]>
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> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
e6e6685a GC |
77 | #ifdef CONFIG_PARAVIRT |
78 | #include <asm/paravirt.h> | |
79 | #endif | |
1da177e4 | 80 | |
029632fb | 81 | #include "sched.h" |
391e43da | 82 | #include "../workqueue_sched.h" |
6e0534f2 | 83 | |
a8d154b0 | 84 | #define CREATE_TRACE_POINTS |
ad8d75ff | 85 | #include <trace/events/sched.h> |
a8d154b0 | 86 | |
029632fb | 87 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 88 | { |
58088ad0 PT |
89 | unsigned long delta; |
90 | ktime_t soft, hard, now; | |
d0b27fa7 | 91 | |
58088ad0 PT |
92 | for (;;) { |
93 | if (hrtimer_active(period_timer)) | |
94 | break; | |
95 | ||
96 | now = hrtimer_cb_get_time(period_timer); | |
97 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 98 | |
58088ad0 PT |
99 | soft = hrtimer_get_softexpires(period_timer); |
100 | hard = hrtimer_get_expires(period_timer); | |
101 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
102 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
103 | HRTIMER_MODE_ABS_PINNED, 0); | |
104 | } | |
105 | } | |
106 | ||
029632fb PZ |
107 | DEFINE_MUTEX(sched_domains_mutex); |
108 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 109 | |
fe44d621 | 110 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 111 | |
029632fb | 112 | void update_rq_clock(struct rq *rq) |
3e51f33f | 113 | { |
fe44d621 | 114 | s64 delta; |
305e6835 | 115 | |
61eadef6 | 116 | if (rq->skip_clock_update > 0) |
f26f9aff | 117 | return; |
aa483808 | 118 | |
fe44d621 PZ |
119 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
120 | rq->clock += delta; | |
121 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
122 | } |
123 | ||
bf5c91ba IM |
124 | /* |
125 | * Debugging: various feature bits | |
126 | */ | |
f00b45c1 | 127 | |
f00b45c1 PZ |
128 | #define SCHED_FEAT(name, enabled) \ |
129 | (1UL << __SCHED_FEAT_##name) * enabled | | |
130 | ||
bf5c91ba | 131 | const_debug unsigned int sysctl_sched_features = |
391e43da | 132 | #include "features.h" |
f00b45c1 PZ |
133 | 0; |
134 | ||
135 | #undef SCHED_FEAT | |
136 | ||
137 | #ifdef CONFIG_SCHED_DEBUG | |
138 | #define SCHED_FEAT(name, enabled) \ | |
139 | #name , | |
140 | ||
983ed7a6 | 141 | static __read_mostly char *sched_feat_names[] = { |
391e43da | 142 | #include "features.h" |
f00b45c1 PZ |
143 | NULL |
144 | }; | |
145 | ||
146 | #undef SCHED_FEAT | |
147 | ||
34f3a814 | 148 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 149 | { |
f00b45c1 PZ |
150 | int i; |
151 | ||
152 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
153 | if (!(sysctl_sched_features & (1UL << i))) |
154 | seq_puts(m, "NO_"); | |
155 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 156 | } |
34f3a814 | 157 | seq_puts(m, "\n"); |
f00b45c1 | 158 | |
34f3a814 | 159 | return 0; |
f00b45c1 PZ |
160 | } |
161 | ||
162 | static ssize_t | |
163 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
164 | size_t cnt, loff_t *ppos) | |
165 | { | |
166 | char buf[64]; | |
7740191c | 167 | char *cmp; |
f00b45c1 PZ |
168 | int neg = 0; |
169 | int i; | |
170 | ||
171 | if (cnt > 63) | |
172 | cnt = 63; | |
173 | ||
174 | if (copy_from_user(&buf, ubuf, cnt)) | |
175 | return -EFAULT; | |
176 | ||
177 | buf[cnt] = 0; | |
7740191c | 178 | cmp = strstrip(buf); |
f00b45c1 | 179 | |
524429c3 | 180 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
181 | neg = 1; |
182 | cmp += 3; | |
183 | } | |
184 | ||
185 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 186 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
187 | if (neg) |
188 | sysctl_sched_features &= ~(1UL << i); | |
189 | else | |
190 | sysctl_sched_features |= (1UL << i); | |
191 | break; | |
192 | } | |
193 | } | |
194 | ||
195 | if (!sched_feat_names[i]) | |
196 | return -EINVAL; | |
197 | ||
42994724 | 198 | *ppos += cnt; |
f00b45c1 PZ |
199 | |
200 | return cnt; | |
201 | } | |
202 | ||
34f3a814 LZ |
203 | static int sched_feat_open(struct inode *inode, struct file *filp) |
204 | { | |
205 | return single_open(filp, sched_feat_show, NULL); | |
206 | } | |
207 | ||
828c0950 | 208 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
209 | .open = sched_feat_open, |
210 | .write = sched_feat_write, | |
211 | .read = seq_read, | |
212 | .llseek = seq_lseek, | |
213 | .release = single_release, | |
f00b45c1 PZ |
214 | }; |
215 | ||
216 | static __init int sched_init_debug(void) | |
217 | { | |
f00b45c1 PZ |
218 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
219 | &sched_feat_fops); | |
220 | ||
221 | return 0; | |
222 | } | |
223 | late_initcall(sched_init_debug); | |
224 | ||
225 | #endif | |
226 | ||
b82d9fdd PZ |
227 | /* |
228 | * Number of tasks to iterate in a single balance run. | |
229 | * Limited because this is done with IRQs disabled. | |
230 | */ | |
231 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
232 | ||
e9e9250b PZ |
233 | /* |
234 | * period over which we average the RT time consumption, measured | |
235 | * in ms. | |
236 | * | |
237 | * default: 1s | |
238 | */ | |
239 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
240 | ||
029632fb PZ |
241 | /* |
242 | * period over which we measure -rt task cpu usage in us. | |
243 | * default: 1s | |
244 | */ | |
245 | unsigned int sysctl_sched_rt_period = 1000000; | |
8a25d5de | 246 | |
029632fb PZ |
247 | __read_mostly int scheduler_running; |
248 | ||
249 | /* | |
250 | * part of the period that we allow rt tasks to run in us. | |
251 | * default: 0.95s | |
252 | */ | |
253 | int sysctl_sched_rt_runtime = 950000; | |
4866cde0 | 254 | |
4866cde0 | 255 | |
1da177e4 | 256 | |
0970d299 | 257 | /* |
0122ec5b | 258 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 259 | */ |
70b97a7f | 260 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
261 | __acquires(rq->lock) |
262 | { | |
0970d299 PZ |
263 | struct rq *rq; |
264 | ||
0122ec5b PZ |
265 | lockdep_assert_held(&p->pi_lock); |
266 | ||
3a5c359a | 267 | for (;;) { |
0970d299 | 268 | rq = task_rq(p); |
05fa785c | 269 | raw_spin_lock(&rq->lock); |
65cc8e48 | 270 | if (likely(rq == task_rq(p))) |
3a5c359a | 271 | return rq; |
05fa785c | 272 | raw_spin_unlock(&rq->lock); |
b29739f9 | 273 | } |
b29739f9 IM |
274 | } |
275 | ||
1da177e4 | 276 | /* |
0122ec5b | 277 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 278 | */ |
70b97a7f | 279 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 280 | __acquires(p->pi_lock) |
1da177e4 LT |
281 | __acquires(rq->lock) |
282 | { | |
70b97a7f | 283 | struct rq *rq; |
1da177e4 | 284 | |
3a5c359a | 285 | for (;;) { |
0122ec5b | 286 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 287 | rq = task_rq(p); |
05fa785c | 288 | raw_spin_lock(&rq->lock); |
65cc8e48 | 289 | if (likely(rq == task_rq(p))) |
3a5c359a | 290 | return rq; |
0122ec5b PZ |
291 | raw_spin_unlock(&rq->lock); |
292 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 293 | } |
1da177e4 LT |
294 | } |
295 | ||
a9957449 | 296 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
297 | __releases(rq->lock) |
298 | { | |
05fa785c | 299 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
300 | } |
301 | ||
0122ec5b PZ |
302 | static inline void |
303 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 304 | __releases(rq->lock) |
0122ec5b | 305 | __releases(p->pi_lock) |
1da177e4 | 306 | { |
0122ec5b PZ |
307 | raw_spin_unlock(&rq->lock); |
308 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
309 | } |
310 | ||
1da177e4 | 311 | /* |
cc2a73b5 | 312 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 313 | */ |
a9957449 | 314 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
315 | __acquires(rq->lock) |
316 | { | |
70b97a7f | 317 | struct rq *rq; |
1da177e4 LT |
318 | |
319 | local_irq_disable(); | |
320 | rq = this_rq(); | |
05fa785c | 321 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
322 | |
323 | return rq; | |
324 | } | |
325 | ||
8f4d37ec PZ |
326 | #ifdef CONFIG_SCHED_HRTICK |
327 | /* | |
328 | * Use HR-timers to deliver accurate preemption points. | |
329 | * | |
330 | * Its all a bit involved since we cannot program an hrt while holding the | |
331 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
332 | * reschedule event. | |
333 | * | |
334 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
335 | * rq->lock. | |
336 | */ | |
8f4d37ec | 337 | |
8f4d37ec PZ |
338 | static void hrtick_clear(struct rq *rq) |
339 | { | |
340 | if (hrtimer_active(&rq->hrtick_timer)) | |
341 | hrtimer_cancel(&rq->hrtick_timer); | |
342 | } | |
343 | ||
8f4d37ec PZ |
344 | /* |
345 | * High-resolution timer tick. | |
346 | * Runs from hardirq context with interrupts disabled. | |
347 | */ | |
348 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
349 | { | |
350 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
351 | ||
352 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
353 | ||
05fa785c | 354 | raw_spin_lock(&rq->lock); |
3e51f33f | 355 | update_rq_clock(rq); |
8f4d37ec | 356 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 357 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
358 | |
359 | return HRTIMER_NORESTART; | |
360 | } | |
361 | ||
95e904c7 | 362 | #ifdef CONFIG_SMP |
31656519 PZ |
363 | /* |
364 | * called from hardirq (IPI) context | |
365 | */ | |
366 | static void __hrtick_start(void *arg) | |
b328ca18 | 367 | { |
31656519 | 368 | struct rq *rq = arg; |
b328ca18 | 369 | |
05fa785c | 370 | raw_spin_lock(&rq->lock); |
31656519 PZ |
371 | hrtimer_restart(&rq->hrtick_timer); |
372 | rq->hrtick_csd_pending = 0; | |
05fa785c | 373 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
374 | } |
375 | ||
31656519 PZ |
376 | /* |
377 | * Called to set the hrtick timer state. | |
378 | * | |
379 | * called with rq->lock held and irqs disabled | |
380 | */ | |
029632fb | 381 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 382 | { |
31656519 PZ |
383 | struct hrtimer *timer = &rq->hrtick_timer; |
384 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 385 | |
cc584b21 | 386 | hrtimer_set_expires(timer, time); |
31656519 PZ |
387 | |
388 | if (rq == this_rq()) { | |
389 | hrtimer_restart(timer); | |
390 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 391 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
392 | rq->hrtick_csd_pending = 1; |
393 | } | |
b328ca18 PZ |
394 | } |
395 | ||
396 | static int | |
397 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
398 | { | |
399 | int cpu = (int)(long)hcpu; | |
400 | ||
401 | switch (action) { | |
402 | case CPU_UP_CANCELED: | |
403 | case CPU_UP_CANCELED_FROZEN: | |
404 | case CPU_DOWN_PREPARE: | |
405 | case CPU_DOWN_PREPARE_FROZEN: | |
406 | case CPU_DEAD: | |
407 | case CPU_DEAD_FROZEN: | |
31656519 | 408 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
409 | return NOTIFY_OK; |
410 | } | |
411 | ||
412 | return NOTIFY_DONE; | |
413 | } | |
414 | ||
fa748203 | 415 | static __init void init_hrtick(void) |
b328ca18 PZ |
416 | { |
417 | hotcpu_notifier(hotplug_hrtick, 0); | |
418 | } | |
31656519 PZ |
419 | #else |
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) |
31656519 | 426 | { |
7f1e2ca9 | 427 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 428 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 429 | } |
b328ca18 | 430 | |
006c75f1 | 431 | static inline void init_hrtick(void) |
8f4d37ec | 432 | { |
8f4d37ec | 433 | } |
31656519 | 434 | #endif /* CONFIG_SMP */ |
8f4d37ec | 435 | |
31656519 | 436 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 437 | { |
31656519 PZ |
438 | #ifdef CONFIG_SMP |
439 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 440 | |
31656519 PZ |
441 | rq->hrtick_csd.flags = 0; |
442 | rq->hrtick_csd.func = __hrtick_start; | |
443 | rq->hrtick_csd.info = rq; | |
444 | #endif | |
8f4d37ec | 445 | |
31656519 PZ |
446 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
447 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 448 | } |
006c75f1 | 449 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
450 | static inline void hrtick_clear(struct rq *rq) |
451 | { | |
452 | } | |
453 | ||
8f4d37ec PZ |
454 | static inline void init_rq_hrtick(struct rq *rq) |
455 | { | |
456 | } | |
457 | ||
b328ca18 PZ |
458 | static inline void init_hrtick(void) |
459 | { | |
460 | } | |
006c75f1 | 461 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 462 | |
c24d20db IM |
463 | /* |
464 | * resched_task - mark a task 'to be rescheduled now'. | |
465 | * | |
466 | * On UP this means the setting of the need_resched flag, on SMP it | |
467 | * might also involve a cross-CPU call to trigger the scheduler on | |
468 | * the target CPU. | |
469 | */ | |
470 | #ifdef CONFIG_SMP | |
471 | ||
472 | #ifndef tsk_is_polling | |
473 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
474 | #endif | |
475 | ||
029632fb | 476 | void resched_task(struct task_struct *p) |
c24d20db IM |
477 | { |
478 | int cpu; | |
479 | ||
05fa785c | 480 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 481 | |
5ed0cec0 | 482 | if (test_tsk_need_resched(p)) |
c24d20db IM |
483 | return; |
484 | ||
5ed0cec0 | 485 | set_tsk_need_resched(p); |
c24d20db IM |
486 | |
487 | cpu = task_cpu(p); | |
488 | if (cpu == smp_processor_id()) | |
489 | return; | |
490 | ||
491 | /* NEED_RESCHED must be visible before we test polling */ | |
492 | smp_mb(); | |
493 | if (!tsk_is_polling(p)) | |
494 | smp_send_reschedule(cpu); | |
495 | } | |
496 | ||
029632fb | 497 | void resched_cpu(int cpu) |
c24d20db IM |
498 | { |
499 | struct rq *rq = cpu_rq(cpu); | |
500 | unsigned long flags; | |
501 | ||
05fa785c | 502 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
503 | return; |
504 | resched_task(cpu_curr(cpu)); | |
05fa785c | 505 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 506 | } |
06d8308c TG |
507 | |
508 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
509 | /* |
510 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
511 | * from an idle cpu. This is good for power-savings. | |
512 | * | |
513 | * We don't do similar optimization for completely idle system, as | |
514 | * selecting an idle cpu will add more delays to the timers than intended | |
515 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
516 | */ | |
517 | int get_nohz_timer_target(void) | |
518 | { | |
519 | int cpu = smp_processor_id(); | |
520 | int i; | |
521 | struct sched_domain *sd; | |
522 | ||
057f3fad | 523 | rcu_read_lock(); |
83cd4fe2 | 524 | for_each_domain(cpu, sd) { |
057f3fad PZ |
525 | for_each_cpu(i, sched_domain_span(sd)) { |
526 | if (!idle_cpu(i)) { | |
527 | cpu = i; | |
528 | goto unlock; | |
529 | } | |
530 | } | |
83cd4fe2 | 531 | } |
057f3fad PZ |
532 | unlock: |
533 | rcu_read_unlock(); | |
83cd4fe2 VP |
534 | return cpu; |
535 | } | |
06d8308c TG |
536 | /* |
537 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
538 | * idle CPU then this timer might expire before the next timer event | |
539 | * which is scheduled to wake up that CPU. In case of a completely | |
540 | * idle system the next event might even be infinite time into the | |
541 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
542 | * leaves the inner idle loop so the newly added timer is taken into | |
543 | * account when the CPU goes back to idle and evaluates the timer | |
544 | * wheel for the next timer event. | |
545 | */ | |
546 | void wake_up_idle_cpu(int cpu) | |
547 | { | |
548 | struct rq *rq = cpu_rq(cpu); | |
549 | ||
550 | if (cpu == smp_processor_id()) | |
551 | return; | |
552 | ||
553 | /* | |
554 | * This is safe, as this function is called with the timer | |
555 | * wheel base lock of (cpu) held. When the CPU is on the way | |
556 | * to idle and has not yet set rq->curr to idle then it will | |
557 | * be serialized on the timer wheel base lock and take the new | |
558 | * timer into account automatically. | |
559 | */ | |
560 | if (rq->curr != rq->idle) | |
561 | return; | |
562 | ||
563 | /* | |
564 | * We can set TIF_RESCHED on the idle task of the other CPU | |
565 | * lockless. The worst case is that the other CPU runs the | |
566 | * idle task through an additional NOOP schedule() | |
567 | */ | |
5ed0cec0 | 568 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
569 | |
570 | /* NEED_RESCHED must be visible before we test polling */ | |
571 | smp_mb(); | |
572 | if (!tsk_is_polling(rq->idle)) | |
573 | smp_send_reschedule(cpu); | |
574 | } | |
39c0cbe2 | 575 | |
ca38062e SS |
576 | static inline bool got_nohz_idle_kick(void) |
577 | { | |
578 | return idle_cpu(smp_processor_id()) && this_rq()->nohz_balance_kick; | |
579 | } | |
580 | ||
581 | #else /* CONFIG_NO_HZ */ | |
582 | ||
583 | static inline bool got_nohz_idle_kick(void) | |
584 | { | |
585 | return false; | |
586 | } | |
587 | ||
6d6bc0ad | 588 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 589 | |
029632fb | 590 | void sched_avg_update(struct rq *rq) |
e9e9250b PZ |
591 | { |
592 | s64 period = sched_avg_period(); | |
593 | ||
594 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
595 | /* |
596 | * Inline assembly required to prevent the compiler | |
597 | * optimising this loop into a divmod call. | |
598 | * See __iter_div_u64_rem() for another example of this. | |
599 | */ | |
600 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
601 | rq->age_stamp += period; |
602 | rq->rt_avg /= 2; | |
603 | } | |
604 | } | |
605 | ||
6d6bc0ad | 606 | #else /* !CONFIG_SMP */ |
029632fb | 607 | void resched_task(struct task_struct *p) |
c24d20db | 608 | { |
05fa785c | 609 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 610 | set_tsk_need_resched(p); |
c24d20db | 611 | } |
6d6bc0ad | 612 | #endif /* CONFIG_SMP */ |
c24d20db | 613 | |
a790de99 PT |
614 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
615 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 616 | /* |
8277434e PT |
617 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
618 | * node and @up when leaving it for the final time. | |
619 | * | |
620 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 621 | */ |
029632fb | 622 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 623 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
624 | { |
625 | struct task_group *parent, *child; | |
eb755805 | 626 | int ret; |
c09595f6 | 627 | |
8277434e PT |
628 | parent = from; |
629 | ||
c09595f6 | 630 | down: |
eb755805 PZ |
631 | ret = (*down)(parent, data); |
632 | if (ret) | |
8277434e | 633 | goto out; |
c09595f6 PZ |
634 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
635 | parent = child; | |
636 | goto down; | |
637 | ||
638 | up: | |
029632fb PZ |
639 | continue; |
640 | } | |
641 | ret = (*up)(parent, data); | |
642 | if (ret || parent == from) | |
643 | goto out; | |
1e3c88bd | 644 | |
029632fb PZ |
645 | child = parent; |
646 | parent = parent->parent; | |
647 | if (parent) | |
648 | goto up; | |
649 | out: | |
650 | return ret; | |
9c217245 IM |
651 | } |
652 | ||
029632fb | 653 | int tg_nop(struct task_group *tg, void *data) |
9c217245 | 654 | { |
029632fb | 655 | return 0; |
9c217245 | 656 | } |
029632fb PZ |
657 | #endif |
658 | ||
659 | void update_cpu_load(struct rq *this_rq); | |
9c217245 | 660 | |
45bf76df IM |
661 | static void set_load_weight(struct task_struct *p) |
662 | { | |
f05998d4 NR |
663 | int prio = p->static_prio - MAX_RT_PRIO; |
664 | struct load_weight *load = &p->se.load; | |
665 | ||
dd41f596 IM |
666 | /* |
667 | * SCHED_IDLE tasks get minimal weight: | |
668 | */ | |
669 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 670 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 671 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
672 | return; |
673 | } | |
71f8bd46 | 674 | |
c8b28116 | 675 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 676 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
677 | } |
678 | ||
371fd7e7 | 679 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 680 | { |
a64692a3 | 681 | update_rq_clock(rq); |
dd41f596 | 682 | sched_info_queued(p); |
371fd7e7 | 683 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
684 | } |
685 | ||
371fd7e7 | 686 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 687 | { |
a64692a3 | 688 | update_rq_clock(rq); |
46ac22ba | 689 | sched_info_dequeued(p); |
371fd7e7 | 690 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
691 | } |
692 | ||
1e3c88bd PZ |
693 | /* |
694 | * activate_task - move a task to the runqueue. | |
695 | */ | |
029632fb | 696 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
697 | { |
698 | if (task_contributes_to_load(p)) | |
699 | rq->nr_uninterruptible--; | |
700 | ||
371fd7e7 | 701 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
702 | } |
703 | ||
704 | /* | |
705 | * deactivate_task - remove a task from the runqueue. | |
706 | */ | |
029632fb | 707 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
708 | { |
709 | if (task_contributes_to_load(p)) | |
710 | rq->nr_uninterruptible++; | |
711 | ||
371fd7e7 | 712 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
713 | } |
714 | ||
b52bfee4 VP |
715 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
716 | ||
305e6835 VP |
717 | /* |
718 | * There are no locks covering percpu hardirq/softirq time. | |
719 | * They are only modified in account_system_vtime, on corresponding CPU | |
720 | * with interrupts disabled. So, writes are safe. | |
721 | * They are read and saved off onto struct rq in update_rq_clock(). | |
722 | * This may result in other CPU reading this CPU's irq time and can | |
723 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
724 | * or new value with a side effect of accounting a slice of irq time to wrong |
725 | * task when irq is in progress while we read rq->clock. That is a worthy | |
726 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 727 | */ |
b52bfee4 VP |
728 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
729 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
730 | ||
731 | static DEFINE_PER_CPU(u64, irq_start_time); | |
732 | static int sched_clock_irqtime; | |
733 | ||
734 | void enable_sched_clock_irqtime(void) | |
735 | { | |
736 | sched_clock_irqtime = 1; | |
737 | } | |
738 | ||
739 | void disable_sched_clock_irqtime(void) | |
740 | { | |
741 | sched_clock_irqtime = 0; | |
742 | } | |
743 | ||
8e92c201 PZ |
744 | #ifndef CONFIG_64BIT |
745 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
746 | ||
747 | static inline void irq_time_write_begin(void) | |
748 | { | |
749 | __this_cpu_inc(irq_time_seq.sequence); | |
750 | smp_wmb(); | |
751 | } | |
752 | ||
753 | static inline void irq_time_write_end(void) | |
754 | { | |
755 | smp_wmb(); | |
756 | __this_cpu_inc(irq_time_seq.sequence); | |
757 | } | |
758 | ||
759 | static inline u64 irq_time_read(int cpu) | |
760 | { | |
761 | u64 irq_time; | |
762 | unsigned seq; | |
763 | ||
764 | do { | |
765 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
766 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
767 | per_cpu(cpu_hardirq_time, cpu); | |
768 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
769 | ||
770 | return irq_time; | |
771 | } | |
772 | #else /* CONFIG_64BIT */ | |
773 | static inline void irq_time_write_begin(void) | |
774 | { | |
775 | } | |
776 | ||
777 | static inline void irq_time_write_end(void) | |
778 | { | |
779 | } | |
780 | ||
781 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 782 | { |
305e6835 VP |
783 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
784 | } | |
8e92c201 | 785 | #endif /* CONFIG_64BIT */ |
305e6835 | 786 | |
fe44d621 PZ |
787 | /* |
788 | * Called before incrementing preempt_count on {soft,}irq_enter | |
789 | * and before decrementing preempt_count on {soft,}irq_exit. | |
790 | */ | |
b52bfee4 VP |
791 | void account_system_vtime(struct task_struct *curr) |
792 | { | |
793 | unsigned long flags; | |
fe44d621 | 794 | s64 delta; |
b52bfee4 | 795 | int cpu; |
b52bfee4 VP |
796 | |
797 | if (!sched_clock_irqtime) | |
798 | return; | |
799 | ||
800 | local_irq_save(flags); | |
801 | ||
b52bfee4 | 802 | cpu = smp_processor_id(); |
fe44d621 PZ |
803 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
804 | __this_cpu_add(irq_start_time, delta); | |
805 | ||
8e92c201 | 806 | irq_time_write_begin(); |
b52bfee4 VP |
807 | /* |
808 | * We do not account for softirq time from ksoftirqd here. | |
809 | * We want to continue accounting softirq time to ksoftirqd thread | |
810 | * in that case, so as not to confuse scheduler with a special task | |
811 | * that do not consume any time, but still wants to run. | |
812 | */ | |
813 | if (hardirq_count()) | |
fe44d621 | 814 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 815 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 816 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 817 | |
8e92c201 | 818 | irq_time_write_end(); |
b52bfee4 VP |
819 | local_irq_restore(flags); |
820 | } | |
b7dadc38 | 821 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 822 | |
e6e6685a GC |
823 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
824 | ||
825 | #ifdef CONFIG_PARAVIRT | |
826 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 827 | { |
e6e6685a GC |
828 | if (unlikely(steal > NSEC_PER_SEC)) |
829 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 830 | |
e6e6685a GC |
831 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
832 | } | |
833 | #endif | |
834 | ||
fe44d621 | 835 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 836 | { |
095c0aa8 GC |
837 | /* |
838 | * In theory, the compile should just see 0 here, and optimize out the call | |
839 | * to sched_rt_avg_update. But I don't trust it... | |
840 | */ | |
841 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
842 | s64 steal = 0, irq_delta = 0; | |
843 | #endif | |
844 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 845 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
846 | |
847 | /* | |
848 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
849 | * this case when a previous update_rq_clock() happened inside a | |
850 | * {soft,}irq region. | |
851 | * | |
852 | * When this happens, we stop ->clock_task and only update the | |
853 | * prev_irq_time stamp to account for the part that fit, so that a next | |
854 | * update will consume the rest. This ensures ->clock_task is | |
855 | * monotonic. | |
856 | * | |
857 | * It does however cause some slight miss-attribution of {soft,}irq | |
858 | * time, a more accurate solution would be to update the irq_time using | |
859 | * the current rq->clock timestamp, except that would require using | |
860 | * atomic ops. | |
861 | */ | |
862 | if (irq_delta > delta) | |
863 | irq_delta = delta; | |
864 | ||
865 | rq->prev_irq_time += irq_delta; | |
866 | delta -= irq_delta; | |
095c0aa8 GC |
867 | #endif |
868 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
869 | if (static_branch((¶virt_steal_rq_enabled))) { | |
870 | u64 st; | |
871 | ||
872 | steal = paravirt_steal_clock(cpu_of(rq)); | |
873 | steal -= rq->prev_steal_time_rq; | |
874 | ||
875 | if (unlikely(steal > delta)) | |
876 | steal = delta; | |
877 | ||
878 | st = steal_ticks(steal); | |
879 | steal = st * TICK_NSEC; | |
880 | ||
881 | rq->prev_steal_time_rq += steal; | |
882 | ||
883 | delta -= steal; | |
884 | } | |
885 | #endif | |
886 | ||
fe44d621 PZ |
887 | rq->clock_task += delta; |
888 | ||
095c0aa8 GC |
889 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
890 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
891 | sched_rt_avg_update(rq, irq_delta + steal); | |
892 | #endif | |
aa483808 VP |
893 | } |
894 | ||
095c0aa8 | 895 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
896 | static int irqtime_account_hi_update(void) |
897 | { | |
898 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
899 | unsigned long flags; | |
900 | u64 latest_ns; | |
901 | int ret = 0; | |
902 | ||
903 | local_irq_save(flags); | |
904 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
905 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
906 | ret = 1; | |
907 | local_irq_restore(flags); | |
908 | return ret; | |
909 | } | |
910 | ||
911 | static int irqtime_account_si_update(void) | |
912 | { | |
913 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
914 | unsigned long flags; | |
915 | u64 latest_ns; | |
916 | int ret = 0; | |
917 | ||
918 | local_irq_save(flags); | |
919 | latest_ns = this_cpu_read(cpu_softirq_time); | |
920 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
921 | ret = 1; | |
922 | local_irq_restore(flags); | |
923 | return ret; | |
924 | } | |
925 | ||
fe44d621 | 926 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 927 | |
abb74cef VP |
928 | #define sched_clock_irqtime (0) |
929 | ||
095c0aa8 | 930 | #endif |
b52bfee4 | 931 | |
34f971f6 PZ |
932 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
933 | { | |
934 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
935 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
936 | ||
937 | if (stop) { | |
938 | /* | |
939 | * Make it appear like a SCHED_FIFO task, its something | |
940 | * userspace knows about and won't get confused about. | |
941 | * | |
942 | * Also, it will make PI more or less work without too | |
943 | * much confusion -- but then, stop work should not | |
944 | * rely on PI working anyway. | |
945 | */ | |
946 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
947 | ||
948 | stop->sched_class = &stop_sched_class; | |
949 | } | |
950 | ||
951 | cpu_rq(cpu)->stop = stop; | |
952 | ||
953 | if (old_stop) { | |
954 | /* | |
955 | * Reset it back to a normal scheduling class so that | |
956 | * it can die in pieces. | |
957 | */ | |
958 | old_stop->sched_class = &rt_sched_class; | |
959 | } | |
960 | } | |
961 | ||
14531189 | 962 | /* |
dd41f596 | 963 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 964 | */ |
14531189 IM |
965 | static inline int __normal_prio(struct task_struct *p) |
966 | { | |
dd41f596 | 967 | return p->static_prio; |
14531189 IM |
968 | } |
969 | ||
b29739f9 IM |
970 | /* |
971 | * Calculate the expected normal priority: i.e. priority | |
972 | * without taking RT-inheritance into account. Might be | |
973 | * boosted by interactivity modifiers. Changes upon fork, | |
974 | * setprio syscalls, and whenever the interactivity | |
975 | * estimator recalculates. | |
976 | */ | |
36c8b586 | 977 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
978 | { |
979 | int prio; | |
980 | ||
e05606d3 | 981 | if (task_has_rt_policy(p)) |
b29739f9 IM |
982 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
983 | else | |
984 | prio = __normal_prio(p); | |
985 | return prio; | |
986 | } | |
987 | ||
988 | /* | |
989 | * Calculate the current priority, i.e. the priority | |
990 | * taken into account by the scheduler. This value might | |
991 | * be boosted by RT tasks, or might be boosted by | |
992 | * interactivity modifiers. Will be RT if the task got | |
993 | * RT-boosted. If not then it returns p->normal_prio. | |
994 | */ | |
36c8b586 | 995 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
996 | { |
997 | p->normal_prio = normal_prio(p); | |
998 | /* | |
999 | * If we are RT tasks or we were boosted to RT priority, | |
1000 | * keep the priority unchanged. Otherwise, update priority | |
1001 | * to the normal priority: | |
1002 | */ | |
1003 | if (!rt_prio(p->prio)) | |
1004 | return p->normal_prio; | |
1005 | return p->prio; | |
1006 | } | |
1007 | ||
1da177e4 LT |
1008 | /** |
1009 | * task_curr - is this task currently executing on a CPU? | |
1010 | * @p: the task in question. | |
1011 | */ | |
36c8b586 | 1012 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1013 | { |
1014 | return cpu_curr(task_cpu(p)) == p; | |
1015 | } | |
1016 | ||
cb469845 SR |
1017 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1018 | const struct sched_class *prev_class, | |
da7a735e | 1019 | int oldprio) |
cb469845 SR |
1020 | { |
1021 | if (prev_class != p->sched_class) { | |
1022 | if (prev_class->switched_from) | |
da7a735e PZ |
1023 | prev_class->switched_from(rq, p); |
1024 | p->sched_class->switched_to(rq, p); | |
1025 | } else if (oldprio != p->prio) | |
1026 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
1027 | } |
1028 | ||
029632fb | 1029 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1030 | { |
1031 | const struct sched_class *class; | |
1032 | ||
1033 | if (p->sched_class == rq->curr->sched_class) { | |
1034 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1035 | } else { | |
1036 | for_each_class(class) { | |
1037 | if (class == rq->curr->sched_class) | |
1038 | break; | |
1039 | if (class == p->sched_class) { | |
1040 | resched_task(rq->curr); | |
1041 | break; | |
1042 | } | |
1043 | } | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * A queue event has occurred, and we're going to schedule. In | |
1048 | * this case, we can save a useless back to back clock update. | |
1049 | */ | |
fd2f4419 | 1050 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1051 | rq->skip_clock_update = 1; |
1052 | } | |
1053 | ||
1da177e4 | 1054 | #ifdef CONFIG_SMP |
dd41f596 | 1055 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1056 | { |
e2912009 PZ |
1057 | #ifdef CONFIG_SCHED_DEBUG |
1058 | /* | |
1059 | * We should never call set_task_cpu() on a blocked task, | |
1060 | * ttwu() will sort out the placement. | |
1061 | */ | |
077614ee PZ |
1062 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
1063 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
1064 | |
1065 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1066 | /* |
1067 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1068 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1069 | * | |
1070 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
1071 | * see set_task_rq(). | |
1072 | * | |
1073 | * Furthermore, all task_rq users should acquire both locks, see | |
1074 | * task_rq_lock(). | |
1075 | */ | |
0122ec5b PZ |
1076 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1077 | lockdep_is_held(&task_rq(p)->lock))); | |
1078 | #endif | |
e2912009 PZ |
1079 | #endif |
1080 | ||
de1d7286 | 1081 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1082 | |
0c69774e PZ |
1083 | if (task_cpu(p) != new_cpu) { |
1084 | p->se.nr_migrations++; | |
a8b0ca17 | 1085 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1086 | } |
dd41f596 IM |
1087 | |
1088 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1089 | } |
1090 | ||
969c7921 | 1091 | struct migration_arg { |
36c8b586 | 1092 | struct task_struct *task; |
1da177e4 | 1093 | int dest_cpu; |
70b97a7f | 1094 | }; |
1da177e4 | 1095 | |
969c7921 TH |
1096 | static int migration_cpu_stop(void *data); |
1097 | ||
1da177e4 LT |
1098 | /* |
1099 | * wait_task_inactive - wait for a thread to unschedule. | |
1100 | * | |
85ba2d86 RM |
1101 | * If @match_state is nonzero, it's the @p->state value just checked and |
1102 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1103 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1104 | * we return a positive number (its total switch count). If a second call | |
1105 | * a short while later returns the same number, the caller can be sure that | |
1106 | * @p has remained unscheduled the whole time. | |
1107 | * | |
1da177e4 LT |
1108 | * The caller must ensure that the task *will* unschedule sometime soon, |
1109 | * else this function might spin for a *long* time. This function can't | |
1110 | * be called with interrupts off, or it may introduce deadlock with | |
1111 | * smp_call_function() if an IPI is sent by the same process we are | |
1112 | * waiting to become inactive. | |
1113 | */ | |
85ba2d86 | 1114 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1115 | { |
1116 | unsigned long flags; | |
dd41f596 | 1117 | int running, on_rq; |
85ba2d86 | 1118 | unsigned long ncsw; |
70b97a7f | 1119 | struct rq *rq; |
1da177e4 | 1120 | |
3a5c359a AK |
1121 | for (;;) { |
1122 | /* | |
1123 | * We do the initial early heuristics without holding | |
1124 | * any task-queue locks at all. We'll only try to get | |
1125 | * the runqueue lock when things look like they will | |
1126 | * work out! | |
1127 | */ | |
1128 | rq = task_rq(p); | |
fa490cfd | 1129 | |
3a5c359a AK |
1130 | /* |
1131 | * If the task is actively running on another CPU | |
1132 | * still, just relax and busy-wait without holding | |
1133 | * any locks. | |
1134 | * | |
1135 | * NOTE! Since we don't hold any locks, it's not | |
1136 | * even sure that "rq" stays as the right runqueue! | |
1137 | * But we don't care, since "task_running()" will | |
1138 | * return false if the runqueue has changed and p | |
1139 | * is actually now running somewhere else! | |
1140 | */ | |
85ba2d86 RM |
1141 | while (task_running(rq, p)) { |
1142 | if (match_state && unlikely(p->state != match_state)) | |
1143 | return 0; | |
3a5c359a | 1144 | cpu_relax(); |
85ba2d86 | 1145 | } |
fa490cfd | 1146 | |
3a5c359a AK |
1147 | /* |
1148 | * Ok, time to look more closely! We need the rq | |
1149 | * lock now, to be *sure*. If we're wrong, we'll | |
1150 | * just go back and repeat. | |
1151 | */ | |
1152 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1153 | trace_sched_wait_task(p); |
3a5c359a | 1154 | running = task_running(rq, p); |
fd2f4419 | 1155 | on_rq = p->on_rq; |
85ba2d86 | 1156 | ncsw = 0; |
f31e11d8 | 1157 | if (!match_state || p->state == match_state) |
93dcf55f | 1158 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1159 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1160 | |
85ba2d86 RM |
1161 | /* |
1162 | * If it changed from the expected state, bail out now. | |
1163 | */ | |
1164 | if (unlikely(!ncsw)) | |
1165 | break; | |
1166 | ||
3a5c359a AK |
1167 | /* |
1168 | * Was it really running after all now that we | |
1169 | * checked with the proper locks actually held? | |
1170 | * | |
1171 | * Oops. Go back and try again.. | |
1172 | */ | |
1173 | if (unlikely(running)) { | |
1174 | cpu_relax(); | |
1175 | continue; | |
1176 | } | |
fa490cfd | 1177 | |
3a5c359a AK |
1178 | /* |
1179 | * It's not enough that it's not actively running, | |
1180 | * it must be off the runqueue _entirely_, and not | |
1181 | * preempted! | |
1182 | * | |
80dd99b3 | 1183 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1184 | * running right now), it's preempted, and we should |
1185 | * yield - it could be a while. | |
1186 | */ | |
1187 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1188 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1189 | ||
1190 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1191 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1192 | continue; |
1193 | } | |
fa490cfd | 1194 | |
3a5c359a AK |
1195 | /* |
1196 | * Ahh, all good. It wasn't running, and it wasn't | |
1197 | * runnable, which means that it will never become | |
1198 | * running in the future either. We're all done! | |
1199 | */ | |
1200 | break; | |
1201 | } | |
85ba2d86 RM |
1202 | |
1203 | return ncsw; | |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | /*** | |
1207 | * kick_process - kick a running thread to enter/exit the kernel | |
1208 | * @p: the to-be-kicked thread | |
1209 | * | |
1210 | * Cause a process which is running on another CPU to enter | |
1211 | * kernel-mode, without any delay. (to get signals handled.) | |
1212 | * | |
25985edc | 1213 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1214 | * because all it wants to ensure is that the remote task enters |
1215 | * the kernel. If the IPI races and the task has been migrated | |
1216 | * to another CPU then no harm is done and the purpose has been | |
1217 | * achieved as well. | |
1218 | */ | |
36c8b586 | 1219 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1220 | { |
1221 | int cpu; | |
1222 | ||
1223 | preempt_disable(); | |
1224 | cpu = task_cpu(p); | |
1225 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1226 | smp_send_reschedule(cpu); | |
1227 | preempt_enable(); | |
1228 | } | |
b43e3521 | 1229 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1230 | #endif /* CONFIG_SMP */ |
1da177e4 | 1231 | |
970b13ba | 1232 | #ifdef CONFIG_SMP |
30da688e | 1233 | /* |
013fdb80 | 1234 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1235 | */ |
5da9a0fb PZ |
1236 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1237 | { | |
1238 | int dest_cpu; | |
1239 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
1240 | ||
1241 | /* Look for allowed, online CPU in same node. */ | |
1242 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
fa17b507 | 1243 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb PZ |
1244 | return dest_cpu; |
1245 | ||
1246 | /* Any allowed, online CPU? */ | |
fa17b507 | 1247 | dest_cpu = cpumask_any_and(tsk_cpus_allowed(p), cpu_active_mask); |
5da9a0fb PZ |
1248 | if (dest_cpu < nr_cpu_ids) |
1249 | return dest_cpu; | |
1250 | ||
1251 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
1252 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
1253 | /* | |
1254 | * Don't tell them about moving exiting tasks or | |
1255 | * kernel threads (both mm NULL), since they never | |
1256 | * leave kernel. | |
1257 | */ | |
1258 | if (p->mm && printk_ratelimit()) { | |
1259 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
1260 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
1261 | } |
1262 | ||
1263 | return dest_cpu; | |
1264 | } | |
1265 | ||
e2912009 | 1266 | /* |
013fdb80 | 1267 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1268 | */ |
970b13ba | 1269 | static inline |
7608dec2 | 1270 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1271 | { |
7608dec2 | 1272 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1273 | |
1274 | /* | |
1275 | * In order not to call set_task_cpu() on a blocking task we need | |
1276 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1277 | * cpu. | |
1278 | * | |
1279 | * Since this is common to all placement strategies, this lives here. | |
1280 | * | |
1281 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1282 | * not worry about this generic constraint ] | |
1283 | */ | |
fa17b507 | 1284 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1285 | !cpu_online(cpu))) |
5da9a0fb | 1286 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1287 | |
1288 | return cpu; | |
970b13ba | 1289 | } |
09a40af5 MG |
1290 | |
1291 | static void update_avg(u64 *avg, u64 sample) | |
1292 | { | |
1293 | s64 diff = sample - *avg; | |
1294 | *avg += diff >> 3; | |
1295 | } | |
970b13ba PZ |
1296 | #endif |
1297 | ||
d7c01d27 | 1298 | static void |
b84cb5df | 1299 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1300 | { |
d7c01d27 | 1301 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1302 | struct rq *rq = this_rq(); |
1303 | ||
d7c01d27 PZ |
1304 | #ifdef CONFIG_SMP |
1305 | int this_cpu = smp_processor_id(); | |
1306 | ||
1307 | if (cpu == this_cpu) { | |
1308 | schedstat_inc(rq, ttwu_local); | |
1309 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1310 | } else { | |
1311 | struct sched_domain *sd; | |
1312 | ||
1313 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1314 | rcu_read_lock(); |
d7c01d27 PZ |
1315 | for_each_domain(this_cpu, sd) { |
1316 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1317 | schedstat_inc(sd, ttwu_wake_remote); | |
1318 | break; | |
1319 | } | |
1320 | } | |
057f3fad | 1321 | rcu_read_unlock(); |
d7c01d27 | 1322 | } |
f339b9dc PZ |
1323 | |
1324 | if (wake_flags & WF_MIGRATED) | |
1325 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1326 | ||
d7c01d27 PZ |
1327 | #endif /* CONFIG_SMP */ |
1328 | ||
1329 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1330 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1331 | |
1332 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1333 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1334 | |
d7c01d27 PZ |
1335 | #endif /* CONFIG_SCHEDSTATS */ |
1336 | } | |
1337 | ||
1338 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1339 | { | |
9ed3811a | 1340 | activate_task(rq, p, en_flags); |
fd2f4419 | 1341 | p->on_rq = 1; |
c2f7115e PZ |
1342 | |
1343 | /* if a worker is waking up, notify workqueue */ | |
1344 | if (p->flags & PF_WQ_WORKER) | |
1345 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1346 | } |
1347 | ||
23f41eeb PZ |
1348 | /* |
1349 | * Mark the task runnable and perform wakeup-preemption. | |
1350 | */ | |
89363381 | 1351 | static void |
23f41eeb | 1352 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1353 | { |
89363381 | 1354 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1355 | check_preempt_curr(rq, p, wake_flags); |
1356 | ||
1357 | p->state = TASK_RUNNING; | |
1358 | #ifdef CONFIG_SMP | |
1359 | if (p->sched_class->task_woken) | |
1360 | p->sched_class->task_woken(rq, p); | |
1361 | ||
e69c6341 | 1362 | if (rq->idle_stamp) { |
9ed3811a TH |
1363 | u64 delta = rq->clock - rq->idle_stamp; |
1364 | u64 max = 2*sysctl_sched_migration_cost; | |
1365 | ||
1366 | if (delta > max) | |
1367 | rq->avg_idle = max; | |
1368 | else | |
1369 | update_avg(&rq->avg_idle, delta); | |
1370 | rq->idle_stamp = 0; | |
1371 | } | |
1372 | #endif | |
1373 | } | |
1374 | ||
c05fbafb PZ |
1375 | static void |
1376 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1377 | { | |
1378 | #ifdef CONFIG_SMP | |
1379 | if (p->sched_contributes_to_load) | |
1380 | rq->nr_uninterruptible--; | |
1381 | #endif | |
1382 | ||
1383 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1384 | ttwu_do_wakeup(rq, p, wake_flags); | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1389 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1390 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1391 | * the task is still ->on_rq. | |
1392 | */ | |
1393 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1394 | { | |
1395 | struct rq *rq; | |
1396 | int ret = 0; | |
1397 | ||
1398 | rq = __task_rq_lock(p); | |
1399 | if (p->on_rq) { | |
1400 | ttwu_do_wakeup(rq, p, wake_flags); | |
1401 | ret = 1; | |
1402 | } | |
1403 | __task_rq_unlock(rq); | |
1404 | ||
1405 | return ret; | |
1406 | } | |
1407 | ||
317f3941 | 1408 | #ifdef CONFIG_SMP |
fa14ff4a | 1409 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1410 | { |
1411 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1412 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1413 | struct task_struct *p; | |
317f3941 PZ |
1414 | |
1415 | raw_spin_lock(&rq->lock); | |
1416 | ||
fa14ff4a PZ |
1417 | while (llist) { |
1418 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1419 | llist = llist_next(llist); | |
317f3941 PZ |
1420 | ttwu_do_activate(rq, p, 0); |
1421 | } | |
1422 | ||
1423 | raw_spin_unlock(&rq->lock); | |
1424 | } | |
1425 | ||
1426 | void scheduler_ipi(void) | |
1427 | { | |
ca38062e | 1428 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1429 | return; |
1430 | ||
1431 | /* | |
1432 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1433 | * traditionally all their work was done from the interrupt return | |
1434 | * path. Now that we actually do some work, we need to make sure | |
1435 | * we do call them. | |
1436 | * | |
1437 | * Some archs already do call them, luckily irq_enter/exit nest | |
1438 | * properly. | |
1439 | * | |
1440 | * Arguably we should visit all archs and update all handlers, | |
1441 | * however a fair share of IPIs are still resched only so this would | |
1442 | * somewhat pessimize the simple resched case. | |
1443 | */ | |
1444 | irq_enter(); | |
fa14ff4a | 1445 | sched_ttwu_pending(); |
ca38062e SS |
1446 | |
1447 | /* | |
1448 | * Check if someone kicked us for doing the nohz idle load balance. | |
1449 | */ | |
6eb57e0d SS |
1450 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1451 | this_rq()->idle_balance = 1; | |
ca38062e | 1452 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1453 | } |
c5d753a5 | 1454 | irq_exit(); |
317f3941 PZ |
1455 | } |
1456 | ||
1457 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1458 | { | |
fa14ff4a | 1459 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1460 | smp_send_reschedule(cpu); |
1461 | } | |
d6aa8f85 PZ |
1462 | |
1463 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1464 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
1465 | { | |
1466 | struct rq *rq; | |
1467 | int ret = 0; | |
1468 | ||
1469 | rq = __task_rq_lock(p); | |
1470 | if (p->on_cpu) { | |
1471 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
1472 | ttwu_do_wakeup(rq, p, wake_flags); | |
1473 | ret = 1; | |
1474 | } | |
1475 | __task_rq_unlock(rq); | |
1476 | ||
1477 | return ret; | |
1478 | ||
1479 | } | |
1480 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
1481 | #endif /* CONFIG_SMP */ | |
317f3941 | 1482 | |
c05fbafb PZ |
1483 | static void ttwu_queue(struct task_struct *p, int cpu) |
1484 | { | |
1485 | struct rq *rq = cpu_rq(cpu); | |
1486 | ||
17d9f311 | 1487 | #if defined(CONFIG_SMP) |
317f3941 | 1488 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { |
f01114cb | 1489 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1490 | ttwu_queue_remote(p, cpu); |
1491 | return; | |
1492 | } | |
1493 | #endif | |
1494 | ||
c05fbafb PZ |
1495 | raw_spin_lock(&rq->lock); |
1496 | ttwu_do_activate(rq, p, 0); | |
1497 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1498 | } |
1499 | ||
1500 | /** | |
1da177e4 | 1501 | * try_to_wake_up - wake up a thread |
9ed3811a | 1502 | * @p: the thread to be awakened |
1da177e4 | 1503 | * @state: the mask of task states that can be woken |
9ed3811a | 1504 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1505 | * |
1506 | * Put it on the run-queue if it's not already there. The "current" | |
1507 | * thread is always on the run-queue (except when the actual | |
1508 | * re-schedule is in progress), and as such you're allowed to do | |
1509 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1510 | * runnable without the overhead of this. | |
1511 | * | |
9ed3811a TH |
1512 | * Returns %true if @p was woken up, %false if it was already running |
1513 | * or @state didn't match @p's state. | |
1da177e4 | 1514 | */ |
e4a52bcb PZ |
1515 | static int |
1516 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1517 | { |
1da177e4 | 1518 | unsigned long flags; |
c05fbafb | 1519 | int cpu, success = 0; |
2398f2c6 | 1520 | |
04e2f174 | 1521 | smp_wmb(); |
013fdb80 | 1522 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1523 | if (!(p->state & state)) |
1da177e4 LT |
1524 | goto out; |
1525 | ||
c05fbafb | 1526 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1527 | cpu = task_cpu(p); |
1da177e4 | 1528 | |
c05fbafb PZ |
1529 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1530 | goto stat; | |
1da177e4 | 1531 | |
1da177e4 | 1532 | #ifdef CONFIG_SMP |
e9c84311 | 1533 | /* |
c05fbafb PZ |
1534 | * If the owning (remote) cpu is still in the middle of schedule() with |
1535 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1536 | */ |
e4a52bcb PZ |
1537 | while (p->on_cpu) { |
1538 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1539 | /* | |
d6aa8f85 PZ |
1540 | * In case the architecture enables interrupts in |
1541 | * context_switch(), we cannot busy wait, since that | |
1542 | * would lead to deadlocks when an interrupt hits and | |
1543 | * tries to wake up @prev. So bail and do a complete | |
1544 | * remote wakeup. | |
e4a52bcb | 1545 | */ |
d6aa8f85 | 1546 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 1547 | goto stat; |
d6aa8f85 | 1548 | #else |
e4a52bcb | 1549 | cpu_relax(); |
d6aa8f85 | 1550 | #endif |
371fd7e7 | 1551 | } |
0970d299 | 1552 | /* |
e4a52bcb | 1553 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1554 | */ |
e4a52bcb | 1555 | smp_rmb(); |
1da177e4 | 1556 | |
a8e4f2ea | 1557 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1558 | p->state = TASK_WAKING; |
e7693a36 | 1559 | |
e4a52bcb | 1560 | if (p->sched_class->task_waking) |
74f8e4b2 | 1561 | p->sched_class->task_waking(p); |
efbbd05a | 1562 | |
7608dec2 | 1563 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1564 | if (task_cpu(p) != cpu) { |
1565 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1566 | set_task_cpu(p, cpu); |
f339b9dc | 1567 | } |
1da177e4 | 1568 | #endif /* CONFIG_SMP */ |
1da177e4 | 1569 | |
c05fbafb PZ |
1570 | ttwu_queue(p, cpu); |
1571 | stat: | |
b84cb5df | 1572 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1573 | out: |
013fdb80 | 1574 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1575 | |
1576 | return success; | |
1577 | } | |
1578 | ||
21aa9af0 TH |
1579 | /** |
1580 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1581 | * @p: the thread to be awakened | |
1582 | * | |
2acca55e | 1583 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1584 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1585 | * the current task. |
21aa9af0 TH |
1586 | */ |
1587 | static void try_to_wake_up_local(struct task_struct *p) | |
1588 | { | |
1589 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1590 | |
1591 | BUG_ON(rq != this_rq()); | |
1592 | BUG_ON(p == current); | |
1593 | lockdep_assert_held(&rq->lock); | |
1594 | ||
2acca55e PZ |
1595 | if (!raw_spin_trylock(&p->pi_lock)) { |
1596 | raw_spin_unlock(&rq->lock); | |
1597 | raw_spin_lock(&p->pi_lock); | |
1598 | raw_spin_lock(&rq->lock); | |
1599 | } | |
1600 | ||
21aa9af0 | 1601 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1602 | goto out; |
21aa9af0 | 1603 | |
fd2f4419 | 1604 | if (!p->on_rq) |
d7c01d27 PZ |
1605 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1606 | ||
23f41eeb | 1607 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1608 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1609 | out: |
1610 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1611 | } |
1612 | ||
50fa610a DH |
1613 | /** |
1614 | * wake_up_process - Wake up a specific process | |
1615 | * @p: The process to be woken up. | |
1616 | * | |
1617 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1618 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1619 | * running. | |
1620 | * | |
1621 | * It may be assumed that this function implies a write memory barrier before | |
1622 | * changing the task state if and only if any tasks are woken up. | |
1623 | */ | |
7ad5b3a5 | 1624 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1625 | { |
d9514f6c | 1626 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1627 | } |
1da177e4 LT |
1628 | EXPORT_SYMBOL(wake_up_process); |
1629 | ||
7ad5b3a5 | 1630 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1631 | { |
1632 | return try_to_wake_up(p, state, 0); | |
1633 | } | |
1634 | ||
1da177e4 LT |
1635 | /* |
1636 | * Perform scheduler related setup for a newly forked process p. | |
1637 | * p is forked by current. | |
dd41f596 IM |
1638 | * |
1639 | * __sched_fork() is basic setup used by init_idle() too: | |
1640 | */ | |
1641 | static void __sched_fork(struct task_struct *p) | |
1642 | { | |
fd2f4419 PZ |
1643 | p->on_rq = 0; |
1644 | ||
1645 | p->se.on_rq = 0; | |
dd41f596 IM |
1646 | p->se.exec_start = 0; |
1647 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1648 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1649 | p->se.nr_migrations = 0; |
da7a735e | 1650 | p->se.vruntime = 0; |
fd2f4419 | 1651 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1652 | |
1653 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1654 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1655 | #endif |
476d139c | 1656 | |
fa717060 | 1657 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1658 | |
e107be36 AK |
1659 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1660 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1661 | #endif | |
dd41f596 IM |
1662 | } |
1663 | ||
1664 | /* | |
1665 | * fork()/clone()-time setup: | |
1666 | */ | |
3e51e3ed | 1667 | void sched_fork(struct task_struct *p) |
dd41f596 | 1668 | { |
0122ec5b | 1669 | unsigned long flags; |
dd41f596 IM |
1670 | int cpu = get_cpu(); |
1671 | ||
1672 | __sched_fork(p); | |
06b83b5f | 1673 | /* |
0017d735 | 1674 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1675 | * nobody will actually run it, and a signal or other external |
1676 | * event cannot wake it up and insert it on the runqueue either. | |
1677 | */ | |
0017d735 | 1678 | p->state = TASK_RUNNING; |
dd41f596 | 1679 | |
c350a04e MG |
1680 | /* |
1681 | * Make sure we do not leak PI boosting priority to the child. | |
1682 | */ | |
1683 | p->prio = current->normal_prio; | |
1684 | ||
b9dc29e7 MG |
1685 | /* |
1686 | * Revert to default priority/policy on fork if requested. | |
1687 | */ | |
1688 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1689 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1690 | p->policy = SCHED_NORMAL; |
6c697bdf | 1691 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1692 | p->rt_priority = 0; |
1693 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1694 | p->static_prio = NICE_TO_PRIO(0); | |
1695 | ||
1696 | p->prio = p->normal_prio = __normal_prio(p); | |
1697 | set_load_weight(p); | |
6c697bdf | 1698 | |
b9dc29e7 MG |
1699 | /* |
1700 | * We don't need the reset flag anymore after the fork. It has | |
1701 | * fulfilled its duty: | |
1702 | */ | |
1703 | p->sched_reset_on_fork = 0; | |
1704 | } | |
ca94c442 | 1705 | |
2ddbf952 HS |
1706 | if (!rt_prio(p->prio)) |
1707 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1708 | |
cd29fe6f PZ |
1709 | if (p->sched_class->task_fork) |
1710 | p->sched_class->task_fork(p); | |
1711 | ||
86951599 PZ |
1712 | /* |
1713 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1714 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1715 | * is ran before sched_fork(). | |
1716 | * | |
1717 | * Silence PROVE_RCU. | |
1718 | */ | |
0122ec5b | 1719 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1720 | set_task_cpu(p, cpu); |
0122ec5b | 1721 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1722 | |
52f17b6c | 1723 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1724 | if (likely(sched_info_on())) |
52f17b6c | 1725 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1726 | #endif |
3ca7a440 PZ |
1727 | #if defined(CONFIG_SMP) |
1728 | p->on_cpu = 0; | |
4866cde0 | 1729 | #endif |
bdd4e85d | 1730 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1731 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1732 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1733 | #endif |
806c09a7 | 1734 | #ifdef CONFIG_SMP |
917b627d | 1735 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1736 | #endif |
917b627d | 1737 | |
476d139c | 1738 | put_cpu(); |
1da177e4 LT |
1739 | } |
1740 | ||
1741 | /* | |
1742 | * wake_up_new_task - wake up a newly created task for the first time. | |
1743 | * | |
1744 | * This function will do some initial scheduler statistics housekeeping | |
1745 | * that must be done for every newly created context, then puts the task | |
1746 | * on the runqueue and wakes it. | |
1747 | */ | |
3e51e3ed | 1748 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1749 | { |
1750 | unsigned long flags; | |
dd41f596 | 1751 | struct rq *rq; |
fabf318e | 1752 | |
ab2515c4 | 1753 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1754 | #ifdef CONFIG_SMP |
1755 | /* | |
1756 | * Fork balancing, do it here and not earlier because: | |
1757 | * - cpus_allowed can change in the fork path | |
1758 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1759 | */ |
ab2515c4 | 1760 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1761 | #endif |
1762 | ||
ab2515c4 | 1763 | rq = __task_rq_lock(p); |
cd29fe6f | 1764 | activate_task(rq, p, 0); |
fd2f4419 | 1765 | p->on_rq = 1; |
89363381 | 1766 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1767 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1768 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1769 | if (p->sched_class->task_woken) |
1770 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1771 | #endif |
0122ec5b | 1772 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1773 | } |
1774 | ||
e107be36 AK |
1775 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1776 | ||
1777 | /** | |
80dd99b3 | 1778 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1779 | * @notifier: notifier struct to register |
e107be36 AK |
1780 | */ |
1781 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1782 | { | |
1783 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1784 | } | |
1785 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1786 | ||
1787 | /** | |
1788 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1789 | * @notifier: notifier struct to unregister |
e107be36 AK |
1790 | * |
1791 | * This is safe to call from within a preemption notifier. | |
1792 | */ | |
1793 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1794 | { | |
1795 | hlist_del(¬ifier->link); | |
1796 | } | |
1797 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1798 | ||
1799 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1800 | { | |
1801 | struct preempt_notifier *notifier; | |
1802 | struct hlist_node *node; | |
1803 | ||
1804 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1805 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1806 | } | |
1807 | ||
1808 | static void | |
1809 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1810 | struct task_struct *next) | |
1811 | { | |
1812 | struct preempt_notifier *notifier; | |
1813 | struct hlist_node *node; | |
1814 | ||
1815 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1816 | notifier->ops->sched_out(notifier, next); | |
1817 | } | |
1818 | ||
6d6bc0ad | 1819 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1820 | |
1821 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1822 | { | |
1823 | } | |
1824 | ||
1825 | static void | |
1826 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1827 | struct task_struct *next) | |
1828 | { | |
1829 | } | |
1830 | ||
6d6bc0ad | 1831 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1832 | |
4866cde0 NP |
1833 | /** |
1834 | * prepare_task_switch - prepare to switch tasks | |
1835 | * @rq: the runqueue preparing to switch | |
421cee29 | 1836 | * @prev: the current task that is being switched out |
4866cde0 NP |
1837 | * @next: the task we are going to switch to. |
1838 | * | |
1839 | * This is called with the rq lock held and interrupts off. It must | |
1840 | * be paired with a subsequent finish_task_switch after the context | |
1841 | * switch. | |
1842 | * | |
1843 | * prepare_task_switch sets up locking and calls architecture specific | |
1844 | * hooks. | |
1845 | */ | |
e107be36 AK |
1846 | static inline void |
1847 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1848 | struct task_struct *next) | |
4866cde0 | 1849 | { |
fe4b04fa PZ |
1850 | sched_info_switch(prev, next); |
1851 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1852 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1853 | prepare_lock_switch(rq, next); |
1854 | prepare_arch_switch(next); | |
fe4b04fa | 1855 | trace_sched_switch(prev, next); |
4866cde0 NP |
1856 | } |
1857 | ||
1da177e4 LT |
1858 | /** |
1859 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1860 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1861 | * @prev: the thread we just switched away from. |
1862 | * | |
4866cde0 NP |
1863 | * finish_task_switch must be called after the context switch, paired |
1864 | * with a prepare_task_switch call before the context switch. | |
1865 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1866 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1867 | * |
1868 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1869 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1870 | * with the lock held can cause deadlocks; see schedule() for |
1871 | * details.) | |
1872 | */ | |
a9957449 | 1873 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1874 | __releases(rq->lock) |
1875 | { | |
1da177e4 | 1876 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1877 | long prev_state; |
1da177e4 LT |
1878 | |
1879 | rq->prev_mm = NULL; | |
1880 | ||
1881 | /* | |
1882 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1883 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1884 | * schedule one last time. The schedule call will never return, and |
1885 | * the scheduled task must drop that reference. | |
c394cc9f | 1886 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1887 | * still held, otherwise prev could be scheduled on another cpu, die |
1888 | * there before we look at prev->state, and then the reference would | |
1889 | * be dropped twice. | |
1890 | * Manfred Spraul <manfred@colorfullife.com> | |
1891 | */ | |
55a101f8 | 1892 | prev_state = prev->state; |
4866cde0 | 1893 | finish_arch_switch(prev); |
8381f65d JI |
1894 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1895 | local_irq_disable(); | |
1896 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 1897 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
1898 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1899 | local_irq_enable(); | |
1900 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 1901 | finish_lock_switch(rq, prev); |
e8fa1362 | 1902 | |
e107be36 | 1903 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1904 | if (mm) |
1905 | mmdrop(mm); | |
c394cc9f | 1906 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1907 | /* |
1908 | * Remove function-return probe instances associated with this | |
1909 | * task and put them back on the free list. | |
9761eea8 | 1910 | */ |
c6fd91f0 | 1911 | kprobe_flush_task(prev); |
1da177e4 | 1912 | put_task_struct(prev); |
c6fd91f0 | 1913 | } |
1da177e4 LT |
1914 | } |
1915 | ||
3f029d3c GH |
1916 | #ifdef CONFIG_SMP |
1917 | ||
1918 | /* assumes rq->lock is held */ | |
1919 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1920 | { | |
1921 | if (prev->sched_class->pre_schedule) | |
1922 | prev->sched_class->pre_schedule(rq, prev); | |
1923 | } | |
1924 | ||
1925 | /* rq->lock is NOT held, but preemption is disabled */ | |
1926 | static inline void post_schedule(struct rq *rq) | |
1927 | { | |
1928 | if (rq->post_schedule) { | |
1929 | unsigned long flags; | |
1930 | ||
05fa785c | 1931 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1932 | if (rq->curr->sched_class->post_schedule) |
1933 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1934 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1935 | |
1936 | rq->post_schedule = 0; | |
1937 | } | |
1938 | } | |
1939 | ||
1940 | #else | |
da19ab51 | 1941 | |
3f029d3c GH |
1942 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1943 | { | |
1944 | } | |
1945 | ||
1946 | static inline void post_schedule(struct rq *rq) | |
1947 | { | |
1da177e4 LT |
1948 | } |
1949 | ||
3f029d3c GH |
1950 | #endif |
1951 | ||
1da177e4 LT |
1952 | /** |
1953 | * schedule_tail - first thing a freshly forked thread must call. | |
1954 | * @prev: the thread we just switched away from. | |
1955 | */ | |
36c8b586 | 1956 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1957 | __releases(rq->lock) |
1958 | { | |
70b97a7f IM |
1959 | struct rq *rq = this_rq(); |
1960 | ||
4866cde0 | 1961 | finish_task_switch(rq, prev); |
da19ab51 | 1962 | |
3f029d3c GH |
1963 | /* |
1964 | * FIXME: do we need to worry about rq being invalidated by the | |
1965 | * task_switch? | |
1966 | */ | |
1967 | post_schedule(rq); | |
70b97a7f | 1968 | |
4866cde0 NP |
1969 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
1970 | /* In this case, finish_task_switch does not reenable preemption */ | |
1971 | preempt_enable(); | |
1972 | #endif | |
1da177e4 | 1973 | if (current->set_child_tid) |
b488893a | 1974 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1975 | } |
1976 | ||
1977 | /* | |
1978 | * context_switch - switch to the new MM and the new | |
1979 | * thread's register state. | |
1980 | */ | |
dd41f596 | 1981 | static inline void |
70b97a7f | 1982 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1983 | struct task_struct *next) |
1da177e4 | 1984 | { |
dd41f596 | 1985 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1986 | |
e107be36 | 1987 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 1988 | |
dd41f596 IM |
1989 | mm = next->mm; |
1990 | oldmm = prev->active_mm; | |
9226d125 ZA |
1991 | /* |
1992 | * For paravirt, this is coupled with an exit in switch_to to | |
1993 | * combine the page table reload and the switch backend into | |
1994 | * one hypercall. | |
1995 | */ | |
224101ed | 1996 | arch_start_context_switch(prev); |
9226d125 | 1997 | |
31915ab4 | 1998 | if (!mm) { |
1da177e4 LT |
1999 | next->active_mm = oldmm; |
2000 | atomic_inc(&oldmm->mm_count); | |
2001 | enter_lazy_tlb(oldmm, next); | |
2002 | } else | |
2003 | switch_mm(oldmm, mm, next); | |
2004 | ||
31915ab4 | 2005 | if (!prev->mm) { |
1da177e4 | 2006 | prev->active_mm = NULL; |
1da177e4 LT |
2007 | rq->prev_mm = oldmm; |
2008 | } | |
3a5f5e48 IM |
2009 | /* |
2010 | * Since the runqueue lock will be released by the next | |
2011 | * task (which is an invalid locking op but in the case | |
2012 | * of the scheduler it's an obvious special-case), so we | |
2013 | * do an early lockdep release here: | |
2014 | */ | |
2015 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2016 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2017 | #endif |
1da177e4 LT |
2018 | |
2019 | /* Here we just switch the register state and the stack. */ | |
2020 | switch_to(prev, next, prev); | |
2021 | ||
dd41f596 IM |
2022 | barrier(); |
2023 | /* | |
2024 | * this_rq must be evaluated again because prev may have moved | |
2025 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2026 | * frame will be invalid. | |
2027 | */ | |
2028 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2029 | } |
2030 | ||
2031 | /* | |
2032 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2033 | * | |
2034 | * externally visible scheduler statistics: current number of runnable | |
2035 | * threads, current number of uninterruptible-sleeping threads, total | |
2036 | * number of context switches performed since bootup. | |
2037 | */ | |
2038 | unsigned long nr_running(void) | |
2039 | { | |
2040 | unsigned long i, sum = 0; | |
2041 | ||
2042 | for_each_online_cpu(i) | |
2043 | sum += cpu_rq(i)->nr_running; | |
2044 | ||
2045 | return sum; | |
f711f609 | 2046 | } |
1da177e4 LT |
2047 | |
2048 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2049 | { |
1da177e4 | 2050 | unsigned long i, sum = 0; |
f711f609 | 2051 | |
0a945022 | 2052 | for_each_possible_cpu(i) |
1da177e4 | 2053 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2054 | |
2055 | /* | |
1da177e4 LT |
2056 | * Since we read the counters lockless, it might be slightly |
2057 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2058 | */ |
1da177e4 LT |
2059 | if (unlikely((long)sum < 0)) |
2060 | sum = 0; | |
f711f609 | 2061 | |
1da177e4 | 2062 | return sum; |
f711f609 | 2063 | } |
f711f609 | 2064 | |
1da177e4 | 2065 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2066 | { |
cc94abfc SR |
2067 | int i; |
2068 | unsigned long long sum = 0; | |
46cb4b7c | 2069 | |
0a945022 | 2070 | for_each_possible_cpu(i) |
1da177e4 | 2071 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2072 | |
1da177e4 LT |
2073 | return sum; |
2074 | } | |
483b4ee6 | 2075 | |
1da177e4 LT |
2076 | unsigned long nr_iowait(void) |
2077 | { | |
2078 | unsigned long i, sum = 0; | |
483b4ee6 | 2079 | |
0a945022 | 2080 | for_each_possible_cpu(i) |
1da177e4 | 2081 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2082 | |
1da177e4 LT |
2083 | return sum; |
2084 | } | |
483b4ee6 | 2085 | |
8c215bd3 | 2086 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2087 | { |
8c215bd3 | 2088 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2089 | return atomic_read(&this->nr_iowait); |
2090 | } | |
46cb4b7c | 2091 | |
69d25870 AV |
2092 | unsigned long this_cpu_load(void) |
2093 | { | |
2094 | struct rq *this = this_rq(); | |
2095 | return this->cpu_load[0]; | |
2096 | } | |
e790fb0b | 2097 | |
46cb4b7c | 2098 | |
dce48a84 TG |
2099 | /* Variables and functions for calc_load */ |
2100 | static atomic_long_t calc_load_tasks; | |
2101 | static unsigned long calc_load_update; | |
2102 | unsigned long avenrun[3]; | |
2103 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2104 | |
74f5187a PZ |
2105 | static long calc_load_fold_active(struct rq *this_rq) |
2106 | { | |
2107 | long nr_active, delta = 0; | |
2108 | ||
2109 | nr_active = this_rq->nr_running; | |
2110 | nr_active += (long) this_rq->nr_uninterruptible; | |
2111 | ||
2112 | if (nr_active != this_rq->calc_load_active) { | |
2113 | delta = nr_active - this_rq->calc_load_active; | |
2114 | this_rq->calc_load_active = nr_active; | |
2115 | } | |
2116 | ||
2117 | return delta; | |
2118 | } | |
2119 | ||
0f004f5a PZ |
2120 | static unsigned long |
2121 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2122 | { | |
2123 | load *= exp; | |
2124 | load += active * (FIXED_1 - exp); | |
2125 | load += 1UL << (FSHIFT - 1); | |
2126 | return load >> FSHIFT; | |
2127 | } | |
2128 | ||
74f5187a PZ |
2129 | #ifdef CONFIG_NO_HZ |
2130 | /* | |
2131 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2132 | * | |
2133 | * When making the ILB scale, we should try to pull this in as well. | |
2134 | */ | |
2135 | static atomic_long_t calc_load_tasks_idle; | |
2136 | ||
029632fb | 2137 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2138 | { |
2139 | long delta; | |
2140 | ||
2141 | delta = calc_load_fold_active(this_rq); | |
2142 | if (delta) | |
2143 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2144 | } | |
2145 | ||
2146 | static long calc_load_fold_idle(void) | |
2147 | { | |
2148 | long delta = 0; | |
2149 | ||
2150 | /* | |
2151 | * Its got a race, we don't care... | |
2152 | */ | |
2153 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2154 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2155 | ||
2156 | return delta; | |
2157 | } | |
0f004f5a PZ |
2158 | |
2159 | /** | |
2160 | * fixed_power_int - compute: x^n, in O(log n) time | |
2161 | * | |
2162 | * @x: base of the power | |
2163 | * @frac_bits: fractional bits of @x | |
2164 | * @n: power to raise @x to. | |
2165 | * | |
2166 | * By exploiting the relation between the definition of the natural power | |
2167 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2168 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2169 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2170 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2171 | * of course trivially computable in O(log_2 n), the length of our binary | |
2172 | * vector. | |
2173 | */ | |
2174 | static unsigned long | |
2175 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2176 | { | |
2177 | unsigned long result = 1UL << frac_bits; | |
2178 | ||
2179 | if (n) for (;;) { | |
2180 | if (n & 1) { | |
2181 | result *= x; | |
2182 | result += 1UL << (frac_bits - 1); | |
2183 | result >>= frac_bits; | |
2184 | } | |
2185 | n >>= 1; | |
2186 | if (!n) | |
2187 | break; | |
2188 | x *= x; | |
2189 | x += 1UL << (frac_bits - 1); | |
2190 | x >>= frac_bits; | |
2191 | } | |
2192 | ||
2193 | return result; | |
2194 | } | |
2195 | ||
2196 | /* | |
2197 | * a1 = a0 * e + a * (1 - e) | |
2198 | * | |
2199 | * a2 = a1 * e + a * (1 - e) | |
2200 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2201 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2202 | * | |
2203 | * a3 = a2 * e + a * (1 - e) | |
2204 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2205 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2206 | * | |
2207 | * ... | |
2208 | * | |
2209 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2210 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2211 | * = a0 * e^n + a * (1 - e^n) | |
2212 | * | |
2213 | * [1] application of the geometric series: | |
2214 | * | |
2215 | * n 1 - x^(n+1) | |
2216 | * S_n := \Sum x^i = ------------- | |
2217 | * i=0 1 - x | |
2218 | */ | |
2219 | static unsigned long | |
2220 | calc_load_n(unsigned long load, unsigned long exp, | |
2221 | unsigned long active, unsigned int n) | |
2222 | { | |
2223 | ||
2224 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2225 | } | |
2226 | ||
2227 | /* | |
2228 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2229 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2230 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2231 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2232 | * | |
2233 | * Once we've updated the global active value, we need to apply the exponential | |
2234 | * weights adjusted to the number of cycles missed. | |
2235 | */ | |
2236 | static void calc_global_nohz(unsigned long ticks) | |
2237 | { | |
2238 | long delta, active, n; | |
2239 | ||
2240 | if (time_before(jiffies, calc_load_update)) | |
2241 | return; | |
2242 | ||
2243 | /* | |
2244 | * If we crossed a calc_load_update boundary, make sure to fold | |
2245 | * any pending idle changes, the respective CPUs might have | |
2246 | * missed the tick driven calc_load_account_active() update | |
2247 | * due to NO_HZ. | |
2248 | */ | |
2249 | delta = calc_load_fold_idle(); | |
2250 | if (delta) | |
2251 | atomic_long_add(delta, &calc_load_tasks); | |
2252 | ||
2253 | /* | |
2254 | * If we were idle for multiple load cycles, apply them. | |
2255 | */ | |
2256 | if (ticks >= LOAD_FREQ) { | |
2257 | n = ticks / LOAD_FREQ; | |
2258 | ||
2259 | active = atomic_long_read(&calc_load_tasks); | |
2260 | active = active > 0 ? active * FIXED_1 : 0; | |
2261 | ||
2262 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
2263 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2264 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
2265 | ||
2266 | calc_load_update += n * LOAD_FREQ; | |
2267 | } | |
2268 | ||
2269 | /* | |
2270 | * Its possible the remainder of the above division also crosses | |
2271 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
2272 | * which comes after this will take care of that. | |
2273 | * | |
2274 | * Consider us being 11 ticks before a cycle completion, and us | |
2275 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
2276 | * age us 4 cycles, and the test in calc_global_load() will | |
2277 | * pick up the final one. | |
2278 | */ | |
2279 | } | |
74f5187a | 2280 | #else |
029632fb | 2281 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2282 | { |
2283 | } | |
2284 | ||
2285 | static inline long calc_load_fold_idle(void) | |
2286 | { | |
2287 | return 0; | |
2288 | } | |
0f004f5a PZ |
2289 | |
2290 | static void calc_global_nohz(unsigned long ticks) | |
2291 | { | |
2292 | } | |
74f5187a PZ |
2293 | #endif |
2294 | ||
2d02494f TG |
2295 | /** |
2296 | * get_avenrun - get the load average array | |
2297 | * @loads: pointer to dest load array | |
2298 | * @offset: offset to add | |
2299 | * @shift: shift count to shift the result left | |
2300 | * | |
2301 | * These values are estimates at best, so no need for locking. | |
2302 | */ | |
2303 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2304 | { | |
2305 | loads[0] = (avenrun[0] + offset) << shift; | |
2306 | loads[1] = (avenrun[1] + offset) << shift; | |
2307 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2308 | } |
46cb4b7c | 2309 | |
46cb4b7c | 2310 | /* |
dce48a84 TG |
2311 | * calc_load - update the avenrun load estimates 10 ticks after the |
2312 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2313 | */ |
0f004f5a | 2314 | void calc_global_load(unsigned long ticks) |
7835b98b | 2315 | { |
dce48a84 | 2316 | long active; |
1da177e4 | 2317 | |
0f004f5a PZ |
2318 | calc_global_nohz(ticks); |
2319 | ||
2320 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 2321 | return; |
1da177e4 | 2322 | |
dce48a84 TG |
2323 | active = atomic_long_read(&calc_load_tasks); |
2324 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2325 | |
dce48a84 TG |
2326 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2327 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2328 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2329 | |
dce48a84 TG |
2330 | calc_load_update += LOAD_FREQ; |
2331 | } | |
1da177e4 | 2332 | |
dce48a84 | 2333 | /* |
74f5187a PZ |
2334 | * Called from update_cpu_load() to periodically update this CPU's |
2335 | * active count. | |
dce48a84 TG |
2336 | */ |
2337 | static void calc_load_account_active(struct rq *this_rq) | |
2338 | { | |
74f5187a | 2339 | long delta; |
08c183f3 | 2340 | |
74f5187a PZ |
2341 | if (time_before(jiffies, this_rq->calc_load_update)) |
2342 | return; | |
783609c6 | 2343 | |
74f5187a PZ |
2344 | delta = calc_load_fold_active(this_rq); |
2345 | delta += calc_load_fold_idle(); | |
2346 | if (delta) | |
dce48a84 | 2347 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2348 | |
2349 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2350 | } |
2351 | ||
fdf3e95d VP |
2352 | /* |
2353 | * The exact cpuload at various idx values, calculated at every tick would be | |
2354 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2355 | * | |
2356 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2357 | * on nth tick when cpu may be busy, then we have: | |
2358 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2359 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2360 | * | |
2361 | * decay_load_missed() below does efficient calculation of | |
2362 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2363 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2364 | * | |
2365 | * The calculation is approximated on a 128 point scale. | |
2366 | * degrade_zero_ticks is the number of ticks after which load at any | |
2367 | * particular idx is approximated to be zero. | |
2368 | * degrade_factor is a precomputed table, a row for each load idx. | |
2369 | * Each column corresponds to degradation factor for a power of two ticks, | |
2370 | * based on 128 point scale. | |
2371 | * Example: | |
2372 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2373 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2374 | * | |
2375 | * With this power of 2 load factors, we can degrade the load n times | |
2376 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2377 | * n mult/shifts needed by the exact degradation. | |
2378 | */ | |
2379 | #define DEGRADE_SHIFT 7 | |
2380 | static const unsigned char | |
2381 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2382 | static const unsigned char | |
2383 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2384 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2385 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2386 | {96, 72, 40, 12, 1, 0, 0}, | |
2387 | {112, 98, 75, 43, 15, 1, 0}, | |
2388 | {120, 112, 98, 76, 45, 16, 2} }; | |
2389 | ||
2390 | /* | |
2391 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2392 | * would be when CPU is idle and so we just decay the old load without | |
2393 | * adding any new load. | |
2394 | */ | |
2395 | static unsigned long | |
2396 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2397 | { | |
2398 | int j = 0; | |
2399 | ||
2400 | if (!missed_updates) | |
2401 | return load; | |
2402 | ||
2403 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2404 | return 0; | |
2405 | ||
2406 | if (idx == 1) | |
2407 | return load >> missed_updates; | |
2408 | ||
2409 | while (missed_updates) { | |
2410 | if (missed_updates % 2) | |
2411 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2412 | ||
2413 | missed_updates >>= 1; | |
2414 | j++; | |
2415 | } | |
2416 | return load; | |
2417 | } | |
2418 | ||
46cb4b7c | 2419 | /* |
dd41f596 | 2420 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2421 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2422 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2423 | */ |
029632fb | 2424 | void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 2425 | { |
495eca49 | 2426 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
2427 | unsigned long curr_jiffies = jiffies; |
2428 | unsigned long pending_updates; | |
dd41f596 | 2429 | int i, scale; |
46cb4b7c | 2430 | |
dd41f596 | 2431 | this_rq->nr_load_updates++; |
46cb4b7c | 2432 | |
fdf3e95d VP |
2433 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
2434 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2435 | return; | |
2436 | ||
2437 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2438 | this_rq->last_load_update_tick = curr_jiffies; | |
2439 | ||
dd41f596 | 2440 | /* Update our load: */ |
fdf3e95d VP |
2441 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2442 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2443 | unsigned long old_load, new_load; |
7d1e6a9b | 2444 | |
dd41f596 | 2445 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2446 | |
dd41f596 | 2447 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2448 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2449 | new_load = this_load; |
a25707f3 IM |
2450 | /* |
2451 | * Round up the averaging division if load is increasing. This | |
2452 | * prevents us from getting stuck on 9 if the load is 10, for | |
2453 | * example. | |
2454 | */ | |
2455 | if (new_load > old_load) | |
fdf3e95d VP |
2456 | new_load += scale - 1; |
2457 | ||
2458 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2459 | } |
da2b71ed SS |
2460 | |
2461 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2462 | } |
2463 | ||
2464 | static void update_cpu_load_active(struct rq *this_rq) | |
2465 | { | |
2466 | update_cpu_load(this_rq); | |
46cb4b7c | 2467 | |
74f5187a | 2468 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2469 | } |
2470 | ||
dd41f596 | 2471 | #ifdef CONFIG_SMP |
8a0be9ef | 2472 | |
46cb4b7c | 2473 | /* |
38022906 PZ |
2474 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2475 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2476 | */ |
38022906 | 2477 | void sched_exec(void) |
46cb4b7c | 2478 | { |
38022906 | 2479 | struct task_struct *p = current; |
1da177e4 | 2480 | unsigned long flags; |
0017d735 | 2481 | int dest_cpu; |
46cb4b7c | 2482 | |
8f42ced9 | 2483 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2484 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2485 | if (dest_cpu == smp_processor_id()) |
2486 | goto unlock; | |
38022906 | 2487 | |
8f42ced9 | 2488 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2489 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2490 | |
8f42ced9 PZ |
2491 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2492 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2493 | return; |
2494 | } | |
0017d735 | 2495 | unlock: |
8f42ced9 | 2496 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2497 | } |
dd41f596 | 2498 | |
1da177e4 LT |
2499 | #endif |
2500 | ||
1da177e4 LT |
2501 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
2502 | ||
2503 | EXPORT_PER_CPU_SYMBOL(kstat); | |
2504 | ||
2505 | /* | |
c5f8d995 | 2506 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2507 | * @p in case that task is currently running. |
c5f8d995 HS |
2508 | * |
2509 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2510 | */ |
c5f8d995 HS |
2511 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2512 | { | |
2513 | u64 ns = 0; | |
2514 | ||
2515 | if (task_current(rq, p)) { | |
2516 | update_rq_clock(rq); | |
305e6835 | 2517 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2518 | if ((s64)ns < 0) |
2519 | ns = 0; | |
2520 | } | |
2521 | ||
2522 | return ns; | |
2523 | } | |
2524 | ||
bb34d92f | 2525 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2526 | { |
1da177e4 | 2527 | unsigned long flags; |
41b86e9c | 2528 | struct rq *rq; |
bb34d92f | 2529 | u64 ns = 0; |
48f24c4d | 2530 | |
41b86e9c | 2531 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2532 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2533 | task_rq_unlock(rq, p, &flags); |
1508487e | 2534 | |
c5f8d995 HS |
2535 | return ns; |
2536 | } | |
f06febc9 | 2537 | |
c5f8d995 HS |
2538 | /* |
2539 | * Return accounted runtime for the task. | |
2540 | * In case the task is currently running, return the runtime plus current's | |
2541 | * pending runtime that have not been accounted yet. | |
2542 | */ | |
2543 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2544 | { | |
2545 | unsigned long flags; | |
2546 | struct rq *rq; | |
2547 | u64 ns = 0; | |
2548 | ||
2549 | rq = task_rq_lock(p, &flags); | |
2550 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2551 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2552 | |
2553 | return ns; | |
2554 | } | |
48f24c4d | 2555 | |
1da177e4 LT |
2556 | /* |
2557 | * Account user cpu time to a process. | |
2558 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 2559 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 2560 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 2561 | */ |
457533a7 MS |
2562 | void account_user_time(struct task_struct *p, cputime_t cputime, |
2563 | cputime_t cputime_scaled) | |
1da177e4 LT |
2564 | { |
2565 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2566 | cputime64_t tmp; | |
2567 | ||
457533a7 | 2568 | /* Add user time to process. */ |
1da177e4 | 2569 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 2570 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 2571 | account_group_user_time(p, cputime); |
1da177e4 LT |
2572 | |
2573 | /* Add user time to cpustat. */ | |
2574 | tmp = cputime_to_cputime64(cputime); | |
2575 | if (TASK_NICE(p) > 0) | |
2576 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
2577 | else | |
2578 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
2579 | |
2580 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
2581 | /* Account for user time used */ |
2582 | acct_update_integrals(p); | |
1da177e4 LT |
2583 | } |
2584 | ||
94886b84 LV |
2585 | /* |
2586 | * Account guest cpu time to a process. | |
2587 | * @p: the process that the cpu time gets accounted to | |
2588 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 2589 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 2590 | */ |
457533a7 MS |
2591 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
2592 | cputime_t cputime_scaled) | |
94886b84 LV |
2593 | { |
2594 | cputime64_t tmp; | |
2595 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2596 | ||
2597 | tmp = cputime_to_cputime64(cputime); | |
2598 | ||
457533a7 | 2599 | /* Add guest time to process. */ |
94886b84 | 2600 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 2601 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 2602 | account_group_user_time(p, cputime); |
94886b84 LV |
2603 | p->gtime = cputime_add(p->gtime, cputime); |
2604 | ||
457533a7 | 2605 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
2606 | if (TASK_NICE(p) > 0) { |
2607 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
2608 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
2609 | } else { | |
2610 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
2611 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
2612 | } | |
94886b84 LV |
2613 | } |
2614 | ||
70a89a66 VP |
2615 | /* |
2616 | * Account system cpu time to a process and desired cpustat field | |
2617 | * @p: the process that the cpu time gets accounted to | |
2618 | * @cputime: the cpu time spent in kernel space since the last update | |
2619 | * @cputime_scaled: cputime scaled by cpu frequency | |
2620 | * @target_cputime64: pointer to cpustat field that has to be updated | |
2621 | */ | |
2622 | static inline | |
2623 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
2624 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
2625 | { | |
2626 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
2627 | ||
2628 | /* Add system time to process. */ | |
2629 | p->stime = cputime_add(p->stime, cputime); | |
2630 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
2631 | account_group_system_time(p, cputime); | |
2632 | ||
2633 | /* Add system time to cpustat. */ | |
2634 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
2635 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
2636 | ||
2637 | /* Account for system time used */ | |
2638 | acct_update_integrals(p); | |
2639 | } | |
2640 | ||
1da177e4 LT |
2641 | /* |
2642 | * Account system cpu time to a process. | |
2643 | * @p: the process that the cpu time gets accounted to | |
2644 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
2645 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 2646 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
2647 | */ |
2648 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 2649 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
2650 | { |
2651 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 2652 | cputime64_t *target_cputime64; |
1da177e4 | 2653 | |
983ed7a6 | 2654 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 2655 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
2656 | return; |
2657 | } | |
94886b84 | 2658 | |
1da177e4 | 2659 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 2660 | target_cputime64 = &cpustat->irq; |
75e1056f | 2661 | else if (in_serving_softirq()) |
70a89a66 | 2662 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 2663 | else |
70a89a66 | 2664 | target_cputime64 = &cpustat->system; |
ef12fefa | 2665 | |
70a89a66 | 2666 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
2667 | } |
2668 | ||
c66f08be | 2669 | /* |
1da177e4 | 2670 | * Account for involuntary wait time. |
544b4a1f | 2671 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 2672 | */ |
79741dd3 | 2673 | void account_steal_time(cputime_t cputime) |
c66f08be | 2674 | { |
79741dd3 MS |
2675 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
2676 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
2677 | ||
2678 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
2679 | } |
2680 | ||
1da177e4 | 2681 | /* |
79741dd3 MS |
2682 | * Account for idle time. |
2683 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 2684 | */ |
79741dd3 | 2685 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
2686 | { |
2687 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 2688 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 2689 | struct rq *rq = this_rq(); |
1da177e4 | 2690 | |
79741dd3 MS |
2691 | if (atomic_read(&rq->nr_iowait) > 0) |
2692 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
2693 | else | |
2694 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
2695 | } |
2696 | ||
e6e6685a GC |
2697 | static __always_inline bool steal_account_process_tick(void) |
2698 | { | |
2699 | #ifdef CONFIG_PARAVIRT | |
2700 | if (static_branch(¶virt_steal_enabled)) { | |
2701 | u64 steal, st = 0; | |
2702 | ||
2703 | steal = paravirt_steal_clock(smp_processor_id()); | |
2704 | steal -= this_rq()->prev_steal_time; | |
2705 | ||
2706 | st = steal_ticks(steal); | |
2707 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
2708 | ||
2709 | account_steal_time(st); | |
2710 | return st; | |
2711 | } | |
2712 | #endif | |
2713 | return false; | |
2714 | } | |
2715 | ||
79741dd3 MS |
2716 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
2717 | ||
abb74cef VP |
2718 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
2719 | /* | |
2720 | * Account a tick to a process and cpustat | |
2721 | * @p: the process that the cpu time gets accounted to | |
2722 | * @user_tick: is the tick from userspace | |
2723 | * @rq: the pointer to rq | |
2724 | * | |
2725 | * Tick demultiplexing follows the order | |
2726 | * - pending hardirq update | |
2727 | * - pending softirq update | |
2728 | * - user_time | |
2729 | * - idle_time | |
2730 | * - system time | |
2731 | * - check for guest_time | |
2732 | * - else account as system_time | |
2733 | * | |
2734 | * Check for hardirq is done both for system and user time as there is | |
2735 | * no timer going off while we are on hardirq and hence we may never get an | |
2736 | * opportunity to update it solely in system time. | |
2737 | * p->stime and friends are only updated on system time and not on irq | |
2738 | * softirq as those do not count in task exec_runtime any more. | |
2739 | */ | |
2740 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2741 | struct rq *rq) | |
2742 | { | |
2743 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
2744 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
2745 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2746 | ||
e6e6685a GC |
2747 | if (steal_account_process_tick()) |
2748 | return; | |
2749 | ||
abb74cef VP |
2750 | if (irqtime_account_hi_update()) { |
2751 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
2752 | } else if (irqtime_account_si_update()) { | |
2753 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
2754 | } else if (this_cpu_ksoftirqd() == p) { |
2755 | /* | |
2756 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
2757 | * So, we have to handle it separately here. | |
2758 | * Also, p->stime needs to be updated for ksoftirqd. | |
2759 | */ | |
2760 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
2761 | &cpustat->softirq); | |
abb74cef VP |
2762 | } else if (user_tick) { |
2763 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2764 | } else if (p == rq->idle) { | |
2765 | account_idle_time(cputime_one_jiffy); | |
2766 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
2767 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2768 | } else { | |
2769 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
2770 | &cpustat->system); | |
2771 | } | |
2772 | } | |
2773 | ||
2774 | static void irqtime_account_idle_ticks(int ticks) | |
2775 | { | |
2776 | int i; | |
2777 | struct rq *rq = this_rq(); | |
2778 | ||
2779 | for (i = 0; i < ticks; i++) | |
2780 | irqtime_account_process_tick(current, 0, rq); | |
2781 | } | |
544b4a1f | 2782 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
2783 | static void irqtime_account_idle_ticks(int ticks) {} |
2784 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2785 | struct rq *rq) {} | |
544b4a1f | 2786 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
2787 | |
2788 | /* | |
2789 | * Account a single tick of cpu time. | |
2790 | * @p: the process that the cpu time gets accounted to | |
2791 | * @user_tick: indicates if the tick is a user or a system tick | |
2792 | */ | |
2793 | void account_process_tick(struct task_struct *p, int user_tick) | |
2794 | { | |
a42548a1 | 2795 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
2796 | struct rq *rq = this_rq(); |
2797 | ||
abb74cef VP |
2798 | if (sched_clock_irqtime) { |
2799 | irqtime_account_process_tick(p, user_tick, rq); | |
2800 | return; | |
2801 | } | |
2802 | ||
e6e6685a GC |
2803 | if (steal_account_process_tick()) |
2804 | return; | |
2805 | ||
79741dd3 | 2806 | if (user_tick) |
a42548a1 | 2807 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 2808 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 2809 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
2810 | one_jiffy_scaled); |
2811 | else | |
a42548a1 | 2812 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
2813 | } |
2814 | ||
2815 | /* | |
2816 | * Account multiple ticks of steal time. | |
2817 | * @p: the process from which the cpu time has been stolen | |
2818 | * @ticks: number of stolen ticks | |
2819 | */ | |
2820 | void account_steal_ticks(unsigned long ticks) | |
2821 | { | |
2822 | account_steal_time(jiffies_to_cputime(ticks)); | |
2823 | } | |
2824 | ||
2825 | /* | |
2826 | * Account multiple ticks of idle time. | |
2827 | * @ticks: number of stolen ticks | |
2828 | */ | |
2829 | void account_idle_ticks(unsigned long ticks) | |
2830 | { | |
abb74cef VP |
2831 | |
2832 | if (sched_clock_irqtime) { | |
2833 | irqtime_account_idle_ticks(ticks); | |
2834 | return; | |
2835 | } | |
2836 | ||
79741dd3 | 2837 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
2838 | } |
2839 | ||
79741dd3 MS |
2840 | #endif |
2841 | ||
49048622 BS |
2842 | /* |
2843 | * Use precise platform statistics if available: | |
2844 | */ | |
2845 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 2846 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2847 | { |
d99ca3b9 HS |
2848 | *ut = p->utime; |
2849 | *st = p->stime; | |
49048622 BS |
2850 | } |
2851 | ||
0cf55e1e | 2852 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2853 | { |
0cf55e1e HS |
2854 | struct task_cputime cputime; |
2855 | ||
2856 | thread_group_cputime(p, &cputime); | |
2857 | ||
2858 | *ut = cputime.utime; | |
2859 | *st = cputime.stime; | |
49048622 BS |
2860 | } |
2861 | #else | |
761b1d26 HS |
2862 | |
2863 | #ifndef nsecs_to_cputime | |
b7b20df9 | 2864 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
2865 | #endif |
2866 | ||
d180c5bc | 2867 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2868 | { |
d99ca3b9 | 2869 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
2870 | |
2871 | /* | |
2872 | * Use CFS's precise accounting: | |
2873 | */ | |
d180c5bc | 2874 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
2875 | |
2876 | if (total) { | |
e75e863d | 2877 | u64 temp = rtime; |
d180c5bc | 2878 | |
e75e863d | 2879 | temp *= utime; |
49048622 | 2880 | do_div(temp, total); |
d180c5bc HS |
2881 | utime = (cputime_t)temp; |
2882 | } else | |
2883 | utime = rtime; | |
49048622 | 2884 | |
d180c5bc HS |
2885 | /* |
2886 | * Compare with previous values, to keep monotonicity: | |
2887 | */ | |
761b1d26 | 2888 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 2889 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 2890 | |
d99ca3b9 HS |
2891 | *ut = p->prev_utime; |
2892 | *st = p->prev_stime; | |
49048622 BS |
2893 | } |
2894 | ||
0cf55e1e HS |
2895 | /* |
2896 | * Must be called with siglock held. | |
2897 | */ | |
2898 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 2899 | { |
0cf55e1e HS |
2900 | struct signal_struct *sig = p->signal; |
2901 | struct task_cputime cputime; | |
2902 | cputime_t rtime, utime, total; | |
49048622 | 2903 | |
0cf55e1e | 2904 | thread_group_cputime(p, &cputime); |
49048622 | 2905 | |
0cf55e1e HS |
2906 | total = cputime_add(cputime.utime, cputime.stime); |
2907 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 2908 | |
0cf55e1e | 2909 | if (total) { |
e75e863d | 2910 | u64 temp = rtime; |
49048622 | 2911 | |
e75e863d | 2912 | temp *= cputime.utime; |
0cf55e1e HS |
2913 | do_div(temp, total); |
2914 | utime = (cputime_t)temp; | |
2915 | } else | |
2916 | utime = rtime; | |
2917 | ||
2918 | sig->prev_utime = max(sig->prev_utime, utime); | |
2919 | sig->prev_stime = max(sig->prev_stime, | |
2920 | cputime_sub(rtime, sig->prev_utime)); | |
2921 | ||
2922 | *ut = sig->prev_utime; | |
2923 | *st = sig->prev_stime; | |
49048622 | 2924 | } |
49048622 | 2925 | #endif |
49048622 | 2926 | |
7835b98b CL |
2927 | /* |
2928 | * This function gets called by the timer code, with HZ frequency. | |
2929 | * We call it with interrupts disabled. | |
7835b98b CL |
2930 | */ |
2931 | void scheduler_tick(void) | |
2932 | { | |
7835b98b CL |
2933 | int cpu = smp_processor_id(); |
2934 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2935 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2936 | |
2937 | sched_clock_tick(); | |
dd41f596 | 2938 | |
05fa785c | 2939 | raw_spin_lock(&rq->lock); |
3e51f33f | 2940 | update_rq_clock(rq); |
fdf3e95d | 2941 | update_cpu_load_active(rq); |
fa85ae24 | 2942 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2943 | raw_spin_unlock(&rq->lock); |
7835b98b | 2944 | |
e9d2b064 | 2945 | perf_event_task_tick(); |
e220d2dc | 2946 | |
e418e1c2 | 2947 | #ifdef CONFIG_SMP |
6eb57e0d | 2948 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2949 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2950 | #endif |
1da177e4 LT |
2951 | } |
2952 | ||
132380a0 | 2953 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2954 | { |
2955 | if (in_lock_functions(addr)) { | |
2956 | addr = CALLER_ADDR2; | |
2957 | if (in_lock_functions(addr)) | |
2958 | addr = CALLER_ADDR3; | |
2959 | } | |
2960 | return addr; | |
2961 | } | |
1da177e4 | 2962 | |
7e49fcce SR |
2963 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2964 | defined(CONFIG_PREEMPT_TRACER)) | |
2965 | ||
43627582 | 2966 | void __kprobes add_preempt_count(int val) |
1da177e4 | 2967 | { |
6cd8a4bb | 2968 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2969 | /* |
2970 | * Underflow? | |
2971 | */ | |
9a11b49a IM |
2972 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2973 | return; | |
6cd8a4bb | 2974 | #endif |
1da177e4 | 2975 | preempt_count() += val; |
6cd8a4bb | 2976 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2977 | /* |
2978 | * Spinlock count overflowing soon? | |
2979 | */ | |
33859f7f MOS |
2980 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2981 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2982 | #endif |
2983 | if (preempt_count() == val) | |
2984 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2985 | } |
2986 | EXPORT_SYMBOL(add_preempt_count); | |
2987 | ||
43627582 | 2988 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 2989 | { |
6cd8a4bb | 2990 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2991 | /* |
2992 | * Underflow? | |
2993 | */ | |
01e3eb82 | 2994 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2995 | return; |
1da177e4 LT |
2996 | /* |
2997 | * Is the spinlock portion underflowing? | |
2998 | */ | |
9a11b49a IM |
2999 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3000 | !(preempt_count() & PREEMPT_MASK))) | |
3001 | return; | |
6cd8a4bb | 3002 | #endif |
9a11b49a | 3003 | |
6cd8a4bb SR |
3004 | if (preempt_count() == val) |
3005 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3006 | preempt_count() -= val; |
3007 | } | |
3008 | EXPORT_SYMBOL(sub_preempt_count); | |
3009 | ||
3010 | #endif | |
3011 | ||
3012 | /* | |
dd41f596 | 3013 | * Print scheduling while atomic bug: |
1da177e4 | 3014 | */ |
dd41f596 | 3015 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3016 | { |
838225b4 SS |
3017 | struct pt_regs *regs = get_irq_regs(); |
3018 | ||
3df0fc5b PZ |
3019 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3020 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3021 | |
dd41f596 | 3022 | debug_show_held_locks(prev); |
e21f5b15 | 3023 | print_modules(); |
dd41f596 IM |
3024 | if (irqs_disabled()) |
3025 | print_irqtrace_events(prev); | |
838225b4 SS |
3026 | |
3027 | if (regs) | |
3028 | show_regs(regs); | |
3029 | else | |
3030 | dump_stack(); | |
dd41f596 | 3031 | } |
1da177e4 | 3032 | |
dd41f596 IM |
3033 | /* |
3034 | * Various schedule()-time debugging checks and statistics: | |
3035 | */ | |
3036 | static inline void schedule_debug(struct task_struct *prev) | |
3037 | { | |
1da177e4 | 3038 | /* |
41a2d6cf | 3039 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3040 | * schedule() atomically, we ignore that path for now. |
3041 | * Otherwise, whine if we are scheduling when we should not be. | |
3042 | */ | |
3f33a7ce | 3043 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 3044 | __schedule_bug(prev); |
b3fbab05 | 3045 | rcu_sleep_check(); |
dd41f596 | 3046 | |
1da177e4 LT |
3047 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3048 | ||
2d72376b | 3049 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3050 | } |
3051 | ||
6cecd084 | 3052 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3053 | { |
61eadef6 | 3054 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 3055 | update_rq_clock(rq); |
6cecd084 | 3056 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3057 | } |
3058 | ||
dd41f596 IM |
3059 | /* |
3060 | * Pick up the highest-prio task: | |
3061 | */ | |
3062 | static inline struct task_struct * | |
b67802ea | 3063 | pick_next_task(struct rq *rq) |
dd41f596 | 3064 | { |
5522d5d5 | 3065 | const struct sched_class *class; |
dd41f596 | 3066 | struct task_struct *p; |
1da177e4 LT |
3067 | |
3068 | /* | |
dd41f596 IM |
3069 | * Optimization: we know that if all tasks are in |
3070 | * the fair class we can call that function directly: | |
1da177e4 | 3071 | */ |
953bfcd1 | 3072 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 3073 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3074 | if (likely(p)) |
3075 | return p; | |
1da177e4 LT |
3076 | } |
3077 | ||
34f971f6 | 3078 | for_each_class(class) { |
fb8d4724 | 3079 | p = class->pick_next_task(rq); |
dd41f596 IM |
3080 | if (p) |
3081 | return p; | |
dd41f596 | 3082 | } |
34f971f6 PZ |
3083 | |
3084 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3085 | } |
1da177e4 | 3086 | |
dd41f596 | 3087 | /* |
c259e01a | 3088 | * __schedule() is the main scheduler function. |
dd41f596 | 3089 | */ |
c259e01a | 3090 | static void __sched __schedule(void) |
dd41f596 IM |
3091 | { |
3092 | struct task_struct *prev, *next; | |
67ca7bde | 3093 | unsigned long *switch_count; |
dd41f596 | 3094 | struct rq *rq; |
31656519 | 3095 | int cpu; |
dd41f596 | 3096 | |
ff743345 PZ |
3097 | need_resched: |
3098 | preempt_disable(); | |
dd41f596 IM |
3099 | cpu = smp_processor_id(); |
3100 | rq = cpu_rq(cpu); | |
25502a6c | 3101 | rcu_note_context_switch(cpu); |
dd41f596 | 3102 | prev = rq->curr; |
dd41f596 | 3103 | |
dd41f596 | 3104 | schedule_debug(prev); |
1da177e4 | 3105 | |
31656519 | 3106 | if (sched_feat(HRTICK)) |
f333fdc9 | 3107 | hrtick_clear(rq); |
8f4d37ec | 3108 | |
05fa785c | 3109 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3110 | |
246d86b5 | 3111 | switch_count = &prev->nivcsw; |
1da177e4 | 3112 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3113 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3114 | prev->state = TASK_RUNNING; |
21aa9af0 | 3115 | } else { |
2acca55e PZ |
3116 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3117 | prev->on_rq = 0; | |
3118 | ||
21aa9af0 | 3119 | /* |
2acca55e PZ |
3120 | * If a worker went to sleep, notify and ask workqueue |
3121 | * whether it wants to wake up a task to maintain | |
3122 | * concurrency. | |
21aa9af0 TH |
3123 | */ |
3124 | if (prev->flags & PF_WQ_WORKER) { | |
3125 | struct task_struct *to_wakeup; | |
3126 | ||
3127 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3128 | if (to_wakeup) | |
3129 | try_to_wake_up_local(to_wakeup); | |
3130 | } | |
21aa9af0 | 3131 | } |
dd41f596 | 3132 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3133 | } |
3134 | ||
3f029d3c | 3135 | pre_schedule(rq, prev); |
f65eda4f | 3136 | |
dd41f596 | 3137 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3138 | idle_balance(cpu, rq); |
1da177e4 | 3139 | |
df1c99d4 | 3140 | put_prev_task(rq, prev); |
b67802ea | 3141 | next = pick_next_task(rq); |
f26f9aff MG |
3142 | clear_tsk_need_resched(prev); |
3143 | rq->skip_clock_update = 0; | |
1da177e4 | 3144 | |
1da177e4 | 3145 | if (likely(prev != next)) { |
1da177e4 LT |
3146 | rq->nr_switches++; |
3147 | rq->curr = next; | |
3148 | ++*switch_count; | |
3149 | ||
dd41f596 | 3150 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3151 | /* |
246d86b5 ON |
3152 | * The context switch have flipped the stack from under us |
3153 | * and restored the local variables which were saved when | |
3154 | * this task called schedule() in the past. prev == current | |
3155 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3156 | */ |
3157 | cpu = smp_processor_id(); | |
3158 | rq = cpu_rq(cpu); | |
1da177e4 | 3159 | } else |
05fa785c | 3160 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3161 | |
3f029d3c | 3162 | post_schedule(rq); |
1da177e4 | 3163 | |
1da177e4 | 3164 | preempt_enable_no_resched(); |
ff743345 | 3165 | if (need_resched()) |
1da177e4 LT |
3166 | goto need_resched; |
3167 | } | |
c259e01a | 3168 | |
9c40cef2 TG |
3169 | static inline void sched_submit_work(struct task_struct *tsk) |
3170 | { | |
3171 | if (!tsk->state) | |
3172 | return; | |
3173 | /* | |
3174 | * If we are going to sleep and we have plugged IO queued, | |
3175 | * make sure to submit it to avoid deadlocks. | |
3176 | */ | |
3177 | if (blk_needs_flush_plug(tsk)) | |
3178 | blk_schedule_flush_plug(tsk); | |
3179 | } | |
3180 | ||
6ebbe7a0 | 3181 | asmlinkage void __sched schedule(void) |
c259e01a | 3182 | { |
9c40cef2 TG |
3183 | struct task_struct *tsk = current; |
3184 | ||
3185 | sched_submit_work(tsk); | |
c259e01a TG |
3186 | __schedule(); |
3187 | } | |
1da177e4 LT |
3188 | EXPORT_SYMBOL(schedule); |
3189 | ||
c08f7829 | 3190 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3191 | |
c6eb3dda PZ |
3192 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3193 | { | |
c6eb3dda | 3194 | if (lock->owner != owner) |
307bf980 | 3195 | return false; |
0d66bf6d PZ |
3196 | |
3197 | /* | |
c6eb3dda PZ |
3198 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3199 | * lock->owner still matches owner, if that fails, owner might | |
3200 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3201 | * ensures the memory stays valid. | |
0d66bf6d | 3202 | */ |
c6eb3dda | 3203 | barrier(); |
0d66bf6d | 3204 | |
307bf980 | 3205 | return owner->on_cpu; |
c6eb3dda | 3206 | } |
0d66bf6d | 3207 | |
c6eb3dda PZ |
3208 | /* |
3209 | * Look out! "owner" is an entirely speculative pointer | |
3210 | * access and not reliable. | |
3211 | */ | |
3212 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3213 | { | |
3214 | if (!sched_feat(OWNER_SPIN)) | |
3215 | return 0; | |
0d66bf6d | 3216 | |
307bf980 | 3217 | rcu_read_lock(); |
c6eb3dda PZ |
3218 | while (owner_running(lock, owner)) { |
3219 | if (need_resched()) | |
307bf980 | 3220 | break; |
0d66bf6d | 3221 | |
335d7afb | 3222 | arch_mutex_cpu_relax(); |
0d66bf6d | 3223 | } |
307bf980 | 3224 | rcu_read_unlock(); |
4b402210 | 3225 | |
c6eb3dda | 3226 | /* |
307bf980 TG |
3227 | * We break out the loop above on need_resched() and when the |
3228 | * owner changed, which is a sign for heavy contention. Return | |
3229 | * success only when lock->owner is NULL. | |
c6eb3dda | 3230 | */ |
307bf980 | 3231 | return lock->owner == NULL; |
0d66bf6d PZ |
3232 | } |
3233 | #endif | |
3234 | ||
1da177e4 LT |
3235 | #ifdef CONFIG_PREEMPT |
3236 | /* | |
2ed6e34f | 3237 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3238 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3239 | * occur there and call schedule directly. |
3240 | */ | |
d1f74e20 | 3241 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3242 | { |
3243 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3244 | |
1da177e4 LT |
3245 | /* |
3246 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3247 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3248 | */ |
beed33a8 | 3249 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3250 | return; |
3251 | ||
3a5c359a | 3252 | do { |
d1f74e20 | 3253 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3254 | __schedule(); |
d1f74e20 | 3255 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3256 | |
3a5c359a AK |
3257 | /* |
3258 | * Check again in case we missed a preemption opportunity | |
3259 | * between schedule and now. | |
3260 | */ | |
3261 | barrier(); | |
5ed0cec0 | 3262 | } while (need_resched()); |
1da177e4 | 3263 | } |
1da177e4 LT |
3264 | EXPORT_SYMBOL(preempt_schedule); |
3265 | ||
3266 | /* | |
2ed6e34f | 3267 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3268 | * off of irq context. |
3269 | * Note, that this is called and return with irqs disabled. This will | |
3270 | * protect us against recursive calling from irq. | |
3271 | */ | |
3272 | asmlinkage void __sched preempt_schedule_irq(void) | |
3273 | { | |
3274 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3275 | |
2ed6e34f | 3276 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3277 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3278 | ||
3a5c359a AK |
3279 | do { |
3280 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3281 | local_irq_enable(); |
c259e01a | 3282 | __schedule(); |
3a5c359a | 3283 | local_irq_disable(); |
3a5c359a | 3284 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3285 | |
3a5c359a AK |
3286 | /* |
3287 | * Check again in case we missed a preemption opportunity | |
3288 | * between schedule and now. | |
3289 | */ | |
3290 | barrier(); | |
5ed0cec0 | 3291 | } while (need_resched()); |
1da177e4 LT |
3292 | } |
3293 | ||
3294 | #endif /* CONFIG_PREEMPT */ | |
3295 | ||
63859d4f | 3296 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3297 | void *key) |
1da177e4 | 3298 | { |
63859d4f | 3299 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3300 | } |
1da177e4 LT |
3301 | EXPORT_SYMBOL(default_wake_function); |
3302 | ||
3303 | /* | |
41a2d6cf IM |
3304 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3305 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3306 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3307 | * | |
3308 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3309 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3310 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3311 | */ | |
78ddb08f | 3312 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3313 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3314 | { |
2e45874c | 3315 | wait_queue_t *curr, *next; |
1da177e4 | 3316 | |
2e45874c | 3317 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3318 | unsigned flags = curr->flags; |
3319 | ||
63859d4f | 3320 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3321 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3322 | break; |
3323 | } | |
3324 | } | |
3325 | ||
3326 | /** | |
3327 | * __wake_up - wake up threads blocked on a waitqueue. | |
3328 | * @q: the waitqueue | |
3329 | * @mode: which threads | |
3330 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3331 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3332 | * |
3333 | * It may be assumed that this function implies a write memory barrier before | |
3334 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3335 | */ |
7ad5b3a5 | 3336 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3337 | int nr_exclusive, void *key) |
1da177e4 LT |
3338 | { |
3339 | unsigned long flags; | |
3340 | ||
3341 | spin_lock_irqsave(&q->lock, flags); | |
3342 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3343 | spin_unlock_irqrestore(&q->lock, flags); | |
3344 | } | |
1da177e4 LT |
3345 | EXPORT_SYMBOL(__wake_up); |
3346 | ||
3347 | /* | |
3348 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3349 | */ | |
7ad5b3a5 | 3350 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3351 | { |
3352 | __wake_up_common(q, mode, 1, 0, NULL); | |
3353 | } | |
22c43c81 | 3354 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3355 | |
4ede816a DL |
3356 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3357 | { | |
3358 | __wake_up_common(q, mode, 1, 0, key); | |
3359 | } | |
bf294b41 | 3360 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3361 | |
1da177e4 | 3362 | /** |
4ede816a | 3363 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3364 | * @q: the waitqueue |
3365 | * @mode: which threads | |
3366 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3367 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3368 | * |
3369 | * The sync wakeup differs that the waker knows that it will schedule | |
3370 | * away soon, so while the target thread will be woken up, it will not | |
3371 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3372 | * with each other. This can prevent needless bouncing between CPUs. | |
3373 | * | |
3374 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3375 | * |
3376 | * It may be assumed that this function implies a write memory barrier before | |
3377 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3378 | */ |
4ede816a DL |
3379 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3380 | int nr_exclusive, void *key) | |
1da177e4 LT |
3381 | { |
3382 | unsigned long flags; | |
7d478721 | 3383 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3384 | |
3385 | if (unlikely(!q)) | |
3386 | return; | |
3387 | ||
3388 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3389 | wake_flags = 0; |
1da177e4 LT |
3390 | |
3391 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3392 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3393 | spin_unlock_irqrestore(&q->lock, flags); |
3394 | } | |
4ede816a DL |
3395 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3396 | ||
3397 | /* | |
3398 | * __wake_up_sync - see __wake_up_sync_key() | |
3399 | */ | |
3400 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3401 | { | |
3402 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3403 | } | |
1da177e4 LT |
3404 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3405 | ||
65eb3dc6 KD |
3406 | /** |
3407 | * complete: - signals a single thread waiting on this completion | |
3408 | * @x: holds the state of this particular completion | |
3409 | * | |
3410 | * This will wake up a single thread waiting on this completion. Threads will be | |
3411 | * awakened in the same order in which they were queued. | |
3412 | * | |
3413 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3414 | * |
3415 | * It may be assumed that this function implies a write memory barrier before | |
3416 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3417 | */ |
b15136e9 | 3418 | void complete(struct completion *x) |
1da177e4 LT |
3419 | { |
3420 | unsigned long flags; | |
3421 | ||
3422 | spin_lock_irqsave(&x->wait.lock, flags); | |
3423 | x->done++; | |
d9514f6c | 3424 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3425 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3426 | } | |
3427 | EXPORT_SYMBOL(complete); | |
3428 | ||
65eb3dc6 KD |
3429 | /** |
3430 | * complete_all: - signals all threads waiting on this completion | |
3431 | * @x: holds the state of this particular completion | |
3432 | * | |
3433 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3434 | * |
3435 | * It may be assumed that this function implies a write memory barrier before | |
3436 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3437 | */ |
b15136e9 | 3438 | void complete_all(struct completion *x) |
1da177e4 LT |
3439 | { |
3440 | unsigned long flags; | |
3441 | ||
3442 | spin_lock_irqsave(&x->wait.lock, flags); | |
3443 | x->done += UINT_MAX/2; | |
d9514f6c | 3444 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3445 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3446 | } | |
3447 | EXPORT_SYMBOL(complete_all); | |
3448 | ||
8cbbe86d AK |
3449 | static inline long __sched |
3450 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3451 | { |
1da177e4 LT |
3452 | if (!x->done) { |
3453 | DECLARE_WAITQUEUE(wait, current); | |
3454 | ||
a93d2f17 | 3455 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3456 | do { |
94d3d824 | 3457 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3458 | timeout = -ERESTARTSYS; |
3459 | break; | |
8cbbe86d AK |
3460 | } |
3461 | __set_current_state(state); | |
1da177e4 LT |
3462 | spin_unlock_irq(&x->wait.lock); |
3463 | timeout = schedule_timeout(timeout); | |
3464 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3465 | } while (!x->done && timeout); |
1da177e4 | 3466 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3467 | if (!x->done) |
3468 | return timeout; | |
1da177e4 LT |
3469 | } |
3470 | x->done--; | |
ea71a546 | 3471 | return timeout ?: 1; |
1da177e4 | 3472 | } |
1da177e4 | 3473 | |
8cbbe86d AK |
3474 | static long __sched |
3475 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3476 | { |
1da177e4 LT |
3477 | might_sleep(); |
3478 | ||
3479 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3480 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3481 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3482 | return timeout; |
3483 | } | |
1da177e4 | 3484 | |
65eb3dc6 KD |
3485 | /** |
3486 | * wait_for_completion: - waits for completion of a task | |
3487 | * @x: holds the state of this particular completion | |
3488 | * | |
3489 | * This waits to be signaled for completion of a specific task. It is NOT | |
3490 | * interruptible and there is no timeout. | |
3491 | * | |
3492 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3493 | * and interrupt capability. Also see complete(). | |
3494 | */ | |
b15136e9 | 3495 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3496 | { |
3497 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3498 | } |
8cbbe86d | 3499 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3500 | |
65eb3dc6 KD |
3501 | /** |
3502 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3503 | * @x: holds the state of this particular completion | |
3504 | * @timeout: timeout value in jiffies | |
3505 | * | |
3506 | * This waits for either a completion of a specific task to be signaled or for a | |
3507 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3508 | * interruptible. | |
c6dc7f05 BF |
3509 | * |
3510 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3511 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3512 | */ |
b15136e9 | 3513 | unsigned long __sched |
8cbbe86d | 3514 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3515 | { |
8cbbe86d | 3516 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3517 | } |
8cbbe86d | 3518 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3519 | |
65eb3dc6 KD |
3520 | /** |
3521 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3522 | * @x: holds the state of this particular completion | |
3523 | * | |
3524 | * This waits for completion of a specific task to be signaled. It is | |
3525 | * interruptible. | |
c6dc7f05 BF |
3526 | * |
3527 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3528 | */ |
8cbbe86d | 3529 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3530 | { |
51e97990 AK |
3531 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3532 | if (t == -ERESTARTSYS) | |
3533 | return t; | |
3534 | return 0; | |
0fec171c | 3535 | } |
8cbbe86d | 3536 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3537 | |
65eb3dc6 KD |
3538 | /** |
3539 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3540 | * @x: holds the state of this particular completion | |
3541 | * @timeout: timeout value in jiffies | |
3542 | * | |
3543 | * This waits for either a completion of a specific task to be signaled or for a | |
3544 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3545 | * |
3546 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3547 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3548 | */ |
6bf41237 | 3549 | long __sched |
8cbbe86d AK |
3550 | wait_for_completion_interruptible_timeout(struct completion *x, |
3551 | unsigned long timeout) | |
0fec171c | 3552 | { |
8cbbe86d | 3553 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3554 | } |
8cbbe86d | 3555 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3556 | |
65eb3dc6 KD |
3557 | /** |
3558 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3559 | * @x: holds the state of this particular completion | |
3560 | * | |
3561 | * This waits to be signaled for completion of a specific task. It can be | |
3562 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3563 | * |
3564 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3565 | */ |
009e577e MW |
3566 | int __sched wait_for_completion_killable(struct completion *x) |
3567 | { | |
3568 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3569 | if (t == -ERESTARTSYS) | |
3570 | return t; | |
3571 | return 0; | |
3572 | } | |
3573 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3574 | ||
0aa12fb4 SW |
3575 | /** |
3576 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3577 | * @x: holds the state of this particular completion | |
3578 | * @timeout: timeout value in jiffies | |
3579 | * | |
3580 | * This waits for either a completion of a specific task to be | |
3581 | * signaled or for a specified timeout to expire. It can be | |
3582 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3583 | * |
3584 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3585 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3586 | */ |
6bf41237 | 3587 | long __sched |
0aa12fb4 SW |
3588 | wait_for_completion_killable_timeout(struct completion *x, |
3589 | unsigned long timeout) | |
3590 | { | |
3591 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3592 | } | |
3593 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3594 | ||
be4de352 DC |
3595 | /** |
3596 | * try_wait_for_completion - try to decrement a completion without blocking | |
3597 | * @x: completion structure | |
3598 | * | |
3599 | * Returns: 0 if a decrement cannot be done without blocking | |
3600 | * 1 if a decrement succeeded. | |
3601 | * | |
3602 | * If a completion is being used as a counting completion, | |
3603 | * attempt to decrement the counter without blocking. This | |
3604 | * enables us to avoid waiting if the resource the completion | |
3605 | * is protecting is not available. | |
3606 | */ | |
3607 | bool try_wait_for_completion(struct completion *x) | |
3608 | { | |
7539a3b3 | 3609 | unsigned long flags; |
be4de352 DC |
3610 | int ret = 1; |
3611 | ||
7539a3b3 | 3612 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3613 | if (!x->done) |
3614 | ret = 0; | |
3615 | else | |
3616 | x->done--; | |
7539a3b3 | 3617 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3618 | return ret; |
3619 | } | |
3620 | EXPORT_SYMBOL(try_wait_for_completion); | |
3621 | ||
3622 | /** | |
3623 | * completion_done - Test to see if a completion has any waiters | |
3624 | * @x: completion structure | |
3625 | * | |
3626 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3627 | * 1 if there are no waiters. | |
3628 | * | |
3629 | */ | |
3630 | bool completion_done(struct completion *x) | |
3631 | { | |
7539a3b3 | 3632 | unsigned long flags; |
be4de352 DC |
3633 | int ret = 1; |
3634 | ||
7539a3b3 | 3635 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3636 | if (!x->done) |
3637 | ret = 0; | |
7539a3b3 | 3638 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3639 | return ret; |
3640 | } | |
3641 | EXPORT_SYMBOL(completion_done); | |
3642 | ||
8cbbe86d AK |
3643 | static long __sched |
3644 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3645 | { |
0fec171c IM |
3646 | unsigned long flags; |
3647 | wait_queue_t wait; | |
3648 | ||
3649 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3650 | |
8cbbe86d | 3651 | __set_current_state(state); |
1da177e4 | 3652 | |
8cbbe86d AK |
3653 | spin_lock_irqsave(&q->lock, flags); |
3654 | __add_wait_queue(q, &wait); | |
3655 | spin_unlock(&q->lock); | |
3656 | timeout = schedule_timeout(timeout); | |
3657 | spin_lock_irq(&q->lock); | |
3658 | __remove_wait_queue(q, &wait); | |
3659 | spin_unlock_irqrestore(&q->lock, flags); | |
3660 | ||
3661 | return timeout; | |
3662 | } | |
3663 | ||
3664 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3665 | { | |
3666 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3667 | } |
1da177e4 LT |
3668 | EXPORT_SYMBOL(interruptible_sleep_on); |
3669 | ||
0fec171c | 3670 | long __sched |
95cdf3b7 | 3671 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3672 | { |
8cbbe86d | 3673 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3674 | } |
1da177e4 LT |
3675 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3676 | ||
0fec171c | 3677 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3678 | { |
8cbbe86d | 3679 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3680 | } |
1da177e4 LT |
3681 | EXPORT_SYMBOL(sleep_on); |
3682 | ||
0fec171c | 3683 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3684 | { |
8cbbe86d | 3685 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3686 | } |
1da177e4 LT |
3687 | EXPORT_SYMBOL(sleep_on_timeout); |
3688 | ||
b29739f9 IM |
3689 | #ifdef CONFIG_RT_MUTEXES |
3690 | ||
3691 | /* | |
3692 | * rt_mutex_setprio - set the current priority of a task | |
3693 | * @p: task | |
3694 | * @prio: prio value (kernel-internal form) | |
3695 | * | |
3696 | * This function changes the 'effective' priority of a task. It does | |
3697 | * not touch ->normal_prio like __setscheduler(). | |
3698 | * | |
3699 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3700 | */ | |
36c8b586 | 3701 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3702 | { |
83b699ed | 3703 | int oldprio, on_rq, running; |
70b97a7f | 3704 | struct rq *rq; |
83ab0aa0 | 3705 | const struct sched_class *prev_class; |
b29739f9 IM |
3706 | |
3707 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3708 | ||
0122ec5b | 3709 | rq = __task_rq_lock(p); |
b29739f9 | 3710 | |
a8027073 | 3711 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3712 | oldprio = p->prio; |
83ab0aa0 | 3713 | prev_class = p->sched_class; |
fd2f4419 | 3714 | on_rq = p->on_rq; |
051a1d1a | 3715 | running = task_current(rq, p); |
0e1f3483 | 3716 | if (on_rq) |
69be72c1 | 3717 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3718 | if (running) |
3719 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3720 | |
3721 | if (rt_prio(prio)) | |
3722 | p->sched_class = &rt_sched_class; | |
3723 | else | |
3724 | p->sched_class = &fair_sched_class; | |
3725 | ||
b29739f9 IM |
3726 | p->prio = prio; |
3727 | ||
0e1f3483 HS |
3728 | if (running) |
3729 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3730 | if (on_rq) |
371fd7e7 | 3731 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3732 | |
da7a735e | 3733 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3734 | __task_rq_unlock(rq); |
b29739f9 IM |
3735 | } |
3736 | ||
3737 | #endif | |
3738 | ||
36c8b586 | 3739 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3740 | { |
dd41f596 | 3741 | int old_prio, delta, on_rq; |
1da177e4 | 3742 | unsigned long flags; |
70b97a7f | 3743 | struct rq *rq; |
1da177e4 LT |
3744 | |
3745 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3746 | return; | |
3747 | /* | |
3748 | * We have to be careful, if called from sys_setpriority(), | |
3749 | * the task might be in the middle of scheduling on another CPU. | |
3750 | */ | |
3751 | rq = task_rq_lock(p, &flags); | |
3752 | /* | |
3753 | * The RT priorities are set via sched_setscheduler(), but we still | |
3754 | * allow the 'normal' nice value to be set - but as expected | |
3755 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3756 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3757 | */ |
e05606d3 | 3758 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3759 | p->static_prio = NICE_TO_PRIO(nice); |
3760 | goto out_unlock; | |
3761 | } | |
fd2f4419 | 3762 | on_rq = p->on_rq; |
c09595f6 | 3763 | if (on_rq) |
69be72c1 | 3764 | dequeue_task(rq, p, 0); |
1da177e4 | 3765 | |
1da177e4 | 3766 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3767 | set_load_weight(p); |
b29739f9 IM |
3768 | old_prio = p->prio; |
3769 | p->prio = effective_prio(p); | |
3770 | delta = p->prio - old_prio; | |
1da177e4 | 3771 | |
dd41f596 | 3772 | if (on_rq) { |
371fd7e7 | 3773 | enqueue_task(rq, p, 0); |
1da177e4 | 3774 | /* |
d5f9f942 AM |
3775 | * If the task increased its priority or is running and |
3776 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3777 | */ |
d5f9f942 | 3778 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3779 | resched_task(rq->curr); |
3780 | } | |
3781 | out_unlock: | |
0122ec5b | 3782 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3783 | } |
1da177e4 LT |
3784 | EXPORT_SYMBOL(set_user_nice); |
3785 | ||
e43379f1 MM |
3786 | /* |
3787 | * can_nice - check if a task can reduce its nice value | |
3788 | * @p: task | |
3789 | * @nice: nice value | |
3790 | */ | |
36c8b586 | 3791 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3792 | { |
024f4747 MM |
3793 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3794 | int nice_rlim = 20 - nice; | |
48f24c4d | 3795 | |
78d7d407 | 3796 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3797 | capable(CAP_SYS_NICE)); |
3798 | } | |
3799 | ||
1da177e4 LT |
3800 | #ifdef __ARCH_WANT_SYS_NICE |
3801 | ||
3802 | /* | |
3803 | * sys_nice - change the priority of the current process. | |
3804 | * @increment: priority increment | |
3805 | * | |
3806 | * sys_setpriority is a more generic, but much slower function that | |
3807 | * does similar things. | |
3808 | */ | |
5add95d4 | 3809 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3810 | { |
48f24c4d | 3811 | long nice, retval; |
1da177e4 LT |
3812 | |
3813 | /* | |
3814 | * Setpriority might change our priority at the same moment. | |
3815 | * We don't have to worry. Conceptually one call occurs first | |
3816 | * and we have a single winner. | |
3817 | */ | |
e43379f1 MM |
3818 | if (increment < -40) |
3819 | increment = -40; | |
1da177e4 LT |
3820 | if (increment > 40) |
3821 | increment = 40; | |
3822 | ||
2b8f836f | 3823 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3824 | if (nice < -20) |
3825 | nice = -20; | |
3826 | if (nice > 19) | |
3827 | nice = 19; | |
3828 | ||
e43379f1 MM |
3829 | if (increment < 0 && !can_nice(current, nice)) |
3830 | return -EPERM; | |
3831 | ||
1da177e4 LT |
3832 | retval = security_task_setnice(current, nice); |
3833 | if (retval) | |
3834 | return retval; | |
3835 | ||
3836 | set_user_nice(current, nice); | |
3837 | return 0; | |
3838 | } | |
3839 | ||
3840 | #endif | |
3841 | ||
3842 | /** | |
3843 | * task_prio - return the priority value of a given task. | |
3844 | * @p: the task in question. | |
3845 | * | |
3846 | * This is the priority value as seen by users in /proc. | |
3847 | * RT tasks are offset by -200. Normal tasks are centered | |
3848 | * around 0, value goes from -16 to +15. | |
3849 | */ | |
36c8b586 | 3850 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3851 | { |
3852 | return p->prio - MAX_RT_PRIO; | |
3853 | } | |
3854 | ||
3855 | /** | |
3856 | * task_nice - return the nice value of a given task. | |
3857 | * @p: the task in question. | |
3858 | */ | |
36c8b586 | 3859 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3860 | { |
3861 | return TASK_NICE(p); | |
3862 | } | |
150d8bed | 3863 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3864 | |
3865 | /** | |
3866 | * idle_cpu - is a given cpu idle currently? | |
3867 | * @cpu: the processor in question. | |
3868 | */ | |
3869 | int idle_cpu(int cpu) | |
3870 | { | |
908a3283 TG |
3871 | struct rq *rq = cpu_rq(cpu); |
3872 | ||
3873 | if (rq->curr != rq->idle) | |
3874 | return 0; | |
3875 | ||
3876 | if (rq->nr_running) | |
3877 | return 0; | |
3878 | ||
3879 | #ifdef CONFIG_SMP | |
3880 | if (!llist_empty(&rq->wake_list)) | |
3881 | return 0; | |
3882 | #endif | |
3883 | ||
3884 | return 1; | |
1da177e4 LT |
3885 | } |
3886 | ||
1da177e4 LT |
3887 | /** |
3888 | * idle_task - return the idle task for a given cpu. | |
3889 | * @cpu: the processor in question. | |
3890 | */ | |
36c8b586 | 3891 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3892 | { |
3893 | return cpu_rq(cpu)->idle; | |
3894 | } | |
3895 | ||
3896 | /** | |
3897 | * find_process_by_pid - find a process with a matching PID value. | |
3898 | * @pid: the pid in question. | |
3899 | */ | |
a9957449 | 3900 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3901 | { |
228ebcbe | 3902 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3903 | } |
3904 | ||
3905 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3906 | static void |
3907 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3908 | { |
1da177e4 LT |
3909 | p->policy = policy; |
3910 | p->rt_priority = prio; | |
b29739f9 IM |
3911 | p->normal_prio = normal_prio(p); |
3912 | /* we are holding p->pi_lock already */ | |
3913 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3914 | if (rt_prio(p->prio)) |
3915 | p->sched_class = &rt_sched_class; | |
3916 | else | |
3917 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3918 | set_load_weight(p); |
1da177e4 LT |
3919 | } |
3920 | ||
c69e8d9c DH |
3921 | /* |
3922 | * check the target process has a UID that matches the current process's | |
3923 | */ | |
3924 | static bool check_same_owner(struct task_struct *p) | |
3925 | { | |
3926 | const struct cred *cred = current_cred(), *pcred; | |
3927 | bool match; | |
3928 | ||
3929 | rcu_read_lock(); | |
3930 | pcred = __task_cred(p); | |
b0e77598 SH |
3931 | if (cred->user->user_ns == pcred->user->user_ns) |
3932 | match = (cred->euid == pcred->euid || | |
3933 | cred->euid == pcred->uid); | |
3934 | else | |
3935 | match = false; | |
c69e8d9c DH |
3936 | rcu_read_unlock(); |
3937 | return match; | |
3938 | } | |
3939 | ||
961ccddd | 3940 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3941 | const struct sched_param *param, bool user) |
1da177e4 | 3942 | { |
83b699ed | 3943 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3944 | unsigned long flags; |
83ab0aa0 | 3945 | const struct sched_class *prev_class; |
70b97a7f | 3946 | struct rq *rq; |
ca94c442 | 3947 | int reset_on_fork; |
1da177e4 | 3948 | |
66e5393a SR |
3949 | /* may grab non-irq protected spin_locks */ |
3950 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3951 | recheck: |
3952 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3953 | if (policy < 0) { |
3954 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3955 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3956 | } else { |
3957 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3958 | policy &= ~SCHED_RESET_ON_FORK; | |
3959 | ||
3960 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3961 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3962 | policy != SCHED_IDLE) | |
3963 | return -EINVAL; | |
3964 | } | |
3965 | ||
1da177e4 LT |
3966 | /* |
3967 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3968 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3969 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3970 | */ |
3971 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3972 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3973 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3974 | return -EINVAL; |
e05606d3 | 3975 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3976 | return -EINVAL; |
3977 | ||
37e4ab3f OC |
3978 | /* |
3979 | * Allow unprivileged RT tasks to decrease priority: | |
3980 | */ | |
961ccddd | 3981 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3982 | if (rt_policy(policy)) { |
a44702e8 ON |
3983 | unsigned long rlim_rtprio = |
3984 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3985 | |
3986 | /* can't set/change the rt policy */ | |
3987 | if (policy != p->policy && !rlim_rtprio) | |
3988 | return -EPERM; | |
3989 | ||
3990 | /* can't increase priority */ | |
3991 | if (param->sched_priority > p->rt_priority && | |
3992 | param->sched_priority > rlim_rtprio) | |
3993 | return -EPERM; | |
3994 | } | |
c02aa73b | 3995 | |
dd41f596 | 3996 | /* |
c02aa73b DH |
3997 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3998 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3999 | */ |
c02aa73b DH |
4000 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4001 | if (!can_nice(p, TASK_NICE(p))) | |
4002 | return -EPERM; | |
4003 | } | |
5fe1d75f | 4004 | |
37e4ab3f | 4005 | /* can't change other user's priorities */ |
c69e8d9c | 4006 | if (!check_same_owner(p)) |
37e4ab3f | 4007 | return -EPERM; |
ca94c442 LP |
4008 | |
4009 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4010 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4011 | return -EPERM; | |
37e4ab3f | 4012 | } |
1da177e4 | 4013 | |
725aad24 | 4014 | if (user) { |
b0ae1981 | 4015 | retval = security_task_setscheduler(p); |
725aad24 JF |
4016 | if (retval) |
4017 | return retval; | |
4018 | } | |
4019 | ||
b29739f9 IM |
4020 | /* |
4021 | * make sure no PI-waiters arrive (or leave) while we are | |
4022 | * changing the priority of the task: | |
0122ec5b | 4023 | * |
25985edc | 4024 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4025 | * runqueue lock must be held. |
4026 | */ | |
0122ec5b | 4027 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 4028 | |
34f971f6 PZ |
4029 | /* |
4030 | * Changing the policy of the stop threads its a very bad idea | |
4031 | */ | |
4032 | if (p == rq->stop) { | |
0122ec5b | 4033 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
4034 | return -EINVAL; |
4035 | } | |
4036 | ||
a51e9198 DF |
4037 | /* |
4038 | * If not changing anything there's no need to proceed further: | |
4039 | */ | |
4040 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
4041 | param->sched_priority == p->rt_priority))) { | |
4042 | ||
4043 | __task_rq_unlock(rq); | |
4044 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4045 | return 0; | |
4046 | } | |
4047 | ||
dc61b1d6 PZ |
4048 | #ifdef CONFIG_RT_GROUP_SCHED |
4049 | if (user) { | |
4050 | /* | |
4051 | * Do not allow realtime tasks into groups that have no runtime | |
4052 | * assigned. | |
4053 | */ | |
4054 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4055 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4056 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4057 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4058 | return -EPERM; |
4059 | } | |
4060 | } | |
4061 | #endif | |
4062 | ||
1da177e4 LT |
4063 | /* recheck policy now with rq lock held */ |
4064 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4065 | policy = oldpolicy = -1; | |
0122ec5b | 4066 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4067 | goto recheck; |
4068 | } | |
fd2f4419 | 4069 | on_rq = p->on_rq; |
051a1d1a | 4070 | running = task_current(rq, p); |
0e1f3483 | 4071 | if (on_rq) |
2e1cb74a | 4072 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4073 | if (running) |
4074 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4075 | |
ca94c442 LP |
4076 | p->sched_reset_on_fork = reset_on_fork; |
4077 | ||
1da177e4 | 4078 | oldprio = p->prio; |
83ab0aa0 | 4079 | prev_class = p->sched_class; |
dd41f596 | 4080 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4081 | |
0e1f3483 HS |
4082 | if (running) |
4083 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4084 | if (on_rq) |
dd41f596 | 4085 | activate_task(rq, p, 0); |
cb469845 | 4086 | |
da7a735e | 4087 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4088 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4089 | |
95e02ca9 TG |
4090 | rt_mutex_adjust_pi(p); |
4091 | ||
1da177e4 LT |
4092 | return 0; |
4093 | } | |
961ccddd RR |
4094 | |
4095 | /** | |
4096 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4097 | * @p: the task in question. | |
4098 | * @policy: new policy. | |
4099 | * @param: structure containing the new RT priority. | |
4100 | * | |
4101 | * NOTE that the task may be already dead. | |
4102 | */ | |
4103 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4104 | const struct sched_param *param) |
961ccddd RR |
4105 | { |
4106 | return __sched_setscheduler(p, policy, param, true); | |
4107 | } | |
1da177e4 LT |
4108 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4109 | ||
961ccddd RR |
4110 | /** |
4111 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4112 | * @p: the task in question. | |
4113 | * @policy: new policy. | |
4114 | * @param: structure containing the new RT priority. | |
4115 | * | |
4116 | * Just like sched_setscheduler, only don't bother checking if the | |
4117 | * current context has permission. For example, this is needed in | |
4118 | * stop_machine(): we create temporary high priority worker threads, | |
4119 | * but our caller might not have that capability. | |
4120 | */ | |
4121 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4122 | const struct sched_param *param) |
961ccddd RR |
4123 | { |
4124 | return __sched_setscheduler(p, policy, param, false); | |
4125 | } | |
4126 | ||
95cdf3b7 IM |
4127 | static int |
4128 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4129 | { |
1da177e4 LT |
4130 | struct sched_param lparam; |
4131 | struct task_struct *p; | |
36c8b586 | 4132 | int retval; |
1da177e4 LT |
4133 | |
4134 | if (!param || pid < 0) | |
4135 | return -EINVAL; | |
4136 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4137 | return -EFAULT; | |
5fe1d75f ON |
4138 | |
4139 | rcu_read_lock(); | |
4140 | retval = -ESRCH; | |
1da177e4 | 4141 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4142 | if (p != NULL) |
4143 | retval = sched_setscheduler(p, policy, &lparam); | |
4144 | rcu_read_unlock(); | |
36c8b586 | 4145 | |
1da177e4 LT |
4146 | return retval; |
4147 | } | |
4148 | ||
4149 | /** | |
4150 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4151 | * @pid: the pid in question. | |
4152 | * @policy: new policy. | |
4153 | * @param: structure containing the new RT priority. | |
4154 | */ | |
5add95d4 HC |
4155 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4156 | struct sched_param __user *, param) | |
1da177e4 | 4157 | { |
c21761f1 JB |
4158 | /* negative values for policy are not valid */ |
4159 | if (policy < 0) | |
4160 | return -EINVAL; | |
4161 | ||
1da177e4 LT |
4162 | return do_sched_setscheduler(pid, policy, param); |
4163 | } | |
4164 | ||
4165 | /** | |
4166 | * sys_sched_setparam - set/change the RT priority of a thread | |
4167 | * @pid: the pid in question. | |
4168 | * @param: structure containing the new RT priority. | |
4169 | */ | |
5add95d4 | 4170 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4171 | { |
4172 | return do_sched_setscheduler(pid, -1, param); | |
4173 | } | |
4174 | ||
4175 | /** | |
4176 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4177 | * @pid: the pid in question. | |
4178 | */ | |
5add95d4 | 4179 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4180 | { |
36c8b586 | 4181 | struct task_struct *p; |
3a5c359a | 4182 | int retval; |
1da177e4 LT |
4183 | |
4184 | if (pid < 0) | |
3a5c359a | 4185 | return -EINVAL; |
1da177e4 LT |
4186 | |
4187 | retval = -ESRCH; | |
5fe85be0 | 4188 | rcu_read_lock(); |
1da177e4 LT |
4189 | p = find_process_by_pid(pid); |
4190 | if (p) { | |
4191 | retval = security_task_getscheduler(p); | |
4192 | if (!retval) | |
ca94c442 LP |
4193 | retval = p->policy |
4194 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4195 | } |
5fe85be0 | 4196 | rcu_read_unlock(); |
1da177e4 LT |
4197 | return retval; |
4198 | } | |
4199 | ||
4200 | /** | |
ca94c442 | 4201 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4202 | * @pid: the pid in question. |
4203 | * @param: structure containing the RT priority. | |
4204 | */ | |
5add95d4 | 4205 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4206 | { |
4207 | struct sched_param lp; | |
36c8b586 | 4208 | struct task_struct *p; |
3a5c359a | 4209 | int retval; |
1da177e4 LT |
4210 | |
4211 | if (!param || pid < 0) | |
3a5c359a | 4212 | return -EINVAL; |
1da177e4 | 4213 | |
5fe85be0 | 4214 | rcu_read_lock(); |
1da177e4 LT |
4215 | p = find_process_by_pid(pid); |
4216 | retval = -ESRCH; | |
4217 | if (!p) | |
4218 | goto out_unlock; | |
4219 | ||
4220 | retval = security_task_getscheduler(p); | |
4221 | if (retval) | |
4222 | goto out_unlock; | |
4223 | ||
4224 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4225 | rcu_read_unlock(); |
1da177e4 LT |
4226 | |
4227 | /* | |
4228 | * This one might sleep, we cannot do it with a spinlock held ... | |
4229 | */ | |
4230 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4231 | ||
1da177e4 LT |
4232 | return retval; |
4233 | ||
4234 | out_unlock: | |
5fe85be0 | 4235 | rcu_read_unlock(); |
1da177e4 LT |
4236 | return retval; |
4237 | } | |
4238 | ||
96f874e2 | 4239 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4240 | { |
5a16f3d3 | 4241 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4242 | struct task_struct *p; |
4243 | int retval; | |
1da177e4 | 4244 | |
95402b38 | 4245 | get_online_cpus(); |
23f5d142 | 4246 | rcu_read_lock(); |
1da177e4 LT |
4247 | |
4248 | p = find_process_by_pid(pid); | |
4249 | if (!p) { | |
23f5d142 | 4250 | rcu_read_unlock(); |
95402b38 | 4251 | put_online_cpus(); |
1da177e4 LT |
4252 | return -ESRCH; |
4253 | } | |
4254 | ||
23f5d142 | 4255 | /* Prevent p going away */ |
1da177e4 | 4256 | get_task_struct(p); |
23f5d142 | 4257 | rcu_read_unlock(); |
1da177e4 | 4258 | |
5a16f3d3 RR |
4259 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4260 | retval = -ENOMEM; | |
4261 | goto out_put_task; | |
4262 | } | |
4263 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4264 | retval = -ENOMEM; | |
4265 | goto out_free_cpus_allowed; | |
4266 | } | |
1da177e4 | 4267 | retval = -EPERM; |
b0e77598 | 4268 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
4269 | goto out_unlock; |
4270 | ||
b0ae1981 | 4271 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4272 | if (retval) |
4273 | goto out_unlock; | |
4274 | ||
5a16f3d3 RR |
4275 | cpuset_cpus_allowed(p, cpus_allowed); |
4276 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4277 | again: |
5a16f3d3 | 4278 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4279 | |
8707d8b8 | 4280 | if (!retval) { |
5a16f3d3 RR |
4281 | cpuset_cpus_allowed(p, cpus_allowed); |
4282 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4283 | /* |
4284 | * We must have raced with a concurrent cpuset | |
4285 | * update. Just reset the cpus_allowed to the | |
4286 | * cpuset's cpus_allowed | |
4287 | */ | |
5a16f3d3 | 4288 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4289 | goto again; |
4290 | } | |
4291 | } | |
1da177e4 | 4292 | out_unlock: |
5a16f3d3 RR |
4293 | free_cpumask_var(new_mask); |
4294 | out_free_cpus_allowed: | |
4295 | free_cpumask_var(cpus_allowed); | |
4296 | out_put_task: | |
1da177e4 | 4297 | put_task_struct(p); |
95402b38 | 4298 | put_online_cpus(); |
1da177e4 LT |
4299 | return retval; |
4300 | } | |
4301 | ||
4302 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4303 | struct cpumask *new_mask) |
1da177e4 | 4304 | { |
96f874e2 RR |
4305 | if (len < cpumask_size()) |
4306 | cpumask_clear(new_mask); | |
4307 | else if (len > cpumask_size()) | |
4308 | len = cpumask_size(); | |
4309 | ||
1da177e4 LT |
4310 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4311 | } | |
4312 | ||
4313 | /** | |
4314 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4315 | * @pid: pid of the process | |
4316 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4317 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4318 | */ | |
5add95d4 HC |
4319 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4320 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4321 | { |
5a16f3d3 | 4322 | cpumask_var_t new_mask; |
1da177e4 LT |
4323 | int retval; |
4324 | ||
5a16f3d3 RR |
4325 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4326 | return -ENOMEM; | |
1da177e4 | 4327 | |
5a16f3d3 RR |
4328 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4329 | if (retval == 0) | |
4330 | retval = sched_setaffinity(pid, new_mask); | |
4331 | free_cpumask_var(new_mask); | |
4332 | return retval; | |
1da177e4 LT |
4333 | } |
4334 | ||
96f874e2 | 4335 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4336 | { |
36c8b586 | 4337 | struct task_struct *p; |
31605683 | 4338 | unsigned long flags; |
1da177e4 | 4339 | int retval; |
1da177e4 | 4340 | |
95402b38 | 4341 | get_online_cpus(); |
23f5d142 | 4342 | rcu_read_lock(); |
1da177e4 LT |
4343 | |
4344 | retval = -ESRCH; | |
4345 | p = find_process_by_pid(pid); | |
4346 | if (!p) | |
4347 | goto out_unlock; | |
4348 | ||
e7834f8f DQ |
4349 | retval = security_task_getscheduler(p); |
4350 | if (retval) | |
4351 | goto out_unlock; | |
4352 | ||
013fdb80 | 4353 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4354 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4355 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4356 | |
4357 | out_unlock: | |
23f5d142 | 4358 | rcu_read_unlock(); |
95402b38 | 4359 | put_online_cpus(); |
1da177e4 | 4360 | |
9531b62f | 4361 | return retval; |
1da177e4 LT |
4362 | } |
4363 | ||
4364 | /** | |
4365 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4366 | * @pid: pid of the process | |
4367 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4368 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4369 | */ | |
5add95d4 HC |
4370 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4371 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4372 | { |
4373 | int ret; | |
f17c8607 | 4374 | cpumask_var_t mask; |
1da177e4 | 4375 | |
84fba5ec | 4376 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4377 | return -EINVAL; |
4378 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4379 | return -EINVAL; |
4380 | ||
f17c8607 RR |
4381 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4382 | return -ENOMEM; | |
1da177e4 | 4383 | |
f17c8607 RR |
4384 | ret = sched_getaffinity(pid, mask); |
4385 | if (ret == 0) { | |
8bc037fb | 4386 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4387 | |
4388 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4389 | ret = -EFAULT; |
4390 | else | |
cd3d8031 | 4391 | ret = retlen; |
f17c8607 RR |
4392 | } |
4393 | free_cpumask_var(mask); | |
1da177e4 | 4394 | |
f17c8607 | 4395 | return ret; |
1da177e4 LT |
4396 | } |
4397 | ||
4398 | /** | |
4399 | * sys_sched_yield - yield the current processor to other threads. | |
4400 | * | |
dd41f596 IM |
4401 | * This function yields the current CPU to other tasks. If there are no |
4402 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4403 | */ |
5add95d4 | 4404 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4405 | { |
70b97a7f | 4406 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4407 | |
2d72376b | 4408 | schedstat_inc(rq, yld_count); |
4530d7ab | 4409 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4410 | |
4411 | /* | |
4412 | * Since we are going to call schedule() anyway, there's | |
4413 | * no need to preempt or enable interrupts: | |
4414 | */ | |
4415 | __release(rq->lock); | |
8a25d5de | 4416 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4417 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4418 | preempt_enable_no_resched(); |
4419 | ||
4420 | schedule(); | |
4421 | ||
4422 | return 0; | |
4423 | } | |
4424 | ||
d86ee480 PZ |
4425 | static inline int should_resched(void) |
4426 | { | |
4427 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4428 | } | |
4429 | ||
e7b38404 | 4430 | static void __cond_resched(void) |
1da177e4 | 4431 | { |
e7aaaa69 | 4432 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4433 | __schedule(); |
e7aaaa69 | 4434 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4435 | } |
4436 | ||
02b67cc3 | 4437 | int __sched _cond_resched(void) |
1da177e4 | 4438 | { |
d86ee480 | 4439 | if (should_resched()) { |
1da177e4 LT |
4440 | __cond_resched(); |
4441 | return 1; | |
4442 | } | |
4443 | return 0; | |
4444 | } | |
02b67cc3 | 4445 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4446 | |
4447 | /* | |
613afbf8 | 4448 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4449 | * call schedule, and on return reacquire the lock. |
4450 | * | |
41a2d6cf | 4451 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4452 | * operations here to prevent schedule() from being called twice (once via |
4453 | * spin_unlock(), once by hand). | |
4454 | */ | |
613afbf8 | 4455 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4456 | { |
d86ee480 | 4457 | int resched = should_resched(); |
6df3cecb JK |
4458 | int ret = 0; |
4459 | ||
f607c668 PZ |
4460 | lockdep_assert_held(lock); |
4461 | ||
95c354fe | 4462 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4463 | spin_unlock(lock); |
d86ee480 | 4464 | if (resched) |
95c354fe NP |
4465 | __cond_resched(); |
4466 | else | |
4467 | cpu_relax(); | |
6df3cecb | 4468 | ret = 1; |
1da177e4 | 4469 | spin_lock(lock); |
1da177e4 | 4470 | } |
6df3cecb | 4471 | return ret; |
1da177e4 | 4472 | } |
613afbf8 | 4473 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4474 | |
613afbf8 | 4475 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4476 | { |
4477 | BUG_ON(!in_softirq()); | |
4478 | ||
d86ee480 | 4479 | if (should_resched()) { |
98d82567 | 4480 | local_bh_enable(); |
1da177e4 LT |
4481 | __cond_resched(); |
4482 | local_bh_disable(); | |
4483 | return 1; | |
4484 | } | |
4485 | return 0; | |
4486 | } | |
613afbf8 | 4487 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4488 | |
1da177e4 LT |
4489 | /** |
4490 | * yield - yield the current processor to other threads. | |
4491 | * | |
72fd4a35 | 4492 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4493 | * thread runnable and calls sys_sched_yield(). |
4494 | */ | |
4495 | void __sched yield(void) | |
4496 | { | |
4497 | set_current_state(TASK_RUNNING); | |
4498 | sys_sched_yield(); | |
4499 | } | |
1da177e4 LT |
4500 | EXPORT_SYMBOL(yield); |
4501 | ||
d95f4122 MG |
4502 | /** |
4503 | * yield_to - yield the current processor to another thread in | |
4504 | * your thread group, or accelerate that thread toward the | |
4505 | * processor it's on. | |
16addf95 RD |
4506 | * @p: target task |
4507 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4508 | * |
4509 | * It's the caller's job to ensure that the target task struct | |
4510 | * can't go away on us before we can do any checks. | |
4511 | * | |
4512 | * Returns true if we indeed boosted the target task. | |
4513 | */ | |
4514 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4515 | { | |
4516 | struct task_struct *curr = current; | |
4517 | struct rq *rq, *p_rq; | |
4518 | unsigned long flags; | |
4519 | bool yielded = 0; | |
4520 | ||
4521 | local_irq_save(flags); | |
4522 | rq = this_rq(); | |
4523 | ||
4524 | again: | |
4525 | p_rq = task_rq(p); | |
4526 | double_rq_lock(rq, p_rq); | |
4527 | while (task_rq(p) != p_rq) { | |
4528 | double_rq_unlock(rq, p_rq); | |
4529 | goto again; | |
4530 | } | |
4531 | ||
4532 | if (!curr->sched_class->yield_to_task) | |
4533 | goto out; | |
4534 | ||
4535 | if (curr->sched_class != p->sched_class) | |
4536 | goto out; | |
4537 | ||
4538 | if (task_running(p_rq, p) || p->state) | |
4539 | goto out; | |
4540 | ||
4541 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4542 | if (yielded) { |
d95f4122 | 4543 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4544 | /* |
4545 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4546 | * fairness. | |
4547 | */ | |
4548 | if (preempt && rq != p_rq) | |
4549 | resched_task(p_rq->curr); | |
4550 | } | |
d95f4122 MG |
4551 | |
4552 | out: | |
4553 | double_rq_unlock(rq, p_rq); | |
4554 | local_irq_restore(flags); | |
4555 | ||
4556 | if (yielded) | |
4557 | schedule(); | |
4558 | ||
4559 | return yielded; | |
4560 | } | |
4561 | EXPORT_SYMBOL_GPL(yield_to); | |
4562 | ||
1da177e4 | 4563 | /* |
41a2d6cf | 4564 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4565 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4566 | */ |
4567 | void __sched io_schedule(void) | |
4568 | { | |
54d35f29 | 4569 | struct rq *rq = raw_rq(); |
1da177e4 | 4570 | |
0ff92245 | 4571 | delayacct_blkio_start(); |
1da177e4 | 4572 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4573 | blk_flush_plug(current); |
8f0dfc34 | 4574 | current->in_iowait = 1; |
1da177e4 | 4575 | schedule(); |
8f0dfc34 | 4576 | current->in_iowait = 0; |
1da177e4 | 4577 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4578 | delayacct_blkio_end(); |
1da177e4 | 4579 | } |
1da177e4 LT |
4580 | EXPORT_SYMBOL(io_schedule); |
4581 | ||
4582 | long __sched io_schedule_timeout(long timeout) | |
4583 | { | |
54d35f29 | 4584 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4585 | long ret; |
4586 | ||
0ff92245 | 4587 | delayacct_blkio_start(); |
1da177e4 | 4588 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4589 | blk_flush_plug(current); |
8f0dfc34 | 4590 | current->in_iowait = 1; |
1da177e4 | 4591 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4592 | current->in_iowait = 0; |
1da177e4 | 4593 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4594 | delayacct_blkio_end(); |
1da177e4 LT |
4595 | return ret; |
4596 | } | |
4597 | ||
4598 | /** | |
4599 | * sys_sched_get_priority_max - return maximum RT priority. | |
4600 | * @policy: scheduling class. | |
4601 | * | |
4602 | * this syscall returns the maximum rt_priority that can be used | |
4603 | * by a given scheduling class. | |
4604 | */ | |
5add95d4 | 4605 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4606 | { |
4607 | int ret = -EINVAL; | |
4608 | ||
4609 | switch (policy) { | |
4610 | case SCHED_FIFO: | |
4611 | case SCHED_RR: | |
4612 | ret = MAX_USER_RT_PRIO-1; | |
4613 | break; | |
4614 | case SCHED_NORMAL: | |
b0a9499c | 4615 | case SCHED_BATCH: |
dd41f596 | 4616 | case SCHED_IDLE: |
1da177e4 LT |
4617 | ret = 0; |
4618 | break; | |
4619 | } | |
4620 | return ret; | |
4621 | } | |
4622 | ||
4623 | /** | |
4624 | * sys_sched_get_priority_min - return minimum RT priority. | |
4625 | * @policy: scheduling class. | |
4626 | * | |
4627 | * this syscall returns the minimum rt_priority that can be used | |
4628 | * by a given scheduling class. | |
4629 | */ | |
5add95d4 | 4630 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4631 | { |
4632 | int ret = -EINVAL; | |
4633 | ||
4634 | switch (policy) { | |
4635 | case SCHED_FIFO: | |
4636 | case SCHED_RR: | |
4637 | ret = 1; | |
4638 | break; | |
4639 | case SCHED_NORMAL: | |
b0a9499c | 4640 | case SCHED_BATCH: |
dd41f596 | 4641 | case SCHED_IDLE: |
1da177e4 LT |
4642 | ret = 0; |
4643 | } | |
4644 | return ret; | |
4645 | } | |
4646 | ||
4647 | /** | |
4648 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4649 | * @pid: pid of the process. | |
4650 | * @interval: userspace pointer to the timeslice value. | |
4651 | * | |
4652 | * this syscall writes the default timeslice value of a given process | |
4653 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4654 | */ | |
17da2bd9 | 4655 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4656 | struct timespec __user *, interval) |
1da177e4 | 4657 | { |
36c8b586 | 4658 | struct task_struct *p; |
a4ec24b4 | 4659 | unsigned int time_slice; |
dba091b9 TG |
4660 | unsigned long flags; |
4661 | struct rq *rq; | |
3a5c359a | 4662 | int retval; |
1da177e4 | 4663 | struct timespec t; |
1da177e4 LT |
4664 | |
4665 | if (pid < 0) | |
3a5c359a | 4666 | return -EINVAL; |
1da177e4 LT |
4667 | |
4668 | retval = -ESRCH; | |
1a551ae7 | 4669 | rcu_read_lock(); |
1da177e4 LT |
4670 | p = find_process_by_pid(pid); |
4671 | if (!p) | |
4672 | goto out_unlock; | |
4673 | ||
4674 | retval = security_task_getscheduler(p); | |
4675 | if (retval) | |
4676 | goto out_unlock; | |
4677 | ||
dba091b9 TG |
4678 | rq = task_rq_lock(p, &flags); |
4679 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4680 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4681 | |
1a551ae7 | 4682 | rcu_read_unlock(); |
a4ec24b4 | 4683 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4684 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4685 | return retval; |
3a5c359a | 4686 | |
1da177e4 | 4687 | out_unlock: |
1a551ae7 | 4688 | rcu_read_unlock(); |
1da177e4 LT |
4689 | return retval; |
4690 | } | |
4691 | ||
7c731e0a | 4692 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4693 | |
82a1fcb9 | 4694 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4695 | { |
1da177e4 | 4696 | unsigned long free = 0; |
36c8b586 | 4697 | unsigned state; |
1da177e4 | 4698 | |
1da177e4 | 4699 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4700 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4701 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4702 | #if BITS_PER_LONG == 32 |
1da177e4 | 4703 | if (state == TASK_RUNNING) |
3df0fc5b | 4704 | printk(KERN_CONT " running "); |
1da177e4 | 4705 | else |
3df0fc5b | 4706 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4707 | #else |
4708 | if (state == TASK_RUNNING) | |
3df0fc5b | 4709 | printk(KERN_CONT " running task "); |
1da177e4 | 4710 | else |
3df0fc5b | 4711 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4712 | #endif |
4713 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4714 | free = stack_not_used(p); |
1da177e4 | 4715 | #endif |
3df0fc5b | 4716 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
4717 | task_pid_nr(p), task_pid_nr(p->real_parent), |
4718 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 4719 | |
5fb5e6de | 4720 | show_stack(p, NULL); |
1da177e4 LT |
4721 | } |
4722 | ||
e59e2ae2 | 4723 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4724 | { |
36c8b586 | 4725 | struct task_struct *g, *p; |
1da177e4 | 4726 | |
4bd77321 | 4727 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4728 | printk(KERN_INFO |
4729 | " task PC stack pid father\n"); | |
1da177e4 | 4730 | #else |
3df0fc5b PZ |
4731 | printk(KERN_INFO |
4732 | " task PC stack pid father\n"); | |
1da177e4 | 4733 | #endif |
510f5acc | 4734 | rcu_read_lock(); |
1da177e4 LT |
4735 | do_each_thread(g, p) { |
4736 | /* | |
4737 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4738 | * console might take a lot of time: |
1da177e4 LT |
4739 | */ |
4740 | touch_nmi_watchdog(); | |
39bc89fd | 4741 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4742 | sched_show_task(p); |
1da177e4 LT |
4743 | } while_each_thread(g, p); |
4744 | ||
04c9167f JF |
4745 | touch_all_softlockup_watchdogs(); |
4746 | ||
dd41f596 IM |
4747 | #ifdef CONFIG_SCHED_DEBUG |
4748 | sysrq_sched_debug_show(); | |
4749 | #endif | |
510f5acc | 4750 | rcu_read_unlock(); |
e59e2ae2 IM |
4751 | /* |
4752 | * Only show locks if all tasks are dumped: | |
4753 | */ | |
93335a21 | 4754 | if (!state_filter) |
e59e2ae2 | 4755 | debug_show_all_locks(); |
1da177e4 LT |
4756 | } |
4757 | ||
1df21055 IM |
4758 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4759 | { | |
dd41f596 | 4760 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4761 | } |
4762 | ||
f340c0d1 IM |
4763 | /** |
4764 | * init_idle - set up an idle thread for a given CPU | |
4765 | * @idle: task in question | |
4766 | * @cpu: cpu the idle task belongs to | |
4767 | * | |
4768 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4769 | * flag, to make booting more robust. | |
4770 | */ | |
5c1e1767 | 4771 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4772 | { |
70b97a7f | 4773 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4774 | unsigned long flags; |
4775 | ||
05fa785c | 4776 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4777 | |
dd41f596 | 4778 | __sched_fork(idle); |
06b83b5f | 4779 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4780 | idle->se.exec_start = sched_clock(); |
4781 | ||
1e1b6c51 | 4782 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4783 | /* |
4784 | * We're having a chicken and egg problem, even though we are | |
4785 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4786 | * lockdep check in task_group() will fail. | |
4787 | * | |
4788 | * Similar case to sched_fork(). / Alternatively we could | |
4789 | * use task_rq_lock() here and obtain the other rq->lock. | |
4790 | * | |
4791 | * Silence PROVE_RCU | |
4792 | */ | |
4793 | rcu_read_lock(); | |
dd41f596 | 4794 | __set_task_cpu(idle, cpu); |
6506cf6c | 4795 | rcu_read_unlock(); |
1da177e4 | 4796 | |
1da177e4 | 4797 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4798 | #if defined(CONFIG_SMP) |
4799 | idle->on_cpu = 1; | |
4866cde0 | 4800 | #endif |
05fa785c | 4801 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4802 | |
4803 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4804 | task_thread_info(idle)->preempt_count = 0; |
625f2a37 | 4805 | |
dd41f596 IM |
4806 | /* |
4807 | * The idle tasks have their own, simple scheduling class: | |
4808 | */ | |
4809 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4810 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
4811 | #if defined(CONFIG_SMP) |
4812 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4813 | #endif | |
1da177e4 LT |
4814 | } |
4815 | ||
1da177e4 | 4816 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4817 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4818 | { | |
4819 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4820 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4821 | |
4822 | cpumask_copy(&p->cpus_allowed, new_mask); | |
4823 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
1e1b6c51 KM |
4824 | } |
4825 | ||
1da177e4 LT |
4826 | /* |
4827 | * This is how migration works: | |
4828 | * | |
969c7921 TH |
4829 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4830 | * stop_one_cpu(). | |
4831 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4832 | * off the CPU) | |
4833 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4834 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4835 | * it and puts it into the right queue. |
969c7921 TH |
4836 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4837 | * is done. | |
1da177e4 LT |
4838 | */ |
4839 | ||
4840 | /* | |
4841 | * Change a given task's CPU affinity. Migrate the thread to a | |
4842 | * proper CPU and schedule it away if the CPU it's executing on | |
4843 | * is removed from the allowed bitmask. | |
4844 | * | |
4845 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4846 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4847 | * call is not atomic; no spinlocks may be held. |
4848 | */ | |
96f874e2 | 4849 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4850 | { |
4851 | unsigned long flags; | |
70b97a7f | 4852 | struct rq *rq; |
969c7921 | 4853 | unsigned int dest_cpu; |
48f24c4d | 4854 | int ret = 0; |
1da177e4 LT |
4855 | |
4856 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4857 | |
db44fc01 YZ |
4858 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4859 | goto out; | |
4860 | ||
6ad4c188 | 4861 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4862 | ret = -EINVAL; |
4863 | goto out; | |
4864 | } | |
4865 | ||
db44fc01 | 4866 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4867 | ret = -EINVAL; |
4868 | goto out; | |
4869 | } | |
4870 | ||
1e1b6c51 | 4871 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4872 | |
1da177e4 | 4873 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4874 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4875 | goto out; |
4876 | ||
969c7921 | 4877 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4878 | if (p->on_rq) { |
969c7921 | 4879 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4880 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4881 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4882 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4883 | tlb_migrate_finish(p->mm); |
4884 | return 0; | |
4885 | } | |
4886 | out: | |
0122ec5b | 4887 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4888 | |
1da177e4 LT |
4889 | return ret; |
4890 | } | |
cd8ba7cd | 4891 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4892 | |
4893 | /* | |
41a2d6cf | 4894 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4895 | * this because either it can't run here any more (set_cpus_allowed() |
4896 | * away from this CPU, or CPU going down), or because we're | |
4897 | * attempting to rebalance this task on exec (sched_exec). | |
4898 | * | |
4899 | * So we race with normal scheduler movements, but that's OK, as long | |
4900 | * as the task is no longer on this CPU. | |
efc30814 KK |
4901 | * |
4902 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4903 | */ |
efc30814 | 4904 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4905 | { |
70b97a7f | 4906 | struct rq *rq_dest, *rq_src; |
e2912009 | 4907 | int ret = 0; |
1da177e4 | 4908 | |
e761b772 | 4909 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4910 | return ret; |
1da177e4 LT |
4911 | |
4912 | rq_src = cpu_rq(src_cpu); | |
4913 | rq_dest = cpu_rq(dest_cpu); | |
4914 | ||
0122ec5b | 4915 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4916 | double_rq_lock(rq_src, rq_dest); |
4917 | /* Already moved. */ | |
4918 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4919 | goto done; |
1da177e4 | 4920 | /* Affinity changed (again). */ |
fa17b507 | 4921 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4922 | goto fail; |
1da177e4 | 4923 | |
e2912009 PZ |
4924 | /* |
4925 | * If we're not on a rq, the next wake-up will ensure we're | |
4926 | * placed properly. | |
4927 | */ | |
fd2f4419 | 4928 | if (p->on_rq) { |
2e1cb74a | 4929 | deactivate_task(rq_src, p, 0); |
e2912009 | 4930 | set_task_cpu(p, dest_cpu); |
dd41f596 | 4931 | activate_task(rq_dest, p, 0); |
15afe09b | 4932 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4933 | } |
b1e38734 | 4934 | done: |
efc30814 | 4935 | ret = 1; |
b1e38734 | 4936 | fail: |
1da177e4 | 4937 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4938 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4939 | return ret; |
1da177e4 LT |
4940 | } |
4941 | ||
4942 | /* | |
969c7921 TH |
4943 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4944 | * and performs thread migration by bumping thread off CPU then | |
4945 | * 'pushing' onto another runqueue. | |
1da177e4 | 4946 | */ |
969c7921 | 4947 | static int migration_cpu_stop(void *data) |
1da177e4 | 4948 | { |
969c7921 | 4949 | struct migration_arg *arg = data; |
f7b4cddc | 4950 | |
969c7921 TH |
4951 | /* |
4952 | * The original target cpu might have gone down and we might | |
4953 | * be on another cpu but it doesn't matter. | |
4954 | */ | |
f7b4cddc | 4955 | local_irq_disable(); |
969c7921 | 4956 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4957 | local_irq_enable(); |
1da177e4 | 4958 | return 0; |
f7b4cddc ON |
4959 | } |
4960 | ||
1da177e4 | 4961 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4962 | |
054b9108 | 4963 | /* |
48c5ccae PZ |
4964 | * Ensures that the idle task is using init_mm right before its cpu goes |
4965 | * offline. | |
054b9108 | 4966 | */ |
48c5ccae | 4967 | void idle_task_exit(void) |
1da177e4 | 4968 | { |
48c5ccae | 4969 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4970 | |
48c5ccae | 4971 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4972 | |
48c5ccae PZ |
4973 | if (mm != &init_mm) |
4974 | switch_mm(mm, &init_mm, current); | |
4975 | mmdrop(mm); | |
1da177e4 LT |
4976 | } |
4977 | ||
4978 | /* | |
4979 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
4980 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
4981 | * for performance reasons the counter is not stricly tracking tasks to | |
4982 | * their home CPUs. So we just add the counter to another CPU's counter, | |
4983 | * to keep the global sum constant after CPU-down: | |
4984 | */ | |
70b97a7f | 4985 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 4986 | { |
6ad4c188 | 4987 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 4988 | |
1da177e4 LT |
4989 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
4990 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
4991 | } |
4992 | ||
dd41f596 | 4993 | /* |
48c5ccae | 4994 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 4995 | */ |
48c5ccae | 4996 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 4997 | { |
48c5ccae PZ |
4998 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
4999 | rq->calc_load_active = 0; | |
1da177e4 LT |
5000 | } |
5001 | ||
48f24c4d | 5002 | /* |
48c5ccae PZ |
5003 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5004 | * try_to_wake_up()->select_task_rq(). | |
5005 | * | |
5006 | * Called with rq->lock held even though we'er in stop_machine() and | |
5007 | * there's no concurrency possible, we hold the required locks anyway | |
5008 | * because of lock validation efforts. | |
1da177e4 | 5009 | */ |
48c5ccae | 5010 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5011 | { |
70b97a7f | 5012 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5013 | struct task_struct *next, *stop = rq->stop; |
5014 | int dest_cpu; | |
1da177e4 LT |
5015 | |
5016 | /* | |
48c5ccae PZ |
5017 | * Fudge the rq selection such that the below task selection loop |
5018 | * doesn't get stuck on the currently eligible stop task. | |
5019 | * | |
5020 | * We're currently inside stop_machine() and the rq is either stuck | |
5021 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5022 | * either way we should never end up calling schedule() until we're | |
5023 | * done here. | |
1da177e4 | 5024 | */ |
48c5ccae | 5025 | rq->stop = NULL; |
48f24c4d | 5026 | |
8cb120d3 PT |
5027 | /* Ensure any throttled groups are reachable by pick_next_task */ |
5028 | unthrottle_offline_cfs_rqs(rq); | |
5029 | ||
dd41f596 | 5030 | for ( ; ; ) { |
48c5ccae PZ |
5031 | /* |
5032 | * There's this thread running, bail when that's the only | |
5033 | * remaining thread. | |
5034 | */ | |
5035 | if (rq->nr_running == 1) | |
dd41f596 | 5036 | break; |
48c5ccae | 5037 | |
b67802ea | 5038 | next = pick_next_task(rq); |
48c5ccae | 5039 | BUG_ON(!next); |
79c53799 | 5040 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5041 | |
48c5ccae PZ |
5042 | /* Find suitable destination for @next, with force if needed. */ |
5043 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5044 | raw_spin_unlock(&rq->lock); | |
5045 | ||
5046 | __migrate_task(next, dead_cpu, dest_cpu); | |
5047 | ||
5048 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5049 | } |
dce48a84 | 5050 | |
48c5ccae | 5051 | rq->stop = stop; |
dce48a84 | 5052 | } |
48c5ccae | 5053 | |
1da177e4 LT |
5054 | #endif /* CONFIG_HOTPLUG_CPU */ |
5055 | ||
e692ab53 NP |
5056 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5057 | ||
5058 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5059 | { |
5060 | .procname = "sched_domain", | |
c57baf1e | 5061 | .mode = 0555, |
e0361851 | 5062 | }, |
56992309 | 5063 | {} |
e692ab53 NP |
5064 | }; |
5065 | ||
5066 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5067 | { |
5068 | .procname = "kernel", | |
c57baf1e | 5069 | .mode = 0555, |
e0361851 AD |
5070 | .child = sd_ctl_dir, |
5071 | }, | |
56992309 | 5072 | {} |
e692ab53 NP |
5073 | }; |
5074 | ||
5075 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5076 | { | |
5077 | struct ctl_table *entry = | |
5cf9f062 | 5078 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5079 | |
e692ab53 NP |
5080 | return entry; |
5081 | } | |
5082 | ||
6382bc90 MM |
5083 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5084 | { | |
cd790076 | 5085 | struct ctl_table *entry; |
6382bc90 | 5086 | |
cd790076 MM |
5087 | /* |
5088 | * In the intermediate directories, both the child directory and | |
5089 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5090 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5091 | * static strings and all have proc handlers. |
5092 | */ | |
5093 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5094 | if (entry->child) |
5095 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5096 | if (entry->proc_handler == NULL) |
5097 | kfree(entry->procname); | |
5098 | } | |
6382bc90 MM |
5099 | |
5100 | kfree(*tablep); | |
5101 | *tablep = NULL; | |
5102 | } | |
5103 | ||
e692ab53 | 5104 | static void |
e0361851 | 5105 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5106 | const char *procname, void *data, int maxlen, |
5107 | mode_t mode, proc_handler *proc_handler) | |
5108 | { | |
e692ab53 NP |
5109 | entry->procname = procname; |
5110 | entry->data = data; | |
5111 | entry->maxlen = maxlen; | |
5112 | entry->mode = mode; | |
5113 | entry->proc_handler = proc_handler; | |
5114 | } | |
5115 | ||
5116 | static struct ctl_table * | |
5117 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5118 | { | |
a5d8c348 | 5119 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5120 | |
ad1cdc1d MM |
5121 | if (table == NULL) |
5122 | return NULL; | |
5123 | ||
e0361851 | 5124 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5125 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5126 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5127 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5128 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5129 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5130 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5131 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5132 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5133 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5134 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5135 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5136 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5137 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5138 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5139 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5140 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5141 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5142 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5143 | &sd->cache_nice_tries, |
5144 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5145 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5146 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5147 | set_table_entry(&table[11], "name", sd->name, |
5148 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5149 | /* &table[12] is terminator */ | |
e692ab53 NP |
5150 | |
5151 | return table; | |
5152 | } | |
5153 | ||
9a4e7159 | 5154 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5155 | { |
5156 | struct ctl_table *entry, *table; | |
5157 | struct sched_domain *sd; | |
5158 | int domain_num = 0, i; | |
5159 | char buf[32]; | |
5160 | ||
5161 | for_each_domain(cpu, sd) | |
5162 | domain_num++; | |
5163 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5164 | if (table == NULL) |
5165 | return NULL; | |
e692ab53 NP |
5166 | |
5167 | i = 0; | |
5168 | for_each_domain(cpu, sd) { | |
5169 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5170 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5171 | entry->mode = 0555; |
e692ab53 NP |
5172 | entry->child = sd_alloc_ctl_domain_table(sd); |
5173 | entry++; | |
5174 | i++; | |
5175 | } | |
5176 | return table; | |
5177 | } | |
5178 | ||
5179 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5180 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5181 | { |
6ad4c188 | 5182 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5183 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5184 | char buf[32]; | |
5185 | ||
7378547f MM |
5186 | WARN_ON(sd_ctl_dir[0].child); |
5187 | sd_ctl_dir[0].child = entry; | |
5188 | ||
ad1cdc1d MM |
5189 | if (entry == NULL) |
5190 | return; | |
5191 | ||
6ad4c188 | 5192 | for_each_possible_cpu(i) { |
e692ab53 | 5193 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5194 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5195 | entry->mode = 0555; |
e692ab53 | 5196 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5197 | entry++; |
e692ab53 | 5198 | } |
7378547f MM |
5199 | |
5200 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5201 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5202 | } | |
6382bc90 | 5203 | |
7378547f | 5204 | /* may be called multiple times per register */ |
6382bc90 MM |
5205 | static void unregister_sched_domain_sysctl(void) |
5206 | { | |
7378547f MM |
5207 | if (sd_sysctl_header) |
5208 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5209 | sd_sysctl_header = NULL; |
7378547f MM |
5210 | if (sd_ctl_dir[0].child) |
5211 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5212 | } |
e692ab53 | 5213 | #else |
6382bc90 MM |
5214 | static void register_sched_domain_sysctl(void) |
5215 | { | |
5216 | } | |
5217 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5218 | { |
5219 | } | |
5220 | #endif | |
5221 | ||
1f11eb6a GH |
5222 | static void set_rq_online(struct rq *rq) |
5223 | { | |
5224 | if (!rq->online) { | |
5225 | const struct sched_class *class; | |
5226 | ||
c6c4927b | 5227 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5228 | rq->online = 1; |
5229 | ||
5230 | for_each_class(class) { | |
5231 | if (class->rq_online) | |
5232 | class->rq_online(rq); | |
5233 | } | |
5234 | } | |
5235 | } | |
5236 | ||
5237 | static void set_rq_offline(struct rq *rq) | |
5238 | { | |
5239 | if (rq->online) { | |
5240 | const struct sched_class *class; | |
5241 | ||
5242 | for_each_class(class) { | |
5243 | if (class->rq_offline) | |
5244 | class->rq_offline(rq); | |
5245 | } | |
5246 | ||
c6c4927b | 5247 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5248 | rq->online = 0; |
5249 | } | |
5250 | } | |
5251 | ||
1da177e4 LT |
5252 | /* |
5253 | * migration_call - callback that gets triggered when a CPU is added. | |
5254 | * Here we can start up the necessary migration thread for the new CPU. | |
5255 | */ | |
48f24c4d IM |
5256 | static int __cpuinit |
5257 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5258 | { |
48f24c4d | 5259 | int cpu = (long)hcpu; |
1da177e4 | 5260 | unsigned long flags; |
969c7921 | 5261 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5262 | |
48c5ccae | 5263 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5264 | |
1da177e4 | 5265 | case CPU_UP_PREPARE: |
a468d389 | 5266 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5267 | break; |
48f24c4d | 5268 | |
1da177e4 | 5269 | case CPU_ONLINE: |
1f94ef59 | 5270 | /* Update our root-domain */ |
05fa785c | 5271 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5272 | if (rq->rd) { |
c6c4927b | 5273 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5274 | |
5275 | set_rq_online(rq); | |
1f94ef59 | 5276 | } |
05fa785c | 5277 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5278 | break; |
48f24c4d | 5279 | |
1da177e4 | 5280 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5281 | case CPU_DYING: |
317f3941 | 5282 | sched_ttwu_pending(); |
57d885fe | 5283 | /* Update our root-domain */ |
05fa785c | 5284 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5285 | if (rq->rd) { |
c6c4927b | 5286 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5287 | set_rq_offline(rq); |
57d885fe | 5288 | } |
48c5ccae PZ |
5289 | migrate_tasks(cpu); |
5290 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5291 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5292 | |
5293 | migrate_nr_uninterruptible(rq); | |
5294 | calc_global_load_remove(rq); | |
57d885fe | 5295 | break; |
1da177e4 LT |
5296 | #endif |
5297 | } | |
49c022e6 PZ |
5298 | |
5299 | update_max_interval(); | |
5300 | ||
1da177e4 LT |
5301 | return NOTIFY_OK; |
5302 | } | |
5303 | ||
f38b0820 PM |
5304 | /* |
5305 | * Register at high priority so that task migration (migrate_all_tasks) | |
5306 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5307 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5308 | */ |
26c2143b | 5309 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5310 | .notifier_call = migration_call, |
50a323b7 | 5311 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5312 | }; |
5313 | ||
3a101d05 TH |
5314 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5315 | unsigned long action, void *hcpu) | |
5316 | { | |
5317 | switch (action & ~CPU_TASKS_FROZEN) { | |
5318 | case CPU_ONLINE: | |
5319 | case CPU_DOWN_FAILED: | |
5320 | set_cpu_active((long)hcpu, true); | |
5321 | return NOTIFY_OK; | |
5322 | default: | |
5323 | return NOTIFY_DONE; | |
5324 | } | |
5325 | } | |
5326 | ||
5327 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5328 | unsigned long action, void *hcpu) | |
5329 | { | |
5330 | switch (action & ~CPU_TASKS_FROZEN) { | |
5331 | case CPU_DOWN_PREPARE: | |
5332 | set_cpu_active((long)hcpu, false); | |
5333 | return NOTIFY_OK; | |
5334 | default: | |
5335 | return NOTIFY_DONE; | |
5336 | } | |
5337 | } | |
5338 | ||
7babe8db | 5339 | static int __init migration_init(void) |
1da177e4 LT |
5340 | { |
5341 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5342 | int err; |
48f24c4d | 5343 | |
3a101d05 | 5344 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5345 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5346 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5347 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5348 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5349 | |
3a101d05 TH |
5350 | /* Register cpu active notifiers */ |
5351 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5352 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5353 | ||
a004cd42 | 5354 | return 0; |
1da177e4 | 5355 | } |
7babe8db | 5356 | early_initcall(migration_init); |
1da177e4 LT |
5357 | #endif |
5358 | ||
5359 | #ifdef CONFIG_SMP | |
476f3534 | 5360 | |
4cb98839 PZ |
5361 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5362 | ||
3e9830dc | 5363 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5364 | |
f6630114 MT |
5365 | static __read_mostly int sched_domain_debug_enabled; |
5366 | ||
5367 | static int __init sched_domain_debug_setup(char *str) | |
5368 | { | |
5369 | sched_domain_debug_enabled = 1; | |
5370 | ||
5371 | return 0; | |
5372 | } | |
5373 | early_param("sched_debug", sched_domain_debug_setup); | |
5374 | ||
7c16ec58 | 5375 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5376 | struct cpumask *groupmask) |
1da177e4 | 5377 | { |
4dcf6aff | 5378 | struct sched_group *group = sd->groups; |
434d53b0 | 5379 | char str[256]; |
1da177e4 | 5380 | |
968ea6d8 | 5381 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5382 | cpumask_clear(groupmask); |
4dcf6aff IM |
5383 | |
5384 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5385 | ||
5386 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5387 | printk("does not load-balance\n"); |
4dcf6aff | 5388 | if (sd->parent) |
3df0fc5b PZ |
5389 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5390 | " has parent"); | |
4dcf6aff | 5391 | return -1; |
41c7ce9a NP |
5392 | } |
5393 | ||
3df0fc5b | 5394 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5395 | |
758b2cdc | 5396 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5397 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5398 | "CPU%d\n", cpu); | |
4dcf6aff | 5399 | } |
758b2cdc | 5400 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5401 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5402 | " CPU%d\n", cpu); | |
4dcf6aff | 5403 | } |
1da177e4 | 5404 | |
4dcf6aff | 5405 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5406 | do { |
4dcf6aff | 5407 | if (!group) { |
3df0fc5b PZ |
5408 | printk("\n"); |
5409 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5410 | break; |
5411 | } | |
5412 | ||
9c3f75cb | 5413 | if (!group->sgp->power) { |
3df0fc5b PZ |
5414 | printk(KERN_CONT "\n"); |
5415 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5416 | "set\n"); | |
4dcf6aff IM |
5417 | break; |
5418 | } | |
1da177e4 | 5419 | |
758b2cdc | 5420 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5421 | printk(KERN_CONT "\n"); |
5422 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5423 | break; |
5424 | } | |
1da177e4 | 5425 | |
758b2cdc | 5426 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5427 | printk(KERN_CONT "\n"); |
5428 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5429 | break; |
5430 | } | |
1da177e4 | 5431 | |
758b2cdc | 5432 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5433 | |
968ea6d8 | 5434 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5435 | |
3df0fc5b | 5436 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5437 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5438 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5439 | group->sgp->power); |
381512cf | 5440 | } |
1da177e4 | 5441 | |
4dcf6aff IM |
5442 | group = group->next; |
5443 | } while (group != sd->groups); | |
3df0fc5b | 5444 | printk(KERN_CONT "\n"); |
1da177e4 | 5445 | |
758b2cdc | 5446 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5447 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5448 | |
758b2cdc RR |
5449 | if (sd->parent && |
5450 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5451 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5452 | "of domain->span\n"); | |
4dcf6aff IM |
5453 | return 0; |
5454 | } | |
1da177e4 | 5455 | |
4dcf6aff IM |
5456 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5457 | { | |
5458 | int level = 0; | |
1da177e4 | 5459 | |
f6630114 MT |
5460 | if (!sched_domain_debug_enabled) |
5461 | return; | |
5462 | ||
4dcf6aff IM |
5463 | if (!sd) { |
5464 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5465 | return; | |
5466 | } | |
1da177e4 | 5467 | |
4dcf6aff IM |
5468 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5469 | ||
5470 | for (;;) { | |
4cb98839 | 5471 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5472 | break; |
1da177e4 LT |
5473 | level++; |
5474 | sd = sd->parent; | |
33859f7f | 5475 | if (!sd) |
4dcf6aff IM |
5476 | break; |
5477 | } | |
1da177e4 | 5478 | } |
6d6bc0ad | 5479 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5480 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5481 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5482 | |
1a20ff27 | 5483 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5484 | { |
758b2cdc | 5485 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5486 | return 1; |
5487 | ||
5488 | /* Following flags need at least 2 groups */ | |
5489 | if (sd->flags & (SD_LOAD_BALANCE | | |
5490 | SD_BALANCE_NEWIDLE | | |
5491 | SD_BALANCE_FORK | | |
89c4710e SS |
5492 | SD_BALANCE_EXEC | |
5493 | SD_SHARE_CPUPOWER | | |
5494 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5495 | if (sd->groups != sd->groups->next) |
5496 | return 0; | |
5497 | } | |
5498 | ||
5499 | /* Following flags don't use groups */ | |
c88d5910 | 5500 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5501 | return 0; |
5502 | ||
5503 | return 1; | |
5504 | } | |
5505 | ||
48f24c4d IM |
5506 | static int |
5507 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5508 | { |
5509 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5510 | ||
5511 | if (sd_degenerate(parent)) | |
5512 | return 1; | |
5513 | ||
758b2cdc | 5514 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5515 | return 0; |
5516 | ||
245af2c7 SS |
5517 | /* Flags needing groups don't count if only 1 group in parent */ |
5518 | if (parent->groups == parent->groups->next) { | |
5519 | pflags &= ~(SD_LOAD_BALANCE | | |
5520 | SD_BALANCE_NEWIDLE | | |
5521 | SD_BALANCE_FORK | | |
89c4710e SS |
5522 | SD_BALANCE_EXEC | |
5523 | SD_SHARE_CPUPOWER | | |
5524 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5525 | if (nr_node_ids == 1) |
5526 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5527 | } |
5528 | if (~cflags & pflags) | |
5529 | return 0; | |
5530 | ||
5531 | return 1; | |
5532 | } | |
5533 | ||
dce840a0 | 5534 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5535 | { |
dce840a0 | 5536 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5537 | |
68e74568 | 5538 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5539 | free_cpumask_var(rd->rto_mask); |
5540 | free_cpumask_var(rd->online); | |
5541 | free_cpumask_var(rd->span); | |
5542 | kfree(rd); | |
5543 | } | |
5544 | ||
57d885fe GH |
5545 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5546 | { | |
a0490fa3 | 5547 | struct root_domain *old_rd = NULL; |
57d885fe | 5548 | unsigned long flags; |
57d885fe | 5549 | |
05fa785c | 5550 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5551 | |
5552 | if (rq->rd) { | |
a0490fa3 | 5553 | old_rd = rq->rd; |
57d885fe | 5554 | |
c6c4927b | 5555 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5556 | set_rq_offline(rq); |
57d885fe | 5557 | |
c6c4927b | 5558 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5559 | |
a0490fa3 IM |
5560 | /* |
5561 | * If we dont want to free the old_rt yet then | |
5562 | * set old_rd to NULL to skip the freeing later | |
5563 | * in this function: | |
5564 | */ | |
5565 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5566 | old_rd = NULL; | |
57d885fe GH |
5567 | } |
5568 | ||
5569 | atomic_inc(&rd->refcount); | |
5570 | rq->rd = rd; | |
5571 | ||
c6c4927b | 5572 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5573 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5574 | set_rq_online(rq); |
57d885fe | 5575 | |
05fa785c | 5576 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5577 | |
5578 | if (old_rd) | |
dce840a0 | 5579 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5580 | } |
5581 | ||
68c38fc3 | 5582 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5583 | { |
5584 | memset(rd, 0, sizeof(*rd)); | |
5585 | ||
68c38fc3 | 5586 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5587 | goto out; |
68c38fc3 | 5588 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5589 | goto free_span; |
68c38fc3 | 5590 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5591 | goto free_online; |
6e0534f2 | 5592 | |
68c38fc3 | 5593 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5594 | goto free_rto_mask; |
c6c4927b | 5595 | return 0; |
6e0534f2 | 5596 | |
68e74568 RR |
5597 | free_rto_mask: |
5598 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5599 | free_online: |
5600 | free_cpumask_var(rd->online); | |
5601 | free_span: | |
5602 | free_cpumask_var(rd->span); | |
0c910d28 | 5603 | out: |
c6c4927b | 5604 | return -ENOMEM; |
57d885fe GH |
5605 | } |
5606 | ||
029632fb PZ |
5607 | /* |
5608 | * By default the system creates a single root-domain with all cpus as | |
5609 | * members (mimicking the global state we have today). | |
5610 | */ | |
5611 | struct root_domain def_root_domain; | |
5612 | ||
57d885fe GH |
5613 | static void init_defrootdomain(void) |
5614 | { | |
68c38fc3 | 5615 | init_rootdomain(&def_root_domain); |
c6c4927b | 5616 | |
57d885fe GH |
5617 | atomic_set(&def_root_domain.refcount, 1); |
5618 | } | |
5619 | ||
dc938520 | 5620 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5621 | { |
5622 | struct root_domain *rd; | |
5623 | ||
5624 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5625 | if (!rd) | |
5626 | return NULL; | |
5627 | ||
68c38fc3 | 5628 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5629 | kfree(rd); |
5630 | return NULL; | |
5631 | } | |
57d885fe GH |
5632 | |
5633 | return rd; | |
5634 | } | |
5635 | ||
e3589f6c PZ |
5636 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5637 | { | |
5638 | struct sched_group *tmp, *first; | |
5639 | ||
5640 | if (!sg) | |
5641 | return; | |
5642 | ||
5643 | first = sg; | |
5644 | do { | |
5645 | tmp = sg->next; | |
5646 | ||
5647 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5648 | kfree(sg->sgp); | |
5649 | ||
5650 | kfree(sg); | |
5651 | sg = tmp; | |
5652 | } while (sg != first); | |
5653 | } | |
5654 | ||
dce840a0 PZ |
5655 | static void free_sched_domain(struct rcu_head *rcu) |
5656 | { | |
5657 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5658 | |
5659 | /* | |
5660 | * If its an overlapping domain it has private groups, iterate and | |
5661 | * nuke them all. | |
5662 | */ | |
5663 | if (sd->flags & SD_OVERLAP) { | |
5664 | free_sched_groups(sd->groups, 1); | |
5665 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5666 | kfree(sd->groups->sgp); |
dce840a0 | 5667 | kfree(sd->groups); |
9c3f75cb | 5668 | } |
dce840a0 PZ |
5669 | kfree(sd); |
5670 | } | |
5671 | ||
5672 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5673 | { | |
5674 | call_rcu(&sd->rcu, free_sched_domain); | |
5675 | } | |
5676 | ||
5677 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5678 | { | |
5679 | for (; sd; sd = sd->parent) | |
5680 | destroy_sched_domain(sd, cpu); | |
5681 | } | |
5682 | ||
1da177e4 | 5683 | /* |
0eab9146 | 5684 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5685 | * hold the hotplug lock. |
5686 | */ | |
0eab9146 IM |
5687 | static void |
5688 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5689 | { |
70b97a7f | 5690 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5691 | struct sched_domain *tmp; |
5692 | ||
5693 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5694 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5695 | struct sched_domain *parent = tmp->parent; |
5696 | if (!parent) | |
5697 | break; | |
f29c9b1c | 5698 | |
1a848870 | 5699 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5700 | tmp->parent = parent->parent; |
1a848870 SS |
5701 | if (parent->parent) |
5702 | parent->parent->child = tmp; | |
dce840a0 | 5703 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5704 | } else |
5705 | tmp = tmp->parent; | |
245af2c7 SS |
5706 | } |
5707 | ||
1a848870 | 5708 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5709 | tmp = sd; |
245af2c7 | 5710 | sd = sd->parent; |
dce840a0 | 5711 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5712 | if (sd) |
5713 | sd->child = NULL; | |
5714 | } | |
1da177e4 | 5715 | |
4cb98839 | 5716 | sched_domain_debug(sd, cpu); |
1da177e4 | 5717 | |
57d885fe | 5718 | rq_attach_root(rq, rd); |
dce840a0 | 5719 | tmp = rq->sd; |
674311d5 | 5720 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5721 | destroy_sched_domains(tmp, cpu); |
1da177e4 LT |
5722 | } |
5723 | ||
5724 | /* cpus with isolated domains */ | |
dcc30a35 | 5725 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5726 | |
5727 | /* Setup the mask of cpus configured for isolated domains */ | |
5728 | static int __init isolated_cpu_setup(char *str) | |
5729 | { | |
bdddd296 | 5730 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5731 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5732 | return 1; |
5733 | } | |
5734 | ||
8927f494 | 5735 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5736 | |
9c1cfda2 | 5737 | #ifdef CONFIG_NUMA |
198e2f18 | 5738 | |
9c1cfda2 JH |
5739 | /** |
5740 | * find_next_best_node - find the next node to include in a sched_domain | |
5741 | * @node: node whose sched_domain we're building | |
5742 | * @used_nodes: nodes already in the sched_domain | |
5743 | * | |
41a2d6cf | 5744 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5745 | * finds the closest node not already in the @used_nodes map. |
5746 | * | |
5747 | * Should use nodemask_t. | |
5748 | */ | |
c5f59f08 | 5749 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 5750 | { |
7142d17e | 5751 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
5752 | |
5753 | min_val = INT_MAX; | |
5754 | ||
076ac2af | 5755 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 5756 | /* Start at @node */ |
076ac2af | 5757 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
5758 | |
5759 | if (!nr_cpus_node(n)) | |
5760 | continue; | |
5761 | ||
5762 | /* Skip already used nodes */ | |
c5f59f08 | 5763 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
5764 | continue; |
5765 | ||
5766 | /* Simple min distance search */ | |
5767 | val = node_distance(node, n); | |
5768 | ||
5769 | if (val < min_val) { | |
5770 | min_val = val; | |
5771 | best_node = n; | |
5772 | } | |
5773 | } | |
5774 | ||
7142d17e HD |
5775 | if (best_node != -1) |
5776 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
5777 | return best_node; |
5778 | } | |
5779 | ||
5780 | /** | |
5781 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5782 | * @node: node whose cpumask we're constructing | |
73486722 | 5783 | * @span: resulting cpumask |
9c1cfda2 | 5784 | * |
41a2d6cf | 5785 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
5786 | * should be one that prevents unnecessary balancing, but also spreads tasks |
5787 | * out optimally. | |
5788 | */ | |
96f874e2 | 5789 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 5790 | { |
c5f59f08 | 5791 | nodemask_t used_nodes; |
48f24c4d | 5792 | int i; |
9c1cfda2 | 5793 | |
6ca09dfc | 5794 | cpumask_clear(span); |
c5f59f08 | 5795 | nodes_clear(used_nodes); |
9c1cfda2 | 5796 | |
6ca09dfc | 5797 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 5798 | node_set(node, used_nodes); |
9c1cfda2 JH |
5799 | |
5800 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 5801 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
5802 | if (next_node < 0) |
5803 | break; | |
6ca09dfc | 5804 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 5805 | } |
9c1cfda2 | 5806 | } |
d3081f52 PZ |
5807 | |
5808 | static const struct cpumask *cpu_node_mask(int cpu) | |
5809 | { | |
5810 | lockdep_assert_held(&sched_domains_mutex); | |
5811 | ||
5812 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
5813 | ||
5814 | return sched_domains_tmpmask; | |
5815 | } | |
2c402dc3 PZ |
5816 | |
5817 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
5818 | { | |
5819 | return cpu_possible_mask; | |
5820 | } | |
6d6bc0ad | 5821 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 5822 | |
d3081f52 PZ |
5823 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5824 | { | |
5825 | return cpumask_of_node(cpu_to_node(cpu)); | |
5826 | } | |
5827 | ||
5c45bf27 | 5828 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5829 | |
dce840a0 PZ |
5830 | struct sd_data { |
5831 | struct sched_domain **__percpu sd; | |
5832 | struct sched_group **__percpu sg; | |
9c3f75cb | 5833 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5834 | }; |
5835 | ||
49a02c51 | 5836 | struct s_data { |
21d42ccf | 5837 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5838 | struct root_domain *rd; |
5839 | }; | |
5840 | ||
2109b99e | 5841 | enum s_alloc { |
2109b99e | 5842 | sa_rootdomain, |
21d42ccf | 5843 | sa_sd, |
dce840a0 | 5844 | sa_sd_storage, |
2109b99e AH |
5845 | sa_none, |
5846 | }; | |
5847 | ||
54ab4ff4 PZ |
5848 | struct sched_domain_topology_level; |
5849 | ||
5850 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5851 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5852 | ||
e3589f6c PZ |
5853 | #define SDTL_OVERLAP 0x01 |
5854 | ||
eb7a74e6 | 5855 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5856 | sched_domain_init_f init; |
5857 | sched_domain_mask_f mask; | |
e3589f6c | 5858 | int flags; |
54ab4ff4 | 5859 | struct sd_data data; |
eb7a74e6 PZ |
5860 | }; |
5861 | ||
e3589f6c PZ |
5862 | static int |
5863 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5864 | { | |
5865 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5866 | const struct cpumask *span = sched_domain_span(sd); | |
5867 | struct cpumask *covered = sched_domains_tmpmask; | |
5868 | struct sd_data *sdd = sd->private; | |
5869 | struct sched_domain *child; | |
5870 | int i; | |
5871 | ||
5872 | cpumask_clear(covered); | |
5873 | ||
5874 | for_each_cpu(i, span) { | |
5875 | struct cpumask *sg_span; | |
5876 | ||
5877 | if (cpumask_test_cpu(i, covered)) | |
5878 | continue; | |
5879 | ||
5880 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
5881 | GFP_KERNEL, cpu_to_node(i)); | |
5882 | ||
5883 | if (!sg) | |
5884 | goto fail; | |
5885 | ||
5886 | sg_span = sched_group_cpus(sg); | |
5887 | ||
5888 | child = *per_cpu_ptr(sdd->sd, i); | |
5889 | if (child->child) { | |
5890 | child = child->child; | |
5891 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5892 | } else | |
5893 | cpumask_set_cpu(i, sg_span); | |
5894 | ||
5895 | cpumask_or(covered, covered, sg_span); | |
5896 | ||
5897 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
5898 | atomic_inc(&sg->sgp->ref); | |
5899 | ||
5900 | if (cpumask_test_cpu(cpu, sg_span)) | |
5901 | groups = sg; | |
5902 | ||
5903 | if (!first) | |
5904 | first = sg; | |
5905 | if (last) | |
5906 | last->next = sg; | |
5907 | last = sg; | |
5908 | last->next = first; | |
5909 | } | |
5910 | sd->groups = groups; | |
5911 | ||
5912 | return 0; | |
5913 | ||
5914 | fail: | |
5915 | free_sched_groups(first, 0); | |
5916 | ||
5917 | return -ENOMEM; | |
5918 | } | |
5919 | ||
dce840a0 | 5920 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5921 | { |
dce840a0 PZ |
5922 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5923 | struct sched_domain *child = sd->child; | |
1da177e4 | 5924 | |
dce840a0 PZ |
5925 | if (child) |
5926 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5927 | |
9c3f75cb | 5928 | if (sg) { |
dce840a0 | 5929 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5930 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5931 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5932 | } |
dce840a0 PZ |
5933 | |
5934 | return cpu; | |
1e9f28fa | 5935 | } |
1e9f28fa | 5936 | |
01a08546 | 5937 | /* |
dce840a0 PZ |
5938 | * build_sched_groups will build a circular linked list of the groups |
5939 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5940 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5941 | * |
5942 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5943 | */ |
e3589f6c PZ |
5944 | static int |
5945 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5946 | { |
dce840a0 PZ |
5947 | struct sched_group *first = NULL, *last = NULL; |
5948 | struct sd_data *sdd = sd->private; | |
5949 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5950 | struct cpumask *covered; |
dce840a0 | 5951 | int i; |
9c1cfda2 | 5952 | |
e3589f6c PZ |
5953 | get_group(cpu, sdd, &sd->groups); |
5954 | atomic_inc(&sd->groups->ref); | |
5955 | ||
5956 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
5957 | return 0; | |
5958 | ||
f96225fd PZ |
5959 | lockdep_assert_held(&sched_domains_mutex); |
5960 | covered = sched_domains_tmpmask; | |
5961 | ||
dce840a0 | 5962 | cpumask_clear(covered); |
6711cab4 | 5963 | |
dce840a0 PZ |
5964 | for_each_cpu(i, span) { |
5965 | struct sched_group *sg; | |
5966 | int group = get_group(i, sdd, &sg); | |
5967 | int j; | |
6711cab4 | 5968 | |
dce840a0 PZ |
5969 | if (cpumask_test_cpu(i, covered)) |
5970 | continue; | |
6711cab4 | 5971 | |
dce840a0 | 5972 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5973 | sg->sgp->power = 0; |
0601a88d | 5974 | |
dce840a0 PZ |
5975 | for_each_cpu(j, span) { |
5976 | if (get_group(j, sdd, NULL) != group) | |
5977 | continue; | |
0601a88d | 5978 | |
dce840a0 PZ |
5979 | cpumask_set_cpu(j, covered); |
5980 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5981 | } | |
0601a88d | 5982 | |
dce840a0 PZ |
5983 | if (!first) |
5984 | first = sg; | |
5985 | if (last) | |
5986 | last->next = sg; | |
5987 | last = sg; | |
5988 | } | |
5989 | last->next = first; | |
e3589f6c PZ |
5990 | |
5991 | return 0; | |
0601a88d | 5992 | } |
51888ca2 | 5993 | |
89c4710e SS |
5994 | /* |
5995 | * Initialize sched groups cpu_power. | |
5996 | * | |
5997 | * cpu_power indicates the capacity of sched group, which is used while | |
5998 | * distributing the load between different sched groups in a sched domain. | |
5999 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6000 | * there are asymmetries in the topology. If there are asymmetries, group | |
6001 | * having more cpu_power will pickup more load compared to the group having | |
6002 | * less cpu_power. | |
89c4710e SS |
6003 | */ |
6004 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6005 | { | |
e3589f6c | 6006 | struct sched_group *sg = sd->groups; |
89c4710e | 6007 | |
e3589f6c PZ |
6008 | WARN_ON(!sd || !sg); |
6009 | ||
6010 | do { | |
6011 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6012 | sg = sg->next; | |
6013 | } while (sg != sd->groups); | |
89c4710e | 6014 | |
e3589f6c PZ |
6015 | if (cpu != group_first_cpu(sg)) |
6016 | return; | |
aae6d3dd | 6017 | |
d274cb30 | 6018 | update_group_power(sd, cpu); |
89c4710e SS |
6019 | } |
6020 | ||
029632fb PZ |
6021 | int __weak arch_sd_sibling_asym_packing(void) |
6022 | { | |
6023 | return 0*SD_ASYM_PACKING; | |
6024 | } | |
6025 | ||
7c16ec58 MT |
6026 | /* |
6027 | * Initializers for schedule domains | |
6028 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6029 | */ | |
6030 | ||
a5d8c348 IM |
6031 | #ifdef CONFIG_SCHED_DEBUG |
6032 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6033 | #else | |
6034 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6035 | #endif | |
6036 | ||
54ab4ff4 PZ |
6037 | #define SD_INIT_FUNC(type) \ |
6038 | static noinline struct sched_domain * \ | |
6039 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
6040 | { \ | |
6041 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
6042 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
6043 | SD_INIT_NAME(sd, type); \ |
6044 | sd->private = &tl->data; \ | |
6045 | return sd; \ | |
7c16ec58 MT |
6046 | } |
6047 | ||
6048 | SD_INIT_FUNC(CPU) | |
6049 | #ifdef CONFIG_NUMA | |
6050 | SD_INIT_FUNC(ALLNODES) | |
6051 | SD_INIT_FUNC(NODE) | |
6052 | #endif | |
6053 | #ifdef CONFIG_SCHED_SMT | |
6054 | SD_INIT_FUNC(SIBLING) | |
6055 | #endif | |
6056 | #ifdef CONFIG_SCHED_MC | |
6057 | SD_INIT_FUNC(MC) | |
6058 | #endif | |
01a08546 HC |
6059 | #ifdef CONFIG_SCHED_BOOK |
6060 | SD_INIT_FUNC(BOOK) | |
6061 | #endif | |
7c16ec58 | 6062 | |
1d3504fc | 6063 | static int default_relax_domain_level = -1; |
60495e77 | 6064 | int sched_domain_level_max; |
1d3504fc HS |
6065 | |
6066 | static int __init setup_relax_domain_level(char *str) | |
6067 | { | |
30e0e178 LZ |
6068 | unsigned long val; |
6069 | ||
6070 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 6071 | if (val < sched_domain_level_max) |
30e0e178 LZ |
6072 | default_relax_domain_level = val; |
6073 | ||
1d3504fc HS |
6074 | return 1; |
6075 | } | |
6076 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6077 | ||
6078 | static void set_domain_attribute(struct sched_domain *sd, | |
6079 | struct sched_domain_attr *attr) | |
6080 | { | |
6081 | int request; | |
6082 | ||
6083 | if (!attr || attr->relax_domain_level < 0) { | |
6084 | if (default_relax_domain_level < 0) | |
6085 | return; | |
6086 | else | |
6087 | request = default_relax_domain_level; | |
6088 | } else | |
6089 | request = attr->relax_domain_level; | |
6090 | if (request < sd->level) { | |
6091 | /* turn off idle balance on this domain */ | |
c88d5910 | 6092 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6093 | } else { |
6094 | /* turn on idle balance on this domain */ | |
c88d5910 | 6095 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6096 | } |
6097 | } | |
6098 | ||
54ab4ff4 PZ |
6099 | static void __sdt_free(const struct cpumask *cpu_map); |
6100 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6101 | ||
2109b99e AH |
6102 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6103 | const struct cpumask *cpu_map) | |
6104 | { | |
6105 | switch (what) { | |
2109b99e | 6106 | case sa_rootdomain: |
822ff793 PZ |
6107 | if (!atomic_read(&d->rd->refcount)) |
6108 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6109 | case sa_sd: |
6110 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6111 | case sa_sd_storage: |
54ab4ff4 | 6112 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6113 | case sa_none: |
6114 | break; | |
6115 | } | |
6116 | } | |
3404c8d9 | 6117 | |
2109b99e AH |
6118 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6119 | const struct cpumask *cpu_map) | |
6120 | { | |
dce840a0 PZ |
6121 | memset(d, 0, sizeof(*d)); |
6122 | ||
54ab4ff4 PZ |
6123 | if (__sdt_alloc(cpu_map)) |
6124 | return sa_sd_storage; | |
dce840a0 PZ |
6125 | d->sd = alloc_percpu(struct sched_domain *); |
6126 | if (!d->sd) | |
6127 | return sa_sd_storage; | |
2109b99e | 6128 | d->rd = alloc_rootdomain(); |
dce840a0 | 6129 | if (!d->rd) |
21d42ccf | 6130 | return sa_sd; |
2109b99e AH |
6131 | return sa_rootdomain; |
6132 | } | |
57d885fe | 6133 | |
dce840a0 PZ |
6134 | /* |
6135 | * NULL the sd_data elements we've used to build the sched_domain and | |
6136 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6137 | * will not free the data we're using. | |
6138 | */ | |
6139 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6140 | { | |
6141 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6142 | |
6143 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6144 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6145 | ||
e3589f6c | 6146 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6147 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6148 | |
6149 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6150 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6151 | } |
6152 | ||
2c402dc3 PZ |
6153 | #ifdef CONFIG_SCHED_SMT |
6154 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6155 | { |
2c402dc3 | 6156 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6157 | } |
2c402dc3 | 6158 | #endif |
7f4588f3 | 6159 | |
d069b916 PZ |
6160 | /* |
6161 | * Topology list, bottom-up. | |
6162 | */ | |
2c402dc3 | 6163 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6164 | #ifdef CONFIG_SCHED_SMT |
6165 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6166 | #endif |
1e9f28fa | 6167 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6168 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6169 | #endif |
d069b916 PZ |
6170 | #ifdef CONFIG_SCHED_BOOK |
6171 | { sd_init_BOOK, cpu_book_mask, }, | |
6172 | #endif | |
6173 | { sd_init_CPU, cpu_cpu_mask, }, | |
6174 | #ifdef CONFIG_NUMA | |
e3589f6c | 6175 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, |
d069b916 | 6176 | { sd_init_ALLNODES, cpu_allnodes_mask, }, |
1da177e4 | 6177 | #endif |
eb7a74e6 PZ |
6178 | { NULL, }, |
6179 | }; | |
6180 | ||
6181 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6182 | ||
54ab4ff4 PZ |
6183 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6184 | { | |
6185 | struct sched_domain_topology_level *tl; | |
6186 | int j; | |
6187 | ||
6188 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6189 | struct sd_data *sdd = &tl->data; | |
6190 | ||
6191 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6192 | if (!sdd->sd) | |
6193 | return -ENOMEM; | |
6194 | ||
6195 | sdd->sg = alloc_percpu(struct sched_group *); | |
6196 | if (!sdd->sg) | |
6197 | return -ENOMEM; | |
6198 | ||
9c3f75cb PZ |
6199 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6200 | if (!sdd->sgp) | |
6201 | return -ENOMEM; | |
6202 | ||
54ab4ff4 PZ |
6203 | for_each_cpu(j, cpu_map) { |
6204 | struct sched_domain *sd; | |
6205 | struct sched_group *sg; | |
9c3f75cb | 6206 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6207 | |
6208 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6209 | GFP_KERNEL, cpu_to_node(j)); | |
6210 | if (!sd) | |
6211 | return -ENOMEM; | |
6212 | ||
6213 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6214 | ||
6215 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6216 | GFP_KERNEL, cpu_to_node(j)); | |
6217 | if (!sg) | |
6218 | return -ENOMEM; | |
6219 | ||
6220 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
6221 | |
6222 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
6223 | GFP_KERNEL, cpu_to_node(j)); | |
6224 | if (!sgp) | |
6225 | return -ENOMEM; | |
6226 | ||
6227 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6228 | } |
6229 | } | |
6230 | ||
6231 | return 0; | |
6232 | } | |
6233 | ||
6234 | static void __sdt_free(const struct cpumask *cpu_map) | |
6235 | { | |
6236 | struct sched_domain_topology_level *tl; | |
6237 | int j; | |
6238 | ||
6239 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6240 | struct sd_data *sdd = &tl->data; | |
6241 | ||
6242 | for_each_cpu(j, cpu_map) { | |
e3589f6c PZ |
6243 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); |
6244 | if (sd && (sd->flags & SD_OVERLAP)) | |
6245 | free_sched_groups(sd->groups, 0); | |
feff8fa0 | 6246 | kfree(*per_cpu_ptr(sdd->sd, j)); |
54ab4ff4 | 6247 | kfree(*per_cpu_ptr(sdd->sg, j)); |
9c3f75cb | 6248 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
54ab4ff4 PZ |
6249 | } |
6250 | free_percpu(sdd->sd); | |
6251 | free_percpu(sdd->sg); | |
9c3f75cb | 6252 | free_percpu(sdd->sgp); |
54ab4ff4 PZ |
6253 | } |
6254 | } | |
6255 | ||
2c402dc3 PZ |
6256 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6257 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6258 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6259 | int cpu) |
6260 | { | |
54ab4ff4 | 6261 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6262 | if (!sd) |
d069b916 | 6263 | return child; |
2c402dc3 PZ |
6264 | |
6265 | set_domain_attribute(sd, attr); | |
6266 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
6267 | if (child) { |
6268 | sd->level = child->level + 1; | |
6269 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6270 | child->parent = sd; |
60495e77 | 6271 | } |
d069b916 | 6272 | sd->child = child; |
2c402dc3 PZ |
6273 | |
6274 | return sd; | |
6275 | } | |
6276 | ||
2109b99e AH |
6277 | /* |
6278 | * Build sched domains for a given set of cpus and attach the sched domains | |
6279 | * to the individual cpus | |
6280 | */ | |
dce840a0 PZ |
6281 | static int build_sched_domains(const struct cpumask *cpu_map, |
6282 | struct sched_domain_attr *attr) | |
2109b99e AH |
6283 | { |
6284 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6285 | struct sched_domain *sd; |
2109b99e | 6286 | struct s_data d; |
822ff793 | 6287 | int i, ret = -ENOMEM; |
9c1cfda2 | 6288 | |
2109b99e AH |
6289 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6290 | if (alloc_state != sa_rootdomain) | |
6291 | goto error; | |
9c1cfda2 | 6292 | |
dce840a0 | 6293 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6294 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6295 | struct sched_domain_topology_level *tl; |
6296 | ||
3bd65a80 | 6297 | sd = NULL; |
e3589f6c | 6298 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6299 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6300 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6301 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6302 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6303 | break; | |
e3589f6c | 6304 | } |
d274cb30 | 6305 | |
d069b916 PZ |
6306 | while (sd->child) |
6307 | sd = sd->child; | |
6308 | ||
21d42ccf | 6309 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6310 | } |
6311 | ||
6312 | /* Build the groups for the domains */ | |
6313 | for_each_cpu(i, cpu_map) { | |
6314 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6315 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6316 | if (sd->flags & SD_OVERLAP) { |
6317 | if (build_overlap_sched_groups(sd, i)) | |
6318 | goto error; | |
6319 | } else { | |
6320 | if (build_sched_groups(sd, i)) | |
6321 | goto error; | |
6322 | } | |
1cf51902 | 6323 | } |
a06dadbe | 6324 | } |
9c1cfda2 | 6325 | |
1da177e4 | 6326 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6327 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6328 | if (!cpumask_test_cpu(i, cpu_map)) | |
6329 | continue; | |
9c1cfda2 | 6330 | |
dce840a0 PZ |
6331 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6332 | claim_allocations(i, sd); | |
cd4ea6ae | 6333 | init_sched_groups_power(i, sd); |
dce840a0 | 6334 | } |
f712c0c7 | 6335 | } |
9c1cfda2 | 6336 | |
1da177e4 | 6337 | /* Attach the domains */ |
dce840a0 | 6338 | rcu_read_lock(); |
abcd083a | 6339 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6340 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6341 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6342 | } |
dce840a0 | 6343 | rcu_read_unlock(); |
51888ca2 | 6344 | |
822ff793 | 6345 | ret = 0; |
51888ca2 | 6346 | error: |
2109b99e | 6347 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6348 | return ret; |
1da177e4 | 6349 | } |
029190c5 | 6350 | |
acc3f5d7 | 6351 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6352 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6353 | static struct sched_domain_attr *dattr_cur; |
6354 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6355 | |
6356 | /* | |
6357 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6358 | * cpumask) fails, then fallback to a single sched domain, |
6359 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6360 | */ |
4212823f | 6361 | static cpumask_var_t fallback_doms; |
029190c5 | 6362 | |
ee79d1bd HC |
6363 | /* |
6364 | * arch_update_cpu_topology lets virtualized architectures update the | |
6365 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6366 | * or 0 if it stayed the same. | |
6367 | */ | |
6368 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6369 | { |
ee79d1bd | 6370 | return 0; |
22e52b07 HC |
6371 | } |
6372 | ||
acc3f5d7 RR |
6373 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6374 | { | |
6375 | int i; | |
6376 | cpumask_var_t *doms; | |
6377 | ||
6378 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6379 | if (!doms) | |
6380 | return NULL; | |
6381 | for (i = 0; i < ndoms; i++) { | |
6382 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6383 | free_sched_domains(doms, i); | |
6384 | return NULL; | |
6385 | } | |
6386 | } | |
6387 | return doms; | |
6388 | } | |
6389 | ||
6390 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6391 | { | |
6392 | unsigned int i; | |
6393 | for (i = 0; i < ndoms; i++) | |
6394 | free_cpumask_var(doms[i]); | |
6395 | kfree(doms); | |
6396 | } | |
6397 | ||
1a20ff27 | 6398 | /* |
41a2d6cf | 6399 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6400 | * For now this just excludes isolated cpus, but could be used to |
6401 | * exclude other special cases in the future. | |
1a20ff27 | 6402 | */ |
c4a8849a | 6403 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6404 | { |
7378547f MM |
6405 | int err; |
6406 | ||
22e52b07 | 6407 | arch_update_cpu_topology(); |
029190c5 | 6408 | ndoms_cur = 1; |
acc3f5d7 | 6409 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6410 | if (!doms_cur) |
acc3f5d7 RR |
6411 | doms_cur = &fallback_doms; |
6412 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 6413 | dattr_cur = NULL; |
dce840a0 | 6414 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6415 | register_sched_domain_sysctl(); |
7378547f MM |
6416 | |
6417 | return err; | |
1a20ff27 DG |
6418 | } |
6419 | ||
1a20ff27 DG |
6420 | /* |
6421 | * Detach sched domains from a group of cpus specified in cpu_map | |
6422 | * These cpus will now be attached to the NULL domain | |
6423 | */ | |
96f874e2 | 6424 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6425 | { |
6426 | int i; | |
6427 | ||
dce840a0 | 6428 | rcu_read_lock(); |
abcd083a | 6429 | for_each_cpu(i, cpu_map) |
57d885fe | 6430 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6431 | rcu_read_unlock(); |
1a20ff27 DG |
6432 | } |
6433 | ||
1d3504fc HS |
6434 | /* handle null as "default" */ |
6435 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6436 | struct sched_domain_attr *new, int idx_new) | |
6437 | { | |
6438 | struct sched_domain_attr tmp; | |
6439 | ||
6440 | /* fast path */ | |
6441 | if (!new && !cur) | |
6442 | return 1; | |
6443 | ||
6444 | tmp = SD_ATTR_INIT; | |
6445 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6446 | new ? (new + idx_new) : &tmp, | |
6447 | sizeof(struct sched_domain_attr)); | |
6448 | } | |
6449 | ||
029190c5 PJ |
6450 | /* |
6451 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6452 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6453 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6454 | * It destroys each deleted domain and builds each new domain. | |
6455 | * | |
acc3f5d7 | 6456 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6457 | * The masks don't intersect (don't overlap.) We should setup one |
6458 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6459 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6460 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6461 | * it as it is. | |
6462 | * | |
acc3f5d7 RR |
6463 | * The passed in 'doms_new' should be allocated using |
6464 | * alloc_sched_domains. This routine takes ownership of it and will | |
6465 | * free_sched_domains it when done with it. If the caller failed the | |
6466 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6467 | * and partition_sched_domains() will fallback to the single partition | |
6468 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6469 | * |
96f874e2 | 6470 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6471 | * ndoms_new == 0 is a special case for destroying existing domains, |
6472 | * and it will not create the default domain. | |
dfb512ec | 6473 | * |
029190c5 PJ |
6474 | * Call with hotplug lock held |
6475 | */ | |
acc3f5d7 | 6476 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6477 | struct sched_domain_attr *dattr_new) |
029190c5 | 6478 | { |
dfb512ec | 6479 | int i, j, n; |
d65bd5ec | 6480 | int new_topology; |
029190c5 | 6481 | |
712555ee | 6482 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6483 | |
7378547f MM |
6484 | /* always unregister in case we don't destroy any domains */ |
6485 | unregister_sched_domain_sysctl(); | |
6486 | ||
d65bd5ec HC |
6487 | /* Let architecture update cpu core mappings. */ |
6488 | new_topology = arch_update_cpu_topology(); | |
6489 | ||
dfb512ec | 6490 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6491 | |
6492 | /* Destroy deleted domains */ | |
6493 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6494 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6495 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6496 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6497 | goto match1; |
6498 | } | |
6499 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6500 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6501 | match1: |
6502 | ; | |
6503 | } | |
6504 | ||
e761b772 MK |
6505 | if (doms_new == NULL) { |
6506 | ndoms_cur = 0; | |
acc3f5d7 | 6507 | doms_new = &fallback_doms; |
6ad4c188 | 6508 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6509 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6510 | } |
6511 | ||
029190c5 PJ |
6512 | /* Build new domains */ |
6513 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6514 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6515 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6516 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6517 | goto match2; |
6518 | } | |
6519 | /* no match - add a new doms_new */ | |
dce840a0 | 6520 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6521 | match2: |
6522 | ; | |
6523 | } | |
6524 | ||
6525 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6526 | if (doms_cur != &fallback_doms) |
6527 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6528 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6529 | doms_cur = doms_new; |
1d3504fc | 6530 | dattr_cur = dattr_new; |
029190c5 | 6531 | ndoms_cur = ndoms_new; |
7378547f MM |
6532 | |
6533 | register_sched_domain_sysctl(); | |
a1835615 | 6534 | |
712555ee | 6535 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6536 | } |
6537 | ||
5c45bf27 | 6538 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 6539 | static void reinit_sched_domains(void) |
5c45bf27 | 6540 | { |
95402b38 | 6541 | get_online_cpus(); |
dfb512ec MK |
6542 | |
6543 | /* Destroy domains first to force the rebuild */ | |
6544 | partition_sched_domains(0, NULL, NULL); | |
6545 | ||
e761b772 | 6546 | rebuild_sched_domains(); |
95402b38 | 6547 | put_online_cpus(); |
5c45bf27 SS |
6548 | } |
6549 | ||
6550 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6551 | { | |
afb8a9b7 | 6552 | unsigned int level = 0; |
5c45bf27 | 6553 | |
afb8a9b7 GS |
6554 | if (sscanf(buf, "%u", &level) != 1) |
6555 | return -EINVAL; | |
6556 | ||
6557 | /* | |
6558 | * level is always be positive so don't check for | |
6559 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
6560 | * What happens on 0 or 1 byte write, | |
6561 | * need to check for count as well? | |
6562 | */ | |
6563 | ||
6564 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
6565 | return -EINVAL; |
6566 | ||
6567 | if (smt) | |
afb8a9b7 | 6568 | sched_smt_power_savings = level; |
5c45bf27 | 6569 | else |
afb8a9b7 | 6570 | sched_mc_power_savings = level; |
5c45bf27 | 6571 | |
c4a8849a | 6572 | reinit_sched_domains(); |
5c45bf27 | 6573 | |
c70f22d2 | 6574 | return count; |
5c45bf27 SS |
6575 | } |
6576 | ||
5c45bf27 | 6577 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 6578 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 6579 | struct sysdev_class_attribute *attr, |
f718cd4a | 6580 | char *page) |
5c45bf27 SS |
6581 | { |
6582 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6583 | } | |
f718cd4a | 6584 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 6585 | struct sysdev_class_attribute *attr, |
48f24c4d | 6586 | const char *buf, size_t count) |
5c45bf27 SS |
6587 | { |
6588 | return sched_power_savings_store(buf, count, 0); | |
6589 | } | |
f718cd4a AK |
6590 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
6591 | sched_mc_power_savings_show, | |
6592 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6593 | #endif |
6594 | ||
6595 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 6596 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 6597 | struct sysdev_class_attribute *attr, |
f718cd4a | 6598 | char *page) |
5c45bf27 SS |
6599 | { |
6600 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6601 | } | |
f718cd4a | 6602 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 6603 | struct sysdev_class_attribute *attr, |
48f24c4d | 6604 | const char *buf, size_t count) |
5c45bf27 SS |
6605 | { |
6606 | return sched_power_savings_store(buf, count, 1); | |
6607 | } | |
f718cd4a AK |
6608 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
6609 | sched_smt_power_savings_show, | |
6707de00 AB |
6610 | sched_smt_power_savings_store); |
6611 | #endif | |
6612 | ||
39aac648 | 6613 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
6614 | { |
6615 | int err = 0; | |
6616 | ||
6617 | #ifdef CONFIG_SCHED_SMT | |
6618 | if (smt_capable()) | |
6619 | err = sysfs_create_file(&cls->kset.kobj, | |
6620 | &attr_sched_smt_power_savings.attr); | |
6621 | #endif | |
6622 | #ifdef CONFIG_SCHED_MC | |
6623 | if (!err && mc_capable()) | |
6624 | err = sysfs_create_file(&cls->kset.kobj, | |
6625 | &attr_sched_mc_power_savings.attr); | |
6626 | #endif | |
6627 | return err; | |
6628 | } | |
6d6bc0ad | 6629 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 6630 | |
1da177e4 | 6631 | /* |
3a101d05 TH |
6632 | * Update cpusets according to cpu_active mask. If cpusets are |
6633 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6634 | * around partition_sched_domains(). | |
1da177e4 | 6635 | */ |
0b2e918a TH |
6636 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6637 | void *hcpu) | |
e761b772 | 6638 | { |
3a101d05 | 6639 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 6640 | case CPU_ONLINE: |
6ad4c188 | 6641 | case CPU_DOWN_FAILED: |
3a101d05 | 6642 | cpuset_update_active_cpus(); |
e761b772 | 6643 | return NOTIFY_OK; |
3a101d05 TH |
6644 | default: |
6645 | return NOTIFY_DONE; | |
6646 | } | |
6647 | } | |
e761b772 | 6648 | |
0b2e918a TH |
6649 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6650 | void *hcpu) | |
3a101d05 TH |
6651 | { |
6652 | switch (action & ~CPU_TASKS_FROZEN) { | |
6653 | case CPU_DOWN_PREPARE: | |
6654 | cpuset_update_active_cpus(); | |
6655 | return NOTIFY_OK; | |
e761b772 MK |
6656 | default: |
6657 | return NOTIFY_DONE; | |
6658 | } | |
6659 | } | |
e761b772 | 6660 | |
1da177e4 LT |
6661 | void __init sched_init_smp(void) |
6662 | { | |
dcc30a35 RR |
6663 | cpumask_var_t non_isolated_cpus; |
6664 | ||
6665 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6666 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6667 | |
95402b38 | 6668 | get_online_cpus(); |
712555ee | 6669 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6670 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6671 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6672 | if (cpumask_empty(non_isolated_cpus)) | |
6673 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6674 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6675 | put_online_cpus(); |
e761b772 | 6676 | |
3a101d05 TH |
6677 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6678 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6679 | |
6680 | /* RT runtime code needs to handle some hotplug events */ | |
6681 | hotcpu_notifier(update_runtime, 0); | |
6682 | ||
b328ca18 | 6683 | init_hrtick(); |
5c1e1767 NP |
6684 | |
6685 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6686 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6687 | BUG(); |
19978ca6 | 6688 | sched_init_granularity(); |
dcc30a35 | 6689 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6690 | |
0e3900e6 | 6691 | init_sched_rt_class(); |
1da177e4 LT |
6692 | } |
6693 | #else | |
6694 | void __init sched_init_smp(void) | |
6695 | { | |
19978ca6 | 6696 | sched_init_granularity(); |
1da177e4 LT |
6697 | } |
6698 | #endif /* CONFIG_SMP */ | |
6699 | ||
cd1bb94b AB |
6700 | const_debug unsigned int sysctl_timer_migration = 1; |
6701 | ||
1da177e4 LT |
6702 | int in_sched_functions(unsigned long addr) |
6703 | { | |
1da177e4 LT |
6704 | return in_lock_functions(addr) || |
6705 | (addr >= (unsigned long)__sched_text_start | |
6706 | && addr < (unsigned long)__sched_text_end); | |
6707 | } | |
6708 | ||
029632fb PZ |
6709 | #ifdef CONFIG_CGROUP_SCHED |
6710 | struct task_group root_task_group; | |
052f1dc7 | 6711 | #endif |
6f505b16 | 6712 | |
029632fb | 6713 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6714 | |
1da177e4 LT |
6715 | void __init sched_init(void) |
6716 | { | |
dd41f596 | 6717 | int i, j; |
434d53b0 MT |
6718 | unsigned long alloc_size = 0, ptr; |
6719 | ||
6720 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6721 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6722 | #endif | |
6723 | #ifdef CONFIG_RT_GROUP_SCHED | |
6724 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6725 | #endif |
df7c8e84 | 6726 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6727 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6728 | #endif |
434d53b0 | 6729 | if (alloc_size) { |
36b7b6d4 | 6730 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6731 | |
6732 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6733 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6734 | ptr += nr_cpu_ids * sizeof(void **); |
6735 | ||
07e06b01 | 6736 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6737 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6738 | |
6d6bc0ad | 6739 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6740 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6741 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6742 | ptr += nr_cpu_ids * sizeof(void **); |
6743 | ||
07e06b01 | 6744 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6745 | ptr += nr_cpu_ids * sizeof(void **); |
6746 | ||
6d6bc0ad | 6747 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6748 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6749 | for_each_possible_cpu(i) { | |
6750 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6751 | ptr += cpumask_size(); | |
6752 | } | |
6753 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6754 | } |
dd41f596 | 6755 | |
57d885fe GH |
6756 | #ifdef CONFIG_SMP |
6757 | init_defrootdomain(); | |
6758 | #endif | |
6759 | ||
d0b27fa7 PZ |
6760 | init_rt_bandwidth(&def_rt_bandwidth, |
6761 | global_rt_period(), global_rt_runtime()); | |
6762 | ||
6763 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6764 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6765 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6766 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6767 | |
7c941438 | 6768 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6769 | list_add(&root_task_group.list, &task_groups); |
6770 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6771 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6772 | autogroup_init(&init_task); |
7c941438 | 6773 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6774 | |
0a945022 | 6775 | for_each_possible_cpu(i) { |
70b97a7f | 6776 | struct rq *rq; |
1da177e4 LT |
6777 | |
6778 | rq = cpu_rq(i); | |
05fa785c | 6779 | raw_spin_lock_init(&rq->lock); |
7897986b | 6780 | rq->nr_running = 0; |
dce48a84 TG |
6781 | rq->calc_load_active = 0; |
6782 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6783 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6784 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6785 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6786 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6787 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6788 | /* |
07e06b01 | 6789 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6790 | * |
6791 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6792 | * gets 100% of the cpu resources in the system. This overall | |
6793 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6794 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6795 | * based on each entity's (task or task-group's) weight |
6796 | * (se->load.weight). | |
6797 | * | |
07e06b01 | 6798 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6799 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6800 | * then A0's share of the cpu resource is: | |
6801 | * | |
0d905bca | 6802 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6803 | * |
07e06b01 YZ |
6804 | * We achieve this by letting root_task_group's tasks sit |
6805 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6806 | */ |
ab84d31e | 6807 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6808 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6809 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6810 | ||
6811 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6812 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6813 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6814 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6815 | #endif |
1da177e4 | 6816 | |
dd41f596 IM |
6817 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6818 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6819 | |
6820 | rq->last_load_update_tick = jiffies; | |
6821 | ||
1da177e4 | 6822 | #ifdef CONFIG_SMP |
41c7ce9a | 6823 | rq->sd = NULL; |
57d885fe | 6824 | rq->rd = NULL; |
1399fa78 | 6825 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6826 | rq->post_schedule = 0; |
1da177e4 | 6827 | rq->active_balance = 0; |
dd41f596 | 6828 | rq->next_balance = jiffies; |
1da177e4 | 6829 | rq->push_cpu = 0; |
0a2966b4 | 6830 | rq->cpu = i; |
1f11eb6a | 6831 | rq->online = 0; |
eae0c9df MG |
6832 | rq->idle_stamp = 0; |
6833 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 6834 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
6835 | #ifdef CONFIG_NO_HZ |
6836 | rq->nohz_balance_kick = 0; | |
83cd4fe2 | 6837 | #endif |
1da177e4 | 6838 | #endif |
8f4d37ec | 6839 | init_rq_hrtick(rq); |
1da177e4 | 6840 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6841 | } |
6842 | ||
2dd73a4f | 6843 | set_load_weight(&init_task); |
b50f60ce | 6844 | |
e107be36 AK |
6845 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6846 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6847 | #endif | |
6848 | ||
b50f60ce | 6849 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6850 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6851 | #endif |
6852 | ||
1da177e4 LT |
6853 | /* |
6854 | * The boot idle thread does lazy MMU switching as well: | |
6855 | */ | |
6856 | atomic_inc(&init_mm.mm_count); | |
6857 | enter_lazy_tlb(&init_mm, current); | |
6858 | ||
6859 | /* | |
6860 | * Make us the idle thread. Technically, schedule() should not be | |
6861 | * called from this thread, however somewhere below it might be, | |
6862 | * but because we are the idle thread, we just pick up running again | |
6863 | * when this runqueue becomes "idle". | |
6864 | */ | |
6865 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6866 | |
6867 | calc_load_update = jiffies + LOAD_FREQ; | |
6868 | ||
dd41f596 IM |
6869 | /* |
6870 | * During early bootup we pretend to be a normal task: | |
6871 | */ | |
6872 | current->sched_class = &fair_sched_class; | |
6892b75e | 6873 | |
bf4d83f6 | 6874 | #ifdef CONFIG_SMP |
4cb98839 | 6875 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6876 | /* May be allocated at isolcpus cmdline parse time */ |
6877 | if (cpu_isolated_map == NULL) | |
6878 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
029632fb PZ |
6879 | #endif |
6880 | init_sched_fair_class(); | |
6a7b3dc3 | 6881 | |
6892b75e | 6882 | scheduler_running = 1; |
1da177e4 LT |
6883 | } |
6884 | ||
d902db1e | 6885 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6886 | static inline int preempt_count_equals(int preempt_offset) |
6887 | { | |
234da7bc | 6888 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6889 | |
4ba8216c | 6890 | return (nested == preempt_offset); |
e4aafea2 FW |
6891 | } |
6892 | ||
d894837f | 6893 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6894 | { |
1da177e4 LT |
6895 | static unsigned long prev_jiffy; /* ratelimiting */ |
6896 | ||
b3fbab05 | 6897 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6898 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
6899 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
6900 | return; |
6901 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6902 | return; | |
6903 | prev_jiffy = jiffies; | |
6904 | ||
3df0fc5b PZ |
6905 | printk(KERN_ERR |
6906 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6907 | file, line); | |
6908 | printk(KERN_ERR | |
6909 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6910 | in_atomic(), irqs_disabled(), | |
6911 | current->pid, current->comm); | |
aef745fc IM |
6912 | |
6913 | debug_show_held_locks(current); | |
6914 | if (irqs_disabled()) | |
6915 | print_irqtrace_events(current); | |
6916 | dump_stack(); | |
1da177e4 LT |
6917 | } |
6918 | EXPORT_SYMBOL(__might_sleep); | |
6919 | #endif | |
6920 | ||
6921 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6922 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6923 | { | |
da7a735e PZ |
6924 | const struct sched_class *prev_class = p->sched_class; |
6925 | int old_prio = p->prio; | |
3a5e4dc1 | 6926 | int on_rq; |
3e51f33f | 6927 | |
fd2f4419 | 6928 | on_rq = p->on_rq; |
3a5e4dc1 AK |
6929 | if (on_rq) |
6930 | deactivate_task(rq, p, 0); | |
6931 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6932 | if (on_rq) { | |
6933 | activate_task(rq, p, 0); | |
6934 | resched_task(rq->curr); | |
6935 | } | |
da7a735e PZ |
6936 | |
6937 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
6938 | } |
6939 | ||
1da177e4 LT |
6940 | void normalize_rt_tasks(void) |
6941 | { | |
a0f98a1c | 6942 | struct task_struct *g, *p; |
1da177e4 | 6943 | unsigned long flags; |
70b97a7f | 6944 | struct rq *rq; |
1da177e4 | 6945 | |
4cf5d77a | 6946 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 6947 | do_each_thread(g, p) { |
178be793 IM |
6948 | /* |
6949 | * Only normalize user tasks: | |
6950 | */ | |
6951 | if (!p->mm) | |
6952 | continue; | |
6953 | ||
6cfb0d5d | 6954 | p->se.exec_start = 0; |
6cfb0d5d | 6955 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
6956 | p->se.statistics.wait_start = 0; |
6957 | p->se.statistics.sleep_start = 0; | |
6958 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 6959 | #endif |
dd41f596 IM |
6960 | |
6961 | if (!rt_task(p)) { | |
6962 | /* | |
6963 | * Renice negative nice level userspace | |
6964 | * tasks back to 0: | |
6965 | */ | |
6966 | if (TASK_NICE(p) < 0 && p->mm) | |
6967 | set_user_nice(p, 0); | |
1da177e4 | 6968 | continue; |
dd41f596 | 6969 | } |
1da177e4 | 6970 | |
1d615482 | 6971 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 6972 | rq = __task_rq_lock(p); |
1da177e4 | 6973 | |
178be793 | 6974 | normalize_task(rq, p); |
3a5e4dc1 | 6975 | |
b29739f9 | 6976 | __task_rq_unlock(rq); |
1d615482 | 6977 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
6978 | } while_each_thread(g, p); |
6979 | ||
4cf5d77a | 6980 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
6981 | } |
6982 | ||
6983 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6984 | |
67fc4e0c | 6985 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6986 | /* |
67fc4e0c | 6987 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6988 | * |
6989 | * They can only be called when the whole system has been | |
6990 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6991 | * activity can take place. Using them for anything else would | |
6992 | * be a serious bug, and as a result, they aren't even visible | |
6993 | * under any other configuration. | |
6994 | */ | |
6995 | ||
6996 | /** | |
6997 | * curr_task - return the current task for a given cpu. | |
6998 | * @cpu: the processor in question. | |
6999 | * | |
7000 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7001 | */ | |
36c8b586 | 7002 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7003 | { |
7004 | return cpu_curr(cpu); | |
7005 | } | |
7006 | ||
67fc4e0c JW |
7007 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7008 | ||
7009 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7010 | /** |
7011 | * set_curr_task - set the current task for a given cpu. | |
7012 | * @cpu: the processor in question. | |
7013 | * @p: the task pointer to set. | |
7014 | * | |
7015 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7016 | * are serviced on a separate stack. It allows the architecture to switch the |
7017 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7018 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7019 | * and caller must save the original value of the current task (see | |
7020 | * curr_task() above) and restore that value before reenabling interrupts and | |
7021 | * re-starting the system. | |
7022 | * | |
7023 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7024 | */ | |
36c8b586 | 7025 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7026 | { |
7027 | cpu_curr(cpu) = p; | |
7028 | } | |
7029 | ||
7030 | #endif | |
29f59db3 | 7031 | |
052f1dc7 | 7032 | #ifdef CONFIG_RT_GROUP_SCHED |
6d6bc0ad | 7033 | #else /* !CONFIG_RT_GROUP_SCHED */ |
6d6bc0ad | 7034 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 7035 | |
7c941438 | 7036 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7037 | /* task_group_lock serializes the addition/removal of task groups */ |
7038 | static DEFINE_SPINLOCK(task_group_lock); | |
7039 | ||
bccbe08a PZ |
7040 | static void free_sched_group(struct task_group *tg) |
7041 | { | |
7042 | free_fair_sched_group(tg); | |
7043 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7044 | autogroup_free(tg); |
bccbe08a PZ |
7045 | kfree(tg); |
7046 | } | |
7047 | ||
7048 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7049 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7050 | { |
7051 | struct task_group *tg; | |
7052 | unsigned long flags; | |
bccbe08a PZ |
7053 | |
7054 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7055 | if (!tg) | |
7056 | return ERR_PTR(-ENOMEM); | |
7057 | ||
ec7dc8ac | 7058 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7059 | goto err; |
7060 | ||
ec7dc8ac | 7061 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7062 | goto err; |
7063 | ||
8ed36996 | 7064 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7065 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7066 | |
7067 | WARN_ON(!parent); /* root should already exist */ | |
7068 | ||
7069 | tg->parent = parent; | |
f473aa5e | 7070 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7071 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7072 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7073 | |
9b5b7751 | 7074 | return tg; |
29f59db3 SV |
7075 | |
7076 | err: | |
6f505b16 | 7077 | free_sched_group(tg); |
29f59db3 SV |
7078 | return ERR_PTR(-ENOMEM); |
7079 | } | |
7080 | ||
9b5b7751 | 7081 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7082 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7083 | { |
29f59db3 | 7084 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7085 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7086 | } |
7087 | ||
9b5b7751 | 7088 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7089 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7090 | { |
8ed36996 | 7091 | unsigned long flags; |
9b5b7751 | 7092 | int i; |
29f59db3 | 7093 | |
3d4b47b4 PZ |
7094 | /* end participation in shares distribution */ |
7095 | for_each_possible_cpu(i) | |
bccbe08a | 7096 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7097 | |
7098 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7099 | list_del_rcu(&tg->list); |
f473aa5e | 7100 | list_del_rcu(&tg->siblings); |
8ed36996 | 7101 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7102 | |
9b5b7751 | 7103 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7104 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7105 | } |
7106 | ||
9b5b7751 | 7107 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7108 | * The caller of this function should have put the task in its new group |
7109 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7110 | * reflect its new group. | |
9b5b7751 SV |
7111 | */ |
7112 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7113 | { |
7114 | int on_rq, running; | |
7115 | unsigned long flags; | |
7116 | struct rq *rq; | |
7117 | ||
7118 | rq = task_rq_lock(tsk, &flags); | |
7119 | ||
051a1d1a | 7120 | running = task_current(rq, tsk); |
fd2f4419 | 7121 | on_rq = tsk->on_rq; |
29f59db3 | 7122 | |
0e1f3483 | 7123 | if (on_rq) |
29f59db3 | 7124 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7125 | if (unlikely(running)) |
7126 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7127 | |
810b3817 | 7128 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7129 | if (tsk->sched_class->task_move_group) |
7130 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7131 | else | |
810b3817 | 7132 | #endif |
b2b5ce02 | 7133 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7134 | |
0e1f3483 HS |
7135 | if (unlikely(running)) |
7136 | tsk->sched_class->set_curr_task(rq); | |
7137 | if (on_rq) | |
371fd7e7 | 7138 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7139 | |
0122ec5b | 7140 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7141 | } |
7c941438 | 7142 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7143 | |
052f1dc7 | 7144 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 | 7145 | #endif |
5cb350ba | 7146 | |
a790de99 | 7147 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7148 | static unsigned long to_ratio(u64 period, u64 runtime) |
7149 | { | |
7150 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7151 | return 1ULL << 20; |
9f0c1e56 | 7152 | |
9a7e0b18 | 7153 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7154 | } |
a790de99 PT |
7155 | #endif |
7156 | ||
7157 | #ifdef CONFIG_RT_GROUP_SCHED | |
7158 | /* | |
7159 | * Ensure that the real time constraints are schedulable. | |
7160 | */ | |
7161 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7162 | |
9a7e0b18 PZ |
7163 | /* Must be called with tasklist_lock held */ |
7164 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7165 | { |
9a7e0b18 | 7166 | struct task_struct *g, *p; |
b40b2e8e | 7167 | |
9a7e0b18 | 7168 | do_each_thread(g, p) { |
029632fb | 7169 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7170 | return 1; |
7171 | } while_each_thread(g, p); | |
b40b2e8e | 7172 | |
9a7e0b18 PZ |
7173 | return 0; |
7174 | } | |
b40b2e8e | 7175 | |
9a7e0b18 PZ |
7176 | struct rt_schedulable_data { |
7177 | struct task_group *tg; | |
7178 | u64 rt_period; | |
7179 | u64 rt_runtime; | |
7180 | }; | |
b40b2e8e | 7181 | |
a790de99 | 7182 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7183 | { |
7184 | struct rt_schedulable_data *d = data; | |
7185 | struct task_group *child; | |
7186 | unsigned long total, sum = 0; | |
7187 | u64 period, runtime; | |
b40b2e8e | 7188 | |
9a7e0b18 PZ |
7189 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7190 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7191 | |
9a7e0b18 PZ |
7192 | if (tg == d->tg) { |
7193 | period = d->rt_period; | |
7194 | runtime = d->rt_runtime; | |
b40b2e8e | 7195 | } |
b40b2e8e | 7196 | |
4653f803 PZ |
7197 | /* |
7198 | * Cannot have more runtime than the period. | |
7199 | */ | |
7200 | if (runtime > period && runtime != RUNTIME_INF) | |
7201 | return -EINVAL; | |
6f505b16 | 7202 | |
4653f803 PZ |
7203 | /* |
7204 | * Ensure we don't starve existing RT tasks. | |
7205 | */ | |
9a7e0b18 PZ |
7206 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7207 | return -EBUSY; | |
6f505b16 | 7208 | |
9a7e0b18 | 7209 | total = to_ratio(period, runtime); |
6f505b16 | 7210 | |
4653f803 PZ |
7211 | /* |
7212 | * Nobody can have more than the global setting allows. | |
7213 | */ | |
7214 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7215 | return -EINVAL; | |
6f505b16 | 7216 | |
4653f803 PZ |
7217 | /* |
7218 | * The sum of our children's runtime should not exceed our own. | |
7219 | */ | |
9a7e0b18 PZ |
7220 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7221 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7222 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7223 | |
9a7e0b18 PZ |
7224 | if (child == d->tg) { |
7225 | period = d->rt_period; | |
7226 | runtime = d->rt_runtime; | |
7227 | } | |
6f505b16 | 7228 | |
9a7e0b18 | 7229 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7230 | } |
6f505b16 | 7231 | |
9a7e0b18 PZ |
7232 | if (sum > total) |
7233 | return -EINVAL; | |
7234 | ||
7235 | return 0; | |
6f505b16 PZ |
7236 | } |
7237 | ||
9a7e0b18 | 7238 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7239 | { |
8277434e PT |
7240 | int ret; |
7241 | ||
9a7e0b18 PZ |
7242 | struct rt_schedulable_data data = { |
7243 | .tg = tg, | |
7244 | .rt_period = period, | |
7245 | .rt_runtime = runtime, | |
7246 | }; | |
7247 | ||
8277434e PT |
7248 | rcu_read_lock(); |
7249 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7250 | rcu_read_unlock(); | |
7251 | ||
7252 | return ret; | |
521f1a24 DG |
7253 | } |
7254 | ||
ab84d31e | 7255 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7256 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7257 | { |
ac086bc2 | 7258 | int i, err = 0; |
9f0c1e56 | 7259 | |
9f0c1e56 | 7260 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7261 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7262 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7263 | if (err) | |
9f0c1e56 | 7264 | goto unlock; |
ac086bc2 | 7265 | |
0986b11b | 7266 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7267 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7268 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7269 | |
7270 | for_each_possible_cpu(i) { | |
7271 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7272 | ||
0986b11b | 7273 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7274 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7275 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7276 | } |
0986b11b | 7277 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7278 | unlock: |
521f1a24 | 7279 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7280 | mutex_unlock(&rt_constraints_mutex); |
7281 | ||
7282 | return err; | |
6f505b16 PZ |
7283 | } |
7284 | ||
d0b27fa7 PZ |
7285 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7286 | { | |
7287 | u64 rt_runtime, rt_period; | |
7288 | ||
7289 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7290 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7291 | if (rt_runtime_us < 0) | |
7292 | rt_runtime = RUNTIME_INF; | |
7293 | ||
ab84d31e | 7294 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7295 | } |
7296 | ||
9f0c1e56 PZ |
7297 | long sched_group_rt_runtime(struct task_group *tg) |
7298 | { | |
7299 | u64 rt_runtime_us; | |
7300 | ||
d0b27fa7 | 7301 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7302 | return -1; |
7303 | ||
d0b27fa7 | 7304 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7305 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7306 | return rt_runtime_us; | |
7307 | } | |
d0b27fa7 PZ |
7308 | |
7309 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7310 | { | |
7311 | u64 rt_runtime, rt_period; | |
7312 | ||
7313 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7314 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7315 | ||
619b0488 R |
7316 | if (rt_period == 0) |
7317 | return -EINVAL; | |
7318 | ||
ab84d31e | 7319 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7320 | } |
7321 | ||
7322 | long sched_group_rt_period(struct task_group *tg) | |
7323 | { | |
7324 | u64 rt_period_us; | |
7325 | ||
7326 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7327 | do_div(rt_period_us, NSEC_PER_USEC); | |
7328 | return rt_period_us; | |
7329 | } | |
7330 | ||
7331 | static int sched_rt_global_constraints(void) | |
7332 | { | |
4653f803 | 7333 | u64 runtime, period; |
d0b27fa7 PZ |
7334 | int ret = 0; |
7335 | ||
ec5d4989 HS |
7336 | if (sysctl_sched_rt_period <= 0) |
7337 | return -EINVAL; | |
7338 | ||
4653f803 PZ |
7339 | runtime = global_rt_runtime(); |
7340 | period = global_rt_period(); | |
7341 | ||
7342 | /* | |
7343 | * Sanity check on the sysctl variables. | |
7344 | */ | |
7345 | if (runtime > period && runtime != RUNTIME_INF) | |
7346 | return -EINVAL; | |
10b612f4 | 7347 | |
d0b27fa7 | 7348 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7349 | read_lock(&tasklist_lock); |
4653f803 | 7350 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7351 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7352 | mutex_unlock(&rt_constraints_mutex); |
7353 | ||
7354 | return ret; | |
7355 | } | |
54e99124 DG |
7356 | |
7357 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7358 | { | |
7359 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7360 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7361 | return 0; | |
7362 | ||
7363 | return 1; | |
7364 | } | |
7365 | ||
6d6bc0ad | 7366 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7367 | static int sched_rt_global_constraints(void) |
7368 | { | |
ac086bc2 PZ |
7369 | unsigned long flags; |
7370 | int i; | |
7371 | ||
ec5d4989 HS |
7372 | if (sysctl_sched_rt_period <= 0) |
7373 | return -EINVAL; | |
7374 | ||
60aa605d PZ |
7375 | /* |
7376 | * There's always some RT tasks in the root group | |
7377 | * -- migration, kstopmachine etc.. | |
7378 | */ | |
7379 | if (sysctl_sched_rt_runtime == 0) | |
7380 | return -EBUSY; | |
7381 | ||
0986b11b | 7382 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7383 | for_each_possible_cpu(i) { |
7384 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7385 | ||
0986b11b | 7386 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7387 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7388 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7389 | } |
0986b11b | 7390 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7391 | |
d0b27fa7 PZ |
7392 | return 0; |
7393 | } | |
6d6bc0ad | 7394 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7395 | |
7396 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7397 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7398 | loff_t *ppos) |
7399 | { | |
7400 | int ret; | |
7401 | int old_period, old_runtime; | |
7402 | static DEFINE_MUTEX(mutex); | |
7403 | ||
7404 | mutex_lock(&mutex); | |
7405 | old_period = sysctl_sched_rt_period; | |
7406 | old_runtime = sysctl_sched_rt_runtime; | |
7407 | ||
8d65af78 | 7408 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7409 | |
7410 | if (!ret && write) { | |
7411 | ret = sched_rt_global_constraints(); | |
7412 | if (ret) { | |
7413 | sysctl_sched_rt_period = old_period; | |
7414 | sysctl_sched_rt_runtime = old_runtime; | |
7415 | } else { | |
7416 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7417 | def_rt_bandwidth.rt_period = | |
7418 | ns_to_ktime(global_rt_period()); | |
7419 | } | |
7420 | } | |
7421 | mutex_unlock(&mutex); | |
7422 | ||
7423 | return ret; | |
7424 | } | |
68318b8e | 7425 | |
052f1dc7 | 7426 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7427 | |
7428 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7429 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7430 | { |
2b01dfe3 PM |
7431 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7432 | struct task_group, css); | |
68318b8e SV |
7433 | } |
7434 | ||
7435 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7436 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 7437 | { |
ec7dc8ac | 7438 | struct task_group *tg, *parent; |
68318b8e | 7439 | |
2b01dfe3 | 7440 | if (!cgrp->parent) { |
68318b8e | 7441 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7442 | return &root_task_group.css; |
68318b8e SV |
7443 | } |
7444 | ||
ec7dc8ac DG |
7445 | parent = cgroup_tg(cgrp->parent); |
7446 | tg = sched_create_group(parent); | |
68318b8e SV |
7447 | if (IS_ERR(tg)) |
7448 | return ERR_PTR(-ENOMEM); | |
7449 | ||
68318b8e SV |
7450 | return &tg->css; |
7451 | } | |
7452 | ||
41a2d6cf IM |
7453 | static void |
7454 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7455 | { |
2b01dfe3 | 7456 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7457 | |
7458 | sched_destroy_group(tg); | |
7459 | } | |
7460 | ||
41a2d6cf | 7461 | static int |
be367d09 | 7462 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 7463 | { |
b68aa230 | 7464 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 7465 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
7466 | return -EINVAL; |
7467 | #else | |
68318b8e SV |
7468 | /* We don't support RT-tasks being in separate groups */ |
7469 | if (tsk->sched_class != &fair_sched_class) | |
7470 | return -EINVAL; | |
b68aa230 | 7471 | #endif |
be367d09 BB |
7472 | return 0; |
7473 | } | |
68318b8e | 7474 | |
68318b8e | 7475 | static void |
f780bdb7 | 7476 | cpu_cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e SV |
7477 | { |
7478 | sched_move_task(tsk); | |
7479 | } | |
7480 | ||
068c5cc5 | 7481 | static void |
d41d5a01 PZ |
7482 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7483 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
7484 | { |
7485 | /* | |
7486 | * cgroup_exit() is called in the copy_process() failure path. | |
7487 | * Ignore this case since the task hasn't ran yet, this avoids | |
7488 | * trying to poke a half freed task state from generic code. | |
7489 | */ | |
7490 | if (!(task->flags & PF_EXITING)) | |
7491 | return; | |
7492 | ||
7493 | sched_move_task(task); | |
7494 | } | |
7495 | ||
052f1dc7 | 7496 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7497 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7498 | u64 shareval) |
68318b8e | 7499 | { |
c8b28116 | 7500 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7501 | } |
7502 | ||
f4c753b7 | 7503 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7504 | { |
2b01dfe3 | 7505 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7506 | |
c8b28116 | 7507 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7508 | } |
ab84d31e PT |
7509 | |
7510 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7511 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7512 | ||
ab84d31e PT |
7513 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7514 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7515 | ||
a790de99 PT |
7516 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7517 | ||
ab84d31e PT |
7518 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7519 | { | |
56f570e5 | 7520 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7521 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7522 | |
7523 | if (tg == &root_task_group) | |
7524 | return -EINVAL; | |
7525 | ||
7526 | /* | |
7527 | * Ensure we have at some amount of bandwidth every period. This is | |
7528 | * to prevent reaching a state of large arrears when throttled via | |
7529 | * entity_tick() resulting in prolonged exit starvation. | |
7530 | */ | |
7531 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7532 | return -EINVAL; | |
7533 | ||
7534 | /* | |
7535 | * Likewise, bound things on the otherside by preventing insane quota | |
7536 | * periods. This also allows us to normalize in computing quota | |
7537 | * feasibility. | |
7538 | */ | |
7539 | if (period > max_cfs_quota_period) | |
7540 | return -EINVAL; | |
7541 | ||
a790de99 PT |
7542 | mutex_lock(&cfs_constraints_mutex); |
7543 | ret = __cfs_schedulable(tg, period, quota); | |
7544 | if (ret) | |
7545 | goto out_unlock; | |
7546 | ||
58088ad0 | 7547 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7548 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7549 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7550 | raw_spin_lock_irq(&cfs_b->lock); |
7551 | cfs_b->period = ns_to_ktime(period); | |
7552 | cfs_b->quota = quota; | |
58088ad0 | 7553 | |
a9cf55b2 | 7554 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7555 | /* restart the period timer (if active) to handle new period expiry */ |
7556 | if (runtime_enabled && cfs_b->timer_active) { | |
7557 | /* force a reprogram */ | |
7558 | cfs_b->timer_active = 0; | |
7559 | __start_cfs_bandwidth(cfs_b); | |
7560 | } | |
ab84d31e PT |
7561 | raw_spin_unlock_irq(&cfs_b->lock); |
7562 | ||
7563 | for_each_possible_cpu(i) { | |
7564 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7565 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7566 | |
7567 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7568 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7569 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7570 | |
029632fb | 7571 | if (cfs_rq->throttled) |
671fd9da | 7572 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7573 | raw_spin_unlock_irq(&rq->lock); |
7574 | } | |
a790de99 PT |
7575 | out_unlock: |
7576 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7577 | |
a790de99 | 7578 | return ret; |
ab84d31e PT |
7579 | } |
7580 | ||
7581 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7582 | { | |
7583 | u64 quota, period; | |
7584 | ||
029632fb | 7585 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7586 | if (cfs_quota_us < 0) |
7587 | quota = RUNTIME_INF; | |
7588 | else | |
7589 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7590 | ||
7591 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7592 | } | |
7593 | ||
7594 | long tg_get_cfs_quota(struct task_group *tg) | |
7595 | { | |
7596 | u64 quota_us; | |
7597 | ||
029632fb | 7598 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7599 | return -1; |
7600 | ||
029632fb | 7601 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7602 | do_div(quota_us, NSEC_PER_USEC); |
7603 | ||
7604 | return quota_us; | |
7605 | } | |
7606 | ||
7607 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7608 | { | |
7609 | u64 quota, period; | |
7610 | ||
7611 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7612 | quota = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7613 | |
7614 | if (period <= 0) | |
7615 | return -EINVAL; | |
7616 | ||
7617 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7618 | } | |
7619 | ||
7620 | long tg_get_cfs_period(struct task_group *tg) | |
7621 | { | |
7622 | u64 cfs_period_us; | |
7623 | ||
029632fb | 7624 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7625 | do_div(cfs_period_us, NSEC_PER_USEC); |
7626 | ||
7627 | return cfs_period_us; | |
7628 | } | |
7629 | ||
7630 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7631 | { | |
7632 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7633 | } | |
7634 | ||
7635 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7636 | s64 cfs_quota_us) | |
7637 | { | |
7638 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7639 | } | |
7640 | ||
7641 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7642 | { | |
7643 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7644 | } | |
7645 | ||
7646 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7647 | u64 cfs_period_us) | |
7648 | { | |
7649 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7650 | } | |
7651 | ||
a790de99 PT |
7652 | struct cfs_schedulable_data { |
7653 | struct task_group *tg; | |
7654 | u64 period, quota; | |
7655 | }; | |
7656 | ||
7657 | /* | |
7658 | * normalize group quota/period to be quota/max_period | |
7659 | * note: units are usecs | |
7660 | */ | |
7661 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7662 | struct cfs_schedulable_data *d) | |
7663 | { | |
7664 | u64 quota, period; | |
7665 | ||
7666 | if (tg == d->tg) { | |
7667 | period = d->period; | |
7668 | quota = d->quota; | |
7669 | } else { | |
7670 | period = tg_get_cfs_period(tg); | |
7671 | quota = tg_get_cfs_quota(tg); | |
7672 | } | |
7673 | ||
7674 | /* note: these should typically be equivalent */ | |
7675 | if (quota == RUNTIME_INF || quota == -1) | |
7676 | return RUNTIME_INF; | |
7677 | ||
7678 | return to_ratio(period, quota); | |
7679 | } | |
7680 | ||
7681 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7682 | { | |
7683 | struct cfs_schedulable_data *d = data; | |
029632fb | 7684 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7685 | s64 quota = 0, parent_quota = -1; |
7686 | ||
7687 | if (!tg->parent) { | |
7688 | quota = RUNTIME_INF; | |
7689 | } else { | |
029632fb | 7690 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7691 | |
7692 | quota = normalize_cfs_quota(tg, d); | |
7693 | parent_quota = parent_b->hierarchal_quota; | |
7694 | ||
7695 | /* | |
7696 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7697 | * limit is set | |
7698 | */ | |
7699 | if (quota == RUNTIME_INF) | |
7700 | quota = parent_quota; | |
7701 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7702 | return -EINVAL; | |
7703 | } | |
7704 | cfs_b->hierarchal_quota = quota; | |
7705 | ||
7706 | return 0; | |
7707 | } | |
7708 | ||
7709 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7710 | { | |
8277434e | 7711 | int ret; |
a790de99 PT |
7712 | struct cfs_schedulable_data data = { |
7713 | .tg = tg, | |
7714 | .period = period, | |
7715 | .quota = quota, | |
7716 | }; | |
7717 | ||
7718 | if (quota != RUNTIME_INF) { | |
7719 | do_div(data.period, NSEC_PER_USEC); | |
7720 | do_div(data.quota, NSEC_PER_USEC); | |
7721 | } | |
7722 | ||
8277434e PT |
7723 | rcu_read_lock(); |
7724 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7725 | rcu_read_unlock(); | |
7726 | ||
7727 | return ret; | |
a790de99 | 7728 | } |
e8da1b18 NR |
7729 | |
7730 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7731 | struct cgroup_map_cb *cb) | |
7732 | { | |
7733 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7734 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7735 | |
7736 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7737 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7738 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7739 | ||
7740 | return 0; | |
7741 | } | |
ab84d31e | 7742 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7743 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7744 | |
052f1dc7 | 7745 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7746 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7747 | s64 val) |
6f505b16 | 7748 | { |
06ecb27c | 7749 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7750 | } |
7751 | ||
06ecb27c | 7752 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7753 | { |
06ecb27c | 7754 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7755 | } |
d0b27fa7 PZ |
7756 | |
7757 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7758 | u64 rt_period_us) | |
7759 | { | |
7760 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7761 | } | |
7762 | ||
7763 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7764 | { | |
7765 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7766 | } | |
6d6bc0ad | 7767 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7768 | |
fe5c7cc2 | 7769 | static struct cftype cpu_files[] = { |
052f1dc7 | 7770 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7771 | { |
7772 | .name = "shares", | |
f4c753b7 PM |
7773 | .read_u64 = cpu_shares_read_u64, |
7774 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7775 | }, |
052f1dc7 | 7776 | #endif |
ab84d31e PT |
7777 | #ifdef CONFIG_CFS_BANDWIDTH |
7778 | { | |
7779 | .name = "cfs_quota_us", | |
7780 | .read_s64 = cpu_cfs_quota_read_s64, | |
7781 | .write_s64 = cpu_cfs_quota_write_s64, | |
7782 | }, | |
7783 | { | |
7784 | .name = "cfs_period_us", | |
7785 | .read_u64 = cpu_cfs_period_read_u64, | |
7786 | .write_u64 = cpu_cfs_period_write_u64, | |
7787 | }, | |
e8da1b18 NR |
7788 | { |
7789 | .name = "stat", | |
7790 | .read_map = cpu_stats_show, | |
7791 | }, | |
ab84d31e | 7792 | #endif |
052f1dc7 | 7793 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7794 | { |
9f0c1e56 | 7795 | .name = "rt_runtime_us", |
06ecb27c PM |
7796 | .read_s64 = cpu_rt_runtime_read, |
7797 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7798 | }, |
d0b27fa7 PZ |
7799 | { |
7800 | .name = "rt_period_us", | |
f4c753b7 PM |
7801 | .read_u64 = cpu_rt_period_read_uint, |
7802 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7803 | }, |
052f1dc7 | 7804 | #endif |
68318b8e SV |
7805 | }; |
7806 | ||
7807 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7808 | { | |
fe5c7cc2 | 7809 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7810 | } |
7811 | ||
7812 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7813 | .name = "cpu", |
7814 | .create = cpu_cgroup_create, | |
7815 | .destroy = cpu_cgroup_destroy, | |
f780bdb7 BB |
7816 | .can_attach_task = cpu_cgroup_can_attach_task, |
7817 | .attach_task = cpu_cgroup_attach_task, | |
068c5cc5 | 7818 | .exit = cpu_cgroup_exit, |
38605cae IM |
7819 | .populate = cpu_cgroup_populate, |
7820 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7821 | .early_init = 1, |
7822 | }; | |
7823 | ||
052f1dc7 | 7824 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7825 | |
7826 | #ifdef CONFIG_CGROUP_CPUACCT | |
7827 | ||
7828 | /* | |
7829 | * CPU accounting code for task groups. | |
7830 | * | |
7831 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7832 | * (balbir@in.ibm.com). | |
7833 | */ | |
7834 | ||
934352f2 | 7835 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
7836 | struct cpuacct { |
7837 | struct cgroup_subsys_state css; | |
7838 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 7839 | u64 __percpu *cpuusage; |
ef12fefa | 7840 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 7841 | struct cpuacct *parent; |
d842de87 SV |
7842 | }; |
7843 | ||
7844 | struct cgroup_subsys cpuacct_subsys; | |
7845 | ||
7846 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 7847 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 7848 | { |
32cd756a | 7849 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
7850 | struct cpuacct, css); |
7851 | } | |
7852 | ||
7853 | /* return cpu accounting group to which this task belongs */ | |
7854 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7855 | { | |
7856 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7857 | struct cpuacct, css); | |
7858 | } | |
7859 | ||
7860 | /* create a new cpu accounting group */ | |
7861 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 7862 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
7863 | { |
7864 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 7865 | int i; |
d842de87 SV |
7866 | |
7867 | if (!ca) | |
ef12fefa | 7868 | goto out; |
d842de87 SV |
7869 | |
7870 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
7871 | if (!ca->cpuusage) |
7872 | goto out_free_ca; | |
7873 | ||
7874 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
7875 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
7876 | goto out_free_counters; | |
d842de87 | 7877 | |
934352f2 BR |
7878 | if (cgrp->parent) |
7879 | ca->parent = cgroup_ca(cgrp->parent); | |
7880 | ||
d842de87 | 7881 | return &ca->css; |
ef12fefa BR |
7882 | |
7883 | out_free_counters: | |
7884 | while (--i >= 0) | |
7885 | percpu_counter_destroy(&ca->cpustat[i]); | |
7886 | free_percpu(ca->cpuusage); | |
7887 | out_free_ca: | |
7888 | kfree(ca); | |
7889 | out: | |
7890 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
7891 | } |
7892 | ||
7893 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 7894 | static void |
32cd756a | 7895 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7896 | { |
32cd756a | 7897 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 7898 | int i; |
d842de87 | 7899 | |
ef12fefa BR |
7900 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
7901 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
7902 | free_percpu(ca->cpuusage); |
7903 | kfree(ca); | |
7904 | } | |
7905 | ||
720f5498 KC |
7906 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
7907 | { | |
b36128c8 | 7908 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7909 | u64 data; |
7910 | ||
7911 | #ifndef CONFIG_64BIT | |
7912 | /* | |
7913 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
7914 | */ | |
05fa785c | 7915 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 7916 | data = *cpuusage; |
05fa785c | 7917 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
7918 | #else |
7919 | data = *cpuusage; | |
7920 | #endif | |
7921 | ||
7922 | return data; | |
7923 | } | |
7924 | ||
7925 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
7926 | { | |
b36128c8 | 7927 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7928 | |
7929 | #ifndef CONFIG_64BIT | |
7930 | /* | |
7931 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
7932 | */ | |
05fa785c | 7933 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 7934 | *cpuusage = val; |
05fa785c | 7935 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
7936 | #else |
7937 | *cpuusage = val; | |
7938 | #endif | |
7939 | } | |
7940 | ||
d842de87 | 7941 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 7942 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 7943 | { |
32cd756a | 7944 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
7945 | u64 totalcpuusage = 0; |
7946 | int i; | |
7947 | ||
720f5498 KC |
7948 | for_each_present_cpu(i) |
7949 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
7950 | |
7951 | return totalcpuusage; | |
7952 | } | |
7953 | ||
0297b803 DG |
7954 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
7955 | u64 reset) | |
7956 | { | |
7957 | struct cpuacct *ca = cgroup_ca(cgrp); | |
7958 | int err = 0; | |
7959 | int i; | |
7960 | ||
7961 | if (reset) { | |
7962 | err = -EINVAL; | |
7963 | goto out; | |
7964 | } | |
7965 | ||
720f5498 KC |
7966 | for_each_present_cpu(i) |
7967 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 7968 | |
0297b803 DG |
7969 | out: |
7970 | return err; | |
7971 | } | |
7972 | ||
e9515c3c KC |
7973 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
7974 | struct seq_file *m) | |
7975 | { | |
7976 | struct cpuacct *ca = cgroup_ca(cgroup); | |
7977 | u64 percpu; | |
7978 | int i; | |
7979 | ||
7980 | for_each_present_cpu(i) { | |
7981 | percpu = cpuacct_cpuusage_read(ca, i); | |
7982 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
7983 | } | |
7984 | seq_printf(m, "\n"); | |
7985 | return 0; | |
7986 | } | |
7987 | ||
ef12fefa BR |
7988 | static const char *cpuacct_stat_desc[] = { |
7989 | [CPUACCT_STAT_USER] = "user", | |
7990 | [CPUACCT_STAT_SYSTEM] = "system", | |
7991 | }; | |
7992 | ||
7993 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7994 | struct cgroup_map_cb *cb) | |
7995 | { | |
7996 | struct cpuacct *ca = cgroup_ca(cgrp); | |
7997 | int i; | |
7998 | ||
7999 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8000 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8001 | val = cputime64_to_clock_t(val); | |
8002 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8003 | } | |
8004 | return 0; | |
8005 | } | |
8006 | ||
d842de87 SV |
8007 | static struct cftype files[] = { |
8008 | { | |
8009 | .name = "usage", | |
f4c753b7 PM |
8010 | .read_u64 = cpuusage_read, |
8011 | .write_u64 = cpuusage_write, | |
d842de87 | 8012 | }, |
e9515c3c KC |
8013 | { |
8014 | .name = "usage_percpu", | |
8015 | .read_seq_string = cpuacct_percpu_seq_read, | |
8016 | }, | |
ef12fefa BR |
8017 | { |
8018 | .name = "stat", | |
8019 | .read_map = cpuacct_stats_show, | |
8020 | }, | |
d842de87 SV |
8021 | }; |
8022 | ||
32cd756a | 8023 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8024 | { |
32cd756a | 8025 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8026 | } |
8027 | ||
8028 | /* | |
8029 | * charge this task's execution time to its accounting group. | |
8030 | * | |
8031 | * called with rq->lock held. | |
8032 | */ | |
029632fb | 8033 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8034 | { |
8035 | struct cpuacct *ca; | |
934352f2 | 8036 | int cpu; |
d842de87 | 8037 | |
c40c6f85 | 8038 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8039 | return; |
8040 | ||
934352f2 | 8041 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8042 | |
8043 | rcu_read_lock(); | |
8044 | ||
d842de87 | 8045 | ca = task_ca(tsk); |
d842de87 | 8046 | |
934352f2 | 8047 | for (; ca; ca = ca->parent) { |
b36128c8 | 8048 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8049 | *cpuusage += cputime; |
8050 | } | |
a18b83b7 BR |
8051 | |
8052 | rcu_read_unlock(); | |
d842de87 SV |
8053 | } |
8054 | ||
fa535a77 AB |
8055 | /* |
8056 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8057 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8058 | * percpu_counter_add with values large enough to always overflow the | |
8059 | * per cpu batch limit causing bad SMP scalability. | |
8060 | * | |
8061 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8062 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8063 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8064 | */ | |
8065 | #ifdef CONFIG_SMP | |
8066 | #define CPUACCT_BATCH \ | |
8067 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8068 | #else | |
8069 | #define CPUACCT_BATCH 0 | |
8070 | #endif | |
8071 | ||
ef12fefa BR |
8072 | /* |
8073 | * Charge the system/user time to the task's accounting group. | |
8074 | */ | |
029632fb | 8075 | void cpuacct_update_stats(struct task_struct *tsk, |
ef12fefa BR |
8076 | enum cpuacct_stat_index idx, cputime_t val) |
8077 | { | |
8078 | struct cpuacct *ca; | |
fa535a77 | 8079 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8080 | |
8081 | if (unlikely(!cpuacct_subsys.active)) | |
8082 | return; | |
8083 | ||
8084 | rcu_read_lock(); | |
8085 | ca = task_ca(tsk); | |
8086 | ||
8087 | do { | |
fa535a77 | 8088 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8089 | ca = ca->parent; |
8090 | } while (ca); | |
8091 | rcu_read_unlock(); | |
8092 | } | |
8093 | ||
d842de87 SV |
8094 | struct cgroup_subsys cpuacct_subsys = { |
8095 | .name = "cpuacct", | |
8096 | .create = cpuacct_create, | |
8097 | .destroy = cpuacct_destroy, | |
8098 | .populate = cpuacct_populate, | |
8099 | .subsys_id = cpuacct_subsys_id, | |
8100 | }; | |
8101 | #endif /* CONFIG_CGROUP_CPUACCT */ |