diff options
author | Martin Schwidefsky <schwidefsky@de.ibm.com> | 2011-12-19 13:23:15 -0500 |
---|---|---|
committer | Martin Schwidefsky <schwidefsky@de.ibm.com> | 2011-12-19 13:23:15 -0500 |
commit | 612ef28a045efadb3a98d4492ead7806a146485d (patch) | |
tree | 05621c87b37e91c27b06d450d76adffe97ce9666 /kernel/sched/core.c | |
parent | c3e0ef9a298e028a82ada28101ccd5cf64d209ee (diff) | |
parent | 07cde2608a3b5c66515363f1b53623b1536b9785 (diff) |
Merge branch 'sched/core' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip into cputime-tip
Conflicts:
drivers/cpufreq/cpufreq_conservative.c
drivers/cpufreq/cpufreq_ondemand.c
drivers/macintosh/rack-meter.c
fs/proc/stat.c
fs/proc/uptime.c
kernel/sched/core.c
Diffstat (limited to 'kernel/sched/core.c')
-rw-r--r-- | kernel/sched/core.c | 8119 |
1 files changed, 8119 insertions, 0 deletions
diff --git a/kernel/sched/core.c b/kernel/sched/core.c new file mode 100644 index 00000000000..cdf51a2adc2 --- /dev/null +++ b/kernel/sched/core.c | |||
@@ -0,0 +1,8119 @@ | |||
1 | /* | ||
2 | * kernel/sched/core.c | ||
3 | * | ||
4 | * Kernel scheduler and related syscalls | ||
5 | * | ||
6 | * Copyright (C) 1991-2002 Linus Torvalds | ||
7 | * | ||
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | ||
9 | * make semaphores SMP safe | ||
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | ||
11 | * by Andrea Arcangeli | ||
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | ||
13 | * hybrid priority-list and round-robin design with | ||
14 | * an array-switch method of distributing timeslices | ||
15 | * and per-CPU runqueues. Cleanups and useful suggestions | ||
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | ||
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | ||
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | ||
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 | ||
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, | ||
26 | * Thomas Gleixner, Mike Kravetz | ||
27 | */ | ||
28 | |||
29 | #include <linux/mm.h> | ||
30 | #include <linux/module.h> | ||
31 | #include <linux/nmi.h> | ||
32 | #include <linux/init.h> | ||
33 | #include <linux/uaccess.h> | ||
34 | #include <linux/highmem.h> | ||
35 | #include <asm/mmu_context.h> | ||
36 | #include <linux/interrupt.h> | ||
37 | #include <linux/capability.h> | ||
38 | #include <linux/completion.h> | ||
39 | #include <linux/kernel_stat.h> | ||
40 | #include <linux/debug_locks.h> | ||
41 | #include <linux/perf_event.h> | ||
42 | #include <linux/security.h> | ||
43 | #include <linux/notifier.h> | ||
44 | #include <linux/profile.h> | ||
45 | #include <linux/freezer.h> | ||
46 | #include <linux/vmalloc.h> | ||
47 | #include <linux/blkdev.h> | ||
48 | #include <linux/delay.h> | ||
49 | #include <linux/pid_namespace.h> | ||
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> | ||
57 | #include <linux/proc_fs.h> | ||
58 | #include <linux/seq_file.h> | ||
59 | #include <linux/sysctl.h> | ||
60 | #include <linux/syscalls.h> | ||
61 | #include <linux/times.h> | ||
62 | #include <linux/tsacct_kern.h> | ||
63 | #include <linux/kprobes.h> | ||
64 | #include <linux/delayacct.h> | ||
65 | #include <linux/unistd.h> | ||
66 | #include <linux/pagemap.h> | ||
67 | #include <linux/hrtimer.h> | ||
68 | #include <linux/tick.h> | ||
69 | #include <linux/debugfs.h> | ||
70 | #include <linux/ctype.h> | ||
71 | #include <linux/ftrace.h> | ||
72 | #include <linux/slab.h> | ||
73 | #include <linux/init_task.h> | ||
74 | |||
75 | #include <asm/tlb.h> | ||
76 | #include <asm/irq_regs.h> | ||
77 | #ifdef CONFIG_PARAVIRT | ||
78 | #include <asm/paravirt.h> | ||
79 | #endif | ||
80 | |||
81 | #include "sched.h" | ||
82 | #include "../workqueue_sched.h" | ||
83 | |||
84 | #define CREATE_TRACE_POINTS | ||
85 | #include <trace/events/sched.h> | ||
86 | |||
87 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) | ||
88 | { | ||
89 | unsigned long delta; | ||
90 | ktime_t soft, hard, now; | ||
91 | |||
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); | ||
98 | |||
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 | |||
107 | DEFINE_MUTEX(sched_domains_mutex); | ||
108 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | ||
109 | |||
110 | static void update_rq_clock_task(struct rq *rq, s64 delta); | ||
111 | |||
112 | void update_rq_clock(struct rq *rq) | ||
113 | { | ||
114 | s64 delta; | ||
115 | |||
116 | if (rq->skip_clock_update > 0) | ||
117 | return; | ||
118 | |||
119 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; | ||
120 | rq->clock += delta; | ||
121 | update_rq_clock_task(rq, delta); | ||
122 | } | ||
123 | |||
124 | /* | ||
125 | * Debugging: various feature bits | ||
126 | */ | ||
127 | |||
128 | #define SCHED_FEAT(name, enabled) \ | ||
129 | (1UL << __SCHED_FEAT_##name) * enabled | | ||
130 | |||
131 | const_debug unsigned int sysctl_sched_features = | ||
132 | #include "features.h" | ||
133 | 0; | ||
134 | |||
135 | #undef SCHED_FEAT | ||
136 | |||
137 | #ifdef CONFIG_SCHED_DEBUG | ||
138 | #define SCHED_FEAT(name, enabled) \ | ||
139 | #name , | ||
140 | |||
141 | static __read_mostly char *sched_feat_names[] = { | ||
142 | #include "features.h" | ||
143 | NULL | ||
144 | }; | ||
145 | |||
146 | #undef SCHED_FEAT | ||
147 | |||
148 | static int sched_feat_show(struct seq_file *m, void *v) | ||
149 | { | ||
150 | int i; | ||
151 | |||
152 | for (i = 0; i < __SCHED_FEAT_NR; i++) { | ||
153 | if (!(sysctl_sched_features & (1UL << i))) | ||
154 | seq_puts(m, "NO_"); | ||
155 | seq_printf(m, "%s ", sched_feat_names[i]); | ||
156 | } | ||
157 | seq_puts(m, "\n"); | ||
158 | |||
159 | return 0; | ||
160 | } | ||
161 | |||
162 | #ifdef HAVE_JUMP_LABEL | ||
163 | |||
164 | #define jump_label_key__true jump_label_key_enabled | ||
165 | #define jump_label_key__false jump_label_key_disabled | ||
166 | |||
167 | #define SCHED_FEAT(name, enabled) \ | ||
168 | jump_label_key__##enabled , | ||
169 | |||
170 | struct jump_label_key sched_feat_keys[__SCHED_FEAT_NR] = { | ||
171 | #include "features.h" | ||
172 | }; | ||
173 | |||
174 | #undef SCHED_FEAT | ||
175 | |||
176 | static void sched_feat_disable(int i) | ||
177 | { | ||
178 | if (jump_label_enabled(&sched_feat_keys[i])) | ||
179 | jump_label_dec(&sched_feat_keys[i]); | ||
180 | } | ||
181 | |||
182 | static void sched_feat_enable(int i) | ||
183 | { | ||
184 | if (!jump_label_enabled(&sched_feat_keys[i])) | ||
185 | jump_label_inc(&sched_feat_keys[i]); | ||
186 | } | ||
187 | #else | ||
188 | static void sched_feat_disable(int i) { }; | ||
189 | static void sched_feat_enable(int i) { }; | ||
190 | #endif /* HAVE_JUMP_LABEL */ | ||
191 | |||
192 | static ssize_t | ||
193 | sched_feat_write(struct file *filp, const char __user *ubuf, | ||
194 | size_t cnt, loff_t *ppos) | ||
195 | { | ||
196 | char buf[64]; | ||
197 | char *cmp; | ||
198 | int neg = 0; | ||
199 | int i; | ||
200 | |||
201 | if (cnt > 63) | ||
202 | cnt = 63; | ||
203 | |||
204 | if (copy_from_user(&buf, ubuf, cnt)) | ||
205 | return -EFAULT; | ||
206 | |||
207 | buf[cnt] = 0; | ||
208 | cmp = strstrip(buf); | ||
209 | |||
210 | if (strncmp(cmp, "NO_", 3) == 0) { | ||
211 | neg = 1; | ||
212 | cmp += 3; | ||
213 | } | ||
214 | |||
215 | for (i = 0; i < __SCHED_FEAT_NR; i++) { | ||
216 | if (strcmp(cmp, sched_feat_names[i]) == 0) { | ||
217 | if (neg) { | ||
218 | sysctl_sched_features &= ~(1UL << i); | ||
219 | sched_feat_disable(i); | ||
220 | } else { | ||
221 | sysctl_sched_features |= (1UL << i); | ||
222 | sched_feat_enable(i); | ||
223 | } | ||
224 | break; | ||
225 | } | ||
226 | } | ||
227 | |||
228 | if (i == __SCHED_FEAT_NR) | ||
229 | return -EINVAL; | ||
230 | |||
231 | *ppos += cnt; | ||
232 | |||
233 | return cnt; | ||
234 | } | ||
235 | |||
236 | static int sched_feat_open(struct inode *inode, struct file *filp) | ||
237 | { | ||
238 | return single_open(filp, sched_feat_show, NULL); | ||
239 | } | ||
240 | |||
241 | static const struct file_operations sched_feat_fops = { | ||
242 | .open = sched_feat_open, | ||
243 | .write = sched_feat_write, | ||
244 | .read = seq_read, | ||
245 | .llseek = seq_lseek, | ||
246 | .release = single_release, | ||
247 | }; | ||
248 | |||
249 | static __init int sched_init_debug(void) | ||
250 | { | ||
251 | debugfs_create_file("sched_features", 0644, NULL, NULL, | ||
252 | &sched_feat_fops); | ||
253 | |||
254 | return 0; | ||
255 | } | ||
256 | late_initcall(sched_init_debug); | ||
257 | #endif /* CONFIG_SCHED_DEBUG */ | ||
258 | |||
259 | /* | ||
260 | * Number of tasks to iterate in a single balance run. | ||
261 | * Limited because this is done with IRQs disabled. | ||
262 | */ | ||
263 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | ||
264 | |||
265 | /* | ||
266 | * period over which we average the RT time consumption, measured | ||
267 | * in ms. | ||
268 | * | ||
269 | * default: 1s | ||
270 | */ | ||
271 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | ||
272 | |||
273 | /* | ||
274 | * period over which we measure -rt task cpu usage in us. | ||
275 | * default: 1s | ||
276 | */ | ||
277 | unsigned int sysctl_sched_rt_period = 1000000; | ||
278 | |||
279 | __read_mostly int scheduler_running; | ||
280 | |||
281 | /* | ||
282 | * part of the period that we allow rt tasks to run in us. | ||
283 | * default: 0.95s | ||
284 | */ | ||
285 | int sysctl_sched_rt_runtime = 950000; | ||
286 | |||
287 | |||
288 | |||
289 | /* | ||
290 | * __task_rq_lock - lock the rq @p resides on. | ||
291 | */ | ||
292 | static inline struct rq *__task_rq_lock(struct task_struct *p) | ||
293 | __acquires(rq->lock) | ||
294 | { | ||
295 | struct rq *rq; | ||
296 | |||
297 | lockdep_assert_held(&p->pi_lock); | ||
298 | |||
299 | for (;;) { | ||
300 | rq = task_rq(p); | ||
301 | raw_spin_lock(&rq->lock); | ||
302 | if (likely(rq == task_rq(p))) | ||
303 | return rq; | ||
304 | raw_spin_unlock(&rq->lock); | ||
305 | } | ||
306 | } | ||
307 | |||
308 | /* | ||
309 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | ||
310 | */ | ||
311 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) | ||
312 | __acquires(p->pi_lock) | ||
313 | __acquires(rq->lock) | ||
314 | { | ||
315 | struct rq *rq; | ||
316 | |||
317 | for (;;) { | ||
318 | raw_spin_lock_irqsave(&p->pi_lock, *flags); | ||
319 | rq = task_rq(p); | ||
320 | raw_spin_lock(&rq->lock); | ||
321 | if (likely(rq == task_rq(p))) | ||
322 | return rq; | ||
323 | raw_spin_unlock(&rq->lock); | ||
324 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | ||
325 | } | ||
326 | } | ||
327 | |||
328 | static void __task_rq_unlock(struct rq *rq) | ||
329 | __releases(rq->lock) | ||
330 | { | ||
331 | raw_spin_unlock(&rq->lock); | ||
332 | } | ||
333 | |||
334 | static inline void | ||
335 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | ||
336 | __releases(rq->lock) | ||
337 | __releases(p->pi_lock) | ||
338 | { | ||
339 | raw_spin_unlock(&rq->lock); | ||
340 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | ||
341 | } | ||
342 | |||
343 | /* | ||
344 | * this_rq_lock - lock this runqueue and disable interrupts. | ||
345 | */ | ||
346 | static struct rq *this_rq_lock(void) | ||
347 | __acquires(rq->lock) | ||
348 | { | ||
349 | struct rq *rq; | ||
350 | |||
351 | local_irq_disable(); | ||
352 | rq = this_rq(); | ||
353 | raw_spin_lock(&rq->lock); | ||
354 | |||
355 | return rq; | ||
356 | } | ||
357 | |||
358 | #ifdef CONFIG_SCHED_HRTICK | ||
359 | /* | ||
360 | * Use HR-timers to deliver accurate preemption points. | ||
361 | * | ||
362 | * Its all a bit involved since we cannot program an hrt while holding the | ||
363 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | ||
364 | * reschedule event. | ||
365 | * | ||
366 | * When we get rescheduled we reprogram the hrtick_timer outside of the | ||
367 | * rq->lock. | ||
368 | */ | ||
369 | |||
370 | static void hrtick_clear(struct rq *rq) | ||
371 | { | ||
372 | if (hrtimer_active(&rq->hrtick_timer)) | ||
373 | hrtimer_cancel(&rq->hrtick_timer); | ||
374 | } | ||
375 | |||
376 | /* | ||
377 | * High-resolution timer tick. | ||
378 | * Runs from hardirq context with interrupts disabled. | ||
379 | */ | ||
380 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | ||
381 | { | ||
382 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | ||
383 | |||
384 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | ||
385 | |||
386 | raw_spin_lock(&rq->lock); | ||
387 | update_rq_clock(rq); | ||
388 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); | ||
389 | raw_spin_unlock(&rq->lock); | ||
390 | |||
391 | return HRTIMER_NORESTART; | ||
392 | } | ||
393 | |||
394 | #ifdef CONFIG_SMP | ||
395 | /* | ||
396 | * called from hardirq (IPI) context | ||
397 | */ | ||
398 | static void __hrtick_start(void *arg) | ||
399 | { | ||
400 | struct rq *rq = arg; | ||
401 | |||
402 | raw_spin_lock(&rq->lock); | ||
403 | hrtimer_restart(&rq->hrtick_timer); | ||
404 | rq->hrtick_csd_pending = 0; | ||
405 | raw_spin_unlock(&rq->lock); | ||
406 | } | ||
407 | |||
408 | /* | ||
409 | * Called to set the hrtick timer state. | ||
410 | * | ||
411 | * called with rq->lock held and irqs disabled | ||
412 | */ | ||
413 | void hrtick_start(struct rq *rq, u64 delay) | ||
414 | { | ||
415 | struct hrtimer *timer = &rq->hrtick_timer; | ||
416 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | ||
417 | |||
418 | hrtimer_set_expires(timer, time); | ||
419 | |||
420 | if (rq == this_rq()) { | ||
421 | hrtimer_restart(timer); | ||
422 | } else if (!rq->hrtick_csd_pending) { | ||
423 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); | ||
424 | rq->hrtick_csd_pending = 1; | ||
425 | } | ||
426 | } | ||
427 | |||
428 | static int | ||
429 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | ||
430 | { | ||
431 | int cpu = (int)(long)hcpu; | ||
432 | |||
433 | switch (action) { | ||
434 | case CPU_UP_CANCELED: | ||
435 | case CPU_UP_CANCELED_FROZEN: | ||
436 | case CPU_DOWN_PREPARE: | ||
437 | case CPU_DOWN_PREPARE_FROZEN: | ||
438 | case CPU_DEAD: | ||
439 | case CPU_DEAD_FROZEN: | ||
440 | hrtick_clear(cpu_rq(cpu)); | ||
441 | return NOTIFY_OK; | ||
442 | } | ||
443 | |||
444 | return NOTIFY_DONE; | ||
445 | } | ||
446 | |||
447 | static __init void init_hrtick(void) | ||
448 | { | ||
449 | hotcpu_notifier(hotplug_hrtick, 0); | ||
450 | } | ||
451 | #else | ||
452 | /* | ||
453 | * Called to set the hrtick timer state. | ||
454 | * | ||
455 | * called with rq->lock held and irqs disabled | ||
456 | */ | ||
457 | void hrtick_start(struct rq *rq, u64 delay) | ||
458 | { | ||
459 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, | ||
460 | HRTIMER_MODE_REL_PINNED, 0); | ||
461 | } | ||
462 | |||
463 | static inline void init_hrtick(void) | ||
464 | { | ||
465 | } | ||
466 | #endif /* CONFIG_SMP */ | ||
467 | |||
468 | static void init_rq_hrtick(struct rq *rq) | ||
469 | { | ||
470 | #ifdef CONFIG_SMP | ||
471 | rq->hrtick_csd_pending = 0; | ||
472 | |||
473 | rq->hrtick_csd.flags = 0; | ||
474 | rq->hrtick_csd.func = __hrtick_start; | ||
475 | rq->hrtick_csd.info = rq; | ||
476 | #endif | ||
477 | |||
478 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | ||
479 | rq->hrtick_timer.function = hrtick; | ||
480 | } | ||
481 | #else /* CONFIG_SCHED_HRTICK */ | ||
482 | static inline void hrtick_clear(struct rq *rq) | ||
483 | { | ||
484 | } | ||
485 | |||
486 | static inline void init_rq_hrtick(struct rq *rq) | ||
487 | { | ||
488 | } | ||
489 | |||
490 | static inline void init_hrtick(void) | ||
491 | { | ||
492 | } | ||
493 | #endif /* CONFIG_SCHED_HRTICK */ | ||
494 | |||
495 | /* | ||
496 | * resched_task - mark a task 'to be rescheduled now'. | ||
497 | * | ||
498 | * On UP this means the setting of the need_resched flag, on SMP it | ||
499 | * might also involve a cross-CPU call to trigger the scheduler on | ||
500 | * the target CPU. | ||
501 | */ | ||
502 | #ifdef CONFIG_SMP | ||
503 | |||
504 | #ifndef tsk_is_polling | ||
505 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | ||
506 | #endif | ||
507 | |||
508 | void resched_task(struct task_struct *p) | ||
509 | { | ||
510 | int cpu; | ||
511 | |||
512 | assert_raw_spin_locked(&task_rq(p)->lock); | ||
513 | |||
514 | if (test_tsk_need_resched(p)) | ||
515 | return; | ||
516 | |||
517 | set_tsk_need_resched(p); | ||
518 | |||
519 | cpu = task_cpu(p); | ||
520 | if (cpu == smp_processor_id()) | ||
521 | return; | ||
522 | |||
523 | /* NEED_RESCHED must be visible before we test polling */ | ||
524 | smp_mb(); | ||
525 | if (!tsk_is_polling(p)) | ||
526 | smp_send_reschedule(cpu); | ||
527 | } | ||
528 | |||
529 | void resched_cpu(int cpu) | ||
530 | { | ||
531 | struct rq *rq = cpu_rq(cpu); | ||
532 | unsigned long flags; | ||
533 | |||
534 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) | ||
535 | return; | ||
536 | resched_task(cpu_curr(cpu)); | ||
537 | raw_spin_unlock_irqrestore(&rq->lock, flags); | ||
538 | } | ||
539 | |||
540 | #ifdef CONFIG_NO_HZ | ||
541 | /* | ||
542 | * In the semi idle case, use the nearest busy cpu for migrating timers | ||
543 | * from an idle cpu. This is good for power-savings. | ||
544 | * | ||
545 | * We don't do similar optimization for completely idle system, as | ||
546 | * selecting an idle cpu will add more delays to the timers than intended | ||
547 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | ||
548 | */ | ||
549 | int get_nohz_timer_target(void) | ||
550 | { | ||
551 | int cpu = smp_processor_id(); | ||
552 | int i; | ||
553 | struct sched_domain *sd; | ||
554 | |||
555 | rcu_read_lock(); | ||
556 | for_each_domain(cpu, sd) { | ||
557 | for_each_cpu(i, sched_domain_span(sd)) { | ||
558 | if (!idle_cpu(i)) { | ||
559 | cpu = i; | ||
560 | goto unlock; | ||
561 | } | ||
562 | } | ||
563 | } | ||
564 | unlock: | ||
565 | rcu_read_unlock(); | ||
566 | return cpu; | ||
567 | } | ||
568 | /* | ||
569 | * When add_timer_on() enqueues a timer into the timer wheel of an | ||
570 | * idle CPU then this timer might expire before the next timer event | ||
571 | * which is scheduled to wake up that CPU. In case of a completely | ||
572 | * idle system the next event might even be infinite time into the | ||
573 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | ||
574 | * leaves the inner idle loop so the newly added timer is taken into | ||
575 | * account when the CPU goes back to idle and evaluates the timer | ||
576 | * wheel for the next timer event. | ||
577 | */ | ||
578 | void wake_up_idle_cpu(int cpu) | ||
579 | { | ||
580 | struct rq *rq = cpu_rq(cpu); | ||
581 | |||
582 | if (cpu == smp_processor_id()) | ||
583 | return; | ||
584 | |||
585 | /* | ||
586 | * This is safe, as this function is called with the timer | ||
587 | * wheel base lock of (cpu) held. When the CPU is on the way | ||
588 | * to idle and has not yet set rq->curr to idle then it will | ||
589 | * be serialized on the timer wheel base lock and take the new | ||
590 | * timer into account automatically. | ||
591 | */ | ||
592 | if (rq->curr != rq->idle) | ||
593 | return; | ||
594 | |||
595 | /* | ||
596 | * We can set TIF_RESCHED on the idle task of the other CPU | ||
597 | * lockless. The worst case is that the other CPU runs the | ||
598 | * idle task through an additional NOOP schedule() | ||
599 | */ | ||
600 | set_tsk_need_resched(rq->idle); | ||
601 | |||
602 | /* NEED_RESCHED must be visible before we test polling */ | ||
603 | smp_mb(); | ||
604 | if (!tsk_is_polling(rq->idle)) | ||
605 | smp_send_reschedule(cpu); | ||
606 | } | ||
607 | |||
608 | static inline bool got_nohz_idle_kick(void) | ||
609 | { | ||
610 | int cpu = smp_processor_id(); | ||
611 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | ||
612 | } | ||
613 | |||
614 | #else /* CONFIG_NO_HZ */ | ||
615 | |||
616 | static inline bool got_nohz_idle_kick(void) | ||
617 | { | ||
618 | return false; | ||
619 | } | ||
620 | |||
621 | #endif /* CONFIG_NO_HZ */ | ||
622 | |||
623 | void sched_avg_update(struct rq *rq) | ||
624 | { | ||
625 | s64 period = sched_avg_period(); | ||
626 | |||
627 | while ((s64)(rq->clock - rq->age_stamp) > period) { | ||
628 | /* | ||
629 | * Inline assembly required to prevent the compiler | ||
630 | * optimising this loop into a divmod call. | ||
631 | * See __iter_div_u64_rem() for another example of this. | ||
632 | */ | ||
633 | asm("" : "+rm" (rq->age_stamp)); | ||
634 | rq->age_stamp += period; | ||
635 | rq->rt_avg /= 2; | ||
636 | } | ||
637 | } | ||
638 | |||
639 | #else /* !CONFIG_SMP */ | ||
640 | void resched_task(struct task_struct *p) | ||
641 | { | ||
642 | assert_raw_spin_locked(&task_rq(p)->lock); | ||
643 | set_tsk_need_resched(p); | ||
644 | } | ||
645 | #endif /* CONFIG_SMP */ | ||
646 | |||
647 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ | ||
648 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | ||
649 | /* | ||
650 | * Iterate task_group tree rooted at *from, calling @down when first entering a | ||
651 | * node and @up when leaving it for the final time. | ||
652 | * | ||
653 | * Caller must hold rcu_lock or sufficient equivalent. | ||
654 | */ | ||
655 | int walk_tg_tree_from(struct task_group *from, | ||
656 | tg_visitor down, tg_visitor up, void *data) | ||
657 | { | ||
658 | struct task_group *parent, *child; | ||
659 | int ret; | ||
660 | |||
661 | parent = from; | ||
662 | |||
663 | down: | ||
664 | ret = (*down)(parent, data); | ||
665 | if (ret) | ||
666 | goto out; | ||
667 | list_for_each_entry_rcu(child, &parent->children, siblings) { | ||
668 | parent = child; | ||
669 | goto down; | ||
670 | |||
671 | up: | ||
672 | continue; | ||
673 | } | ||
674 | ret = (*up)(parent, data); | ||
675 | if (ret || parent == from) | ||
676 | goto out; | ||
677 | |||
678 | child = parent; | ||
679 | parent = parent->parent; | ||
680 | if (parent) | ||
681 | goto up; | ||
682 | out: | ||
683 | return ret; | ||
684 | } | ||
685 | |||
686 | int tg_nop(struct task_group *tg, void *data) | ||
687 | { | ||
688 | return 0; | ||
689 | } | ||
690 | #endif | ||
691 | |||
692 | void update_cpu_load(struct rq *this_rq); | ||
693 | |||
694 | static void set_load_weight(struct task_struct *p) | ||
695 | { | ||
696 | int prio = p->static_prio - MAX_RT_PRIO; | ||
697 | struct load_weight *load = &p->se.load; | ||
698 | |||
699 | /* | ||
700 | * SCHED_IDLE tasks get minimal weight: | ||
701 | */ | ||
702 | if (p->policy == SCHED_IDLE) { | ||
703 | load->weight = scale_load(WEIGHT_IDLEPRIO); | ||
704 | load->inv_weight = WMULT_IDLEPRIO; | ||
705 | return; | ||
706 | } | ||
707 | |||
708 | load->weight = scale_load(prio_to_weight[prio]); | ||
709 | load->inv_weight = prio_to_wmult[prio]; | ||
710 | } | ||
711 | |||
712 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) | ||
713 | { | ||
714 | update_rq_clock(rq); | ||
715 | sched_info_queued(p); | ||
716 | p->sched_class->enqueue_task(rq, p, flags); | ||
717 | } | ||
718 | |||
719 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) | ||
720 | { | ||
721 | update_rq_clock(rq); | ||
722 | sched_info_dequeued(p); | ||
723 | p->sched_class->dequeue_task(rq, p, flags); | ||
724 | } | ||
725 | |||
726 | /* | ||
727 | * activate_task - move a task to the runqueue. | ||
728 | */ | ||
729 | void activate_task(struct rq *rq, struct task_struct *p, int flags) | ||
730 | { | ||
731 | if (task_contributes_to_load(p)) | ||
732 | rq->nr_uninterruptible--; | ||
733 | |||
734 | enqueue_task(rq, p, flags); | ||
735 | } | ||
736 | |||
737 | /* | ||
738 | * deactivate_task - remove a task from the runqueue. | ||
739 | */ | ||
740 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) | ||
741 | { | ||
742 | if (task_contributes_to_load(p)) | ||
743 | rq->nr_uninterruptible++; | ||
744 | |||
745 | dequeue_task(rq, p, flags); | ||
746 | } | ||
747 | |||
748 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | ||
749 | |||
750 | /* | ||
751 | * There are no locks covering percpu hardirq/softirq time. | ||
752 | * They are only modified in account_system_vtime, on corresponding CPU | ||
753 | * with interrupts disabled. So, writes are safe. | ||
754 | * They are read and saved off onto struct rq in update_rq_clock(). | ||
755 | * This may result in other CPU reading this CPU's irq time and can | ||
756 | * race with irq/account_system_vtime on this CPU. We would either get old | ||
757 | * or new value with a side effect of accounting a slice of irq time to wrong | ||
758 | * task when irq is in progress while we read rq->clock. That is a worthy | ||
759 | * compromise in place of having locks on each irq in account_system_time. | ||
760 | */ | ||
761 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); | ||
762 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | ||
763 | |||
764 | static DEFINE_PER_CPU(u64, irq_start_time); | ||
765 | static int sched_clock_irqtime; | ||
766 | |||
767 | void enable_sched_clock_irqtime(void) | ||
768 | { | ||
769 | sched_clock_irqtime = 1; | ||
770 | } | ||
771 | |||
772 | void disable_sched_clock_irqtime(void) | ||
773 | { | ||
774 | sched_clock_irqtime = 0; | ||
775 | } | ||
776 | |||
777 | #ifndef CONFIG_64BIT | ||
778 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | ||
779 | |||
780 | static inline void irq_time_write_begin(void) | ||
781 | { | ||
782 | __this_cpu_inc(irq_time_seq.sequence); | ||
783 | smp_wmb(); | ||
784 | } | ||
785 | |||
786 | static inline void irq_time_write_end(void) | ||
787 | { | ||
788 | smp_wmb(); | ||
789 | __this_cpu_inc(irq_time_seq.sequence); | ||
790 | } | ||
791 | |||
792 | static inline u64 irq_time_read(int cpu) | ||
793 | { | ||
794 | u64 irq_time; | ||
795 | unsigned seq; | ||
796 | |||
797 | do { | ||
798 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | ||
799 | irq_time = per_cpu(cpu_softirq_time, cpu) + | ||
800 | per_cpu(cpu_hardirq_time, cpu); | ||
801 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | ||
802 | |||
803 | return irq_time; | ||
804 | } | ||
805 | #else /* CONFIG_64BIT */ | ||
806 | static inline void irq_time_write_begin(void) | ||
807 | { | ||
808 | } | ||
809 | |||
810 | static inline void irq_time_write_end(void) | ||
811 | { | ||
812 | } | ||
813 | |||
814 | static inline u64 irq_time_read(int cpu) | ||
815 | { | ||
816 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | ||
817 | } | ||
818 | #endif /* CONFIG_64BIT */ | ||
819 | |||
820 | /* | ||
821 | * Called before incrementing preempt_count on {soft,}irq_enter | ||
822 | * and before decrementing preempt_count on {soft,}irq_exit. | ||
823 | */ | ||
824 | void account_system_vtime(struct task_struct *curr) | ||
825 | { | ||
826 | unsigned long flags; | ||
827 | s64 delta; | ||
828 | int cpu; | ||
829 | |||
830 | if (!sched_clock_irqtime) | ||
831 | return; | ||
832 | |||
833 | local_irq_save(flags); | ||
834 | |||
835 | cpu = smp_processor_id(); | ||
836 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); | ||
837 | __this_cpu_add(irq_start_time, delta); | ||
838 | |||
839 | irq_time_write_begin(); | ||
840 | /* | ||
841 | * We do not account for softirq time from ksoftirqd here. | ||
842 | * We want to continue accounting softirq time to ksoftirqd thread | ||
843 | * in that case, so as not to confuse scheduler with a special task | ||
844 | * that do not consume any time, but still wants to run. | ||
845 | */ | ||
846 | if (hardirq_count()) | ||
847 | __this_cpu_add(cpu_hardirq_time, delta); | ||
848 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) | ||
849 | __this_cpu_add(cpu_softirq_time, delta); | ||
850 | |||
851 | irq_time_write_end(); | ||
852 | local_irq_restore(flags); | ||
853 | } | ||
854 | EXPORT_SYMBOL_GPL(account_system_vtime); | ||
855 | |||
856 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ | ||
857 | |||
858 | #ifdef CONFIG_PARAVIRT | ||
859 | static inline u64 steal_ticks(u64 steal) | ||
860 | { | ||
861 | if (unlikely(steal > NSEC_PER_SEC)) | ||
862 | return div_u64(steal, TICK_NSEC); | ||
863 | |||
864 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); | ||
865 | } | ||
866 | #endif | ||
867 | |||
868 | static void update_rq_clock_task(struct rq *rq, s64 delta) | ||
869 | { | ||
870 | /* | ||
871 | * In theory, the compile should just see 0 here, and optimize out the call | ||
872 | * to sched_rt_avg_update. But I don't trust it... | ||
873 | */ | ||
874 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | ||
875 | s64 steal = 0, irq_delta = 0; | ||
876 | #endif | ||
877 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | ||
878 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | ||
879 | |||
880 | /* | ||
881 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | ||
882 | * this case when a previous update_rq_clock() happened inside a | ||
883 | * {soft,}irq region. | ||
884 | * | ||
885 | * When this happens, we stop ->clock_task and only update the | ||
886 | * prev_irq_time stamp to account for the part that fit, so that a next | ||
887 | * update will consume the rest. This ensures ->clock_task is | ||
888 | * monotonic. | ||
889 | * | ||
890 | * It does however cause some slight miss-attribution of {soft,}irq | ||
891 | * time, a more accurate solution would be to update the irq_time using | ||
892 | * the current rq->clock timestamp, except that would require using | ||
893 | * atomic ops. | ||
894 | */ | ||
895 | if (irq_delta > delta) | ||
896 | irq_delta = delta; | ||
897 | |||
898 | rq->prev_irq_time += irq_delta; | ||
899 | delta -= irq_delta; | ||
900 | #endif | ||
901 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | ||
902 | if (static_branch((¶virt_steal_rq_enabled))) { | ||
903 | u64 st; | ||
904 | |||
905 | steal = paravirt_steal_clock(cpu_of(rq)); | ||
906 | steal -= rq->prev_steal_time_rq; | ||
907 | |||
908 | if (unlikely(steal > delta)) | ||
909 | steal = delta; | ||
910 | |||
911 | st = steal_ticks(steal); | ||
912 | steal = st * TICK_NSEC; | ||
913 | |||
914 | rq->prev_steal_time_rq += steal; | ||
915 | |||
916 | delta -= steal; | ||
917 | } | ||
918 | #endif | ||
919 | |||
920 | rq->clock_task += delta; | ||
921 | |||
922 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | ||
923 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | ||
924 | sched_rt_avg_update(rq, irq_delta + steal); | ||
925 | #endif | ||
926 | } | ||
927 | |||
928 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | ||
929 | static int irqtime_account_hi_update(void) | ||
930 | { | ||
931 | u64 *cpustat = kcpustat_this_cpu->cpustat; | ||
932 | unsigned long flags; | ||
933 | u64 latest_ns; | ||
934 | int ret = 0; | ||
935 | |||
936 | local_irq_save(flags); | ||
937 | latest_ns = this_cpu_read(cpu_hardirq_time); | ||
938 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) | ||
939 | ret = 1; | ||
940 | local_irq_restore(flags); | ||
941 | return ret; | ||
942 | } | ||
943 | |||
944 | static int irqtime_account_si_update(void) | ||
945 | { | ||
946 | u64 *cpustat = kcpustat_this_cpu->cpustat; | ||
947 | unsigned long flags; | ||
948 | u64 latest_ns; | ||
949 | int ret = 0; | ||
950 | |||
951 | local_irq_save(flags); | ||
952 | latest_ns = this_cpu_read(cpu_softirq_time); | ||
953 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) | ||
954 | ret = 1; | ||
955 | local_irq_restore(flags); | ||
956 | return ret; | ||
957 | } | ||
958 | |||
959 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | ||
960 | |||
961 | #define sched_clock_irqtime (0) | ||
962 | |||
963 | #endif | ||
964 | |||
965 | void sched_set_stop_task(int cpu, struct task_struct *stop) | ||
966 | { | ||
967 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | ||
968 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | ||
969 | |||
970 | if (stop) { | ||
971 | /* | ||
972 | * Make it appear like a SCHED_FIFO task, its something | ||
973 | * userspace knows about and won't get confused about. | ||
974 | * | ||
975 | * Also, it will make PI more or less work without too | ||
976 | * much confusion -- but then, stop work should not | ||
977 | * rely on PI working anyway. | ||
978 | */ | ||
979 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | ||
980 | |||
981 | stop->sched_class = &stop_sched_class; | ||
982 | } | ||
983 | |||
984 | cpu_rq(cpu)->stop = stop; | ||
985 | |||
986 | if (old_stop) { | ||
987 | /* | ||
988 | * Reset it back to a normal scheduling class so that | ||
989 | * it can die in pieces. | ||
990 | */ | ||
991 | old_stop->sched_class = &rt_sched_class; | ||
992 | } | ||
993 | } | ||
994 | |||
995 | /* | ||
996 | * __normal_prio - return the priority that is based on the static prio | ||
997 | */ | ||
998 | static inline int __normal_prio(struct task_struct *p) | ||
999 | { | ||
1000 | return p->static_prio; | ||
1001 | } | ||
1002 | |||
1003 | /* | ||
1004 | * Calculate the expected normal priority: i.e. priority | ||
1005 | * without taking RT-inheritance into account. Might be | ||
1006 | * boosted by interactivity modifiers. Changes upon fork, | ||
1007 | * setprio syscalls, and whenever the interactivity | ||
1008 | * estimator recalculates. | ||
1009 | */ | ||
1010 | static inline int normal_prio(struct task_struct *p) | ||
1011 | { | ||
1012 | int prio; | ||
1013 | |||
1014 | if (task_has_rt_policy(p)) | ||
1015 | prio = MAX_RT_PRIO-1 - p->rt_priority; | ||
1016 | else | ||
1017 | prio = __normal_prio(p); | ||
1018 | return prio; | ||
1019 | } | ||
1020 | |||
1021 | /* | ||
1022 | * Calculate the current priority, i.e. the priority | ||
1023 | * taken into account by the scheduler. This value might | ||
1024 | * be boosted by RT tasks, or might be boosted by | ||
1025 | * interactivity modifiers. Will be RT if the task got | ||
1026 | * RT-boosted. If not then it returns p->normal_prio. | ||
1027 | */ | ||
1028 | static int effective_prio(struct task_struct *p) | ||
1029 | { | ||
1030 | p->normal_prio = normal_prio(p); | ||
1031 | /* | ||
1032 | * If we are RT tasks or we were boosted to RT priority, | ||
1033 | * keep the priority unchanged. Otherwise, update priority | ||
1034 | * to the normal priority: | ||
1035 | */ | ||
1036 | if (!rt_prio(p->prio)) | ||
1037 | return p->normal_prio; | ||
1038 | return p->prio; | ||
1039 | } | ||
1040 | |||
1041 | /** | ||
1042 | * task_curr - is this task currently executing on a CPU? | ||
1043 | * @p: the task in question. | ||
1044 | */ | ||
1045 | inline int task_curr(const struct task_struct *p) | ||
1046 | { | ||
1047 | return cpu_curr(task_cpu(p)) == p; | ||
1048 | } | ||
1049 | |||
1050 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, | ||
1051 | const struct sched_class *prev_class, | ||
1052 | int oldprio) | ||
1053 | { | ||
1054 | if (prev_class != p->sched_class) { | ||
1055 | if (prev_class->switched_from) | ||
1056 | prev_class->switched_from(rq, p); | ||
1057 | p->sched_class->switched_to(rq, p); | ||
1058 | } else if (oldprio != p->prio) | ||
1059 | p->sched_class->prio_changed(rq, p, oldprio); | ||
1060 | } | ||
1061 | |||
1062 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | ||
1063 | { | ||
1064 | const struct sched_class *class; | ||
1065 | |||
1066 | if (p->sched_class == rq->curr->sched_class) { | ||
1067 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | ||
1068 | } else { | ||
1069 | for_each_class(class) { | ||
1070 | if (class == rq->curr->sched_class) | ||
1071 | break; | ||
1072 | if (class == p->sched_class) { | ||
1073 | resched_task(rq->curr); | ||
1074 | break; | ||
1075 | } | ||
1076 | } | ||
1077 | } | ||
1078 | |||
1079 | /* | ||
1080 | * A queue event has occurred, and we're going to schedule. In | ||
1081 | * this case, we can save a useless back to back clock update. | ||
1082 | */ | ||
1083 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) | ||
1084 | rq->skip_clock_update = 1; | ||
1085 | } | ||
1086 | |||
1087 | #ifdef CONFIG_SMP | ||
1088 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) | ||
1089 | { | ||
1090 | #ifdef CONFIG_SCHED_DEBUG | ||
1091 | /* | ||
1092 | * We should never call set_task_cpu() on a blocked task, | ||
1093 | * ttwu() will sort out the placement. | ||
1094 | */ | ||
1095 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && | ||
1096 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | ||
1097 | |||
1098 | #ifdef CONFIG_LOCKDEP | ||
1099 | /* | ||
1100 | * The caller should hold either p->pi_lock or rq->lock, when changing | ||
1101 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | ||
1102 | * | ||
1103 | * sched_move_task() holds both and thus holding either pins the cgroup, | ||
1104 | * see set_task_rq(). | ||
1105 | * | ||
1106 | * Furthermore, all task_rq users should acquire both locks, see | ||
1107 | * task_rq_lock(). | ||
1108 | */ | ||
1109 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || | ||
1110 | lockdep_is_held(&task_rq(p)->lock))); | ||
1111 | #endif | ||
1112 | #endif | ||
1113 | |||
1114 | trace_sched_migrate_task(p, new_cpu); | ||
1115 | |||
1116 | if (task_cpu(p) != new_cpu) { | ||
1117 | p->se.nr_migrations++; | ||
1118 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); | ||
1119 | } | ||
1120 | |||
1121 | __set_task_cpu(p, new_cpu); | ||
1122 | } | ||
1123 | |||
1124 | struct migration_arg { | ||
1125 | struct task_struct *task; | ||
1126 | int dest_cpu; | ||
1127 | }; | ||
1128 | |||
1129 | static int migration_cpu_stop(void *data); | ||
1130 | |||
1131 | /* | ||
1132 | * wait_task_inactive - wait for a thread to unschedule. | ||
1133 | * | ||
1134 | * If @match_state is nonzero, it's the @p->state value just checked and | ||
1135 | * not expected to change. If it changes, i.e. @p might have woken up, | ||
1136 | * then return zero. When we succeed in waiting for @p to be off its CPU, | ||
1137 | * we return a positive number (its total switch count). If a second call | ||
1138 | * a short while later returns the same number, the caller can be sure that | ||
1139 | * @p has remained unscheduled the whole time. | ||
1140 | * | ||
1141 | * The caller must ensure that the task *will* unschedule sometime soon, | ||
1142 | * else this function might spin for a *long* time. This function can't | ||
1143 | * be called with interrupts off, or it may introduce deadlock with | ||
1144 | * smp_call_function() if an IPI is sent by the same process we are | ||
1145 | * waiting to become inactive. | ||
1146 | */ | ||
1147 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) | ||
1148 | { | ||
1149 | unsigned long flags; | ||
1150 | int running, on_rq; | ||
1151 | unsigned long ncsw; | ||
1152 | struct rq *rq; | ||
1153 | |||
1154 | for (;;) { | ||
1155 | /* | ||
1156 | * We do the initial early heuristics without holding | ||
1157 | * any task-queue locks at all. We'll only try to get | ||
1158 | * the runqueue lock when things look like they will | ||
1159 | * work out! | ||
1160 | */ | ||
1161 | rq = task_rq(p); | ||
1162 | |||
1163 | /* | ||
1164 | * If the task is actively running on another CPU | ||
1165 | * still, just relax and busy-wait without holding | ||
1166 | * any locks. | ||
1167 | * | ||
1168 | * NOTE! Since we don't hold any locks, it's not | ||
1169 | * even sure that "rq" stays as the right runqueue! | ||
1170 | * But we don't care, since "task_running()" will | ||
1171 | * return false if the runqueue has changed and p | ||
1172 | * is actually now running somewhere else! | ||
1173 | */ | ||
1174 | while (task_running(rq, p)) { | ||
1175 | if (match_state && unlikely(p->state != match_state)) | ||
1176 | return 0; | ||
1177 | cpu_relax(); | ||
1178 | } | ||
1179 | |||
1180 | /* | ||
1181 | * Ok, time to look more closely! We need the rq | ||
1182 | * lock now, to be *sure*. If we're wrong, we'll | ||
1183 | * just go back and repeat. | ||
1184 | */ | ||
1185 | rq = task_rq_lock(p, &flags); | ||
1186 | trace_sched_wait_task(p); | ||
1187 | running = task_running(rq, p); | ||
1188 | on_rq = p->on_rq; | ||
1189 | ncsw = 0; | ||
1190 | if (!match_state || p->state == match_state) | ||
1191 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ | ||
1192 | task_rq_unlock(rq, p, &flags); | ||
1193 | |||
1194 | /* | ||
1195 | * If it changed from the expected state, bail out now. | ||
1196 | */ | ||
1197 | if (unlikely(!ncsw)) | ||
1198 | break; | ||
1199 | |||
1200 | /* | ||
1201 | * Was it really running after all now that we | ||
1202 | * checked with the proper locks actually held? | ||
1203 | * | ||
1204 | * Oops. Go back and try again.. | ||
1205 | */ | ||
1206 | if (unlikely(running)) { | ||
1207 | cpu_relax(); | ||
1208 | continue; | ||
1209 | } | ||
1210 | |||
1211 | /* | ||
1212 | * It's not enough that it's not actively running, | ||
1213 | * it must be off the runqueue _entirely_, and not | ||
1214 | * preempted! | ||
1215 | * | ||
1216 | * So if it was still runnable (but just not actively | ||
1217 | * running right now), it's preempted, and we should | ||
1218 | * yield - it could be a while. | ||
1219 | */ | ||
1220 | if (unlikely(on_rq)) { | ||
1221 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); | ||
1222 | |||
1223 | set_current_state(TASK_UNINTERRUPTIBLE); | ||
1224 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | ||
1225 | continue; | ||
1226 | } | ||
1227 | |||
1228 | /* | ||
1229 | * Ahh, all good. It wasn't running, and it wasn't | ||
1230 | * runnable, which means that it will never become | ||
1231 | * running in the future either. We're all done! | ||
1232 | */ | ||
1233 | break; | ||
1234 | } | ||
1235 | |||
1236 | return ncsw; | ||
1237 | } | ||
1238 | |||
1239 | /*** | ||
1240 | * kick_process - kick a running thread to enter/exit the kernel | ||
1241 | * @p: the to-be-kicked thread | ||
1242 | * | ||
1243 | * Cause a process which is running on another CPU to enter | ||
1244 | * kernel-mode, without any delay. (to get signals handled.) | ||
1245 | * | ||
1246 | * NOTE: this function doesn't have to take the runqueue lock, | ||
1247 | * because all it wants to ensure is that the remote task enters | ||
1248 | * the kernel. If the IPI races and the task has been migrated | ||
1249 | * to another CPU then no harm is done and the purpose has been | ||
1250 | * achieved as well. | ||
1251 | */ | ||
1252 | void kick_process(struct task_struct *p) | ||
1253 | { | ||
1254 | int cpu; | ||
1255 | |||
1256 | preempt_disable(); | ||
1257 | cpu = task_cpu(p); | ||
1258 | if ((cpu != smp_processor_id()) && task_curr(p)) | ||
1259 | smp_send_reschedule(cpu); | ||
1260 | preempt_enable(); | ||
1261 | } | ||
1262 | EXPORT_SYMBOL_GPL(kick_process); | ||
1263 | #endif /* CONFIG_SMP */ | ||
1264 | |||
1265 | #ifdef CONFIG_SMP | ||
1266 | /* | ||
1267 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock | ||
1268 | */ | ||
1269 | static int select_fallback_rq(int cpu, struct task_struct *p) | ||
1270 | { | ||
1271 | int dest_cpu; | ||
1272 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | ||
1273 | |||
1274 | /* Look for allowed, online CPU in same node. */ | ||
1275 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | ||
1276 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | ||
1277 | return dest_cpu; | ||
1278 | |||
1279 | /* Any allowed, online CPU? */ | ||
1280 | dest_cpu = cpumask_any_and(tsk_cpus_allowed(p), cpu_active_mask); | ||
1281 | if (dest_cpu < nr_cpu_ids) | ||
1282 | return dest_cpu; | ||
1283 | |||
1284 | /* No more Mr. Nice Guy. */ | ||
1285 | dest_cpu = cpuset_cpus_allowed_fallback(p); | ||
1286 | /* | ||
1287 | * Don't tell them about moving exiting tasks or | ||
1288 | * kernel threads (both mm NULL), since they never | ||
1289 | * leave kernel. | ||
1290 | */ | ||
1291 | if (p->mm && printk_ratelimit()) { | ||
1292 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | ||
1293 | task_pid_nr(p), p->comm, cpu); | ||
1294 | } | ||
1295 | |||
1296 | return dest_cpu; | ||
1297 | } | ||
1298 | |||
1299 | /* | ||
1300 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. | ||
1301 | */ | ||
1302 | static inline | ||
1303 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | ||
1304 | { | ||
1305 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); | ||
1306 | |||
1307 | /* | ||
1308 | * In order not to call set_task_cpu() on a blocking task we need | ||
1309 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | ||
1310 | * cpu. | ||
1311 | * | ||
1312 | * Since this is common to all placement strategies, this lives here. | ||
1313 | * | ||
1314 | * [ this allows ->select_task() to simply return task_cpu(p) and | ||
1315 | * not worry about this generic constraint ] | ||
1316 | */ | ||
1317 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || | ||
1318 | !cpu_online(cpu))) | ||
1319 | cpu = select_fallback_rq(task_cpu(p), p); | ||
1320 | |||
1321 | return cpu; | ||
1322 | } | ||
1323 | |||
1324 | static void update_avg(u64 *avg, u64 sample) | ||
1325 | { | ||
1326 | s64 diff = sample - *avg; | ||
1327 | *avg += diff >> 3; | ||
1328 | } | ||
1329 | #endif | ||
1330 | |||
1331 | static void | ||
1332 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) | ||
1333 | { | ||
1334 | #ifdef CONFIG_SCHEDSTATS | ||
1335 | struct rq *rq = this_rq(); | ||
1336 | |||
1337 | #ifdef CONFIG_SMP | ||
1338 | int this_cpu = smp_processor_id(); | ||
1339 | |||
1340 | if (cpu == this_cpu) { | ||
1341 | schedstat_inc(rq, ttwu_local); | ||
1342 | schedstat_inc(p, se.statistics.nr_wakeups_local); | ||
1343 | } else { | ||
1344 | struct sched_domain *sd; | ||
1345 | |||
1346 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | ||
1347 | rcu_read_lock(); | ||
1348 | for_each_domain(this_cpu, sd) { | ||
1349 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | ||
1350 | schedstat_inc(sd, ttwu_wake_remote); | ||
1351 | break; | ||
1352 | } | ||
1353 | } | ||
1354 | rcu_read_unlock(); | ||
1355 | } | ||
1356 | |||
1357 | if (wake_flags & WF_MIGRATED) | ||
1358 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | ||
1359 | |||
1360 | #endif /* CONFIG_SMP */ | ||
1361 | |||
1362 | schedstat_inc(rq, ttwu_count); | ||
1363 | schedstat_inc(p, se.statistics.nr_wakeups); | ||
1364 | |||
1365 | if (wake_flags & WF_SYNC) | ||
1366 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | ||
1367 | |||
1368 | #endif /* CONFIG_SCHEDSTATS */ | ||
1369 | } | ||
1370 | |||
1371 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | ||
1372 | { | ||
1373 | activate_task(rq, p, en_flags); | ||
1374 | p->on_rq = 1; | ||
1375 | |||
1376 | /* if a worker is waking up, notify workqueue */ | ||
1377 | if (p->flags & PF_WQ_WORKER) | ||
1378 | wq_worker_waking_up(p, cpu_of(rq)); | ||
1379 | } | ||
1380 | |||
1381 | /* | ||
1382 | * Mark the task runnable and perform wakeup-preemption. | ||
1383 | */ | ||
1384 | static void | ||
1385 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) | ||
1386 | { | ||
1387 | trace_sched_wakeup(p, true); | ||
1388 | check_preempt_curr(rq, p, wake_flags); | ||
1389 | |||
1390 | p->state = TASK_RUNNING; | ||
1391 | #ifdef CONFIG_SMP | ||
1392 | if (p->sched_class->task_woken) | ||
1393 | p->sched_class->task_woken(rq, p); | ||
1394 | |||
1395 | if (rq->idle_stamp) { | ||
1396 | u64 delta = rq->clock - rq->idle_stamp; | ||
1397 | u64 max = 2*sysctl_sched_migration_cost; | ||
1398 | |||
1399 | if (delta > max) | ||
1400 | rq->avg_idle = max; | ||
1401 | else | ||
1402 | update_avg(&rq->avg_idle, delta); | ||
1403 | rq->idle_stamp = 0; | ||
1404 | } | ||
1405 | #endif | ||
1406 | } | ||
1407 | |||
1408 | static void | ||
1409 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | ||
1410 | { | ||
1411 | #ifdef CONFIG_SMP | ||
1412 | if (p->sched_contributes_to_load) | ||
1413 | rq->nr_uninterruptible--; | ||
1414 | #endif | ||
1415 | |||
1416 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | ||
1417 | ttwu_do_wakeup(rq, p, wake_flags); | ||
1418 | } | ||
1419 | |||
1420 | /* | ||
1421 | * Called in case the task @p isn't fully descheduled from its runqueue, | ||
1422 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | ||
1423 | * since all we need to do is flip p->state to TASK_RUNNING, since | ||
1424 | * the task is still ->on_rq. | ||
1425 | */ | ||
1426 | static int ttwu_remote(struct task_struct *p, int wake_flags) | ||
1427 | { | ||
1428 | struct rq *rq; | ||
1429 | int ret = 0; | ||
1430 | |||
1431 | rq = __task_rq_lock(p); | ||
1432 | if (p->on_rq) { | ||
1433 | ttwu_do_wakeup(rq, p, wake_flags); | ||
1434 | ret = 1; | ||
1435 | } | ||
1436 | __task_rq_unlock(rq); | ||
1437 | |||
1438 | return ret; | ||
1439 | } | ||
1440 | |||
1441 | #ifdef CONFIG_SMP | ||
1442 | static void sched_ttwu_pending(void) | ||
1443 | { | ||
1444 | struct rq *rq = this_rq(); | ||
1445 | struct llist_node *llist = llist_del_all(&rq->wake_list); | ||
1446 | struct task_struct *p; | ||
1447 | |||
1448 | raw_spin_lock(&rq->lock); | ||
1449 | |||
1450 | while (llist) { | ||
1451 | p = llist_entry(llist, struct task_struct, wake_entry); | ||
1452 | llist = llist_next(llist); | ||
1453 | ttwu_do_activate(rq, p, 0); | ||
1454 | } | ||
1455 | |||
1456 | raw_spin_unlock(&rq->lock); | ||
1457 | } | ||
1458 | |||
1459 | void scheduler_ipi(void) | ||
1460 | { | ||
1461 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) | ||
1462 | return; | ||
1463 | |||
1464 | /* | ||
1465 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | ||
1466 | * traditionally all their work was done from the interrupt return | ||
1467 | * path. Now that we actually do some work, we need to make sure | ||
1468 | * we do call them. | ||
1469 | * | ||
1470 | * Some archs already do call them, luckily irq_enter/exit nest | ||
1471 | * properly. | ||
1472 | * | ||
1473 | * Arguably we should visit all archs and update all handlers, | ||
1474 | * however a fair share of IPIs are still resched only so this would | ||
1475 | * somewhat pessimize the simple resched case. | ||
1476 | */ | ||
1477 | irq_enter(); | ||
1478 | sched_ttwu_pending(); | ||
1479 | |||
1480 | /* | ||
1481 | * Check if someone kicked us for doing the nohz idle load balance. | ||
1482 | */ | ||
1483 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { | ||
1484 | this_rq()->idle_balance = 1; | ||
1485 | raise_softirq_irqoff(SCHED_SOFTIRQ); | ||
1486 | } | ||
1487 | irq_exit(); | ||
1488 | } | ||
1489 | |||
1490 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | ||
1491 | { | ||
1492 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) | ||
1493 | smp_send_reschedule(cpu); | ||
1494 | } | ||
1495 | |||
1496 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | ||
1497 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | ||
1498 | { | ||
1499 | struct rq *rq; | ||
1500 | int ret = 0; | ||
1501 | |||
1502 | rq = __task_rq_lock(p); | ||
1503 | if (p->on_cpu) { | ||
1504 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | ||
1505 | ttwu_do_wakeup(rq, p, wake_flags); | ||
1506 | ret = 1; | ||
1507 | } | ||
1508 | __task_rq_unlock(rq); | ||
1509 | |||
1510 | return ret; | ||
1511 | |||
1512 | } | ||
1513 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | ||
1514 | #endif /* CONFIG_SMP */ | ||
1515 | |||
1516 | static void ttwu_queue(struct task_struct *p, int cpu) | ||
1517 | { | ||
1518 | struct rq *rq = cpu_rq(cpu); | ||
1519 | |||
1520 | #if defined(CONFIG_SMP) | ||
1521 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { | ||
1522 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ | ||
1523 | ttwu_queue_remote(p, cpu); | ||
1524 | return; | ||
1525 | } | ||
1526 | #endif | ||
1527 | |||
1528 | raw_spin_lock(&rq->lock); | ||
1529 | ttwu_do_activate(rq, p, 0); | ||
1530 | raw_spin_unlock(&rq->lock); | ||
1531 | } | ||
1532 | |||
1533 | /** | ||
1534 | * try_to_wake_up - wake up a thread | ||
1535 | * @p: the thread to be awakened | ||
1536 | * @state: the mask of task states that can be woken | ||
1537 | * @wake_flags: wake modifier flags (WF_*) | ||
1538 | * | ||
1539 | * Put it on the run-queue if it's not already there. The "current" | ||
1540 | * thread is always on the run-queue (except when the actual | ||
1541 | * re-schedule is in progress), and as such you're allowed to do | ||
1542 | * the simpler "current->state = TASK_RUNNING" to mark yourself | ||
1543 | * runnable without the overhead of this. | ||
1544 | * | ||
1545 | * Returns %true if @p was woken up, %false if it was already running | ||
1546 | * or @state didn't match @p's state. | ||
1547 | */ | ||
1548 | static int | ||
1549 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | ||
1550 | { | ||
1551 | unsigned long flags; | ||
1552 | int cpu, success = 0; | ||
1553 | |||
1554 | smp_wmb(); | ||
1555 | raw_spin_lock_irqsave(&p->pi_lock, flags); | ||
1556 | if (!(p->state & state)) | ||
1557 | goto out; | ||
1558 | |||
1559 | success = 1; /* we're going to change ->state */ | ||
1560 | cpu = task_cpu(p); | ||
1561 | |||
1562 | if (p->on_rq && ttwu_remote(p, wake_flags)) | ||
1563 | goto stat; | ||
1564 | |||
1565 | #ifdef CONFIG_SMP | ||
1566 | /* | ||
1567 | * If the owning (remote) cpu is still in the middle of schedule() with | ||
1568 | * this task as prev, wait until its done referencing the task. | ||
1569 | */ | ||
1570 | while (p->on_cpu) { | ||
1571 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | ||
1572 | /* | ||
1573 | * In case the architecture enables interrupts in | ||
1574 | * context_switch(), we cannot busy wait, since that | ||
1575 | * would lead to deadlocks when an interrupt hits and | ||
1576 | * tries to wake up @prev. So bail and do a complete | ||
1577 | * remote wakeup. | ||
1578 | */ | ||
1579 | if (ttwu_activate_remote(p, wake_flags)) | ||
1580 | goto stat; | ||
1581 | #else | ||
1582 | cpu_relax(); | ||
1583 | #endif | ||
1584 | } | ||
1585 | /* | ||
1586 | * Pairs with the smp_wmb() in finish_lock_switch(). | ||
1587 | */ | ||
1588 | smp_rmb(); | ||
1589 | |||
1590 | p->sched_contributes_to_load = !!task_contributes_to_load(p); | ||
1591 | p->state = TASK_WAKING; | ||
1592 | |||
1593 | if (p->sched_class->task_waking) | ||
1594 | p->sched_class->task_waking(p); | ||
1595 | |||
1596 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); | ||
1597 | if (task_cpu(p) != cpu) { | ||
1598 | wake_flags |= WF_MIGRATED; | ||
1599 | set_task_cpu(p, cpu); | ||
1600 | } | ||
1601 | #endif /* CONFIG_SMP */ | ||
1602 | |||
1603 | ttwu_queue(p, cpu); | ||
1604 | stat: | ||
1605 | ttwu_stat(p, cpu, wake_flags); | ||
1606 | out: | ||
1607 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | ||
1608 | |||
1609 | return success; | ||
1610 | } | ||
1611 | |||
1612 | /** | ||
1613 | * try_to_wake_up_local - try to wake up a local task with rq lock held | ||
1614 | * @p: the thread to be awakened | ||
1615 | * | ||
1616 | * Put @p on the run-queue if it's not already there. The caller must | ||
1617 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | ||
1618 | * the current task. | ||
1619 | */ | ||
1620 | static void try_to_wake_up_local(struct task_struct *p) | ||
1621 | { | ||
1622 | struct rq *rq = task_rq(p); | ||
1623 | |||
1624 | BUG_ON(rq != this_rq()); | ||
1625 | BUG_ON(p == current); | ||
1626 | lockdep_assert_held(&rq->lock); | ||
1627 | |||
1628 | if (!raw_spin_trylock(&p->pi_lock)) { | ||
1629 | raw_spin_unlock(&rq->lock); | ||
1630 | raw_spin_lock(&p->pi_lock); | ||
1631 | raw_spin_lock(&rq->lock); | ||
1632 | } | ||
1633 | |||
1634 | if (!(p->state & TASK_NORMAL)) | ||
1635 | goto out; | ||
1636 | |||
1637 | if (!p->on_rq) | ||
1638 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | ||
1639 | |||
1640 | ttwu_do_wakeup(rq, p, 0); | ||
1641 | ttwu_stat(p, smp_processor_id(), 0); | ||
1642 | out: | ||
1643 | raw_spin_unlock(&p->pi_lock); | ||
1644 | } | ||
1645 | |||
1646 | /** | ||
1647 | * wake_up_process - Wake up a specific process | ||
1648 | * @p: The process to be woken up. | ||
1649 | * | ||
1650 | * Attempt to wake up the nominated process and move it to the set of runnable | ||
1651 | * processes. Returns 1 if the process was woken up, 0 if it was already | ||
1652 | * running. | ||
1653 | * | ||
1654 | * It may be assumed that this function implies a write memory barrier before | ||
1655 | * changing the task state if and only if any tasks are woken up. | ||
1656 | */ | ||
1657 | int wake_up_process(struct task_struct *p) | ||
1658 | { | ||
1659 | return try_to_wake_up(p, TASK_ALL, 0); | ||
1660 | } | ||
1661 | EXPORT_SYMBOL(wake_up_process); | ||
1662 | |||
1663 | int wake_up_state(struct task_struct *p, unsigned int state) | ||
1664 | { | ||
1665 | return try_to_wake_up(p, state, 0); | ||
1666 | } | ||
1667 | |||
1668 | /* | ||
1669 | * Perform scheduler related setup for a newly forked process p. | ||
1670 | * p is forked by current. | ||
1671 | * | ||
1672 | * __sched_fork() is basic setup used by init_idle() too: | ||
1673 | */ | ||
1674 | static void __sched_fork(struct task_struct *p) | ||
1675 | { | ||
1676 | p->on_rq = 0; | ||
1677 | |||
1678 | p->se.on_rq = 0; | ||
1679 | p->se.exec_start = 0; | ||
1680 | p->se.sum_exec_runtime = 0; | ||
1681 | p->se.prev_sum_exec_runtime = 0; | ||
1682 | p->se.nr_migrations = 0; | ||
1683 | p->se.vruntime = 0; | ||
1684 | INIT_LIST_HEAD(&p->se.group_node); | ||
1685 | |||
1686 | #ifdef CONFIG_SCHEDSTATS | ||
1687 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); | ||
1688 | #endif | ||
1689 | |||
1690 | INIT_LIST_HEAD(&p->rt.run_list); | ||
1691 | |||
1692 | #ifdef CONFIG_PREEMPT_NOTIFIERS | ||
1693 | INIT_HLIST_HEAD(&p->preempt_notifiers); | ||
1694 | #endif | ||
1695 | } | ||
1696 | |||
1697 | /* | ||
1698 | * fork()/clone()-time setup: | ||
1699 | */ | ||
1700 | void sched_fork(struct task_struct *p) | ||
1701 | { | ||
1702 | unsigned long flags; | ||
1703 | int cpu = get_cpu(); | ||
1704 | |||
1705 | __sched_fork(p); | ||
1706 | /* | ||
1707 | * We mark the process as running here. This guarantees that | ||
1708 | * nobody will actually run it, and a signal or other external | ||
1709 | * event cannot wake it up and insert it on the runqueue either. | ||
1710 | */ | ||
1711 | p->state = TASK_RUNNING; | ||
1712 | |||
1713 | /* | ||
1714 | * Make sure we do not leak PI boosting priority to the child. | ||
1715 | */ | ||
1716 | p->prio = current->normal_prio; | ||
1717 | |||
1718 | /* | ||
1719 | * Revert to default priority/policy on fork if requested. | ||
1720 | */ | ||
1721 | if (unlikely(p->sched_reset_on_fork)) { | ||
1722 | if (task_has_rt_policy(p)) { | ||
1723 | p->policy = SCHED_NORMAL; | ||
1724 | p->static_prio = NICE_TO_PRIO(0); | ||
1725 | p->rt_priority = 0; | ||
1726 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | ||
1727 | p->static_prio = NICE_TO_PRIO(0); | ||
1728 | |||
1729 | p->prio = p->normal_prio = __normal_prio(p); | ||
1730 | set_load_weight(p); | ||
1731 | |||
1732 | /* | ||
1733 | * We don't need the reset flag anymore after the fork. It has | ||
1734 | * fulfilled its duty: | ||
1735 | */ | ||
1736 | p->sched_reset_on_fork = 0; | ||
1737 | } | ||
1738 | |||
1739 | if (!rt_prio(p->prio)) | ||
1740 | p->sched_class = &fair_sched_class; | ||
1741 | |||
1742 | if (p->sched_class->task_fork) | ||
1743 | p->sched_class->task_fork(p); | ||
1744 | |||
1745 | /* | ||
1746 | * The child is not yet in the pid-hash so no cgroup attach races, | ||
1747 | * and the cgroup is pinned to this child due to cgroup_fork() | ||
1748 | * is ran before sched_fork(). | ||
1749 | * | ||
1750 | * Silence PROVE_RCU. | ||
1751 | */ | ||
1752 | raw_spin_lock_irqsave(&p->pi_lock, flags); | ||
1753 | set_task_cpu(p, cpu); | ||
1754 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | ||
1755 | |||
1756 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) | ||
1757 | if (likely(sched_info_on())) | ||
1758 | memset(&p->sched_info, 0, sizeof(p->sched_info)); | ||
1759 | #endif | ||
1760 | #if defined(CONFIG_SMP) | ||
1761 | p->on_cpu = 0; | ||
1762 | #endif | ||
1763 | #ifdef CONFIG_PREEMPT_COUNT | ||
1764 | /* Want to start with kernel preemption disabled. */ | ||
1765 | task_thread_info(p)->preempt_count = 1; | ||
1766 | #endif | ||
1767 | #ifdef CONFIG_SMP | ||
1768 | plist_node_init(&p->pushable_tasks, MAX_PRIO); | ||
1769 | #endif | ||
1770 | |||
1771 | put_cpu(); | ||
1772 | } | ||
1773 | |||
1774 | /* | ||
1775 | * wake_up_new_task - wake up a newly created task for the first time. | ||
1776 | * | ||
1777 | * This function will do some initial scheduler statistics housekeeping | ||
1778 | * that must be done for every newly created context, then puts the task | ||
1779 | * on the runqueue and wakes it. | ||
1780 | */ | ||
1781 | void wake_up_new_task(struct task_struct *p) | ||
1782 | { | ||
1783 | unsigned long flags; | ||
1784 | struct rq *rq; | ||
1785 | |||
1786 | raw_spin_lock_irqsave(&p->pi_lock, flags); | ||
1787 | #ifdef CONFIG_SMP | ||
1788 | /* | ||
1789 | * Fork balancing, do it here and not earlier because: | ||
1790 | * - cpus_allowed can change in the fork path | ||
1791 | * - any previously selected cpu might disappear through hotplug | ||
1792 | */ | ||
1793 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); | ||
1794 | #endif | ||
1795 | |||
1796 | rq = __task_rq_lock(p); | ||
1797 | activate_task(rq, p, 0); | ||
1798 | p->on_rq = 1; | ||
1799 | trace_sched_wakeup_new(p, true); | ||
1800 | check_preempt_curr(rq, p, WF_FORK); | ||
1801 | #ifdef CONFIG_SMP | ||
1802 | if (p->sched_class->task_woken) | ||
1803 | p->sched_class->task_woken(rq, p); | ||
1804 | #endif | ||
1805 | task_rq_unlock(rq, p, &flags); | ||
1806 | } | ||
1807 | |||
1808 | #ifdef CONFIG_PREEMPT_NOTIFIERS | ||
1809 | |||
1810 | /** | ||
1811 | * preempt_notifier_register - tell me when current is being preempted & rescheduled | ||
1812 | * @notifier: notifier struct to register | ||
1813 | */ | ||
1814 | void preempt_notifier_register(struct preempt_notifier *notifier) | ||
1815 | { | ||
1816 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | ||
1817 | } | ||
1818 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | ||
1819 | |||
1820 | /** | ||
1821 | * preempt_notifier_unregister - no longer interested in preemption notifications | ||
1822 | * @notifier: notifier struct to unregister | ||
1823 | * | ||
1824 | * This is safe to call from within a preemption notifier. | ||
1825 | */ | ||
1826 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | ||
1827 | { | ||
1828 | hlist_del(¬ifier->link); | ||
1829 | } | ||
1830 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | ||
1831 | |||
1832 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | ||
1833 | { | ||
1834 | struct preempt_notifier *notifier; | ||
1835 | struct hlist_node *node; | ||
1836 | |||
1837 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | ||
1838 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | ||
1839 | } | ||
1840 | |||
1841 | static void | ||
1842 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | ||
1843 | struct task_struct *next) | ||
1844 | { | ||
1845 | struct preempt_notifier *notifier; | ||
1846 | struct hlist_node *node; | ||
1847 | |||
1848 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | ||
1849 | notifier->ops->sched_out(notifier, next); | ||
1850 | } | ||
1851 | |||
1852 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ | ||
1853 | |||
1854 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | ||
1855 | { | ||
1856 | } | ||
1857 | |||
1858 | static void | ||
1859 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | ||
1860 | struct task_struct *next) | ||
1861 | { | ||
1862 | } | ||
1863 | |||
1864 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ | ||
1865 | |||
1866 | /** | ||
1867 | * prepare_task_switch - prepare to switch tasks | ||
1868 | * @rq: the runqueue preparing to switch | ||
1869 | * @prev: the current task that is being switched out | ||
1870 | * @next: the task we are going to switch to. | ||
1871 | * | ||
1872 | * This is called with the rq lock held and interrupts off. It must | ||
1873 | * be paired with a subsequent finish_task_switch after the context | ||
1874 | * switch. | ||
1875 | * | ||
1876 | * prepare_task_switch sets up locking and calls architecture specific | ||
1877 | * hooks. | ||
1878 | */ | ||
1879 | static inline void | ||
1880 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | ||
1881 | struct task_struct *next) | ||
1882 | { | ||
1883 | sched_info_switch(prev, next); | ||
1884 | perf_event_task_sched_out(prev, next); | ||
1885 | fire_sched_out_preempt_notifiers(prev, next); | ||
1886 | prepare_lock_switch(rq, next); | ||
1887 | prepare_arch_switch(next); | ||
1888 | trace_sched_switch(prev, next); | ||
1889 | } | ||
1890 | |||
1891 | /** | ||
1892 | * finish_task_switch - clean up after a task-switch | ||
1893 | * @rq: runqueue associated with task-switch | ||
1894 | * @prev: the thread we just switched away from. | ||
1895 | * | ||
1896 | * finish_task_switch must be called after the context switch, paired | ||
1897 | * with a prepare_task_switch call before the context switch. | ||
1898 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | ||
1899 | * and do any other architecture-specific cleanup actions. | ||
1900 | * | ||
1901 | * Note that we may have delayed dropping an mm in context_switch(). If | ||
1902 | * so, we finish that here outside of the runqueue lock. (Doing it | ||
1903 | * with the lock held can cause deadlocks; see schedule() for | ||
1904 | * details.) | ||
1905 | */ | ||
1906 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) | ||
1907 | __releases(rq->lock) | ||
1908 | { | ||
1909 | struct mm_struct *mm = rq->prev_mm; | ||
1910 | long prev_state; | ||
1911 | |||
1912 | rq->prev_mm = NULL; | ||
1913 | |||
1914 | /* | ||
1915 | * A task struct has one reference for the use as "current". | ||
1916 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls | ||
1917 | * schedule one last time. The schedule call will never return, and | ||
1918 | * the scheduled task must drop that reference. | ||
1919 | * The test for TASK_DEAD must occur while the runqueue locks are | ||
1920 | * still held, otherwise prev could be scheduled on another cpu, die | ||
1921 | * there before we look at prev->state, and then the reference would | ||
1922 | * be dropped twice. | ||
1923 | * Manfred Spraul <manfred@colorfullife.com> | ||
1924 | */ | ||
1925 | prev_state = prev->state; | ||
1926 | finish_arch_switch(prev); | ||
1927 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | ||
1928 | local_irq_disable(); | ||
1929 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | ||
1930 | perf_event_task_sched_in(prev, current); | ||
1931 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | ||
1932 | local_irq_enable(); | ||
1933 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | ||
1934 | finish_lock_switch(rq, prev); | ||
1935 | |||
1936 | fire_sched_in_preempt_notifiers(current); | ||
1937 | if (mm) | ||
1938 | mmdrop(mm); | ||
1939 | if (unlikely(prev_state == TASK_DEAD)) { | ||
1940 | /* | ||
1941 | * Remove function-return probe instances associated with this | ||
1942 | * task and put them back on the free list. | ||
1943 | */ | ||
1944 | kprobe_flush_task(prev); | ||
1945 | put_task_struct(prev); | ||
1946 | } | ||
1947 | } | ||
1948 | |||
1949 | #ifdef CONFIG_SMP | ||
1950 | |||
1951 | /* assumes rq->lock is held */ | ||
1952 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | ||
1953 | { | ||
1954 | if (prev->sched_class->pre_schedule) | ||
1955 | prev->sched_class->pre_schedule(rq, prev); | ||
1956 | } | ||
1957 | |||
1958 | /* rq->lock is NOT held, but preemption is disabled */ | ||
1959 | static inline void post_schedule(struct rq *rq) | ||
1960 | { | ||
1961 | if (rq->post_schedule) { | ||
1962 | unsigned long flags; | ||
1963 | |||
1964 | raw_spin_lock_irqsave(&rq->lock, flags); | ||
1965 | if (rq->curr->sched_class->post_schedule) | ||
1966 | rq->curr->sched_class->post_schedule(rq); | ||
1967 | raw_spin_unlock_irqrestore(&rq->lock, flags); | ||
1968 | |||
1969 | rq->post_schedule = 0; | ||
1970 | } | ||
1971 | } | ||
1972 | |||
1973 | #else | ||
1974 | |||
1975 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) | ||
1976 | { | ||
1977 | } | ||
1978 | |||
1979 | static inline void post_schedule(struct rq *rq) | ||
1980 | { | ||
1981 | } | ||
1982 | |||
1983 | #endif | ||
1984 | |||
1985 | /** | ||
1986 | * schedule_tail - first thing a freshly forked thread must call. | ||
1987 | * @prev: the thread we just switched away from. | ||
1988 | */ | ||
1989 | asmlinkage void schedule_tail(struct task_struct *prev) | ||
1990 | __releases(rq->lock) | ||
1991 | { | ||
1992 | struct rq *rq = this_rq(); | ||
1993 | |||
1994 | finish_task_switch(rq, prev); | ||
1995 | |||
1996 | /* | ||
1997 | * FIXME: do we need to worry about rq being invalidated by the | ||
1998 | * task_switch? | ||
1999 | */ | ||
2000 | post_schedule(rq); | ||
2001 | |||
2002 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | ||
2003 | /* In this case, finish_task_switch does not reenable preemption */ | ||
2004 | preempt_enable(); | ||
2005 | #endif | ||
2006 | if (current->set_child_tid) | ||
2007 | put_user(task_pid_vnr(current), current->set_child_tid); | ||
2008 | } | ||
2009 | |||
2010 | /* | ||
2011 | * context_switch - switch to the new MM and the new | ||
2012 | * thread's register state. | ||
2013 | */ | ||
2014 | static inline void | ||
2015 | context_switch(struct rq *rq, struct task_struct *prev, | ||
2016 | struct task_struct *next) | ||
2017 | { | ||
2018 | struct mm_struct *mm, *oldmm; | ||
2019 | |||
2020 | prepare_task_switch(rq, prev, next); | ||
2021 | |||
2022 | mm = next->mm; | ||
2023 | oldmm = prev->active_mm; | ||
2024 | /* | ||
2025 | * For paravirt, this is coupled with an exit in switch_to to | ||
2026 | * combine the page table reload and the switch backend into | ||
2027 | * one hypercall. | ||
2028 | */ | ||
2029 | arch_start_context_switch(prev); | ||
2030 | |||
2031 | if (!mm) { | ||
2032 | next->active_mm = oldmm; | ||
2033 | atomic_inc(&oldmm->mm_count); | ||
2034 | enter_lazy_tlb(oldmm, next); | ||
2035 | } else | ||
2036 | switch_mm(oldmm, mm, next); | ||
2037 | |||
2038 | if (!prev->mm) { | ||
2039 | prev->active_mm = NULL; | ||
2040 | rq->prev_mm = oldmm; | ||
2041 | } | ||
2042 | /* | ||
2043 | * Since the runqueue lock will be released by the next | ||
2044 | * task (which is an invalid locking op but in the case | ||
2045 | * of the scheduler it's an obvious special-case), so we | ||
2046 | * do an early lockdep release here: | ||
2047 | */ | ||
2048 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | ||
2049 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); | ||
2050 | #endif | ||
2051 | |||
2052 | /* Here we just switch the register state and the stack. */ | ||
2053 | switch_to(prev, next, prev); | ||
2054 | |||
2055 | barrier(); | ||
2056 | /* | ||
2057 | * this_rq must be evaluated again because prev may have moved | ||
2058 | * CPUs since it called schedule(), thus the 'rq' on its stack | ||
2059 | * frame will be invalid. | ||
2060 | */ | ||
2061 | finish_task_switch(this_rq(), prev); | ||
2062 | } | ||
2063 | |||
2064 | /* | ||
2065 | * nr_running, nr_uninterruptible and nr_context_switches: | ||
2066 | * | ||
2067 | * externally visible scheduler statistics: current number of runnable | ||
2068 | * threads, current number of uninterruptible-sleeping threads, total | ||
2069 | * number of context switches performed since bootup. | ||
2070 | */ | ||
2071 | unsigned long nr_running(void) | ||
2072 | { | ||
2073 | unsigned long i, sum = 0; | ||
2074 | |||
2075 | for_each_online_cpu(i) | ||
2076 | sum += cpu_rq(i)->nr_running; | ||
2077 | |||
2078 | return sum; | ||
2079 | } | ||
2080 | |||
2081 | unsigned long nr_uninterruptible(void) | ||
2082 | { | ||
2083 | unsigned long i, sum = 0; | ||
2084 | |||
2085 | for_each_possible_cpu(i) | ||
2086 | sum += cpu_rq(i)->nr_uninterruptible; | ||
2087 | |||
2088 | /* | ||
2089 | * Since we read the counters lockless, it might be slightly | ||
2090 | * inaccurate. Do not allow it to go below zero though: | ||
2091 | */ | ||
2092 | if (unlikely((long)sum < 0)) | ||
2093 | sum = 0; | ||
2094 | |||
2095 | return sum; | ||
2096 | } | ||
2097 | |||
2098 | unsigned long long nr_context_switches(void) | ||
2099 | { | ||
2100 | int i; | ||
2101 | unsigned long long sum = 0; | ||
2102 | |||
2103 | for_each_possible_cpu(i) | ||
2104 | sum += cpu_rq(i)->nr_switches; | ||
2105 | |||
2106 | return sum; | ||
2107 | } | ||
2108 | |||
2109 | unsigned long nr_iowait(void) | ||
2110 | { | ||
2111 | unsigned long i, sum = 0; | ||
2112 | |||
2113 | for_each_possible_cpu(i) | ||
2114 | sum += atomic_read(&cpu_rq(i)->nr_iowait); | ||
2115 | |||
2116 | return sum; | ||
2117 | } | ||
2118 | |||
2119 | unsigned long nr_iowait_cpu(int cpu) | ||
2120 | { | ||
2121 | struct rq *this = cpu_rq(cpu); | ||
2122 | return atomic_read(&this->nr_iowait); | ||
2123 | } | ||
2124 | |||
2125 | unsigned long this_cpu_load(void) | ||
2126 | { | ||
2127 | struct rq *this = this_rq(); | ||
2128 | return this->cpu_load[0]; | ||
2129 | } | ||
2130 | |||
2131 | |||
2132 | /* Variables and functions for calc_load */ | ||
2133 | static atomic_long_t calc_load_tasks; | ||
2134 | static unsigned long calc_load_update; | ||
2135 | unsigned long avenrun[3]; | ||
2136 | EXPORT_SYMBOL(avenrun); | ||
2137 | |||
2138 | static long calc_load_fold_active(struct rq *this_rq) | ||
2139 | { | ||
2140 | long nr_active, delta = 0; | ||
2141 | |||
2142 | nr_active = this_rq->nr_running; | ||
2143 | nr_active += (long) this_rq->nr_uninterruptible; | ||
2144 | |||
2145 | if (nr_active != this_rq->calc_load_active) { | ||
2146 | delta = nr_active - this_rq->calc_load_active; | ||
2147 | this_rq->calc_load_active = nr_active; | ||
2148 | } | ||
2149 | |||
2150 | return delta; | ||
2151 | } | ||
2152 | |||
2153 | static unsigned long | ||
2154 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | ||
2155 | { | ||
2156 | load *= exp; | ||
2157 | load += active * (FIXED_1 - exp); | ||
2158 | load += 1UL << (FSHIFT - 1); | ||
2159 | return load >> FSHIFT; | ||
2160 | } | ||
2161 | |||
2162 | #ifdef CONFIG_NO_HZ | ||
2163 | /* | ||
2164 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | ||
2165 | * | ||
2166 | * When making the ILB scale, we should try to pull this in as well. | ||
2167 | */ | ||
2168 | static atomic_long_t calc_load_tasks_idle; | ||
2169 | |||
2170 | void calc_load_account_idle(struct rq *this_rq) | ||
2171 | { | ||
2172 | long delta; | ||
2173 | |||
2174 | delta = calc_load_fold_active(this_rq); | ||
2175 | if (delta) | ||
2176 | atomic_long_add(delta, &calc_load_tasks_idle); | ||
2177 | } | ||
2178 | |||
2179 | static long calc_load_fold_idle(void) | ||
2180 | { | ||
2181 | long delta = 0; | ||
2182 | |||
2183 | /* | ||
2184 | * Its got a race, we don't care... | ||
2185 | */ | ||
2186 | if (atomic_long_read(&calc_load_tasks_idle)) | ||
2187 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | ||
2188 | |||
2189 | return delta; | ||
2190 | } | ||
2191 | |||
2192 | /** | ||
2193 | * fixed_power_int - compute: x^n, in O(log n) time | ||
2194 | * | ||
2195 | * @x: base of the power | ||
2196 | * @frac_bits: fractional bits of @x | ||
2197 | * @n: power to raise @x to. | ||
2198 | * | ||
2199 | * By exploiting the relation between the definition of the natural power | ||
2200 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | ||
2201 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | ||
2202 | * (where: n_i \elem {0, 1}, the binary vector representing n), | ||
2203 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | ||
2204 | * of course trivially computable in O(log_2 n), the length of our binary | ||
2205 | * vector. | ||
2206 | */ | ||
2207 | static unsigned long | ||
2208 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | ||
2209 | { | ||
2210 | unsigned long result = 1UL << frac_bits; | ||
2211 | |||
2212 | if (n) for (;;) { | ||
2213 | if (n & 1) { | ||
2214 | result *= x; | ||
2215 | result += 1UL << (frac_bits - 1); | ||
2216 | result >>= frac_bits; | ||
2217 | } | ||
2218 | n >>= 1; | ||
2219 | if (!n) | ||
2220 | break; | ||
2221 | x *= x; | ||
2222 | x += 1UL << (frac_bits - 1); | ||
2223 | x >>= frac_bits; | ||
2224 | } | ||
2225 | |||
2226 | return result; | ||
2227 | } | ||
2228 | |||
2229 | /* | ||
2230 | * a1 = a0 * e + a * (1 - e) | ||
2231 | * | ||
2232 | * a2 = a1 * e + a * (1 - e) | ||
2233 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | ||
2234 | * = a0 * e^2 + a * (1 - e) * (1 + e) | ||
2235 | * | ||
2236 | * a3 = a2 * e + a * (1 - e) | ||
2237 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | ||
2238 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | ||
2239 | * | ||
2240 | * ... | ||
2241 | * | ||
2242 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | ||
2243 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | ||
2244 | * = a0 * e^n + a * (1 - e^n) | ||
2245 | * | ||
2246 | * [1] application of the geometric series: | ||
2247 | * | ||
2248 | * n 1 - x^(n+1) | ||
2249 | * S_n := \Sum x^i = ------------- | ||
2250 | * i=0 1 - x | ||
2251 | */ | ||
2252 | static unsigned long | ||
2253 | calc_load_n(unsigned long load, unsigned long exp, | ||
2254 | unsigned long active, unsigned int n) | ||
2255 | { | ||
2256 | |||
2257 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | ||
2258 | } | ||
2259 | |||
2260 | /* | ||
2261 | * NO_HZ can leave us missing all per-cpu ticks calling | ||
2262 | * calc_load_account_active(), but since an idle CPU folds its delta into | ||
2263 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | ||
2264 | * in the pending idle delta if our idle period crossed a load cycle boundary. | ||
2265 | * | ||
2266 | * Once we've updated the global active value, we need to apply the exponential | ||
2267 | * weights adjusted to the number of cycles missed. | ||
2268 | */ | ||
2269 | static void calc_global_nohz(unsigned long ticks) | ||
2270 | { | ||
2271 | long delta, active, n; | ||
2272 | |||
2273 | if (time_before(jiffies, calc_load_update)) | ||
2274 | return; | ||
2275 | |||
2276 | /* | ||
2277 | * If we crossed a calc_load_update boundary, make sure to fold | ||
2278 | * any pending idle changes, the respective CPUs might have | ||
2279 | * missed the tick driven calc_load_account_active() update | ||
2280 | * due to NO_HZ. | ||
2281 | */ | ||
2282 | delta = calc_load_fold_idle(); | ||
2283 | if (delta) | ||
2284 | atomic_long_add(delta, &calc_load_tasks); | ||
2285 | |||
2286 | /* | ||
2287 | * If we were idle for multiple load cycles, apply them. | ||
2288 | */ | ||
2289 | if (ticks >= LOAD_FREQ) { | ||
2290 | n = ticks / LOAD_FREQ; | ||
2291 | |||
2292 | active = atomic_long_read(&calc_load_tasks); | ||
2293 | active = active > 0 ? active * FIXED_1 : 0; | ||
2294 | |||
2295 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | ||
2296 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | ||
2297 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | ||
2298 | |||
2299 | calc_load_update += n * LOAD_FREQ; | ||
2300 | } | ||
2301 | |||
2302 | /* | ||
2303 | * Its possible the remainder of the above division also crosses | ||
2304 | * a LOAD_FREQ period, the regular check in calc_global_load() | ||
2305 | * which comes after this will take care of that. | ||
2306 | * | ||
2307 | * Consider us being 11 ticks before a cycle completion, and us | ||
2308 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | ||
2309 | * age us 4 cycles, and the test in calc_global_load() will | ||
2310 | * pick up the final one. | ||
2311 | */ | ||
2312 | } | ||
2313 | #else | ||
2314 | void calc_load_account_idle(struct rq *this_rq) | ||
2315 | { | ||
2316 | } | ||
2317 | |||
2318 | static inline long calc_load_fold_idle(void) | ||
2319 | { | ||
2320 | return 0; | ||
2321 | } | ||
2322 | |||
2323 | static void calc_global_nohz(unsigned long ticks) | ||
2324 | { | ||
2325 | } | ||
2326 | #endif | ||
2327 | |||
2328 | /** | ||
2329 | * get_avenrun - get the load average array | ||
2330 | * @loads: pointer to dest load array | ||
2331 | * @offset: offset to add | ||
2332 | * @shift: shift count to shift the result left | ||
2333 | * | ||
2334 | * These values are estimates at best, so no need for locking. | ||
2335 | */ | ||
2336 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | ||
2337 | { | ||
2338 | loads[0] = (avenrun[0] + offset) << shift; | ||
2339 | loads[1] = (avenrun[1] + offset) << shift; | ||
2340 | loads[2] = (avenrun[2] + offset) << shift; | ||
2341 | } | ||
2342 | |||
2343 | /* | ||
2344 | * calc_load - update the avenrun load estimates 10 ticks after the | ||
2345 | * CPUs have updated calc_load_tasks. | ||
2346 | */ | ||
2347 | void calc_global_load(unsigned long ticks) | ||
2348 | { | ||
2349 | long active; | ||
2350 | |||
2351 | calc_global_nohz(ticks); | ||
2352 | |||
2353 | if (time_before(jiffies, calc_load_update + 10)) | ||
2354 | return; | ||
2355 | |||
2356 | active = atomic_long_read(&calc_load_tasks); | ||
2357 | active = active > 0 ? active * FIXED_1 : 0; | ||
2358 | |||
2359 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); | ||
2360 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | ||
2361 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | ||
2362 | |||
2363 | calc_load_update += LOAD_FREQ; | ||
2364 | } | ||
2365 | |||
2366 | /* | ||
2367 | * Called from update_cpu_load() to periodically update this CPU's | ||
2368 | * active count. | ||
2369 | */ | ||
2370 | static void calc_load_account_active(struct rq *this_rq) | ||
2371 | { | ||
2372 | long delta; | ||
2373 | |||
2374 | if (time_before(jiffies, this_rq->calc_load_update)) | ||
2375 | return; | ||
2376 | |||
2377 | delta = calc_load_fold_active(this_rq); | ||
2378 | delta += calc_load_fold_idle(); | ||
2379 | if (delta) | ||
2380 | atomic_long_add(delta, &calc_load_tasks); | ||
2381 | |||
2382 | this_rq->calc_load_update += LOAD_FREQ; | ||
2383 | } | ||
2384 | |||
2385 | /* | ||
2386 | * The exact cpuload at various idx values, calculated at every tick would be | ||
2387 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | ||
2388 | * | ||
2389 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | ||
2390 | * on nth tick when cpu may be busy, then we have: | ||
2391 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | ||
2392 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | ||
2393 | * | ||
2394 | * decay_load_missed() below does efficient calculation of | ||
2395 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | ||
2396 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | ||
2397 | * | ||
2398 | * The calculation is approximated on a 128 point scale. | ||
2399 | * degrade_zero_ticks is the number of ticks after which load at any | ||
2400 | * particular idx is approximated to be zero. | ||
2401 | * degrade_factor is a precomputed table, a row for each load idx. | ||
2402 | * Each column corresponds to degradation factor for a power of two ticks, | ||
2403 | * based on 128 point scale. | ||
2404 | * Example: | ||
2405 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | ||
2406 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | ||
2407 | * | ||
2408 | * With this power of 2 load factors, we can degrade the load n times | ||
2409 | * by looking at 1 bits in n and doing as many mult/shift instead of | ||
2410 | * n mult/shifts needed by the exact degradation. | ||
2411 | */ | ||
2412 | #define DEGRADE_SHIFT 7 | ||
2413 | static const unsigned char | ||
2414 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | ||
2415 | static const unsigned char | ||
2416 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | ||
2417 | {0, 0, 0, 0, 0, 0, 0, 0}, | ||
2418 | {64, 32, 8, 0, 0, 0, 0, 0}, | ||
2419 | {96, 72, 40, 12, 1, 0, 0}, | ||
2420 | {112, 98, 75, 43, 15, 1, 0}, | ||
2421 | {120, 112, 98, 76, 45, 16, 2} }; | ||
2422 | |||
2423 | /* | ||
2424 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | ||
2425 | * would be when CPU is idle and so we just decay the old load without | ||
2426 | * adding any new load. | ||
2427 | */ | ||
2428 | static unsigned long | ||
2429 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | ||
2430 | { | ||
2431 | int j = 0; | ||
2432 | |||
2433 | if (!missed_updates) | ||
2434 | return load; | ||
2435 | |||
2436 | if (missed_updates >= degrade_zero_ticks[idx]) | ||
2437 | return 0; | ||
2438 | |||
2439 | if (idx == 1) | ||
2440 | return load >> missed_updates; | ||
2441 | |||
2442 | while (missed_updates) { | ||
2443 | if (missed_updates % 2) | ||
2444 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | ||
2445 | |||
2446 | missed_updates >>= 1; | ||
2447 | j++; | ||
2448 | } | ||
2449 | return load; | ||
2450 | } | ||
2451 | |||
2452 | /* | ||
2453 | * Update rq->cpu_load[] statistics. This function is usually called every | ||
2454 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called | ||
2455 | * every tick. We fix it up based on jiffies. | ||
2456 | */ | ||
2457 | void update_cpu_load(struct rq *this_rq) | ||
2458 | { | ||
2459 | unsigned long this_load = this_rq->load.weight; | ||
2460 | unsigned long curr_jiffies = jiffies; | ||
2461 | unsigned long pending_updates; | ||
2462 | int i, scale; | ||
2463 | |||
2464 | this_rq->nr_load_updates++; | ||
2465 | |||
2466 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ | ||
2467 | if (curr_jiffies == this_rq->last_load_update_tick) | ||
2468 | return; | ||
2469 | |||
2470 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | ||
2471 | this_rq->last_load_update_tick = curr_jiffies; | ||
2472 | |||
2473 | /* Update our load: */ | ||
2474 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ | ||
2475 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | ||
2476 | unsigned long old_load, new_load; | ||
2477 | |||
2478 | /* scale is effectively 1 << i now, and >> i divides by scale */ | ||
2479 | |||
2480 | old_load = this_rq->cpu_load[i]; | ||
2481 | old_load = decay_load_missed(old_load, pending_updates - 1, i); | ||
2482 | new_load = this_load; | ||
2483 | /* | ||
2484 | * Round up the averaging division if load is increasing. This | ||
2485 | * prevents us from getting stuck on 9 if the load is 10, for | ||
2486 | * example. | ||
2487 | */ | ||
2488 | if (new_load > old_load) | ||
2489 | new_load += scale - 1; | ||
2490 | |||
2491 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | ||
2492 | } | ||
2493 | |||
2494 | sched_avg_update(this_rq); | ||
2495 | } | ||
2496 | |||
2497 | static void update_cpu_load_active(struct rq *this_rq) | ||
2498 | { | ||
2499 | update_cpu_load(this_rq); | ||
2500 | |||
2501 | calc_load_account_active(this_rq); | ||
2502 | } | ||
2503 | |||
2504 | #ifdef CONFIG_SMP | ||
2505 | |||
2506 | /* | ||
2507 | * sched_exec - execve() is a valuable balancing opportunity, because at | ||
2508 | * this point the task has the smallest effective memory and cache footprint. | ||
2509 | */ | ||
2510 | void sched_exec(void) | ||
2511 | { | ||
2512 | struct task_struct *p = current; | ||
2513 | unsigned long flags; | ||
2514 | int dest_cpu; | ||
2515 | |||
2516 | raw_spin_lock_irqsave(&p->pi_lock, flags); | ||
2517 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); | ||
2518 | if (dest_cpu == smp_processor_id()) | ||
2519 | goto unlock; | ||
2520 | |||
2521 | if (likely(cpu_active(dest_cpu))) { | ||
2522 | struct migration_arg arg = { p, dest_cpu }; | ||
2523 | |||
2524 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | ||
2525 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | ||
2526 | return; | ||
2527 | } | ||
2528 | unlock: | ||
2529 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | ||
2530 | } | ||
2531 | |||
2532 | #endif | ||
2533 | |||
2534 | DEFINE_PER_CPU(struct kernel_stat, kstat); | ||
2535 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); | ||
2536 | |||
2537 | EXPORT_PER_CPU_SYMBOL(kstat); | ||
2538 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); | ||
2539 | |||
2540 | /* | ||
2541 | * Return any ns on the sched_clock that have not yet been accounted in | ||
2542 | * @p in case that task is currently running. | ||
2543 | * | ||
2544 | * Called with task_rq_lock() held on @rq. | ||
2545 | */ | ||
2546 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) | ||
2547 | { | ||
2548 | u64 ns = 0; | ||
2549 | |||
2550 | if (task_current(rq, p)) { | ||
2551 | update_rq_clock(rq); | ||
2552 | ns = rq->clock_task - p->se.exec_start; | ||
2553 | if ((s64)ns < 0) | ||
2554 | ns = 0; | ||
2555 | } | ||
2556 | |||
2557 | return ns; | ||
2558 | } | ||
2559 | |||
2560 | unsigned long long task_delta_exec(struct task_struct *p) | ||
2561 | { | ||
2562 | unsigned long flags; | ||
2563 | struct rq *rq; | ||
2564 | u64 ns = 0; | ||
2565 | |||
2566 | rq = task_rq_lock(p, &flags); | ||
2567 | ns = do_task_delta_exec(p, rq); | ||
2568 | task_rq_unlock(rq, p, &flags); | ||
2569 | |||
2570 | return ns; | ||
2571 | } | ||
2572 | |||
2573 | /* | ||
2574 | * Return accounted runtime for the task. | ||
2575 | * In case the task is currently running, return the runtime plus current's | ||
2576 | * pending runtime that have not been accounted yet. | ||
2577 | */ | ||
2578 | unsigned long long task_sched_runtime(struct task_struct *p) | ||
2579 | { | ||
2580 | unsigned long flags; | ||
2581 | struct rq *rq; | ||
2582 | u64 ns = 0; | ||
2583 | |||
2584 | rq = task_rq_lock(p, &flags); | ||
2585 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | ||
2586 | task_rq_unlock(rq, p, &flags); | ||
2587 | |||
2588 | return ns; | ||
2589 | } | ||
2590 | |||
2591 | #ifdef CONFIG_CGROUP_CPUACCT | ||
2592 | struct cgroup_subsys cpuacct_subsys; | ||
2593 | struct cpuacct root_cpuacct; | ||
2594 | #endif | ||
2595 | |||
2596 | static inline void task_group_account_field(struct task_struct *p, int index, | ||
2597 | u64 tmp) | ||
2598 | { | ||
2599 | #ifdef CONFIG_CGROUP_CPUACCT | ||
2600 | struct kernel_cpustat *kcpustat; | ||
2601 | struct cpuacct *ca; | ||
2602 | #endif | ||
2603 | /* | ||
2604 | * Since all updates are sure to touch the root cgroup, we | ||
2605 | * get ourselves ahead and touch it first. If the root cgroup | ||
2606 | * is the only cgroup, then nothing else should be necessary. | ||
2607 | * | ||
2608 | */ | ||
2609 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | ||
2610 | |||
2611 | #ifdef CONFIG_CGROUP_CPUACCT | ||
2612 | if (unlikely(!cpuacct_subsys.active)) | ||
2613 | return; | ||
2614 | |||
2615 | rcu_read_lock(); | ||
2616 | ca = task_ca(p); | ||
2617 | while (ca && (ca != &root_cpuacct)) { | ||
2618 | kcpustat = this_cpu_ptr(ca->cpustat); | ||
2619 | kcpustat->cpustat[index] += tmp; | ||
2620 | ca = parent_ca(ca); | ||
2621 | } | ||
2622 | rcu_read_unlock(); | ||
2623 | #endif | ||
2624 | } | ||
2625 | |||
2626 | |||
2627 | /* | ||
2628 | * Account user cpu time to a process. | ||
2629 | * @p: the process that the cpu time gets accounted to | ||
2630 | * @cputime: the cpu time spent in user space since the last update | ||
2631 | * @cputime_scaled: cputime scaled by cpu frequency | ||
2632 | */ | ||
2633 | void account_user_time(struct task_struct *p, cputime_t cputime, | ||
2634 | cputime_t cputime_scaled) | ||
2635 | { | ||
2636 | int index; | ||
2637 | |||
2638 | /* Add user time to process. */ | ||
2639 | p->utime += cputime; | ||
2640 | p->utimescaled += cputime_scaled; | ||
2641 | account_group_user_time(p, cputime); | ||
2642 | |||
2643 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; | ||
2644 | |||
2645 | /* Add user time to cpustat. */ | ||
2646 | task_group_account_field(p, index, (__force u64) cputime); | ||
2647 | |||
2648 | /* Account for user time used */ | ||
2649 | acct_update_integrals(p); | ||
2650 | } | ||
2651 | |||
2652 | /* | ||
2653 | * Account guest cpu time to a process. | ||
2654 | * @p: the process that the cpu time gets accounted to | ||
2655 | * @cputime: the cpu time spent in virtual machine since the last update | ||
2656 | * @cputime_scaled: cputime scaled by cpu frequency | ||
2657 | */ | ||
2658 | static void account_guest_time(struct task_struct *p, cputime_t cputime, | ||
2659 | cputime_t cputime_scaled) | ||
2660 | { | ||
2661 | u64 *cpustat = kcpustat_this_cpu->cpustat; | ||
2662 | |||
2663 | /* Add guest time to process. */ | ||
2664 | p->utime += cputime; | ||
2665 | p->utimescaled += cputime_scaled; | ||
2666 | account_group_user_time(p, cputime); | ||
2667 | p->gtime += cputime; | ||
2668 | |||
2669 | /* Add guest time to cpustat. */ | ||
2670 | if (TASK_NICE(p) > 0) { | ||
2671 | cpustat[CPUTIME_NICE] += (__force u64) cputime; | ||
2672 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | ||
2673 | } else { | ||
2674 | cpustat[CPUTIME_USER] += (__force u64) cputime; | ||
2675 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | ||
2676 | } | ||
2677 | } | ||
2678 | |||
2679 | /* | ||
2680 | * Account system cpu time to a process and desired cpustat field | ||
2681 | * @p: the process that the cpu time gets accounted to | ||
2682 | * @cputime: the cpu time spent in kernel space since the last update | ||
2683 | * @cputime_scaled: cputime scaled by cpu frequency | ||
2684 | * @target_cputime64: pointer to cpustat field that has to be updated | ||
2685 | */ | ||
2686 | static inline | ||
2687 | void __account_system_time(struct task_struct *p, cputime_t cputime, | ||
2688 | cputime_t cputime_scaled, int index) | ||
2689 | { | ||
2690 | /* Add system time to process. */ | ||
2691 | p->stime += cputime; | ||
2692 | p->stimescaled += cputime_scaled; | ||
2693 | account_group_system_time(p, cputime); | ||
2694 | |||
2695 | /* Add system time to cpustat. */ | ||
2696 | task_group_account_field(p, index, (__force u64) cputime); | ||
2697 | |||
2698 | /* Account for system time used */ | ||
2699 | acct_update_integrals(p); | ||
2700 | } | ||
2701 | |||
2702 | /* | ||
2703 | * Account system cpu time to a process. | ||
2704 | * @p: the process that the cpu time gets accounted to | ||
2705 | * @hardirq_offset: the offset to subtract from hardirq_count() | ||
2706 | * @cputime: the cpu time spent in kernel space since the last update | ||
2707 | * @cputime_scaled: cputime scaled by cpu frequency | ||
2708 | */ | ||
2709 | void account_system_time(struct task_struct *p, int hardirq_offset, | ||
2710 | cputime_t cputime, cputime_t cputime_scaled) | ||
2711 | { | ||
2712 | int index; | ||
2713 | |||
2714 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | ||
2715 | account_guest_time(p, cputime, cputime_scaled); | ||
2716 | return; | ||
2717 | } | ||
2718 | |||
2719 | if (hardirq_count() - hardirq_offset) | ||
2720 | index = CPUTIME_IRQ; | ||
2721 | else if (in_serving_softirq()) | ||
2722 | index = CPUTIME_SOFTIRQ; | ||
2723 | else | ||
2724 | index = CPUTIME_SYSTEM; | ||
2725 | |||
2726 | __account_system_time(p, cputime, cputime_scaled, index); | ||
2727 | } | ||
2728 | |||
2729 | /* | ||
2730 | * Account for involuntary wait time. | ||
2731 | * @cputime: the cpu time spent in involuntary wait | ||
2732 | */ | ||
2733 | void account_steal_time(cputime_t cputime) | ||
2734 | { | ||
2735 | u64 *cpustat = kcpustat_this_cpu->cpustat; | ||
2736 | |||
2737 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; | ||
2738 | } | ||
2739 | |||
2740 | /* | ||
2741 | * Account for idle time. | ||
2742 | * @cputime: the cpu time spent in idle wait | ||
2743 | */ | ||
2744 | void account_idle_time(cputime_t cputime) | ||
2745 | { | ||
2746 | u64 *cpustat = kcpustat_this_cpu->cpustat; | ||
2747 | struct rq *rq = this_rq(); | ||
2748 | |||
2749 | if (atomic_read(&rq->nr_iowait) > 0) | ||
2750 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; | ||
2751 | else | ||
2752 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; | ||
2753 | } | ||
2754 | |||
2755 | static __always_inline bool steal_account_process_tick(void) | ||
2756 | { | ||
2757 | #ifdef CONFIG_PARAVIRT | ||
2758 | if (static_branch(¶virt_steal_enabled)) { | ||
2759 | u64 steal, st = 0; | ||
2760 | |||
2761 | steal = paravirt_steal_clock(smp_processor_id()); | ||
2762 | steal -= this_rq()->prev_steal_time; | ||
2763 | |||
2764 | st = steal_ticks(steal); | ||
2765 | this_rq()->prev_steal_time += st * TICK_NSEC; | ||
2766 | |||
2767 | account_steal_time(st); | ||
2768 | return st; | ||
2769 | } | ||
2770 | #endif | ||
2771 | return false; | ||
2772 | } | ||
2773 | |||
2774 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING | ||
2775 | |||
2776 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | ||
2777 | /* | ||
2778 | * Account a tick to a process and cpustat | ||
2779 | * @p: the process that the cpu time gets accounted to | ||
2780 | * @user_tick: is the tick from userspace | ||
2781 | * @rq: the pointer to rq | ||
2782 | * | ||
2783 | * Tick demultiplexing follows the order | ||
2784 | * - pending hardirq update | ||
2785 | * - pending softirq update | ||
2786 | * - user_time | ||
2787 | * - idle_time | ||
2788 | * - system time | ||
2789 | * - check for guest_time | ||
2790 | * - else account as system_time | ||
2791 | * | ||
2792 | * Check for hardirq is done both for system and user time as there is | ||
2793 | * no timer going off while we are on hardirq and hence we may never get an | ||
2794 | * opportunity to update it solely in system time. | ||
2795 | * p->stime and friends are only updated on system time and not on irq | ||
2796 | * softirq as those do not count in task exec_runtime any more. | ||
2797 | */ | ||
2798 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | ||
2799 | struct rq *rq) | ||
2800 | { | ||
2801 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | ||
2802 | u64 *cpustat = kcpustat_this_cpu->cpustat; | ||
2803 | |||
2804 | if (steal_account_process_tick()) | ||
2805 | return; | ||
2806 | |||
2807 | if (irqtime_account_hi_update()) { | ||
2808 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; | ||
2809 | } else if (irqtime_account_si_update()) { | ||
2810 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; | ||
2811 | } else if (this_cpu_ksoftirqd() == p) { | ||
2812 | /* | ||
2813 | * ksoftirqd time do not get accounted in cpu_softirq_time. | ||
2814 | * So, we have to handle it separately here. | ||
2815 | * Also, p->stime needs to be updated for ksoftirqd. | ||
2816 | */ | ||
2817 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | ||
2818 | CPUTIME_SOFTIRQ); | ||
2819 | } else if (user_tick) { | ||
2820 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | ||
2821 | } else if (p == rq->idle) { | ||
2822 | account_idle_time(cputime_one_jiffy); | ||
2823 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | ||
2824 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | ||
2825 | } else { | ||
2826 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | ||
2827 | CPUTIME_SYSTEM); | ||
2828 | } | ||
2829 | } | ||
2830 | |||
2831 | static void irqtime_account_idle_ticks(int ticks) | ||
2832 | { | ||
2833 | int i; | ||
2834 | struct rq *rq = this_rq(); | ||
2835 | |||
2836 | for (i = 0; i < ticks; i++) | ||
2837 | irqtime_account_process_tick(current, 0, rq); | ||
2838 | } | ||
2839 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | ||
2840 | static void irqtime_account_idle_ticks(int ticks) {} | ||
2841 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | ||
2842 | struct rq *rq) {} | ||
2843 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ | ||
2844 | |||
2845 | /* | ||
2846 | * Account a single tick of cpu time. | ||
2847 | * @p: the process that the cpu time gets accounted to | ||
2848 | * @user_tick: indicates if the tick is a user or a system tick | ||
2849 | */ | ||
2850 | void account_process_tick(struct task_struct *p, int user_tick) | ||
2851 | { | ||
2852 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | ||
2853 | struct rq *rq = this_rq(); | ||
2854 | |||
2855 | if (sched_clock_irqtime) { | ||
2856 | irqtime_account_process_tick(p, user_tick, rq); | ||
2857 | return; | ||
2858 | } | ||
2859 | |||
2860 | if (steal_account_process_tick()) | ||
2861 | return; | ||
2862 | |||
2863 | if (user_tick) | ||
2864 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | ||
2865 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) | ||
2866 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, | ||
2867 | one_jiffy_scaled); | ||
2868 | else | ||
2869 | account_idle_time(cputime_one_jiffy); | ||
2870 | } | ||
2871 | |||
2872 | /* | ||
2873 | * Account multiple ticks of steal time. | ||
2874 | * @p: the process from which the cpu time has been stolen | ||
2875 | * @ticks: number of stolen ticks | ||
2876 | */ | ||
2877 | void account_steal_ticks(unsigned long ticks) | ||
2878 | { | ||
2879 | account_steal_time(jiffies_to_cputime(ticks)); | ||
2880 | } | ||
2881 | |||
2882 | /* | ||
2883 | * Account multiple ticks of idle time. | ||
2884 | * @ticks: number of stolen ticks | ||
2885 | */ | ||
2886 | void account_idle_ticks(unsigned long ticks) | ||
2887 | { | ||
2888 | |||
2889 | if (sched_clock_irqtime) { | ||
2890 | irqtime_account_idle_ticks(ticks); | ||
2891 | return; | ||
2892 | } | ||
2893 | |||
2894 | account_idle_time(jiffies_to_cputime(ticks)); | ||
2895 | } | ||
2896 | |||
2897 | #endif | ||
2898 | |||
2899 | /* | ||
2900 | * Use precise platform statistics if available: | ||
2901 | */ | ||
2902 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | ||
2903 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | ||
2904 | { | ||
2905 | *ut = p->utime; | ||
2906 | *st = p->stime; | ||
2907 | } | ||
2908 | |||
2909 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | ||
2910 | { | ||
2911 | struct task_cputime cputime; | ||
2912 | |||
2913 | thread_group_cputime(p, &cputime); | ||
2914 | |||
2915 | *ut = cputime.utime; | ||
2916 | *st = cputime.stime; | ||
2917 | } | ||
2918 | #else | ||
2919 | |||
2920 | #ifndef nsecs_to_cputime | ||
2921 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) | ||
2922 | #endif | ||
2923 | |||
2924 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | ||
2925 | { | ||
2926 | cputime_t rtime, utime = p->utime, total = utime + p->stime; | ||
2927 | |||
2928 | /* | ||
2929 | * Use CFS's precise accounting: | ||
2930 | */ | ||
2931 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); | ||
2932 | |||
2933 | if (total) { | ||
2934 | u64 temp = (__force u64) rtime; | ||
2935 | |||
2936 | temp *= (__force u64) utime; | ||
2937 | do_div(temp, (__force u32) total); | ||
2938 | utime = (__force cputime_t) temp; | ||
2939 | } else | ||
2940 | utime = rtime; | ||
2941 | |||
2942 | /* | ||
2943 | * Compare with previous values, to keep monotonicity: | ||
2944 | */ | ||
2945 | p->prev_utime = max(p->prev_utime, utime); | ||
2946 | p->prev_stime = max(p->prev_stime, rtime - p->prev_utime); | ||
2947 | |||
2948 | *ut = p->prev_utime; | ||
2949 | *st = p->prev_stime; | ||
2950 | } | ||
2951 | |||
2952 | /* | ||
2953 | * Must be called with siglock held. | ||
2954 | */ | ||
2955 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | ||
2956 | { | ||
2957 | struct signal_struct *sig = p->signal; | ||
2958 | struct task_cputime cputime; | ||
2959 | cputime_t rtime, utime, total; | ||
2960 | |||
2961 | thread_group_cputime(p, &cputime); | ||
2962 | |||
2963 | total = cputime.utime + cputime.stime; | ||
2964 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | ||
2965 | |||
2966 | if (total) { | ||
2967 | u64 temp = (__force u64) rtime; | ||
2968 | |||
2969 | temp *= (__force u64) cputime.utime; | ||
2970 | do_div(temp, (__force u32) total); | ||
2971 | utime = (__force cputime_t) temp; | ||
2972 | } else | ||
2973 | utime = rtime; | ||
2974 | |||
2975 | sig->prev_utime = max(sig->prev_utime, utime); | ||
2976 | sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime); | ||
2977 | |||
2978 | *ut = sig->prev_utime; | ||
2979 | *st = sig->prev_stime; | ||
2980 | } | ||
2981 | #endif | ||
2982 | |||
2983 | /* | ||
2984 | * This function gets called by the timer code, with HZ frequency. | ||
2985 | * We call it with interrupts disabled. | ||
2986 | */ | ||
2987 | void scheduler_tick(void) | ||
2988 | { | ||
2989 | int cpu = smp_processor_id(); | ||
2990 | struct rq *rq = cpu_rq(cpu); | ||
2991 | struct task_struct *curr = rq->curr; | ||
2992 | |||
2993 | sched_clock_tick(); | ||
2994 | |||
2995 | raw_spin_lock(&rq->lock); | ||
2996 | update_rq_clock(rq); | ||
2997 | update_cpu_load_active(rq); | ||
2998 | curr->sched_class->task_tick(rq, curr, 0); | ||
2999 | raw_spin_unlock(&rq->lock); | ||
3000 | |||
3001 | perf_event_task_tick(); | ||
3002 | |||
3003 | #ifdef CONFIG_SMP | ||
3004 | rq->idle_balance = idle_cpu(cpu); | ||
3005 | trigger_load_balance(rq, cpu); | ||
3006 | #endif | ||
3007 | } | ||
3008 | |||
3009 | notrace unsigned long get_parent_ip(unsigned long addr) | ||
3010 | { | ||
3011 | if (in_lock_functions(addr)) { | ||
3012 | addr = CALLER_ADDR2; | ||
3013 | if (in_lock_functions(addr)) | ||
3014 | addr = CALLER_ADDR3; | ||
3015 | } | ||
3016 | return addr; | ||
3017 | } | ||
3018 | |||
3019 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ | ||
3020 | defined(CONFIG_PREEMPT_TRACER)) | ||
3021 | |||
3022 | void __kprobes add_preempt_count(int val) | ||
3023 | { | ||
3024 | #ifdef CONFIG_DEBUG_PREEMPT | ||
3025 | /* | ||
3026 | * Underflow? | ||
3027 | */ | ||
3028 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) | ||
3029 | return; | ||
3030 | #endif | ||
3031 | preempt_count() += val; | ||
3032 | #ifdef CONFIG_DEBUG_PREEMPT | ||
3033 | /* | ||
3034 | * Spinlock count overflowing soon? | ||
3035 | */ | ||
3036 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= | ||
3037 | PREEMPT_MASK - 10); | ||
3038 | #endif | ||
3039 | if (preempt_count() == val) | ||
3040 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | ||
3041 | } | ||
3042 | EXPORT_SYMBOL(add_preempt_count); | ||
3043 | |||
3044 | void __kprobes sub_preempt_count(int val) | ||
3045 | { | ||
3046 | #ifdef CONFIG_DEBUG_PREEMPT | ||
3047 | /* | ||
3048 | * Underflow? | ||
3049 | */ | ||
3050 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) | ||
3051 | return; | ||
3052 | /* | ||
3053 | * Is the spinlock portion underflowing? | ||
3054 | */ | ||
3055 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && | ||
3056 | !(preempt_count() & PREEMPT_MASK))) | ||
3057 | return; | ||
3058 | #endif | ||
3059 | |||
3060 | if (preempt_count() == val) | ||
3061 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | ||
3062 | preempt_count() -= val; | ||
3063 | } | ||
3064 | EXPORT_SYMBOL(sub_preempt_count); | ||
3065 | |||
3066 | #endif | ||
3067 | |||
3068 | /* | ||
3069 | * Print scheduling while atomic bug: | ||
3070 | */ | ||
3071 | static noinline void __schedule_bug(struct task_struct *prev) | ||
3072 | { | ||
3073 | struct pt_regs *regs = get_irq_regs(); | ||
3074 | |||
3075 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | ||
3076 | prev->comm, prev->pid, preempt_count()); | ||
3077 | |||
3078 | debug_show_held_locks(prev); | ||
3079 | print_modules(); | ||
3080 | if (irqs_disabled()) | ||
3081 | print_irqtrace_events(prev); | ||
3082 | |||
3083 | if (regs) | ||
3084 | show_regs(regs); | ||
3085 | else | ||
3086 | dump_stack(); | ||
3087 | } | ||
3088 | |||
3089 | /* | ||
3090 | * Various schedule()-time debugging checks and statistics: | ||
3091 | */ | ||
3092 | static inline void schedule_debug(struct task_struct *prev) | ||
3093 | { | ||
3094 | /* | ||
3095 | * Test if we are atomic. Since do_exit() needs to call into | ||
3096 | * schedule() atomically, we ignore that path for now. | ||
3097 | * Otherwise, whine if we are scheduling when we should not be. | ||
3098 | */ | ||
3099 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) | ||
3100 | __schedule_bug(prev); | ||
3101 | rcu_sleep_check(); | ||
3102 | |||
3103 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); | ||
3104 | |||
3105 | schedstat_inc(this_rq(), sched_count); | ||
3106 | } | ||
3107 | |||
3108 | static void put_prev_task(struct rq *rq, struct task_struct *prev) | ||
3109 | { | ||
3110 | if (prev->on_rq || rq->skip_clock_update < 0) | ||
3111 | update_rq_clock(rq); | ||
3112 | prev->sched_class->put_prev_task(rq, prev); | ||
3113 | } | ||
3114 | |||
3115 | /* | ||
3116 | * Pick up the highest-prio task: | ||
3117 | */ | ||
3118 | static inline struct task_struct * | ||
3119 | pick_next_task(struct rq *rq) | ||
3120 | { | ||
3121 | const struct sched_class *class; | ||
3122 | struct task_struct *p; | ||
3123 | |||
3124 | /* | ||
3125 | * Optimization: we know that if all tasks are in | ||
3126 | * the fair class we can call that function directly: | ||
3127 | */ | ||
3128 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { | ||
3129 | p = fair_sched_class.pick_next_task(rq); | ||
3130 | if (likely(p)) | ||
3131 | return p; | ||
3132 | } | ||
3133 | |||
3134 | for_each_class(class) { | ||
3135 | p = class->pick_next_task(rq); | ||
3136 | if (p) | ||
3137 | return p; | ||
3138 | } | ||
3139 | |||
3140 | BUG(); /* the idle class will always have a runnable task */ | ||
3141 | } | ||
3142 | |||
3143 | /* | ||
3144 | * __schedule() is the main scheduler function. | ||
3145 | */ | ||
3146 | static void __sched __schedule(void) | ||
3147 | { | ||
3148 | struct task_struct *prev, *next; | ||
3149 | unsigned long *switch_count; | ||
3150 | struct rq *rq; | ||
3151 | int cpu; | ||
3152 | |||
3153 | need_resched: | ||
3154 | preempt_disable(); | ||
3155 | cpu = smp_processor_id(); | ||
3156 | rq = cpu_rq(cpu); | ||
3157 | rcu_note_context_switch(cpu); | ||
3158 | prev = rq->curr; | ||
3159 | |||
3160 | schedule_debug(prev); | ||
3161 | |||
3162 | if (sched_feat(HRTICK)) | ||
3163 | hrtick_clear(rq); | ||
3164 | |||
3165 | raw_spin_lock_irq(&rq->lock); | ||
3166 | |||
3167 | switch_count = &prev->nivcsw; | ||
3168 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { | ||
3169 | if (unlikely(signal_pending_state(prev->state, prev))) { | ||
3170 | prev->state = TASK_RUNNING; | ||
3171 | } else { | ||
3172 | deactivate_task(rq, prev, DEQUEUE_SLEEP); | ||
3173 | prev->on_rq = 0; | ||
3174 | |||
3175 | /* | ||
3176 | * If a worker went to sleep, notify and ask workqueue | ||
3177 | * whether it wants to wake up a task to maintain | ||
3178 | * concurrency. | ||
3179 | */ | ||
3180 | if (prev->flags & PF_WQ_WORKER) { | ||
3181 | struct task_struct *to_wakeup; | ||
3182 | |||
3183 | to_wakeup = wq_worker_sleeping(prev, cpu); | ||
3184 | if (to_wakeup) | ||
3185 | try_to_wake_up_local(to_wakeup); | ||
3186 | } | ||
3187 | } | ||
3188 | switch_count = &prev->nvcsw; | ||
3189 | } | ||
3190 | |||
3191 | pre_schedule(rq, prev); | ||
3192 | |||
3193 | if (unlikely(!rq->nr_running)) | ||
3194 | idle_balance(cpu, rq); | ||
3195 | |||
3196 | put_prev_task(rq, prev); | ||
3197 | next = pick_next_task(rq); | ||
3198 | clear_tsk_need_resched(prev); | ||
3199 | rq->skip_clock_update = 0; | ||
3200 | |||
3201 | if (likely(prev != next)) { | ||
3202 | rq->nr_switches++; | ||
3203 | rq->curr = next; | ||
3204 | ++*switch_count; | ||
3205 | |||
3206 | context_switch(rq, prev, next); /* unlocks the rq */ | ||
3207 | /* | ||
3208 | * The context switch have flipped the stack from under us | ||
3209 | * and restored the local variables which were saved when | ||
3210 | * this task called schedule() in the past. prev == current | ||
3211 | * is still correct, but it can be moved to another cpu/rq. | ||
3212 | */ | ||
3213 | cpu = smp_processor_id(); | ||
3214 | rq = cpu_rq(cpu); | ||
3215 | } else | ||
3216 | raw_spin_unlock_irq(&rq->lock); | ||
3217 | |||
3218 | post_schedule(rq); | ||
3219 | |||
3220 | preempt_enable_no_resched(); | ||
3221 | if (need_resched()) | ||
3222 | goto need_resched; | ||
3223 | } | ||
3224 | |||
3225 | static inline void sched_submit_work(struct task_struct *tsk) | ||
3226 | { | ||
3227 | if (!tsk->state) | ||
3228 | return; | ||
3229 | /* | ||
3230 | * If we are going to sleep and we have plugged IO queued, | ||
3231 | * make sure to submit it to avoid deadlocks. | ||
3232 | */ | ||
3233 | if (blk_needs_flush_plug(tsk)) | ||
3234 | blk_schedule_flush_plug(tsk); | ||
3235 | } | ||
3236 | |||
3237 | asmlinkage void __sched schedule(void) | ||
3238 | { | ||
3239 | struct task_struct *tsk = current; | ||
3240 | |||
3241 | sched_submit_work(tsk); | ||
3242 | __schedule(); | ||
3243 | } | ||
3244 | EXPORT_SYMBOL(schedule); | ||
3245 | |||
3246 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER | ||
3247 | |||
3248 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) | ||
3249 | { | ||
3250 | if (lock->owner != owner) | ||
3251 | return false; | ||
3252 | |||
3253 | /* | ||
3254 | * Ensure we emit the owner->on_cpu, dereference _after_ checking | ||
3255 | * lock->owner still matches owner, if that fails, owner might | ||
3256 | * point to free()d memory, if it still matches, the rcu_read_lock() | ||
3257 | * ensures the memory stays valid. | ||
3258 | */ | ||
3259 | barrier(); | ||
3260 | |||
3261 | return owner->on_cpu; | ||
3262 | } | ||
3263 | |||
3264 | /* | ||
3265 | * Look out! "owner" is an entirely speculative pointer | ||
3266 | * access and not reliable. | ||
3267 | */ | ||
3268 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | ||
3269 | { | ||
3270 | if (!sched_feat(OWNER_SPIN)) | ||
3271 | return 0; | ||
3272 | |||
3273 | rcu_read_lock(); | ||
3274 | while (owner_running(lock, owner)) { | ||
3275 | if (need_resched()) | ||
3276 | break; | ||
3277 | |||
3278 | arch_mutex_cpu_relax(); | ||
3279 | } | ||
3280 | rcu_read_unlock(); | ||
3281 | |||
3282 | /* | ||
3283 | * We break out the loop above on need_resched() and when the | ||
3284 | * owner changed, which is a sign for heavy contention. Return | ||
3285 | * success only when lock->owner is NULL. | ||
3286 | */ | ||
3287 | return lock->owner == NULL; | ||
3288 | } | ||
3289 | #endif | ||
3290 | |||
3291 | #ifdef CONFIG_PREEMPT | ||
3292 | /* | ||
3293 | * this is the entry point to schedule() from in-kernel preemption | ||
3294 | * off of preempt_enable. Kernel preemptions off return from interrupt | ||
3295 | * occur there and call schedule directly. | ||
3296 | */ | ||
3297 | asmlinkage void __sched notrace preempt_schedule(void) | ||
3298 | { | ||
3299 | struct thread_info *ti = current_thread_info(); | ||
3300 | |||
3301 | /* | ||
3302 | * If there is a non-zero preempt_count or interrupts are disabled, | ||
3303 | * we do not want to preempt the current task. Just return.. | ||
3304 | */ | ||
3305 | if (likely(ti->preempt_count || irqs_disabled())) | ||
3306 | return; | ||
3307 | |||
3308 | do { | ||
3309 | add_preempt_count_notrace(PREEMPT_ACTIVE); | ||
3310 | __schedule(); | ||
3311 | sub_preempt_count_notrace(PREEMPT_ACTIVE); | ||
3312 | |||
3313 | /* | ||
3314 | * Check again in case we missed a preemption opportunity | ||
3315 | * between schedule and now. | ||
3316 | */ | ||
3317 | barrier(); | ||
3318 | } while (need_resched()); | ||
3319 | } | ||
3320 | EXPORT_SYMBOL(preempt_schedule); | ||
3321 | |||
3322 | /* | ||
3323 | * this is the entry point to schedule() from kernel preemption | ||
3324 | * off of irq context. | ||
3325 | * Note, that this is called and return with irqs disabled. This will | ||
3326 | * protect us against recursive calling from irq. | ||
3327 | */ | ||
3328 | asmlinkage void __sched preempt_schedule_irq(void) | ||
3329 | { | ||
3330 | struct thread_info *ti = current_thread_info(); | ||
3331 | |||
3332 | /* Catch callers which need to be fixed */ | ||
3333 | BUG_ON(ti->preempt_count || !irqs_disabled()); | ||
3334 | |||
3335 | do { | ||
3336 | add_preempt_count(PREEMPT_ACTIVE); | ||
3337 | local_irq_enable(); | ||
3338 | __schedule(); | ||
3339 | local_irq_disable(); | ||
3340 | sub_preempt_count(PREEMPT_ACTIVE); | ||
3341 | |||
3342 | /* | ||
3343 | * Check again in case we missed a preemption opportunity | ||
3344 | * between schedule and now. | ||
3345 | */ | ||
3346 | barrier(); | ||
3347 | } while (need_resched()); | ||
3348 | } | ||
3349 | |||
3350 | #endif /* CONFIG_PREEMPT */ | ||
3351 | |||
3352 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, | ||
3353 | void *key) | ||
3354 | { | ||
3355 | return try_to_wake_up(curr->private, mode, wake_flags); | ||
3356 | } | ||
3357 | EXPORT_SYMBOL(default_wake_function); | ||
3358 | |||
3359 | /* | ||
3360 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | ||
3361 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | ||
3362 | * number) then we wake all the non-exclusive tasks and one exclusive task. | ||
3363 | * | ||
3364 | * There are circumstances in which we can try to wake a task which has already | ||
3365 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | ||
3366 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | ||
3367 | */ | ||
3368 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | ||
3369 | int nr_exclusive, int wake_flags, void *key) | ||
3370 | { | ||
3371 | wait_queue_t *curr, *next; | ||
3372 | |||
3373 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { | ||
3374 | unsigned flags = curr->flags; | ||
3375 | |||
3376 | if (curr->func(curr, mode, wake_flags, key) && | ||
3377 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) | ||
3378 | break; | ||
3379 | } | ||
3380 | } | ||
3381 | |||
3382 | /** | ||
3383 | * __wake_up - wake up threads blocked on a waitqueue. | ||
3384 | * @q: the waitqueue | ||
3385 | * @mode: which threads | ||
3386 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | ||
3387 | * @key: is directly passed to the wakeup function | ||
3388 | * | ||
3389 | * It may be assumed that this function implies a write memory barrier before | ||
3390 | * changing the task state if and only if any tasks are woken up. | ||
3391 | */ | ||
3392 | void __wake_up(wait_queue_head_t *q, unsigned int mode, | ||
3393 | int nr_exclusive, void *key) | ||
3394 | { | ||
3395 | unsigned long flags; | ||
3396 | |||
3397 | spin_lock_irqsave(&q->lock, flags); | ||
3398 | __wake_up_common(q, mode, nr_exclusive, 0, key); | ||
3399 | spin_unlock_irqrestore(&q->lock, flags); | ||
3400 | } | ||
3401 | EXPORT_SYMBOL(__wake_up); | ||
3402 | |||
3403 | /* | ||
3404 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | ||
3405 | */ | ||
3406 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | ||
3407 | { | ||
3408 | __wake_up_common(q, mode, 1, 0, NULL); | ||
3409 | } | ||
3410 | EXPORT_SYMBOL_GPL(__wake_up_locked); | ||
3411 | |||
3412 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) | ||
3413 | { | ||
3414 | __wake_up_common(q, mode, 1, 0, key); | ||
3415 | } | ||
3416 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); | ||
3417 | |||
3418 | /** | ||
3419 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. | ||
3420 | * @q: the waitqueue | ||
3421 | * @mode: which threads | ||
3422 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | ||
3423 | * @key: opaque value to be passed to wakeup targets | ||
3424 | * | ||
3425 | * The sync wakeup differs that the waker knows that it will schedule | ||
3426 | * away soon, so while the target thread will be woken up, it will not | ||
3427 | * be migrated to another CPU - ie. the two threads are 'synchronized' | ||
3428 | * with each other. This can prevent needless bouncing between CPUs. | ||
3429 | * | ||
3430 | * On UP it can prevent extra preemption. | ||
3431 | * | ||
3432 | * It may be assumed that this function implies a write memory barrier before | ||
3433 | * changing the task state if and only if any tasks are woken up. | ||
3434 | */ | ||
3435 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, | ||
3436 | int nr_exclusive, void *key) | ||
3437 | { | ||
3438 | unsigned long flags; | ||
3439 | int wake_flags = WF_SYNC; | ||
3440 | |||
3441 | if (unlikely(!q)) | ||
3442 | return; | ||
3443 | |||
3444 | if (unlikely(!nr_exclusive)) | ||
3445 | wake_flags = 0; | ||
3446 | |||
3447 | spin_lock_irqsave(&q->lock, flags); | ||
3448 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); | ||
3449 | spin_unlock_irqrestore(&q->lock, flags); | ||
3450 | } | ||
3451 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); | ||
3452 | |||
3453 | /* | ||
3454 | * __wake_up_sync - see __wake_up_sync_key() | ||
3455 | */ | ||
3456 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | ||
3457 | { | ||
3458 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | ||
3459 | } | ||
3460 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | ||
3461 | |||
3462 | /** | ||
3463 | * complete: - signals a single thread waiting on this completion | ||
3464 | * @x: holds the state of this particular completion | ||
3465 | * | ||
3466 | * This will wake up a single thread waiting on this completion. Threads will be | ||
3467 | * awakened in the same order in which they were queued. | ||
3468 | * | ||
3469 | * See also complete_all(), wait_for_completion() and related routines. | ||
3470 | * | ||
3471 | * It may be assumed that this function implies a write memory barrier before | ||
3472 | * changing the task state if and only if any tasks are woken up. | ||
3473 | */ | ||
3474 | void complete(struct completion *x) | ||
3475 | { | ||
3476 | unsigned long flags; | ||
3477 | |||
3478 | spin_lock_irqsave(&x->wait.lock, flags); | ||
3479 | x->done++; | ||
3480 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); | ||
3481 | spin_unlock_irqrestore(&x->wait.lock, flags); | ||
3482 | } | ||
3483 | EXPORT_SYMBOL(complete); | ||
3484 | |||
3485 | /** | ||
3486 | * complete_all: - signals all threads waiting on this completion | ||
3487 | * @x: holds the state of this particular completion | ||
3488 | * | ||
3489 | * This will wake up all threads waiting on this particular completion event. | ||
3490 | * | ||
3491 | * It may be assumed that this function implies a write memory barrier before | ||
3492 | * changing the task state if and only if any tasks are woken up. | ||
3493 | */ | ||
3494 | void complete_all(struct completion *x) | ||
3495 | { | ||
3496 | unsigned long flags; | ||
3497 | |||
3498 | spin_lock_irqsave(&x->wait.lock, flags); | ||
3499 | x->done += UINT_MAX/2; | ||
3500 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); | ||
3501 | spin_unlock_irqrestore(&x->wait.lock, flags); | ||
3502 | } | ||
3503 | EXPORT_SYMBOL(complete_all); | ||
3504 | |||
3505 | static inline long __sched | ||
3506 | do_wait_for_common(struct completion *x, long timeout, int state) | ||
3507 | { | ||
3508 | if (!x->done) { | ||
3509 | DECLARE_WAITQUEUE(wait, current); | ||
3510 | |||
3511 | __add_wait_queue_tail_exclusive(&x->wait, &wait); | ||
3512 | do { | ||
3513 | if (signal_pending_state(state, current)) { | ||
3514 | timeout = -ERESTARTSYS; | ||
3515 | break; | ||
3516 | } | ||
3517 | __set_current_state(state); | ||
3518 | spin_unlock_irq(&x->wait.lock); | ||
3519 | timeout = schedule_timeout(timeout); | ||
3520 | spin_lock_irq(&x->wait.lock); | ||
3521 | } while (!x->done && timeout); | ||
3522 | __remove_wait_queue(&x->wait, &wait); | ||
3523 | if (!x->done) | ||
3524 | return timeout; | ||
3525 | } | ||
3526 | x->done--; | ||
3527 | return timeout ?: 1; | ||
3528 | } | ||
3529 | |||
3530 | static long __sched | ||
3531 | wait_for_common(struct completion *x, long timeout, int state) | ||
3532 | { | ||
3533 | might_sleep(); | ||
3534 | |||
3535 | spin_lock_irq(&x->wait.lock); | ||
3536 | timeout = do_wait_for_common(x, timeout, state); | ||
3537 | spin_unlock_irq(&x->wait.lock); | ||
3538 | return timeout; | ||
3539 | } | ||
3540 | |||
3541 | /** | ||
3542 | * wait_for_completion: - waits for completion of a task | ||
3543 | * @x: holds the state of this particular completion | ||
3544 | * | ||
3545 | * This waits to be signaled for completion of a specific task. It is NOT | ||
3546 | * interruptible and there is no timeout. | ||
3547 | * | ||
3548 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | ||
3549 | * and interrupt capability. Also see complete(). | ||
3550 | */ | ||
3551 | void __sched wait_for_completion(struct completion *x) | ||
3552 | { | ||
3553 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | ||
3554 | } | ||
3555 | EXPORT_SYMBOL(wait_for_completion); | ||
3556 | |||
3557 | /** | ||
3558 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | ||
3559 | * @x: holds the state of this particular completion | ||
3560 | * @timeout: timeout value in jiffies | ||
3561 | * | ||
3562 | * This waits for either a completion of a specific task to be signaled or for a | ||
3563 | * specified timeout to expire. The timeout is in jiffies. It is not | ||
3564 | * interruptible. | ||
3565 | * | ||
3566 | * The return value is 0 if timed out, and positive (at least 1, or number of | ||
3567 | * jiffies left till timeout) if completed. | ||
3568 | */ | ||
3569 | unsigned long __sched | ||
3570 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | ||
3571 | { | ||
3572 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); | ||
3573 | } | ||
3574 | EXPORT_SYMBOL(wait_for_completion_timeout); | ||
3575 | |||
3576 | /** | ||
3577 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | ||
3578 | * @x: holds the state of this particular completion | ||
3579 | * | ||
3580 | * This waits for completion of a specific task to be signaled. It is | ||
3581 | * interruptible. | ||
3582 | * | ||
3583 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | ||
3584 | */ | ||
3585 | int __sched wait_for_completion_interruptible(struct completion *x) | ||
3586 | { | ||
3587 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); | ||
3588 | if (t == -ERESTARTSYS) | ||
3589 | return t; | ||
3590 | return 0; | ||
3591 | } | ||
3592 | EXPORT_SYMBOL(wait_for_completion_interruptible); | ||
3593 | |||
3594 | /** | ||
3595 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | ||
3596 | * @x: holds the state of this particular completion | ||
3597 | * @timeout: timeout value in jiffies | ||
3598 | * | ||
3599 | * This waits for either a completion of a specific task to be signaled or for a | ||
3600 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | ||
3601 | * | ||
3602 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | ||
3603 | * positive (at least 1, or number of jiffies left till timeout) if completed. | ||
3604 | */ | ||
3605 | long __sched | ||
3606 | wait_for_completion_interruptible_timeout(struct completion *x, | ||
3607 | unsigned long timeout) | ||
3608 | { | ||
3609 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); | ||
3610 | } | ||
3611 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | ||
3612 | |||
3613 | /** | ||
3614 | * wait_for_completion_killable: - waits for completion of a task (killable) | ||
3615 | * @x: holds the state of this particular completion | ||
3616 | * | ||
3617 | * This waits to be signaled for completion of a specific task. It can be | ||
3618 | * interrupted by a kill signal. | ||
3619 | * | ||
3620 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | ||
3621 | */ | ||
3622 | int __sched wait_for_completion_killable(struct completion *x) | ||
3623 | { | ||
3624 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | ||
3625 | if (t == -ERESTARTSYS) | ||
3626 | return t; | ||
3627 | return 0; | ||
3628 | } | ||
3629 | EXPORT_SYMBOL(wait_for_completion_killable); | ||
3630 | |||
3631 | /** | ||
3632 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | ||
3633 | * @x: holds the state of this particular completion | ||
3634 | * @timeout: timeout value in jiffies | ||
3635 | * | ||
3636 | * This waits for either a completion of a specific task to be | ||
3637 | * signaled or for a specified timeout to expire. It can be | ||
3638 | * interrupted by a kill signal. The timeout is in jiffies. | ||
3639 | * | ||
3640 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | ||
3641 | * positive (at least 1, or number of jiffies left till timeout) if completed. | ||
3642 | */ | ||
3643 | long __sched | ||
3644 | wait_for_completion_killable_timeout(struct completion *x, | ||
3645 | unsigned long timeout) | ||
3646 | { | ||
3647 | return wait_for_common(x, timeout, TASK_KILLABLE); | ||
3648 | } | ||
3649 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | ||
3650 | |||
3651 | /** | ||
3652 | * try_wait_for_completion - try to decrement a completion without blocking | ||
3653 | * @x: completion structure | ||
3654 | * | ||
3655 | * Returns: 0 if a decrement cannot be done without blocking | ||
3656 | * 1 if a decrement succeeded. | ||
3657 | * | ||
3658 | * If a completion is being used as a counting completion, | ||
3659 | * attempt to decrement the counter without blocking. This | ||
3660 | * enables us to avoid waiting if the resource the completion | ||
3661 | * is protecting is not available. | ||
3662 | */ | ||
3663 | bool try_wait_for_completion(struct completion *x) | ||
3664 | { | ||
3665 | unsigned long flags; | ||
3666 | int ret = 1; | ||
3667 | |||
3668 | spin_lock_irqsave(&x->wait.lock, flags); | ||
3669 | if (!x->done) | ||
3670 | ret = 0; | ||
3671 | else | ||
3672 | x->done--; | ||
3673 | spin_unlock_irqrestore(&x->wait.lock, flags); | ||
3674 | return ret; | ||
3675 | } | ||
3676 | EXPORT_SYMBOL(try_wait_for_completion); | ||
3677 | |||
3678 | /** | ||
3679 | * completion_done - Test to see if a completion has any waiters | ||
3680 | * @x: completion structure | ||
3681 | * | ||
3682 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | ||
3683 | * 1 if there are no waiters. | ||
3684 | * | ||
3685 | */ | ||
3686 | bool completion_done(struct completion *x) | ||
3687 | { | ||
3688 | unsigned long flags; | ||
3689 | int ret = 1; | ||
3690 | |||
3691 | spin_lock_irqsave(&x->wait.lock, flags); | ||
3692 | if (!x->done) | ||
3693 | ret = 0; | ||
3694 | spin_unlock_irqrestore(&x->wait.lock, flags); | ||
3695 | return ret; | ||
3696 | } | ||
3697 | EXPORT_SYMBOL(completion_done); | ||
3698 | |||
3699 | static long __sched | ||
3700 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | ||
3701 | { | ||
3702 | unsigned long flags; | ||
3703 | wait_queue_t wait; | ||
3704 | |||
3705 | init_waitqueue_entry(&wait, current); | ||
3706 | |||
3707 | __set_current_state(state); | ||
3708 | |||
3709 | spin_lock_irqsave(&q->lock, flags); | ||
3710 | __add_wait_queue(q, &wait); | ||
3711 | spin_unlock(&q->lock); | ||
3712 | timeout = schedule_timeout(timeout); | ||
3713 | spin_lock_irq(&q->lock); | ||
3714 | __remove_wait_queue(q, &wait); | ||
3715 | spin_unlock_irqrestore(&q->lock, flags); | ||
3716 | |||
3717 | return timeout; | ||
3718 | } | ||
3719 | |||
3720 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | ||
3721 | { | ||
3722 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | ||
3723 | } | ||
3724 | EXPORT_SYMBOL(interruptible_sleep_on); | ||
3725 | |||
3726 | long __sched | ||
3727 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) | ||
3728 | { | ||
3729 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); | ||
3730 | } | ||
3731 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); | ||
3732 | |||
3733 | void __sched sleep_on(wait_queue_head_t *q) | ||
3734 | { | ||
3735 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | ||
3736 | } | ||
3737 | EXPORT_SYMBOL(sleep_on); | ||
3738 | |||
3739 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) | ||
3740 | { | ||
3741 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); | ||
3742 | } | ||
3743 | EXPORT_SYMBOL(sleep_on_timeout); | ||
3744 | |||
3745 | #ifdef CONFIG_RT_MUTEXES | ||
3746 | |||
3747 | /* | ||
3748 | * rt_mutex_setprio - set the current priority of a task | ||
3749 | * @p: task | ||
3750 | * @prio: prio value (kernel-internal form) | ||
3751 | * | ||
3752 | * This function changes the 'effective' priority of a task. It does | ||
3753 | * not touch ->normal_prio like __setscheduler(). | ||
3754 | * | ||
3755 | * Used by the rt_mutex code to implement priority inheritance logic. | ||
3756 | */ | ||
3757 | void rt_mutex_setprio(struct task_struct *p, int prio) | ||
3758 | { | ||
3759 | int oldprio, on_rq, running; | ||
3760 | struct rq *rq; | ||
3761 | const struct sched_class *prev_class; | ||
3762 | |||
3763 | BUG_ON(prio < 0 || prio > MAX_PRIO); | ||
3764 | |||
3765 | rq = __task_rq_lock(p); | ||
3766 | |||
3767 | trace_sched_pi_setprio(p, prio); | ||
3768 | oldprio = p->prio; | ||
3769 | prev_class = p->sched_class; | ||
3770 | on_rq = p->on_rq; | ||
3771 | running = task_current(rq, p); | ||
3772 | if (on_rq) | ||
3773 | dequeue_task(rq, p, 0); | ||
3774 | if (running) | ||
3775 | p->sched_class->put_prev_task(rq, p); | ||
3776 | |||
3777 | if (rt_prio(prio)) | ||
3778 | p->sched_class = &rt_sched_class; | ||
3779 | else | ||
3780 | p->sched_class = &fair_sched_class; | ||
3781 | |||
3782 | p->prio = prio; | ||
3783 | |||
3784 | if (running) | ||
3785 | p->sched_class->set_curr_task(rq); | ||
3786 | if (on_rq) | ||
3787 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); | ||
3788 | |||
3789 | check_class_changed(rq, p, prev_class, oldprio); | ||
3790 | __task_rq_unlock(rq); | ||
3791 | } | ||
3792 | |||
3793 | #endif | ||
3794 | |||
3795 | void set_user_nice(struct task_struct *p, long nice) | ||
3796 | { | ||
3797 | int old_prio, delta, on_rq; | ||
3798 | unsigned long flags; | ||
3799 | struct rq *rq; | ||
3800 | |||
3801 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | ||
3802 | return; | ||
3803 | /* | ||
3804 | * We have to be careful, if called from sys_setpriority(), | ||
3805 | * the task might be in the middle of scheduling on another CPU. | ||
3806 | */ | ||
3807 | rq = task_rq_lock(p, &flags); | ||
3808 | /* | ||
3809 | * The RT priorities are set via sched_setscheduler(), but we still | ||
3810 | * allow the 'normal' nice value to be set - but as expected | ||
3811 | * it wont have any effect on scheduling until the task is | ||
3812 | * SCHED_FIFO/SCHED_RR: | ||
3813 | */ | ||
3814 | if (task_has_rt_policy(p)) { | ||
3815 | p->static_prio = NICE_TO_PRIO(nice); | ||
3816 | goto out_unlock; | ||
3817 | } | ||
3818 | on_rq = p->on_rq; | ||
3819 | if (on_rq) | ||
3820 | dequeue_task(rq, p, 0); | ||
3821 | |||
3822 | p->static_prio = NICE_TO_PRIO(nice); | ||
3823 | set_load_weight(p); | ||
3824 | old_prio = p->prio; | ||
3825 | p->prio = effective_prio(p); | ||
3826 | delta = p->prio - old_prio; | ||
3827 | |||
3828 | if (on_rq) { | ||
3829 | enqueue_task(rq, p, 0); | ||
3830 | /* | ||
3831 | * If the task increased its priority or is running and | ||
3832 | * lowered its priority, then reschedule its CPU: | ||
3833 | */ | ||
3834 | if (delta < 0 || (delta > 0 && task_running(rq, p))) | ||
3835 | resched_task(rq->curr); | ||
3836 | } | ||
3837 | out_unlock: | ||
3838 | task_rq_unlock(rq, p, &flags); | ||
3839 | } | ||
3840 | EXPORT_SYMBOL(set_user_nice); | ||
3841 | |||
3842 | /* | ||
3843 | * can_nice - check if a task can reduce its nice value | ||
3844 | * @p: task | ||
3845 | * @nice: nice value | ||
3846 | */ | ||
3847 | int can_nice(const struct task_struct *p, const int nice) | ||
3848 | { | ||
3849 | /* convert nice value [19,-20] to rlimit style value [1,40] */ | ||
3850 | int nice_rlim = 20 - nice; | ||
3851 | |||
3852 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || | ||
3853 | capable(CAP_SYS_NICE)); | ||
3854 | } | ||
3855 | |||
3856 | #ifdef __ARCH_WANT_SYS_NICE | ||
3857 | |||
3858 | /* | ||
3859 | * sys_nice - change the priority of the current process. | ||
3860 | * @increment: priority increment | ||
3861 | * | ||
3862 | * sys_setpriority is a more generic, but much slower function that | ||
3863 | * does similar things. | ||
3864 | */ | ||
3865 | SYSCALL_DEFINE1(nice, int, increment) | ||
3866 | { | ||
3867 | long nice, retval; | ||
3868 | |||
3869 | /* | ||
3870 | * Setpriority might change our priority at the same moment. | ||
3871 | * We don't have to worry. Conceptually one call occurs first | ||
3872 | * and we have a single winner. | ||
3873 | */ | ||
3874 | if (increment < -40) | ||
3875 | increment = -40; | ||
3876 | if (increment > 40) | ||
3877 | increment = 40; | ||
3878 | |||
3879 | nice = TASK_NICE(current) + increment; | ||
3880 | if (nice < -20) | ||
3881 | nice = -20; | ||
3882 | if (nice > 19) | ||
3883 | nice = 19; | ||
3884 | |||
3885 | if (increment < 0 && !can_nice(current, nice)) | ||
3886 | return -EPERM; | ||
3887 | |||
3888 | retval = security_task_setnice(current, nice); | ||
3889 | if (retval) | ||
3890 | return retval; | ||
3891 | |||
3892 | set_user_nice(current, nice); | ||
3893 | return 0; | ||
3894 | } | ||
3895 | |||
3896 | #endif | ||
3897 | |||
3898 | /** | ||
3899 | * task_prio - return the priority value of a given task. | ||
3900 | * @p: the task in question. | ||
3901 | * | ||
3902 | * This is the priority value as seen by users in /proc. | ||
3903 | * RT tasks are offset by -200. Normal tasks are centered | ||
3904 | * around 0, value goes from -16 to +15. | ||
3905 | */ | ||
3906 | int task_prio(const struct task_struct *p) | ||
3907 | { | ||
3908 | return p->prio - MAX_RT_PRIO; | ||
3909 | } | ||
3910 | |||
3911 | /** | ||
3912 | * task_nice - return the nice value of a given task. | ||
3913 | * @p: the task in question. | ||
3914 | */ | ||
3915 | int task_nice(const struct task_struct *p) | ||
3916 | { | ||
3917 | return TASK_NICE(p); | ||
3918 | } | ||
3919 | EXPORT_SYMBOL(task_nice); | ||
3920 | |||
3921 | /** | ||
3922 | * idle_cpu - is a given cpu idle currently? | ||
3923 | * @cpu: the processor in question. | ||
3924 | */ | ||
3925 | int idle_cpu(int cpu) | ||
3926 | { | ||
3927 | struct rq *rq = cpu_rq(cpu); | ||
3928 | |||
3929 | if (rq->curr != rq->idle) | ||
3930 | return 0; | ||
3931 | |||
3932 | if (rq->nr_running) | ||
3933 | return 0; | ||
3934 | |||
3935 | #ifdef CONFIG_SMP | ||
3936 | if (!llist_empty(&rq->wake_list)) | ||
3937 | return 0; | ||
3938 | #endif | ||
3939 | |||
3940 | return 1; | ||
3941 | } | ||
3942 | |||
3943 | /** | ||
3944 | * idle_task - return the idle task for a given cpu. | ||
3945 | * @cpu: the processor in question. | ||
3946 | */ | ||
3947 | struct task_struct *idle_task(int cpu) | ||
3948 | { | ||
3949 | return cpu_rq(cpu)->idle; | ||
3950 | } | ||
3951 | |||
3952 | /** | ||
3953 | * find_process_by_pid - find a process with a matching PID value. | ||
3954 | * @pid: the pid in question. | ||
3955 | */ | ||
3956 | static struct task_struct *find_process_by_pid(pid_t pid) | ||
3957 | { | ||
3958 | return pid ? find_task_by_vpid(pid) : current; | ||
3959 | } | ||
3960 | |||
3961 | /* Actually do priority change: must hold rq lock. */ | ||
3962 | static void | ||
3963 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | ||
3964 | { | ||
3965 | p->policy = policy; | ||
3966 | p->rt_priority = prio; | ||
3967 | p->normal_prio = normal_prio(p); | ||
3968 | /* we are holding p->pi_lock already */ | ||
3969 | p->prio = rt_mutex_getprio(p); | ||
3970 | if (rt_prio(p->prio)) | ||
3971 | p->sched_class = &rt_sched_class; | ||
3972 | else | ||
3973 | p->sched_class = &fair_sched_class; | ||
3974 | set_load_weight(p); | ||
3975 | } | ||
3976 | |||
3977 | /* | ||
3978 | * check the target process has a UID that matches the current process's | ||
3979 | */ | ||
3980 | static bool check_same_owner(struct task_struct *p) | ||
3981 | { | ||
3982 | const struct cred *cred = current_cred(), *pcred; | ||
3983 | bool match; | ||
3984 | |||
3985 | rcu_read_lock(); | ||
3986 | pcred = __task_cred(p); | ||
3987 | if (cred->user->user_ns == pcred->user->user_ns) | ||
3988 | match = (cred->euid == pcred->euid || | ||
3989 | cred->euid == pcred->uid); | ||
3990 | else | ||
3991 | match = false; | ||
3992 | rcu_read_unlock(); | ||
3993 | return match; | ||
3994 | } | ||
3995 | |||
3996 | static int __sched_setscheduler(struct task_struct *p, int policy, | ||
3997 | const struct sched_param *param, bool user) | ||
3998 | { | ||
3999 | int retval, oldprio, oldpolicy = -1, on_rq, running; | ||
4000 | unsigned long flags; | ||
4001 | const struct sched_class *prev_class; | ||
4002 | struct rq *rq; | ||
4003 | int reset_on_fork; | ||
4004 | |||
4005 | /* may grab non-irq protected spin_locks */ | ||
4006 | BUG_ON(in_interrupt()); | ||
4007 | recheck: | ||
4008 | /* double check policy once rq lock held */ | ||
4009 | if (policy < 0) { | ||
4010 | reset_on_fork = p->sched_reset_on_fork; | ||
4011 | policy = oldpolicy = p->policy; | ||
4012 | } else { | ||
4013 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | ||
4014 | policy &= ~SCHED_RESET_ON_FORK; | ||
4015 | |||
4016 | if (policy != SCHED_FIFO && policy != SCHED_RR && | ||
4017 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | ||
4018 | policy != SCHED_IDLE) | ||
4019 | return -EINVAL; | ||
4020 | } | ||
4021 | |||
4022 | /* | ||
4023 | * Valid priorities for SCHED_FIFO and SCHED_RR are | ||
4024 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, | ||
4025 | * SCHED_BATCH and SCHED_IDLE is 0. | ||
4026 | */ | ||
4027 | if (param->sched_priority < 0 || | ||
4028 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || | ||
4029 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) | ||
4030 | return -EINVAL; | ||
4031 | if (rt_policy(policy) != (param->sched_priority != 0)) | ||
4032 | return -EINVAL; | ||
4033 | |||
4034 | /* | ||
4035 | * Allow unprivileged RT tasks to decrease priority: | ||
4036 | */ | ||
4037 | if (user && !capable(CAP_SYS_NICE)) { | ||
4038 | if (rt_policy(policy)) { | ||
4039 | unsigned long rlim_rtprio = | ||
4040 | task_rlimit(p, RLIMIT_RTPRIO); | ||
4041 | |||
4042 | /* can't set/change the rt policy */ | ||
4043 | if (policy != p->policy && !rlim_rtprio) | ||
4044 | return -EPERM; | ||
4045 | |||
4046 | /* can't increase priority */ | ||
4047 | if (param->sched_priority > p->rt_priority && | ||
4048 | param->sched_priority > rlim_rtprio) | ||
4049 | return -EPERM; | ||
4050 | } | ||
4051 | |||
4052 | /* | ||
4053 | * Treat SCHED_IDLE as nice 20. Only allow a switch to | ||
4054 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | ||
4055 | */ | ||
4056 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { | ||
4057 | if (!can_nice(p, TASK_NICE(p))) | ||
4058 | return -EPERM; | ||
4059 | } | ||
4060 | |||
4061 | /* can't change other user's priorities */ | ||
4062 | if (!check_same_owner(p)) | ||
4063 | return -EPERM; | ||
4064 | |||
4065 | /* Normal users shall not reset the sched_reset_on_fork flag */ | ||
4066 | if (p->sched_reset_on_fork && !reset_on_fork) | ||
4067 | return -EPERM; | ||
4068 | } | ||
4069 | |||
4070 | if (user) { | ||
4071 | retval = security_task_setscheduler(p); | ||
4072 | if (retval) | ||
4073 | return retval; | ||
4074 | } | ||
4075 | |||
4076 | /* | ||
4077 | * make sure no PI-waiters arrive (or leave) while we are | ||
4078 | * changing the priority of the task: | ||
4079 | * | ||
4080 | * To be able to change p->policy safely, the appropriate | ||
4081 | * runqueue lock must be held. | ||
4082 | */ | ||
4083 | rq = task_rq_lock(p, &flags); | ||
4084 | |||
4085 | /* | ||
4086 | * Changing the policy of the stop threads its a very bad idea | ||
4087 | */ | ||
4088 | if (p == rq->stop) { | ||
4089 | task_rq_unlock(rq, p, &flags); | ||
4090 | return -EINVAL; | ||
4091 | } | ||
4092 | |||
4093 | /* | ||
4094 | * If not changing anything there's no need to proceed further: | ||
4095 | */ | ||
4096 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | ||
4097 | param->sched_priority == p->rt_priority))) { | ||
4098 | |||
4099 | __task_rq_unlock(rq); | ||
4100 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | ||
4101 | return 0; | ||
4102 | } | ||
4103 | |||
4104 | #ifdef CONFIG_RT_GROUP_SCHED | ||
4105 | if (user) { | ||
4106 | /* | ||
4107 | * Do not allow realtime tasks into groups that have no runtime | ||
4108 | * assigned. | ||
4109 | */ | ||
4110 | if (rt_bandwidth_enabled() && rt_policy(policy) && | ||
4111 | task_group(p)->rt_bandwidth.rt_runtime == 0 && | ||
4112 | !task_group_is_autogroup(task_group(p))) { | ||
4113 | task_rq_unlock(rq, p, &flags); | ||
4114 | return -EPERM; | ||
4115 | } | ||
4116 | } | ||
4117 | #endif | ||
4118 | |||
4119 | /* recheck policy now with rq lock held */ | ||
4120 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | ||
4121 | policy = oldpolicy = -1; | ||
4122 | task_rq_unlock(rq, p, &flags); | ||
4123 | goto recheck; | ||
4124 | } | ||
4125 | on_rq = p->on_rq; | ||
4126 | running = task_current(rq, p); | ||
4127 | if (on_rq) | ||
4128 | deactivate_task(rq, p, 0); | ||
4129 | if (running) | ||
4130 | p->sched_class->put_prev_task(rq, p); | ||
4131 | |||
4132 | p->sched_reset_on_fork = reset_on_fork; | ||
4133 | |||
4134 | oldprio = p->prio; | ||
4135 | prev_class = p->sched_class; | ||
4136 | __setscheduler(rq, p, policy, param->sched_priority); | ||
4137 | |||
4138 | if (running) | ||
4139 | p->sched_class->set_curr_task(rq); | ||
4140 | if (on_rq) | ||
4141 | activate_task(rq, p, 0); | ||
4142 | |||
4143 | check_class_changed(rq, p, prev_class, oldprio); | ||
4144 | task_rq_unlock(rq, p, &flags); | ||
4145 | |||
4146 | rt_mutex_adjust_pi(p); | ||
4147 | |||
4148 | return 0; | ||
4149 | } | ||
4150 | |||
4151 | /** | ||
4152 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | ||
4153 | * @p: the task in question. | ||
4154 | * @policy: new policy. | ||
4155 | * @param: structure containing the new RT priority. | ||
4156 | * | ||
4157 | * NOTE that the task may be already dead. | ||
4158 | */ | ||
4159 | int sched_setscheduler(struct task_struct *p, int policy, | ||
4160 | const struct sched_param *param) | ||
4161 | { | ||
4162 | return __sched_setscheduler(p, policy, param, true); | ||
4163 | } | ||
4164 | EXPORT_SYMBOL_GPL(sched_setscheduler); | ||
4165 | |||
4166 | /** | ||
4167 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | ||
4168 | * @p: the task in question. | ||
4169 | * @policy: new policy. | ||
4170 | * @param: structure containing the new RT priority. | ||
4171 | * | ||
4172 | * Just like sched_setscheduler, only don't bother checking if the | ||
4173 | * current context has permission. For example, this is needed in | ||
4174 | * stop_machine(): we create temporary high priority worker threads, | ||
4175 | * but our caller might not have that capability. | ||
4176 | */ | ||
4177 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | ||
4178 | const struct sched_param *param) | ||
4179 | { | ||
4180 | return __sched_setscheduler(p, policy, param, false); | ||
4181 | } | ||
4182 | |||
4183 | static int | ||
4184 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | ||
4185 | { | ||
4186 | struct sched_param lparam; | ||
4187 | struct task_struct *p; | ||
4188 | int retval; | ||
4189 | |||
4190 | if (!param || pid < 0) | ||
4191 | return -EINVAL; | ||
4192 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | ||
4193 | return -EFAULT; | ||
4194 | |||
4195 | rcu_read_lock(); | ||
4196 | retval = -ESRCH; | ||
4197 | p = find_process_by_pid(pid); | ||
4198 | if (p != NULL) | ||
4199 | retval = sched_setscheduler(p, policy, &lparam); | ||
4200 | rcu_read_unlock(); | ||
4201 | |||
4202 | return retval; | ||
4203 | } | ||
4204 | |||
4205 | /** | ||
4206 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | ||
4207 | * @pid: the pid in question. | ||
4208 | * @policy: new policy. | ||
4209 | * @param: structure containing the new RT priority. | ||
4210 | */ | ||
4211 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, | ||
4212 | struct sched_param __user *, param) | ||
4213 | { | ||
4214 | /* negative values for policy are not valid */ | ||
4215 | if (policy < 0) | ||
4216 | return -EINVAL; | ||
4217 | |||
4218 | return do_sched_setscheduler(pid, policy, param); | ||
4219 | } | ||
4220 | |||
4221 | /** | ||
4222 | * sys_sched_setparam - set/change the RT priority of a thread | ||
4223 | * @pid: the pid in question. | ||
4224 | * @param: structure containing the new RT priority. | ||
4225 | */ | ||
4226 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) | ||
4227 | { | ||
4228 | return do_sched_setscheduler(pid, -1, param); | ||
4229 | } | ||
4230 | |||
4231 | /** | ||
4232 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | ||
4233 | * @pid: the pid in question. | ||
4234 | */ | ||
4235 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) | ||
4236 | { | ||
4237 | struct task_struct *p; | ||
4238 | int retval; | ||
4239 | |||
4240 | if (pid < 0) | ||
4241 | return -EINVAL; | ||
4242 | |||
4243 | retval = -ESRCH; | ||
4244 | rcu_read_lock(); | ||
4245 | p = find_process_by_pid(pid); | ||
4246 | if (p) { | ||
4247 | retval = security_task_getscheduler(p); | ||
4248 | if (!retval) | ||
4249 | retval = p->policy | ||
4250 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | ||
4251 | } | ||
4252 | rcu_read_unlock(); | ||
4253 | return retval; | ||
4254 | } | ||
4255 | |||
4256 | /** | ||
4257 | * sys_sched_getparam - get the RT priority of a thread | ||
4258 | * @pid: the pid in question. | ||
4259 | * @param: structure containing the RT priority. | ||
4260 | */ | ||
4261 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) | ||
4262 | { | ||
4263 | struct sched_param lp; | ||
4264 | struct task_struct *p; | ||
4265 | int retval; | ||
4266 | |||
4267 | if (!param || pid < 0) | ||
4268 | return -EINVAL; | ||
4269 | |||
4270 | rcu_read_lock(); | ||
4271 | p = find_process_by_pid(pid); | ||
4272 | retval = -ESRCH; | ||
4273 | if (!p) | ||
4274 | goto out_unlock; | ||
4275 | |||
4276 | retval = security_task_getscheduler(p); | ||
4277 | if (retval) | ||
4278 | goto out_unlock; | ||
4279 | |||
4280 | lp.sched_priority = p->rt_priority; | ||
4281 | rcu_read_unlock(); | ||
4282 | |||
4283 | /* | ||
4284 | * This one might sleep, we cannot do it with a spinlock held ... | ||
4285 | */ | ||
4286 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | ||
4287 | |||
4288 | return retval; | ||
4289 | |||
4290 | out_unlock: | ||
4291 | rcu_read_unlock(); | ||
4292 | return retval; | ||
4293 | } | ||
4294 | |||
4295 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) | ||
4296 | { | ||
4297 | cpumask_var_t cpus_allowed, new_mask; | ||
4298 | struct task_struct *p; | ||
4299 | int retval; | ||
4300 | |||
4301 | get_online_cpus(); | ||
4302 | rcu_read_lock(); | ||
4303 | |||
4304 | p = find_process_by_pid(pid); | ||
4305 | if (!p) { | ||
4306 | rcu_read_unlock(); | ||
4307 | put_online_cpus(); | ||
4308 | return -ESRCH; | ||
4309 | } | ||
4310 | |||
4311 | /* Prevent p going away */ | ||
4312 | get_task_struct(p); | ||
4313 | rcu_read_unlock(); | ||
4314 | |||
4315 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { | ||
4316 | retval = -ENOMEM; | ||
4317 | goto out_put_task; | ||
4318 | } | ||
4319 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | ||
4320 | retval = -ENOMEM; | ||
4321 | goto out_free_cpus_allowed; | ||
4322 | } | ||
4323 | retval = -EPERM; | ||
4324 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) | ||
4325 | goto out_unlock; | ||
4326 | |||
4327 | retval = security_task_setscheduler(p); | ||
4328 | if (retval) | ||
4329 | goto out_unlock; | ||
4330 | |||
4331 | cpuset_cpus_allowed(p, cpus_allowed); | ||
4332 | cpumask_and(new_mask, in_mask, cpus_allowed); | ||
4333 | again: | ||
4334 | retval = set_cpus_allowed_ptr(p, new_mask); | ||
4335 | |||
4336 | if (!retval) { | ||
4337 | cpuset_cpus_allowed(p, cpus_allowed); | ||
4338 | if (!cpumask_subset(new_mask, cpus_allowed)) { | ||
4339 | /* | ||
4340 | * We must have raced with a concurrent cpuset | ||
4341 | * update. Just reset the cpus_allowed to the | ||
4342 | * cpuset's cpus_allowed | ||
4343 | */ | ||
4344 | cpumask_copy(new_mask, cpus_allowed); | ||
4345 | goto again; | ||
4346 | } | ||
4347 | } | ||
4348 | out_unlock: | ||
4349 | free_cpumask_var(new_mask); | ||
4350 | out_free_cpus_allowed: | ||
4351 | free_cpumask_var(cpus_allowed); | ||
4352 | out_put_task: | ||
4353 | put_task_struct(p); | ||
4354 | put_online_cpus(); | ||
4355 | return retval; | ||
4356 | } | ||
4357 | |||
4358 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | ||
4359 | struct cpumask *new_mask) | ||
4360 | { | ||
4361 | if (len < cpumask_size()) | ||
4362 | cpumask_clear(new_mask); | ||
4363 | else if (len > cpumask_size()) | ||
4364 | len = cpumask_size(); | ||
4365 | |||
4366 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | ||
4367 | } | ||
4368 | |||
4369 | /** | ||
4370 | * sys_sched_setaffinity - set the cpu affinity of a process | ||
4371 | * @pid: pid of the process | ||
4372 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | ||
4373 | * @user_mask_ptr: user-space pointer to the new cpu mask | ||
4374 | */ | ||
4375 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, | ||
4376 | unsigned long __user *, user_mask_ptr) | ||
4377 | { | ||
4378 | cpumask_var_t new_mask; | ||
4379 | int retval; | ||
4380 | |||
4381 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) | ||
4382 | return -ENOMEM; | ||
4383 | |||
4384 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); | ||
4385 | if (retval == 0) | ||
4386 | retval = sched_setaffinity(pid, new_mask); | ||
4387 | free_cpumask_var(new_mask); | ||
4388 | return retval; | ||
4389 | } | ||
4390 | |||
4391 | long sched_getaffinity(pid_t pid, struct cpumask *mask) | ||
4392 | { | ||
4393 | struct task_struct *p; | ||
4394 | unsigned long flags; | ||
4395 | int retval; | ||
4396 | |||
4397 | get_online_cpus(); | ||
4398 | rcu_read_lock(); | ||
4399 | |||
4400 | retval = -ESRCH; | ||
4401 | p = find_process_by_pid(pid); | ||
4402 | if (!p) | ||
4403 | goto out_unlock; | ||
4404 | |||
4405 | retval = security_task_getscheduler(p); | ||
4406 | if (retval) | ||
4407 | goto out_unlock; | ||
4408 | |||
4409 | raw_spin_lock_irqsave(&p->pi_lock, flags); | ||
4410 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); | ||
4411 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | ||
4412 | |||
4413 | out_unlock: | ||
4414 | rcu_read_unlock(); | ||
4415 | put_online_cpus(); | ||
4416 | |||
4417 | return retval; | ||
4418 | } | ||
4419 | |||
4420 | /** | ||
4421 | * sys_sched_getaffinity - get the cpu affinity of a process | ||
4422 | * @pid: pid of the process | ||
4423 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | ||
4424 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | ||
4425 | */ | ||
4426 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, | ||
4427 | unsigned long __user *, user_mask_ptr) | ||
4428 | { | ||
4429 | int ret; | ||
4430 | cpumask_var_t mask; | ||
4431 | |||
4432 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) | ||
4433 | return -EINVAL; | ||
4434 | if (len & (sizeof(unsigned long)-1)) | ||
4435 | return -EINVAL; | ||
4436 | |||
4437 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) | ||
4438 | return -ENOMEM; | ||
4439 | |||
4440 | ret = sched_getaffinity(pid, mask); | ||
4441 | if (ret == 0) { | ||
4442 | size_t retlen = min_t(size_t, len, cpumask_size()); | ||
4443 | |||
4444 | if (copy_to_user(user_mask_ptr, mask, retlen)) | ||
4445 | ret = -EFAULT; | ||
4446 | else | ||
4447 | ret = retlen; | ||
4448 | } | ||
4449 | free_cpumask_var(mask); | ||
4450 | |||
4451 | return ret; | ||
4452 | } | ||
4453 | |||
4454 | /** | ||
4455 | * sys_sched_yield - yield the current processor to other threads. | ||
4456 | * | ||
4457 | * This function yields the current CPU to other tasks. If there are no | ||
4458 | * other threads running on this CPU then this function will return. | ||
4459 | */ | ||
4460 | SYSCALL_DEFINE0(sched_yield) | ||
4461 | { | ||
4462 | struct rq *rq = this_rq_lock(); | ||
4463 | |||
4464 | schedstat_inc(rq, yld_count); | ||
4465 | current->sched_class->yield_task(rq); | ||
4466 | |||
4467 | /* | ||
4468 | * Since we are going to call schedule() anyway, there's | ||
4469 | * no need to preempt or enable interrupts: | ||
4470 | */ | ||
4471 | __release(rq->lock); | ||
4472 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); | ||
4473 | do_raw_spin_unlock(&rq->lock); | ||
4474 | preempt_enable_no_resched(); | ||
4475 | |||
4476 | schedule(); | ||
4477 | |||
4478 | return 0; | ||
4479 | } | ||
4480 | |||
4481 | static inline int should_resched(void) | ||
4482 | { | ||
4483 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | ||
4484 | } | ||
4485 | |||
4486 | static void __cond_resched(void) | ||
4487 | { | ||
4488 | add_preempt_count(PREEMPT_ACTIVE); | ||
4489 | __schedule(); | ||
4490 | sub_preempt_count(PREEMPT_ACTIVE); | ||
4491 | } | ||
4492 | |||
4493 | int __sched _cond_resched(void) | ||
4494 | { | ||
4495 | if (should_resched()) { | ||
4496 | __cond_resched(); | ||
4497 | return 1; | ||
4498 | } | ||
4499 | return 0; | ||
4500 | } | ||
4501 | EXPORT_SYMBOL(_cond_resched); | ||
4502 | |||
4503 | /* | ||
4504 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, | ||
4505 | * call schedule, and on return reacquire the lock. | ||
4506 | * | ||
4507 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | ||
4508 | * operations here to prevent schedule() from being called twice (once via | ||
4509 | * spin_unlock(), once by hand). | ||
4510 | */ | ||
4511 | int __cond_resched_lock(spinlock_t *lock) | ||
4512 | { | ||
4513 | int resched = should_resched(); | ||
4514 | int ret = 0; | ||
4515 | |||
4516 | lockdep_assert_held(lock); | ||
4517 | |||
4518 | if (spin_needbreak(lock) || resched) { | ||
4519 | spin_unlock(lock); | ||
4520 | if (resched) | ||
4521 | __cond_resched(); | ||
4522 | else | ||
4523 | cpu_relax(); | ||
4524 | ret = 1; | ||
4525 | spin_lock(lock); | ||
4526 | } | ||
4527 | return ret; | ||
4528 | } | ||
4529 | EXPORT_SYMBOL(__cond_resched_lock); | ||
4530 | |||
4531 | int __sched __cond_resched_softirq(void) | ||
4532 | { | ||
4533 | BUG_ON(!in_softirq()); | ||
4534 | |||
4535 | if (should_resched()) { | ||
4536 | local_bh_enable(); | ||
4537 | __cond_resched(); | ||
4538 | local_bh_disable(); | ||
4539 | return 1; | ||
4540 | } | ||
4541 | return 0; | ||
4542 | } | ||
4543 | EXPORT_SYMBOL(__cond_resched_softirq); | ||
4544 | |||
4545 | /** | ||
4546 | * yield - yield the current processor to other threads. | ||
4547 | * | ||
4548 | * This is a shortcut for kernel-space yielding - it marks the | ||
4549 | * thread runnable and calls sys_sched_yield(). | ||
4550 | */ | ||
4551 | void __sched yield(void) | ||
4552 | { | ||
4553 | set_current_state(TASK_RUNNING); | ||
4554 | sys_sched_yield(); | ||
4555 | } | ||
4556 | EXPORT_SYMBOL(yield); | ||
4557 | |||
4558 | /** | ||
4559 | * yield_to - yield the current processor to another thread in | ||
4560 | * your thread group, or accelerate that thread toward the | ||
4561 | * processor it's on. | ||
4562 | * @p: target task | ||
4563 | * @preempt: whether task preemption is allowed or not | ||
4564 | * | ||
4565 | * It's the caller's job to ensure that the target task struct | ||
4566 | * can't go away on us before we can do any checks. | ||
4567 | * | ||
4568 | * Returns true if we indeed boosted the target task. | ||
4569 | */ | ||
4570 | bool __sched yield_to(struct task_struct *p, bool preempt) | ||
4571 | { | ||
4572 | struct task_struct *curr = current; | ||
4573 | struct rq *rq, *p_rq; | ||
4574 | unsigned long flags; | ||
4575 | bool yielded = 0; | ||
4576 | |||
4577 | local_irq_save(flags); | ||
4578 | rq = this_rq(); | ||
4579 | |||
4580 | again: | ||
4581 | p_rq = task_rq(p); | ||
4582 | double_rq_lock(rq, p_rq); | ||
4583 | while (task_rq(p) != p_rq) { | ||
4584 | double_rq_unlock(rq, p_rq); | ||
4585 | goto again; | ||
4586 | } | ||
4587 | |||
4588 | if (!curr->sched_class->yield_to_task) | ||
4589 | goto out; | ||
4590 | |||
4591 | if (curr->sched_class != p->sched_class) | ||
4592 | goto out; | ||
4593 | |||
4594 | if (task_running(p_rq, p) || p->state) | ||
4595 | goto out; | ||
4596 | |||
4597 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | ||
4598 | if (yielded) { | ||
4599 | schedstat_inc(rq, yld_count); | ||
4600 | /* | ||
4601 | * Make p's CPU reschedule; pick_next_entity takes care of | ||
4602 | * fairness. | ||
4603 | */ | ||
4604 | if (preempt && rq != p_rq) | ||
4605 | resched_task(p_rq->curr); | ||
4606 | } else { | ||
4607 | /* | ||
4608 | * We might have set it in task_yield_fair(), but are | ||
4609 | * not going to schedule(), so don't want to skip | ||
4610 | * the next update. | ||
4611 | */ | ||
4612 | rq->skip_clock_update = 0; | ||
4613 | } | ||
4614 | |||
4615 | out: | ||
4616 | double_rq_unlock(rq, p_rq); | ||
4617 | local_irq_restore(flags); | ||
4618 | |||
4619 | if (yielded) | ||
4620 | schedule(); | ||
4621 | |||
4622 | return yielded; | ||
4623 | } | ||
4624 | EXPORT_SYMBOL_GPL(yield_to); | ||
4625 | |||
4626 | /* | ||
4627 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | ||
4628 | * that process accounting knows that this is a task in IO wait state. | ||
4629 | */ | ||
4630 | void __sched io_schedule(void) | ||
4631 | { | ||
4632 | struct rq *rq = raw_rq(); | ||
4633 | |||
4634 | delayacct_blkio_start(); | ||
4635 | atomic_inc(&rq->nr_iowait); | ||
4636 | blk_flush_plug(current); | ||
4637 | current->in_iowait = 1; | ||
4638 | schedule(); | ||
4639 | current->in_iowait = 0; | ||
4640 | atomic_dec(&rq->nr_iowait); | ||
4641 | delayacct_blkio_end(); | ||
4642 | } | ||
4643 | EXPORT_SYMBOL(io_schedule); | ||
4644 | |||
4645 | long __sched io_schedule_timeout(long timeout) | ||
4646 | { | ||
4647 | struct rq *rq = raw_rq(); | ||
4648 | long ret; | ||
4649 | |||
4650 | delayacct_blkio_start(); | ||
4651 | atomic_inc(&rq->nr_iowait); | ||
4652 | blk_flush_plug(current); | ||
4653 | current->in_iowait = 1; | ||
4654 | ret = schedule_timeout(timeout); | ||
4655 | current->in_iowait = 0; | ||
4656 | atomic_dec(&rq->nr_iowait); | ||
4657 | delayacct_blkio_end(); | ||
4658 | return ret; | ||
4659 | } | ||
4660 | |||
4661 | /** | ||
4662 | * sys_sched_get_priority_max - return maximum RT priority. | ||
4663 | * @policy: scheduling class. | ||
4664 | * | ||
4665 | * this syscall returns the maximum rt_priority that can be used | ||
4666 | * by a given scheduling class. | ||
4667 | */ | ||
4668 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) | ||
4669 | { | ||
4670 | int ret = -EINVAL; | ||
4671 | |||
4672 | switch (policy) { | ||
4673 | case SCHED_FIFO: | ||
4674 | case SCHED_RR: | ||
4675 | ret = MAX_USER_RT_PRIO-1; | ||
4676 | break; | ||
4677 | case SCHED_NORMAL: | ||
4678 | case SCHED_BATCH: | ||
4679 | case SCHED_IDLE: | ||
4680 | ret = 0; | ||
4681 | break; | ||
4682 | } | ||
4683 | return ret; | ||
4684 | } | ||
4685 | |||
4686 | /** | ||
4687 | * sys_sched_get_priority_min - return minimum RT priority. | ||
4688 | * @policy: scheduling class. | ||
4689 | * | ||
4690 | * this syscall returns the minimum rt_priority that can be used | ||
4691 | * by a given scheduling class. | ||
4692 | */ | ||
4693 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) | ||
4694 | { | ||
4695 | int ret = -EINVAL; | ||
4696 | |||
4697 | switch (policy) { | ||
4698 | case SCHED_FIFO: | ||
4699 | case SCHED_RR: | ||
4700 | ret = 1; | ||
4701 | break; | ||
4702 | case SCHED_NORMAL: | ||
4703 | case SCHED_BATCH: | ||
4704 | case SCHED_IDLE: | ||
4705 | ret = 0; | ||
4706 | } | ||
4707 | return ret; | ||
4708 | } | ||
4709 | |||
4710 | /** | ||
4711 | * sys_sched_rr_get_interval - return the default timeslice of a process. | ||
4712 | * @pid: pid of the process. | ||
4713 | * @interval: userspace pointer to the timeslice value. | ||
4714 | * | ||
4715 | * this syscall writes the default timeslice value of a given process | ||
4716 | * into the user-space timespec buffer. A value of '0' means infinity. | ||
4717 | */ | ||
4718 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, | ||
4719 | struct timespec __user *, interval) | ||
4720 | { | ||
4721 | struct task_struct *p; | ||
4722 | unsigned int time_slice; | ||
4723 | unsigned long flags; | ||
4724 | struct rq *rq; | ||
4725 | int retval; | ||
4726 | struct timespec t; | ||
4727 | |||
4728 | if (pid < 0) | ||
4729 | return -EINVAL; | ||
4730 | |||
4731 | retval = -ESRCH; | ||
4732 | rcu_read_lock(); | ||
4733 | p = find_process_by_pid(pid); | ||
4734 | if (!p) | ||
4735 | goto out_unlock; | ||
4736 | |||
4737 | retval = security_task_getscheduler(p); | ||
4738 | if (retval) | ||
4739 | goto out_unlock; | ||
4740 | |||
4741 | rq = task_rq_lock(p, &flags); | ||
4742 | time_slice = p->sched_class->get_rr_interval(rq, p); | ||
4743 | task_rq_unlock(rq, p, &flags); | ||
4744 | |||
4745 | rcu_read_unlock(); | ||
4746 | jiffies_to_timespec(time_slice, &t); | ||
4747 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | ||
4748 | return retval; | ||
4749 | |||
4750 | out_unlock: | ||
4751 | rcu_read_unlock(); | ||
4752 | return retval; | ||
4753 | } | ||
4754 | |||
4755 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; | ||
4756 | |||
4757 | void sched_show_task(struct task_struct *p) | ||
4758 | { | ||
4759 | unsigned long free = 0; | ||
4760 | unsigned state; | ||
4761 | |||
4762 | state = p->state ? __ffs(p->state) + 1 : 0; | ||
4763 | printk(KERN_INFO "%-15.15s %c", p->comm, | ||
4764 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | ||
4765 | #if BITS_PER_LONG == 32 | ||
4766 | if (state == TASK_RUNNING) | ||
4767 | printk(KERN_CONT " running "); | ||
4768 | else | ||
4769 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); | ||
4770 | #else | ||
4771 | if (state == TASK_RUNNING) | ||
4772 | printk(KERN_CONT " running task "); | ||
4773 | else | ||
4774 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); | ||
4775 | #endif | ||
4776 | #ifdef CONFIG_DEBUG_STACK_USAGE | ||
4777 | free = stack_not_used(p); | ||
4778 | #endif | ||
4779 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, | ||
4780 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), | ||
4781 | (unsigned long)task_thread_info(p)->flags); | ||
4782 | |||
4783 | show_stack(p, NULL); | ||
4784 | } | ||
4785 | |||
4786 | void show_state_filter(unsigned long state_filter) | ||
4787 | { | ||
4788 | struct task_struct *g, *p; | ||
4789 | |||
4790 | #if BITS_PER_LONG == 32 | ||
4791 | printk(KERN_INFO | ||
4792 | " task PC stack pid father\n"); | ||
4793 | #else | ||
4794 | printk(KERN_INFO | ||
4795 | " task PC stack pid father\n"); | ||
4796 | #endif | ||
4797 | rcu_read_lock(); | ||
4798 | do_each_thread(g, p) { | ||
4799 | /* | ||
4800 | * reset the NMI-timeout, listing all files on a slow | ||
4801 | * console might take a lot of time: | ||
4802 | */ | ||
4803 | touch_nmi_watchdog(); | ||
4804 | if (!state_filter || (p->state & state_filter)) | ||
4805 | sched_show_task(p); | ||
4806 | } while_each_thread(g, p); | ||
4807 | |||
4808 | touch_all_softlockup_watchdogs(); | ||
4809 | |||
4810 | #ifdef CONFIG_SCHED_DEBUG | ||
4811 | sysrq_sched_debug_show(); | ||
4812 | #endif | ||
4813 | rcu_read_unlock(); | ||
4814 | /* | ||
4815 | * Only show locks if all tasks are dumped: | ||
4816 | */ | ||
4817 | if (!state_filter) | ||
4818 | debug_show_all_locks(); | ||
4819 | } | ||
4820 | |||
4821 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) | ||
4822 | { | ||
4823 | idle->sched_class = &idle_sched_class; | ||
4824 | } | ||
4825 | |||
4826 | /** | ||
4827 | * init_idle - set up an idle thread for a given CPU | ||
4828 | * @idle: task in question | ||
4829 | * @cpu: cpu the idle task belongs to | ||
4830 | * | ||
4831 | * NOTE: this function does not set the idle thread's NEED_RESCHED | ||
4832 | * flag, to make booting more robust. | ||
4833 | */ | ||
4834 | void __cpuinit init_idle(struct task_struct *idle, int cpu) | ||
4835 | { | ||
4836 | struct rq *rq = cpu_rq(cpu); | ||
4837 | unsigned long flags; | ||
4838 | |||
4839 | raw_spin_lock_irqsave(&rq->lock, flags); | ||
4840 | |||
4841 | __sched_fork(idle); | ||
4842 | idle->state = TASK_RUNNING; | ||
4843 | idle->se.exec_start = sched_clock(); | ||
4844 | |||
4845 | do_set_cpus_allowed(idle, cpumask_of(cpu)); | ||
4846 | /* | ||
4847 | * We're having a chicken and egg problem, even though we are | ||
4848 | * holding rq->lock, the cpu isn't yet set to this cpu so the | ||
4849 | * lockdep check in task_group() will fail. | ||
4850 | * | ||
4851 | * Similar case to sched_fork(). / Alternatively we could | ||
4852 | * use task_rq_lock() here and obtain the other rq->lock. | ||
4853 | * | ||
4854 | * Silence PROVE_RCU | ||
4855 | */ | ||
4856 | rcu_read_lock(); | ||
4857 | __set_task_cpu(idle, cpu); | ||
4858 | rcu_read_unlock(); | ||
4859 | |||
4860 | rq->curr = rq->idle = idle; | ||
4861 | #if defined(CONFIG_SMP) | ||
4862 | idle->on_cpu = 1; | ||
4863 | #endif | ||
4864 | raw_spin_unlock_irqrestore(&rq->lock, flags); | ||
4865 | |||
4866 | /* Set the preempt count _outside_ the spinlocks! */ | ||
4867 | task_thread_info(idle)->preempt_count = 0; | ||
4868 | |||
4869 | /* | ||
4870 | * The idle tasks have their own, simple scheduling class: | ||
4871 | */ | ||
4872 | idle->sched_class = &idle_sched_class; | ||
4873 | ftrace_graph_init_idle_task(idle, cpu); | ||
4874 | #if defined(CONFIG_SMP) | ||
4875 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | ||
4876 | #endif | ||
4877 | } | ||
4878 | |||
4879 | #ifdef CONFIG_SMP | ||
4880 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) | ||
4881 | { | ||
4882 | if (p->sched_class && p->sched_class->set_cpus_allowed) | ||
4883 | p->sched_class->set_cpus_allowed(p, new_mask); | ||
4884 | |||
4885 | cpumask_copy(&p->cpus_allowed, new_mask); | ||
4886 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | ||
4887 | } | ||
4888 | |||
4889 | /* | ||
4890 | * This is how migration works: | ||
4891 | * | ||
4892 | * 1) we invoke migration_cpu_stop() on the target CPU using | ||
4893 | * stop_one_cpu(). | ||
4894 | * 2) stopper starts to run (implicitly forcing the migrated thread | ||
4895 | * off the CPU) | ||
4896 | * 3) it checks whether the migrated task is still in the wrong runqueue. | ||
4897 | * 4) if it's in the wrong runqueue then the migration thread removes | ||
4898 | * it and puts it into the right queue. | ||
4899 | * 5) stopper completes and stop_one_cpu() returns and the migration | ||
4900 | * is done. | ||
4901 | */ | ||
4902 | |||
4903 | /* | ||
4904 | * Change a given task's CPU affinity. Migrate the thread to a | ||
4905 | * proper CPU and schedule it away if the CPU it's executing on | ||
4906 | * is removed from the allowed bitmask. | ||
4907 | * | ||
4908 | * NOTE: the caller must have a valid reference to the task, the | ||
4909 | * task must not exit() & deallocate itself prematurely. The | ||
4910 | * call is not atomic; no spinlocks may be held. | ||
4911 | */ | ||
4912 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | ||
4913 | { | ||
4914 | unsigned long flags; | ||
4915 | struct rq *rq; | ||
4916 | unsigned int dest_cpu; | ||
4917 | int ret = 0; | ||
4918 | |||
4919 | rq = task_rq_lock(p, &flags); | ||
4920 | |||
4921 | if (cpumask_equal(&p->cpus_allowed, new_mask)) | ||
4922 | goto out; | ||
4923 | |||
4924 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { | ||
4925 | ret = -EINVAL; | ||
4926 | goto out; | ||
4927 | } | ||
4928 | |||
4929 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { | ||
4930 | ret = -EINVAL; | ||
4931 | goto out; | ||
4932 | } | ||
4933 | |||
4934 | do_set_cpus_allowed(p, new_mask); | ||
4935 | |||
4936 | /* Can the task run on the task's current CPU? If so, we're done */ | ||
4937 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | ||
4938 | goto out; | ||
4939 | |||
4940 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); | ||
4941 | if (p->on_rq) { | ||
4942 | struct migration_arg arg = { p, dest_cpu }; | ||
4943 | /* Need help from migration thread: drop lock and wait. */ | ||
4944 | task_rq_unlock(rq, p, &flags); | ||
4945 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | ||
4946 | tlb_migrate_finish(p->mm); | ||
4947 | return 0; | ||
4948 | } | ||
4949 | out: | ||
4950 | task_rq_unlock(rq, p, &flags); | ||
4951 | |||
4952 | return ret; | ||
4953 | } | ||
4954 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); | ||
4955 | |||
4956 | /* | ||
4957 | * Move (not current) task off this cpu, onto dest cpu. We're doing | ||
4958 | * this because either it can't run here any more (set_cpus_allowed() | ||
4959 | * away from this CPU, or CPU going down), or because we're | ||
4960 | * attempting to rebalance this task on exec (sched_exec). | ||
4961 | * | ||
4962 | * So we race with normal scheduler movements, but that's OK, as long | ||
4963 | * as the task is no longer on this CPU. | ||
4964 | * | ||
4965 | * Returns non-zero if task was successfully migrated. | ||
4966 | */ | ||
4967 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) | ||
4968 | { | ||
4969 | struct rq *rq_dest, *rq_src; | ||
4970 | int ret = 0; | ||
4971 | |||
4972 | if (unlikely(!cpu_active(dest_cpu))) | ||
4973 | return ret; | ||
4974 | |||
4975 | rq_src = cpu_rq(src_cpu); | ||
4976 | rq_dest = cpu_rq(dest_cpu); | ||
4977 | |||
4978 | raw_spin_lock(&p->pi_lock); | ||
4979 | double_rq_lock(rq_src, rq_dest); | ||
4980 | /* Already moved. */ | ||
4981 | if (task_cpu(p) != src_cpu) | ||
4982 | goto done; | ||
4983 | /* Affinity changed (again). */ | ||
4984 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | ||
4985 | goto fail; | ||
4986 | |||
4987 | /* | ||
4988 | * If we're not on a rq, the next wake-up will ensure we're | ||
4989 | * placed properly. | ||
4990 | */ | ||
4991 | if (p->on_rq) { | ||
4992 | deactivate_task(rq_src, p, 0); | ||
4993 | set_task_cpu(p, dest_cpu); | ||
4994 | activate_task(rq_dest, p, 0); | ||
4995 | check_preempt_curr(rq_dest, p, 0); | ||
4996 | } | ||
4997 | done: | ||
4998 | ret = 1; | ||
4999 | fail: | ||
5000 | double_rq_unlock(rq_src, rq_dest); | ||
5001 | raw_spin_unlock(&p->pi_lock); | ||
5002 | return ret; | ||
5003 | } | ||
5004 | |||
5005 | /* | ||
5006 | * migration_cpu_stop - this will be executed by a highprio stopper thread | ||
5007 | * and performs thread migration by bumping thread off CPU then | ||
5008 | * 'pushing' onto another runqueue. | ||
5009 | */ | ||
5010 | static int migration_cpu_stop(void *data) | ||
5011 | { | ||
5012 | struct migration_arg *arg = data; | ||
5013 | |||
5014 | /* | ||
5015 | * The original target cpu might have gone down and we might | ||
5016 | * be on another cpu but it doesn't matter. | ||
5017 | */ | ||
5018 | local_irq_disable(); | ||
5019 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); | ||
5020 | local_irq_enable(); | ||
5021 | return 0; | ||
5022 | } | ||
5023 | |||
5024 | #ifdef CONFIG_HOTPLUG_CPU | ||
5025 | |||
5026 | /* | ||
5027 | * Ensures that the idle task is using init_mm right before its cpu goes | ||
5028 | * offline. | ||
5029 | */ | ||
5030 | void idle_task_exit(void) | ||
5031 | { | ||
5032 | struct mm_struct *mm = current->active_mm; | ||
5033 | |||
5034 | BUG_ON(cpu_online(smp_processor_id())); | ||
5035 | |||
5036 | if (mm != &init_mm) | ||
5037 | switch_mm(mm, &init_mm, current); | ||
5038 | mmdrop(mm); | ||
5039 | } | ||
5040 | |||
5041 | /* | ||
5042 | * While a dead CPU has no uninterruptible tasks queued at this point, | ||
5043 | * it might still have a nonzero ->nr_uninterruptible counter, because | ||
5044 | * for performance reasons the counter is not stricly tracking tasks to | ||
5045 | * their home CPUs. So we just add the counter to another CPU's counter, | ||
5046 | * to keep the global sum constant after CPU-down: | ||
5047 | */ | ||
5048 | static void migrate_nr_uninterruptible(struct rq *rq_src) | ||
5049 | { | ||
5050 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); | ||
5051 | |||
5052 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | ||
5053 | rq_src->nr_uninterruptible = 0; | ||
5054 | } | ||
5055 | |||
5056 | /* | ||
5057 | * remove the tasks which were accounted by rq from calc_load_tasks. | ||
5058 | */ | ||
5059 | static void calc_global_load_remove(struct rq *rq) | ||
5060 | { | ||
5061 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | ||
5062 | rq->calc_load_active = 0; | ||
5063 | } | ||
5064 | |||
5065 | /* | ||
5066 | * Migrate all tasks from the rq, sleeping tasks will be migrated by | ||
5067 | * try_to_wake_up()->select_task_rq(). | ||
5068 | * | ||
5069 | * Called with rq->lock held even though we'er in stop_machine() and | ||
5070 | * there's no concurrency possible, we hold the required locks anyway | ||
5071 | * because of lock validation efforts. | ||
5072 | */ | ||
5073 | static void migrate_tasks(unsigned int dead_cpu) | ||
5074 | { | ||
5075 | struct rq *rq = cpu_rq(dead_cpu); | ||
5076 | struct task_struct *next, *stop = rq->stop; | ||
5077 | int dest_cpu; | ||
5078 | |||
5079 | /* | ||
5080 | * Fudge the rq selection such that the below task selection loop | ||
5081 | * doesn't get stuck on the currently eligible stop task. | ||
5082 | * | ||
5083 | * We're currently inside stop_machine() and the rq is either stuck | ||
5084 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | ||
5085 | * either way we should never end up calling schedule() until we're | ||
5086 | * done here. | ||
5087 | */ | ||
5088 | rq->stop = NULL; | ||
5089 | |||
5090 | /* Ensure any throttled groups are reachable by pick_next_task */ | ||
5091 | unthrottle_offline_cfs_rqs(rq); | ||
5092 | |||
5093 | for ( ; ; ) { | ||
5094 | /* | ||
5095 | * There's this thread running, bail when that's the only | ||
5096 | * remaining thread. | ||
5097 | */ | ||
5098 | if (rq->nr_running == 1) | ||
5099 | break; | ||
5100 | |||
5101 | next = pick_next_task(rq); | ||
5102 | BUG_ON(!next); | ||
5103 | next->sched_class->put_prev_task(rq, next); | ||
5104 | |||
5105 | /* Find suitable destination for @next, with force if needed. */ | ||
5106 | dest_cpu = select_fallback_rq(dead_cpu, next); | ||
5107 | raw_spin_unlock(&rq->lock); | ||
5108 | |||
5109 | __migrate_task(next, dead_cpu, dest_cpu); | ||
5110 | |||
5111 | raw_spin_lock(&rq->lock); | ||
5112 | } | ||
5113 | |||
5114 | rq->stop = stop; | ||
5115 | } | ||
5116 | |||
5117 | #endif /* CONFIG_HOTPLUG_CPU */ | ||
5118 | |||
5119 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) | ||
5120 | |||
5121 | static struct ctl_table sd_ctl_dir[] = { | ||
5122 | { | ||
5123 | .procname = "sched_domain", | ||
5124 | .mode = 0555, | ||
5125 | }, | ||
5126 | {} | ||
5127 | }; | ||
5128 | |||
5129 | static struct ctl_table sd_ctl_root[] = { | ||
5130 | { | ||
5131 | .procname = "kernel", | ||
5132 | .mode = 0555, | ||
5133 | .child = sd_ctl_dir, | ||
5134 | }, | ||
5135 | {} | ||
5136 | }; | ||
5137 | |||
5138 | static struct ctl_table *sd_alloc_ctl_entry(int n) | ||
5139 | { | ||
5140 | struct ctl_table *entry = | ||
5141 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); | ||
5142 | |||
5143 | return entry; | ||
5144 | } | ||
5145 | |||
5146 | static void sd_free_ctl_entry(struct ctl_table **tablep) | ||
5147 | { | ||
5148 | struct ctl_table *entry; | ||
5149 | |||
5150 | /* | ||
5151 | * In the intermediate directories, both the child directory and | ||
5152 | * procname are dynamically allocated and could fail but the mode | ||
5153 | * will always be set. In the lowest directory the names are | ||
5154 | * static strings and all have proc handlers. | ||
5155 | */ | ||
5156 | for (entry = *tablep; entry->mode; entry++) { | ||
5157 | if (entry->child) | ||
5158 | sd_free_ctl_entry(&entry->child); | ||
5159 | if (entry->proc_handler == NULL) | ||
5160 | kfree(entry->procname); | ||
5161 | } | ||
5162 | |||
5163 | kfree(*tablep); | ||
5164 | *tablep = NULL; | ||
5165 | } | ||
5166 | |||
5167 | static void | ||
5168 | set_table_entry(struct ctl_table *entry, | ||
5169 | const char *procname, void *data, int maxlen, | ||
5170 | mode_t mode, proc_handler *proc_handler) | ||
5171 | { | ||
5172 | entry->procname = procname; | ||
5173 | entry->data = data; | ||
5174 | entry->maxlen = maxlen; | ||
5175 | entry->mode = mode; | ||
5176 | entry->proc_handler = proc_handler; | ||
5177 | } | ||
5178 | |||
5179 | static struct ctl_table * | ||
5180 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | ||
5181 | { | ||
5182 | struct ctl_table *table = sd_alloc_ctl_entry(13); | ||
5183 | |||
5184 | if (table == NULL) | ||
5185 | return NULL; | ||
5186 | |||
5187 | set_table_entry(&table[0], "min_interval", &sd->min_interval, | ||
5188 | sizeof(long), 0644, proc_doulongvec_minmax); | ||
5189 | set_table_entry(&table[1], "max_interval", &sd->max_interval, | ||
5190 | sizeof(long), 0644, proc_doulongvec_minmax); | ||
5191 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, | ||
5192 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5193 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, | ||
5194 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5195 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, | ||
5196 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5197 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, | ||
5198 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5199 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, | ||
5200 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5201 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, | ||
5202 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5203 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, | ||
5204 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5205 | set_table_entry(&table[9], "cache_nice_tries", | ||
5206 | &sd->cache_nice_tries, | ||
5207 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5208 | set_table_entry(&table[10], "flags", &sd->flags, | ||
5209 | sizeof(int), 0644, proc_dointvec_minmax); | ||
5210 | set_table_entry(&table[11], "name", sd->name, | ||
5211 | CORENAME_MAX_SIZE, 0444, proc_dostring); | ||
5212 | /* &table[12] is terminator */ | ||
5213 | |||
5214 | return table; | ||
5215 | } | ||
5216 | |||
5217 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) | ||
5218 | { | ||
5219 | struct ctl_table *entry, *table; | ||
5220 | struct sched_domain *sd; | ||
5221 | int domain_num = 0, i; | ||
5222 | char buf[32]; | ||
5223 | |||
5224 | for_each_domain(cpu, sd) | ||
5225 | domain_num++; | ||
5226 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | ||
5227 | if (table == NULL) | ||
5228 | return NULL; | ||
5229 | |||
5230 | i = 0; | ||
5231 | for_each_domain(cpu, sd) { | ||
5232 | snprintf(buf, 32, "domain%d", i); | ||
5233 | entry->procname = kstrdup(buf, GFP_KERNEL); | ||
5234 | entry->mode = 0555; | ||
5235 | entry->child = sd_alloc_ctl_domain_table(sd); | ||
5236 | entry++; | ||
5237 | i++; | ||
5238 | } | ||
5239 | return table; | ||
5240 | } | ||
5241 | |||
5242 | static struct ctl_table_header *sd_sysctl_header; | ||
5243 | static void register_sched_domain_sysctl(void) | ||
5244 | { | ||
5245 | int i, cpu_num = num_possible_cpus(); | ||
5246 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | ||
5247 | char buf[32]; | ||
5248 | |||
5249 | WARN_ON(sd_ctl_dir[0].child); | ||
5250 | sd_ctl_dir[0].child = entry; | ||
5251 | |||
5252 | if (entry == NULL) | ||
5253 | return; | ||
5254 | |||
5255 | for_each_possible_cpu(i) { | ||
5256 | snprintf(buf, 32, "cpu%d", i); | ||
5257 | entry->procname = kstrdup(buf, GFP_KERNEL); | ||
5258 | entry->mode = 0555; | ||
5259 | entry->child = sd_alloc_ctl_cpu_table(i); | ||
5260 | entry++; | ||
5261 | } | ||
5262 | |||
5263 | WARN_ON(sd_sysctl_header); | ||
5264 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | ||
5265 | } | ||
5266 | |||
5267 | /* may be called multiple times per register */ | ||
5268 | static void unregister_sched_domain_sysctl(void) | ||
5269 | { | ||
5270 | if (sd_sysctl_header) | ||
5271 | unregister_sysctl_table(sd_sysctl_header); | ||
5272 | sd_sysctl_header = NULL; | ||
5273 | if (sd_ctl_dir[0].child) | ||
5274 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | ||
5275 | } | ||
5276 | #else | ||
5277 | static void register_sched_domain_sysctl(void) | ||
5278 | { | ||
5279 | } | ||
5280 | static void unregister_sched_domain_sysctl(void) | ||
5281 | { | ||
5282 | } | ||
5283 | #endif | ||
5284 | |||
5285 | static void set_rq_online(struct rq *rq) | ||
5286 | { | ||
5287 | if (!rq->online) { | ||
5288 | const struct sched_class *class; | ||
5289 | |||
5290 | cpumask_set_cpu(rq->cpu, rq->rd->online); | ||
5291 | rq->online = 1; | ||
5292 | |||
5293 | for_each_class(class) { | ||
5294 | if (class->rq_online) | ||
5295 | class->rq_online(rq); | ||
5296 | } | ||
5297 | } | ||
5298 | } | ||
5299 | |||
5300 | static void set_rq_offline(struct rq *rq) | ||
5301 | { | ||
5302 | if (rq->online) { | ||
5303 | const struct sched_class *class; | ||
5304 | |||
5305 | for_each_class(class) { | ||
5306 | if (class->rq_offline) | ||
5307 | class->rq_offline(rq); | ||
5308 | } | ||
5309 | |||
5310 | cpumask_clear_cpu(rq->cpu, rq->rd->online); | ||
5311 | rq->online = 0; | ||
5312 | } | ||
5313 | } | ||
5314 | |||
5315 | /* | ||
5316 | * migration_call - callback that gets triggered when a CPU is added. | ||
5317 | * Here we can start up the necessary migration thread for the new CPU. | ||
5318 | */ | ||
5319 | static int __cpuinit | ||
5320 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | ||
5321 | { | ||
5322 | int cpu = (long)hcpu; | ||
5323 | unsigned long flags; | ||
5324 | struct rq *rq = cpu_rq(cpu); | ||
5325 | |||
5326 | switch (action & ~CPU_TASKS_FROZEN) { | ||
5327 | |||
5328 | case CPU_UP_PREPARE: | ||
5329 | rq->calc_load_update = calc_load_update; | ||
5330 | break; | ||
5331 | |||
5332 | case CPU_ONLINE: | ||
5333 | /* Update our root-domain */ | ||
5334 | raw_spin_lock_irqsave(&rq->lock, flags); | ||
5335 | if (rq->rd) { | ||
5336 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | ||
5337 | |||
5338 | set_rq_online(rq); | ||
5339 | } | ||
5340 | raw_spin_unlock_irqrestore(&rq->lock, flags); | ||
5341 | break; | ||
5342 | |||
5343 | #ifdef CONFIG_HOTPLUG_CPU | ||
5344 | case CPU_DYING: | ||
5345 | sched_ttwu_pending(); | ||
5346 | /* Update our root-domain */ | ||
5347 | raw_spin_lock_irqsave(&rq->lock, flags); | ||
5348 | if (rq->rd) { | ||
5349 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | ||
5350 | set_rq_offline(rq); | ||
5351 | } | ||
5352 | migrate_tasks(cpu); | ||
5353 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | ||
5354 | raw_spin_unlock_irqrestore(&rq->lock, flags); | ||
5355 | |||
5356 | migrate_nr_uninterruptible(rq); | ||
5357 | calc_global_load_remove(rq); | ||
5358 | break; | ||
5359 | #endif | ||
5360 | } | ||
5361 | |||
5362 | update_max_interval(); | ||
5363 | |||
5364 | return NOTIFY_OK; | ||
5365 | } | ||
5366 | |||
5367 | /* | ||
5368 | * Register at high priority so that task migration (migrate_all_tasks) | ||
5369 | * happens before everything else. This has to be lower priority than | ||
5370 | * the notifier in the perf_event subsystem, though. | ||
5371 | */ | ||
5372 | static struct notifier_block __cpuinitdata migration_notifier = { | ||
5373 | .notifier_call = migration_call, | ||
5374 | .priority = CPU_PRI_MIGRATION, | ||
5375 | }; | ||
5376 | |||
5377 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, | ||
5378 | unsigned long action, void *hcpu) | ||
5379 | { | ||
5380 | switch (action & ~CPU_TASKS_FROZEN) { | ||
5381 | case CPU_ONLINE: | ||
5382 | case CPU_DOWN_FAILED: | ||
5383 | set_cpu_active((long)hcpu, true); | ||
5384 | return NOTIFY_OK; | ||
5385 | default: | ||
5386 | return NOTIFY_DONE; | ||
5387 | } | ||
5388 | } | ||
5389 | |||
5390 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | ||
5391 | unsigned long action, void *hcpu) | ||
5392 | { | ||
5393 | switch (action & ~CPU_TASKS_FROZEN) { | ||
5394 | case CPU_DOWN_PREPARE: | ||
5395 | set_cpu_active((long)hcpu, false); | ||
5396 | return NOTIFY_OK; | ||
5397 | default: | ||
5398 | return NOTIFY_DONE; | ||
5399 | } | ||
5400 | } | ||
5401 | |||
5402 | static int __init migration_init(void) | ||
5403 | { | ||
5404 | void *cpu = (void *)(long)smp_processor_id(); | ||
5405 | int err; | ||
5406 | |||
5407 | /* Initialize migration for the boot CPU */ | ||
5408 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); | ||
5409 | BUG_ON(err == NOTIFY_BAD); | ||
5410 | migration_call(&migration_notifier, CPU_ONLINE, cpu); | ||
5411 | register_cpu_notifier(&migration_notifier); | ||
5412 | |||
5413 | /* Register cpu active notifiers */ | ||
5414 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | ||
5415 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | ||
5416 | |||
5417 | return 0; | ||
5418 | } | ||
5419 | early_initcall(migration_init); | ||
5420 | #endif | ||
5421 | |||
5422 | #ifdef CONFIG_SMP | ||
5423 | |||
5424 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ | ||
5425 | |||
5426 | #ifdef CONFIG_SCHED_DEBUG | ||
5427 | |||
5428 | static __read_mostly int sched_domain_debug_enabled; | ||
5429 | |||
5430 | static int __init sched_domain_debug_setup(char *str) | ||
5431 | { | ||
5432 | sched_domain_debug_enabled = 1; | ||
5433 | |||
5434 | return 0; | ||
5435 | } | ||
5436 | early_param("sched_debug", sched_domain_debug_setup); | ||
5437 | |||
5438 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, | ||
5439 | struct cpumask *groupmask) | ||
5440 | { | ||
5441 | struct sched_group *group = sd->groups; | ||
5442 | char str[256]; | ||
5443 | |||
5444 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); | ||
5445 | cpumask_clear(groupmask); | ||
5446 | |||
5447 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | ||
5448 | |||
5449 | if (!(sd->flags & SD_LOAD_BALANCE)) { | ||
5450 | printk("does not load-balance\n"); | ||
5451 | if (sd->parent) | ||
5452 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | ||
5453 | " has parent"); | ||
5454 | return -1; | ||
5455 | } | ||
5456 | |||
5457 | printk(KERN_CONT "span %s level %s\n", str, sd->name); | ||
5458 | |||
5459 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { | ||
5460 | printk(KERN_ERR "ERROR: domain->span does not contain " | ||
5461 | "CPU%d\n", cpu); | ||
5462 | } | ||
5463 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { | ||
5464 | printk(KERN_ERR "ERROR: domain->groups does not contain" | ||
5465 | " CPU%d\n", cpu); | ||
5466 | } | ||
5467 | |||
5468 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); | ||
5469 | do { | ||
5470 | if (!group) { | ||
5471 | printk("\n"); | ||
5472 | printk(KERN_ERR "ERROR: group is NULL\n"); | ||
5473 | break; | ||
5474 | } | ||
5475 | |||
5476 | if (!group->sgp->power) { | ||
5477 | printk(KERN_CONT "\n"); | ||
5478 | printk(KERN_ERR "ERROR: domain->cpu_power not " | ||
5479 | "set\n"); | ||
5480 | break; | ||
5481 | } | ||
5482 | |||
5483 | if (!cpumask_weight(sched_group_cpus(group))) { | ||
5484 | printk(KERN_CONT "\n"); | ||
5485 | printk(KERN_ERR "ERROR: empty group\n"); | ||
5486 | break; | ||
5487 | } | ||
5488 | |||
5489 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { | ||
5490 | printk(KERN_CONT "\n"); | ||
5491 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | ||
5492 | break; | ||
5493 | } | ||
5494 | |||
5495 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); | ||
5496 | |||
5497 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); | ||
5498 | |||
5499 | printk(KERN_CONT " %s", str); | ||
5500 | if (group->sgp->power != SCHED_POWER_SCALE) { | ||
5501 | printk(KERN_CONT " (cpu_power = %d)", | ||
5502 | group->sgp->power); | ||
5503 | } | ||
5504 | |||
5505 | group = group->next; | ||
5506 | } while (group != sd->groups); | ||
5507 | printk(KERN_CONT "\n"); | ||
5508 | |||
5509 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) | ||
5510 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | ||
5511 | |||
5512 | if (sd->parent && | ||
5513 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | ||
5514 | printk(KERN_ERR "ERROR: parent span is not a superset " | ||
5515 | "of domain->span\n"); | ||
5516 | return 0; | ||
5517 | } | ||
5518 | |||
5519 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | ||
5520 | { | ||
5521 | int level = 0; | ||
5522 | |||
5523 | if (!sched_domain_debug_enabled) | ||
5524 | return; | ||
5525 | |||
5526 | if (!sd) { | ||
5527 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | ||
5528 | return; | ||
5529 | } | ||
5530 | |||
5531 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); | ||
5532 | |||
5533 | for (;;) { | ||
5534 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) | ||
5535 | break; | ||
5536 | level++; | ||
5537 | sd = sd->parent; | ||
5538 | if (!sd) | ||
5539 | break; | ||
5540 | } | ||
5541 | } | ||
5542 | #else /* !CONFIG_SCHED_DEBUG */ | ||
5543 | # define sched_domain_debug(sd, cpu) do { } while (0) | ||
5544 | #endif /* CONFIG_SCHED_DEBUG */ | ||
5545 | |||
5546 | static int sd_degenerate(struct sched_domain *sd) | ||
5547 | { | ||
5548 | if (cpumask_weight(sched_domain_span(sd)) == 1) | ||
5549 | return 1; | ||
5550 | |||
5551 | /* Following flags need at least 2 groups */ | ||
5552 | if (sd->flags & (SD_LOAD_BALANCE | | ||
5553 | SD_BALANCE_NEWIDLE | | ||
5554 | SD_BALANCE_FORK | | ||
5555 | SD_BALANCE_EXEC | | ||
5556 | SD_SHARE_CPUPOWER | | ||
5557 | SD_SHARE_PKG_RESOURCES)) { | ||
5558 | if (sd->groups != sd->groups->next) | ||
5559 | return 0; | ||
5560 | } | ||
5561 | |||
5562 | /* Following flags don't use groups */ | ||
5563 | if (sd->flags & (SD_WAKE_AFFINE)) | ||
5564 | return 0; | ||
5565 | |||
5566 | return 1; | ||
5567 | } | ||
5568 | |||
5569 | static int | ||
5570 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | ||
5571 | { | ||
5572 | unsigned long cflags = sd->flags, pflags = parent->flags; | ||
5573 | |||
5574 | if (sd_degenerate(parent)) | ||
5575 | return 1; | ||
5576 | |||
5577 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) | ||
5578 | return 0; | ||
5579 | |||
5580 | /* Flags needing groups don't count if only 1 group in parent */ | ||
5581 | if (parent->groups == parent->groups->next) { | ||
5582 | pflags &= ~(SD_LOAD_BALANCE | | ||
5583 | SD_BALANCE_NEWIDLE | | ||
5584 | SD_BALANCE_FORK | | ||
5585 | SD_BALANCE_EXEC | | ||
5586 | SD_SHARE_CPUPOWER | | ||
5587 | SD_SHARE_PKG_RESOURCES); | ||
5588 | if (nr_node_ids == 1) | ||
5589 | pflags &= ~SD_SERIALIZE; | ||
5590 | } | ||
5591 | if (~cflags & pflags) | ||
5592 | return 0; | ||
5593 | |||
5594 | return 1; | ||
5595 | } | ||
5596 | |||
5597 | static void free_rootdomain(struct rcu_head *rcu) | ||
5598 | { | ||
5599 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); | ||
5600 | |||
5601 | cpupri_cleanup(&rd->cpupri); | ||
5602 | free_cpumask_var(rd->rto_mask); | ||
5603 | free_cpumask_var(rd->online); | ||
5604 | free_cpumask_var(rd->span); | ||
5605 | kfree(rd); | ||
5606 | } | ||
5607 | |||
5608 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) | ||
5609 | { | ||
5610 | struct root_domain *old_rd = NULL; | ||
5611 | unsigned long flags; | ||
5612 | |||
5613 | raw_spin_lock_irqsave(&rq->lock, flags); | ||
5614 | |||
5615 | if (rq->rd) { | ||
5616 | old_rd = rq->rd; | ||
5617 | |||
5618 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) | ||
5619 | set_rq_offline(rq); | ||
5620 | |||
5621 | cpumask_clear_cpu(rq->cpu, old_rd->span); | ||
5622 | |||
5623 | /* | ||
5624 | * If we dont want to free the old_rt yet then | ||
5625 | * set old_rd to NULL to skip the freeing later | ||
5626 | * in this function: | ||
5627 | */ | ||
5628 | if (!atomic_dec_and_test(&old_rd->refcount)) | ||
5629 | old_rd = NULL; | ||
5630 | } | ||
5631 | |||
5632 | atomic_inc(&rd->refcount); | ||
5633 | rq->rd = rd; | ||
5634 | |||
5635 | cpumask_set_cpu(rq->cpu, rd->span); | ||
5636 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) | ||
5637 | set_rq_online(rq); | ||
5638 | |||
5639 | raw_spin_unlock_irqrestore(&rq->lock, flags); | ||
5640 | |||
5641 | if (old_rd) | ||
5642 | call_rcu_sched(&old_rd->rcu, free_rootdomain); | ||
5643 | } | ||
5644 | |||
5645 | static int init_rootdomain(struct root_domain *rd) | ||
5646 | { | ||
5647 | memset(rd, 0, sizeof(*rd)); | ||
5648 | |||
5649 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | ||
5650 | goto out; | ||
5651 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) | ||
5652 | goto free_span; | ||
5653 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | ||
5654 | goto free_online; | ||
5655 | |||
5656 | if (cpupri_init(&rd->cpupri) != 0) | ||
5657 | goto free_rto_mask; | ||
5658 | return 0; | ||
5659 | |||
5660 | free_rto_mask: | ||
5661 | free_cpumask_var(rd->rto_mask); | ||
5662 | free_online: | ||
5663 | free_cpumask_var(rd->online); | ||
5664 | free_span: | ||
5665 | free_cpumask_var(rd->span); | ||
5666 | out: | ||
5667 | return -ENOMEM; | ||
5668 | } | ||
5669 | |||
5670 | /* | ||
5671 | * By default the system creates a single root-domain with all cpus as | ||
5672 | * members (mimicking the global state we have today). | ||
5673 | */ | ||
5674 | struct root_domain def_root_domain; | ||
5675 | |||
5676 | static void init_defrootdomain(void) | ||
5677 | { | ||
5678 | init_rootdomain(&def_root_domain); | ||
5679 | |||
5680 | atomic_set(&def_root_domain.refcount, 1); | ||
5681 | } | ||
5682 | |||
5683 | static struct root_domain *alloc_rootdomain(void) | ||
5684 | { | ||
5685 | struct root_domain *rd; | ||
5686 | |||
5687 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | ||
5688 | if (!rd) | ||
5689 | return NULL; | ||
5690 | |||
5691 | if (init_rootdomain(rd) != 0) { | ||
5692 | kfree(rd); | ||
5693 | return NULL; | ||
5694 | } | ||
5695 | |||
5696 | return rd; | ||
5697 | } | ||
5698 | |||
5699 | static void free_sched_groups(struct sched_group *sg, int free_sgp) | ||
5700 | { | ||
5701 | struct sched_group *tmp, *first; | ||
5702 | |||
5703 | if (!sg) | ||
5704 | return; | ||
5705 | |||
5706 | first = sg; | ||
5707 | do { | ||
5708 | tmp = sg->next; | ||
5709 | |||
5710 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | ||
5711 | kfree(sg->sgp); | ||
5712 | |||
5713 | kfree(sg); | ||
5714 | sg = tmp; | ||
5715 | } while (sg != first); | ||
5716 | } | ||
5717 | |||
5718 | static void free_sched_domain(struct rcu_head *rcu) | ||
5719 | { | ||
5720 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | ||
5721 | |||
5722 | /* | ||
5723 | * If its an overlapping domain it has private groups, iterate and | ||
5724 | * nuke them all. | ||
5725 | */ | ||
5726 | if (sd->flags & SD_OVERLAP) { | ||
5727 | free_sched_groups(sd->groups, 1); | ||
5728 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | ||
5729 | kfree(sd->groups->sgp); | ||
5730 | kfree(sd->groups); | ||
5731 | } | ||
5732 | kfree(sd); | ||
5733 | } | ||
5734 | |||
5735 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | ||
5736 | { | ||
5737 | call_rcu(&sd->rcu, free_sched_domain); | ||
5738 | } | ||
5739 | |||
5740 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | ||
5741 | { | ||
5742 | for (; sd; sd = sd->parent) | ||
5743 | destroy_sched_domain(sd, cpu); | ||
5744 | } | ||
5745 | |||
5746 | /* | ||
5747 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | ||
5748 | * hold the hotplug lock. | ||
5749 | */ | ||
5750 | static void | ||
5751 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | ||
5752 | { | ||
5753 | struct rq *rq = cpu_rq(cpu); | ||
5754 | struct sched_domain *tmp; | ||
5755 | |||
5756 | /* Remove the sched domains which do not contribute to scheduling. */ | ||
5757 | for (tmp = sd; tmp; ) { | ||
5758 | struct sched_domain *parent = tmp->parent; | ||
5759 | if (!parent) | ||
5760 | break; | ||
5761 | |||
5762 | if (sd_parent_degenerate(tmp, parent)) { | ||
5763 | tmp->parent = parent->parent; | ||
5764 | if (parent->parent) | ||
5765 | parent->parent->child = tmp; | ||
5766 | destroy_sched_domain(parent, cpu); | ||
5767 | } else | ||
5768 | tmp = tmp->parent; | ||
5769 | } | ||
5770 | |||
5771 | if (sd && sd_degenerate(sd)) { | ||
5772 | tmp = sd; | ||
5773 | sd = sd->parent; | ||
5774 | destroy_sched_domain(tmp, cpu); | ||
5775 | if (sd) | ||
5776 | sd->child = NULL; | ||
5777 | } | ||
5778 | |||
5779 | sched_domain_debug(sd, cpu); | ||
5780 | |||
5781 | rq_attach_root(rq, rd); | ||
5782 | tmp = rq->sd; | ||
5783 | rcu_assign_pointer(rq->sd, sd); | ||
5784 | destroy_sched_domains(tmp, cpu); | ||
5785 | } | ||
5786 | |||
5787 | /* cpus with isolated domains */ | ||
5788 | static cpumask_var_t cpu_isolated_map; | ||
5789 | |||
5790 | /* Setup the mask of cpus configured for isolated domains */ | ||
5791 | static int __init isolated_cpu_setup(char *str) | ||
5792 | { | ||
5793 | alloc_bootmem_cpumask_var(&cpu_isolated_map); | ||
5794 | cpulist_parse(str, cpu_isolated_map); | ||
5795 | return 1; | ||
5796 | } | ||
5797 | |||
5798 | __setup("isolcpus=", isolated_cpu_setup); | ||
5799 | |||
5800 | #ifdef CONFIG_NUMA | ||
5801 | |||
5802 | /** | ||
5803 | * find_next_best_node - find the next node to include in a sched_domain | ||
5804 | * @node: node whose sched_domain we're building | ||
5805 | * @used_nodes: nodes already in the sched_domain | ||
5806 | * | ||
5807 | * Find the next node to include in a given scheduling domain. Simply | ||
5808 | * finds the closest node not already in the @used_nodes map. | ||
5809 | * | ||
5810 | * Should use nodemask_t. | ||
5811 | */ | ||
5812 | static int find_next_best_node(int node, nodemask_t *used_nodes) | ||
5813 | { | ||
5814 | int i, n, val, min_val, best_node = -1; | ||
5815 | |||
5816 | min_val = INT_MAX; | ||
5817 | |||
5818 | for (i = 0; i < nr_node_ids; i++) { | ||
5819 | /* Start at @node */ | ||
5820 | n = (node + i) % nr_node_ids; | ||
5821 | |||
5822 | if (!nr_cpus_node(n)) | ||
5823 | continue; | ||
5824 | |||
5825 | /* Skip already used nodes */ | ||
5826 | if (node_isset(n, *used_nodes)) | ||
5827 | continue; | ||
5828 | |||
5829 | /* Simple min distance search */ | ||
5830 | val = node_distance(node, n); | ||
5831 | |||
5832 | if (val < min_val) { | ||
5833 | min_val = val; | ||
5834 | best_node = n; | ||
5835 | } | ||
5836 | } | ||
5837 | |||
5838 | if (best_node != -1) | ||
5839 | node_set(best_node, *used_nodes); | ||
5840 | return best_node; | ||
5841 | } | ||
5842 | |||
5843 | /** | ||
5844 | * sched_domain_node_span - get a cpumask for a node's sched_domain | ||
5845 | * @node: node whose cpumask we're constructing | ||
5846 | * @span: resulting cpumask | ||
5847 | * | ||
5848 | * Given a node, construct a good cpumask for its sched_domain to span. It | ||
5849 | * should be one that prevents unnecessary balancing, but also spreads tasks | ||
5850 | * out optimally. | ||
5851 | */ | ||
5852 | static void sched_domain_node_span(int node, struct cpumask *span) | ||
5853 | { | ||
5854 | nodemask_t used_nodes; | ||
5855 | int i; | ||
5856 | |||
5857 | cpumask_clear(span); | ||
5858 | nodes_clear(used_nodes); | ||
5859 | |||
5860 | cpumask_or(span, span, cpumask_of_node(node)); | ||
5861 | node_set(node, used_nodes); | ||
5862 | |||
5863 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | ||
5864 | int next_node = find_next_best_node(node, &used_nodes); | ||
5865 | if (next_node < 0) | ||
5866 | break; | ||
5867 | cpumask_or(span, span, cpumask_of_node(next_node)); | ||
5868 | } | ||
5869 | } | ||
5870 | |||
5871 | static const struct cpumask *cpu_node_mask(int cpu) | ||
5872 | { | ||
5873 | lockdep_assert_held(&sched_domains_mutex); | ||
5874 | |||
5875 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | ||
5876 | |||
5877 | return sched_domains_tmpmask; | ||
5878 | } | ||
5879 | |||
5880 | static const struct cpumask *cpu_allnodes_mask(int cpu) | ||
5881 | { | ||
5882 | return cpu_possible_mask; | ||
5883 | } | ||
5884 | #endif /* CONFIG_NUMA */ | ||
5885 | |||
5886 | static const struct cpumask *cpu_cpu_mask(int cpu) | ||
5887 | { | ||
5888 | return cpumask_of_node(cpu_to_node(cpu)); | ||
5889 | } | ||
5890 | |||
5891 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; | ||
5892 | |||
5893 | struct sd_data { | ||
5894 | struct sched_domain **__percpu sd; | ||
5895 | struct sched_group **__percpu sg; | ||
5896 | struct sched_group_power **__percpu sgp; | ||
5897 | }; | ||
5898 | |||
5899 | struct s_data { | ||
5900 | struct sched_domain ** __percpu sd; | ||
5901 | struct root_domain *rd; | ||
5902 | }; | ||
5903 | |||
5904 | enum s_alloc { | ||
5905 | sa_rootdomain, | ||
5906 | sa_sd, | ||
5907 | sa_sd_storage, | ||
5908 | sa_none, | ||
5909 | }; | ||
5910 | |||
5911 | struct sched_domain_topology_level; | ||
5912 | |||
5913 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | ||
5914 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); | ||
5915 | |||
5916 | #define SDTL_OVERLAP 0x01 | ||
5917 | |||
5918 | struct sched_domain_topology_level { | ||
5919 | sched_domain_init_f init; | ||
5920 | sched_domain_mask_f mask; | ||
5921 | int flags; | ||
5922 | struct sd_data data; | ||
5923 | }; | ||
5924 | |||
5925 | static int | ||
5926 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | ||
5927 | { | ||
5928 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | ||
5929 | const struct cpumask *span = sched_domain_span(sd); | ||
5930 | struct cpumask *covered = sched_domains_tmpmask; | ||
5931 | struct sd_data *sdd = sd->private; | ||
5932 | struct sched_domain *child; | ||
5933 | int i; | ||
5934 | |||
5935 | cpumask_clear(covered); | ||
5936 | |||
5937 | for_each_cpu(i, span) { | ||
5938 | struct cpumask *sg_span; | ||
5939 | |||
5940 | if (cpumask_test_cpu(i, covered)) | ||
5941 | continue; | ||
5942 | |||
5943 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | ||
5944 | GFP_KERNEL, cpu_to_node(cpu)); | ||
5945 | |||
5946 | if (!sg) | ||
5947 | goto fail; | ||
5948 | |||
5949 | sg_span = sched_group_cpus(sg); | ||
5950 | |||
5951 | child = *per_cpu_ptr(sdd->sd, i); | ||
5952 | if (child->child) { | ||
5953 | child = child->child; | ||
5954 | cpumask_copy(sg_span, sched_domain_span(child)); | ||
5955 | } else | ||
5956 | cpumask_set_cpu(i, sg_span); | ||
5957 | |||
5958 | cpumask_or(covered, covered, sg_span); | ||
5959 | |||
5960 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | ||
5961 | atomic_inc(&sg->sgp->ref); | ||
5962 | |||
5963 | if (cpumask_test_cpu(cpu, sg_span)) | ||
5964 | groups = sg; | ||
5965 | |||
5966 | if (!first) | ||
5967 | first = sg; | ||
5968 | if (last) | ||
5969 | last->next = sg; | ||
5970 | last = sg; | ||
5971 | last->next = first; | ||
5972 | } | ||
5973 | sd->groups = groups; | ||
5974 | |||
5975 | return 0; | ||
5976 | |||
5977 | fail: | ||
5978 | free_sched_groups(first, 0); | ||
5979 | |||
5980 | return -ENOMEM; | ||
5981 | } | ||
5982 | |||
5983 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) | ||
5984 | { | ||
5985 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); | ||
5986 | struct sched_domain *child = sd->child; | ||
5987 | |||
5988 | if (child) | ||
5989 | cpu = cpumask_first(sched_domain_span(child)); | ||
5990 | |||
5991 | if (sg) { | ||
5992 | *sg = *per_cpu_ptr(sdd->sg, cpu); | ||
5993 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); | ||
5994 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ | ||
5995 | } | ||
5996 | |||
5997 | return cpu; | ||
5998 | } | ||
5999 | |||
6000 | /* | ||
6001 | * build_sched_groups will build a circular linked list of the groups | ||
6002 | * covered by the given span, and will set each group's ->cpumask correctly, | ||
6003 | * and ->cpu_power to 0. | ||
6004 | * | ||
6005 | * Assumes the sched_domain tree is fully constructed | ||
6006 | */ | ||
6007 | static int | ||
6008 | build_sched_groups(struct sched_domain *sd, int cpu) | ||
6009 | { | ||
6010 | struct sched_group *first = NULL, *last = NULL; | ||
6011 | struct sd_data *sdd = sd->private; | ||
6012 | const struct cpumask *span = sched_domain_span(sd); | ||
6013 | struct cpumask *covered; | ||
6014 | int i; | ||
6015 | |||
6016 | get_group(cpu, sdd, &sd->groups); | ||
6017 | atomic_inc(&sd->groups->ref); | ||
6018 | |||
6019 | if (cpu != cpumask_first(sched_domain_span(sd))) | ||
6020 | return 0; | ||
6021 | |||
6022 | lockdep_assert_held(&sched_domains_mutex); | ||
6023 | covered = sched_domains_tmpmask; | ||
6024 | |||
6025 | cpumask_clear(covered); | ||
6026 | |||
6027 | for_each_cpu(i, span) { | ||
6028 | struct sched_group *sg; | ||
6029 | int group = get_group(i, sdd, &sg); | ||
6030 | int j; | ||
6031 | |||
6032 | if (cpumask_test_cpu(i, covered)) | ||
6033 | continue; | ||
6034 | |||
6035 | cpumask_clear(sched_group_cpus(sg)); | ||
6036 | sg->sgp->power = 0; | ||
6037 | |||
6038 | for_each_cpu(j, span) { | ||
6039 | if (get_group(j, sdd, NULL) != group) | ||
6040 | continue; | ||
6041 | |||
6042 | cpumask_set_cpu(j, covered); | ||
6043 | cpumask_set_cpu(j, sched_group_cpus(sg)); | ||
6044 | } | ||
6045 | |||
6046 | if (!first) | ||
6047 | first = sg; | ||
6048 | if (last) | ||
6049 | last->next = sg; | ||
6050 | last = sg; | ||
6051 | } | ||
6052 | last->next = first; | ||
6053 | |||
6054 | return 0; | ||
6055 | } | ||
6056 | |||
6057 | /* | ||
6058 | * Initialize sched groups cpu_power. | ||
6059 | * | ||
6060 | * cpu_power indicates the capacity of sched group, which is used while | ||
6061 | * distributing the load between different sched groups in a sched domain. | ||
6062 | * Typically cpu_power for all the groups in a sched domain will be same unless | ||
6063 | * there are asymmetries in the topology. If there are asymmetries, group | ||
6064 | * having more cpu_power will pickup more load compared to the group having | ||
6065 | * less cpu_power. | ||
6066 | */ | ||
6067 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | ||
6068 | { | ||
6069 | struct sched_group *sg = sd->groups; | ||
6070 | |||
6071 | WARN_ON(!sd || !sg); | ||
6072 | |||
6073 | do { | ||
6074 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | ||
6075 | sg = sg->next; | ||
6076 | } while (sg != sd->groups); | ||
6077 | |||
6078 | if (cpu != group_first_cpu(sg)) | ||
6079 | return; | ||
6080 | |||
6081 | update_group_power(sd, cpu); | ||
6082 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); | ||
6083 | } | ||
6084 | |||
6085 | int __weak arch_sd_sibling_asym_packing(void) | ||
6086 | { | ||
6087 | return 0*SD_ASYM_PACKING; | ||
6088 | } | ||
6089 | |||
6090 | /* | ||
6091 | * Initializers for schedule domains | ||
6092 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | ||
6093 | */ | ||
6094 | |||
6095 | #ifdef CONFIG_SCHED_DEBUG | ||
6096 | # define SD_INIT_NAME(sd, type) sd->name = #type | ||
6097 | #else | ||
6098 | # define SD_INIT_NAME(sd, type) do { } while (0) | ||
6099 | #endif | ||
6100 | |||
6101 | #define SD_INIT_FUNC(type) \ | ||
6102 | static noinline struct sched_domain * \ | ||
6103 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | ||
6104 | { \ | ||
6105 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | ||
6106 | *sd = SD_##type##_INIT; \ | ||
6107 | SD_INIT_NAME(sd, type); \ | ||
6108 | sd->private = &tl->data; \ | ||
6109 | return sd; \ | ||
6110 | } | ||
6111 | |||
6112 | SD_INIT_FUNC(CPU) | ||
6113 | #ifdef CONFIG_NUMA | ||
6114 | SD_INIT_FUNC(ALLNODES) | ||
6115 | SD_INIT_FUNC(NODE) | ||
6116 | #endif | ||
6117 | #ifdef CONFIG_SCHED_SMT | ||
6118 | SD_INIT_FUNC(SIBLING) | ||
6119 | #endif | ||
6120 | #ifdef CONFIG_SCHED_MC | ||
6121 | SD_INIT_FUNC(MC) | ||
6122 | #endif | ||
6123 | #ifdef CONFIG_SCHED_BOOK | ||
6124 | SD_INIT_FUNC(BOOK) | ||
6125 | #endif | ||
6126 | |||
6127 | static int default_relax_domain_level = -1; | ||
6128 | int sched_domain_level_max; | ||
6129 | |||
6130 | static int __init setup_relax_domain_level(char *str) | ||
6131 | { | ||
6132 | unsigned long val; | ||
6133 | |||
6134 | val = simple_strtoul(str, NULL, 0); | ||
6135 | if (val < sched_domain_level_max) | ||
6136 | default_relax_domain_level = val; | ||
6137 | |||
6138 | return 1; | ||
6139 | } | ||
6140 | __setup("relax_domain_level=", setup_relax_domain_level); | ||
6141 | |||
6142 | static void set_domain_attribute(struct sched_domain *sd, | ||
6143 | struct sched_domain_attr *attr) | ||
6144 | { | ||
6145 | int request; | ||
6146 | |||
6147 | if (!attr || attr->relax_domain_level < 0) { | ||
6148 | if (default_relax_domain_level < 0) | ||
6149 | return; | ||
6150 | else | ||
6151 | request = default_relax_domain_level; | ||
6152 | } else | ||
6153 | request = attr->relax_domain_level; | ||
6154 | if (request < sd->level) { | ||
6155 | /* turn off idle balance on this domain */ | ||
6156 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); | ||
6157 | } else { | ||
6158 | /* turn on idle balance on this domain */ | ||
6159 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); | ||
6160 | } | ||
6161 | } | ||
6162 | |||
6163 | static void __sdt_free(const struct cpumask *cpu_map); | ||
6164 | static int __sdt_alloc(const struct cpumask *cpu_map); | ||
6165 | |||
6166 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, | ||
6167 | const struct cpumask *cpu_map) | ||
6168 | { | ||
6169 | switch (what) { | ||
6170 | case sa_rootdomain: | ||
6171 | if (!atomic_read(&d->rd->refcount)) | ||
6172 | free_rootdomain(&d->rd->rcu); /* fall through */ | ||
6173 | case sa_sd: | ||
6174 | free_percpu(d->sd); /* fall through */ | ||
6175 | case sa_sd_storage: | ||
6176 | __sdt_free(cpu_map); /* fall through */ | ||
6177 | case sa_none: | ||
6178 | break; | ||
6179 | } | ||
6180 | } | ||
6181 | |||
6182 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, | ||
6183 | const struct cpumask *cpu_map) | ||
6184 | { | ||
6185 | memset(d, 0, sizeof(*d)); | ||
6186 | |||
6187 | if (__sdt_alloc(cpu_map)) | ||
6188 | return sa_sd_storage; | ||
6189 | d->sd = alloc_percpu(struct sched_domain *); | ||
6190 | if (!d->sd) | ||
6191 | return sa_sd_storage; | ||
6192 | d->rd = alloc_rootdomain(); | ||
6193 | if (!d->rd) | ||
6194 | return sa_sd; | ||
6195 | return sa_rootdomain; | ||
6196 | } | ||
6197 | |||
6198 | /* | ||
6199 | * NULL the sd_data elements we've used to build the sched_domain and | ||
6200 | * sched_group structure so that the subsequent __free_domain_allocs() | ||
6201 | * will not free the data we're using. | ||
6202 | */ | ||
6203 | static void claim_allocations(int cpu, struct sched_domain *sd) | ||
6204 | { | ||
6205 | struct sd_data *sdd = sd->private; | ||
6206 | |||
6207 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | ||
6208 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | ||
6209 | |||
6210 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) | ||
6211 | *per_cpu_ptr(sdd->sg, cpu) = NULL; | ||
6212 | |||
6213 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | ||
6214 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; | ||
6215 | } | ||
6216 | |||
6217 | #ifdef CONFIG_SCHED_SMT | ||
6218 | static const struct cpumask *cpu_smt_mask(int cpu) | ||
6219 | { | ||
6220 | return topology_thread_cpumask(cpu); | ||
6221 | } | ||
6222 | #endif | ||
6223 | |||
6224 | /* | ||
6225 | * Topology list, bottom-up. | ||
6226 | */ | ||
6227 | static struct sched_domain_topology_level default_topology[] = { | ||
6228 | #ifdef CONFIG_SCHED_SMT | ||
6229 | { sd_init_SIBLING, cpu_smt_mask, }, | ||
6230 | #endif | ||
6231 | #ifdef CONFIG_SCHED_MC | ||
6232 | { sd_init_MC, cpu_coregroup_mask, }, | ||
6233 | #endif | ||
6234 | #ifdef CONFIG_SCHED_BOOK | ||
6235 | { sd_init_BOOK, cpu_book_mask, }, | ||
6236 | #endif | ||
6237 | { sd_init_CPU, cpu_cpu_mask, }, | ||
6238 | #ifdef CONFIG_NUMA | ||
6239 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, | ||
6240 | { sd_init_ALLNODES, cpu_allnodes_mask, }, | ||
6241 | #endif | ||
6242 | { NULL, }, | ||
6243 | }; | ||
6244 | |||
6245 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | ||
6246 | |||
6247 | static int __sdt_alloc(const struct cpumask *cpu_map) | ||
6248 | { | ||
6249 | struct sched_domain_topology_level *tl; | ||
6250 | int j; | ||
6251 | |||
6252 | for (tl = sched_domain_topology; tl->init; tl++) { | ||
6253 | struct sd_data *sdd = &tl->data; | ||
6254 | |||
6255 | sdd->sd = alloc_percpu(struct sched_domain *); | ||
6256 | if (!sdd->sd) | ||
6257 | return -ENOMEM; | ||
6258 | |||
6259 | sdd->sg = alloc_percpu(struct sched_group *); | ||
6260 | if (!sdd->sg) | ||
6261 | return -ENOMEM; | ||
6262 | |||
6263 | sdd->sgp = alloc_percpu(struct sched_group_power *); | ||
6264 | if (!sdd->sgp) | ||
6265 | return -ENOMEM; | ||
6266 | |||
6267 | for_each_cpu(j, cpu_map) { | ||
6268 | struct sched_domain *sd; | ||
6269 | struct sched_group *sg; | ||
6270 | struct sched_group_power *sgp; | ||
6271 | |||
6272 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | ||
6273 | GFP_KERNEL, cpu_to_node(j)); | ||
6274 | if (!sd) | ||
6275 | return -ENOMEM; | ||
6276 | |||
6277 | *per_cpu_ptr(sdd->sd, j) = sd; | ||
6278 | |||
6279 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | ||
6280 | GFP_KERNEL, cpu_to_node(j)); | ||
6281 | if (!sg) | ||
6282 | return -ENOMEM; | ||
6283 | |||
6284 | *per_cpu_ptr(sdd->sg, j) = sg; | ||
6285 | |||
6286 | sgp = kzalloc_node(sizeof(struct sched_group_power), | ||
6287 | GFP_KERNEL, cpu_to_node(j)); | ||
6288 | if (!sgp) | ||
6289 | return -ENOMEM; | ||
6290 | |||
6291 | *per_cpu_ptr(sdd->sgp, j) = sgp; | ||
6292 | } | ||
6293 | } | ||
6294 | |||
6295 | return 0; | ||
6296 | } | ||
6297 | |||
6298 | static void __sdt_free(const struct cpumask *cpu_map) | ||
6299 | { | ||
6300 | struct sched_domain_topology_level *tl; | ||
6301 | int j; | ||
6302 | |||
6303 | for (tl = sched_domain_topology; tl->init; tl++) { | ||
6304 | struct sd_data *sdd = &tl->data; | ||
6305 | |||
6306 | for_each_cpu(j, cpu_map) { | ||
6307 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); | ||
6308 | if (sd && (sd->flags & SD_OVERLAP)) | ||
6309 | free_sched_groups(sd->groups, 0); | ||
6310 | kfree(*per_cpu_ptr(sdd->sd, j)); | ||
6311 | kfree(*per_cpu_ptr(sdd->sg, j)); | ||
6312 | kfree(*per_cpu_ptr(sdd->sgp, j)); | ||
6313 | } | ||
6314 | free_percpu(sdd->sd); | ||
6315 | free_percpu(sdd->sg); | ||
6316 | free_percpu(sdd->sgp); | ||
6317 | } | ||
6318 | } | ||
6319 | |||
6320 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, | ||
6321 | struct s_data *d, const struct cpumask *cpu_map, | ||
6322 | struct sched_domain_attr *attr, struct sched_domain *child, | ||
6323 | int cpu) | ||
6324 | { | ||
6325 | struct sched_domain *sd = tl->init(tl, cpu); | ||
6326 | if (!sd) | ||
6327 | return child; | ||
6328 | |||
6329 | set_domain_attribute(sd, attr); | ||
6330 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | ||
6331 | if (child) { | ||
6332 | sd->level = child->level + 1; | ||
6333 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | ||
6334 | child->parent = sd; | ||
6335 | } | ||
6336 | sd->child = child; | ||
6337 | |||
6338 | return sd; | ||
6339 | } | ||
6340 | |||
6341 | /* | ||
6342 | * Build sched domains for a given set of cpus and attach the sched domains | ||
6343 | * to the individual cpus | ||
6344 | */ | ||
6345 | static int build_sched_domains(const struct cpumask *cpu_map, | ||
6346 | struct sched_domain_attr *attr) | ||
6347 | { | ||
6348 | enum s_alloc alloc_state = sa_none; | ||
6349 | struct sched_domain *sd; | ||
6350 | struct s_data d; | ||
6351 | int i, ret = -ENOMEM; | ||
6352 | |||
6353 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); | ||
6354 | if (alloc_state != sa_rootdomain) | ||
6355 | goto error; | ||
6356 | |||
6357 | /* Set up domains for cpus specified by the cpu_map. */ | ||
6358 | for_each_cpu(i, cpu_map) { | ||
6359 | struct sched_domain_topology_level *tl; | ||
6360 | |||
6361 | sd = NULL; | ||
6362 | for (tl = sched_domain_topology; tl->init; tl++) { | ||
6363 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); | ||
6364 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) | ||
6365 | sd->flags |= SD_OVERLAP; | ||
6366 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) | ||
6367 | break; | ||
6368 | } | ||
6369 | |||
6370 | while (sd->child) | ||
6371 | sd = sd->child; | ||
6372 | |||
6373 | *per_cpu_ptr(d.sd, i) = sd; | ||
6374 | } | ||
6375 | |||
6376 | /* Build the groups for the domains */ | ||
6377 | for_each_cpu(i, cpu_map) { | ||
6378 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | ||
6379 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | ||
6380 | if (sd->flags & SD_OVERLAP) { | ||
6381 | if (build_overlap_sched_groups(sd, i)) | ||
6382 | goto error; | ||
6383 | } else { | ||
6384 | if (build_sched_groups(sd, i)) | ||
6385 | goto error; | ||
6386 | } | ||
6387 | } | ||
6388 | } | ||
6389 | |||
6390 | /* Calculate CPU power for physical packages and nodes */ | ||
6391 | for (i = nr_cpumask_bits-1; i >= 0; i--) { | ||
6392 | if (!cpumask_test_cpu(i, cpu_map)) | ||
6393 | continue; | ||
6394 | |||
6395 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | ||
6396 | claim_allocations(i, sd); | ||
6397 | init_sched_groups_power(i, sd); | ||
6398 | } | ||
6399 | } | ||
6400 | |||
6401 | /* Attach the domains */ | ||
6402 | rcu_read_lock(); | ||
6403 | for_each_cpu(i, cpu_map) { | ||
6404 | sd = *per_cpu_ptr(d.sd, i); | ||
6405 | cpu_attach_domain(sd, d.rd, i); | ||
6406 | } | ||
6407 | rcu_read_unlock(); | ||
6408 | |||
6409 | ret = 0; | ||
6410 | error: | ||
6411 | __free_domain_allocs(&d, alloc_state, cpu_map); | ||
6412 | return ret; | ||
6413 | } | ||
6414 | |||
6415 | static cpumask_var_t *doms_cur; /* current sched domains */ | ||
6416 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | ||
6417 | static struct sched_domain_attr *dattr_cur; | ||
6418 | /* attribues of custom domains in 'doms_cur' */ | ||
6419 | |||
6420 | /* | ||
6421 | * Special case: If a kmalloc of a doms_cur partition (array of | ||
6422 | * cpumask) fails, then fallback to a single sched domain, | ||
6423 | * as determined by the single cpumask fallback_doms. | ||
6424 | */ | ||
6425 | static cpumask_var_t fallback_doms; | ||
6426 | |||
6427 | /* | ||
6428 | * arch_update_cpu_topology lets virtualized architectures update the | ||
6429 | * cpu core maps. It is supposed to return 1 if the topology changed | ||
6430 | * or 0 if it stayed the same. | ||
6431 | */ | ||
6432 | int __attribute__((weak)) arch_update_cpu_topology(void) | ||
6433 | { | ||
6434 | return 0; | ||
6435 | } | ||
6436 | |||
6437 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) | ||
6438 | { | ||
6439 | int i; | ||
6440 | cpumask_var_t *doms; | ||
6441 | |||
6442 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | ||
6443 | if (!doms) | ||
6444 | return NULL; | ||
6445 | for (i = 0; i < ndoms; i++) { | ||
6446 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | ||
6447 | free_sched_domains(doms, i); | ||
6448 | return NULL; | ||
6449 | } | ||
6450 | } | ||
6451 | return doms; | ||
6452 | } | ||
6453 | |||
6454 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | ||
6455 | { | ||
6456 | unsigned int i; | ||
6457 | for (i = 0; i < ndoms; i++) | ||
6458 | free_cpumask_var(doms[i]); | ||
6459 | kfree(doms); | ||
6460 | } | ||
6461 | |||
6462 | /* | ||
6463 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | ||
6464 | * For now this just excludes isolated cpus, but could be used to | ||
6465 | * exclude other special cases in the future. | ||
6466 | */ | ||
6467 | static int init_sched_domains(const struct cpumask *cpu_map) | ||
6468 | { | ||
6469 | int err; | ||
6470 | |||
6471 | arch_update_cpu_topology(); | ||
6472 | ndoms_cur = 1; | ||
6473 | doms_cur = alloc_sched_domains(ndoms_cur); | ||
6474 | if (!doms_cur) | ||
6475 | doms_cur = &fallback_doms; | ||
6476 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | ||
6477 | dattr_cur = NULL; | ||
6478 | err = build_sched_domains(doms_cur[0], NULL); | ||
6479 | register_sched_domain_sysctl(); | ||
6480 | |||
6481 | return err; | ||
6482 | } | ||
6483 | |||
6484 | /* | ||
6485 | * Detach sched domains from a group of cpus specified in cpu_map | ||
6486 | * These cpus will now be attached to the NULL domain | ||
6487 | */ | ||
6488 | static void detach_destroy_domains(const struct cpumask *cpu_map) | ||
6489 | { | ||
6490 | int i; | ||
6491 | |||
6492 | rcu_read_lock(); | ||
6493 | for_each_cpu(i, cpu_map) | ||
6494 | cpu_attach_domain(NULL, &def_root_domain, i); | ||
6495 | rcu_read_unlock(); | ||
6496 | } | ||
6497 | |||
6498 | /* handle null as "default" */ | ||
6499 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | ||
6500 | struct sched_domain_attr *new, int idx_new) | ||
6501 | { | ||
6502 | struct sched_domain_attr tmp; | ||
6503 | |||
6504 | /* fast path */ | ||
6505 | if (!new && !cur) | ||
6506 | return 1; | ||
6507 | |||
6508 | tmp = SD_ATTR_INIT; | ||
6509 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | ||
6510 | new ? (new + idx_new) : &tmp, | ||
6511 | sizeof(struct sched_domain_attr)); | ||
6512 | } | ||
6513 | |||
6514 | /* | ||
6515 | * Partition sched domains as specified by the 'ndoms_new' | ||
6516 | * cpumasks in the array doms_new[] of cpumasks. This compares | ||
6517 | * doms_new[] to the current sched domain partitioning, doms_cur[]. | ||
6518 | * It destroys each deleted domain and builds each new domain. | ||
6519 | * | ||
6520 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. | ||
6521 | * The masks don't intersect (don't overlap.) We should setup one | ||
6522 | * sched domain for each mask. CPUs not in any of the cpumasks will | ||
6523 | * not be load balanced. If the same cpumask appears both in the | ||
6524 | * current 'doms_cur' domains and in the new 'doms_new', we can leave | ||
6525 | * it as it is. | ||
6526 | * | ||
6527 | * The passed in 'doms_new' should be allocated using | ||
6528 | * alloc_sched_domains. This routine takes ownership of it and will | ||
6529 | * free_sched_domains it when done with it. If the caller failed the | ||
6530 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | ||
6531 | * and partition_sched_domains() will fallback to the single partition | ||
6532 | * 'fallback_doms', it also forces the domains to be rebuilt. | ||
6533 | * | ||
6534 | * If doms_new == NULL it will be replaced with cpu_online_mask. | ||
6535 | * ndoms_new == 0 is a special case for destroying existing domains, | ||
6536 | * and it will not create the default domain. | ||
6537 | * | ||
6538 | * Call with hotplug lock held | ||
6539 | */ | ||
6540 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], | ||
6541 | struct sched_domain_attr *dattr_new) | ||
6542 | { | ||
6543 | int i, j, n; | ||
6544 | int new_topology; | ||
6545 | |||
6546 | mutex_lock(&sched_domains_mutex); | ||
6547 | |||
6548 | /* always unregister in case we don't destroy any domains */ | ||
6549 | unregister_sched_domain_sysctl(); | ||
6550 | |||
6551 | /* Let architecture update cpu core mappings. */ | ||
6552 | new_topology = arch_update_cpu_topology(); | ||
6553 | |||
6554 | n = doms_new ? ndoms_new : 0; | ||
6555 | |||
6556 | /* Destroy deleted domains */ | ||
6557 | for (i = 0; i < ndoms_cur; i++) { | ||
6558 | for (j = 0; j < n && !new_topology; j++) { | ||
6559 | if (cpumask_equal(doms_cur[i], doms_new[j]) | ||
6560 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | ||
6561 | goto match1; | ||
6562 | } | ||
6563 | /* no match - a current sched domain not in new doms_new[] */ | ||
6564 | detach_destroy_domains(doms_cur[i]); | ||
6565 | match1: | ||
6566 | ; | ||
6567 | } | ||
6568 | |||
6569 | if (doms_new == NULL) { | ||
6570 | ndoms_cur = 0; | ||
6571 | doms_new = &fallback_doms; | ||
6572 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); | ||
6573 | WARN_ON_ONCE(dattr_new); | ||
6574 | } | ||
6575 | |||
6576 | /* Build new domains */ | ||
6577 | for (i = 0; i < ndoms_new; i++) { | ||
6578 | for (j = 0; j < ndoms_cur && !new_topology; j++) { | ||
6579 | if (cpumask_equal(doms_new[i], doms_cur[j]) | ||
6580 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | ||
6581 | goto match2; | ||
6582 | } | ||
6583 | /* no match - add a new doms_new */ | ||
6584 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); | ||
6585 | match2: | ||
6586 | ; | ||
6587 | } | ||
6588 | |||
6589 | /* Remember the new sched domains */ | ||
6590 | if (doms_cur != &fallback_doms) | ||
6591 | free_sched_domains(doms_cur, ndoms_cur); | ||
6592 | kfree(dattr_cur); /* kfree(NULL) is safe */ | ||
6593 | doms_cur = doms_new; | ||
6594 | dattr_cur = dattr_new; | ||
6595 | ndoms_cur = ndoms_new; | ||
6596 | |||
6597 | register_sched_domain_sysctl(); | ||
6598 | |||
6599 | mutex_unlock(&sched_domains_mutex); | ||
6600 | } | ||
6601 | |||
6602 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | ||
6603 | static void reinit_sched_domains(void) | ||
6604 | { | ||
6605 | get_online_cpus(); | ||
6606 | |||
6607 | /* Destroy domains first to force the rebuild */ | ||
6608 | partition_sched_domains(0, NULL, NULL); | ||
6609 | |||
6610 | rebuild_sched_domains(); | ||
6611 | put_online_cpus(); | ||
6612 | } | ||
6613 | |||
6614 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | ||
6615 | { | ||
6616 | unsigned int level = 0; | ||
6617 | |||
6618 | if (sscanf(buf, "%u", &level) != 1) | ||
6619 | return -EINVAL; | ||
6620 | |||
6621 | /* | ||
6622 | * level is always be positive so don't check for | ||
6623 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | ||
6624 | * What happens on 0 or 1 byte write, | ||
6625 | * need to check for count as well? | ||
6626 | */ | ||
6627 | |||
6628 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | ||
6629 | return -EINVAL; | ||
6630 | |||
6631 | if (smt) | ||
6632 | sched_smt_power_savings = level; | ||
6633 | else | ||
6634 | sched_mc_power_savings = level; | ||
6635 | |||
6636 | reinit_sched_domains(); | ||
6637 | |||
6638 | return count; | ||
6639 | } | ||
6640 | |||
6641 | #ifdef CONFIG_SCHED_MC | ||
6642 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, | ||
6643 | struct sysdev_class_attribute *attr, | ||
6644 | char *page) | ||
6645 | { | ||
6646 | return sprintf(page, "%u\n", sched_mc_power_savings); | ||
6647 | } | ||
6648 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, | ||
6649 | struct sysdev_class_attribute *attr, | ||
6650 | const char *buf, size_t count) | ||
6651 | { | ||
6652 | return sched_power_savings_store(buf, count, 0); | ||
6653 | } | ||
6654 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, | ||
6655 | sched_mc_power_savings_show, | ||
6656 | sched_mc_power_savings_store); | ||
6657 | #endif | ||
6658 | |||
6659 | #ifdef CONFIG_SCHED_SMT | ||
6660 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, | ||
6661 | struct sysdev_class_attribute *attr, | ||
6662 | char *page) | ||
6663 | { | ||
6664 | return sprintf(page, "%u\n", sched_smt_power_savings); | ||
6665 | } | ||
6666 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, | ||
6667 | struct sysdev_class_attribute *attr, | ||
6668 | const char *buf, size_t count) | ||
6669 | { | ||
6670 | return sched_power_savings_store(buf, count, 1); | ||
6671 | } | ||
6672 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, | ||
6673 | sched_smt_power_savings_show, | ||
6674 | sched_smt_power_savings_store); | ||
6675 | #endif | ||
6676 | |||
6677 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | ||
6678 | { | ||
6679 | int err = 0; | ||
6680 | |||
6681 | #ifdef CONFIG_SCHED_SMT | ||
6682 | if (smt_capable()) | ||
6683 | err = sysfs_create_file(&cls->kset.kobj, | ||
6684 | &attr_sched_smt_power_savings.attr); | ||
6685 | #endif | ||
6686 | #ifdef CONFIG_SCHED_MC | ||
6687 | if (!err && mc_capable()) | ||
6688 | err = sysfs_create_file(&cls->kset.kobj, | ||
6689 | &attr_sched_mc_power_savings.attr); | ||
6690 | #endif | ||
6691 | return err; | ||
6692 | } | ||
6693 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | ||
6694 | |||
6695 | /* | ||
6696 | * Update cpusets according to cpu_active mask. If cpusets are | ||
6697 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | ||
6698 | * around partition_sched_domains(). | ||
6699 | */ | ||
6700 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, | ||
6701 | void *hcpu) | ||
6702 | { | ||
6703 | switch (action & ~CPU_TASKS_FROZEN) { | ||
6704 | case CPU_ONLINE: | ||
6705 | case CPU_DOWN_FAILED: | ||
6706 | cpuset_update_active_cpus(); | ||
6707 | return NOTIFY_OK; | ||
6708 | default: | ||
6709 | return NOTIFY_DONE; | ||
6710 | } | ||
6711 | } | ||
6712 | |||
6713 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, | ||
6714 | void *hcpu) | ||
6715 | { | ||
6716 | switch (action & ~CPU_TASKS_FROZEN) { | ||
6717 | case CPU_DOWN_PREPARE: | ||
6718 | cpuset_update_active_cpus(); | ||
6719 | return NOTIFY_OK; | ||
6720 | default: | ||
6721 | return NOTIFY_DONE; | ||
6722 | } | ||
6723 | } | ||
6724 | |||
6725 | void __init sched_init_smp(void) | ||
6726 | { | ||
6727 | cpumask_var_t non_isolated_cpus; | ||
6728 | |||
6729 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | ||
6730 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | ||
6731 | |||
6732 | get_online_cpus(); | ||
6733 | mutex_lock(&sched_domains_mutex); | ||
6734 | init_sched_domains(cpu_active_mask); | ||
6735 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | ||
6736 | if (cpumask_empty(non_isolated_cpus)) | ||
6737 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | ||
6738 | mutex_unlock(&sched_domains_mutex); | ||
6739 | put_online_cpus(); | ||
6740 | |||
6741 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); | ||
6742 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | ||
6743 | |||
6744 | /* RT runtime code needs to handle some hotplug events */ | ||
6745 | hotcpu_notifier(update_runtime, 0); | ||
6746 | |||
6747 | init_hrtick(); | ||
6748 | |||
6749 | /* Move init over to a non-isolated CPU */ | ||
6750 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) | ||
6751 | BUG(); | ||
6752 | sched_init_granularity(); | ||
6753 | free_cpumask_var(non_isolated_cpus); | ||
6754 | |||
6755 | init_sched_rt_class(); | ||
6756 | } | ||
6757 | #else | ||
6758 | void __init sched_init_smp(void) | ||
6759 | { | ||
6760 | sched_init_granularity(); | ||
6761 | } | ||
6762 | #endif /* CONFIG_SMP */ | ||
6763 | |||
6764 | const_debug unsigned int sysctl_timer_migration = 1; | ||
6765 | |||
6766 | int in_sched_functions(unsigned long addr) | ||
6767 | { | ||
6768 | return in_lock_functions(addr) || | ||
6769 | (addr >= (unsigned long)__sched_text_start | ||
6770 | && addr < (unsigned long)__sched_text_end); | ||
6771 | } | ||
6772 | |||
6773 | #ifdef CONFIG_CGROUP_SCHED | ||
6774 | struct task_group root_task_group; | ||
6775 | #endif | ||
6776 | |||
6777 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | ||
6778 | |||
6779 | void __init sched_init(void) | ||
6780 | { | ||
6781 | int i, j; | ||
6782 | unsigned long alloc_size = 0, ptr; | ||
6783 | |||
6784 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
6785 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | ||
6786 | #endif | ||
6787 | #ifdef CONFIG_RT_GROUP_SCHED | ||
6788 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | ||
6789 | #endif | ||
6790 | #ifdef CONFIG_CPUMASK_OFFSTACK | ||
6791 | alloc_size += num_possible_cpus() * cpumask_size(); | ||
6792 | #endif | ||
6793 | if (alloc_size) { | ||
6794 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); | ||
6795 | |||
6796 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
6797 | root_task_group.se = (struct sched_entity **)ptr; | ||
6798 | ptr += nr_cpu_ids * sizeof(void **); | ||
6799 | |||
6800 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | ||
6801 | ptr += nr_cpu_ids * sizeof(void **); | ||
6802 | |||
6803 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | ||
6804 | #ifdef CONFIG_RT_GROUP_SCHED | ||
6805 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | ||
6806 | ptr += nr_cpu_ids * sizeof(void **); | ||
6807 | |||
6808 | root_task_group.rt_rq = (struct rt_rq **)ptr; | ||
6809 | ptr += nr_cpu_ids * sizeof(void **); | ||
6810 | |||
6811 | #endif /* CONFIG_RT_GROUP_SCHED */ | ||
6812 | #ifdef CONFIG_CPUMASK_OFFSTACK | ||
6813 | for_each_possible_cpu(i) { | ||
6814 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | ||
6815 | ptr += cpumask_size(); | ||
6816 | } | ||
6817 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | ||
6818 | } | ||
6819 | |||
6820 | #ifdef CONFIG_SMP | ||
6821 | init_defrootdomain(); | ||
6822 | #endif | ||
6823 | |||
6824 | init_rt_bandwidth(&def_rt_bandwidth, | ||
6825 | global_rt_period(), global_rt_runtime()); | ||
6826 | |||
6827 | #ifdef CONFIG_RT_GROUP_SCHED | ||
6828 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | ||
6829 | global_rt_period(), global_rt_runtime()); | ||
6830 | #endif /* CONFIG_RT_GROUP_SCHED */ | ||
6831 | |||
6832 | #ifdef CONFIG_CGROUP_SCHED | ||
6833 | list_add(&root_task_group.list, &task_groups); | ||
6834 | INIT_LIST_HEAD(&root_task_group.children); | ||
6835 | INIT_LIST_HEAD(&root_task_group.siblings); | ||
6836 | autogroup_init(&init_task); | ||
6837 | |||
6838 | #endif /* CONFIG_CGROUP_SCHED */ | ||
6839 | |||
6840 | #ifdef CONFIG_CGROUP_CPUACCT | ||
6841 | root_cpuacct.cpustat = &kernel_cpustat; | ||
6842 | root_cpuacct.cpuusage = alloc_percpu(u64); | ||
6843 | /* Too early, not expected to fail */ | ||
6844 | BUG_ON(!root_cpuacct.cpuusage); | ||
6845 | #endif | ||
6846 | for_each_possible_cpu(i) { | ||
6847 | struct rq *rq; | ||
6848 | |||
6849 | rq = cpu_rq(i); | ||
6850 | raw_spin_lock_init(&rq->lock); | ||
6851 | rq->nr_running = 0; | ||
6852 | rq->calc_load_active = 0; | ||
6853 | rq->calc_load_update = jiffies + LOAD_FREQ; | ||
6854 | init_cfs_rq(&rq->cfs); | ||
6855 | init_rt_rq(&rq->rt, rq); | ||
6856 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
6857 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; | ||
6858 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | ||
6859 | /* | ||
6860 | * How much cpu bandwidth does root_task_group get? | ||
6861 | * | ||
6862 | * In case of task-groups formed thr' the cgroup filesystem, it | ||
6863 | * gets 100% of the cpu resources in the system. This overall | ||
6864 | * system cpu resource is divided among the tasks of | ||
6865 | * root_task_group and its child task-groups in a fair manner, | ||
6866 | * based on each entity's (task or task-group's) weight | ||
6867 | * (se->load.weight). | ||
6868 | * | ||
6869 | * In other words, if root_task_group has 10 tasks of weight | ||
6870 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | ||
6871 | * then A0's share of the cpu resource is: | ||
6872 | * | ||
6873 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | ||
6874 | * | ||
6875 | * We achieve this by letting root_task_group's tasks sit | ||
6876 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | ||
6877 | */ | ||
6878 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | ||
6879 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); | ||
6880 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | ||
6881 | |||
6882 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | ||
6883 | #ifdef CONFIG_RT_GROUP_SCHED | ||
6884 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); | ||
6885 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); | ||
6886 | #endif | ||
6887 | |||
6888 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) | ||
6889 | rq->cpu_load[j] = 0; | ||
6890 | |||
6891 | rq->last_load_update_tick = jiffies; | ||
6892 | |||
6893 | #ifdef CONFIG_SMP | ||
6894 | rq->sd = NULL; | ||
6895 | rq->rd = NULL; | ||
6896 | rq->cpu_power = SCHED_POWER_SCALE; | ||
6897 | rq->post_schedule = 0; | ||
6898 | rq->active_balance = 0; | ||
6899 | rq->next_balance = jiffies; | ||
6900 | rq->push_cpu = 0; | ||
6901 | rq->cpu = i; | ||
6902 | rq->online = 0; | ||
6903 | rq->idle_stamp = 0; | ||
6904 | rq->avg_idle = 2*sysctl_sched_migration_cost; | ||
6905 | rq_attach_root(rq, &def_root_domain); | ||
6906 | #ifdef CONFIG_NO_HZ | ||
6907 | rq->nohz_flags = 0; | ||
6908 | #endif | ||
6909 | #endif | ||
6910 | init_rq_hrtick(rq); | ||
6911 | atomic_set(&rq->nr_iowait, 0); | ||
6912 | } | ||
6913 | |||
6914 | set_load_weight(&init_task); | ||
6915 | |||
6916 | #ifdef CONFIG_PREEMPT_NOTIFIERS | ||
6917 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | ||
6918 | #endif | ||
6919 | |||
6920 | #ifdef CONFIG_RT_MUTEXES | ||
6921 | plist_head_init(&init_task.pi_waiters); | ||
6922 | #endif | ||
6923 | |||
6924 | /* | ||
6925 | * The boot idle thread does lazy MMU switching as well: | ||
6926 | */ | ||
6927 | atomic_inc(&init_mm.mm_count); | ||
6928 | enter_lazy_tlb(&init_mm, current); | ||
6929 | |||
6930 | /* | ||
6931 | * Make us the idle thread. Technically, schedule() should not be | ||
6932 | * called from this thread, however somewhere below it might be, | ||
6933 | * but because we are the idle thread, we just pick up running again | ||
6934 | * when this runqueue becomes "idle". | ||
6935 | */ | ||
6936 | init_idle(current, smp_processor_id()); | ||
6937 | |||
6938 | calc_load_update = jiffies + LOAD_FREQ; | ||
6939 | |||
6940 | /* | ||
6941 | * During early bootup we pretend to be a normal task: | ||
6942 | */ | ||
6943 | current->sched_class = &fair_sched_class; | ||
6944 | |||
6945 | #ifdef CONFIG_SMP | ||
6946 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); | ||
6947 | /* May be allocated at isolcpus cmdline parse time */ | ||
6948 | if (cpu_isolated_map == NULL) | ||
6949 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | ||
6950 | #endif | ||
6951 | init_sched_fair_class(); | ||
6952 | |||
6953 | scheduler_running = 1; | ||
6954 | } | ||
6955 | |||
6956 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP | ||
6957 | static inline int preempt_count_equals(int preempt_offset) | ||
6958 | { | ||
6959 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); | ||
6960 | |||
6961 | return (nested == preempt_offset); | ||
6962 | } | ||
6963 | |||
6964 | void __might_sleep(const char *file, int line, int preempt_offset) | ||
6965 | { | ||
6966 | static unsigned long prev_jiffy; /* ratelimiting */ | ||
6967 | |||
6968 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ | ||
6969 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || | ||
6970 | system_state != SYSTEM_RUNNING || oops_in_progress) | ||
6971 | return; | ||
6972 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | ||
6973 | return; | ||
6974 | prev_jiffy = jiffies; | ||
6975 | |||
6976 | printk(KERN_ERR | ||
6977 | "BUG: sleeping function called from invalid context at %s:%d\n", | ||
6978 | file, line); | ||
6979 | printk(KERN_ERR | ||
6980 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | ||
6981 | in_atomic(), irqs_disabled(), | ||
6982 | current->pid, current->comm); | ||
6983 | |||
6984 | debug_show_held_locks(current); | ||
6985 | if (irqs_disabled()) | ||
6986 | print_irqtrace_events(current); | ||
6987 | dump_stack(); | ||
6988 | } | ||
6989 | EXPORT_SYMBOL(__might_sleep); | ||
6990 | #endif | ||
6991 | |||
6992 | #ifdef CONFIG_MAGIC_SYSRQ | ||
6993 | static void normalize_task(struct rq *rq, struct task_struct *p) | ||
6994 | { | ||
6995 | const struct sched_class *prev_class = p->sched_class; | ||
6996 | int old_prio = p->prio; | ||
6997 | int on_rq; | ||
6998 | |||
6999 | on_rq = p->on_rq; | ||
7000 | if (on_rq) | ||
7001 | deactivate_task(rq, p, 0); | ||
7002 | __setscheduler(rq, p, SCHED_NORMAL, 0); | ||
7003 | if (on_rq) { | ||
7004 | activate_task(rq, p, 0); | ||
7005 | resched_task(rq->curr); | ||
7006 | } | ||
7007 | |||
7008 | check_class_changed(rq, p, prev_class, old_prio); | ||
7009 | } | ||
7010 | |||
7011 | void normalize_rt_tasks(void) | ||
7012 | { | ||
7013 | struct task_struct *g, *p; | ||
7014 | unsigned long flags; | ||
7015 | struct rq *rq; | ||
7016 | |||
7017 | read_lock_irqsave(&tasklist_lock, flags); | ||
7018 | do_each_thread(g, p) { | ||
7019 | /* | ||
7020 | * Only normalize user tasks: | ||
7021 | */ | ||
7022 | if (!p->mm) | ||
7023 | continue; | ||
7024 | |||
7025 | p->se.exec_start = 0; | ||
7026 | #ifdef CONFIG_SCHEDSTATS | ||
7027 | p->se.statistics.wait_start = 0; | ||
7028 | p->se.statistics.sleep_start = 0; | ||
7029 | p->se.statistics.block_start = 0; | ||
7030 | #endif | ||
7031 | |||
7032 | if (!rt_task(p)) { | ||
7033 | /* | ||
7034 | * Renice negative nice level userspace | ||
7035 | * tasks back to 0: | ||
7036 | */ | ||
7037 | if (TASK_NICE(p) < 0 && p->mm) | ||
7038 | set_user_nice(p, 0); | ||
7039 | continue; | ||
7040 | } | ||
7041 | |||
7042 | raw_spin_lock(&p->pi_lock); | ||
7043 | rq = __task_rq_lock(p); | ||
7044 | |||
7045 | normalize_task(rq, p); | ||
7046 | |||
7047 | __task_rq_unlock(rq); | ||
7048 | raw_spin_unlock(&p->pi_lock); | ||
7049 | } while_each_thread(g, p); | ||
7050 | |||
7051 | read_unlock_irqrestore(&tasklist_lock, flags); | ||
7052 | } | ||
7053 | |||
7054 | #endif /* CONFIG_MAGIC_SYSRQ */ | ||
7055 | |||
7056 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) | ||
7057 | /* | ||
7058 | * These functions are only useful for the IA64 MCA handling, or kdb. | ||
7059 | * | ||
7060 | * They can only be called when the whole system has been | ||
7061 | * stopped - every CPU needs to be quiescent, and no scheduling | ||
7062 | * activity can take place. Using them for anything else would | ||
7063 | * be a serious bug, and as a result, they aren't even visible | ||
7064 | * under any other configuration. | ||
7065 | */ | ||
7066 | |||
7067 | /** | ||
7068 | * curr_task - return the current task for a given cpu. | ||
7069 | * @cpu: the processor in question. | ||
7070 | * | ||
7071 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | ||
7072 | */ | ||
7073 | struct task_struct *curr_task(int cpu) | ||
7074 | { | ||
7075 | return cpu_curr(cpu); | ||
7076 | } | ||
7077 | |||
7078 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ | ||
7079 | |||
7080 | #ifdef CONFIG_IA64 | ||
7081 | /** | ||
7082 | * set_curr_task - set the current task for a given cpu. | ||
7083 | * @cpu: the processor in question. | ||
7084 | * @p: the task pointer to set. | ||
7085 | * | ||
7086 | * Description: This function must only be used when non-maskable interrupts | ||
7087 | * are serviced on a separate stack. It allows the architecture to switch the | ||
7088 | * notion of the current task on a cpu in a non-blocking manner. This function | ||
7089 | * must be called with all CPU's synchronized, and interrupts disabled, the | ||
7090 | * and caller must save the original value of the current task (see | ||
7091 | * curr_task() above) and restore that value before reenabling interrupts and | ||
7092 | * re-starting the system. | ||
7093 | * | ||
7094 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | ||
7095 | */ | ||
7096 | void set_curr_task(int cpu, struct task_struct *p) | ||
7097 | { | ||
7098 | cpu_curr(cpu) = p; | ||
7099 | } | ||
7100 | |||
7101 | #endif | ||
7102 | |||
7103 | #ifdef CONFIG_RT_GROUP_SCHED | ||
7104 | #else /* !CONFIG_RT_GROUP_SCHED */ | ||
7105 | #endif /* CONFIG_RT_GROUP_SCHED */ | ||
7106 | |||
7107 | #ifdef CONFIG_CGROUP_SCHED | ||
7108 | /* task_group_lock serializes the addition/removal of task groups */ | ||
7109 | static DEFINE_SPINLOCK(task_group_lock); | ||
7110 | |||
7111 | static void free_sched_group(struct task_group *tg) | ||
7112 | { | ||
7113 | free_fair_sched_group(tg); | ||
7114 | free_rt_sched_group(tg); | ||
7115 | autogroup_free(tg); | ||
7116 | kfree(tg); | ||
7117 | } | ||
7118 | |||
7119 | /* allocate runqueue etc for a new task group */ | ||
7120 | struct task_group *sched_create_group(struct task_group *parent) | ||
7121 | { | ||
7122 | struct task_group *tg; | ||
7123 | unsigned long flags; | ||
7124 | |||
7125 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | ||
7126 | if (!tg) | ||
7127 | return ERR_PTR(-ENOMEM); | ||
7128 | |||
7129 | if (!alloc_fair_sched_group(tg, parent)) | ||
7130 | goto err; | ||
7131 | |||
7132 | if (!alloc_rt_sched_group(tg, parent)) | ||
7133 | goto err; | ||
7134 | |||
7135 | spin_lock_irqsave(&task_group_lock, flags); | ||
7136 | list_add_rcu(&tg->list, &task_groups); | ||
7137 | |||
7138 | WARN_ON(!parent); /* root should already exist */ | ||
7139 | |||
7140 | tg->parent = parent; | ||
7141 | INIT_LIST_HEAD(&tg->children); | ||
7142 | list_add_rcu(&tg->siblings, &parent->children); | ||
7143 | spin_unlock_irqrestore(&task_group_lock, flags); | ||
7144 | |||
7145 | return tg; | ||
7146 | |||
7147 | err: | ||
7148 | free_sched_group(tg); | ||
7149 | return ERR_PTR(-ENOMEM); | ||
7150 | } | ||
7151 | |||
7152 | /* rcu callback to free various structures associated with a task group */ | ||
7153 | static void free_sched_group_rcu(struct rcu_head *rhp) | ||
7154 | { | ||
7155 | /* now it should be safe to free those cfs_rqs */ | ||
7156 | free_sched_group(container_of(rhp, struct task_group, rcu)); | ||
7157 | } | ||
7158 | |||
7159 | /* Destroy runqueue etc associated with a task group */ | ||
7160 | void sched_destroy_group(struct task_group *tg) | ||
7161 | { | ||
7162 | unsigned long flags; | ||
7163 | int i; | ||
7164 | |||
7165 | /* end participation in shares distribution */ | ||
7166 | for_each_possible_cpu(i) | ||
7167 | unregister_fair_sched_group(tg, i); | ||
7168 | |||
7169 | spin_lock_irqsave(&task_group_lock, flags); | ||
7170 | list_del_rcu(&tg->list); | ||
7171 | list_del_rcu(&tg->siblings); | ||
7172 | spin_unlock_irqrestore(&task_group_lock, flags); | ||
7173 | |||
7174 | /* wait for possible concurrent references to cfs_rqs complete */ | ||
7175 | call_rcu(&tg->rcu, free_sched_group_rcu); | ||
7176 | } | ||
7177 | |||
7178 | /* change task's runqueue when it moves between groups. | ||
7179 | * The caller of this function should have put the task in its new group | ||
7180 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | ||
7181 | * reflect its new group. | ||
7182 | */ | ||
7183 | void sched_move_task(struct task_struct *tsk) | ||
7184 | { | ||
7185 | int on_rq, running; | ||
7186 | unsigned long flags; | ||
7187 | struct rq *rq; | ||
7188 | |||
7189 | rq = task_rq_lock(tsk, &flags); | ||
7190 | |||
7191 | running = task_current(rq, tsk); | ||
7192 | on_rq = tsk->on_rq; | ||
7193 | |||
7194 | if (on_rq) | ||
7195 | dequeue_task(rq, tsk, 0); | ||
7196 | if (unlikely(running)) | ||
7197 | tsk->sched_class->put_prev_task(rq, tsk); | ||
7198 | |||
7199 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
7200 | if (tsk->sched_class->task_move_group) | ||
7201 | tsk->sched_class->task_move_group(tsk, on_rq); | ||
7202 | else | ||
7203 | #endif | ||
7204 | set_task_rq(tsk, task_cpu(tsk)); | ||
7205 | |||
7206 | if (unlikely(running)) | ||
7207 | tsk->sched_class->set_curr_task(rq); | ||
7208 | if (on_rq) | ||
7209 | enqueue_task(rq, tsk, 0); | ||
7210 | |||
7211 | task_rq_unlock(rq, tsk, &flags); | ||
7212 | } | ||
7213 | #endif /* CONFIG_CGROUP_SCHED */ | ||
7214 | |||
7215 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
7216 | #endif | ||
7217 | |||
7218 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) | ||
7219 | static unsigned long to_ratio(u64 period, u64 runtime) | ||
7220 | { | ||
7221 | if (runtime == RUNTIME_INF) | ||
7222 | return 1ULL << 20; | ||
7223 | |||
7224 | return div64_u64(runtime << 20, period); | ||
7225 | } | ||
7226 | #endif | ||
7227 | |||
7228 | #ifdef CONFIG_RT_GROUP_SCHED | ||
7229 | /* | ||
7230 | * Ensure that the real time constraints are schedulable. | ||
7231 | */ | ||
7232 | static DEFINE_MUTEX(rt_constraints_mutex); | ||
7233 | |||
7234 | /* Must be called with tasklist_lock held */ | ||
7235 | static inline int tg_has_rt_tasks(struct task_group *tg) | ||
7236 | { | ||
7237 | struct task_struct *g, *p; | ||
7238 | |||
7239 | do_each_thread(g, p) { | ||
7240 | if (rt_task(p) && task_rq(p)->rt.tg == tg) | ||
7241 | return 1; | ||
7242 | } while_each_thread(g, p); | ||
7243 | |||
7244 | return 0; | ||
7245 | } | ||
7246 | |||
7247 | struct rt_schedulable_data { | ||
7248 | struct task_group *tg; | ||
7249 | u64 rt_period; | ||
7250 | u64 rt_runtime; | ||
7251 | }; | ||
7252 | |||
7253 | static int tg_rt_schedulable(struct task_group *tg, void *data) | ||
7254 | { | ||
7255 | struct rt_schedulable_data *d = data; | ||
7256 | struct task_group *child; | ||
7257 | unsigned long total, sum = 0; | ||
7258 | u64 period, runtime; | ||
7259 | |||
7260 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); | ||
7261 | runtime = tg->rt_bandwidth.rt_runtime; | ||
7262 | |||
7263 | if (tg == d->tg) { | ||
7264 | period = d->rt_period; | ||
7265 | runtime = d->rt_runtime; | ||
7266 | } | ||
7267 | |||
7268 | /* | ||
7269 | * Cannot have more runtime than the period. | ||
7270 | */ | ||
7271 | if (runtime > period && runtime != RUNTIME_INF) | ||
7272 | return -EINVAL; | ||
7273 | |||
7274 | /* | ||
7275 | * Ensure we don't starve existing RT tasks. | ||
7276 | */ | ||
7277 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) | ||
7278 | return -EBUSY; | ||
7279 | |||
7280 | total = to_ratio(period, runtime); | ||
7281 | |||
7282 | /* | ||
7283 | * Nobody can have more than the global setting allows. | ||
7284 | */ | ||
7285 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | ||
7286 | return -EINVAL; | ||
7287 | |||
7288 | /* | ||
7289 | * The sum of our children's runtime should not exceed our own. | ||
7290 | */ | ||
7291 | list_for_each_entry_rcu(child, &tg->children, siblings) { | ||
7292 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | ||
7293 | runtime = child->rt_bandwidth.rt_runtime; | ||
7294 | |||
7295 | if (child == d->tg) { | ||
7296 | period = d->rt_period; | ||
7297 | runtime = d->rt_runtime; | ||
7298 | } | ||
7299 | |||
7300 | sum += to_ratio(period, runtime); | ||
7301 | } | ||
7302 | |||
7303 | if (sum > total) | ||
7304 | return -EINVAL; | ||
7305 | |||
7306 | return 0; | ||
7307 | } | ||
7308 | |||
7309 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | ||
7310 | { | ||
7311 | int ret; | ||
7312 | |||
7313 | struct rt_schedulable_data data = { | ||
7314 | .tg = tg, | ||
7315 | .rt_period = period, | ||
7316 | .rt_runtime = runtime, | ||
7317 | }; | ||
7318 | |||
7319 | rcu_read_lock(); | ||
7320 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | ||
7321 | rcu_read_unlock(); | ||
7322 | |||
7323 | return ret; | ||
7324 | } | ||
7325 | |||
7326 | static int tg_set_rt_bandwidth(struct task_group *tg, | ||
7327 | u64 rt_period, u64 rt_runtime) | ||
7328 | { | ||
7329 | int i, err = 0; | ||
7330 | |||
7331 | mutex_lock(&rt_constraints_mutex); | ||
7332 | read_lock(&tasklist_lock); | ||
7333 | err = __rt_schedulable(tg, rt_period, rt_runtime); | ||
7334 | if (err) | ||
7335 | goto unlock; | ||
7336 | |||
7337 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | ||
7338 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); | ||
7339 | tg->rt_bandwidth.rt_runtime = rt_runtime; | ||
7340 | |||
7341 | for_each_possible_cpu(i) { | ||
7342 | struct rt_rq *rt_rq = tg->rt_rq[i]; | ||
7343 | |||
7344 | raw_spin_lock(&rt_rq->rt_runtime_lock); | ||
7345 | rt_rq->rt_runtime = rt_runtime; | ||
7346 | raw_spin_unlock(&rt_rq->rt_runtime_lock); | ||
7347 | } | ||
7348 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | ||
7349 | unlock: | ||
7350 | read_unlock(&tasklist_lock); | ||
7351 | mutex_unlock(&rt_constraints_mutex); | ||
7352 | |||
7353 | return err; | ||
7354 | } | ||
7355 | |||
7356 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) | ||
7357 | { | ||
7358 | u64 rt_runtime, rt_period; | ||
7359 | |||
7360 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | ||
7361 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | ||
7362 | if (rt_runtime_us < 0) | ||
7363 | rt_runtime = RUNTIME_INF; | ||
7364 | |||
7365 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); | ||
7366 | } | ||
7367 | |||
7368 | long sched_group_rt_runtime(struct task_group *tg) | ||
7369 | { | ||
7370 | u64 rt_runtime_us; | ||
7371 | |||
7372 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) | ||
7373 | return -1; | ||
7374 | |||
7375 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; | ||
7376 | do_div(rt_runtime_us, NSEC_PER_USEC); | ||
7377 | return rt_runtime_us; | ||
7378 | } | ||
7379 | |||
7380 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | ||
7381 | { | ||
7382 | u64 rt_runtime, rt_period; | ||
7383 | |||
7384 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | ||
7385 | rt_runtime = tg->rt_bandwidth.rt_runtime; | ||
7386 | |||
7387 | if (rt_period == 0) | ||
7388 | return -EINVAL; | ||
7389 | |||
7390 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); | ||
7391 | } | ||
7392 | |||
7393 | long sched_group_rt_period(struct task_group *tg) | ||
7394 | { | ||
7395 | u64 rt_period_us; | ||
7396 | |||
7397 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | ||
7398 | do_div(rt_period_us, NSEC_PER_USEC); | ||
7399 | return rt_period_us; | ||
7400 | } | ||
7401 | |||
7402 | static int sched_rt_global_constraints(void) | ||
7403 | { | ||
7404 | u64 runtime, period; | ||
7405 | int ret = 0; | ||
7406 | |||
7407 | if (sysctl_sched_rt_period <= 0) | ||
7408 | return -EINVAL; | ||
7409 | |||
7410 | runtime = global_rt_runtime(); | ||
7411 | period = global_rt_period(); | ||
7412 | |||
7413 | /* | ||
7414 | * Sanity check on the sysctl variables. | ||
7415 | */ | ||
7416 | if (runtime > period && runtime != RUNTIME_INF) | ||
7417 | return -EINVAL; | ||
7418 | |||
7419 | mutex_lock(&rt_constraints_mutex); | ||
7420 | read_lock(&tasklist_lock); | ||
7421 | ret = __rt_schedulable(NULL, 0, 0); | ||
7422 | read_unlock(&tasklist_lock); | ||
7423 | mutex_unlock(&rt_constraints_mutex); | ||
7424 | |||
7425 | return ret; | ||
7426 | } | ||
7427 | |||
7428 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | ||
7429 | { | ||
7430 | /* Don't accept realtime tasks when there is no way for them to run */ | ||
7431 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | ||
7432 | return 0; | ||
7433 | |||
7434 | return 1; | ||
7435 | } | ||
7436 | |||
7437 | #else /* !CONFIG_RT_GROUP_SCHED */ | ||
7438 | static int sched_rt_global_constraints(void) | ||
7439 | { | ||
7440 | unsigned long flags; | ||
7441 | int i; | ||
7442 | |||
7443 | if (sysctl_sched_rt_period <= 0) | ||
7444 | return -EINVAL; | ||
7445 | |||
7446 | /* | ||
7447 | * There's always some RT tasks in the root group | ||
7448 | * -- migration, kstopmachine etc.. | ||
7449 | */ | ||
7450 | if (sysctl_sched_rt_runtime == 0) | ||
7451 | return -EBUSY; | ||
7452 | |||
7453 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | ||
7454 | for_each_possible_cpu(i) { | ||
7455 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | ||
7456 | |||
7457 | raw_spin_lock(&rt_rq->rt_runtime_lock); | ||
7458 | rt_rq->rt_runtime = global_rt_runtime(); | ||
7459 | raw_spin_unlock(&rt_rq->rt_runtime_lock); | ||
7460 | } | ||
7461 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | ||
7462 | |||
7463 | return 0; | ||
7464 | } | ||
7465 | #endif /* CONFIG_RT_GROUP_SCHED */ | ||
7466 | |||
7467 | int sched_rt_handler(struct ctl_table *table, int write, | ||
7468 | void __user *buffer, size_t *lenp, | ||
7469 | loff_t *ppos) | ||
7470 | { | ||
7471 | int ret; | ||
7472 | int old_period, old_runtime; | ||
7473 | static DEFINE_MUTEX(mutex); | ||
7474 | |||
7475 | mutex_lock(&mutex); | ||
7476 | old_period = sysctl_sched_rt_period; | ||
7477 | old_runtime = sysctl_sched_rt_runtime; | ||
7478 | |||
7479 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | ||
7480 | |||
7481 | if (!ret && write) { | ||
7482 | ret = sched_rt_global_constraints(); | ||
7483 | if (ret) { | ||
7484 | sysctl_sched_rt_period = old_period; | ||
7485 | sysctl_sched_rt_runtime = old_runtime; | ||
7486 | } else { | ||
7487 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | ||
7488 | def_rt_bandwidth.rt_period = | ||
7489 | ns_to_ktime(global_rt_period()); | ||
7490 | } | ||
7491 | } | ||
7492 | mutex_unlock(&mutex); | ||
7493 | |||
7494 | return ret; | ||
7495 | } | ||
7496 | |||
7497 | #ifdef CONFIG_CGROUP_SCHED | ||
7498 | |||
7499 | /* return corresponding task_group object of a cgroup */ | ||
7500 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) | ||
7501 | { | ||
7502 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), | ||
7503 | struct task_group, css); | ||
7504 | } | ||
7505 | |||
7506 | static struct cgroup_subsys_state * | ||
7507 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) | ||
7508 | { | ||
7509 | struct task_group *tg, *parent; | ||
7510 | |||
7511 | if (!cgrp->parent) { | ||
7512 | /* This is early initialization for the top cgroup */ | ||
7513 | return &root_task_group.css; | ||
7514 | } | ||
7515 | |||
7516 | parent = cgroup_tg(cgrp->parent); | ||
7517 | tg = sched_create_group(parent); | ||
7518 | if (IS_ERR(tg)) | ||
7519 | return ERR_PTR(-ENOMEM); | ||
7520 | |||
7521 | return &tg->css; | ||
7522 | } | ||
7523 | |||
7524 | static void | ||
7525 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | ||
7526 | { | ||
7527 | struct task_group *tg = cgroup_tg(cgrp); | ||
7528 | |||
7529 | sched_destroy_group(tg); | ||
7530 | } | ||
7531 | |||
7532 | static int | ||
7533 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) | ||
7534 | { | ||
7535 | #ifdef CONFIG_RT_GROUP_SCHED | ||
7536 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) | ||
7537 | return -EINVAL; | ||
7538 | #else | ||
7539 | /* We don't support RT-tasks being in separate groups */ | ||
7540 | if (tsk->sched_class != &fair_sched_class) | ||
7541 | return -EINVAL; | ||
7542 | #endif | ||
7543 | return 0; | ||
7544 | } | ||
7545 | |||
7546 | static void | ||
7547 | cpu_cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) | ||
7548 | { | ||
7549 | sched_move_task(tsk); | ||
7550 | } | ||
7551 | |||
7552 | static void | ||
7553 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, | ||
7554 | struct cgroup *old_cgrp, struct task_struct *task) | ||
7555 | { | ||
7556 | /* | ||
7557 | * cgroup_exit() is called in the copy_process() failure path. | ||
7558 | * Ignore this case since the task hasn't ran yet, this avoids | ||
7559 | * trying to poke a half freed task state from generic code. | ||
7560 | */ | ||
7561 | if (!(task->flags & PF_EXITING)) | ||
7562 | return; | ||
7563 | |||
7564 | sched_move_task(task); | ||
7565 | } | ||
7566 | |||
7567 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
7568 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, | ||
7569 | u64 shareval) | ||
7570 | { | ||
7571 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); | ||
7572 | } | ||
7573 | |||
7574 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) | ||
7575 | { | ||
7576 | struct task_group *tg = cgroup_tg(cgrp); | ||
7577 | |||
7578 | return (u64) scale_load_down(tg->shares); | ||
7579 | } | ||
7580 | |||
7581 | #ifdef CONFIG_CFS_BANDWIDTH | ||
7582 | static DEFINE_MUTEX(cfs_constraints_mutex); | ||
7583 | |||
7584 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ | ||
7585 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | ||
7586 | |||
7587 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); | ||
7588 | |||
7589 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) | ||
7590 | { | ||
7591 | int i, ret = 0, runtime_enabled, runtime_was_enabled; | ||
7592 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; | ||
7593 | |||
7594 | if (tg == &root_task_group) | ||
7595 | return -EINVAL; | ||
7596 | |||
7597 | /* | ||
7598 | * Ensure we have at some amount of bandwidth every period. This is | ||
7599 | * to prevent reaching a state of large arrears when throttled via | ||
7600 | * entity_tick() resulting in prolonged exit starvation. | ||
7601 | */ | ||
7602 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | ||
7603 | return -EINVAL; | ||
7604 | |||
7605 | /* | ||
7606 | * Likewise, bound things on the otherside by preventing insane quota | ||
7607 | * periods. This also allows us to normalize in computing quota | ||
7608 | * feasibility. | ||
7609 | */ | ||
7610 | if (period > max_cfs_quota_period) | ||
7611 | return -EINVAL; | ||
7612 | |||
7613 | mutex_lock(&cfs_constraints_mutex); | ||
7614 | ret = __cfs_schedulable(tg, period, quota); | ||
7615 | if (ret) | ||
7616 | goto out_unlock; | ||
7617 | |||
7618 | runtime_enabled = quota != RUNTIME_INF; | ||
7619 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; | ||
7620 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | ||
7621 | raw_spin_lock_irq(&cfs_b->lock); | ||
7622 | cfs_b->period = ns_to_ktime(period); | ||
7623 | cfs_b->quota = quota; | ||
7624 | |||
7625 | __refill_cfs_bandwidth_runtime(cfs_b); | ||
7626 | /* restart the period timer (if active) to handle new period expiry */ | ||
7627 | if (runtime_enabled && cfs_b->timer_active) { | ||
7628 | /* force a reprogram */ | ||
7629 | cfs_b->timer_active = 0; | ||
7630 | __start_cfs_bandwidth(cfs_b); | ||
7631 | } | ||
7632 | raw_spin_unlock_irq(&cfs_b->lock); | ||
7633 | |||
7634 | for_each_possible_cpu(i) { | ||
7635 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | ||
7636 | struct rq *rq = cfs_rq->rq; | ||
7637 | |||
7638 | raw_spin_lock_irq(&rq->lock); | ||
7639 | cfs_rq->runtime_enabled = runtime_enabled; | ||
7640 | cfs_rq->runtime_remaining = 0; | ||
7641 | |||
7642 | if (cfs_rq->throttled) | ||
7643 | unthrottle_cfs_rq(cfs_rq); | ||
7644 | raw_spin_unlock_irq(&rq->lock); | ||
7645 | } | ||
7646 | out_unlock: | ||
7647 | mutex_unlock(&cfs_constraints_mutex); | ||
7648 | |||
7649 | return ret; | ||
7650 | } | ||
7651 | |||
7652 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | ||
7653 | { | ||
7654 | u64 quota, period; | ||
7655 | |||
7656 | period = ktime_to_ns(tg->cfs_bandwidth.period); | ||
7657 | if (cfs_quota_us < 0) | ||
7658 | quota = RUNTIME_INF; | ||
7659 | else | ||
7660 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | ||
7661 | |||
7662 | return tg_set_cfs_bandwidth(tg, period, quota); | ||
7663 | } | ||
7664 | |||
7665 | long tg_get_cfs_quota(struct task_group *tg) | ||
7666 | { | ||
7667 | u64 quota_us; | ||
7668 | |||
7669 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) | ||
7670 | return -1; | ||
7671 | |||
7672 | quota_us = tg->cfs_bandwidth.quota; | ||
7673 | do_div(quota_us, NSEC_PER_USEC); | ||
7674 | |||
7675 | return quota_us; | ||
7676 | } | ||
7677 | |||
7678 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | ||
7679 | { | ||
7680 | u64 quota, period; | ||
7681 | |||
7682 | period = (u64)cfs_period_us * NSEC_PER_USEC; | ||
7683 | quota = tg->cfs_bandwidth.quota; | ||
7684 | |||
7685 | if (period <= 0) | ||
7686 | return -EINVAL; | ||
7687 | |||
7688 | return tg_set_cfs_bandwidth(tg, period, quota); | ||
7689 | } | ||
7690 | |||
7691 | long tg_get_cfs_period(struct task_group *tg) | ||
7692 | { | ||
7693 | u64 cfs_period_us; | ||
7694 | |||
7695 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); | ||
7696 | do_div(cfs_period_us, NSEC_PER_USEC); | ||
7697 | |||
7698 | return cfs_period_us; | ||
7699 | } | ||
7700 | |||
7701 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | ||
7702 | { | ||
7703 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | ||
7704 | } | ||
7705 | |||
7706 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | ||
7707 | s64 cfs_quota_us) | ||
7708 | { | ||
7709 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | ||
7710 | } | ||
7711 | |||
7712 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | ||
7713 | { | ||
7714 | return tg_get_cfs_period(cgroup_tg(cgrp)); | ||
7715 | } | ||
7716 | |||
7717 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | ||
7718 | u64 cfs_period_us) | ||
7719 | { | ||
7720 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | ||
7721 | } | ||
7722 | |||
7723 | struct cfs_schedulable_data { | ||
7724 | struct task_group *tg; | ||
7725 | u64 period, quota; | ||
7726 | }; | ||
7727 | |||
7728 | /* | ||
7729 | * normalize group quota/period to be quota/max_period | ||
7730 | * note: units are usecs | ||
7731 | */ | ||
7732 | static u64 normalize_cfs_quota(struct task_group *tg, | ||
7733 | struct cfs_schedulable_data *d) | ||
7734 | { | ||
7735 | u64 quota, period; | ||
7736 | |||
7737 | if (tg == d->tg) { | ||
7738 | period = d->period; | ||
7739 | quota = d->quota; | ||
7740 | } else { | ||
7741 | period = tg_get_cfs_period(tg); | ||
7742 | quota = tg_get_cfs_quota(tg); | ||
7743 | } | ||
7744 | |||
7745 | /* note: these should typically be equivalent */ | ||
7746 | if (quota == RUNTIME_INF || quota == -1) | ||
7747 | return RUNTIME_INF; | ||
7748 | |||
7749 | return to_ratio(period, quota); | ||
7750 | } | ||
7751 | |||
7752 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | ||
7753 | { | ||
7754 | struct cfs_schedulable_data *d = data; | ||
7755 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; | ||
7756 | s64 quota = 0, parent_quota = -1; | ||
7757 | |||
7758 | if (!tg->parent) { | ||
7759 | quota = RUNTIME_INF; | ||
7760 | } else { | ||
7761 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; | ||
7762 | |||
7763 | quota = normalize_cfs_quota(tg, d); | ||
7764 | parent_quota = parent_b->hierarchal_quota; | ||
7765 | |||
7766 | /* | ||
7767 | * ensure max(child_quota) <= parent_quota, inherit when no | ||
7768 | * limit is set | ||
7769 | */ | ||
7770 | if (quota == RUNTIME_INF) | ||
7771 | quota = parent_quota; | ||
7772 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | ||
7773 | return -EINVAL; | ||
7774 | } | ||
7775 | cfs_b->hierarchal_quota = quota; | ||
7776 | |||
7777 | return 0; | ||
7778 | } | ||
7779 | |||
7780 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | ||
7781 | { | ||
7782 | int ret; | ||
7783 | struct cfs_schedulable_data data = { | ||
7784 | .tg = tg, | ||
7785 | .period = period, | ||
7786 | .quota = quota, | ||
7787 | }; | ||
7788 | |||
7789 | if (quota != RUNTIME_INF) { | ||
7790 | do_div(data.period, NSEC_PER_USEC); | ||
7791 | do_div(data.quota, NSEC_PER_USEC); | ||
7792 | } | ||
7793 | |||
7794 | rcu_read_lock(); | ||
7795 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | ||
7796 | rcu_read_unlock(); | ||
7797 | |||
7798 | return ret; | ||
7799 | } | ||
7800 | |||
7801 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | ||
7802 | struct cgroup_map_cb *cb) | ||
7803 | { | ||
7804 | struct task_group *tg = cgroup_tg(cgrp); | ||
7805 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; | ||
7806 | |||
7807 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | ||
7808 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | ||
7809 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | ||
7810 | |||
7811 | return 0; | ||
7812 | } | ||
7813 | #endif /* CONFIG_CFS_BANDWIDTH */ | ||
7814 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | ||
7815 | |||
7816 | #ifdef CONFIG_RT_GROUP_SCHED | ||
7817 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, | ||
7818 | s64 val) | ||
7819 | { | ||
7820 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); | ||
7821 | } | ||
7822 | |||
7823 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) | ||
7824 | { | ||
7825 | return sched_group_rt_runtime(cgroup_tg(cgrp)); | ||
7826 | } | ||
7827 | |||
7828 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | ||
7829 | u64 rt_period_us) | ||
7830 | { | ||
7831 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | ||
7832 | } | ||
7833 | |||
7834 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | ||
7835 | { | ||
7836 | return sched_group_rt_period(cgroup_tg(cgrp)); | ||
7837 | } | ||
7838 | #endif /* CONFIG_RT_GROUP_SCHED */ | ||
7839 | |||
7840 | static struct cftype cpu_files[] = { | ||
7841 | #ifdef CONFIG_FAIR_GROUP_SCHED | ||
7842 | { | ||
7843 | .name = "shares", | ||
7844 | .read_u64 = cpu_shares_read_u64, | ||
7845 | .write_u64 = cpu_shares_write_u64, | ||
7846 | }, | ||
7847 | #endif | ||
7848 | #ifdef CONFIG_CFS_BANDWIDTH | ||
7849 | { | ||
7850 | .name = "cfs_quota_us", | ||
7851 | .read_s64 = cpu_cfs_quota_read_s64, | ||
7852 | .write_s64 = cpu_cfs_quota_write_s64, | ||
7853 | }, | ||
7854 | { | ||
7855 | .name = "cfs_period_us", | ||
7856 | .read_u64 = cpu_cfs_period_read_u64, | ||
7857 | .write_u64 = cpu_cfs_period_write_u64, | ||
7858 | }, | ||
7859 | { | ||
7860 | .name = "stat", | ||
7861 | .read_map = cpu_stats_show, | ||
7862 | }, | ||
7863 | #endif | ||
7864 | #ifdef CONFIG_RT_GROUP_SCHED | ||
7865 | { | ||
7866 | .name = "rt_runtime_us", | ||
7867 | .read_s64 = cpu_rt_runtime_read, | ||
7868 | .write_s64 = cpu_rt_runtime_write, | ||
7869 | }, | ||
7870 | { | ||
7871 | .name = "rt_period_us", | ||
7872 | .read_u64 = cpu_rt_period_read_uint, | ||
7873 | .write_u64 = cpu_rt_period_write_uint, | ||
7874 | }, | ||
7875 | #endif | ||
7876 | }; | ||
7877 | |||
7878 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | ||
7879 | { | ||
7880 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); | ||
7881 | } | ||
7882 | |||
7883 | struct cgroup_subsys cpu_cgroup_subsys = { | ||
7884 | .name = "cpu", | ||
7885 | .create = cpu_cgroup_create, | ||
7886 | .destroy = cpu_cgroup_destroy, | ||
7887 | .can_attach_task = cpu_cgroup_can_attach_task, | ||
7888 | .attach_task = cpu_cgroup_attach_task, | ||
7889 | .exit = cpu_cgroup_exit, | ||
7890 | .populate = cpu_cgroup_populate, | ||
7891 | .subsys_id = cpu_cgroup_subsys_id, | ||
7892 | .early_init = 1, | ||
7893 | }; | ||
7894 | |||
7895 | #endif /* CONFIG_CGROUP_SCHED */ | ||
7896 | |||
7897 | #ifdef CONFIG_CGROUP_CPUACCT | ||
7898 | |||
7899 | /* | ||
7900 | * CPU accounting code for task groups. | ||
7901 | * | ||
7902 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | ||
7903 | * (balbir@in.ibm.com). | ||
7904 | */ | ||
7905 | |||
7906 | /* create a new cpu accounting group */ | ||
7907 | static struct cgroup_subsys_state *cpuacct_create( | ||
7908 | struct cgroup_subsys *ss, struct cgroup *cgrp) | ||
7909 | { | ||
7910 | struct cpuacct *ca; | ||
7911 | |||
7912 | if (!cgrp->parent) | ||
7913 | return &root_cpuacct.css; | ||
7914 | |||
7915 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | ||
7916 | if (!ca) | ||
7917 | goto out; | ||
7918 | |||
7919 | ca->cpuusage = alloc_percpu(u64); | ||
7920 | if (!ca->cpuusage) | ||
7921 | goto out_free_ca; | ||
7922 | |||
7923 | ca->cpustat = alloc_percpu(struct kernel_cpustat); | ||
7924 | if (!ca->cpustat) | ||
7925 | goto out_free_cpuusage; | ||
7926 | |||
7927 | return &ca->css; | ||
7928 | |||
7929 | out_free_cpuusage: | ||
7930 | free_percpu(ca->cpuusage); | ||
7931 | out_free_ca: | ||
7932 | kfree(ca); | ||
7933 | out: | ||
7934 | return ERR_PTR(-ENOMEM); | ||
7935 | } | ||
7936 | |||
7937 | /* destroy an existing cpu accounting group */ | ||
7938 | static void | ||
7939 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | ||
7940 | { | ||
7941 | struct cpuacct *ca = cgroup_ca(cgrp); | ||
7942 | |||
7943 | free_percpu(ca->cpustat); | ||
7944 | free_percpu(ca->cpuusage); | ||
7945 | kfree(ca); | ||
7946 | } | ||
7947 | |||
7948 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) | ||
7949 | { | ||
7950 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); | ||
7951 | u64 data; | ||
7952 | |||
7953 | #ifndef CONFIG_64BIT | ||
7954 | /* | ||
7955 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | ||
7956 | */ | ||
7957 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); | ||
7958 | data = *cpuusage; | ||
7959 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); | ||
7960 | #else | ||
7961 | data = *cpuusage; | ||
7962 | #endif | ||
7963 | |||
7964 | return data; | ||
7965 | } | ||
7966 | |||
7967 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | ||
7968 | { | ||
7969 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); | ||
7970 | |||
7971 | #ifndef CONFIG_64BIT | ||
7972 | /* | ||
7973 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | ||
7974 | */ | ||
7975 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); | ||
7976 | *cpuusage = val; | ||
7977 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); | ||
7978 | #else | ||
7979 | *cpuusage = val; | ||
7980 | #endif | ||
7981 | } | ||
7982 | |||
7983 | /* return total cpu usage (in nanoseconds) of a group */ | ||
7984 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) | ||
7985 | { | ||
7986 | struct cpuacct *ca = cgroup_ca(cgrp); | ||
7987 | u64 totalcpuusage = 0; | ||
7988 | int i; | ||
7989 | |||
7990 | for_each_present_cpu(i) | ||
7991 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | ||
7992 | |||
7993 | return totalcpuusage; | ||
7994 | } | ||
7995 | |||
7996 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, | ||
7997 | u64 reset) | ||
7998 | { | ||
7999 | struct cpuacct *ca = cgroup_ca(cgrp); | ||
8000 | int err = 0; | ||
8001 | int i; | ||
8002 | |||
8003 | if (reset) { | ||
8004 | err = -EINVAL; | ||
8005 | goto out; | ||
8006 | } | ||
8007 | |||
8008 | for_each_present_cpu(i) | ||
8009 | cpuacct_cpuusage_write(ca, i, 0); | ||
8010 | |||
8011 | out: | ||
8012 | return err; | ||
8013 | } | ||
8014 | |||
8015 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, | ||
8016 | struct seq_file *m) | ||
8017 | { | ||
8018 | struct cpuacct *ca = cgroup_ca(cgroup); | ||
8019 | u64 percpu; | ||
8020 | int i; | ||
8021 | |||
8022 | for_each_present_cpu(i) { | ||
8023 | percpu = cpuacct_cpuusage_read(ca, i); | ||
8024 | seq_printf(m, "%llu ", (unsigned long long) percpu); | ||
8025 | } | ||
8026 | seq_printf(m, "\n"); | ||
8027 | return 0; | ||
8028 | } | ||
8029 | |||
8030 | static const char *cpuacct_stat_desc[] = { | ||
8031 | [CPUACCT_STAT_USER] = "user", | ||
8032 | [CPUACCT_STAT_SYSTEM] = "system", | ||
8033 | }; | ||
8034 | |||
8035 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | ||
8036 | struct cgroup_map_cb *cb) | ||
8037 | { | ||
8038 | struct cpuacct *ca = cgroup_ca(cgrp); | ||
8039 | int cpu; | ||
8040 | s64 val = 0; | ||
8041 | |||
8042 | for_each_online_cpu(cpu) { | ||
8043 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | ||
8044 | val += kcpustat->cpustat[CPUTIME_USER]; | ||
8045 | val += kcpustat->cpustat[CPUTIME_NICE]; | ||
8046 | } | ||
8047 | val = cputime64_to_clock_t(val); | ||
8048 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | ||
8049 | |||
8050 | val = 0; | ||
8051 | for_each_online_cpu(cpu) { | ||
8052 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | ||
8053 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | ||
8054 | val += kcpustat->cpustat[CPUTIME_IRQ]; | ||
8055 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | ||
8056 | } | ||
8057 | |||
8058 | val = cputime64_to_clock_t(val); | ||
8059 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | ||
8060 | |||
8061 | return 0; | ||
8062 | } | ||
8063 | |||
8064 | static struct cftype files[] = { | ||
8065 | { | ||
8066 | .name = "usage", | ||
8067 | .read_u64 = cpuusage_read, | ||
8068 | .write_u64 = cpuusage_write, | ||
8069 | }, | ||
8070 | { | ||
8071 | .name = "usage_percpu", | ||
8072 | .read_seq_string = cpuacct_percpu_seq_read, | ||
8073 | }, | ||
8074 | { | ||
8075 | .name = "stat", | ||
8076 | .read_map = cpuacct_stats_show, | ||
8077 | }, | ||
8078 | }; | ||
8079 | |||
8080 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) | ||
8081 | { | ||
8082 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); | ||
8083 | } | ||
8084 | |||
8085 | /* | ||
8086 | * charge this task's execution time to its accounting group. | ||
8087 | * | ||
8088 | * called with rq->lock held. | ||
8089 | */ | ||
8090 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) | ||
8091 | { | ||
8092 | struct cpuacct *ca; | ||
8093 | int cpu; | ||
8094 | |||
8095 | if (unlikely(!cpuacct_subsys.active)) | ||
8096 | return; | ||
8097 | |||
8098 | cpu = task_cpu(tsk); | ||
8099 | |||
8100 | rcu_read_lock(); | ||
8101 | |||
8102 | ca = task_ca(tsk); | ||
8103 | |||
8104 | for (; ca; ca = parent_ca(ca)) { | ||
8105 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); | ||
8106 | *cpuusage += cputime; | ||
8107 | } | ||
8108 | |||
8109 | rcu_read_unlock(); | ||
8110 | } | ||
8111 | |||
8112 | struct cgroup_subsys cpuacct_subsys = { | ||
8113 | .name = "cpuacct", | ||
8114 | .create = cpuacct_create, | ||
8115 | .destroy = cpuacct_destroy, | ||
8116 | .populate = cpuacct_populate, | ||
8117 | .subsys_id = cpuacct_subsys_id, | ||
8118 | }; | ||
8119 | #endif /* CONFIG_CGROUP_CPUACCT */ | ||