668 PZ |
4363 | lockdep_assert_held(lock); |
4364 | ||
95c354fe | 4365 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4366 | spin_unlock(lock); |
d86ee480 | 4367 | if (resched) |
95c354fe NP |
4368 | __cond_resched(); |
4369 | else | |
4370 | cpu_relax(); | |
6df3cecb | 4371 | ret = 1; |
1da177e4 | 4372 | spin_lock(lock); |
1da177e4 | 4373 | } |
6df3cecb | 4374 | return ret; |
1da177e4 | 4375 | } |
613afbf8 | 4376 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4377 | |
613afbf8 | 4378 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4379 | { |
4380 | BUG_ON(!in_softirq()); | |
4381 | ||
d86ee480 | 4382 | if (should_resched()) { |
98d82567 | 4383 | local_bh_enable(); |
1da177e4 LT |
4384 | __cond_resched(); |
4385 | local_bh_disable(); | |
4386 | return 1; | |
4387 | } | |
4388 | return 0; | |
4389 | } | |
613afbf8 | 4390 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4391 | |
1da177e4 LT |
4392 | /** |
4393 | * yield - yield the current processor to other threads. | |
4394 | * | |
8e3fabfd PZ |
4395 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4396 | * | |
4397 | * The scheduler is at all times free to pick the calling task as the most | |
4398 | * eligible task to run, if removing the yield() call from your code breaks | |
4399 | * it, its already broken. | |
4400 | * | |
4401 | * Typical broken usage is: | |
4402 | * | |
4403 | * while (!event) | |
4404 | * yield(); | |
4405 | * | |
4406 | * where one assumes that yield() will let 'the other' process run that will | |
4407 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4408 | * happen. Never use yield() as a progress guarantee!! | |
4409 | * | |
4410 | * If you want to use yield() to wait for something, use wait_event(). | |
4411 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4412 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4413 | */ |
4414 | void __sched yield(void) | |
4415 | { | |
4416 | set_current_state(TASK_RUNNING); | |
4417 | sys_sched_yield(); | |
4418 | } | |
1da177e4 LT |
4419 | EXPORT_SYMBOL(yield); |
4420 | ||
d95f4122 MG |
4421 | /** |
4422 | * yield_to - yield the current processor to another thread in | |
4423 | * your thread group, or accelerate that thread toward the | |
4424 | * processor it's on. | |
16addf95 RD |
4425 | * @p: target task |
4426 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4427 | * |
4428 | * It's the caller's job to ensure that the target task struct | |
4429 | * can't go away on us before we can do any checks. | |
4430 | * | |
7b270f60 PZ |
4431 | * Returns: |
4432 | * true (>0) if we indeed boosted the target task. | |
4433 | * false (0) if we failed to boost the target. | |
4434 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4435 | */ |
4436 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4437 | { | |
4438 | struct task_struct *curr = current; | |
4439 | struct rq *rq, *p_rq; | |
4440 | unsigned long flags; | |
c3c18640 | 4441 | int yielded = 0; |
d95f4122 MG |
4442 | |
4443 | local_irq_save(flags); | |
4444 | rq = this_rq(); | |
4445 | ||
4446 | again: | |
4447 | p_rq = task_rq(p); | |
7b270f60 PZ |
4448 | /* |
4449 | * If we're the only runnable task on the rq and target rq also | |
4450 | * has only one task, there's absolutely no point in yielding. | |
4451 | */ | |
4452 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4453 | yielded = -ESRCH; | |
4454 | goto out_irq; | |
4455 | } | |
4456 | ||
d95f4122 MG |
4457 | double_rq_lock(rq, p_rq); |
4458 | while (task_rq(p) != p_rq) { | |
4459 | double_rq_unlock(rq, p_rq); | |
4460 | goto again; | |
4461 | } | |
4462 | ||
4463 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4464 | goto out_unlock; |
d95f4122 MG |
4465 | |
4466 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4467 | goto out_unlock; |
d95f4122 MG |
4468 | |
4469 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4470 | goto out_unlock; |
d95f4122 MG |
4471 | |
4472 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4473 | if (yielded) { |
d95f4122 | 4474 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4475 | /* |
4476 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4477 | * fairness. | |
4478 | */ | |
4479 | if (preempt && rq != p_rq) | |
4480 | resched_task(p_rq->curr); | |
4481 | } | |
d95f4122 | 4482 | |
7b270f60 | 4483 | out_unlock: |
d95f4122 | 4484 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4485 | out_irq: |
d95f4122 MG |
4486 | local_irq_restore(flags); |
4487 | ||
7b270f60 | 4488 | if (yielded > 0) |
d95f4122 MG |
4489 | schedule(); |
4490 | ||
4491 | return yielded; | |
4492 | } | |
4493 | EXPORT_SYMBOL_GPL(yield_to); | |
4494 | ||
1da177e4 | 4495 | /* |
41a2d6cf | 4496 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4497 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4498 | */ |
4499 | void __sched io_schedule(void) | |
4500 | { | |
54d35f29 | 4501 | struct rq *rq = raw_rq(); |
1da177e4 | 4502 | |
0ff92245 | 4503 | delayacct_blkio_start(); |
1da177e4 | 4504 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4505 | blk_flush_plug(current); |
8f0dfc34 | 4506 | current->in_iowait = 1; |
1da177e4 | 4507 | schedule(); |
8f0dfc34 | 4508 | current->in_iowait = 0; |
1da177e4 | 4509 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4510 | delayacct_blkio_end(); |
1da177e4 | 4511 | } |
1da177e4 LT |
4512 | EXPORT_SYMBOL(io_schedule); |
4513 | ||
4514 | long __sched io_schedule_timeout(long timeout) | |
4515 | { | |
54d35f29 | 4516 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4517 | long ret; |
4518 | ||
0ff92245 | 4519 | delayacct_blkio_start(); |
1da177e4 | 4520 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4521 | blk_flush_plug(current); |
8f0dfc34 | 4522 | current->in_iowait = 1; |
1da177e4 | 4523 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4524 | current->in_iowait = 0; |
1da177e4 | 4525 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4526 | delayacct_blkio_end(); |
1da177e4 LT |
4527 | return ret; |
4528 | } | |
4529 | ||
4530 | /** | |
4531 | * sys_sched_get_priority_max - return maximum RT priority. | |
4532 | * @policy: scheduling class. | |
4533 | * | |
4534 | * this syscall returns the maximum rt_priority that can be used | |
4535 | * by a given scheduling class. | |
4536 | */ | |
5add95d4 | 4537 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4538 | { |
4539 | int ret = -EINVAL; | |
4540 | ||
4541 | switch (policy) { | |
4542 | case SCHED_FIFO: | |
4543 | case SCHED_RR: | |
4544 | ret = MAX_USER_RT_PRIO-1; | |
4545 | break; | |
4546 | case SCHED_NORMAL: | |
b0a9499c | 4547 | case SCHED_BATCH: |
dd41f596 | 4548 | case SCHED_IDLE: |
1da177e4 LT |
4549 | ret = 0; |
4550 | break; | |
4551 | } | |
4552 | return ret; | |
4553 | } | |
4554 | ||
4555 | /** | |
4556 | * sys_sched_get_priority_min - return minimum RT priority. | |
4557 | * @policy: scheduling class. | |
4558 | * | |
4559 | * this syscall returns the minimum rt_priority that can be used | |
4560 | * by a given scheduling class. | |
4561 | */ | |
5add95d4 | 4562 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4563 | { |
4564 | int ret = -EINVAL; | |
4565 | ||
4566 | switch (policy) { | |
4567 | case SCHED_FIFO: | |
4568 | case SCHED_RR: | |
4569 | ret = 1; | |
4570 | break; | |
4571 | case SCHED_NORMAL: | |
b0a9499c | 4572 | case SCHED_BATCH: |
dd41f596 | 4573 | case SCHED_IDLE: |
1da177e4 LT |
4574 | ret = 0; |
4575 | } | |
4576 | return ret; | |
4577 | } | |
4578 | ||
4579 | /** | |
4580 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4581 | * @pid: pid of the process. | |
4582 | * @interval: userspace pointer to the timeslice value. | |
4583 | * | |
4584 | * this syscall writes the default timeslice value of a given process | |
4585 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4586 | */ | |
17da2bd9 | 4587 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4588 | struct timespec __user *, interval) |
1da177e4 | 4589 | { |
36c8b586 | 4590 | struct task_struct *p; |
a4ec24b4 | 4591 | unsigned int time_slice; |
dba091b9 TG |
4592 | unsigned long flags; |
4593 | struct rq *rq; | |
3a5c359a | 4594 | int retval; |
1da177e4 | 4595 | struct timespec t; |
1da177e4 LT |
4596 | |
4597 | if (pid < 0) | |
3a5c359a | 4598 | return -EINVAL; |
1da177e4 LT |
4599 | |
4600 | retval = -ESRCH; | |
1a551ae7 | 4601 | rcu_read_lock(); |
1da177e4 LT |
4602 | p = find_process_by_pid(pid); |
4603 | if (!p) | |
4604 | goto out_unlock; | |
4605 | ||
4606 | retval = security_task_getscheduler(p); | |
4607 | if (retval) | |
4608 | goto out_unlock; | |
4609 | ||
dba091b9 TG |
4610 | rq = task_rq_lock(p, &flags); |
4611 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4612 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4613 | |
1a551ae7 | 4614 | rcu_read_unlock(); |
a4ec24b4 | 4615 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4616 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4617 | return retval; |
3a5c359a | 4618 | |
1da177e4 | 4619 | out_unlock: |
1a551ae7 | 4620 | rcu_read_unlock(); |
1da177e4 LT |
4621 | return retval; |
4622 | } | |
4623 | ||
7c731e0a | 4624 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4625 | |
82a1fcb9 | 4626 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4627 | { |
1da177e4 | 4628 | unsigned long free = 0; |
4e79752c | 4629 | int ppid; |
36c8b586 | 4630 | unsigned state; |
1da177e4 | 4631 | |
1da177e4 | 4632 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4633 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4634 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4635 | #if BITS_PER_LONG == 32 |
1da177e4 | 4636 | if (state == TASK_RUNNING) |
3df0fc5b | 4637 | printk(KERN_CONT " running "); |
1da177e4 | 4638 | else |
3df0fc5b | 4639 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4640 | #else |
4641 | if (state == TASK_RUNNING) | |
3df0fc5b | 4642 | printk(KERN_CONT " running task "); |
1da177e4 | 4643 | else |
3df0fc5b | 4644 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4645 | #endif |
4646 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4647 | free = stack_not_used(p); |
1da177e4 | 4648 | #endif |
4e79752c PM |
4649 | rcu_read_lock(); |
4650 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4651 | rcu_read_unlock(); | |
3df0fc5b | 4652 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4653 | task_pid_nr(p), ppid, |
aa47b7e0 | 4654 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4655 | |
3d1cb205 | 4656 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4657 | show_stack(p, NULL); |
1da177e4 LT |
4658 | } |
4659 | ||
e59e2ae2 | 4660 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4661 | { |
36c8b586 | 4662 | struct task_struct *g, *p; |
1da177e4 | 4663 | |
4bd77321 | 4664 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4665 | printk(KERN_INFO |
4666 | " task PC stack pid father\n"); | |
1da177e4 | 4667 | #else |
3df0fc5b PZ |
4668 | printk(KERN_INFO |
4669 | " task PC stack pid father\n"); | |
1da177e4 | 4670 | #endif |
510f5acc | 4671 | rcu_read_lock(); |
1da177e4 LT |
4672 | do_each_thread(g, p) { |
4673 | /* | |
4674 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4675 | * console might take a lot of time: |
1da177e4 LT |
4676 | */ |
4677 | touch_nmi_watchdog(); | |
39bc89fd | 4678 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4679 | sched_show_task(p); |
1da177e4 LT |
4680 | } while_each_thread(g, p); |
4681 | ||
04c9167f JF |
4682 | touch_all_softlockup_watchdogs(); |
4683 | ||
dd41f596 IM |
4684 | #ifdef CONFIG_SCHED_DEBUG |
4685 | sysrq_sched_debug_show(); | |
4686 | #endif | |
510f5acc | 4687 | rcu_read_unlock(); |
e59e2ae2 IM |
4688 | /* |
4689 | * Only show locks if all tasks are dumped: | |
4690 | */ | |
93335a21 | 4691 | if (!state_filter) |
e59e2ae2 | 4692 | debug_show_all_locks(); |
1da177e4 LT |
4693 | } |
4694 | ||
1df21055 IM |
4695 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4696 | { | |
dd41f596 | 4697 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4698 | } |
4699 | ||
f340c0d1 IM |
4700 | /** |
4701 | * init_idle - set up an idle thread for a given CPU | |
4702 | * @idle: task in question | |
4703 | * @cpu: cpu the idle task belongs to | |
4704 | * | |
4705 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4706 | * flag, to make booting more robust. | |
4707 | */ | |
5c1e1767 | 4708 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4709 | { |
70b97a7f | 4710 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4711 | unsigned long flags; |
4712 | ||
05fa785c | 4713 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4714 | |
dd41f596 | 4715 | __sched_fork(idle); |
06b83b5f | 4716 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4717 | idle->se.exec_start = sched_clock(); |
4718 | ||
1e1b6c51 | 4719 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4720 | /* |
4721 | * We're having a chicken and egg problem, even though we are | |
4722 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4723 | * lockdep check in task_group() will fail. | |
4724 | * | |
4725 | * Similar case to sched_fork(). / Alternatively we could | |
4726 | * use task_rq_lock() here and obtain the other rq->lock. | |
4727 | * | |
4728 | * Silence PROVE_RCU | |
4729 | */ | |
4730 | rcu_read_lock(); | |
dd41f596 | 4731 | __set_task_cpu(idle, cpu); |
6506cf6c | 4732 | rcu_read_unlock(); |
1da177e4 | 4733 | |
1da177e4 | 4734 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4735 | #if defined(CONFIG_SMP) |
4736 | idle->on_cpu = 1; | |
4866cde0 | 4737 | #endif |
05fa785c | 4738 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4739 | |
4740 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4741 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4742 | |
dd41f596 IM |
4743 | /* |
4744 | * The idle tasks have their own, simple scheduling class: | |
4745 | */ | |
4746 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4747 | ftrace_graph_init_idle_task(idle, cpu); |
6a61671b | 4748 | vtime_init_idle(idle); |
f1c6f1a7 CE |
4749 | #if defined(CONFIG_SMP) |
4750 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4751 | #endif | |
19978ca6 IM |
4752 | } |
4753 | ||
1da177e4 | 4754 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4755 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4756 | { | |
4757 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4758 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4759 | |
4760 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4761 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4762 | } |
4763 | ||
1da177e4 LT |
4764 | /* |
4765 | * This is how migration works: | |
4766 | * | |
969c7921 TH |
4767 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4768 | * stop_one_cpu(). | |
4769 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4770 | * off the CPU) | |
4771 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4772 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4773 | * it and puts it into the right queue. |
969c7921 TH |
4774 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4775 | * is done. | |
1da177e4 LT |
4776 | */ |
4777 | ||
4778 | /* | |
4779 | * Change a given task's CPU affinity. Migrate the thread to a | |
4780 | * proper CPU and schedule it away if the CPU it's executing on | |
4781 | * is removed from the allowed bitmask. | |
4782 | * | |
4783 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4784 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4785 | * call is not atomic; no spinlocks may be held. |
4786 | */ | |
96f874e2 | 4787 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4788 | { |
4789 | unsigned long flags; | |
70b97a7f | 4790 | struct rq *rq; |
969c7921 | 4791 | unsigned int dest_cpu; |
48f24c4d | 4792 | int ret = 0; |
1da177e4 LT |
4793 | |
4794 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4795 | |
db44fc01 YZ |
4796 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4797 | goto out; | |
4798 | ||
6ad4c188 | 4799 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4800 | ret = -EINVAL; |
4801 | goto out; | |
4802 | } | |
4803 | ||
1e1b6c51 | 4804 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4805 | |
1da177e4 | 4806 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4807 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4808 | goto out; |
4809 | ||
969c7921 | 4810 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4811 | if (p->on_rq) { |
969c7921 | 4812 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4813 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4814 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4815 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4816 | tlb_migrate_finish(p->mm); |
4817 | return 0; | |
4818 | } | |
4819 | out: | |
0122ec5b | 4820 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4821 | |
1da177e4 LT |
4822 | return ret; |
4823 | } | |
cd8ba7cd | 4824 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4825 | |
4826 | /* | |
41a2d6cf | 4827 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4828 | * this because either it can't run here any more (set_cpus_allowed() |
4829 | * away from this CPU, or CPU going down), or because we're | |
4830 | * attempting to rebalance this task on exec (sched_exec). | |
4831 | * | |
4832 | * So we race with normal scheduler movements, but that's OK, as long | |
4833 | * as the task is no longer on this CPU. | |
efc30814 KK |
4834 | * |
4835 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4836 | */ |
efc30814 | 4837 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4838 | { |
70b97a7f | 4839 | struct rq *rq_dest, *rq_src; |
e2912009 | 4840 | int ret = 0; |
1da177e4 | 4841 | |
e761b772 | 4842 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4843 | return ret; |
1da177e4 LT |
4844 | |
4845 | rq_src = cpu_rq(src_cpu); | |
4846 | rq_dest = cpu_rq(dest_cpu); | |
4847 | ||
0122ec5b | 4848 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4849 | double_rq_lock(rq_src, rq_dest); |
4850 | /* Already moved. */ | |
4851 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4852 | goto done; |
1da177e4 | 4853 | /* Affinity changed (again). */ |
fa17b507 | 4854 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4855 | goto fail; |
1da177e4 | 4856 | |
e2912009 PZ |
4857 | /* |
4858 | * If we're not on a rq, the next wake-up will ensure we're | |
4859 | * placed properly. | |
4860 | */ | |
fd2f4419 | 4861 | if (p->on_rq) { |
4ca9b72b | 4862 | dequeue_task(rq_src, p, 0); |
e2912009 | 4863 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4864 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4865 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4866 | } |
b1e38734 | 4867 | done: |
efc30814 | 4868 | ret = 1; |
b1e38734 | 4869 | fail: |
1da177e4 | 4870 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4871 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4872 | return ret; |
1da177e4 LT |
4873 | } |
4874 | ||
4875 | /* | |
969c7921 TH |
4876 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4877 | * and performs thread migration by bumping thread off CPU then | |
4878 | * 'pushing' onto another runqueue. | |
1da177e4 | 4879 | */ |
969c7921 | 4880 | static int migration_cpu_stop(void *data) |
1da177e4 | 4881 | { |
969c7921 | 4882 | struct migration_arg *arg = data; |
f7b4cddc | 4883 | |
969c7921 TH |
4884 | /* |
4885 | * The original target cpu might have gone down and we might | |
4886 | * be on another cpu but it doesn't matter. | |
4887 | */ | |
f7b4cddc | 4888 | local_irq_disable(); |
969c7921 | 4889 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4890 | local_irq_enable(); |
1da177e4 | 4891 | return 0; |
f7b4cddc ON |
4892 | } |
4893 | ||
1da177e4 | 4894 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4895 | |
054b9108 | 4896 | /* |
48c5ccae PZ |
4897 | * Ensures that the idle task is using init_mm right before its cpu goes |
4898 | * offline. | |
054b9108 | 4899 | */ |
48c5ccae | 4900 | void idle_task_exit(void) |
1da177e4 | 4901 | { |
48c5ccae | 4902 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4903 | |
48c5ccae | 4904 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4905 | |
48c5ccae PZ |
4906 | if (mm != &init_mm) |
4907 | switch_mm(mm, &init_mm, current); | |
4908 | mmdrop(mm); | |
1da177e4 LT |
4909 | } |
4910 | ||
4911 | /* | |
5d180232 PZ |
4912 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4913 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4914 | * nr_active count is stable. | |
4915 | * | |
4916 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4917 | */ |
5d180232 | 4918 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4919 | { |
5d180232 PZ |
4920 | long delta = calc_load_fold_active(rq); |
4921 | if (delta) | |
4922 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4923 | } |
4924 | ||
48f24c4d | 4925 | /* |
48c5ccae PZ |
4926 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4927 | * try_to_wake_up()->select_task_rq(). | |
4928 | * | |
4929 | * Called with rq->lock held even though we'er in stop_machine() and | |
4930 | * there's no concurrency possible, we hold the required locks anyway | |
4931 | * because of lock validation efforts. | |
1da177e4 | 4932 | */ |
48c5ccae | 4933 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4934 | { |
70b97a7f | 4935 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4936 | struct task_struct *next, *stop = rq->stop; |
4937 | int dest_cpu; | |
1da177e4 LT |
4938 | |
4939 | /* | |
48c5ccae PZ |
4940 | * Fudge the rq selection such that the below task selection loop |
4941 | * doesn't get stuck on the currently eligible stop task. | |
4942 | * | |
4943 | * We're currently inside stop_machine() and the rq is either stuck | |
4944 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4945 | * either way we should never end up calling schedule() until we're | |
4946 | * done here. | |
1da177e4 | 4947 | */ |
48c5ccae | 4948 | rq->stop = NULL; |
48f24c4d | 4949 | |
dd41f596 | 4950 | for ( ; ; ) { |
48c5ccae PZ |
4951 | /* |
4952 | * There's this thread running, bail when that's the only | |
4953 | * remaining thread. | |
4954 | */ | |
4955 | if (rq->nr_running == 1) | |
dd41f596 | 4956 | break; |
48c5ccae | 4957 | |
b67802ea | 4958 | next = pick_next_task(rq); |
48c5ccae | 4959 | BUG_ON(!next); |
79c53799 | 4960 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4961 | |
48c5ccae PZ |
4962 | /* Find suitable destination for @next, with force if needed. */ |
4963 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4964 | raw_spin_unlock(&rq->lock); | |
4965 | ||
4966 | __migrate_task(next, dead_cpu, dest_cpu); | |
4967 | ||
4968 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4969 | } |
dce48a84 | 4970 | |
48c5ccae | 4971 | rq->stop = stop; |
dce48a84 | 4972 | } |
48c5ccae | 4973 | |
1da177e4 LT |
4974 | #endif /* CONFIG_HOTPLUG_CPU */ |
4975 | ||
e692ab53 NP |
4976 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4977 | ||
4978 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4979 | { |
4980 | .procname = "sched_domain", | |
c57baf1e | 4981 | .mode = 0555, |
e0361851 | 4982 | }, |
56992309 | 4983 | {} |
e692ab53 NP |
4984 | }; |
4985 | ||
4986 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4987 | { |
4988 | .procname = "kernel", | |
c57baf1e | 4989 | .mode = 0555, |
e0361851 AD |
4990 | .child = sd_ctl_dir, |
4991 | }, | |
56992309 | 4992 | {} |
e692ab53 NP |
4993 | }; |
4994 | ||
4995 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4996 | { | |
4997 | struct ctl_table *entry = | |
5cf9f062 | 4998 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4999 | |
e692ab53 NP |
5000 | return entry; |
5001 | } | |
5002 | ||
6382bc90 MM |
5003 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5004 | { | |
cd790076 | 5005 | struct ctl_table *entry; |
6382bc90 | 5006 | |
cd790076 MM |
5007 | /* |
5008 | * In the intermediate directories, both the child directory and | |
5009 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5010 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5011 | * static strings and all have proc handlers. |
5012 | */ | |
5013 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5014 | if (entry->child) |
5015 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5016 | if (entry->proc_handler == NULL) |
5017 | kfree(entry->procname); | |
5018 | } | |
6382bc90 MM |
5019 | |
5020 | kfree(*tablep); | |
5021 | *tablep = NULL; | |
5022 | } | |
5023 | ||
201c373e | 5024 | static int min_load_idx = 0; |
fd9b86d3 | 5025 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 5026 | |
e692ab53 | 5027 | static void |
e0361851 | 5028 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5029 | const char *procname, void *data, int maxlen, |
201c373e NK |
5030 | umode_t mode, proc_handler *proc_handler, |
5031 | bool load_idx) | |
e692ab53 | 5032 | { |
e692ab53 NP |
5033 | entry->procname = procname; |
5034 | entry->data = data; | |
5035 | entry->maxlen = maxlen; | |
5036 | entry->mode = mode; | |
5037 | entry->proc_handler = proc_handler; | |
201c373e NK |
5038 | |
5039 | if (load_idx) { | |
5040 | entry->extra1 = &min_load_idx; | |
5041 | entry->extra2 = &max_load_idx; | |
5042 | } | |
e692ab53 NP |
5043 | } |
5044 | ||
5045 | static struct ctl_table * | |
5046 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5047 | { | |
a5d8c348 | 5048 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5049 | |
ad1cdc1d MM |
5050 | if (table == NULL) |
5051 | return NULL; | |
5052 | ||
e0361851 | 5053 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 5054 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5055 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 5056 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 5057 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 5058 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5059 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 5060 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5061 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 5062 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5063 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 5064 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5065 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 5066 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 5067 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5068 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5069 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5070 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5071 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5072 | &sd->cache_nice_tries, |
201c373e | 5073 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5074 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5075 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 5076 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 5077 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 5078 | /* &table[12] is terminator */ |
e692ab53 NP |
5079 | |
5080 | return table; | |
5081 | } | |
5082 | ||
9a4e7159 | 5083 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5084 | { |
5085 | struct ctl_table *entry, *table; | |
5086 | struct sched_domain *sd; | |
5087 | int domain_num = 0, i; | |
5088 | char buf[32]; | |
5089 | ||
5090 | for_each_domain(cpu, sd) | |
5091 | domain_num++; | |
5092 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5093 | if (table == NULL) |
5094 | return NULL; | |
e692ab53 NP |
5095 | |
5096 | i = 0; | |
5097 | for_each_domain(cpu, sd) { | |
5098 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5099 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5100 | entry->mode = 0555; |
e692ab53 NP |
5101 | entry->child = sd_alloc_ctl_domain_table(sd); |
5102 | entry++; | |
5103 | i++; | |
5104 | } | |
5105 | return table; | |
5106 | } | |
5107 | ||
5108 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5109 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5110 | { |
6ad4c188 | 5111 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5112 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5113 | char buf[32]; | |
5114 | ||
7378547f MM |
5115 | WARN_ON(sd_ctl_dir[0].child); |
5116 | sd_ctl_dir[0].child = entry; | |
5117 | ||
ad1cdc1d MM |
5118 | if (entry == NULL) |
5119 | return; | |
5120 | ||
6ad4c188 | 5121 | for_each_possible_cpu(i) { |
e692ab53 | 5122 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5123 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5124 | entry->mode = 0555; |
e692ab53 | 5125 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5126 | entry++; |
e692ab53 | 5127 | } |
7378547f MM |
5128 | |
5129 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5130 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5131 | } | |
6382bc90 | 5132 | |
7378547f | 5133 | /* may be called multiple times per register */ |
6382bc90 MM |
5134 | static void unregister_sched_domain_sysctl(void) |
5135 | { | |
7378547f MM |
5136 | if (sd_sysctl_header) |
5137 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5138 | sd_sysctl_header = NULL; |
7378547f MM |
5139 | if (sd_ctl_dir[0].child) |
5140 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5141 | } |
e692ab53 | 5142 | #else |
6382bc90 MM |
5143 | static void register_sched_domain_sysctl(void) |
5144 | { | |
5145 | } | |
5146 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5147 | { |
5148 | } | |
5149 | #endif | |
5150 | ||
1f11eb6a GH |
5151 | static void set_rq_online(struct rq *rq) |
5152 | { | |
5153 | if (!rq->online) { | |
5154 | const struct sched_class *class; | |
5155 | ||
c6c4927b | 5156 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5157 | rq->online = 1; |
5158 | ||
5159 | for_each_class(class) { | |
5160 | if (class->rq_online) | |
5161 | class->rq_online(rq); | |
5162 | } | |
5163 | } | |
5164 | } | |
5165 | ||
5166 | static void set_rq_offline(struct rq *rq) | |
5167 | { | |
5168 | if (rq->online) { | |
5169 | const struct sched_class *class; | |
5170 | ||
5171 | for_each_class(class) { | |
5172 | if (class->rq_offline) | |
5173 | class->rq_offline(rq); | |
5174 | } | |
5175 | ||
c6c4927b | 5176 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5177 | rq->online = 0; |
5178 | } | |
5179 | } | |
5180 | ||
1da177e4 LT |
5181 | /* |
5182 | * migration_call - callback that gets triggered when a CPU is added. | |
5183 | * Here we can start up the necessary migration thread for the new CPU. | |
5184 | */ | |
48f24c4d IM |
5185 | static int __cpuinit |
5186 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5187 | { |
48f24c4d | 5188 | int cpu = (long)hcpu; |
1da177e4 | 5189 | unsigned long flags; |
969c7921 | 5190 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5191 | |
48c5ccae | 5192 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5193 | |
1da177e4 | 5194 | case CPU_UP_PREPARE: |
a468d389 | 5195 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5196 | break; |
48f24c4d | 5197 | |
1da177e4 | 5198 | case CPU_ONLINE: |
1f94ef59 | 5199 | /* Update our root-domain */ |
05fa785c | 5200 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5201 | if (rq->rd) { |
c6c4927b | 5202 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5203 | |
5204 | set_rq_online(rq); | |
1f94ef59 | 5205 | } |
05fa785c | 5206 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5207 | break; |
48f24c4d | 5208 | |
1da177e4 | 5209 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5210 | case CPU_DYING: |
317f3941 | 5211 | sched_ttwu_pending(); |
57d885fe | 5212 | /* Update our root-domain */ |
05fa785c | 5213 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5214 | if (rq->rd) { |
c6c4927b | 5215 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5216 | set_rq_offline(rq); |
57d885fe | 5217 | } |
48c5ccae PZ |
5218 | migrate_tasks(cpu); |
5219 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5220 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5221 | break; |
48c5ccae | 5222 | |
5d180232 | 5223 | case CPU_DEAD: |
f319da0c | 5224 | calc_load_migrate(rq); |
57d885fe | 5225 | break; |
1da177e4 LT |
5226 | #endif |
5227 | } | |
49c022e6 PZ |
5228 | |
5229 | update_max_interval(); | |
5230 | ||
1da177e4 LT |
5231 | return NOTIFY_OK; |
5232 | } | |
5233 | ||
f38b0820 PM |
5234 | /* |
5235 | * Register at high priority so that task migration (migrate_all_tasks) | |
5236 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5237 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5238 | */ |
26c2143b | 5239 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5240 | .notifier_call = migration_call, |
50a323b7 | 5241 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5242 | }; |
5243 | ||
3a101d05 TH |
5244 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5245 | unsigned long action, void *hcpu) | |
5246 | { | |
5247 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5248 | case CPU_STARTING: |
3a101d05 TH |
5249 | case CPU_DOWN_FAILED: |
5250 | set_cpu_active((long)hcpu, true); | |
5251 | return NOTIFY_OK; | |
5252 | default: | |
5253 | return NOTIFY_DONE; | |
5254 | } | |
5255 | } | |
5256 | ||
5257 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5258 | unsigned long action, void *hcpu) | |
5259 | { | |
5260 | switch (action & ~CPU_TASKS_FROZEN) { | |
5261 | case CPU_DOWN_PREPARE: | |
5262 | set_cpu_active((long)hcpu, false); | |
5263 | return NOTIFY_OK; | |
5264 | default: | |
5265 | return NOTIFY_DONE; | |
5266 | } | |
5267 | } | |
5268 | ||
7babe8db | 5269 | static int __init migration_init(void) |
1da177e4 LT |
5270 | { |
5271 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5272 | int err; |
48f24c4d | 5273 | |
3a101d05 | 5274 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5275 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5276 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5277 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5278 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5279 | |
3a101d05 TH |
5280 | /* Register cpu active notifiers */ |
5281 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5282 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5283 | ||
a004cd42 | 5284 | return 0; |
1da177e4 | 5285 | } |
7babe8db | 5286 | early_initcall(migration_init); |
1da177e4 LT |
5287 | #endif |
5288 | ||
5289 | #ifdef CONFIG_SMP | |
476f3534 | 5290 | |
4cb98839 PZ |
5291 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5292 | ||
3e9830dc | 5293 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5294 | |
d039ac60 | 5295 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5296 | |
d039ac60 | 5297 | static int __init sched_debug_setup(char *str) |
f6630114 | 5298 | { |
d039ac60 | 5299 | sched_debug_enabled = 1; |
f6630114 MT |
5300 | |
5301 | return 0; | |
5302 | } | |
d039ac60 PZ |
5303 | early_param("sched_debug", sched_debug_setup); |
5304 | ||
5305 | static inline bool sched_debug(void) | |
5306 | { | |
5307 | return sched_debug_enabled; | |
5308 | } | |
f6630114 | 5309 | |
7c16ec58 | 5310 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5311 | struct cpumask *groupmask) |
1da177e4 | 5312 | { |
4dcf6aff | 5313 | struct sched_group *group = sd->groups; |
434d53b0 | 5314 | char str[256]; |
1da177e4 | 5315 | |
968ea6d8 | 5316 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5317 | cpumask_clear(groupmask); |
4dcf6aff IM |
5318 | |
5319 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5320 | ||
5321 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5322 | printk("does not load-balance\n"); |
4dcf6aff | 5323 | if (sd->parent) |
3df0fc5b PZ |
5324 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5325 | " has parent"); | |
4dcf6aff | 5326 | return -1; |
41c7ce9a NP |
5327 | } |
5328 | ||
3df0fc5b | 5329 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5330 | |
758b2cdc | 5331 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5332 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5333 | "CPU%d\n", cpu); | |
4dcf6aff | 5334 | } |
758b2cdc | 5335 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5336 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5337 | " CPU%d\n", cpu); | |
4dcf6aff | 5338 | } |
1da177e4 | 5339 | |
4dcf6aff | 5340 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5341 | do { |
4dcf6aff | 5342 | if (!group) { |
3df0fc5b PZ |
5343 | printk("\n"); |
5344 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5345 | break; |
5346 | } | |
5347 | ||
c3decf0d PZ |
5348 | /* |
5349 | * Even though we initialize ->power to something semi-sane, | |
5350 | * we leave power_orig unset. This allows us to detect if | |
5351 | * domain iteration is still funny without causing /0 traps. | |
5352 | */ | |
5353 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5354 | printk(KERN_CONT "\n"); |
5355 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5356 | "set\n"); | |
4dcf6aff IM |
5357 | break; |
5358 | } | |
1da177e4 | 5359 | |
758b2cdc | 5360 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5361 | printk(KERN_CONT "\n"); |
5362 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5363 | break; |
5364 | } | |
1da177e4 | 5365 | |
cb83b629 PZ |
5366 | if (!(sd->flags & SD_OVERLAP) && |
5367 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5368 | printk(KERN_CONT "\n"); |
5369 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5370 | break; |
5371 | } | |
1da177e4 | 5372 | |
758b2cdc | 5373 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5374 | |
968ea6d8 | 5375 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5376 | |
3df0fc5b | 5377 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5378 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5379 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5380 | group->sgp->power); |
381512cf | 5381 | } |
1da177e4 | 5382 | |
4dcf6aff IM |
5383 | group = group->next; |
5384 | } while (group != sd->groups); | |
3df0fc5b | 5385 | printk(KERN_CONT "\n"); |
1da177e4 | 5386 | |
758b2cdc | 5387 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5388 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5389 | |
758b2cdc RR |
5390 | if (sd->parent && |
5391 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5392 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5393 | "of domain->span\n"); | |
4dcf6aff IM |
5394 | return 0; |
5395 | } | |
1da177e4 | 5396 | |
4dcf6aff IM |
5397 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5398 | { | |
5399 | int level = 0; | |
1da177e4 | 5400 | |
d039ac60 | 5401 | if (!sched_debug_enabled) |
f6630114 MT |
5402 | return; |
5403 | ||
4dcf6aff IM |
5404 | if (!sd) { |
5405 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5406 | return; | |
5407 | } | |
1da177e4 | 5408 | |
4dcf6aff IM |
5409 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5410 | ||
5411 | for (;;) { | |
4cb98839 | 5412 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5413 | break; |
1da177e4 LT |
5414 | level++; |
5415 | sd = sd->parent; | |
33859f7f | 5416 | if (!sd) |
4dcf6aff IM |
5417 | break; |
5418 | } | |
1da177e4 | 5419 | } |
6d6bc0ad | 5420 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5421 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5422 | static inline bool sched_debug(void) |
5423 | { | |
5424 | return false; | |
5425 | } | |
6d6bc0ad | 5426 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5427 | |
1a20ff27 | 5428 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5429 | { |
758b2cdc | 5430 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5431 | return 1; |
5432 | ||
5433 | /* Following flags need at least 2 groups */ | |
5434 | if (sd->flags & (SD_LOAD_BALANCE | | |
5435 | SD_BALANCE_NEWIDLE | | |
5436 | SD_BALANCE_FORK | | |
89c4710e SS |
5437 | SD_BALANCE_EXEC | |
5438 | SD_SHARE_CPUPOWER | | |
5439 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5440 | if (sd->groups != sd->groups->next) |
5441 | return 0; | |
5442 | } | |
5443 | ||
5444 | /* Following flags don't use groups */ | |
c88d5910 | 5445 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5446 | return 0; |
5447 | ||
5448 | return 1; | |
5449 | } | |
5450 | ||
48f24c4d IM |
5451 | static int |
5452 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5453 | { |
5454 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5455 | ||
5456 | if (sd_degenerate(parent)) | |
5457 | return 1; | |
5458 | ||
758b2cdc | 5459 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5460 | return 0; |
5461 | ||
245af2c7 SS |
5462 | /* Flags needing groups don't count if only 1 group in parent */ |
5463 | if (parent->groups == parent->groups->next) { | |
5464 | pflags &= ~(SD_LOAD_BALANCE | | |
5465 | SD_BALANCE_NEWIDLE | | |
5466 | SD_BALANCE_FORK | | |
89c4710e SS |
5467 | SD_BALANCE_EXEC | |
5468 | SD_SHARE_CPUPOWER | | |
5469 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5470 | if (nr_node_ids == 1) |
5471 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5472 | } |
5473 | if (~cflags & pflags) | |
5474 | return 0; | |
5475 | ||
5476 | return 1; | |
5477 | } | |
5478 | ||
dce840a0 | 5479 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5480 | { |
dce840a0 | 5481 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5482 | |
68e74568 | 5483 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5484 | free_cpumask_var(rd->rto_mask); |
5485 | free_cpumask_var(rd->online); | |
5486 | free_cpumask_var(rd->span); | |
5487 | kfree(rd); | |
5488 | } | |
5489 | ||
57d885fe GH |
5490 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5491 | { | |
a0490fa3 | 5492 | struct root_domain *old_rd = NULL; |
57d885fe | 5493 | unsigned long flags; |
57d885fe | 5494 | |
05fa785c | 5495 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5496 | |
5497 | if (rq->rd) { | |
a0490fa3 | 5498 | old_rd = rq->rd; |
57d885fe | 5499 | |
c6c4927b | 5500 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5501 | set_rq_offline(rq); |
57d885fe | 5502 | |
c6c4927b | 5503 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5504 | |
a0490fa3 IM |
5505 | /* |
5506 | * If we dont want to free the old_rt yet then | |
5507 | * set old_rd to NULL to skip the freeing later | |
5508 | * in this function: | |
5509 | */ | |
5510 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5511 | old_rd = NULL; | |
57d885fe GH |
5512 | } |
5513 | ||
5514 | atomic_inc(&rd->refcount); | |
5515 | rq->rd = rd; | |
5516 | ||
c6c4927b | 5517 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5518 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5519 | set_rq_online(rq); |
57d885fe | 5520 | |
05fa785c | 5521 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5522 | |
5523 | if (old_rd) | |
dce840a0 | 5524 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5525 | } |
5526 | ||
68c38fc3 | 5527 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5528 | { |
5529 | memset(rd, 0, sizeof(*rd)); | |
5530 | ||
68c38fc3 | 5531 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5532 | goto out; |
68c38fc3 | 5533 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5534 | goto free_span; |
68c38fc3 | 5535 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5536 | goto free_online; |
6e0534f2 | 5537 | |
68c38fc3 | 5538 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5539 | goto free_rto_mask; |
c6c4927b | 5540 | return 0; |
6e0534f2 | 5541 | |
68e74568 RR |
5542 | free_rto_mask: |
5543 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5544 | free_online: |
5545 | free_cpumask_var(rd->online); | |
5546 | free_span: | |
5547 | free_cpumask_var(rd->span); | |
0c910d28 | 5548 | out: |
c6c4927b | 5549 | return -ENOMEM; |
57d885fe GH |
5550 | } |
5551 | ||
029632fb PZ |
5552 | /* |
5553 | * By default the system creates a single root-domain with all cpus as | |
5554 | * members (mimicking the global state we have today). | |
5555 | */ | |
5556 | struct root_domain def_root_domain; | |
5557 | ||
57d885fe GH |
5558 | static void init_defrootdomain(void) |
5559 | { | |
68c38fc3 | 5560 | init_rootdomain(&def_root_domain); |
c6c4927b | 5561 | |
57d885fe GH |
5562 | atomic_set(&def_root_domain.refcount, 1); |
5563 | } | |
5564 | ||
dc938520 | 5565 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5566 | { |
5567 | struct root_domain *rd; | |
5568 | ||
5569 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5570 | if (!rd) | |
5571 | return NULL; | |
5572 | ||
68c38fc3 | 5573 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5574 | kfree(rd); |
5575 | return NULL; | |
5576 | } | |
57d885fe GH |
5577 | |
5578 | return rd; | |
5579 | } | |
5580 | ||
e3589f6c PZ |
5581 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5582 | { | |
5583 | struct sched_group *tmp, *first; | |
5584 | ||
5585 | if (!sg) | |
5586 | return; | |
5587 | ||
5588 | first = sg; | |
5589 | do { | |
5590 | tmp = sg->next; | |
5591 | ||
5592 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5593 | kfree(sg->sgp); | |
5594 | ||
5595 | kfree(sg); | |
5596 | sg = tmp; | |
5597 | } while (sg != first); | |
5598 | } | |
5599 | ||
dce840a0 PZ |
5600 | static void free_sched_domain(struct rcu_head *rcu) |
5601 | { | |
5602 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5603 | |
5604 | /* | |
5605 | * If its an overlapping domain it has private groups, iterate and | |
5606 | * nuke them all. | |
5607 | */ | |
5608 | if (sd->flags & SD_OVERLAP) { | |
5609 | free_sched_groups(sd->groups, 1); | |
5610 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5611 | kfree(sd->groups->sgp); |
dce840a0 | 5612 | kfree(sd->groups); |
9c3f75cb | 5613 | } |
dce840a0 PZ |
5614 | kfree(sd); |
5615 | } | |
5616 | ||
5617 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5618 | { | |
5619 | call_rcu(&sd->rcu, free_sched_domain); | |
5620 | } | |
5621 | ||
5622 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5623 | { | |
5624 | for (; sd; sd = sd->parent) | |
5625 | destroy_sched_domain(sd, cpu); | |
5626 | } | |
5627 | ||
518cd623 PZ |
5628 | /* |
5629 | * Keep a special pointer to the highest sched_domain that has | |
5630 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5631 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5632 | * | |
5633 | * Also keep a unique ID per domain (we use the first cpu number in | |
5634 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5635 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5636 | */ |
5637 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5638 | DEFINE_PER_CPU(int, sd_llc_id); | |
5639 | ||
5640 | static void update_top_cache_domain(int cpu) | |
5641 | { | |
5642 | struct sched_domain *sd; | |
5643 | int id = cpu; | |
5644 | ||
5645 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
37407ea7 | 5646 | if (sd) |
518cd623 PZ |
5647 | id = cpumask_first(sched_domain_span(sd)); |
5648 | ||
5649 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5650 | per_cpu(sd_llc_id, cpu) = id; | |
5651 | } | |
5652 | ||
1da177e4 | 5653 | /* |
0eab9146 | 5654 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5655 | * hold the hotplug lock. |
5656 | */ | |
0eab9146 IM |
5657 | static void |
5658 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5659 | { |
70b97a7f | 5660 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5661 | struct sched_domain *tmp; |
5662 | ||
5663 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5664 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5665 | struct sched_domain *parent = tmp->parent; |
5666 | if (!parent) | |
5667 | break; | |
f29c9b1c | 5668 | |
1a848870 | 5669 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5670 | tmp->parent = parent->parent; |
1a848870 SS |
5671 | if (parent->parent) |
5672 | parent->parent->child = tmp; | |
dce840a0 | 5673 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5674 | } else |
5675 | tmp = tmp->parent; | |
245af2c7 SS |
5676 | } |
5677 | ||
1a848870 | 5678 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5679 | tmp = sd; |
245af2c7 | 5680 | sd = sd->parent; |
dce840a0 | 5681 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5682 | if (sd) |
5683 | sd->child = NULL; | |
5684 | } | |
1da177e4 | 5685 | |
4cb98839 | 5686 | sched_domain_debug(sd, cpu); |
1da177e4 | 5687 | |
57d885fe | 5688 | rq_attach_root(rq, rd); |
dce840a0 | 5689 | tmp = rq->sd; |
674311d5 | 5690 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5691 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5692 | |
5693 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5694 | } |
5695 | ||
5696 | /* cpus with isolated domains */ | |
dcc30a35 | 5697 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5698 | |
5699 | /* Setup the mask of cpus configured for isolated domains */ | |
5700 | static int __init isolated_cpu_setup(char *str) | |
5701 | { | |
bdddd296 | 5702 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5703 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5704 | return 1; |
5705 | } | |
5706 | ||
8927f494 | 5707 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5708 | |
d3081f52 PZ |
5709 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5710 | { | |
5711 | return cpumask_of_node(cpu_to_node(cpu)); | |
5712 | } | |
5713 | ||
dce840a0 PZ |
5714 | struct sd_data { |
5715 | struct sched_domain **__percpu sd; | |
5716 | struct sched_group **__percpu sg; | |
9c3f75cb | 5717 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5718 | }; |
5719 | ||
49a02c51 | 5720 | struct s_data { |
21d42ccf | 5721 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5722 | struct root_domain *rd; |
5723 | }; | |
5724 | ||
2109b99e | 5725 | enum s_alloc { |
2109b99e | 5726 | sa_rootdomain, |
21d42ccf | 5727 | sa_sd, |
dce840a0 | 5728 | sa_sd_storage, |
2109b99e AH |
5729 | sa_none, |
5730 | }; | |
5731 | ||
54ab4ff4 PZ |
5732 | struct sched_domain_topology_level; |
5733 | ||
5734 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5735 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5736 | ||
e3589f6c PZ |
5737 | #define SDTL_OVERLAP 0x01 |
5738 | ||
eb7a74e6 | 5739 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5740 | sched_domain_init_f init; |
5741 | sched_domain_mask_f mask; | |
e3589f6c | 5742 | int flags; |
cb83b629 | 5743 | int numa_level; |
54ab4ff4 | 5744 | struct sd_data data; |
eb7a74e6 PZ |
5745 | }; |
5746 | ||
c1174876 PZ |
5747 | /* |
5748 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5749 | * domain traversal. | |
5750 | * | |
5751 | * Asymmetric node setups can result in situations where the domain tree is of | |
5752 | * unequal depth, make sure to skip domains that already cover the entire | |
5753 | * range. | |
5754 | * | |
5755 | * In that case build_sched_domains() will have terminated the iteration early | |
5756 | * and our sibling sd spans will be empty. Domains should always include the | |
5757 | * cpu they're built on, so check that. | |
5758 | * | |
5759 | */ | |
5760 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5761 | { | |
5762 | const struct cpumask *span = sched_domain_span(sd); | |
5763 | struct sd_data *sdd = sd->private; | |
5764 | struct sched_domain *sibling; | |
5765 | int i; | |
5766 | ||
5767 | for_each_cpu(i, span) { | |
5768 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5769 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5770 | continue; | |
5771 | ||
5772 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5773 | } | |
5774 | } | |
5775 | ||
5776 | /* | |
5777 | * Return the canonical balance cpu for this group, this is the first cpu | |
5778 | * of this group that's also in the iteration mask. | |
5779 | */ | |
5780 | int group_balance_cpu(struct sched_group *sg) | |
5781 | { | |
5782 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5783 | } | |
5784 | ||
e3589f6c PZ |
5785 | static int |
5786 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5787 | { | |
5788 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5789 | const struct cpumask *span = sched_domain_span(sd); | |
5790 | struct cpumask *covered = sched_domains_tmpmask; | |
5791 | struct sd_data *sdd = sd->private; | |
5792 | struct sched_domain *child; | |
5793 | int i; | |
5794 | ||
5795 | cpumask_clear(covered); | |
5796 | ||
5797 | for_each_cpu(i, span) { | |
5798 | struct cpumask *sg_span; | |
5799 | ||
5800 | if (cpumask_test_cpu(i, covered)) | |
5801 | continue; | |
5802 | ||
c1174876 PZ |
5803 | child = *per_cpu_ptr(sdd->sd, i); |
5804 | ||
5805 | /* See the comment near build_group_mask(). */ | |
5806 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5807 | continue; | |
5808 | ||
e3589f6c | 5809 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5810 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5811 | |
5812 | if (!sg) | |
5813 | goto fail; | |
5814 | ||
5815 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5816 | if (child->child) { |
5817 | child = child->child; | |
5818 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5819 | } else | |
5820 | cpumask_set_cpu(i, sg_span); | |
5821 | ||
5822 | cpumask_or(covered, covered, sg_span); | |
5823 | ||
74a5ce20 | 5824 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5825 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5826 | build_group_mask(sd, sg); | |
5827 | ||
c3decf0d PZ |
5828 | /* |
5829 | * Initialize sgp->power such that even if we mess up the | |
5830 | * domains and no possible iteration will get us here, we won't | |
5831 | * die on a /0 trap. | |
5832 | */ | |
5833 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 5834 | |
c1174876 PZ |
5835 | /* |
5836 | * Make sure the first group of this domain contains the | |
5837 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5838 | * breaks. See update_sg_lb_stats(). | |
5839 | */ | |
74a5ce20 | 5840 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5841 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5842 | groups = sg; |
5843 | ||
5844 | if (!first) | |
5845 | first = sg; | |
5846 | if (last) | |
5847 | last->next = sg; | |
5848 | last = sg; | |
5849 | last->next = first; | |
5850 | } | |
5851 | sd->groups = groups; | |
5852 | ||
5853 | return 0; | |
5854 | ||
5855 | fail: | |
5856 | free_sched_groups(first, 0); | |
5857 | ||
5858 | return -ENOMEM; | |
5859 | } | |
5860 | ||
dce840a0 | 5861 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5862 | { |
dce840a0 PZ |
5863 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5864 | struct sched_domain *child = sd->child; | |
1da177e4 | 5865 | |
dce840a0 PZ |
5866 | if (child) |
5867 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5868 | |
9c3f75cb | 5869 | if (sg) { |
dce840a0 | 5870 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5871 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5872 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5873 | } |
dce840a0 PZ |
5874 | |
5875 | return cpu; | |
1e9f28fa | 5876 | } |
1e9f28fa | 5877 | |
01a08546 | 5878 | /* |
dce840a0 PZ |
5879 | * build_sched_groups will build a circular linked list of the groups |
5880 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5881 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5882 | * |
5883 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5884 | */ |
e3589f6c PZ |
5885 | static int |
5886 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5887 | { |
dce840a0 PZ |
5888 | struct sched_group *first = NULL, *last = NULL; |
5889 | struct sd_data *sdd = sd->private; | |
5890 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5891 | struct cpumask *covered; |
dce840a0 | 5892 | int i; |
9c1cfda2 | 5893 | |
e3589f6c PZ |
5894 | get_group(cpu, sdd, &sd->groups); |
5895 | atomic_inc(&sd->groups->ref); | |
5896 | ||
5897 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
5898 | return 0; | |
5899 | ||
f96225fd PZ |
5900 | lockdep_assert_held(&sched_domains_mutex); |
5901 | covered = sched_domains_tmpmask; | |
5902 | ||
dce840a0 | 5903 | cpumask_clear(covered); |
6711cab4 | 5904 | |
dce840a0 PZ |
5905 | for_each_cpu(i, span) { |
5906 | struct sched_group *sg; | |
5907 | int group = get_group(i, sdd, &sg); | |
5908 | int j; | |
6711cab4 | 5909 | |
dce840a0 PZ |
5910 | if (cpumask_test_cpu(i, covered)) |
5911 | continue; | |
6711cab4 | 5912 | |
dce840a0 | 5913 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5914 | sg->sgp->power = 0; |
c1174876 | 5915 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5916 | |
dce840a0 PZ |
5917 | for_each_cpu(j, span) { |
5918 | if (get_group(j, sdd, NULL) != group) | |
5919 | continue; | |
0601a88d | 5920 | |
dce840a0 PZ |
5921 | cpumask_set_cpu(j, covered); |
5922 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5923 | } | |
0601a88d | 5924 | |
dce840a0 PZ |
5925 | if (!first) |
5926 | first = sg; | |
5927 | if (last) | |
5928 | last->next = sg; | |
5929 | last = sg; | |
5930 | } | |
5931 | last->next = first; | |
e3589f6c PZ |
5932 | |
5933 | return 0; | |
0601a88d | 5934 | } |
51888ca2 | 5935 | |
89c4710e SS |
5936 | /* |
5937 | * Initialize sched groups cpu_power. | |
5938 | * | |
5939 | * cpu_power indicates the capacity of sched group, which is used while | |
5940 | * distributing the load between different sched groups in a sched domain. | |
5941 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5942 | * there are asymmetries in the topology. If there are asymmetries, group | |
5943 | * having more cpu_power will pickup more load compared to the group having | |
5944 | * less cpu_power. | |
89c4710e SS |
5945 | */ |
5946 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5947 | { | |
e3589f6c | 5948 | struct sched_group *sg = sd->groups; |
89c4710e | 5949 | |
e3589f6c PZ |
5950 | WARN_ON(!sd || !sg); |
5951 | ||
5952 | do { | |
5953 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5954 | sg = sg->next; | |
5955 | } while (sg != sd->groups); | |
89c4710e | 5956 | |
c1174876 | 5957 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5958 | return; |
aae6d3dd | 5959 | |
d274cb30 | 5960 | update_group_power(sd, cpu); |
69e1e811 | 5961 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5962 | } |
5963 | ||
029632fb PZ |
5964 | int __weak arch_sd_sibling_asym_packing(void) |
5965 | { | |
5966 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5967 | } |
5968 | ||
7c16ec58 MT |
5969 | /* |
5970 | * Initializers for schedule domains | |
5971 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5972 | */ | |
5973 | ||
a5d8c348 IM |
5974 | #ifdef CONFIG_SCHED_DEBUG |
5975 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5976 | #else | |
5977 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5978 | #endif | |
5979 | ||
54ab4ff4 PZ |
5980 | #define SD_INIT_FUNC(type) \ |
5981 | static noinline struct sched_domain * \ | |
5982 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5983 | { \ | |
5984 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5985 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5986 | SD_INIT_NAME(sd, type); \ |
5987 | sd->private = &tl->data; \ | |
5988 | return sd; \ | |
7c16ec58 MT |
5989 | } |
5990 | ||
5991 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5992 | #ifdef CONFIG_SCHED_SMT |
5993 | SD_INIT_FUNC(SIBLING) | |
5994 | #endif | |
5995 | #ifdef CONFIG_SCHED_MC | |
5996 | SD_INIT_FUNC(MC) | |
5997 | #endif | |
01a08546 HC |
5998 | #ifdef CONFIG_SCHED_BOOK |
5999 | SD_INIT_FUNC(BOOK) | |
6000 | #endif | |
7c16ec58 | 6001 | |
1d3504fc | 6002 | static int default_relax_domain_level = -1; |
60495e77 | 6003 | int sched_domain_level_max; |
1d3504fc HS |
6004 | |
6005 | static int __init setup_relax_domain_level(char *str) | |
6006 | { | |
a841f8ce DS |
6007 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6008 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6009 | |
1d3504fc HS |
6010 | return 1; |
6011 | } | |
6012 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6013 | ||
6014 | static void set_domain_attribute(struct sched_domain *sd, | |
6015 | struct sched_domain_attr *attr) | |
6016 | { | |
6017 | int request; | |
6018 | ||
6019 | if (!attr || attr->relax_domain_level < 0) { | |
6020 | if (default_relax_domain_level < 0) | |
6021 | return; | |
6022 | else | |
6023 | request = default_relax_domain_level; | |
6024 | } else | |
6025 | request = attr->relax_domain_level; | |
6026 | if (request < sd->level) { | |
6027 | /* turn off idle balance on this domain */ | |
c88d5910 | 6028 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6029 | } else { |
6030 | /* turn on idle balance on this domain */ | |
c88d5910 | 6031 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6032 | } |
6033 | } | |
6034 | ||
54ab4ff4 PZ |
6035 | static void __sdt_free(const struct cpumask *cpu_map); |
6036 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6037 | ||
2109b99e AH |
6038 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6039 | const struct cpumask *cpu_map) | |
6040 | { | |
6041 | switch (what) { | |
2109b99e | 6042 | case sa_rootdomain: |
822ff793 PZ |
6043 | if (!atomic_read(&d->rd->refcount)) |
6044 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6045 | case sa_sd: |
6046 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6047 | case sa_sd_storage: |
54ab4ff4 | 6048 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6049 | case sa_none: |
6050 | break; | |
6051 | } | |
6052 | } | |
3404c8d9 | 6053 | |
2109b99e AH |
6054 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6055 | const struct cpumask *cpu_map) | |
6056 | { | |
dce840a0 PZ |
6057 | memset(d, 0, sizeof(*d)); |
6058 | ||
54ab4ff4 PZ |
6059 | if (__sdt_alloc(cpu_map)) |
6060 | return sa_sd_storage; | |
dce840a0 PZ |
6061 | d->sd = alloc_percpu(struct sched_domain *); |
6062 | if (!d->sd) | |
6063 | return sa_sd_storage; | |
2109b99e | 6064 | d->rd = alloc_rootdomain(); |
dce840a0 | 6065 | if (!d->rd) |
21d42ccf | 6066 | return sa_sd; |
2109b99e AH |
6067 | return sa_rootdomain; |
6068 | } | |
57d885fe | 6069 | |
dce840a0 PZ |
6070 | /* |
6071 | * NULL the sd_data elements we've used to build the sched_domain and | |
6072 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6073 | * will not free the data we're using. | |
6074 | */ | |
6075 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6076 | { | |
6077 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6078 | |
6079 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6080 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6081 | ||
e3589f6c | 6082 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6083 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6084 | |
6085 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6086 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6087 | } |
6088 | ||
2c402dc3 PZ |
6089 | #ifdef CONFIG_SCHED_SMT |
6090 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6091 | { |
2c402dc3 | 6092 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6093 | } |
2c402dc3 | 6094 | #endif |
7f4588f3 | 6095 | |
d069b916 PZ |
6096 | /* |
6097 | * Topology list, bottom-up. | |
6098 | */ | |
2c402dc3 | 6099 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6100 | #ifdef CONFIG_SCHED_SMT |
6101 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6102 | #endif |
1e9f28fa | 6103 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6104 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6105 | #endif |
d069b916 PZ |
6106 | #ifdef CONFIG_SCHED_BOOK |
6107 | { sd_init_BOOK, cpu_book_mask, }, | |
6108 | #endif | |
6109 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
6110 | { NULL, }, |
6111 | }; | |
6112 | ||
6113 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6114 | ||
cb83b629 PZ |
6115 | #ifdef CONFIG_NUMA |
6116 | ||
6117 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
6118 | static int *sched_domains_numa_distance; |
6119 | static struct cpumask ***sched_domains_numa_masks; | |
6120 | static int sched_domains_curr_level; | |
6121 | ||
cb83b629 PZ |
6122 | static inline int sd_local_flags(int level) |
6123 | { | |
10717dcd | 6124 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
6125 | return 0; |
6126 | ||
6127 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6128 | } | |
6129 | ||
6130 | static struct sched_domain * | |
6131 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6132 | { | |
6133 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6134 | int level = tl->numa_level; | |
6135 | int sd_weight = cpumask_weight( | |
6136 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6137 | ||
6138 | *sd = (struct sched_domain){ | |
6139 | .min_interval = sd_weight, | |
6140 | .max_interval = 2*sd_weight, | |
6141 | .busy_factor = 32, | |
870a0bb5 | 6142 | .imbalance_pct = 125, |
cb83b629 PZ |
6143 | .cache_nice_tries = 2, |
6144 | .busy_idx = 3, | |
6145 | .idle_idx = 2, | |
6146 | .newidle_idx = 0, | |
6147 | .wake_idx = 0, | |
6148 | .forkexec_idx = 0, | |
6149 | ||
6150 | .flags = 1*SD_LOAD_BALANCE | |
6151 | | 1*SD_BALANCE_NEWIDLE | |
6152 | | 0*SD_BALANCE_EXEC | |
6153 | | 0*SD_BALANCE_FORK | |
6154 | | 0*SD_BALANCE_WAKE | |
6155 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 6156 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
6157 | | 0*SD_SHARE_PKG_RESOURCES |
6158 | | 1*SD_SERIALIZE | |
6159 | | 0*SD_PREFER_SIBLING | |
6160 | | sd_local_flags(level) | |
6161 | , | |
6162 | .last_balance = jiffies, | |
6163 | .balance_interval = sd_weight, | |
6164 | }; | |
6165 | SD_INIT_NAME(sd, NUMA); | |
6166 | sd->private = &tl->data; | |
6167 | ||
6168 | /* | |
6169 | * Ugly hack to pass state to sd_numa_mask()... | |
6170 | */ | |
6171 | sched_domains_curr_level = tl->numa_level; | |
6172 | ||
6173 | return sd; | |
6174 | } | |
6175 | ||
6176 | static const struct cpumask *sd_numa_mask(int cpu) | |
6177 | { | |
6178 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6179 | } | |
6180 | ||
d039ac60 PZ |
6181 | static void sched_numa_warn(const char *str) |
6182 | { | |
6183 | static int done = false; | |
6184 | int i,j; | |
6185 | ||
6186 | if (done) | |
6187 | return; | |
6188 | ||
6189 | done = true; | |
6190 | ||
6191 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6192 | ||
6193 | for (i = 0; i < nr_node_ids; i++) { | |
6194 | printk(KERN_WARNING " "); | |
6195 | for (j = 0; j < nr_node_ids; j++) | |
6196 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6197 | printk(KERN_CONT "\n"); | |
6198 | } | |
6199 | printk(KERN_WARNING "\n"); | |
6200 | } | |
6201 | ||
6202 | static bool find_numa_distance(int distance) | |
6203 | { | |
6204 | int i; | |
6205 | ||
6206 | if (distance == node_distance(0, 0)) | |
6207 | return true; | |
6208 | ||
6209 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6210 | if (sched_domains_numa_distance[i] == distance) | |
6211 | return true; | |
6212 | } | |
6213 | ||
6214 | return false; | |
6215 | } | |
6216 | ||
cb83b629 PZ |
6217 | static void sched_init_numa(void) |
6218 | { | |
6219 | int next_distance, curr_distance = node_distance(0, 0); | |
6220 | struct sched_domain_topology_level *tl; | |
6221 | int level = 0; | |
6222 | int i, j, k; | |
6223 | ||
cb83b629 PZ |
6224 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6225 | if (!sched_domains_numa_distance) | |
6226 | return; | |
6227 | ||
6228 | /* | |
6229 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6230 | * unique distances in the node_distance() table. | |
6231 | * | |
6232 | * Assumes node_distance(0,j) includes all distances in | |
6233 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6234 | */ |
6235 | next_distance = curr_distance; | |
6236 | for (i = 0; i < nr_node_ids; i++) { | |
6237 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6238 | for (k = 0; k < nr_node_ids; k++) { |
6239 | int distance = node_distance(i, k); | |
6240 | ||
6241 | if (distance > curr_distance && | |
6242 | (distance < next_distance || | |
6243 | next_distance == curr_distance)) | |
6244 | next_distance = distance; | |
6245 | ||
6246 | /* | |
6247 | * While not a strong assumption it would be nice to know | |
6248 | * about cases where if node A is connected to B, B is not | |
6249 | * equally connected to A. | |
6250 | */ | |
6251 | if (sched_debug() && node_distance(k, i) != distance) | |
6252 | sched_numa_warn("Node-distance not symmetric"); | |
6253 | ||
6254 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6255 | sched_numa_warn("Node-0 not representative"); | |
6256 | } | |
6257 | if (next_distance != curr_distance) { | |
6258 | sched_domains_numa_distance[level++] = next_distance; | |
6259 | sched_domains_numa_levels = level; | |
6260 | curr_distance = next_distance; | |
6261 | } else break; | |
cb83b629 | 6262 | } |
d039ac60 PZ |
6263 | |
6264 | /* | |
6265 | * In case of sched_debug() we verify the above assumption. | |
6266 | */ | |
6267 | if (!sched_debug()) | |
6268 | break; | |
cb83b629 PZ |
6269 | } |
6270 | /* | |
6271 | * 'level' contains the number of unique distances, excluding the | |
6272 | * identity distance node_distance(i,i). | |
6273 | * | |
28b4a521 | 6274 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6275 | * numbers. |
6276 | */ | |
6277 | ||
5f7865f3 TC |
6278 | /* |
6279 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6280 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6281 | * the array will contain less then 'level' members. This could be | |
6282 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6283 | * in other functions. | |
6284 | * | |
6285 | * We reset it to 'level' at the end of this function. | |
6286 | */ | |
6287 | sched_domains_numa_levels = 0; | |
6288 | ||
cb83b629 PZ |
6289 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6290 | if (!sched_domains_numa_masks) | |
6291 | return; | |
6292 | ||
6293 | /* | |
6294 | * Now for each level, construct a mask per node which contains all | |
6295 | * cpus of nodes that are that many hops away from us. | |
6296 | */ | |
6297 | for (i = 0; i < level; i++) { | |
6298 | sched_domains_numa_masks[i] = | |
6299 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6300 | if (!sched_domains_numa_masks[i]) | |
6301 | return; | |
6302 | ||
6303 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6304 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6305 | if (!mask) |
6306 | return; | |
6307 | ||
6308 | sched_domains_numa_masks[i][j] = mask; | |
6309 | ||
6310 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6311 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6312 | continue; |
6313 | ||
6314 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6315 | } | |
6316 | } | |
6317 | } | |
6318 | ||
6319 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6320 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6321 | if (!tl) | |
6322 | return; | |
6323 | ||
6324 | /* | |
6325 | * Copy the default topology bits.. | |
6326 | */ | |
6327 | for (i = 0; default_topology[i].init; i++) | |
6328 | tl[i] = default_topology[i]; | |
6329 | ||
6330 | /* | |
6331 | * .. and append 'j' levels of NUMA goodness. | |
6332 | */ | |
6333 | for (j = 0; j < level; i++, j++) { | |
6334 | tl[i] = (struct sched_domain_topology_level){ | |
6335 | .init = sd_numa_init, | |
6336 | .mask = sd_numa_mask, | |
6337 | .flags = SDTL_OVERLAP, | |
6338 | .numa_level = j, | |
6339 | }; | |
6340 | } | |
6341 | ||
6342 | sched_domain_topology = tl; | |
5f7865f3 TC |
6343 | |
6344 | sched_domains_numa_levels = level; | |
cb83b629 | 6345 | } |
301a5cba TC |
6346 | |
6347 | static void sched_domains_numa_masks_set(int cpu) | |
6348 | { | |
6349 | int i, j; | |
6350 | int node = cpu_to_node(cpu); | |
6351 | ||
6352 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6353 | for (j = 0; j < nr_node_ids; j++) { | |
6354 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6355 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6356 | } | |
6357 | } | |
6358 | } | |
6359 | ||
6360 | static void sched_domains_numa_masks_clear(int cpu) | |
6361 | { | |
6362 | int i, j; | |
6363 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6364 | for (j = 0; j < nr_node_ids; j++) | |
6365 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6366 | } | |
6367 | } | |
6368 | ||
6369 | /* | |
6370 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6371 | * are onlined. | |
6372 | */ | |
6373 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6374 | unsigned long action, | |
6375 | void *hcpu) | |
6376 | { | |
6377 | int cpu = (long)hcpu; | |
6378 | ||
6379 | switch (action & ~CPU_TASKS_FROZEN) { | |
6380 | case CPU_ONLINE: | |
6381 | sched_domains_numa_masks_set(cpu); | |
6382 | break; | |
6383 | ||
6384 | case CPU_DEAD: | |
6385 | sched_domains_numa_masks_clear(cpu); | |
6386 | break; | |
6387 | ||
6388 | default: | |
6389 | return NOTIFY_DONE; | |
6390 | } | |
6391 | ||
6392 | return NOTIFY_OK; | |
cb83b629 PZ |
6393 | } |
6394 | #else | |
6395 | static inline void sched_init_numa(void) | |
6396 | { | |
6397 | } | |
301a5cba TC |
6398 | |
6399 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6400 | unsigned long action, | |
6401 | void *hcpu) | |
6402 | { | |
6403 | return 0; | |
6404 | } | |
cb83b629 PZ |
6405 | #endif /* CONFIG_NUMA */ |
6406 | ||
54ab4ff4 PZ |
6407 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6408 | { | |
6409 | struct sched_domain_topology_level *tl; | |
6410 | int j; | |
6411 | ||
6412 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6413 | struct sd_data *sdd = &tl->data; | |
6414 | ||
6415 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6416 | if (!sdd->sd) | |
6417 | return -ENOMEM; | |
6418 | ||
6419 | sdd->sg = alloc_percpu(struct sched_group *); | |
6420 | if (!sdd->sg) | |
6421 | return -ENOMEM; | |
6422 | ||
9c3f75cb PZ |
6423 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6424 | if (!sdd->sgp) | |
6425 | return -ENOMEM; | |
6426 | ||
54ab4ff4 PZ |
6427 | for_each_cpu(j, cpu_map) { |
6428 | struct sched_domain *sd; | |
6429 | struct sched_group *sg; | |
9c3f75cb | 6430 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6431 | |
6432 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6433 | GFP_KERNEL, cpu_to_node(j)); | |
6434 | if (!sd) | |
6435 | return -ENOMEM; | |
6436 | ||
6437 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6438 | ||
6439 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6440 | GFP_KERNEL, cpu_to_node(j)); | |
6441 | if (!sg) | |
6442 | return -ENOMEM; | |
6443 | ||
30b4e9eb IM |
6444 | sg->next = sg; |
6445 | ||
54ab4ff4 | 6446 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6447 | |
c1174876 | 6448 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6449 | GFP_KERNEL, cpu_to_node(j)); |
6450 | if (!sgp) | |
6451 | return -ENOMEM; | |
6452 | ||
6453 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6454 | } |
6455 | } | |
6456 | ||
6457 | return 0; | |
6458 | } | |
6459 | ||
6460 | static void __sdt_free(const struct cpumask *cpu_map) | |
6461 | { | |
6462 | struct sched_domain_topology_level *tl; | |
6463 | int j; | |
6464 | ||
6465 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6466 | struct sd_data *sdd = &tl->data; | |
6467 | ||
6468 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6469 | struct sched_domain *sd; |
6470 | ||
6471 | if (sdd->sd) { | |
6472 | sd = *per_cpu_ptr(sdd->sd, j); | |
6473 | if (sd && (sd->flags & SD_OVERLAP)) | |
6474 | free_sched_groups(sd->groups, 0); | |
6475 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6476 | } | |
6477 | ||
6478 | if (sdd->sg) | |
6479 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6480 | if (sdd->sgp) | |
6481 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6482 | } |
6483 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6484 | sdd->sd = NULL; |
54ab4ff4 | 6485 | free_percpu(sdd->sg); |
fb2cf2c6 | 6486 | sdd->sg = NULL; |
9c3f75cb | 6487 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6488 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6489 | } |
6490 | } | |
6491 | ||
2c402dc3 PZ |
6492 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6493 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6494 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6495 | int cpu) |
6496 | { | |
54ab4ff4 | 6497 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6498 | if (!sd) |
d069b916 | 6499 | return child; |
2c402dc3 | 6500 | |
2c402dc3 | 6501 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6502 | if (child) { |
6503 | sd->level = child->level + 1; | |
6504 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6505 | child->parent = sd; |
60495e77 | 6506 | } |
d069b916 | 6507 | sd->child = child; |
a841f8ce | 6508 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6509 | |
6510 | return sd; | |
6511 | } | |
6512 | ||
2109b99e AH |
6513 | /* |
6514 | * Build sched domains for a given set of cpus and attach the sched domains | |
6515 | * to the individual cpus | |
6516 | */ | |
dce840a0 PZ |
6517 | static int build_sched_domains(const struct cpumask *cpu_map, |
6518 | struct sched_domain_attr *attr) | |
2109b99e AH |
6519 | { |
6520 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6521 | struct sched_domain *sd; |
2109b99e | 6522 | struct s_data d; |
822ff793 | 6523 | int i, ret = -ENOMEM; |
9c1cfda2 | 6524 | |
2109b99e AH |
6525 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6526 | if (alloc_state != sa_rootdomain) | |
6527 | goto error; | |
9c1cfda2 | 6528 | |
dce840a0 | 6529 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6530 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6531 | struct sched_domain_topology_level *tl; |
6532 | ||
3bd65a80 | 6533 | sd = NULL; |
e3589f6c | 6534 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6535 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6536 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6537 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6538 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6539 | break; | |
e3589f6c | 6540 | } |
d274cb30 | 6541 | |
d069b916 PZ |
6542 | while (sd->child) |
6543 | sd = sd->child; | |
6544 | ||
21d42ccf | 6545 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6546 | } |
6547 | ||
6548 | /* Build the groups for the domains */ | |
6549 | for_each_cpu(i, cpu_map) { | |
6550 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6551 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6552 | if (sd->flags & SD_OVERLAP) { |
6553 | if (build_overlap_sched_groups(sd, i)) | |
6554 | goto error; | |
6555 | } else { | |
6556 | if (build_sched_groups(sd, i)) | |
6557 | goto error; | |
6558 | } | |
1cf51902 | 6559 | } |
a06dadbe | 6560 | } |
9c1cfda2 | 6561 | |
1da177e4 | 6562 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6563 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6564 | if (!cpumask_test_cpu(i, cpu_map)) | |
6565 | continue; | |
9c1cfda2 | 6566 | |
dce840a0 PZ |
6567 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6568 | claim_allocations(i, sd); | |
cd4ea6ae | 6569 | init_sched_groups_power(i, sd); |
dce840a0 | 6570 | } |
f712c0c7 | 6571 | } |
9c1cfda2 | 6572 | |
1da177e4 | 6573 | /* Attach the domains */ |
dce840a0 | 6574 | rcu_read_lock(); |
abcd083a | 6575 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6576 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6577 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6578 | } |
dce840a0 | 6579 | rcu_read_unlock(); |
51888ca2 | 6580 | |
822ff793 | 6581 | ret = 0; |
51888ca2 | 6582 | error: |
2109b99e | 6583 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6584 | return ret; |
1da177e4 | 6585 | } |
029190c5 | 6586 | |
acc3f5d7 | 6587 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6588 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6589 | static struct sched_domain_attr *dattr_cur; |
6590 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6591 | |
6592 | /* | |
6593 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6594 | * cpumask) fails, then fallback to a single sched domain, |
6595 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6596 | */ |
4212823f | 6597 | static cpumask_var_t fallback_doms; |
029190c5 | 6598 | |
ee79d1bd HC |
6599 | /* |
6600 | * arch_update_cpu_topology lets virtualized architectures update the | |
6601 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6602 | * or 0 if it stayed the same. | |
6603 | */ | |
6604 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6605 | { |
ee79d1bd | 6606 | return 0; |
22e52b07 HC |
6607 | } |
6608 | ||
acc3f5d7 RR |
6609 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6610 | { | |
6611 | int i; | |
6612 | cpumask_var_t *doms; | |
6613 | ||
6614 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6615 | if (!doms) | |
6616 | return NULL; | |
6617 | for (i = 0; i < ndoms; i++) { | |
6618 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6619 | free_sched_domains(doms, i); | |
6620 | return NULL; | |
6621 | } | |
6622 | } | |
6623 | return doms; | |
6624 | } | |
6625 | ||
6626 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6627 | { | |
6628 | unsigned int i; | |
6629 | for (i = 0; i < ndoms; i++) | |
6630 | free_cpumask_var(doms[i]); | |
6631 | kfree(doms); | |
6632 | } | |
6633 | ||
1a20ff27 | 6634 | /* |
41a2d6cf | 6635 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6636 | * For now this just excludes isolated cpus, but could be used to |
6637 | * exclude other special cases in the future. | |
1a20ff27 | 6638 | */ |
c4a8849a | 6639 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6640 | { |
7378547f MM |
6641 | int err; |
6642 | ||
22e52b07 | 6643 | arch_update_cpu_topology(); |
029190c5 | 6644 | ndoms_cur = 1; |
acc3f5d7 | 6645 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6646 | if (!doms_cur) |
acc3f5d7 RR |
6647 | doms_cur = &fallback_doms; |
6648 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6649 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6650 | register_sched_domain_sysctl(); |
7378547f MM |
6651 | |
6652 | return err; | |
1a20ff27 DG |
6653 | } |
6654 | ||
1a20ff27 DG |
6655 | /* |
6656 | * Detach sched domains from a group of cpus specified in cpu_map | |
6657 | * These cpus will now be attached to the NULL domain | |
6658 | */ | |
96f874e2 | 6659 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6660 | { |
6661 | int i; | |
6662 | ||
dce840a0 | 6663 | rcu_read_lock(); |
abcd083a | 6664 | for_each_cpu(i, cpu_map) |
57d885fe | 6665 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6666 | rcu_read_unlock(); |
1a20ff27 DG |
6667 | } |
6668 | ||
1d3504fc HS |
6669 | /* handle null as "default" */ |
6670 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6671 | struct sched_domain_attr *new, int idx_new) | |
6672 | { | |
6673 | struct sched_domain_attr tmp; | |
6674 | ||
6675 | /* fast path */ | |
6676 | if (!new && !cur) | |
6677 | return 1; | |
6678 | ||
6679 | tmp = SD_ATTR_INIT; | |
6680 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6681 | new ? (new + idx_new) : &tmp, | |
6682 | sizeof(struct sched_domain_attr)); | |
6683 | } | |
6684 | ||
029190c5 PJ |
6685 | /* |
6686 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6687 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6688 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6689 | * It destroys each deleted domain and builds each new domain. | |
6690 | * | |
acc3f5d7 | 6691 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6692 | * The masks don't intersect (don't overlap.) We should setup one |
6693 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6694 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6695 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6696 | * it as it is. | |
6697 | * | |
acc3f5d7 RR |
6698 | * The passed in 'doms_new' should be allocated using |
6699 | * alloc_sched_domains. This routine takes ownership of it and will | |
6700 | * free_sched_domains it when done with it. If the caller failed the | |
6701 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6702 | * and partition_sched_domains() will fallback to the single partition | |
6703 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6704 | * |
96f874e2 | 6705 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6706 | * ndoms_new == 0 is a special case for destroying existing domains, |
6707 | * and it will not create the default domain. | |
dfb512ec | 6708 | * |
029190c5 PJ |
6709 | * Call with hotplug lock held |
6710 | */ | |
acc3f5d7 | 6711 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6712 | struct sched_domain_attr *dattr_new) |
029190c5 | 6713 | { |
dfb512ec | 6714 | int i, j, n; |
d65bd5ec | 6715 | int new_topology; |
029190c5 | 6716 | |
712555ee | 6717 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6718 | |
7378547f MM |
6719 | /* always unregister in case we don't destroy any domains */ |
6720 | unregister_sched_domain_sysctl(); | |
6721 | ||
d65bd5ec HC |
6722 | /* Let architecture update cpu core mappings. */ |
6723 | new_topology = arch_update_cpu_topology(); | |
6724 | ||
dfb512ec | 6725 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6726 | |
6727 | /* Destroy deleted domains */ | |
6728 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6729 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6730 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6731 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6732 | goto match1; |
6733 | } | |
6734 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6735 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6736 | match1: |
6737 | ; | |
6738 | } | |
6739 | ||
e761b772 MK |
6740 | if (doms_new == NULL) { |
6741 | ndoms_cur = 0; | |
acc3f5d7 | 6742 | doms_new = &fallback_doms; |
6ad4c188 | 6743 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6744 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6745 | } |
6746 | ||
029190c5 PJ |
6747 | /* Build new domains */ |
6748 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6749 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6750 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6751 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6752 | goto match2; |
6753 | } | |
6754 | /* no match - add a new doms_new */ | |
dce840a0 | 6755 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6756 | match2: |
6757 | ; | |
6758 | } | |
6759 | ||
6760 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6761 | if (doms_cur != &fallback_doms) |
6762 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6763 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6764 | doms_cur = doms_new; |
1d3504fc | 6765 | dattr_cur = dattr_new; |
029190c5 | 6766 | ndoms_cur = ndoms_new; |
7378547f MM |
6767 | |
6768 | register_sched_domain_sysctl(); | |
a1835615 | 6769 | |
712555ee | 6770 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6771 | } |
6772 | ||
d35be8ba SB |
6773 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6774 | ||
1da177e4 | 6775 | /* |
3a101d05 TH |
6776 | * Update cpusets according to cpu_active mask. If cpusets are |
6777 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6778 | * around partition_sched_domains(). | |
d35be8ba SB |
6779 | * |
6780 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6781 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6782 | */ |
0b2e918a TH |
6783 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6784 | void *hcpu) | |
e761b772 | 6785 | { |
d35be8ba SB |
6786 | switch (action) { |
6787 | case CPU_ONLINE_FROZEN: | |
6788 | case CPU_DOWN_FAILED_FROZEN: | |
6789 | ||
6790 | /* | |
6791 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6792 | * resume sequence. As long as this is not the last online | |
6793 | * operation in the resume sequence, just build a single sched | |
6794 | * domain, ignoring cpusets. | |
6795 | */ | |
6796 | num_cpus_frozen--; | |
6797 | if (likely(num_cpus_frozen)) { | |
6798 | partition_sched_domains(1, NULL, NULL); | |
6799 | break; | |
6800 | } | |
6801 | ||
6802 | /* | |
6803 | * This is the last CPU online operation. So fall through and | |
6804 | * restore the original sched domains by considering the | |
6805 | * cpuset configurations. | |
6806 | */ | |
6807 | ||
e761b772 | 6808 | case CPU_ONLINE: |
6ad4c188 | 6809 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6810 | cpuset_update_active_cpus(true); |
d35be8ba | 6811 | break; |
3a101d05 TH |
6812 | default: |
6813 | return NOTIFY_DONE; | |
6814 | } | |
d35be8ba | 6815 | return NOTIFY_OK; |
3a101d05 | 6816 | } |
e761b772 | 6817 | |
0b2e918a TH |
6818 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6819 | void *hcpu) | |
3a101d05 | 6820 | { |
d35be8ba | 6821 | switch (action) { |
3a101d05 | 6822 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6823 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6824 | break; |
6825 | case CPU_DOWN_PREPARE_FROZEN: | |
6826 | num_cpus_frozen++; | |
6827 | partition_sched_domains(1, NULL, NULL); | |
6828 | break; | |
e761b772 MK |
6829 | default: |
6830 | return NOTIFY_DONE; | |
6831 | } | |
d35be8ba | 6832 | return NOTIFY_OK; |
e761b772 | 6833 | } |
e761b772 | 6834 | |
1da177e4 LT |
6835 | void __init sched_init_smp(void) |
6836 | { | |
dcc30a35 RR |
6837 | cpumask_var_t non_isolated_cpus; |
6838 | ||
6839 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6840 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6841 | |
cb83b629 PZ |
6842 | sched_init_numa(); |
6843 | ||
95402b38 | 6844 | get_online_cpus(); |
712555ee | 6845 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6846 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6847 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6848 | if (cpumask_empty(non_isolated_cpus)) | |
6849 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6850 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6851 | put_online_cpus(); |
e761b772 | 6852 | |
301a5cba | 6853 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6854 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6855 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6856 | |
6857 | /* RT runtime code needs to handle some hotplug events */ | |
6858 | hotcpu_notifier(update_runtime, 0); | |
6859 | ||
b328ca18 | 6860 | init_hrtick(); |
5c1e1767 NP |
6861 | |
6862 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6863 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6864 | BUG(); |
19978ca6 | 6865 | sched_init_granularity(); |
dcc30a35 | 6866 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6867 | |
0e3900e6 | 6868 | init_sched_rt_class(); |
1da177e4 LT |
6869 | } |
6870 | #else | |
6871 | void __init sched_init_smp(void) | |
6872 | { | |
19978ca6 | 6873 | sched_init_granularity(); |
1da177e4 LT |
6874 | } |
6875 | #endif /* CONFIG_SMP */ | |
6876 | ||
cd1bb94b AB |
6877 | const_debug unsigned int sysctl_timer_migration = 1; |
6878 | ||
1da177e4 LT |
6879 | int in_sched_functions(unsigned long addr) |
6880 | { | |
1da177e4 LT |
6881 | return in_lock_functions(addr) || |
6882 | (addr >= (unsigned long)__sched_text_start | |
6883 | && addr < (unsigned long)__sched_text_end); | |
6884 | } | |
6885 | ||
029632fb | 6886 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6887 | /* |
6888 | * Default task group. | |
6889 | * Every task in system belongs to this group at bootup. | |
6890 | */ | |
029632fb | 6891 | struct task_group root_task_group; |
35cf4e50 | 6892 | LIST_HEAD(task_groups); |
052f1dc7 | 6893 | #endif |
6f505b16 | 6894 | |
e6252c3e | 6895 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6896 | |
1da177e4 LT |
6897 | void __init sched_init(void) |
6898 | { | |
dd41f596 | 6899 | int i, j; |
434d53b0 MT |
6900 | unsigned long alloc_size = 0, ptr; |
6901 | ||
6902 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6903 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6904 | #endif | |
6905 | #ifdef CONFIG_RT_GROUP_SCHED | |
6906 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6907 | #endif |
df7c8e84 | 6908 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6909 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6910 | #endif |
434d53b0 | 6911 | if (alloc_size) { |
36b7b6d4 | 6912 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6913 | |
6914 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6915 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6916 | ptr += nr_cpu_ids * sizeof(void **); |
6917 | ||
07e06b01 | 6918 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6919 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6920 | |
6d6bc0ad | 6921 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6922 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6923 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6924 | ptr += nr_cpu_ids * sizeof(void **); |
6925 | ||
07e06b01 | 6926 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6927 | ptr += nr_cpu_ids * sizeof(void **); |
6928 | ||
6d6bc0ad | 6929 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6930 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6931 | for_each_possible_cpu(i) { | |
e6252c3e | 6932 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6933 | ptr += cpumask_size(); |
6934 | } | |
6935 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6936 | } |
dd41f596 | 6937 | |
57d885fe GH |
6938 | #ifdef CONFIG_SMP |
6939 | init_defrootdomain(); | |
6940 | #endif | |
6941 | ||
d0b27fa7 PZ |
6942 | init_rt_bandwidth(&def_rt_bandwidth, |
6943 | global_rt_period(), global_rt_runtime()); | |
6944 | ||
6945 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6946 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6947 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6948 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6949 | |
7c941438 | 6950 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6951 | list_add(&root_task_group.list, &task_groups); |
6952 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6953 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6954 | autogroup_init(&init_task); |
54c707e9 | 6955 | |
7c941438 | 6956 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6957 | |
0a945022 | 6958 | for_each_possible_cpu(i) { |
70b97a7f | 6959 | struct rq *rq; |
1da177e4 LT |
6960 | |
6961 | rq = cpu_rq(i); | |
05fa785c | 6962 | raw_spin_lock_init(&rq->lock); |
7897986b | 6963 | rq->nr_running = 0; |
dce48a84 TG |
6964 | rq->calc_load_active = 0; |
6965 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6966 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6967 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6968 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6969 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6970 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6971 | /* |
07e06b01 | 6972 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6973 | * |
6974 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6975 | * gets 100% of the cpu resources in the system. This overall | |
6976 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6977 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6978 | * based on each entity's (task or task-group's) weight |
6979 | * (se->load.weight). | |
6980 | * | |
07e06b01 | 6981 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6982 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6983 | * then A0's share of the cpu resource is: | |
6984 | * | |
0d905bca | 6985 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6986 | * |
07e06b01 YZ |
6987 | * We achieve this by letting root_task_group's tasks sit |
6988 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6989 | */ |
ab84d31e | 6990 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6991 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6992 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6993 | ||
6994 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6995 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6996 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6997 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6998 | #endif |
1da177e4 | 6999 | |
dd41f596 IM |
7000 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7001 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7002 | |
7003 | rq->last_load_update_tick = jiffies; | |
7004 | ||
1da177e4 | 7005 | #ifdef CONFIG_SMP |
41c7ce9a | 7006 | rq->sd = NULL; |
57d885fe | 7007 | rq->rd = NULL; |
1399fa78 | 7008 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 7009 | rq->post_schedule = 0; |
1da177e4 | 7010 | rq->active_balance = 0; |
dd41f596 | 7011 | rq->next_balance = jiffies; |
1da177e4 | 7012 | rq->push_cpu = 0; |
0a2966b4 | 7013 | rq->cpu = i; |
1f11eb6a | 7014 | rq->online = 0; |
eae0c9df MG |
7015 | rq->idle_stamp = 0; |
7016 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
7017 | |
7018 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7019 | ||
dc938520 | 7020 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7021 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7022 | rq->nohz_flags = 0; |
83cd4fe2 | 7023 | #endif |
265f22a9 FW |
7024 | #ifdef CONFIG_NO_HZ_FULL |
7025 | rq->last_sched_tick = 0; | |
7026 | #endif | |
1da177e4 | 7027 | #endif |
8f4d37ec | 7028 | init_rq_hrtick(rq); |
1da177e4 | 7029 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7030 | } |
7031 | ||
2dd73a4f | 7032 | set_load_weight(&init_task); |
b50f60ce | 7033 | |
e107be36 AK |
7034 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7035 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7036 | #endif | |
7037 | ||
b50f60ce | 7038 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 7039 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
7040 | #endif |
7041 | ||
1da177e4 LT |
7042 | /* |
7043 | * The boot idle thread does lazy MMU switching as well: | |
7044 | */ | |
7045 | atomic_inc(&init_mm.mm_count); | |
7046 | enter_lazy_tlb(&init_mm, current); | |
7047 | ||
7048 | /* | |
7049 | * Make us the idle thread. Technically, schedule() should not be | |
7050 | * called from this thread, however somewhere below it might be, | |
7051 | * but because we are the idle thread, we just pick up running again | |
7052 | * when this runqueue becomes "idle". | |
7053 | */ | |
7054 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7055 | |
7056 | calc_load_update = jiffies + LOAD_FREQ; | |
7057 | ||
dd41f596 IM |
7058 | /* |
7059 | * During early bootup we pretend to be a normal task: | |
7060 | */ | |
7061 | current->sched_class = &fair_sched_class; | |
6892b75e | 7062 | |
bf4d83f6 | 7063 | #ifdef CONFIG_SMP |
4cb98839 | 7064 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7065 | /* May be allocated at isolcpus cmdline parse time */ |
7066 | if (cpu_isolated_map == NULL) | |
7067 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7068 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7069 | #endif |
7070 | init_sched_fair_class(); | |
6a7b3dc3 | 7071 | |
6892b75e | 7072 | scheduler_running = 1; |
1da177e4 LT |
7073 | } |
7074 | ||
d902db1e | 7075 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7076 | static inline int preempt_count_equals(int preempt_offset) |
7077 | { | |
234da7bc | 7078 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7079 | |
4ba8216c | 7080 | return (nested == preempt_offset); |
e4aafea2 FW |
7081 | } |
7082 | ||
d894837f | 7083 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7084 | { |
1da177e4 LT |
7085 | static unsigned long prev_jiffy; /* ratelimiting */ |
7086 | ||
b3fbab05 | 7087 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7088 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7089 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7090 | return; |
7091 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7092 | return; | |
7093 | prev_jiffy = jiffies; | |
7094 | ||
3df0fc5b PZ |
7095 | printk(KERN_ERR |
7096 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7097 | file, line); | |
7098 | printk(KERN_ERR | |
7099 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7100 | in_atomic(), irqs_disabled(), | |
7101 | current->pid, current->comm); | |
aef745fc IM |
7102 | |
7103 | debug_show_held_locks(current); | |
7104 | if (irqs_disabled()) | |
7105 | print_irqtrace_events(current); | |
7106 | dump_stack(); | |
1da177e4 LT |
7107 | } |
7108 | EXPORT_SYMBOL(__might_sleep); | |
7109 | #endif | |
7110 | ||
7111 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7112 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7113 | { | |
da7a735e PZ |
7114 | const struct sched_class *prev_class = p->sched_class; |
7115 | int old_prio = p->prio; | |
3a5e4dc1 | 7116 | int on_rq; |
3e51f33f | 7117 | |
fd2f4419 | 7118 | on_rq = p->on_rq; |
3a5e4dc1 | 7119 | if (on_rq) |
4ca9b72b | 7120 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7121 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7122 | if (on_rq) { | |
4ca9b72b | 7123 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7124 | resched_task(rq->curr); |
7125 | } | |
da7a735e PZ |
7126 | |
7127 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7128 | } |
7129 | ||
1da177e4 LT |
7130 | void normalize_rt_tasks(void) |
7131 | { | |
a0f98a1c | 7132 | struct task_struct *g, *p; |
1da177e4 | 7133 | unsigned long flags; |
70b97a7f | 7134 | struct rq *rq; |
1da177e4 | 7135 | |
4cf5d77a | 7136 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7137 | do_each_thread(g, p) { |
178be793 IM |
7138 | /* |
7139 | * Only normalize user tasks: | |
7140 | */ | |
7141 | if (!p->mm) | |
7142 | continue; | |
7143 | ||
6cfb0d5d | 7144 | p->se.exec_start = 0; |
6cfb0d5d | 7145 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7146 | p->se.statistics.wait_start = 0; |
7147 | p->se.statistics.sleep_start = 0; | |
7148 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7149 | #endif |
dd41f596 IM |
7150 | |
7151 | if (!rt_task(p)) { | |
7152 | /* | |
7153 | * Renice negative nice level userspace | |
7154 | * tasks back to 0: | |
7155 | */ | |
7156 | if (TASK_NICE(p) < 0 && p->mm) | |
7157 | set_user_nice(p, 0); | |
1da177e4 | 7158 | continue; |
dd41f596 | 7159 | } |
1da177e4 | 7160 | |
1d615482 | 7161 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7162 | rq = __task_rq_lock(p); |
1da177e4 | 7163 | |
178be793 | 7164 | normalize_task(rq, p); |
3a5e4dc1 | 7165 | |
b29739f9 | 7166 | __task_rq_unlock(rq); |
1d615482 | 7167 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7168 | } while_each_thread(g, p); |
7169 | ||
4cf5d77a | 7170 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7171 | } |
7172 | ||
7173 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7174 | |
67fc4e0c | 7175 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7176 | /* |
67fc4e0c | 7177 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7178 | * |
7179 | * They can only be called when the whole system has been | |
7180 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7181 | * activity can take place. Using them for anything else would | |
7182 | * be a serious bug, and as a result, they aren't even visible | |
7183 | * under any other configuration. | |
7184 | */ | |
7185 | ||
7186 | /** | |
7187 | * curr_task - return the current task for a given cpu. | |
7188 | * @cpu: the processor in question. | |
7189 | * | |
7190 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7191 | */ | |
36c8b586 | 7192 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7193 | { |
7194 | return cpu_curr(cpu); | |
7195 | } | |
7196 | ||
67fc4e0c JW |
7197 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7198 | ||
7199 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7200 | /** |
7201 | * set_curr_task - set the current task for a given cpu. | |
7202 | * @cpu: the processor in question. | |
7203 | * @p: the task pointer to set. | |
7204 | * | |
7205 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7206 | * are serviced on a separate stack. It allows the architecture to switch the |
7207 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7208 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7209 | * and caller must save the original value of the current task (see | |
7210 | * curr_task() above) and restore that value before reenabling interrupts and | |
7211 | * re-starting the system. | |
7212 | * | |
7213 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7214 | */ | |
36c8b586 | 7215 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7216 | { |
7217 | cpu_curr(cpu) = p; | |
7218 | } | |
7219 | ||
7220 | #endif | |
29f59db3 | 7221 | |
7c941438 | 7222 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7223 | /* task_group_lock serializes the addition/removal of task groups */ |
7224 | static DEFINE_SPINLOCK(task_group_lock); | |
7225 | ||
bccbe08a PZ |
7226 | static void free_sched_group(struct task_group *tg) |
7227 | { | |
7228 | free_fair_sched_group(tg); | |
7229 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7230 | autogroup_free(tg); |
bccbe08a PZ |
7231 | kfree(tg); |
7232 | } | |
7233 | ||
7234 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7235 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7236 | { |
7237 | struct task_group *tg; | |
bccbe08a PZ |
7238 | |
7239 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7240 | if (!tg) | |
7241 | return ERR_PTR(-ENOMEM); | |
7242 | ||
ec7dc8ac | 7243 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7244 | goto err; |
7245 | ||
ec7dc8ac | 7246 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7247 | goto err; |
7248 | ||
ace783b9 LZ |
7249 | return tg; |
7250 | ||
7251 | err: | |
7252 | free_sched_group(tg); | |
7253 | return ERR_PTR(-ENOMEM); | |
7254 | } | |
7255 | ||
7256 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7257 | { | |
7258 | unsigned long flags; | |
7259 | ||
8ed36996 | 7260 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7261 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7262 | |
7263 | WARN_ON(!parent); /* root should already exist */ | |
7264 | ||
7265 | tg->parent = parent; | |
f473aa5e | 7266 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7267 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7268 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7269 | } |
7270 | ||
9b5b7751 | 7271 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7272 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7273 | { |
29f59db3 | 7274 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7275 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7276 | } |
7277 | ||
9b5b7751 | 7278 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7279 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7280 | { |
7281 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7282 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7283 | } | |
7284 | ||
7285 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7286 | { |
8ed36996 | 7287 | unsigned long flags; |
9b5b7751 | 7288 | int i; |
29f59db3 | 7289 | |
3d4b47b4 PZ |
7290 | /* end participation in shares distribution */ |
7291 | for_each_possible_cpu(i) | |
bccbe08a | 7292 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7293 | |
7294 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7295 | list_del_rcu(&tg->list); |
f473aa5e | 7296 | list_del_rcu(&tg->siblings); |
8ed36996 | 7297 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7298 | } |
7299 | ||
9b5b7751 | 7300 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7301 | * The caller of this function should have put the task in its new group |
7302 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7303 | * reflect its new group. | |
9b5b7751 SV |
7304 | */ |
7305 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7306 | { |
8323f26c | 7307 | struct task_group *tg; |
29f59db3 SV |
7308 | int on_rq, running; |
7309 | unsigned long flags; | |
7310 | struct rq *rq; | |
7311 | ||
7312 | rq = task_rq_lock(tsk, &flags); | |
7313 | ||
051a1d1a | 7314 | running = task_current(rq, tsk); |
fd2f4419 | 7315 | on_rq = tsk->on_rq; |
29f59db3 | 7316 | |
0e1f3483 | 7317 | if (on_rq) |
29f59db3 | 7318 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7319 | if (unlikely(running)) |
7320 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7321 | |
8323f26c PZ |
7322 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
7323 | lockdep_is_held(&tsk->sighand->siglock)), | |
7324 | struct task_group, css); | |
7325 | tg = autogroup_task_group(tsk, tg); | |
7326 | tsk->sched_task_group = tg; | |
7327 | ||
810b3817 | 7328 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7329 | if (tsk->sched_class->task_move_group) |
7330 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7331 | else | |
810b3817 | 7332 | #endif |
b2b5ce02 | 7333 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7334 | |
0e1f3483 HS |
7335 | if (unlikely(running)) |
7336 | tsk->sched_class->set_curr_task(rq); | |
7337 | if (on_rq) | |
371fd7e7 | 7338 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7339 | |
0122ec5b | 7340 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7341 | } |
7c941438 | 7342 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7343 | |
a790de99 | 7344 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7345 | static unsigned long to_ratio(u64 period, u64 runtime) |
7346 | { | |
7347 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7348 | return 1ULL << 20; |
9f0c1e56 | 7349 | |
9a7e0b18 | 7350 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7351 | } |
a790de99 PT |
7352 | #endif |
7353 | ||
7354 | #ifdef CONFIG_RT_GROUP_SCHED | |
7355 | /* | |
7356 | * Ensure that the real time constraints are schedulable. | |
7357 | */ | |
7358 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7359 | |
9a7e0b18 PZ |
7360 | /* Must be called with tasklist_lock held */ |
7361 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7362 | { |
9a7e0b18 | 7363 | struct task_struct *g, *p; |
b40b2e8e | 7364 | |
9a7e0b18 | 7365 | do_each_thread(g, p) { |
029632fb | 7366 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7367 | return 1; |
7368 | } while_each_thread(g, p); | |
b40b2e8e | 7369 | |
9a7e0b18 PZ |
7370 | return 0; |
7371 | } | |
b40b2e8e | 7372 | |
9a7e0b18 PZ |
7373 | struct rt_schedulable_data { |
7374 | struct task_group *tg; | |
7375 | u64 rt_period; | |
7376 | u64 rt_runtime; | |
7377 | }; | |
b40b2e8e | 7378 | |
a790de99 | 7379 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7380 | { |
7381 | struct rt_schedulable_data *d = data; | |
7382 | struct task_group *child; | |
7383 | unsigned long total, sum = 0; | |
7384 | u64 period, runtime; | |
b40b2e8e | 7385 | |
9a7e0b18 PZ |
7386 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7387 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7388 | |
9a7e0b18 PZ |
7389 | if (tg == d->tg) { |
7390 | period = d->rt_period; | |
7391 | runtime = d->rt_runtime; | |
b40b2e8e | 7392 | } |
b40b2e8e | 7393 | |
4653f803 PZ |
7394 | /* |
7395 | * Cannot have more runtime than the period. | |
7396 | */ | |
7397 | if (runtime > period && runtime != RUNTIME_INF) | |
7398 | return -EINVAL; | |
6f505b16 | 7399 | |
4653f803 PZ |
7400 | /* |
7401 | * Ensure we don't starve existing RT tasks. | |
7402 | */ | |
9a7e0b18 PZ |
7403 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7404 | return -EBUSY; | |
6f505b16 | 7405 | |
9a7e0b18 | 7406 | total = to_ratio(period, runtime); |
6f505b16 | 7407 | |
4653f803 PZ |
7408 | /* |
7409 | * Nobody can have more than the global setting allows. | |
7410 | */ | |
7411 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7412 | return -EINVAL; | |
6f505b16 | 7413 | |
4653f803 PZ |
7414 | /* |
7415 | * The sum of our children's runtime should not exceed our own. | |
7416 | */ | |
9a7e0b18 PZ |
7417 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7418 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7419 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7420 | |
9a7e0b18 PZ |
7421 | if (child == d->tg) { |
7422 | period = d->rt_period; | |
7423 | runtime = d->rt_runtime; | |
7424 | } | |
6f505b16 | 7425 | |
9a7e0b18 | 7426 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7427 | } |
6f505b16 | 7428 | |
9a7e0b18 PZ |
7429 | if (sum > total) |
7430 | return -EINVAL; | |
7431 | ||
7432 | return 0; | |
6f505b16 PZ |
7433 | } |
7434 | ||
9a7e0b18 | 7435 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7436 | { |
8277434e PT |
7437 | int ret; |
7438 | ||
9a7e0b18 PZ |
7439 | struct rt_schedulable_data data = { |
7440 | .tg = tg, | |
7441 | .rt_period = period, | |
7442 | .rt_runtime = runtime, | |
7443 | }; | |
7444 | ||
8277434e PT |
7445 | rcu_read_lock(); |
7446 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7447 | rcu_read_unlock(); | |
7448 | ||
7449 | return ret; | |
521f1a24 DG |
7450 | } |
7451 | ||
ab84d31e | 7452 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7453 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7454 | { |
ac086bc2 | 7455 | int i, err = 0; |
9f0c1e56 | 7456 | |
9f0c1e56 | 7457 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7458 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7459 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7460 | if (err) | |
9f0c1e56 | 7461 | goto unlock; |
ac086bc2 | 7462 | |
0986b11b | 7463 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7464 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7465 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7466 | |
7467 | for_each_possible_cpu(i) { | |
7468 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7469 | ||
0986b11b | 7470 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7471 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7472 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7473 | } |
0986b11b | 7474 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7475 | unlock: |
521f1a24 | 7476 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7477 | mutex_unlock(&rt_constraints_mutex); |
7478 | ||
7479 | return err; | |
6f505b16 PZ |
7480 | } |
7481 | ||
25cc7da7 | 7482 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7483 | { |
7484 | u64 rt_runtime, rt_period; | |
7485 | ||
7486 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7487 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7488 | if (rt_runtime_us < 0) | |
7489 | rt_runtime = RUNTIME_INF; | |
7490 | ||
ab84d31e | 7491 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7492 | } |
7493 | ||
25cc7da7 | 7494 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7495 | { |
7496 | u64 rt_runtime_us; | |
7497 | ||
d0b27fa7 | 7498 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7499 | return -1; |
7500 | ||
d0b27fa7 | 7501 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7502 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7503 | return rt_runtime_us; | |
7504 | } | |
d0b27fa7 | 7505 | |
25cc7da7 | 7506 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7507 | { |
7508 | u64 rt_runtime, rt_period; | |
7509 | ||
7510 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7511 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7512 | ||
619b0488 R |
7513 | if (rt_period == 0) |
7514 | return -EINVAL; | |
7515 | ||
ab84d31e | 7516 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7517 | } |
7518 | ||
25cc7da7 | 7519 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7520 | { |
7521 | u64 rt_period_us; | |
7522 | ||
7523 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7524 | do_div(rt_period_us, NSEC_PER_USEC); | |
7525 | return rt_period_us; | |
7526 | } | |
7527 | ||
7528 | static int sched_rt_global_constraints(void) | |
7529 | { | |
4653f803 | 7530 | u64 runtime, period; |
d0b27fa7 PZ |
7531 | int ret = 0; |
7532 | ||
ec5d4989 HS |
7533 | if (sysctl_sched_rt_period <= 0) |
7534 | return -EINVAL; | |
7535 | ||
4653f803 PZ |
7536 | runtime = global_rt_runtime(); |
7537 | period = global_rt_period(); | |
7538 | ||
7539 | /* | |
7540 | * Sanity check on the sysctl variables. | |
7541 | */ | |
7542 | if (runtime > period && runtime != RUNTIME_INF) | |
7543 | return -EINVAL; | |
10b612f4 | 7544 | |
d0b27fa7 | 7545 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7546 | read_lock(&tasklist_lock); |
4653f803 | 7547 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7548 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7549 | mutex_unlock(&rt_constraints_mutex); |
7550 | ||
7551 | return ret; | |
7552 | } | |
54e99124 | 7553 | |
25cc7da7 | 7554 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7555 | { |
7556 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7557 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7558 | return 0; | |
7559 | ||
7560 | return 1; | |
7561 | } | |
7562 | ||
6d6bc0ad | 7563 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7564 | static int sched_rt_global_constraints(void) |
7565 | { | |
ac086bc2 PZ |
7566 | unsigned long flags; |
7567 | int i; | |
7568 | ||
ec5d4989 HS |
7569 | if (sysctl_sched_rt_period <= 0) |
7570 | return -EINVAL; | |
7571 | ||
60aa605d PZ |
7572 | /* |
7573 | * There's always some RT tasks in the root group | |
7574 | * -- migration, kstopmachine etc.. | |
7575 | */ | |
7576 | if (sysctl_sched_rt_runtime == 0) | |
7577 | return -EBUSY; | |
7578 | ||
0986b11b | 7579 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7580 | for_each_possible_cpu(i) { |
7581 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7582 | ||
0986b11b | 7583 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7584 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7585 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7586 | } |
0986b11b | 7587 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7588 | |
d0b27fa7 PZ |
7589 | return 0; |
7590 | } | |
6d6bc0ad | 7591 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7592 | |
ce0dbbbb CW |
7593 | int sched_rr_handler(struct ctl_table *table, int write, |
7594 | void __user *buffer, size_t *lenp, | |
7595 | loff_t *ppos) | |
7596 | { | |
7597 | int ret; | |
7598 | static DEFINE_MUTEX(mutex); | |
7599 | ||
7600 | mutex_lock(&mutex); | |
7601 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7602 | /* make sure that internally we keep jiffies */ | |
7603 | /* also, writing zero resets timeslice to default */ | |
7604 | if (!ret && write) { | |
7605 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7606 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7607 | } | |
7608 | mutex_unlock(&mutex); | |
7609 | return ret; | |
7610 | } | |
7611 | ||
d0b27fa7 | 7612 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7613 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7614 | loff_t *ppos) |
7615 | { | |
7616 | int ret; | |
7617 | int old_period, old_runtime; | |
7618 | static DEFINE_MUTEX(mutex); | |
7619 | ||
7620 | mutex_lock(&mutex); | |
7621 | old_period = sysctl_sched_rt_period; | |
7622 | old_runtime = sysctl_sched_rt_runtime; | |
7623 | ||
8d65af78 | 7624 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7625 | |
7626 | if (!ret && write) { | |
7627 | ret = sched_rt_global_constraints(); | |
7628 | if (ret) { | |
7629 | sysctl_sched_rt_period = old_period; | |
7630 | sysctl_sched_rt_runtime = old_runtime; | |
7631 | } else { | |
7632 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7633 | def_rt_bandwidth.rt_period = | |
7634 | ns_to_ktime(global_rt_period()); | |
7635 | } | |
7636 | } | |
7637 | mutex_unlock(&mutex); | |
7638 | ||
7639 | return ret; | |
7640 | } | |
68318b8e | 7641 | |
052f1dc7 | 7642 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7643 | |
7644 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7645 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7646 | { |
2b01dfe3 PM |
7647 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7648 | struct task_group, css); | |
68318b8e SV |
7649 | } |
7650 | ||
92fb9748 | 7651 | static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp) |
68318b8e | 7652 | { |
ec7dc8ac | 7653 | struct task_group *tg, *parent; |
68318b8e | 7654 | |
2b01dfe3 | 7655 | if (!cgrp->parent) { |
68318b8e | 7656 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7657 | return &root_task_group.css; |
68318b8e SV |
7658 | } |
7659 | ||
ec7dc8ac DG |
7660 | parent = cgroup_tg(cgrp->parent); |
7661 | tg = sched_create_group(parent); | |
68318b8e SV |
7662 | if (IS_ERR(tg)) |
7663 | return ERR_PTR(-ENOMEM); | |
7664 | ||
68318b8e SV |
7665 | return &tg->css; |
7666 | } | |
7667 | ||
ace783b9 LZ |
7668 | static int cpu_cgroup_css_online(struct cgroup *cgrp) |
7669 | { | |
7670 | struct task_group *tg = cgroup_tg(cgrp); | |
7671 | struct task_group *parent; | |
7672 | ||
7673 | if (!cgrp->parent) | |
7674 | return 0; | |
7675 | ||
7676 | parent = cgroup_tg(cgrp->parent); | |
7677 | sched_online_group(tg, parent); | |
7678 | return 0; | |
7679 | } | |
7680 | ||
92fb9748 | 7681 | static void cpu_cgroup_css_free(struct cgroup *cgrp) |
68318b8e | 7682 | { |
2b01dfe3 | 7683 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7684 | |
7685 | sched_destroy_group(tg); | |
7686 | } | |
7687 | ||
ace783b9 LZ |
7688 | static void cpu_cgroup_css_offline(struct cgroup *cgrp) |
7689 | { | |
7690 | struct task_group *tg = cgroup_tg(cgrp); | |
7691 | ||
7692 | sched_offline_group(tg); | |
7693 | } | |
7694 | ||
761b3ef5 | 7695 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 7696 | struct cgroup_taskset *tset) |
68318b8e | 7697 | { |
bb9d97b6 TH |
7698 | struct task_struct *task; |
7699 | ||
7700 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7701 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7702 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7703 | return -EINVAL; | |
b68aa230 | 7704 | #else |
bb9d97b6 TH |
7705 | /* We don't support RT-tasks being in separate groups */ |
7706 | if (task->sched_class != &fair_sched_class) | |
7707 | return -EINVAL; | |
b68aa230 | 7708 | #endif |
bb9d97b6 | 7709 | } |
be367d09 BB |
7710 | return 0; |
7711 | } | |
68318b8e | 7712 | |
761b3ef5 | 7713 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 7714 | struct cgroup_taskset *tset) |
68318b8e | 7715 | { |
bb9d97b6 TH |
7716 | struct task_struct *task; |
7717 | ||
7718 | cgroup_taskset_for_each(task, cgrp, tset) | |
7719 | sched_move_task(task); | |
68318b8e SV |
7720 | } |
7721 | ||
068c5cc5 | 7722 | static void |
761b3ef5 LZ |
7723 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
7724 | struct task_struct *task) | |
068c5cc5 PZ |
7725 | { |
7726 | /* | |
7727 | * cgroup_exit() is called in the copy_process() failure path. | |
7728 | * Ignore this case since the task hasn't ran yet, this avoids | |
7729 | * trying to poke a half freed task state from generic code. | |
7730 | */ | |
7731 | if (!(task->flags & PF_EXITING)) | |
7732 | return; | |
7733 | ||
7734 | sched_move_task(task); | |
7735 | } | |
7736 | ||
052f1dc7 | 7737 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7738 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7739 | u64 shareval) |
68318b8e | 7740 | { |
c8b28116 | 7741 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7742 | } |
7743 | ||
f4c753b7 | 7744 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7745 | { |
2b01dfe3 | 7746 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7747 | |
c8b28116 | 7748 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7749 | } |
ab84d31e PT |
7750 | |
7751 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7752 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7753 | ||
ab84d31e PT |
7754 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7755 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7756 | ||
a790de99 PT |
7757 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7758 | ||
ab84d31e PT |
7759 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7760 | { | |
56f570e5 | 7761 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7762 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7763 | |
7764 | if (tg == &root_task_group) | |
7765 | return -EINVAL; | |
7766 | ||
7767 | /* | |
7768 | * Ensure we have at some amount of bandwidth every period. This is | |
7769 | * to prevent reaching a state of large arrears when throttled via | |
7770 | * entity_tick() resulting in prolonged exit starvation. | |
7771 | */ | |
7772 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7773 | return -EINVAL; | |
7774 | ||
7775 | /* | |
7776 | * Likewise, bound things on the otherside by preventing insane quota | |
7777 | * periods. This also allows us to normalize in computing quota | |
7778 | * feasibility. | |
7779 | */ | |
7780 | if (period > max_cfs_quota_period) | |
7781 | return -EINVAL; | |
7782 | ||
a790de99 PT |
7783 | mutex_lock(&cfs_constraints_mutex); |
7784 | ret = __cfs_schedulable(tg, period, quota); | |
7785 | if (ret) | |
7786 | goto out_unlock; | |
7787 | ||
58088ad0 | 7788 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7789 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7790 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7791 | raw_spin_lock_irq(&cfs_b->lock); |
7792 | cfs_b->period = ns_to_ktime(period); | |
7793 | cfs_b->quota = quota; | |
58088ad0 | 7794 | |
a9cf55b2 | 7795 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7796 | /* restart the period timer (if active) to handle new period expiry */ |
7797 | if (runtime_enabled && cfs_b->timer_active) { | |
7798 | /* force a reprogram */ | |
7799 | cfs_b->timer_active = 0; | |
7800 | __start_cfs_bandwidth(cfs_b); | |
7801 | } | |
ab84d31e PT |
7802 | raw_spin_unlock_irq(&cfs_b->lock); |
7803 | ||
7804 | for_each_possible_cpu(i) { | |
7805 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7806 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7807 | |
7808 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7809 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7810 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7811 | |
029632fb | 7812 | if (cfs_rq->throttled) |
671fd9da | 7813 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7814 | raw_spin_unlock_irq(&rq->lock); |
7815 | } | |
a790de99 PT |
7816 | out_unlock: |
7817 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7818 | |
a790de99 | 7819 | return ret; |
ab84d31e PT |
7820 | } |
7821 | ||
7822 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7823 | { | |
7824 | u64 quota, period; | |
7825 | ||
029632fb | 7826 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7827 | if (cfs_quota_us < 0) |
7828 | quota = RUNTIME_INF; | |
7829 | else | |
7830 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7831 | ||
7832 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7833 | } | |
7834 | ||
7835 | long tg_get_cfs_quota(struct task_group *tg) | |
7836 | { | |
7837 | u64 quota_us; | |
7838 | ||
029632fb | 7839 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7840 | return -1; |
7841 | ||
029632fb | 7842 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7843 | do_div(quota_us, NSEC_PER_USEC); |
7844 | ||
7845 | return quota_us; | |
7846 | } | |
7847 | ||
7848 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7849 | { | |
7850 | u64 quota, period; | |
7851 | ||
7852 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7853 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7854 | |
ab84d31e PT |
7855 | return tg_set_cfs_bandwidth(tg, period, quota); |
7856 | } | |
7857 | ||
7858 | long tg_get_cfs_period(struct task_group *tg) | |
7859 | { | |
7860 | u64 cfs_period_us; | |
7861 | ||
029632fb | 7862 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7863 | do_div(cfs_period_us, NSEC_PER_USEC); |
7864 | ||
7865 | return cfs_period_us; | |
7866 | } | |
7867 | ||
7868 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7869 | { | |
7870 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7871 | } | |
7872 | ||
7873 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7874 | s64 cfs_quota_us) | |
7875 | { | |
7876 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7877 | } | |
7878 | ||
7879 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7880 | { | |
7881 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7882 | } | |
7883 | ||
7884 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7885 | u64 cfs_period_us) | |
7886 | { | |
7887 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7888 | } | |
7889 | ||
a790de99 PT |
7890 | struct cfs_schedulable_data { |
7891 | struct task_group *tg; | |
7892 | u64 period, quota; | |
7893 | }; | |
7894 | ||
7895 | /* | |
7896 | * normalize group quota/period to be quota/max_period | |
7897 | * note: units are usecs | |
7898 | */ | |
7899 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7900 | struct cfs_schedulable_data *d) | |
7901 | { | |
7902 | u64 quota, period; | |
7903 | ||
7904 | if (tg == d->tg) { | |
7905 | period = d->period; | |
7906 | quota = d->quota; | |
7907 | } else { | |
7908 | period = tg_get_cfs_period(tg); | |
7909 | quota = tg_get_cfs_quota(tg); | |
7910 | } | |
7911 | ||
7912 | /* note: these should typically be equivalent */ | |
7913 | if (quota == RUNTIME_INF || quota == -1) | |
7914 | return RUNTIME_INF; | |
7915 | ||
7916 | return to_ratio(period, quota); | |
7917 | } | |
7918 | ||
7919 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7920 | { | |
7921 | struct cfs_schedulable_data *d = data; | |
029632fb | 7922 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7923 | s64 quota = 0, parent_quota = -1; |
7924 | ||
7925 | if (!tg->parent) { | |
7926 | quota = RUNTIME_INF; | |
7927 | } else { | |
029632fb | 7928 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7929 | |
7930 | quota = normalize_cfs_quota(tg, d); | |
7931 | parent_quota = parent_b->hierarchal_quota; | |
7932 | ||
7933 | /* | |
7934 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7935 | * limit is set | |
7936 | */ | |
7937 | if (quota == RUNTIME_INF) | |
7938 | quota = parent_quota; | |
7939 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7940 | return -EINVAL; | |
7941 | } | |
7942 | cfs_b->hierarchal_quota = quota; | |
7943 | ||
7944 | return 0; | |
7945 | } | |
7946 | ||
7947 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7948 | { | |
8277434e | 7949 | int ret; |
a790de99 PT |
7950 | struct cfs_schedulable_data data = { |
7951 | .tg = tg, | |
7952 | .period = period, | |
7953 | .quota = quota, | |
7954 | }; | |
7955 | ||
7956 | if (quota != RUNTIME_INF) { | |
7957 | do_div(data.period, NSEC_PER_USEC); | |
7958 | do_div(data.quota, NSEC_PER_USEC); | |
7959 | } | |
7960 | ||
8277434e PT |
7961 | rcu_read_lock(); |
7962 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7963 | rcu_read_unlock(); | |
7964 | ||
7965 | return ret; | |
a790de99 | 7966 | } |
e8da1b18 NR |
7967 | |
7968 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7969 | struct cgroup_map_cb *cb) | |
7970 | { | |
7971 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7972 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7973 | |
7974 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7975 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7976 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7977 | ||
7978 | return 0; | |
7979 | } | |
ab84d31e | 7980 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7981 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7982 | |
052f1dc7 | 7983 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7984 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7985 | s64 val) |
6f505b16 | 7986 | { |
06ecb27c | 7987 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7988 | } |
7989 | ||
06ecb27c | 7990 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7991 | { |
06ecb27c | 7992 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7993 | } |
d0b27fa7 PZ |
7994 | |
7995 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7996 | u64 rt_period_us) | |
7997 | { | |
7998 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7999 | } | |
8000 | ||
8001 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8002 | { | |
8003 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8004 | } | |
6d6bc0ad | 8005 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8006 | |
fe5c7cc2 | 8007 | static struct cftype cpu_files[] = { |
052f1dc7 | 8008 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8009 | { |
8010 | .name = "shares", | |
f4c753b7 PM |
8011 | .read_u64 = cpu_shares_read_u64, |
8012 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8013 | }, |
052f1dc7 | 8014 | #endif |
ab84d31e PT |
8015 | #ifdef CONFIG_CFS_BANDWIDTH |
8016 | { | |
8017 | .name = "cfs_quota_us", | |
8018 | .read_s64 = cpu_cfs_quota_read_s64, | |
8019 | .write_s64 = cpu_cfs_quota_write_s64, | |
8020 | }, | |
8021 | { | |
8022 | .name = "cfs_period_us", | |
8023 | .read_u64 = cpu_cfs_period_read_u64, | |
8024 | .write_u64 = cpu_cfs_period_write_u64, | |
8025 | }, | |
e8da1b18 NR |
8026 | { |
8027 | .name = "stat", | |
8028 | .read_map = cpu_stats_show, | |
8029 | }, | |
ab84d31e | 8030 | #endif |
052f1dc7 | 8031 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8032 | { |
9f0c1e56 | 8033 | .name = "rt_runtime_us", |
06ecb27c PM |
8034 | .read_s64 = cpu_rt_runtime_read, |
8035 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8036 | }, |
d0b27fa7 PZ |
8037 | { |
8038 | .name = "rt_period_us", | |
f4c753b7 PM |
8039 | .read_u64 = cpu_rt_period_read_uint, |
8040 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8041 | }, |
052f1dc7 | 8042 | #endif |
4baf6e33 | 8043 | { } /* terminate */ |
68318b8e SV |
8044 | }; |
8045 | ||
68318b8e | 8046 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 8047 | .name = "cpu", |
92fb9748 TH |
8048 | .css_alloc = cpu_cgroup_css_alloc, |
8049 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8050 | .css_online = cpu_cgroup_css_online, |
8051 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8052 | .can_attach = cpu_cgroup_can_attach, |
8053 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8054 | .exit = cpu_cgroup_exit, |
38605cae | 8055 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 8056 | .base_cftypes = cpu_files, |
68318b8e SV |
8057 | .early_init = 1, |
8058 | }; | |
8059 | ||
052f1dc7 | 8060 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8061 | |
b637a328 PM |
8062 | void dump_cpu_task(int cpu) |
8063 | { | |
8064 | pr_info("Task dump for CPU %d:\n", cpu); | |
8065 | sched_show_task(cpu_curr(cpu)); | |
8066 | } |