]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
1da177e4 | 75 | |
96f951ed | 76 | #include <asm/switch_to.h> |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
db7e527d | 79 | #include <asm/mutex.h> |
e6e6685a GC |
80 | #ifdef CONFIG_PARAVIRT |
81 | #include <asm/paravirt.h> | |
82 | #endif | |
1da177e4 | 83 | |
029632fb | 84 | #include "sched.h" |
391e43da | 85 | #include "../workqueue_sched.h" |
29d5e047 | 86 | #include "../smpboot.h" |
6e0534f2 | 87 | |
a8d154b0 | 88 | #define CREATE_TRACE_POINTS |
ad8d75ff | 89 | #include <trace/events/sched.h> |
a8d154b0 | 90 | |
029632fb | 91 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 92 | { |
58088ad0 PT |
93 | unsigned long delta; |
94 | ktime_t soft, hard, now; | |
d0b27fa7 | 95 | |
58088ad0 PT |
96 | for (;;) { |
97 | if (hrtimer_active(period_timer)) | |
98 | break; | |
99 | ||
100 | now = hrtimer_cb_get_time(period_timer); | |
101 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 102 | |
58088ad0 PT |
103 | soft = hrtimer_get_softexpires(period_timer); |
104 | hard = hrtimer_get_expires(period_timer); | |
105 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
106 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
107 | HRTIMER_MODE_ABS_PINNED, 0); | |
108 | } | |
109 | } | |
110 | ||
029632fb PZ |
111 | DEFINE_MUTEX(sched_domains_mutex); |
112 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 113 | |
fe44d621 | 114 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 115 | |
029632fb | 116 | void update_rq_clock(struct rq *rq) |
3e51f33f | 117 | { |
fe44d621 | 118 | s64 delta; |
305e6835 | 119 | |
61eadef6 | 120 | if (rq->skip_clock_update > 0) |
f26f9aff | 121 | return; |
aa483808 | 122 | |
fe44d621 PZ |
123 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
124 | rq->clock += delta; | |
125 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
126 | } |
127 | ||
bf5c91ba IM |
128 | /* |
129 | * Debugging: various feature bits | |
130 | */ | |
f00b45c1 | 131 | |
f00b45c1 PZ |
132 | #define SCHED_FEAT(name, enabled) \ |
133 | (1UL << __SCHED_FEAT_##name) * enabled | | |
134 | ||
bf5c91ba | 135 | const_debug unsigned int sysctl_sched_features = |
391e43da | 136 | #include "features.h" |
f00b45c1 PZ |
137 | 0; |
138 | ||
139 | #undef SCHED_FEAT | |
140 | ||
141 | #ifdef CONFIG_SCHED_DEBUG | |
142 | #define SCHED_FEAT(name, enabled) \ | |
143 | #name , | |
144 | ||
1292531f | 145 | static const char * const sched_feat_names[] = { |
391e43da | 146 | #include "features.h" |
f00b45c1 PZ |
147 | }; |
148 | ||
149 | #undef SCHED_FEAT | |
150 | ||
34f3a814 | 151 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 152 | { |
f00b45c1 PZ |
153 | int i; |
154 | ||
f8b6d1cc | 155 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
156 | if (!(sysctl_sched_features & (1UL << i))) |
157 | seq_puts(m, "NO_"); | |
158 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 159 | } |
34f3a814 | 160 | seq_puts(m, "\n"); |
f00b45c1 | 161 | |
34f3a814 | 162 | return 0; |
f00b45c1 PZ |
163 | } |
164 | ||
f8b6d1cc PZ |
165 | #ifdef HAVE_JUMP_LABEL |
166 | ||
c5905afb IM |
167 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
168 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
169 | |
170 | #define SCHED_FEAT(name, enabled) \ | |
171 | jump_label_key__##enabled , | |
172 | ||
c5905afb | 173 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
174 | #include "features.h" |
175 | }; | |
176 | ||
177 | #undef SCHED_FEAT | |
178 | ||
179 | static void sched_feat_disable(int i) | |
180 | { | |
c5905afb IM |
181 | if (static_key_enabled(&sched_feat_keys[i])) |
182 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
183 | } |
184 | ||
185 | static void sched_feat_enable(int i) | |
186 | { | |
c5905afb IM |
187 | if (!static_key_enabled(&sched_feat_keys[i])) |
188 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
189 | } |
190 | #else | |
191 | static void sched_feat_disable(int i) { }; | |
192 | static void sched_feat_enable(int i) { }; | |
193 | #endif /* HAVE_JUMP_LABEL */ | |
194 | ||
1a687c2e | 195 | static int sched_feat_set(char *cmp) |
f00b45c1 | 196 | { |
f00b45c1 | 197 | int i; |
1a687c2e | 198 | int neg = 0; |
f00b45c1 | 199 | |
524429c3 | 200 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
201 | neg = 1; |
202 | cmp += 3; | |
203 | } | |
204 | ||
f8b6d1cc | 205 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 206 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 207 | if (neg) { |
f00b45c1 | 208 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
209 | sched_feat_disable(i); |
210 | } else { | |
f00b45c1 | 211 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
212 | sched_feat_enable(i); |
213 | } | |
f00b45c1 PZ |
214 | break; |
215 | } | |
216 | } | |
217 | ||
1a687c2e MG |
218 | return i; |
219 | } | |
220 | ||
221 | static ssize_t | |
222 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
223 | size_t cnt, loff_t *ppos) | |
224 | { | |
225 | char buf[64]; | |
226 | char *cmp; | |
227 | int i; | |
228 | ||
229 | if (cnt > 63) | |
230 | cnt = 63; | |
231 | ||
232 | if (copy_from_user(&buf, ubuf, cnt)) | |
233 | return -EFAULT; | |
234 | ||
235 | buf[cnt] = 0; | |
236 | cmp = strstrip(buf); | |
237 | ||
238 | i = sched_feat_set(cmp); | |
f8b6d1cc | 239 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
240 | return -EINVAL; |
241 | ||
42994724 | 242 | *ppos += cnt; |
f00b45c1 PZ |
243 | |
244 | return cnt; | |
245 | } | |
246 | ||
34f3a814 LZ |
247 | static int sched_feat_open(struct inode *inode, struct file *filp) |
248 | { | |
249 | return single_open(filp, sched_feat_show, NULL); | |
250 | } | |
251 | ||
828c0950 | 252 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
253 | .open = sched_feat_open, |
254 | .write = sched_feat_write, | |
255 | .read = seq_read, | |
256 | .llseek = seq_lseek, | |
257 | .release = single_release, | |
f00b45c1 PZ |
258 | }; |
259 | ||
260 | static __init int sched_init_debug(void) | |
261 | { | |
f00b45c1 PZ |
262 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
263 | &sched_feat_fops); | |
264 | ||
265 | return 0; | |
266 | } | |
267 | late_initcall(sched_init_debug); | |
f8b6d1cc | 268 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 269 | |
b82d9fdd PZ |
270 | /* |
271 | * Number of tasks to iterate in a single balance run. | |
272 | * Limited because this is done with IRQs disabled. | |
273 | */ | |
274 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
275 | ||
e9e9250b PZ |
276 | /* |
277 | * period over which we average the RT time consumption, measured | |
278 | * in ms. | |
279 | * | |
280 | * default: 1s | |
281 | */ | |
282 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
283 | ||
fa85ae24 | 284 | /* |
9f0c1e56 | 285 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
286 | * default: 1s |
287 | */ | |
9f0c1e56 | 288 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 289 | |
029632fb | 290 | __read_mostly int scheduler_running; |
6892b75e | 291 | |
9f0c1e56 PZ |
292 | /* |
293 | * part of the period that we allow rt tasks to run in us. | |
294 | * default: 0.95s | |
295 | */ | |
296 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 297 | |
fa85ae24 | 298 | |
1da177e4 | 299 | |
0970d299 | 300 | /* |
0122ec5b | 301 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 302 | */ |
70b97a7f | 303 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
304 | __acquires(rq->lock) |
305 | { | |
0970d299 PZ |
306 | struct rq *rq; |
307 | ||
0122ec5b PZ |
308 | lockdep_assert_held(&p->pi_lock); |
309 | ||
3a5c359a | 310 | for (;;) { |
0970d299 | 311 | rq = task_rq(p); |
05fa785c | 312 | raw_spin_lock(&rq->lock); |
65cc8e48 | 313 | if (likely(rq == task_rq(p))) |
3a5c359a | 314 | return rq; |
05fa785c | 315 | raw_spin_unlock(&rq->lock); |
b29739f9 | 316 | } |
b29739f9 IM |
317 | } |
318 | ||
1da177e4 | 319 | /* |
0122ec5b | 320 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 321 | */ |
70b97a7f | 322 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 323 | __acquires(p->pi_lock) |
1da177e4 LT |
324 | __acquires(rq->lock) |
325 | { | |
70b97a7f | 326 | struct rq *rq; |
1da177e4 | 327 | |
3a5c359a | 328 | for (;;) { |
0122ec5b | 329 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 330 | rq = task_rq(p); |
05fa785c | 331 | raw_spin_lock(&rq->lock); |
65cc8e48 | 332 | if (likely(rq == task_rq(p))) |
3a5c359a | 333 | return rq; |
0122ec5b PZ |
334 | raw_spin_unlock(&rq->lock); |
335 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 336 | } |
1da177e4 LT |
337 | } |
338 | ||
a9957449 | 339 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
340 | __releases(rq->lock) |
341 | { | |
05fa785c | 342 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
343 | } |
344 | ||
0122ec5b PZ |
345 | static inline void |
346 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 347 | __releases(rq->lock) |
0122ec5b | 348 | __releases(p->pi_lock) |
1da177e4 | 349 | { |
0122ec5b PZ |
350 | raw_spin_unlock(&rq->lock); |
351 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
352 | } |
353 | ||
1da177e4 | 354 | /* |
cc2a73b5 | 355 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 356 | */ |
a9957449 | 357 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
358 | __acquires(rq->lock) |
359 | { | |
70b97a7f | 360 | struct rq *rq; |
1da177e4 LT |
361 | |
362 | local_irq_disable(); | |
363 | rq = this_rq(); | |
05fa785c | 364 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
365 | |
366 | return rq; | |
367 | } | |
368 | ||
8f4d37ec PZ |
369 | #ifdef CONFIG_SCHED_HRTICK |
370 | /* | |
371 | * Use HR-timers to deliver accurate preemption points. | |
372 | * | |
373 | * Its all a bit involved since we cannot program an hrt while holding the | |
374 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
375 | * reschedule event. | |
376 | * | |
377 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
378 | * rq->lock. | |
379 | */ | |
8f4d37ec | 380 | |
8f4d37ec PZ |
381 | static void hrtick_clear(struct rq *rq) |
382 | { | |
383 | if (hrtimer_active(&rq->hrtick_timer)) | |
384 | hrtimer_cancel(&rq->hrtick_timer); | |
385 | } | |
386 | ||
8f4d37ec PZ |
387 | /* |
388 | * High-resolution timer tick. | |
389 | * Runs from hardirq context with interrupts disabled. | |
390 | */ | |
391 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
392 | { | |
393 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
394 | ||
395 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
396 | ||
05fa785c | 397 | raw_spin_lock(&rq->lock); |
3e51f33f | 398 | update_rq_clock(rq); |
8f4d37ec | 399 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 400 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
401 | |
402 | return HRTIMER_NORESTART; | |
403 | } | |
404 | ||
95e904c7 | 405 | #ifdef CONFIG_SMP |
31656519 PZ |
406 | /* |
407 | * called from hardirq (IPI) context | |
408 | */ | |
409 | static void __hrtick_start(void *arg) | |
b328ca18 | 410 | { |
31656519 | 411 | struct rq *rq = arg; |
b328ca18 | 412 | |
05fa785c | 413 | raw_spin_lock(&rq->lock); |
31656519 PZ |
414 | hrtimer_restart(&rq->hrtick_timer); |
415 | rq->hrtick_csd_pending = 0; | |
05fa785c | 416 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
417 | } |
418 | ||
31656519 PZ |
419 | /* |
420 | * Called to set the hrtick timer state. | |
421 | * | |
422 | * called with rq->lock held and irqs disabled | |
423 | */ | |
029632fb | 424 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 425 | { |
31656519 PZ |
426 | struct hrtimer *timer = &rq->hrtick_timer; |
427 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 428 | |
cc584b21 | 429 | hrtimer_set_expires(timer, time); |
31656519 PZ |
430 | |
431 | if (rq == this_rq()) { | |
432 | hrtimer_restart(timer); | |
433 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 434 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
435 | rq->hrtick_csd_pending = 1; |
436 | } | |
b328ca18 PZ |
437 | } |
438 | ||
439 | static int | |
440 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
441 | { | |
442 | int cpu = (int)(long)hcpu; | |
443 | ||
444 | switch (action) { | |
445 | case CPU_UP_CANCELED: | |
446 | case CPU_UP_CANCELED_FROZEN: | |
447 | case CPU_DOWN_PREPARE: | |
448 | case CPU_DOWN_PREPARE_FROZEN: | |
449 | case CPU_DEAD: | |
450 | case CPU_DEAD_FROZEN: | |
31656519 | 451 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
452 | return NOTIFY_OK; |
453 | } | |
454 | ||
455 | return NOTIFY_DONE; | |
456 | } | |
457 | ||
fa748203 | 458 | static __init void init_hrtick(void) |
b328ca18 PZ |
459 | { |
460 | hotcpu_notifier(hotplug_hrtick, 0); | |
461 | } | |
31656519 PZ |
462 | #else |
463 | /* | |
464 | * Called to set the hrtick timer state. | |
465 | * | |
466 | * called with rq->lock held and irqs disabled | |
467 | */ | |
029632fb | 468 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 469 | { |
7f1e2ca9 | 470 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 471 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 472 | } |
b328ca18 | 473 | |
006c75f1 | 474 | static inline void init_hrtick(void) |
8f4d37ec | 475 | { |
8f4d37ec | 476 | } |
31656519 | 477 | #endif /* CONFIG_SMP */ |
8f4d37ec | 478 | |
31656519 | 479 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 480 | { |
31656519 PZ |
481 | #ifdef CONFIG_SMP |
482 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 483 | |
31656519 PZ |
484 | rq->hrtick_csd.flags = 0; |
485 | rq->hrtick_csd.func = __hrtick_start; | |
486 | rq->hrtick_csd.info = rq; | |
487 | #endif | |
8f4d37ec | 488 | |
31656519 PZ |
489 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
490 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 491 | } |
006c75f1 | 492 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
493 | static inline void hrtick_clear(struct rq *rq) |
494 | { | |
495 | } | |
496 | ||
8f4d37ec PZ |
497 | static inline void init_rq_hrtick(struct rq *rq) |
498 | { | |
499 | } | |
500 | ||
b328ca18 PZ |
501 | static inline void init_hrtick(void) |
502 | { | |
503 | } | |
006c75f1 | 504 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 505 | |
c24d20db IM |
506 | /* |
507 | * resched_task - mark a task 'to be rescheduled now'. | |
508 | * | |
509 | * On UP this means the setting of the need_resched flag, on SMP it | |
510 | * might also involve a cross-CPU call to trigger the scheduler on | |
511 | * the target CPU. | |
512 | */ | |
513 | #ifdef CONFIG_SMP | |
514 | ||
515 | #ifndef tsk_is_polling | |
16a80163 | 516 | #define tsk_is_polling(t) 0 |
c24d20db IM |
517 | #endif |
518 | ||
029632fb | 519 | void resched_task(struct task_struct *p) |
c24d20db IM |
520 | { |
521 | int cpu; | |
522 | ||
05fa785c | 523 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 524 | |
5ed0cec0 | 525 | if (test_tsk_need_resched(p)) |
c24d20db IM |
526 | return; |
527 | ||
5ed0cec0 | 528 | set_tsk_need_resched(p); |
c24d20db IM |
529 | |
530 | cpu = task_cpu(p); | |
531 | if (cpu == smp_processor_id()) | |
532 | return; | |
533 | ||
534 | /* NEED_RESCHED must be visible before we test polling */ | |
535 | smp_mb(); | |
536 | if (!tsk_is_polling(p)) | |
537 | smp_send_reschedule(cpu); | |
538 | } | |
539 | ||
029632fb | 540 | void resched_cpu(int cpu) |
c24d20db IM |
541 | { |
542 | struct rq *rq = cpu_rq(cpu); | |
543 | unsigned long flags; | |
544 | ||
05fa785c | 545 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
546 | return; |
547 | resched_task(cpu_curr(cpu)); | |
05fa785c | 548 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 549 | } |
06d8308c TG |
550 | |
551 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
552 | /* |
553 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
554 | * from an idle cpu. This is good for power-savings. | |
555 | * | |
556 | * We don't do similar optimization for completely idle system, as | |
557 | * selecting an idle cpu will add more delays to the timers than intended | |
558 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
559 | */ | |
560 | int get_nohz_timer_target(void) | |
561 | { | |
562 | int cpu = smp_processor_id(); | |
563 | int i; | |
564 | struct sched_domain *sd; | |
565 | ||
057f3fad | 566 | rcu_read_lock(); |
83cd4fe2 | 567 | for_each_domain(cpu, sd) { |
057f3fad PZ |
568 | for_each_cpu(i, sched_domain_span(sd)) { |
569 | if (!idle_cpu(i)) { | |
570 | cpu = i; | |
571 | goto unlock; | |
572 | } | |
573 | } | |
83cd4fe2 | 574 | } |
057f3fad PZ |
575 | unlock: |
576 | rcu_read_unlock(); | |
83cd4fe2 VP |
577 | return cpu; |
578 | } | |
06d8308c TG |
579 | /* |
580 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
581 | * idle CPU then this timer might expire before the next timer event | |
582 | * which is scheduled to wake up that CPU. In case of a completely | |
583 | * idle system the next event might even be infinite time into the | |
584 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
585 | * leaves the inner idle loop so the newly added timer is taken into | |
586 | * account when the CPU goes back to idle and evaluates the timer | |
587 | * wheel for the next timer event. | |
588 | */ | |
589 | void wake_up_idle_cpu(int cpu) | |
590 | { | |
591 | struct rq *rq = cpu_rq(cpu); | |
592 | ||
593 | if (cpu == smp_processor_id()) | |
594 | return; | |
595 | ||
596 | /* | |
597 | * This is safe, as this function is called with the timer | |
598 | * wheel base lock of (cpu) held. When the CPU is on the way | |
599 | * to idle and has not yet set rq->curr to idle then it will | |
600 | * be serialized on the timer wheel base lock and take the new | |
601 | * timer into account automatically. | |
602 | */ | |
603 | if (rq->curr != rq->idle) | |
604 | return; | |
45bf76df | 605 | |
45bf76df | 606 | /* |
06d8308c TG |
607 | * We can set TIF_RESCHED on the idle task of the other CPU |
608 | * lockless. The worst case is that the other CPU runs the | |
609 | * idle task through an additional NOOP schedule() | |
45bf76df | 610 | */ |
5ed0cec0 | 611 | set_tsk_need_resched(rq->idle); |
45bf76df | 612 | |
06d8308c TG |
613 | /* NEED_RESCHED must be visible before we test polling */ |
614 | smp_mb(); | |
615 | if (!tsk_is_polling(rq->idle)) | |
616 | smp_send_reschedule(cpu); | |
45bf76df IM |
617 | } |
618 | ||
ca38062e | 619 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 620 | { |
1c792db7 SS |
621 | int cpu = smp_processor_id(); |
622 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
623 | } |
624 | ||
ca38062e | 625 | #else /* CONFIG_NO_HZ */ |
45bf76df | 626 | |
ca38062e | 627 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 628 | { |
ca38062e | 629 | return false; |
2069dd75 PZ |
630 | } |
631 | ||
6d6bc0ad | 632 | #endif /* CONFIG_NO_HZ */ |
d842de87 | 633 | |
029632fb | 634 | void sched_avg_update(struct rq *rq) |
18d95a28 | 635 | { |
e9e9250b PZ |
636 | s64 period = sched_avg_period(); |
637 | ||
638 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
639 | /* |
640 | * Inline assembly required to prevent the compiler | |
641 | * optimising this loop into a divmod call. | |
642 | * See __iter_div_u64_rem() for another example of this. | |
643 | */ | |
644 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
645 | rq->age_stamp += period; |
646 | rq->rt_avg /= 2; | |
647 | } | |
18d95a28 PZ |
648 | } |
649 | ||
6d6bc0ad | 650 | #else /* !CONFIG_SMP */ |
029632fb | 651 | void resched_task(struct task_struct *p) |
18d95a28 | 652 | { |
05fa785c | 653 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 654 | set_tsk_need_resched(p); |
18d95a28 | 655 | } |
6d6bc0ad | 656 | #endif /* CONFIG_SMP */ |
18d95a28 | 657 | |
a790de99 PT |
658 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
659 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 660 | /* |
8277434e PT |
661 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
662 | * node and @up when leaving it for the final time. | |
663 | * | |
664 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 665 | */ |
029632fb | 666 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 667 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
668 | { |
669 | struct task_group *parent, *child; | |
eb755805 | 670 | int ret; |
c09595f6 | 671 | |
8277434e PT |
672 | parent = from; |
673 | ||
c09595f6 | 674 | down: |
eb755805 PZ |
675 | ret = (*down)(parent, data); |
676 | if (ret) | |
8277434e | 677 | goto out; |
c09595f6 PZ |
678 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
679 | parent = child; | |
680 | goto down; | |
681 | ||
682 | up: | |
683 | continue; | |
684 | } | |
eb755805 | 685 | ret = (*up)(parent, data); |
8277434e PT |
686 | if (ret || parent == from) |
687 | goto out; | |
c09595f6 PZ |
688 | |
689 | child = parent; | |
690 | parent = parent->parent; | |
691 | if (parent) | |
692 | goto up; | |
8277434e | 693 | out: |
eb755805 | 694 | return ret; |
c09595f6 PZ |
695 | } |
696 | ||
029632fb | 697 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 698 | { |
e2b245f8 | 699 | return 0; |
eb755805 | 700 | } |
18d95a28 PZ |
701 | #endif |
702 | ||
45bf76df IM |
703 | static void set_load_weight(struct task_struct *p) |
704 | { | |
f05998d4 NR |
705 | int prio = p->static_prio - MAX_RT_PRIO; |
706 | struct load_weight *load = &p->se.load; | |
707 | ||
dd41f596 IM |
708 | /* |
709 | * SCHED_IDLE tasks get minimal weight: | |
710 | */ | |
711 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 712 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 713 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
714 | return; |
715 | } | |
71f8bd46 | 716 | |
c8b28116 | 717 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 718 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
719 | } |
720 | ||
371fd7e7 | 721 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 722 | { |
a64692a3 | 723 | update_rq_clock(rq); |
dd41f596 | 724 | sched_info_queued(p); |
371fd7e7 | 725 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
726 | } |
727 | ||
371fd7e7 | 728 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 729 | { |
a64692a3 | 730 | update_rq_clock(rq); |
46ac22ba | 731 | sched_info_dequeued(p); |
371fd7e7 | 732 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
733 | } |
734 | ||
029632fb | 735 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
736 | { |
737 | if (task_contributes_to_load(p)) | |
738 | rq->nr_uninterruptible--; | |
739 | ||
371fd7e7 | 740 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
741 | } |
742 | ||
029632fb | 743 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
744 | { |
745 | if (task_contributes_to_load(p)) | |
746 | rq->nr_uninterruptible++; | |
747 | ||
371fd7e7 | 748 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
749 | } |
750 | ||
fe44d621 | 751 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 752 | { |
095c0aa8 GC |
753 | /* |
754 | * In theory, the compile should just see 0 here, and optimize out the call | |
755 | * to sched_rt_avg_update. But I don't trust it... | |
756 | */ | |
757 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
758 | s64 steal = 0, irq_delta = 0; | |
759 | #endif | |
760 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 761 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
762 | |
763 | /* | |
764 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
765 | * this case when a previous update_rq_clock() happened inside a | |
766 | * {soft,}irq region. | |
767 | * | |
768 | * When this happens, we stop ->clock_task and only update the | |
769 | * prev_irq_time stamp to account for the part that fit, so that a next | |
770 | * update will consume the rest. This ensures ->clock_task is | |
771 | * monotonic. | |
772 | * | |
773 | * It does however cause some slight miss-attribution of {soft,}irq | |
774 | * time, a more accurate solution would be to update the irq_time using | |
775 | * the current rq->clock timestamp, except that would require using | |
776 | * atomic ops. | |
777 | */ | |
778 | if (irq_delta > delta) | |
779 | irq_delta = delta; | |
780 | ||
781 | rq->prev_irq_time += irq_delta; | |
782 | delta -= irq_delta; | |
095c0aa8 GC |
783 | #endif |
784 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 785 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
786 | u64 st; |
787 | ||
788 | steal = paravirt_steal_clock(cpu_of(rq)); | |
789 | steal -= rq->prev_steal_time_rq; | |
790 | ||
791 | if (unlikely(steal > delta)) | |
792 | steal = delta; | |
793 | ||
794 | st = steal_ticks(steal); | |
795 | steal = st * TICK_NSEC; | |
796 | ||
797 | rq->prev_steal_time_rq += steal; | |
798 | ||
799 | delta -= steal; | |
800 | } | |
801 | #endif | |
802 | ||
fe44d621 PZ |
803 | rq->clock_task += delta; |
804 | ||
095c0aa8 GC |
805 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
806 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
807 | sched_rt_avg_update(rq, irq_delta + steal); | |
808 | #endif | |
aa483808 VP |
809 | } |
810 | ||
34f971f6 PZ |
811 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
812 | { | |
813 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
814 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
815 | ||
816 | if (stop) { | |
817 | /* | |
818 | * Make it appear like a SCHED_FIFO task, its something | |
819 | * userspace knows about and won't get confused about. | |
820 | * | |
821 | * Also, it will make PI more or less work without too | |
822 | * much confusion -- but then, stop work should not | |
823 | * rely on PI working anyway. | |
824 | */ | |
825 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
826 | ||
827 | stop->sched_class = &stop_sched_class; | |
828 | } | |
829 | ||
830 | cpu_rq(cpu)->stop = stop; | |
831 | ||
832 | if (old_stop) { | |
833 | /* | |
834 | * Reset it back to a normal scheduling class so that | |
835 | * it can die in pieces. | |
836 | */ | |
837 | old_stop->sched_class = &rt_sched_class; | |
838 | } | |
839 | } | |
840 | ||
14531189 | 841 | /* |
dd41f596 | 842 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 843 | */ |
14531189 IM |
844 | static inline int __normal_prio(struct task_struct *p) |
845 | { | |
dd41f596 | 846 | return p->static_prio; |
14531189 IM |
847 | } |
848 | ||
b29739f9 IM |
849 | /* |
850 | * Calculate the expected normal priority: i.e. priority | |
851 | * without taking RT-inheritance into account. Might be | |
852 | * boosted by interactivity modifiers. Changes upon fork, | |
853 | * setprio syscalls, and whenever the interactivity | |
854 | * estimator recalculates. | |
855 | */ | |
36c8b586 | 856 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
857 | { |
858 | int prio; | |
859 | ||
e05606d3 | 860 | if (task_has_rt_policy(p)) |
b29739f9 IM |
861 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
862 | else | |
863 | prio = __normal_prio(p); | |
864 | return prio; | |
865 | } | |
866 | ||
867 | /* | |
868 | * Calculate the current priority, i.e. the priority | |
869 | * taken into account by the scheduler. This value might | |
870 | * be boosted by RT tasks, or might be boosted by | |
871 | * interactivity modifiers. Will be RT if the task got | |
872 | * RT-boosted. If not then it returns p->normal_prio. | |
873 | */ | |
36c8b586 | 874 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
875 | { |
876 | p->normal_prio = normal_prio(p); | |
877 | /* | |
878 | * If we are RT tasks or we were boosted to RT priority, | |
879 | * keep the priority unchanged. Otherwise, update priority | |
880 | * to the normal priority: | |
881 | */ | |
882 | if (!rt_prio(p->prio)) | |
883 | return p->normal_prio; | |
884 | return p->prio; | |
885 | } | |
886 | ||
1da177e4 LT |
887 | /** |
888 | * task_curr - is this task currently executing on a CPU? | |
889 | * @p: the task in question. | |
890 | */ | |
36c8b586 | 891 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
892 | { |
893 | return cpu_curr(task_cpu(p)) == p; | |
894 | } | |
895 | ||
cb469845 SR |
896 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
897 | const struct sched_class *prev_class, | |
da7a735e | 898 | int oldprio) |
cb469845 SR |
899 | { |
900 | if (prev_class != p->sched_class) { | |
901 | if (prev_class->switched_from) | |
da7a735e PZ |
902 | prev_class->switched_from(rq, p); |
903 | p->sched_class->switched_to(rq, p); | |
904 | } else if (oldprio != p->prio) | |
905 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
906 | } |
907 | ||
029632fb | 908 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
909 | { |
910 | const struct sched_class *class; | |
911 | ||
912 | if (p->sched_class == rq->curr->sched_class) { | |
913 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
914 | } else { | |
915 | for_each_class(class) { | |
916 | if (class == rq->curr->sched_class) | |
917 | break; | |
918 | if (class == p->sched_class) { | |
919 | resched_task(rq->curr); | |
920 | break; | |
921 | } | |
922 | } | |
923 | } | |
924 | ||
925 | /* | |
926 | * A queue event has occurred, and we're going to schedule. In | |
927 | * this case, we can save a useless back to back clock update. | |
928 | */ | |
fd2f4419 | 929 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
930 | rq->skip_clock_update = 1; |
931 | } | |
932 | ||
1da177e4 | 933 | #ifdef CONFIG_SMP |
dd41f596 | 934 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 935 | { |
e2912009 PZ |
936 | #ifdef CONFIG_SCHED_DEBUG |
937 | /* | |
938 | * We should never call set_task_cpu() on a blocked task, | |
939 | * ttwu() will sort out the placement. | |
940 | */ | |
077614ee PZ |
941 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
942 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
943 | |
944 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
945 | /* |
946 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
947 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
948 | * | |
949 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 950 | * see task_group(). |
6c6c54e1 PZ |
951 | * |
952 | * Furthermore, all task_rq users should acquire both locks, see | |
953 | * task_rq_lock(). | |
954 | */ | |
0122ec5b PZ |
955 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
956 | lockdep_is_held(&task_rq(p)->lock))); | |
957 | #endif | |
e2912009 PZ |
958 | #endif |
959 | ||
de1d7286 | 960 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 961 | |
0c69774e PZ |
962 | if (task_cpu(p) != new_cpu) { |
963 | p->se.nr_migrations++; | |
a8b0ca17 | 964 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 965 | } |
dd41f596 IM |
966 | |
967 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
968 | } |
969 | ||
969c7921 | 970 | struct migration_arg { |
36c8b586 | 971 | struct task_struct *task; |
1da177e4 | 972 | int dest_cpu; |
70b97a7f | 973 | }; |
1da177e4 | 974 | |
969c7921 TH |
975 | static int migration_cpu_stop(void *data); |
976 | ||
1da177e4 LT |
977 | /* |
978 | * wait_task_inactive - wait for a thread to unschedule. | |
979 | * | |
85ba2d86 RM |
980 | * If @match_state is nonzero, it's the @p->state value just checked and |
981 | * not expected to change. If it changes, i.e. @p might have woken up, | |
982 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
983 | * we return a positive number (its total switch count). If a second call | |
984 | * a short while later returns the same number, the caller can be sure that | |
985 | * @p has remained unscheduled the whole time. | |
986 | * | |
1da177e4 LT |
987 | * The caller must ensure that the task *will* unschedule sometime soon, |
988 | * else this function might spin for a *long* time. This function can't | |
989 | * be called with interrupts off, or it may introduce deadlock with | |
990 | * smp_call_function() if an IPI is sent by the same process we are | |
991 | * waiting to become inactive. | |
992 | */ | |
85ba2d86 | 993 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
994 | { |
995 | unsigned long flags; | |
dd41f596 | 996 | int running, on_rq; |
85ba2d86 | 997 | unsigned long ncsw; |
70b97a7f | 998 | struct rq *rq; |
1da177e4 | 999 | |
3a5c359a AK |
1000 | for (;;) { |
1001 | /* | |
1002 | * We do the initial early heuristics without holding | |
1003 | * any task-queue locks at all. We'll only try to get | |
1004 | * the runqueue lock when things look like they will | |
1005 | * work out! | |
1006 | */ | |
1007 | rq = task_rq(p); | |
fa490cfd | 1008 | |
3a5c359a AK |
1009 | /* |
1010 | * If the task is actively running on another CPU | |
1011 | * still, just relax and busy-wait without holding | |
1012 | * any locks. | |
1013 | * | |
1014 | * NOTE! Since we don't hold any locks, it's not | |
1015 | * even sure that "rq" stays as the right runqueue! | |
1016 | * But we don't care, since "task_running()" will | |
1017 | * return false if the runqueue has changed and p | |
1018 | * is actually now running somewhere else! | |
1019 | */ | |
85ba2d86 RM |
1020 | while (task_running(rq, p)) { |
1021 | if (match_state && unlikely(p->state != match_state)) | |
1022 | return 0; | |
3a5c359a | 1023 | cpu_relax(); |
85ba2d86 | 1024 | } |
fa490cfd | 1025 | |
3a5c359a AK |
1026 | /* |
1027 | * Ok, time to look more closely! We need the rq | |
1028 | * lock now, to be *sure*. If we're wrong, we'll | |
1029 | * just go back and repeat. | |
1030 | */ | |
1031 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1032 | trace_sched_wait_task(p); |
3a5c359a | 1033 | running = task_running(rq, p); |
fd2f4419 | 1034 | on_rq = p->on_rq; |
85ba2d86 | 1035 | ncsw = 0; |
f31e11d8 | 1036 | if (!match_state || p->state == match_state) |
93dcf55f | 1037 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1038 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1039 | |
85ba2d86 RM |
1040 | /* |
1041 | * If it changed from the expected state, bail out now. | |
1042 | */ | |
1043 | if (unlikely(!ncsw)) | |
1044 | break; | |
1045 | ||
3a5c359a AK |
1046 | /* |
1047 | * Was it really running after all now that we | |
1048 | * checked with the proper locks actually held? | |
1049 | * | |
1050 | * Oops. Go back and try again.. | |
1051 | */ | |
1052 | if (unlikely(running)) { | |
1053 | cpu_relax(); | |
1054 | continue; | |
1055 | } | |
fa490cfd | 1056 | |
3a5c359a AK |
1057 | /* |
1058 | * It's not enough that it's not actively running, | |
1059 | * it must be off the runqueue _entirely_, and not | |
1060 | * preempted! | |
1061 | * | |
80dd99b3 | 1062 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1063 | * running right now), it's preempted, and we should |
1064 | * yield - it could be a while. | |
1065 | */ | |
1066 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1067 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1068 | ||
1069 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1070 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1071 | continue; |
1072 | } | |
fa490cfd | 1073 | |
3a5c359a AK |
1074 | /* |
1075 | * Ahh, all good. It wasn't running, and it wasn't | |
1076 | * runnable, which means that it will never become | |
1077 | * running in the future either. We're all done! | |
1078 | */ | |
1079 | break; | |
1080 | } | |
85ba2d86 RM |
1081 | |
1082 | return ncsw; | |
1da177e4 LT |
1083 | } |
1084 | ||
1085 | /*** | |
1086 | * kick_process - kick a running thread to enter/exit the kernel | |
1087 | * @p: the to-be-kicked thread | |
1088 | * | |
1089 | * Cause a process which is running on another CPU to enter | |
1090 | * kernel-mode, without any delay. (to get signals handled.) | |
1091 | * | |
25985edc | 1092 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1093 | * because all it wants to ensure is that the remote task enters |
1094 | * the kernel. If the IPI races and the task has been migrated | |
1095 | * to another CPU then no harm is done and the purpose has been | |
1096 | * achieved as well. | |
1097 | */ | |
36c8b586 | 1098 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1099 | { |
1100 | int cpu; | |
1101 | ||
1102 | preempt_disable(); | |
1103 | cpu = task_cpu(p); | |
1104 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1105 | smp_send_reschedule(cpu); | |
1106 | preempt_enable(); | |
1107 | } | |
b43e3521 | 1108 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1109 | #endif /* CONFIG_SMP */ |
1da177e4 | 1110 | |
970b13ba | 1111 | #ifdef CONFIG_SMP |
30da688e | 1112 | /* |
013fdb80 | 1113 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1114 | */ |
5da9a0fb PZ |
1115 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1116 | { | |
5da9a0fb | 1117 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); |
2baab4e9 PZ |
1118 | enum { cpuset, possible, fail } state = cpuset; |
1119 | int dest_cpu; | |
5da9a0fb PZ |
1120 | |
1121 | /* Look for allowed, online CPU in same node. */ | |
e3831edd | 1122 | for_each_cpu(dest_cpu, nodemask) { |
2baab4e9 PZ |
1123 | if (!cpu_online(dest_cpu)) |
1124 | continue; | |
1125 | if (!cpu_active(dest_cpu)) | |
1126 | continue; | |
fa17b507 | 1127 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb | 1128 | return dest_cpu; |
2baab4e9 | 1129 | } |
5da9a0fb | 1130 | |
2baab4e9 PZ |
1131 | for (;;) { |
1132 | /* Any allowed, online CPU? */ | |
e3831edd | 1133 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1134 | if (!cpu_online(dest_cpu)) |
1135 | continue; | |
1136 | if (!cpu_active(dest_cpu)) | |
1137 | continue; | |
1138 | goto out; | |
1139 | } | |
5da9a0fb | 1140 | |
2baab4e9 PZ |
1141 | switch (state) { |
1142 | case cpuset: | |
1143 | /* No more Mr. Nice Guy. */ | |
1144 | cpuset_cpus_allowed_fallback(p); | |
1145 | state = possible; | |
1146 | break; | |
1147 | ||
1148 | case possible: | |
1149 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1150 | state = fail; | |
1151 | break; | |
1152 | ||
1153 | case fail: | |
1154 | BUG(); | |
1155 | break; | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | out: | |
1160 | if (state != cpuset) { | |
1161 | /* | |
1162 | * Don't tell them about moving exiting tasks or | |
1163 | * kernel threads (both mm NULL), since they never | |
1164 | * leave kernel. | |
1165 | */ | |
1166 | if (p->mm && printk_ratelimit()) { | |
1167 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1168 | task_pid_nr(p), p->comm, cpu); | |
1169 | } | |
5da9a0fb PZ |
1170 | } |
1171 | ||
1172 | return dest_cpu; | |
1173 | } | |
1174 | ||
e2912009 | 1175 | /* |
013fdb80 | 1176 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1177 | */ |
970b13ba | 1178 | static inline |
7608dec2 | 1179 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1180 | { |
7608dec2 | 1181 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1182 | |
1183 | /* | |
1184 | * In order not to call set_task_cpu() on a blocking task we need | |
1185 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1186 | * cpu. | |
1187 | * | |
1188 | * Since this is common to all placement strategies, this lives here. | |
1189 | * | |
1190 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1191 | * not worry about this generic constraint ] | |
1192 | */ | |
fa17b507 | 1193 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1194 | !cpu_online(cpu))) |
5da9a0fb | 1195 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1196 | |
1197 | return cpu; | |
970b13ba | 1198 | } |
09a40af5 MG |
1199 | |
1200 | static void update_avg(u64 *avg, u64 sample) | |
1201 | { | |
1202 | s64 diff = sample - *avg; | |
1203 | *avg += diff >> 3; | |
1204 | } | |
970b13ba PZ |
1205 | #endif |
1206 | ||
d7c01d27 | 1207 | static void |
b84cb5df | 1208 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1209 | { |
d7c01d27 | 1210 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1211 | struct rq *rq = this_rq(); |
1212 | ||
d7c01d27 PZ |
1213 | #ifdef CONFIG_SMP |
1214 | int this_cpu = smp_processor_id(); | |
1215 | ||
1216 | if (cpu == this_cpu) { | |
1217 | schedstat_inc(rq, ttwu_local); | |
1218 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1219 | } else { | |
1220 | struct sched_domain *sd; | |
1221 | ||
1222 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1223 | rcu_read_lock(); |
d7c01d27 PZ |
1224 | for_each_domain(this_cpu, sd) { |
1225 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1226 | schedstat_inc(sd, ttwu_wake_remote); | |
1227 | break; | |
1228 | } | |
1229 | } | |
057f3fad | 1230 | rcu_read_unlock(); |
d7c01d27 | 1231 | } |
f339b9dc PZ |
1232 | |
1233 | if (wake_flags & WF_MIGRATED) | |
1234 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1235 | ||
d7c01d27 PZ |
1236 | #endif /* CONFIG_SMP */ |
1237 | ||
1238 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1239 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1240 | |
1241 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1242 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1243 | |
d7c01d27 PZ |
1244 | #endif /* CONFIG_SCHEDSTATS */ |
1245 | } | |
1246 | ||
1247 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1248 | { | |
9ed3811a | 1249 | activate_task(rq, p, en_flags); |
fd2f4419 | 1250 | p->on_rq = 1; |
c2f7115e PZ |
1251 | |
1252 | /* if a worker is waking up, notify workqueue */ | |
1253 | if (p->flags & PF_WQ_WORKER) | |
1254 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1255 | } |
1256 | ||
23f41eeb PZ |
1257 | /* |
1258 | * Mark the task runnable and perform wakeup-preemption. | |
1259 | */ | |
89363381 | 1260 | static void |
23f41eeb | 1261 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1262 | { |
89363381 | 1263 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1264 | check_preempt_curr(rq, p, wake_flags); |
1265 | ||
1266 | p->state = TASK_RUNNING; | |
1267 | #ifdef CONFIG_SMP | |
1268 | if (p->sched_class->task_woken) | |
1269 | p->sched_class->task_woken(rq, p); | |
1270 | ||
e69c6341 | 1271 | if (rq->idle_stamp) { |
9ed3811a TH |
1272 | u64 delta = rq->clock - rq->idle_stamp; |
1273 | u64 max = 2*sysctl_sched_migration_cost; | |
1274 | ||
1275 | if (delta > max) | |
1276 | rq->avg_idle = max; | |
1277 | else | |
1278 | update_avg(&rq->avg_idle, delta); | |
1279 | rq->idle_stamp = 0; | |
1280 | } | |
1281 | #endif | |
1282 | } | |
1283 | ||
c05fbafb PZ |
1284 | static void |
1285 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1286 | { | |
1287 | #ifdef CONFIG_SMP | |
1288 | if (p->sched_contributes_to_load) | |
1289 | rq->nr_uninterruptible--; | |
1290 | #endif | |
1291 | ||
1292 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1293 | ttwu_do_wakeup(rq, p, wake_flags); | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1298 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1299 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1300 | * the task is still ->on_rq. | |
1301 | */ | |
1302 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1303 | { | |
1304 | struct rq *rq; | |
1305 | int ret = 0; | |
1306 | ||
1307 | rq = __task_rq_lock(p); | |
1308 | if (p->on_rq) { | |
1309 | ttwu_do_wakeup(rq, p, wake_flags); | |
1310 | ret = 1; | |
1311 | } | |
1312 | __task_rq_unlock(rq); | |
1313 | ||
1314 | return ret; | |
1315 | } | |
1316 | ||
317f3941 | 1317 | #ifdef CONFIG_SMP |
fa14ff4a | 1318 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1319 | { |
1320 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1321 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1322 | struct task_struct *p; | |
317f3941 PZ |
1323 | |
1324 | raw_spin_lock(&rq->lock); | |
1325 | ||
fa14ff4a PZ |
1326 | while (llist) { |
1327 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1328 | llist = llist_next(llist); | |
317f3941 PZ |
1329 | ttwu_do_activate(rq, p, 0); |
1330 | } | |
1331 | ||
1332 | raw_spin_unlock(&rq->lock); | |
1333 | } | |
1334 | ||
1335 | void scheduler_ipi(void) | |
1336 | { | |
ca38062e | 1337 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1338 | return; |
1339 | ||
1340 | /* | |
1341 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1342 | * traditionally all their work was done from the interrupt return | |
1343 | * path. Now that we actually do some work, we need to make sure | |
1344 | * we do call them. | |
1345 | * | |
1346 | * Some archs already do call them, luckily irq_enter/exit nest | |
1347 | * properly. | |
1348 | * | |
1349 | * Arguably we should visit all archs and update all handlers, | |
1350 | * however a fair share of IPIs are still resched only so this would | |
1351 | * somewhat pessimize the simple resched case. | |
1352 | */ | |
1353 | irq_enter(); | |
fa14ff4a | 1354 | sched_ttwu_pending(); |
ca38062e SS |
1355 | |
1356 | /* | |
1357 | * Check if someone kicked us for doing the nohz idle load balance. | |
1358 | */ | |
6eb57e0d SS |
1359 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1360 | this_rq()->idle_balance = 1; | |
ca38062e | 1361 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1362 | } |
c5d753a5 | 1363 | irq_exit(); |
317f3941 PZ |
1364 | } |
1365 | ||
1366 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1367 | { | |
fa14ff4a | 1368 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1369 | smp_send_reschedule(cpu); |
1370 | } | |
d6aa8f85 | 1371 | |
39be3501 | 1372 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1373 | { |
1374 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1375 | } | |
d6aa8f85 | 1376 | #endif /* CONFIG_SMP */ |
317f3941 | 1377 | |
c05fbafb PZ |
1378 | static void ttwu_queue(struct task_struct *p, int cpu) |
1379 | { | |
1380 | struct rq *rq = cpu_rq(cpu); | |
1381 | ||
17d9f311 | 1382 | #if defined(CONFIG_SMP) |
39be3501 | 1383 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1384 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1385 | ttwu_queue_remote(p, cpu); |
1386 | return; | |
1387 | } | |
1388 | #endif | |
1389 | ||
c05fbafb PZ |
1390 | raw_spin_lock(&rq->lock); |
1391 | ttwu_do_activate(rq, p, 0); | |
1392 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1393 | } |
1394 | ||
1395 | /** | |
1da177e4 | 1396 | * try_to_wake_up - wake up a thread |
9ed3811a | 1397 | * @p: the thread to be awakened |
1da177e4 | 1398 | * @state: the mask of task states that can be woken |
9ed3811a | 1399 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1400 | * |
1401 | * Put it on the run-queue if it's not already there. The "current" | |
1402 | * thread is always on the run-queue (except when the actual | |
1403 | * re-schedule is in progress), and as such you're allowed to do | |
1404 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1405 | * runnable without the overhead of this. | |
1406 | * | |
9ed3811a TH |
1407 | * Returns %true if @p was woken up, %false if it was already running |
1408 | * or @state didn't match @p's state. | |
1da177e4 | 1409 | */ |
e4a52bcb PZ |
1410 | static int |
1411 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1412 | { |
1da177e4 | 1413 | unsigned long flags; |
c05fbafb | 1414 | int cpu, success = 0; |
2398f2c6 | 1415 | |
04e2f174 | 1416 | smp_wmb(); |
013fdb80 | 1417 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1418 | if (!(p->state & state)) |
1da177e4 LT |
1419 | goto out; |
1420 | ||
c05fbafb | 1421 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1422 | cpu = task_cpu(p); |
1da177e4 | 1423 | |
c05fbafb PZ |
1424 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1425 | goto stat; | |
1da177e4 | 1426 | |
1da177e4 | 1427 | #ifdef CONFIG_SMP |
e9c84311 | 1428 | /* |
c05fbafb PZ |
1429 | * If the owning (remote) cpu is still in the middle of schedule() with |
1430 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1431 | */ |
f3e94786 | 1432 | while (p->on_cpu) |
e4a52bcb | 1433 | cpu_relax(); |
0970d299 | 1434 | /* |
e4a52bcb | 1435 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1436 | */ |
e4a52bcb | 1437 | smp_rmb(); |
1da177e4 | 1438 | |
a8e4f2ea | 1439 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1440 | p->state = TASK_WAKING; |
e7693a36 | 1441 | |
e4a52bcb | 1442 | if (p->sched_class->task_waking) |
74f8e4b2 | 1443 | p->sched_class->task_waking(p); |
efbbd05a | 1444 | |
7608dec2 | 1445 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1446 | if (task_cpu(p) != cpu) { |
1447 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1448 | set_task_cpu(p, cpu); |
f339b9dc | 1449 | } |
1da177e4 | 1450 | #endif /* CONFIG_SMP */ |
1da177e4 | 1451 | |
c05fbafb PZ |
1452 | ttwu_queue(p, cpu); |
1453 | stat: | |
b84cb5df | 1454 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1455 | out: |
013fdb80 | 1456 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1457 | |
1458 | return success; | |
1459 | } | |
1460 | ||
21aa9af0 TH |
1461 | /** |
1462 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1463 | * @p: the thread to be awakened | |
1464 | * | |
2acca55e | 1465 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1466 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1467 | * the current task. |
21aa9af0 TH |
1468 | */ |
1469 | static void try_to_wake_up_local(struct task_struct *p) | |
1470 | { | |
1471 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1472 | |
1473 | BUG_ON(rq != this_rq()); | |
1474 | BUG_ON(p == current); | |
1475 | lockdep_assert_held(&rq->lock); | |
1476 | ||
2acca55e PZ |
1477 | if (!raw_spin_trylock(&p->pi_lock)) { |
1478 | raw_spin_unlock(&rq->lock); | |
1479 | raw_spin_lock(&p->pi_lock); | |
1480 | raw_spin_lock(&rq->lock); | |
1481 | } | |
1482 | ||
21aa9af0 | 1483 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1484 | goto out; |
21aa9af0 | 1485 | |
fd2f4419 | 1486 | if (!p->on_rq) |
d7c01d27 PZ |
1487 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1488 | ||
23f41eeb | 1489 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1490 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1491 | out: |
1492 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1493 | } |
1494 | ||
50fa610a DH |
1495 | /** |
1496 | * wake_up_process - Wake up a specific process | |
1497 | * @p: The process to be woken up. | |
1498 | * | |
1499 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1500 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1501 | * running. | |
1502 | * | |
1503 | * It may be assumed that this function implies a write memory barrier before | |
1504 | * changing the task state if and only if any tasks are woken up. | |
1505 | */ | |
7ad5b3a5 | 1506 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1507 | { |
d9514f6c | 1508 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1509 | } |
1da177e4 LT |
1510 | EXPORT_SYMBOL(wake_up_process); |
1511 | ||
7ad5b3a5 | 1512 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1513 | { |
1514 | return try_to_wake_up(p, state, 0); | |
1515 | } | |
1516 | ||
1da177e4 LT |
1517 | /* |
1518 | * Perform scheduler related setup for a newly forked process p. | |
1519 | * p is forked by current. | |
dd41f596 IM |
1520 | * |
1521 | * __sched_fork() is basic setup used by init_idle() too: | |
1522 | */ | |
1523 | static void __sched_fork(struct task_struct *p) | |
1524 | { | |
fd2f4419 PZ |
1525 | p->on_rq = 0; |
1526 | ||
1527 | p->se.on_rq = 0; | |
dd41f596 IM |
1528 | p->se.exec_start = 0; |
1529 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1530 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1531 | p->se.nr_migrations = 0; |
da7a735e | 1532 | p->se.vruntime = 0; |
fd2f4419 | 1533 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1534 | |
1535 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1536 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1537 | #endif |
476d139c | 1538 | |
fa717060 | 1539 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1540 | |
e107be36 AK |
1541 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1542 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1543 | #endif | |
cbee9f88 PZ |
1544 | |
1545 | #ifdef CONFIG_NUMA_BALANCING | |
1546 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
1547 | p->mm->numa_next_scan = jiffies; | |
b8593bfd | 1548 | p->mm->numa_next_reset = jiffies; |
cbee9f88 PZ |
1549 | p->mm->numa_scan_seq = 0; |
1550 | } | |
1551 | ||
1552 | p->node_stamp = 0ULL; | |
1553 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
1554 | p->numa_migrate_seq = p->mm ? p->mm->numa_scan_seq - 1 : 0; | |
4b96a29b | 1555 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 PZ |
1556 | p->numa_work.next = &p->numa_work; |
1557 | #endif /* CONFIG_NUMA_BALANCING */ | |
dd41f596 IM |
1558 | } |
1559 | ||
1a687c2e MG |
1560 | #ifdef CONFIG_NUMA_BALANCING |
1561 | void set_numabalancing_state(bool enabled) | |
1562 | { | |
1563 | if (enabled) | |
1564 | sched_feat_set("NUMA"); | |
1565 | else | |
1566 | sched_feat_set("NO_NUMA"); | |
1567 | } | |
1568 | #endif /* CONFIG_NUMA_BALANCING */ | |
1569 | ||
dd41f596 IM |
1570 | /* |
1571 | * fork()/clone()-time setup: | |
1572 | */ | |
3e51e3ed | 1573 | void sched_fork(struct task_struct *p) |
dd41f596 | 1574 | { |
0122ec5b | 1575 | unsigned long flags; |
dd41f596 IM |
1576 | int cpu = get_cpu(); |
1577 | ||
1578 | __sched_fork(p); | |
06b83b5f | 1579 | /* |
0017d735 | 1580 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1581 | * nobody will actually run it, and a signal or other external |
1582 | * event cannot wake it up and insert it on the runqueue either. | |
1583 | */ | |
0017d735 | 1584 | p->state = TASK_RUNNING; |
dd41f596 | 1585 | |
c350a04e MG |
1586 | /* |
1587 | * Make sure we do not leak PI boosting priority to the child. | |
1588 | */ | |
1589 | p->prio = current->normal_prio; | |
1590 | ||
b9dc29e7 MG |
1591 | /* |
1592 | * Revert to default priority/policy on fork if requested. | |
1593 | */ | |
1594 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1595 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1596 | p->policy = SCHED_NORMAL; |
6c697bdf | 1597 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1598 | p->rt_priority = 0; |
1599 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1600 | p->static_prio = NICE_TO_PRIO(0); | |
1601 | ||
1602 | p->prio = p->normal_prio = __normal_prio(p); | |
1603 | set_load_weight(p); | |
6c697bdf | 1604 | |
b9dc29e7 MG |
1605 | /* |
1606 | * We don't need the reset flag anymore after the fork. It has | |
1607 | * fulfilled its duty: | |
1608 | */ | |
1609 | p->sched_reset_on_fork = 0; | |
1610 | } | |
ca94c442 | 1611 | |
2ddbf952 HS |
1612 | if (!rt_prio(p->prio)) |
1613 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1614 | |
cd29fe6f PZ |
1615 | if (p->sched_class->task_fork) |
1616 | p->sched_class->task_fork(p); | |
1617 | ||
86951599 PZ |
1618 | /* |
1619 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1620 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1621 | * is ran before sched_fork(). | |
1622 | * | |
1623 | * Silence PROVE_RCU. | |
1624 | */ | |
0122ec5b | 1625 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1626 | set_task_cpu(p, cpu); |
0122ec5b | 1627 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1628 | |
52f17b6c | 1629 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1630 | if (likely(sched_info_on())) |
52f17b6c | 1631 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1632 | #endif |
3ca7a440 PZ |
1633 | #if defined(CONFIG_SMP) |
1634 | p->on_cpu = 0; | |
4866cde0 | 1635 | #endif |
bdd4e85d | 1636 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1637 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1638 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1639 | #endif |
806c09a7 | 1640 | #ifdef CONFIG_SMP |
917b627d | 1641 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1642 | #endif |
917b627d | 1643 | |
476d139c | 1644 | put_cpu(); |
1da177e4 LT |
1645 | } |
1646 | ||
1647 | /* | |
1648 | * wake_up_new_task - wake up a newly created task for the first time. | |
1649 | * | |
1650 | * This function will do some initial scheduler statistics housekeeping | |
1651 | * that must be done for every newly created context, then puts the task | |
1652 | * on the runqueue and wakes it. | |
1653 | */ | |
3e51e3ed | 1654 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1655 | { |
1656 | unsigned long flags; | |
dd41f596 | 1657 | struct rq *rq; |
fabf318e | 1658 | |
ab2515c4 | 1659 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1660 | #ifdef CONFIG_SMP |
1661 | /* | |
1662 | * Fork balancing, do it here and not earlier because: | |
1663 | * - cpus_allowed can change in the fork path | |
1664 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1665 | */ |
ab2515c4 | 1666 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1667 | #endif |
1668 | ||
ab2515c4 | 1669 | rq = __task_rq_lock(p); |
cd29fe6f | 1670 | activate_task(rq, p, 0); |
fd2f4419 | 1671 | p->on_rq = 1; |
89363381 | 1672 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1673 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1674 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1675 | if (p->sched_class->task_woken) |
1676 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1677 | #endif |
0122ec5b | 1678 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1679 | } |
1680 | ||
e107be36 AK |
1681 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1682 | ||
1683 | /** | |
80dd99b3 | 1684 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1685 | * @notifier: notifier struct to register |
e107be36 AK |
1686 | */ |
1687 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1688 | { | |
1689 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1690 | } | |
1691 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1692 | ||
1693 | /** | |
1694 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1695 | * @notifier: notifier struct to unregister |
e107be36 AK |
1696 | * |
1697 | * This is safe to call from within a preemption notifier. | |
1698 | */ | |
1699 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1700 | { | |
1701 | hlist_del(¬ifier->link); | |
1702 | } | |
1703 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1704 | ||
1705 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1706 | { | |
1707 | struct preempt_notifier *notifier; | |
1708 | struct hlist_node *node; | |
1709 | ||
1710 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1711 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1712 | } | |
1713 | ||
1714 | static void | |
1715 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1716 | struct task_struct *next) | |
1717 | { | |
1718 | struct preempt_notifier *notifier; | |
1719 | struct hlist_node *node; | |
1720 | ||
1721 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1722 | notifier->ops->sched_out(notifier, next); | |
1723 | } | |
1724 | ||
6d6bc0ad | 1725 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1726 | |
1727 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1728 | { | |
1729 | } | |
1730 | ||
1731 | static void | |
1732 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1733 | struct task_struct *next) | |
1734 | { | |
1735 | } | |
1736 | ||
6d6bc0ad | 1737 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1738 | |
4866cde0 NP |
1739 | /** |
1740 | * prepare_task_switch - prepare to switch tasks | |
1741 | * @rq: the runqueue preparing to switch | |
421cee29 | 1742 | * @prev: the current task that is being switched out |
4866cde0 NP |
1743 | * @next: the task we are going to switch to. |
1744 | * | |
1745 | * This is called with the rq lock held and interrupts off. It must | |
1746 | * be paired with a subsequent finish_task_switch after the context | |
1747 | * switch. | |
1748 | * | |
1749 | * prepare_task_switch sets up locking and calls architecture specific | |
1750 | * hooks. | |
1751 | */ | |
e107be36 AK |
1752 | static inline void |
1753 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1754 | struct task_struct *next) | |
4866cde0 | 1755 | { |
895dd92c | 1756 | trace_sched_switch(prev, next); |
fe4b04fa PZ |
1757 | sched_info_switch(prev, next); |
1758 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1759 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1760 | prepare_lock_switch(rq, next); |
1761 | prepare_arch_switch(next); | |
1762 | } | |
1763 | ||
1da177e4 LT |
1764 | /** |
1765 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1766 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1767 | * @prev: the thread we just switched away from. |
1768 | * | |
4866cde0 NP |
1769 | * finish_task_switch must be called after the context switch, paired |
1770 | * with a prepare_task_switch call before the context switch. | |
1771 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1772 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1773 | * |
1774 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1775 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1776 | * with the lock held can cause deadlocks; see schedule() for |
1777 | * details.) | |
1778 | */ | |
a9957449 | 1779 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1780 | __releases(rq->lock) |
1781 | { | |
1da177e4 | 1782 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1783 | long prev_state; |
1da177e4 LT |
1784 | |
1785 | rq->prev_mm = NULL; | |
1786 | ||
1787 | /* | |
1788 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1789 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1790 | * schedule one last time. The schedule call will never return, and |
1791 | * the scheduled task must drop that reference. | |
c394cc9f | 1792 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1793 | * still held, otherwise prev could be scheduled on another cpu, die |
1794 | * there before we look at prev->state, and then the reference would | |
1795 | * be dropped twice. | |
1796 | * Manfred Spraul <manfred@colorfullife.com> | |
1797 | */ | |
55a101f8 | 1798 | prev_state = prev->state; |
bf9fae9f | 1799 | vtime_task_switch(prev); |
4866cde0 | 1800 | finish_arch_switch(prev); |
a8d757ef | 1801 | perf_event_task_sched_in(prev, current); |
4866cde0 | 1802 | finish_lock_switch(rq, prev); |
01f23e16 | 1803 | finish_arch_post_lock_switch(); |
e8fa1362 | 1804 | |
e107be36 | 1805 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1806 | if (mm) |
1807 | mmdrop(mm); | |
c394cc9f | 1808 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1809 | /* |
1810 | * Remove function-return probe instances associated with this | |
1811 | * task and put them back on the free list. | |
9761eea8 | 1812 | */ |
c6fd91f0 | 1813 | kprobe_flush_task(prev); |
1da177e4 | 1814 | put_task_struct(prev); |
c6fd91f0 | 1815 | } |
1da177e4 LT |
1816 | } |
1817 | ||
3f029d3c GH |
1818 | #ifdef CONFIG_SMP |
1819 | ||
1820 | /* assumes rq->lock is held */ | |
1821 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1822 | { | |
1823 | if (prev->sched_class->pre_schedule) | |
1824 | prev->sched_class->pre_schedule(rq, prev); | |
1825 | } | |
1826 | ||
1827 | /* rq->lock is NOT held, but preemption is disabled */ | |
1828 | static inline void post_schedule(struct rq *rq) | |
1829 | { | |
1830 | if (rq->post_schedule) { | |
1831 | unsigned long flags; | |
1832 | ||
05fa785c | 1833 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1834 | if (rq->curr->sched_class->post_schedule) |
1835 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1836 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1837 | |
1838 | rq->post_schedule = 0; | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | #else | |
da19ab51 | 1843 | |
3f029d3c GH |
1844 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1845 | { | |
1846 | } | |
1847 | ||
1848 | static inline void post_schedule(struct rq *rq) | |
1849 | { | |
1da177e4 LT |
1850 | } |
1851 | ||
3f029d3c GH |
1852 | #endif |
1853 | ||
1da177e4 LT |
1854 | /** |
1855 | * schedule_tail - first thing a freshly forked thread must call. | |
1856 | * @prev: the thread we just switched away from. | |
1857 | */ | |
36c8b586 | 1858 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1859 | __releases(rq->lock) |
1860 | { | |
70b97a7f IM |
1861 | struct rq *rq = this_rq(); |
1862 | ||
4866cde0 | 1863 | finish_task_switch(rq, prev); |
da19ab51 | 1864 | |
3f029d3c GH |
1865 | /* |
1866 | * FIXME: do we need to worry about rq being invalidated by the | |
1867 | * task_switch? | |
1868 | */ | |
1869 | post_schedule(rq); | |
70b97a7f | 1870 | |
4866cde0 NP |
1871 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
1872 | /* In this case, finish_task_switch does not reenable preemption */ | |
1873 | preempt_enable(); | |
1874 | #endif | |
1da177e4 | 1875 | if (current->set_child_tid) |
b488893a | 1876 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1877 | } |
1878 | ||
1879 | /* | |
1880 | * context_switch - switch to the new MM and the new | |
1881 | * thread's register state. | |
1882 | */ | |
dd41f596 | 1883 | static inline void |
70b97a7f | 1884 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1885 | struct task_struct *next) |
1da177e4 | 1886 | { |
dd41f596 | 1887 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1888 | |
e107be36 | 1889 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 1890 | |
dd41f596 IM |
1891 | mm = next->mm; |
1892 | oldmm = prev->active_mm; | |
9226d125 ZA |
1893 | /* |
1894 | * For paravirt, this is coupled with an exit in switch_to to | |
1895 | * combine the page table reload and the switch backend into | |
1896 | * one hypercall. | |
1897 | */ | |
224101ed | 1898 | arch_start_context_switch(prev); |
9226d125 | 1899 | |
31915ab4 | 1900 | if (!mm) { |
1da177e4 LT |
1901 | next->active_mm = oldmm; |
1902 | atomic_inc(&oldmm->mm_count); | |
1903 | enter_lazy_tlb(oldmm, next); | |
1904 | } else | |
1905 | switch_mm(oldmm, mm, next); | |
1906 | ||
31915ab4 | 1907 | if (!prev->mm) { |
1da177e4 | 1908 | prev->active_mm = NULL; |
1da177e4 LT |
1909 | rq->prev_mm = oldmm; |
1910 | } | |
3a5f5e48 IM |
1911 | /* |
1912 | * Since the runqueue lock will be released by the next | |
1913 | * task (which is an invalid locking op but in the case | |
1914 | * of the scheduler it's an obvious special-case), so we | |
1915 | * do an early lockdep release here: | |
1916 | */ | |
1917 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1918 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1919 | #endif |
1da177e4 LT |
1920 | |
1921 | /* Here we just switch the register state and the stack. */ | |
04e7e951 | 1922 | rcu_switch(prev, next); |
1da177e4 LT |
1923 | switch_to(prev, next, prev); |
1924 | ||
dd41f596 IM |
1925 | barrier(); |
1926 | /* | |
1927 | * this_rq must be evaluated again because prev may have moved | |
1928 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1929 | * frame will be invalid. | |
1930 | */ | |
1931 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1932 | } |
1933 | ||
1934 | /* | |
1935 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1936 | * | |
1937 | * externally visible scheduler statistics: current number of runnable | |
1938 | * threads, current number of uninterruptible-sleeping threads, total | |
1939 | * number of context switches performed since bootup. | |
1940 | */ | |
1941 | unsigned long nr_running(void) | |
1942 | { | |
1943 | unsigned long i, sum = 0; | |
1944 | ||
1945 | for_each_online_cpu(i) | |
1946 | sum += cpu_rq(i)->nr_running; | |
1947 | ||
1948 | return sum; | |
f711f609 | 1949 | } |
1da177e4 LT |
1950 | |
1951 | unsigned long nr_uninterruptible(void) | |
f711f609 | 1952 | { |
1da177e4 | 1953 | unsigned long i, sum = 0; |
f711f609 | 1954 | |
0a945022 | 1955 | for_each_possible_cpu(i) |
1da177e4 | 1956 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
1957 | |
1958 | /* | |
1da177e4 LT |
1959 | * Since we read the counters lockless, it might be slightly |
1960 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 1961 | */ |
1da177e4 LT |
1962 | if (unlikely((long)sum < 0)) |
1963 | sum = 0; | |
f711f609 | 1964 | |
1da177e4 | 1965 | return sum; |
f711f609 | 1966 | } |
f711f609 | 1967 | |
1da177e4 | 1968 | unsigned long long nr_context_switches(void) |
46cb4b7c | 1969 | { |
cc94abfc SR |
1970 | int i; |
1971 | unsigned long long sum = 0; | |
46cb4b7c | 1972 | |
0a945022 | 1973 | for_each_possible_cpu(i) |
1da177e4 | 1974 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 1975 | |
1da177e4 LT |
1976 | return sum; |
1977 | } | |
483b4ee6 | 1978 | |
1da177e4 LT |
1979 | unsigned long nr_iowait(void) |
1980 | { | |
1981 | unsigned long i, sum = 0; | |
483b4ee6 | 1982 | |
0a945022 | 1983 | for_each_possible_cpu(i) |
1da177e4 | 1984 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 1985 | |
1da177e4 LT |
1986 | return sum; |
1987 | } | |
483b4ee6 | 1988 | |
8c215bd3 | 1989 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 1990 | { |
8c215bd3 | 1991 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
1992 | return atomic_read(&this->nr_iowait); |
1993 | } | |
46cb4b7c | 1994 | |
69d25870 AV |
1995 | unsigned long this_cpu_load(void) |
1996 | { | |
1997 | struct rq *this = this_rq(); | |
1998 | return this->cpu_load[0]; | |
1999 | } | |
e790fb0b | 2000 | |
46cb4b7c | 2001 | |
5167e8d5 PZ |
2002 | /* |
2003 | * Global load-average calculations | |
2004 | * | |
2005 | * We take a distributed and async approach to calculating the global load-avg | |
2006 | * in order to minimize overhead. | |
2007 | * | |
2008 | * The global load average is an exponentially decaying average of nr_running + | |
2009 | * nr_uninterruptible. | |
2010 | * | |
2011 | * Once every LOAD_FREQ: | |
2012 | * | |
2013 | * nr_active = 0; | |
2014 | * for_each_possible_cpu(cpu) | |
2015 | * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; | |
2016 | * | |
2017 | * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) | |
2018 | * | |
2019 | * Due to a number of reasons the above turns in the mess below: | |
2020 | * | |
2021 | * - for_each_possible_cpu() is prohibitively expensive on machines with | |
2022 | * serious number of cpus, therefore we need to take a distributed approach | |
2023 | * to calculating nr_active. | |
2024 | * | |
2025 | * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 | |
2026 | * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } | |
2027 | * | |
2028 | * So assuming nr_active := 0 when we start out -- true per definition, we | |
2029 | * can simply take per-cpu deltas and fold those into a global accumulate | |
2030 | * to obtain the same result. See calc_load_fold_active(). | |
2031 | * | |
2032 | * Furthermore, in order to avoid synchronizing all per-cpu delta folding | |
2033 | * across the machine, we assume 10 ticks is sufficient time for every | |
2034 | * cpu to have completed this task. | |
2035 | * | |
2036 | * This places an upper-bound on the IRQ-off latency of the machine. Then | |
2037 | * again, being late doesn't loose the delta, just wrecks the sample. | |
2038 | * | |
2039 | * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because | |
2040 | * this would add another cross-cpu cacheline miss and atomic operation | |
2041 | * to the wakeup path. Instead we increment on whatever cpu the task ran | |
2042 | * when it went into uninterruptible state and decrement on whatever cpu | |
2043 | * did the wakeup. This means that only the sum of nr_uninterruptible over | |
2044 | * all cpus yields the correct result. | |
2045 | * | |
2046 | * This covers the NO_HZ=n code, for extra head-aches, see the comment below. | |
2047 | */ | |
2048 | ||
dce48a84 TG |
2049 | /* Variables and functions for calc_load */ |
2050 | static atomic_long_t calc_load_tasks; | |
2051 | static unsigned long calc_load_update; | |
2052 | unsigned long avenrun[3]; | |
5167e8d5 PZ |
2053 | EXPORT_SYMBOL(avenrun); /* should be removed */ |
2054 | ||
2055 | /** | |
2056 | * get_avenrun - get the load average array | |
2057 | * @loads: pointer to dest load array | |
2058 | * @offset: offset to add | |
2059 | * @shift: shift count to shift the result left | |
2060 | * | |
2061 | * These values are estimates at best, so no need for locking. | |
2062 | */ | |
2063 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2064 | { | |
2065 | loads[0] = (avenrun[0] + offset) << shift; | |
2066 | loads[1] = (avenrun[1] + offset) << shift; | |
2067 | loads[2] = (avenrun[2] + offset) << shift; | |
2068 | } | |
46cb4b7c | 2069 | |
74f5187a PZ |
2070 | static long calc_load_fold_active(struct rq *this_rq) |
2071 | { | |
2072 | long nr_active, delta = 0; | |
2073 | ||
2074 | nr_active = this_rq->nr_running; | |
2075 | nr_active += (long) this_rq->nr_uninterruptible; | |
2076 | ||
2077 | if (nr_active != this_rq->calc_load_active) { | |
2078 | delta = nr_active - this_rq->calc_load_active; | |
2079 | this_rq->calc_load_active = nr_active; | |
2080 | } | |
2081 | ||
2082 | return delta; | |
2083 | } | |
2084 | ||
5167e8d5 PZ |
2085 | /* |
2086 | * a1 = a0 * e + a * (1 - e) | |
2087 | */ | |
0f004f5a PZ |
2088 | static unsigned long |
2089 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2090 | { | |
2091 | load *= exp; | |
2092 | load += active * (FIXED_1 - exp); | |
2093 | load += 1UL << (FSHIFT - 1); | |
2094 | return load >> FSHIFT; | |
2095 | } | |
2096 | ||
74f5187a PZ |
2097 | #ifdef CONFIG_NO_HZ |
2098 | /* | |
5167e8d5 PZ |
2099 | * Handle NO_HZ for the global load-average. |
2100 | * | |
2101 | * Since the above described distributed algorithm to compute the global | |
2102 | * load-average relies on per-cpu sampling from the tick, it is affected by | |
2103 | * NO_HZ. | |
2104 | * | |
2105 | * The basic idea is to fold the nr_active delta into a global idle-delta upon | |
2106 | * entering NO_HZ state such that we can include this as an 'extra' cpu delta | |
2107 | * when we read the global state. | |
2108 | * | |
2109 | * Obviously reality has to ruin such a delightfully simple scheme: | |
2110 | * | |
2111 | * - When we go NO_HZ idle during the window, we can negate our sample | |
2112 | * contribution, causing under-accounting. | |
2113 | * | |
2114 | * We avoid this by keeping two idle-delta counters and flipping them | |
2115 | * when the window starts, thus separating old and new NO_HZ load. | |
2116 | * | |
2117 | * The only trick is the slight shift in index flip for read vs write. | |
2118 | * | |
2119 | * 0s 5s 10s 15s | |
2120 | * +10 +10 +10 +10 | |
2121 | * |-|-----------|-|-----------|-|-----------|-| | |
2122 | * r:0 0 1 1 0 0 1 1 0 | |
2123 | * w:0 1 1 0 0 1 1 0 0 | |
2124 | * | |
2125 | * This ensures we'll fold the old idle contribution in this window while | |
2126 | * accumlating the new one. | |
2127 | * | |
2128 | * - When we wake up from NO_HZ idle during the window, we push up our | |
2129 | * contribution, since we effectively move our sample point to a known | |
2130 | * busy state. | |
2131 | * | |
2132 | * This is solved by pushing the window forward, and thus skipping the | |
2133 | * sample, for this cpu (effectively using the idle-delta for this cpu which | |
2134 | * was in effect at the time the window opened). This also solves the issue | |
2135 | * of having to deal with a cpu having been in NOHZ idle for multiple | |
2136 | * LOAD_FREQ intervals. | |
74f5187a PZ |
2137 | * |
2138 | * When making the ILB scale, we should try to pull this in as well. | |
2139 | */ | |
5167e8d5 PZ |
2140 | static atomic_long_t calc_load_idle[2]; |
2141 | static int calc_load_idx; | |
74f5187a | 2142 | |
5167e8d5 | 2143 | static inline int calc_load_write_idx(void) |
74f5187a | 2144 | { |
5167e8d5 PZ |
2145 | int idx = calc_load_idx; |
2146 | ||
2147 | /* | |
2148 | * See calc_global_nohz(), if we observe the new index, we also | |
2149 | * need to observe the new update time. | |
2150 | */ | |
2151 | smp_rmb(); | |
2152 | ||
2153 | /* | |
2154 | * If the folding window started, make sure we start writing in the | |
2155 | * next idle-delta. | |
2156 | */ | |
2157 | if (!time_before(jiffies, calc_load_update)) | |
2158 | idx++; | |
2159 | ||
2160 | return idx & 1; | |
2161 | } | |
2162 | ||
2163 | static inline int calc_load_read_idx(void) | |
2164 | { | |
2165 | return calc_load_idx & 1; | |
2166 | } | |
2167 | ||
2168 | void calc_load_enter_idle(void) | |
2169 | { | |
2170 | struct rq *this_rq = this_rq(); | |
74f5187a PZ |
2171 | long delta; |
2172 | ||
5167e8d5 PZ |
2173 | /* |
2174 | * We're going into NOHZ mode, if there's any pending delta, fold it | |
2175 | * into the pending idle delta. | |
2176 | */ | |
74f5187a | 2177 | delta = calc_load_fold_active(this_rq); |
5167e8d5 PZ |
2178 | if (delta) { |
2179 | int idx = calc_load_write_idx(); | |
2180 | atomic_long_add(delta, &calc_load_idle[idx]); | |
2181 | } | |
74f5187a PZ |
2182 | } |
2183 | ||
5167e8d5 | 2184 | void calc_load_exit_idle(void) |
74f5187a | 2185 | { |
5167e8d5 PZ |
2186 | struct rq *this_rq = this_rq(); |
2187 | ||
2188 | /* | |
2189 | * If we're still before the sample window, we're done. | |
2190 | */ | |
2191 | if (time_before(jiffies, this_rq->calc_load_update)) | |
2192 | return; | |
74f5187a PZ |
2193 | |
2194 | /* | |
5167e8d5 PZ |
2195 | * We woke inside or after the sample window, this means we're already |
2196 | * accounted through the nohz accounting, so skip the entire deal and | |
2197 | * sync up for the next window. | |
74f5187a | 2198 | */ |
5167e8d5 PZ |
2199 | this_rq->calc_load_update = calc_load_update; |
2200 | if (time_before(jiffies, this_rq->calc_load_update + 10)) | |
2201 | this_rq->calc_load_update += LOAD_FREQ; | |
2202 | } | |
2203 | ||
2204 | static long calc_load_fold_idle(void) | |
2205 | { | |
2206 | int idx = calc_load_read_idx(); | |
2207 | long delta = 0; | |
2208 | ||
2209 | if (atomic_long_read(&calc_load_idle[idx])) | |
2210 | delta = atomic_long_xchg(&calc_load_idle[idx], 0); | |
74f5187a PZ |
2211 | |
2212 | return delta; | |
2213 | } | |
0f004f5a PZ |
2214 | |
2215 | /** | |
2216 | * fixed_power_int - compute: x^n, in O(log n) time | |
2217 | * | |
2218 | * @x: base of the power | |
2219 | * @frac_bits: fractional bits of @x | |
2220 | * @n: power to raise @x to. | |
2221 | * | |
2222 | * By exploiting the relation between the definition of the natural power | |
2223 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2224 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2225 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2226 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2227 | * of course trivially computable in O(log_2 n), the length of our binary | |
2228 | * vector. | |
2229 | */ | |
2230 | static unsigned long | |
2231 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2232 | { | |
2233 | unsigned long result = 1UL << frac_bits; | |
2234 | ||
2235 | if (n) for (;;) { | |
2236 | if (n & 1) { | |
2237 | result *= x; | |
2238 | result += 1UL << (frac_bits - 1); | |
2239 | result >>= frac_bits; | |
2240 | } | |
2241 | n >>= 1; | |
2242 | if (!n) | |
2243 | break; | |
2244 | x *= x; | |
2245 | x += 1UL << (frac_bits - 1); | |
2246 | x >>= frac_bits; | |
2247 | } | |
2248 | ||
2249 | return result; | |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * a1 = a0 * e + a * (1 - e) | |
2254 | * | |
2255 | * a2 = a1 * e + a * (1 - e) | |
2256 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2257 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2258 | * | |
2259 | * a3 = a2 * e + a * (1 - e) | |
2260 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2261 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2262 | * | |
2263 | * ... | |
2264 | * | |
2265 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2266 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2267 | * = a0 * e^n + a * (1 - e^n) | |
2268 | * | |
2269 | * [1] application of the geometric series: | |
2270 | * | |
2271 | * n 1 - x^(n+1) | |
2272 | * S_n := \Sum x^i = ------------- | |
2273 | * i=0 1 - x | |
2274 | */ | |
2275 | static unsigned long | |
2276 | calc_load_n(unsigned long load, unsigned long exp, | |
2277 | unsigned long active, unsigned int n) | |
2278 | { | |
2279 | ||
2280 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2281 | } | |
2282 | ||
2283 | /* | |
2284 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2285 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2286 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2287 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2288 | * | |
2289 | * Once we've updated the global active value, we need to apply the exponential | |
2290 | * weights adjusted to the number of cycles missed. | |
2291 | */ | |
c308b56b | 2292 | static void calc_global_nohz(void) |
0f004f5a PZ |
2293 | { |
2294 | long delta, active, n; | |
2295 | ||
5167e8d5 PZ |
2296 | if (!time_before(jiffies, calc_load_update + 10)) { |
2297 | /* | |
2298 | * Catch-up, fold however many we are behind still | |
2299 | */ | |
2300 | delta = jiffies - calc_load_update - 10; | |
2301 | n = 1 + (delta / LOAD_FREQ); | |
0f004f5a | 2302 | |
5167e8d5 PZ |
2303 | active = atomic_long_read(&calc_load_tasks); |
2304 | active = active > 0 ? active * FIXED_1 : 0; | |
0f004f5a | 2305 | |
5167e8d5 PZ |
2306 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
2307 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2308 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
0f004f5a | 2309 | |
5167e8d5 PZ |
2310 | calc_load_update += n * LOAD_FREQ; |
2311 | } | |
74f5187a | 2312 | |
5167e8d5 PZ |
2313 | /* |
2314 | * Flip the idle index... | |
2315 | * | |
2316 | * Make sure we first write the new time then flip the index, so that | |
2317 | * calc_load_write_idx() will see the new time when it reads the new | |
2318 | * index, this avoids a double flip messing things up. | |
2319 | */ | |
2320 | smp_wmb(); | |
2321 | calc_load_idx++; | |
74f5187a | 2322 | } |
5167e8d5 | 2323 | #else /* !CONFIG_NO_HZ */ |
0f004f5a | 2324 | |
5167e8d5 PZ |
2325 | static inline long calc_load_fold_idle(void) { return 0; } |
2326 | static inline void calc_global_nohz(void) { } | |
74f5187a | 2327 | |
5167e8d5 | 2328 | #endif /* CONFIG_NO_HZ */ |
46cb4b7c | 2329 | |
46cb4b7c | 2330 | /* |
dce48a84 TG |
2331 | * calc_load - update the avenrun load estimates 10 ticks after the |
2332 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2333 | */ |
0f004f5a | 2334 | void calc_global_load(unsigned long ticks) |
7835b98b | 2335 | { |
5167e8d5 | 2336 | long active, delta; |
1da177e4 | 2337 | |
0f004f5a | 2338 | if (time_before(jiffies, calc_load_update + 10)) |
dce48a84 | 2339 | return; |
1da177e4 | 2340 | |
5167e8d5 PZ |
2341 | /* |
2342 | * Fold the 'old' idle-delta to include all NO_HZ cpus. | |
2343 | */ | |
2344 | delta = calc_load_fold_idle(); | |
2345 | if (delta) | |
2346 | atomic_long_add(delta, &calc_load_tasks); | |
2347 | ||
dce48a84 TG |
2348 | active = atomic_long_read(&calc_load_tasks); |
2349 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2350 | |
dce48a84 TG |
2351 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2352 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2353 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2354 | |
dce48a84 | 2355 | calc_load_update += LOAD_FREQ; |
c308b56b PZ |
2356 | |
2357 | /* | |
5167e8d5 | 2358 | * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. |
c308b56b PZ |
2359 | */ |
2360 | calc_global_nohz(); | |
dce48a84 | 2361 | } |
1da177e4 | 2362 | |
dce48a84 | 2363 | /* |
74f5187a PZ |
2364 | * Called from update_cpu_load() to periodically update this CPU's |
2365 | * active count. | |
dce48a84 TG |
2366 | */ |
2367 | static void calc_load_account_active(struct rq *this_rq) | |
2368 | { | |
74f5187a | 2369 | long delta; |
08c183f3 | 2370 | |
74f5187a PZ |
2371 | if (time_before(jiffies, this_rq->calc_load_update)) |
2372 | return; | |
783609c6 | 2373 | |
74f5187a | 2374 | delta = calc_load_fold_active(this_rq); |
74f5187a | 2375 | if (delta) |
dce48a84 | 2376 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2377 | |
2378 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2379 | } |
2380 | ||
5167e8d5 PZ |
2381 | /* |
2382 | * End of global load-average stuff | |
2383 | */ | |
2384 | ||
fdf3e95d VP |
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 | ||
46cb4b7c | 2452 | /* |
dd41f596 | 2453 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2454 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2455 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2456 | */ |
556061b0 PZ |
2457 | static void __update_cpu_load(struct rq *this_rq, unsigned long this_load, |
2458 | unsigned long pending_updates) | |
46cb4b7c | 2459 | { |
dd41f596 | 2460 | int i, scale; |
46cb4b7c | 2461 | |
dd41f596 | 2462 | this_rq->nr_load_updates++; |
46cb4b7c | 2463 | |
dd41f596 | 2464 | /* Update our load: */ |
fdf3e95d VP |
2465 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2466 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2467 | unsigned long old_load, new_load; |
7d1e6a9b | 2468 | |
dd41f596 | 2469 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2470 | |
dd41f596 | 2471 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2472 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2473 | new_load = this_load; |
a25707f3 IM |
2474 | /* |
2475 | * Round up the averaging division if load is increasing. This | |
2476 | * prevents us from getting stuck on 9 if the load is 10, for | |
2477 | * example. | |
2478 | */ | |
2479 | if (new_load > old_load) | |
fdf3e95d VP |
2480 | new_load += scale - 1; |
2481 | ||
2482 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2483 | } |
da2b71ed SS |
2484 | |
2485 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2486 | } |
2487 | ||
5aaa0b7a PZ |
2488 | #ifdef CONFIG_NO_HZ |
2489 | /* | |
2490 | * There is no sane way to deal with nohz on smp when using jiffies because the | |
2491 | * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading | |
2492 | * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}. | |
2493 | * | |
2494 | * Therefore we cannot use the delta approach from the regular tick since that | |
2495 | * would seriously skew the load calculation. However we'll make do for those | |
2496 | * updates happening while idle (nohz_idle_balance) or coming out of idle | |
2497 | * (tick_nohz_idle_exit). | |
2498 | * | |
2499 | * This means we might still be one tick off for nohz periods. | |
2500 | */ | |
2501 | ||
556061b0 PZ |
2502 | /* |
2503 | * Called from nohz_idle_balance() to update the load ratings before doing the | |
2504 | * idle balance. | |
2505 | */ | |
2506 | void update_idle_cpu_load(struct rq *this_rq) | |
2507 | { | |
5aaa0b7a | 2508 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
556061b0 PZ |
2509 | unsigned long load = this_rq->load.weight; |
2510 | unsigned long pending_updates; | |
2511 | ||
2512 | /* | |
5aaa0b7a | 2513 | * bail if there's load or we're actually up-to-date. |
556061b0 PZ |
2514 | */ |
2515 | if (load || curr_jiffies == this_rq->last_load_update_tick) | |
2516 | return; | |
2517 | ||
2518 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2519 | this_rq->last_load_update_tick = curr_jiffies; | |
2520 | ||
2521 | __update_cpu_load(this_rq, load, pending_updates); | |
2522 | } | |
2523 | ||
5aaa0b7a PZ |
2524 | /* |
2525 | * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed. | |
2526 | */ | |
2527 | void update_cpu_load_nohz(void) | |
2528 | { | |
2529 | struct rq *this_rq = this_rq(); | |
2530 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); | |
2531 | unsigned long pending_updates; | |
2532 | ||
2533 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2534 | return; | |
2535 | ||
2536 | raw_spin_lock(&this_rq->lock); | |
2537 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2538 | if (pending_updates) { | |
2539 | this_rq->last_load_update_tick = curr_jiffies; | |
2540 | /* | |
2541 | * We were idle, this means load 0, the current load might be | |
2542 | * !0 due to remote wakeups and the sort. | |
2543 | */ | |
2544 | __update_cpu_load(this_rq, 0, pending_updates); | |
2545 | } | |
2546 | raw_spin_unlock(&this_rq->lock); | |
2547 | } | |
2548 | #endif /* CONFIG_NO_HZ */ | |
2549 | ||
556061b0 PZ |
2550 | /* |
2551 | * Called from scheduler_tick() | |
2552 | */ | |
fdf3e95d VP |
2553 | static void update_cpu_load_active(struct rq *this_rq) |
2554 | { | |
556061b0 | 2555 | /* |
5aaa0b7a | 2556 | * See the mess around update_idle_cpu_load() / update_cpu_load_nohz(). |
556061b0 PZ |
2557 | */ |
2558 | this_rq->last_load_update_tick = jiffies; | |
2559 | __update_cpu_load(this_rq, this_rq->load.weight, 1); | |
46cb4b7c | 2560 | |
74f5187a | 2561 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2562 | } |
2563 | ||
dd41f596 | 2564 | #ifdef CONFIG_SMP |
8a0be9ef | 2565 | |
46cb4b7c | 2566 | /* |
38022906 PZ |
2567 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2568 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2569 | */ |
38022906 | 2570 | void sched_exec(void) |
46cb4b7c | 2571 | { |
38022906 | 2572 | struct task_struct *p = current; |
1da177e4 | 2573 | unsigned long flags; |
0017d735 | 2574 | int dest_cpu; |
46cb4b7c | 2575 | |
8f42ced9 | 2576 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2577 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2578 | if (dest_cpu == smp_processor_id()) |
2579 | goto unlock; | |
38022906 | 2580 | |
8f42ced9 | 2581 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2582 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2583 | |
8f42ced9 PZ |
2584 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2585 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2586 | return; |
2587 | } | |
0017d735 | 2588 | unlock: |
8f42ced9 | 2589 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2590 | } |
dd41f596 | 2591 | |
1da177e4 LT |
2592 | #endif |
2593 | ||
1da177e4 | 2594 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2595 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2596 | |
2597 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2598 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2599 | |
2600 | /* | |
c5f8d995 | 2601 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2602 | * @p in case that task is currently running. |
c5f8d995 HS |
2603 | * |
2604 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2605 | */ |
c5f8d995 HS |
2606 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2607 | { | |
2608 | u64 ns = 0; | |
2609 | ||
2610 | if (task_current(rq, p)) { | |
2611 | update_rq_clock(rq); | |
305e6835 | 2612 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2613 | if ((s64)ns < 0) |
2614 | ns = 0; | |
2615 | } | |
2616 | ||
2617 | return ns; | |
2618 | } | |
2619 | ||
bb34d92f | 2620 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2621 | { |
1da177e4 | 2622 | unsigned long flags; |
41b86e9c | 2623 | struct rq *rq; |
bb34d92f | 2624 | u64 ns = 0; |
48f24c4d | 2625 | |
41b86e9c | 2626 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2627 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2628 | task_rq_unlock(rq, p, &flags); |
1508487e | 2629 | |
c5f8d995 HS |
2630 | return ns; |
2631 | } | |
f06febc9 | 2632 | |
c5f8d995 HS |
2633 | /* |
2634 | * Return accounted runtime for the task. | |
2635 | * In case the task is currently running, return the runtime plus current's | |
2636 | * pending runtime that have not been accounted yet. | |
2637 | */ | |
2638 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2639 | { | |
2640 | unsigned long flags; | |
2641 | struct rq *rq; | |
2642 | u64 ns = 0; | |
2643 | ||
2644 | rq = task_rq_lock(p, &flags); | |
2645 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2646 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2647 | |
2648 | return ns; | |
2649 | } | |
48f24c4d | 2650 | |
7835b98b CL |
2651 | /* |
2652 | * This function gets called by the timer code, with HZ frequency. | |
2653 | * We call it with interrupts disabled. | |
7835b98b CL |
2654 | */ |
2655 | void scheduler_tick(void) | |
2656 | { | |
7835b98b CL |
2657 | int cpu = smp_processor_id(); |
2658 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2659 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2660 | |
2661 | sched_clock_tick(); | |
dd41f596 | 2662 | |
05fa785c | 2663 | raw_spin_lock(&rq->lock); |
3e51f33f | 2664 | update_rq_clock(rq); |
fdf3e95d | 2665 | update_cpu_load_active(rq); |
fa85ae24 | 2666 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2667 | raw_spin_unlock(&rq->lock); |
7835b98b | 2668 | |
e9d2b064 | 2669 | perf_event_task_tick(); |
e220d2dc | 2670 | |
e418e1c2 | 2671 | #ifdef CONFIG_SMP |
6eb57e0d | 2672 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2673 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2674 | #endif |
1da177e4 LT |
2675 | } |
2676 | ||
132380a0 | 2677 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2678 | { |
2679 | if (in_lock_functions(addr)) { | |
2680 | addr = CALLER_ADDR2; | |
2681 | if (in_lock_functions(addr)) | |
2682 | addr = CALLER_ADDR3; | |
2683 | } | |
2684 | return addr; | |
2685 | } | |
1da177e4 | 2686 | |
7e49fcce SR |
2687 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2688 | defined(CONFIG_PREEMPT_TRACER)) | |
2689 | ||
43627582 | 2690 | void __kprobes add_preempt_count(int val) |
1da177e4 | 2691 | { |
6cd8a4bb | 2692 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2693 | /* |
2694 | * Underflow? | |
2695 | */ | |
9a11b49a IM |
2696 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2697 | return; | |
6cd8a4bb | 2698 | #endif |
1da177e4 | 2699 | preempt_count() += val; |
6cd8a4bb | 2700 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2701 | /* |
2702 | * Spinlock count overflowing soon? | |
2703 | */ | |
33859f7f MOS |
2704 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2705 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2706 | #endif |
2707 | if (preempt_count() == val) | |
2708 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2709 | } |
2710 | EXPORT_SYMBOL(add_preempt_count); | |
2711 | ||
43627582 | 2712 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 2713 | { |
6cd8a4bb | 2714 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2715 | /* |
2716 | * Underflow? | |
2717 | */ | |
01e3eb82 | 2718 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2719 | return; |
1da177e4 LT |
2720 | /* |
2721 | * Is the spinlock portion underflowing? | |
2722 | */ | |
9a11b49a IM |
2723 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2724 | !(preempt_count() & PREEMPT_MASK))) | |
2725 | return; | |
6cd8a4bb | 2726 | #endif |
9a11b49a | 2727 | |
6cd8a4bb SR |
2728 | if (preempt_count() == val) |
2729 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2730 | preempt_count() -= val; |
2731 | } | |
2732 | EXPORT_SYMBOL(sub_preempt_count); | |
2733 | ||
2734 | #endif | |
2735 | ||
2736 | /* | |
dd41f596 | 2737 | * Print scheduling while atomic bug: |
1da177e4 | 2738 | */ |
dd41f596 | 2739 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2740 | { |
664dfa65 DJ |
2741 | if (oops_in_progress) |
2742 | return; | |
2743 | ||
3df0fc5b PZ |
2744 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2745 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2746 | |
dd41f596 | 2747 | debug_show_held_locks(prev); |
e21f5b15 | 2748 | print_modules(); |
dd41f596 IM |
2749 | if (irqs_disabled()) |
2750 | print_irqtrace_events(prev); | |
6135fc1e | 2751 | dump_stack(); |
1c2927f1 | 2752 | add_taint(TAINT_WARN); |
dd41f596 | 2753 | } |
1da177e4 | 2754 | |
dd41f596 IM |
2755 | /* |
2756 | * Various schedule()-time debugging checks and statistics: | |
2757 | */ | |
2758 | static inline void schedule_debug(struct task_struct *prev) | |
2759 | { | |
1da177e4 | 2760 | /* |
41a2d6cf | 2761 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
2762 | * schedule() atomically, we ignore that path for now. |
2763 | * Otherwise, whine if we are scheduling when we should not be. | |
2764 | */ | |
3f33a7ce | 2765 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 2766 | __schedule_bug(prev); |
b3fbab05 | 2767 | rcu_sleep_check(); |
dd41f596 | 2768 | |
1da177e4 LT |
2769 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2770 | ||
2d72376b | 2771 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2772 | } |
2773 | ||
6cecd084 | 2774 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2775 | { |
61eadef6 | 2776 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2777 | update_rq_clock(rq); |
6cecd084 | 2778 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2779 | } |
2780 | ||
dd41f596 IM |
2781 | /* |
2782 | * Pick up the highest-prio task: | |
2783 | */ | |
2784 | static inline struct task_struct * | |
b67802ea | 2785 | pick_next_task(struct rq *rq) |
dd41f596 | 2786 | { |
5522d5d5 | 2787 | const struct sched_class *class; |
dd41f596 | 2788 | struct task_struct *p; |
1da177e4 LT |
2789 | |
2790 | /* | |
dd41f596 IM |
2791 | * Optimization: we know that if all tasks are in |
2792 | * the fair class we can call that function directly: | |
1da177e4 | 2793 | */ |
953bfcd1 | 2794 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2795 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2796 | if (likely(p)) |
2797 | return p; | |
1da177e4 LT |
2798 | } |
2799 | ||
34f971f6 | 2800 | for_each_class(class) { |
fb8d4724 | 2801 | p = class->pick_next_task(rq); |
dd41f596 IM |
2802 | if (p) |
2803 | return p; | |
dd41f596 | 2804 | } |
34f971f6 PZ |
2805 | |
2806 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2807 | } |
1da177e4 | 2808 | |
dd41f596 | 2809 | /* |
c259e01a | 2810 | * __schedule() is the main scheduler function. |
edde96ea PE |
2811 | * |
2812 | * The main means of driving the scheduler and thus entering this function are: | |
2813 | * | |
2814 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2815 | * | |
2816 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2817 | * paths. For example, see arch/x86/entry_64.S. | |
2818 | * | |
2819 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2820 | * interrupt handler scheduler_tick(). | |
2821 | * | |
2822 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2823 | * task to the run-queue and that's it. | |
2824 | * | |
2825 | * Now, if the new task added to the run-queue preempts the current | |
2826 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2827 | * called on the nearest possible occasion: | |
2828 | * | |
2829 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2830 | * | |
2831 | * - in syscall or exception context, at the next outmost | |
2832 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2833 | * spin_unlock()!) | |
2834 | * | |
2835 | * - in IRQ context, return from interrupt-handler to | |
2836 | * preemptible context | |
2837 | * | |
2838 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2839 | * then at the next: | |
2840 | * | |
2841 | * - cond_resched() call | |
2842 | * - explicit schedule() call | |
2843 | * - return from syscall or exception to user-space | |
2844 | * - return from interrupt-handler to user-space | |
dd41f596 | 2845 | */ |
c259e01a | 2846 | static void __sched __schedule(void) |
dd41f596 IM |
2847 | { |
2848 | struct task_struct *prev, *next; | |
67ca7bde | 2849 | unsigned long *switch_count; |
dd41f596 | 2850 | struct rq *rq; |
31656519 | 2851 | int cpu; |
dd41f596 | 2852 | |
ff743345 PZ |
2853 | need_resched: |
2854 | preempt_disable(); | |
dd41f596 IM |
2855 | cpu = smp_processor_id(); |
2856 | rq = cpu_rq(cpu); | |
25502a6c | 2857 | rcu_note_context_switch(cpu); |
dd41f596 | 2858 | prev = rq->curr; |
dd41f596 | 2859 | |
dd41f596 | 2860 | schedule_debug(prev); |
1da177e4 | 2861 | |
31656519 | 2862 | if (sched_feat(HRTICK)) |
f333fdc9 | 2863 | hrtick_clear(rq); |
8f4d37ec | 2864 | |
05fa785c | 2865 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2866 | |
246d86b5 | 2867 | switch_count = &prev->nivcsw; |
1da177e4 | 2868 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2869 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2870 | prev->state = TASK_RUNNING; |
21aa9af0 | 2871 | } else { |
2acca55e PZ |
2872 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2873 | prev->on_rq = 0; | |
2874 | ||
21aa9af0 | 2875 | /* |
2acca55e PZ |
2876 | * If a worker went to sleep, notify and ask workqueue |
2877 | * whether it wants to wake up a task to maintain | |
2878 | * concurrency. | |
21aa9af0 TH |
2879 | */ |
2880 | if (prev->flags & PF_WQ_WORKER) { | |
2881 | struct task_struct *to_wakeup; | |
2882 | ||
2883 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2884 | if (to_wakeup) | |
2885 | try_to_wake_up_local(to_wakeup); | |
2886 | } | |
21aa9af0 | 2887 | } |
dd41f596 | 2888 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2889 | } |
2890 | ||
3f029d3c | 2891 | pre_schedule(rq, prev); |
f65eda4f | 2892 | |
dd41f596 | 2893 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2894 | idle_balance(cpu, rq); |
1da177e4 | 2895 | |
df1c99d4 | 2896 | put_prev_task(rq, prev); |
b67802ea | 2897 | next = pick_next_task(rq); |
f26f9aff MG |
2898 | clear_tsk_need_resched(prev); |
2899 | rq->skip_clock_update = 0; | |
1da177e4 | 2900 | |
1da177e4 | 2901 | if (likely(prev != next)) { |
1da177e4 LT |
2902 | rq->nr_switches++; |
2903 | rq->curr = next; | |
2904 | ++*switch_count; | |
2905 | ||
dd41f596 | 2906 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2907 | /* |
246d86b5 ON |
2908 | * The context switch have flipped the stack from under us |
2909 | * and restored the local variables which were saved when | |
2910 | * this task called schedule() in the past. prev == current | |
2911 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2912 | */ |
2913 | cpu = smp_processor_id(); | |
2914 | rq = cpu_rq(cpu); | |
1da177e4 | 2915 | } else |
05fa785c | 2916 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2917 | |
3f029d3c | 2918 | post_schedule(rq); |
1da177e4 | 2919 | |
ba74c144 | 2920 | sched_preempt_enable_no_resched(); |
ff743345 | 2921 | if (need_resched()) |
1da177e4 LT |
2922 | goto need_resched; |
2923 | } | |
c259e01a | 2924 | |
9c40cef2 TG |
2925 | static inline void sched_submit_work(struct task_struct *tsk) |
2926 | { | |
3c7d5184 | 2927 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2928 | return; |
2929 | /* | |
2930 | * If we are going to sleep and we have plugged IO queued, | |
2931 | * make sure to submit it to avoid deadlocks. | |
2932 | */ | |
2933 | if (blk_needs_flush_plug(tsk)) | |
2934 | blk_schedule_flush_plug(tsk); | |
2935 | } | |
2936 | ||
6ebbe7a0 | 2937 | asmlinkage void __sched schedule(void) |
c259e01a | 2938 | { |
9c40cef2 TG |
2939 | struct task_struct *tsk = current; |
2940 | ||
2941 | sched_submit_work(tsk); | |
c259e01a TG |
2942 | __schedule(); |
2943 | } | |
1da177e4 LT |
2944 | EXPORT_SYMBOL(schedule); |
2945 | ||
20ab65e3 FW |
2946 | #ifdef CONFIG_RCU_USER_QS |
2947 | asmlinkage void __sched schedule_user(void) | |
2948 | { | |
2949 | /* | |
2950 | * If we come here after a random call to set_need_resched(), | |
2951 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2952 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2953 | * we find a better solution. | |
2954 | */ | |
2955 | rcu_user_exit(); | |
2956 | schedule(); | |
2957 | rcu_user_enter(); | |
2958 | } | |
2959 | #endif | |
2960 | ||
c5491ea7 TG |
2961 | /** |
2962 | * schedule_preempt_disabled - called with preemption disabled | |
2963 | * | |
2964 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2965 | */ | |
2966 | void __sched schedule_preempt_disabled(void) | |
2967 | { | |
ba74c144 | 2968 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2969 | schedule(); |
2970 | preempt_disable(); | |
2971 | } | |
2972 | ||
c08f7829 | 2973 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 2974 | |
c6eb3dda PZ |
2975 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
2976 | { | |
c6eb3dda | 2977 | if (lock->owner != owner) |
307bf980 | 2978 | return false; |
0d66bf6d PZ |
2979 | |
2980 | /* | |
c6eb3dda PZ |
2981 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
2982 | * lock->owner still matches owner, if that fails, owner might | |
2983 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
2984 | * ensures the memory stays valid. | |
0d66bf6d | 2985 | */ |
c6eb3dda | 2986 | barrier(); |
0d66bf6d | 2987 | |
307bf980 | 2988 | return owner->on_cpu; |
c6eb3dda | 2989 | } |
0d66bf6d | 2990 | |
c6eb3dda PZ |
2991 | /* |
2992 | * Look out! "owner" is an entirely speculative pointer | |
2993 | * access and not reliable. | |
2994 | */ | |
2995 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
2996 | { | |
2997 | if (!sched_feat(OWNER_SPIN)) | |
2998 | return 0; | |
0d66bf6d | 2999 | |
307bf980 | 3000 | rcu_read_lock(); |
c6eb3dda PZ |
3001 | while (owner_running(lock, owner)) { |
3002 | if (need_resched()) | |
307bf980 | 3003 | break; |
0d66bf6d | 3004 | |
335d7afb | 3005 | arch_mutex_cpu_relax(); |
0d66bf6d | 3006 | } |
307bf980 | 3007 | rcu_read_unlock(); |
4b402210 | 3008 | |
c6eb3dda | 3009 | /* |
307bf980 TG |
3010 | * We break out the loop above on need_resched() and when the |
3011 | * owner changed, which is a sign for heavy contention. Return | |
3012 | * success only when lock->owner is NULL. | |
c6eb3dda | 3013 | */ |
307bf980 | 3014 | return lock->owner == NULL; |
0d66bf6d PZ |
3015 | } |
3016 | #endif | |
3017 | ||
1da177e4 LT |
3018 | #ifdef CONFIG_PREEMPT |
3019 | /* | |
2ed6e34f | 3020 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3021 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3022 | * occur there and call schedule directly. |
3023 | */ | |
d1f74e20 | 3024 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3025 | { |
3026 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3027 | |
1da177e4 LT |
3028 | /* |
3029 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3030 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3031 | */ |
beed33a8 | 3032 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3033 | return; |
3034 | ||
3a5c359a | 3035 | do { |
d1f74e20 | 3036 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3037 | __schedule(); |
d1f74e20 | 3038 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3039 | |
3a5c359a AK |
3040 | /* |
3041 | * Check again in case we missed a preemption opportunity | |
3042 | * between schedule and now. | |
3043 | */ | |
3044 | barrier(); | |
5ed0cec0 | 3045 | } while (need_resched()); |
1da177e4 | 3046 | } |
1da177e4 LT |
3047 | EXPORT_SYMBOL(preempt_schedule); |
3048 | ||
3049 | /* | |
2ed6e34f | 3050 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3051 | * off of irq context. |
3052 | * Note, that this is called and return with irqs disabled. This will | |
3053 | * protect us against recursive calling from irq. | |
3054 | */ | |
3055 | asmlinkage void __sched preempt_schedule_irq(void) | |
3056 | { | |
3057 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3058 | |
2ed6e34f | 3059 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3060 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3061 | ||
90a340ed | 3062 | rcu_user_exit(); |
3a5c359a AK |
3063 | do { |
3064 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3065 | local_irq_enable(); |
c259e01a | 3066 | __schedule(); |
3a5c359a | 3067 | local_irq_disable(); |
3a5c359a | 3068 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3069 | |
3a5c359a AK |
3070 | /* |
3071 | * Check again in case we missed a preemption opportunity | |
3072 | * between schedule and now. | |
3073 | */ | |
3074 | barrier(); | |
5ed0cec0 | 3075 | } while (need_resched()); |
1da177e4 LT |
3076 | } |
3077 | ||
3078 | #endif /* CONFIG_PREEMPT */ | |
3079 | ||
63859d4f | 3080 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3081 | void *key) |
1da177e4 | 3082 | { |
63859d4f | 3083 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3084 | } |
1da177e4 LT |
3085 | EXPORT_SYMBOL(default_wake_function); |
3086 | ||
3087 | /* | |
41a2d6cf IM |
3088 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3089 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3090 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3091 | * | |
3092 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3093 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3094 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3095 | */ | |
78ddb08f | 3096 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3097 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3098 | { |
2e45874c | 3099 | wait_queue_t *curr, *next; |
1da177e4 | 3100 | |
2e45874c | 3101 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3102 | unsigned flags = curr->flags; |
3103 | ||
63859d4f | 3104 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3105 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3106 | break; |
3107 | } | |
3108 | } | |
3109 | ||
3110 | /** | |
3111 | * __wake_up - wake up threads blocked on a waitqueue. | |
3112 | * @q: the waitqueue | |
3113 | * @mode: which threads | |
3114 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3115 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3116 | * |
3117 | * It may be assumed that this function implies a write memory barrier before | |
3118 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3119 | */ |
7ad5b3a5 | 3120 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3121 | int nr_exclusive, void *key) |
1da177e4 LT |
3122 | { |
3123 | unsigned long flags; | |
3124 | ||
3125 | spin_lock_irqsave(&q->lock, flags); | |
3126 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3127 | spin_unlock_irqrestore(&q->lock, flags); | |
3128 | } | |
1da177e4 LT |
3129 | EXPORT_SYMBOL(__wake_up); |
3130 | ||
3131 | /* | |
3132 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3133 | */ | |
63b20011 | 3134 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3135 | { |
63b20011 | 3136 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3137 | } |
22c43c81 | 3138 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3139 | |
4ede816a DL |
3140 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3141 | { | |
3142 | __wake_up_common(q, mode, 1, 0, key); | |
3143 | } | |
bf294b41 | 3144 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3145 | |
1da177e4 | 3146 | /** |
4ede816a | 3147 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3148 | * @q: the waitqueue |
3149 | * @mode: which threads | |
3150 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3151 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3152 | * |
3153 | * The sync wakeup differs that the waker knows that it will schedule | |
3154 | * away soon, so while the target thread will be woken up, it will not | |
3155 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3156 | * with each other. This can prevent needless bouncing between CPUs. | |
3157 | * | |
3158 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3159 | * |
3160 | * It may be assumed that this function implies a write memory barrier before | |
3161 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3162 | */ |
4ede816a DL |
3163 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3164 | int nr_exclusive, void *key) | |
1da177e4 LT |
3165 | { |
3166 | unsigned long flags; | |
7d478721 | 3167 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3168 | |
3169 | if (unlikely(!q)) | |
3170 | return; | |
3171 | ||
3172 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3173 | wake_flags = 0; |
1da177e4 LT |
3174 | |
3175 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3176 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3177 | spin_unlock_irqrestore(&q->lock, flags); |
3178 | } | |
4ede816a DL |
3179 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3180 | ||
3181 | /* | |
3182 | * __wake_up_sync - see __wake_up_sync_key() | |
3183 | */ | |
3184 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3185 | { | |
3186 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3187 | } | |
1da177e4 LT |
3188 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3189 | ||
65eb3dc6 KD |
3190 | /** |
3191 | * complete: - signals a single thread waiting on this completion | |
3192 | * @x: holds the state of this particular completion | |
3193 | * | |
3194 | * This will wake up a single thread waiting on this completion. Threads will be | |
3195 | * awakened in the same order in which they were queued. | |
3196 | * | |
3197 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3198 | * |
3199 | * It may be assumed that this function implies a write memory barrier before | |
3200 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3201 | */ |
b15136e9 | 3202 | void complete(struct completion *x) |
1da177e4 LT |
3203 | { |
3204 | unsigned long flags; | |
3205 | ||
3206 | spin_lock_irqsave(&x->wait.lock, flags); | |
3207 | x->done++; | |
d9514f6c | 3208 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3209 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3210 | } | |
3211 | EXPORT_SYMBOL(complete); | |
3212 | ||
65eb3dc6 KD |
3213 | /** |
3214 | * complete_all: - signals all threads waiting on this completion | |
3215 | * @x: holds the state of this particular completion | |
3216 | * | |
3217 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3218 | * |
3219 | * It may be assumed that this function implies a write memory barrier before | |
3220 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3221 | */ |
b15136e9 | 3222 | void complete_all(struct completion *x) |
1da177e4 LT |
3223 | { |
3224 | unsigned long flags; | |
3225 | ||
3226 | spin_lock_irqsave(&x->wait.lock, flags); | |
3227 | x->done += UINT_MAX/2; | |
d9514f6c | 3228 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3229 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3230 | } | |
3231 | EXPORT_SYMBOL(complete_all); | |
3232 | ||
8cbbe86d AK |
3233 | static inline long __sched |
3234 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3235 | { |
1da177e4 LT |
3236 | if (!x->done) { |
3237 | DECLARE_WAITQUEUE(wait, current); | |
3238 | ||
a93d2f17 | 3239 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3240 | do { |
94d3d824 | 3241 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3242 | timeout = -ERESTARTSYS; |
3243 | break; | |
8cbbe86d AK |
3244 | } |
3245 | __set_current_state(state); | |
1da177e4 LT |
3246 | spin_unlock_irq(&x->wait.lock); |
3247 | timeout = schedule_timeout(timeout); | |
3248 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3249 | } while (!x->done && timeout); |
1da177e4 | 3250 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3251 | if (!x->done) |
3252 | return timeout; | |
1da177e4 LT |
3253 | } |
3254 | x->done--; | |
ea71a546 | 3255 | return timeout ?: 1; |
1da177e4 | 3256 | } |
1da177e4 | 3257 | |
8cbbe86d AK |
3258 | static long __sched |
3259 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3260 | { |
1da177e4 LT |
3261 | might_sleep(); |
3262 | ||
3263 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3264 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3265 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3266 | return timeout; |
3267 | } | |
1da177e4 | 3268 | |
65eb3dc6 KD |
3269 | /** |
3270 | * wait_for_completion: - waits for completion of a task | |
3271 | * @x: holds the state of this particular completion | |
3272 | * | |
3273 | * This waits to be signaled for completion of a specific task. It is NOT | |
3274 | * interruptible and there is no timeout. | |
3275 | * | |
3276 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3277 | * and interrupt capability. Also see complete(). | |
3278 | */ | |
b15136e9 | 3279 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3280 | { |
3281 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3282 | } |
8cbbe86d | 3283 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3284 | |
65eb3dc6 KD |
3285 | /** |
3286 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3287 | * @x: holds the state of this particular completion | |
3288 | * @timeout: timeout value in jiffies | |
3289 | * | |
3290 | * This waits for either a completion of a specific task to be signaled or for a | |
3291 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3292 | * interruptible. | |
c6dc7f05 BF |
3293 | * |
3294 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3295 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3296 | */ |
b15136e9 | 3297 | unsigned long __sched |
8cbbe86d | 3298 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3299 | { |
8cbbe86d | 3300 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3301 | } |
8cbbe86d | 3302 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3303 | |
65eb3dc6 KD |
3304 | /** |
3305 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3306 | * @x: holds the state of this particular completion | |
3307 | * | |
3308 | * This waits for completion of a specific task to be signaled. It is | |
3309 | * interruptible. | |
c6dc7f05 BF |
3310 | * |
3311 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3312 | */ |
8cbbe86d | 3313 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3314 | { |
51e97990 AK |
3315 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3316 | if (t == -ERESTARTSYS) | |
3317 | return t; | |
3318 | return 0; | |
0fec171c | 3319 | } |
8cbbe86d | 3320 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3321 | |
65eb3dc6 KD |
3322 | /** |
3323 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3324 | * @x: holds the state of this particular completion | |
3325 | * @timeout: timeout value in jiffies | |
3326 | * | |
3327 | * This waits for either a completion of a specific task to be signaled or for a | |
3328 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3329 | * |
3330 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3331 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3332 | */ |
6bf41237 | 3333 | long __sched |
8cbbe86d AK |
3334 | wait_for_completion_interruptible_timeout(struct completion *x, |
3335 | unsigned long timeout) | |
0fec171c | 3336 | { |
8cbbe86d | 3337 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3338 | } |
8cbbe86d | 3339 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3340 | |
65eb3dc6 KD |
3341 | /** |
3342 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3343 | * @x: holds the state of this particular completion | |
3344 | * | |
3345 | * This waits to be signaled for completion of a specific task. It can be | |
3346 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3347 | * |
3348 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3349 | */ |
009e577e MW |
3350 | int __sched wait_for_completion_killable(struct completion *x) |
3351 | { | |
3352 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3353 | if (t == -ERESTARTSYS) | |
3354 | return t; | |
3355 | return 0; | |
3356 | } | |
3357 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3358 | ||
0aa12fb4 SW |
3359 | /** |
3360 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3361 | * @x: holds the state of this particular completion | |
3362 | * @timeout: timeout value in jiffies | |
3363 | * | |
3364 | * This waits for either a completion of a specific task to be | |
3365 | * signaled or for a specified timeout to expire. It can be | |
3366 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3367 | * |
3368 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3369 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3370 | */ |
6bf41237 | 3371 | long __sched |
0aa12fb4 SW |
3372 | wait_for_completion_killable_timeout(struct completion *x, |
3373 | unsigned long timeout) | |
3374 | { | |
3375 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3376 | } | |
3377 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3378 | ||
be4de352 DC |
3379 | /** |
3380 | * try_wait_for_completion - try to decrement a completion without blocking | |
3381 | * @x: completion structure | |
3382 | * | |
3383 | * Returns: 0 if a decrement cannot be done without blocking | |
3384 | * 1 if a decrement succeeded. | |
3385 | * | |
3386 | * If a completion is being used as a counting completion, | |
3387 | * attempt to decrement the counter without blocking. This | |
3388 | * enables us to avoid waiting if the resource the completion | |
3389 | * is protecting is not available. | |
3390 | */ | |
3391 | bool try_wait_for_completion(struct completion *x) | |
3392 | { | |
7539a3b3 | 3393 | unsigned long flags; |
be4de352 DC |
3394 | int ret = 1; |
3395 | ||
7539a3b3 | 3396 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3397 | if (!x->done) |
3398 | ret = 0; | |
3399 | else | |
3400 | x->done--; | |
7539a3b3 | 3401 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3402 | return ret; |
3403 | } | |
3404 | EXPORT_SYMBOL(try_wait_for_completion); | |
3405 | ||
3406 | /** | |
3407 | * completion_done - Test to see if a completion has any waiters | |
3408 | * @x: completion structure | |
3409 | * | |
3410 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3411 | * 1 if there are no waiters. | |
3412 | * | |
3413 | */ | |
3414 | bool completion_done(struct completion *x) | |
3415 | { | |
7539a3b3 | 3416 | unsigned long flags; |
be4de352 DC |
3417 | int ret = 1; |
3418 | ||
7539a3b3 | 3419 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3420 | if (!x->done) |
3421 | ret = 0; | |
7539a3b3 | 3422 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3423 | return ret; |
3424 | } | |
3425 | EXPORT_SYMBOL(completion_done); | |
3426 | ||
8cbbe86d AK |
3427 | static long __sched |
3428 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3429 | { |
0fec171c IM |
3430 | unsigned long flags; |
3431 | wait_queue_t wait; | |
3432 | ||
3433 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3434 | |
8cbbe86d | 3435 | __set_current_state(state); |
1da177e4 | 3436 | |
8cbbe86d AK |
3437 | spin_lock_irqsave(&q->lock, flags); |
3438 | __add_wait_queue(q, &wait); | |
3439 | spin_unlock(&q->lock); | |
3440 | timeout = schedule_timeout(timeout); | |
3441 | spin_lock_irq(&q->lock); | |
3442 | __remove_wait_queue(q, &wait); | |
3443 | spin_unlock_irqrestore(&q->lock, flags); | |
3444 | ||
3445 | return timeout; | |
3446 | } | |
3447 | ||
3448 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3449 | { | |
3450 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3451 | } |
1da177e4 LT |
3452 | EXPORT_SYMBOL(interruptible_sleep_on); |
3453 | ||
0fec171c | 3454 | long __sched |
95cdf3b7 | 3455 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3456 | { |
8cbbe86d | 3457 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3458 | } |
1da177e4 LT |
3459 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3460 | ||
0fec171c | 3461 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3462 | { |
8cbbe86d | 3463 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3464 | } |
1da177e4 LT |
3465 | EXPORT_SYMBOL(sleep_on); |
3466 | ||
0fec171c | 3467 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3468 | { |
8cbbe86d | 3469 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3470 | } |
1da177e4 LT |
3471 | EXPORT_SYMBOL(sleep_on_timeout); |
3472 | ||
b29739f9 IM |
3473 | #ifdef CONFIG_RT_MUTEXES |
3474 | ||
3475 | /* | |
3476 | * rt_mutex_setprio - set the current priority of a task | |
3477 | * @p: task | |
3478 | * @prio: prio value (kernel-internal form) | |
3479 | * | |
3480 | * This function changes the 'effective' priority of a task. It does | |
3481 | * not touch ->normal_prio like __setscheduler(). | |
3482 | * | |
3483 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3484 | */ | |
36c8b586 | 3485 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3486 | { |
83b699ed | 3487 | int oldprio, on_rq, running; |
70b97a7f | 3488 | struct rq *rq; |
83ab0aa0 | 3489 | const struct sched_class *prev_class; |
b29739f9 IM |
3490 | |
3491 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3492 | ||
0122ec5b | 3493 | rq = __task_rq_lock(p); |
b29739f9 | 3494 | |
1c4dd99b TG |
3495 | /* |
3496 | * Idle task boosting is a nono in general. There is one | |
3497 | * exception, when PREEMPT_RT and NOHZ is active: | |
3498 | * | |
3499 | * The idle task calls get_next_timer_interrupt() and holds | |
3500 | * the timer wheel base->lock on the CPU and another CPU wants | |
3501 | * to access the timer (probably to cancel it). We can safely | |
3502 | * ignore the boosting request, as the idle CPU runs this code | |
3503 | * with interrupts disabled and will complete the lock | |
3504 | * protected section without being interrupted. So there is no | |
3505 | * real need to boost. | |
3506 | */ | |
3507 | if (unlikely(p == rq->idle)) { | |
3508 | WARN_ON(p != rq->curr); | |
3509 | WARN_ON(p->pi_blocked_on); | |
3510 | goto out_unlock; | |
3511 | } | |
3512 | ||
a8027073 | 3513 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3514 | oldprio = p->prio; |
83ab0aa0 | 3515 | prev_class = p->sched_class; |
fd2f4419 | 3516 | on_rq = p->on_rq; |
051a1d1a | 3517 | running = task_current(rq, p); |
0e1f3483 | 3518 | if (on_rq) |
69be72c1 | 3519 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3520 | if (running) |
3521 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3522 | |
3523 | if (rt_prio(prio)) | |
3524 | p->sched_class = &rt_sched_class; | |
3525 | else | |
3526 | p->sched_class = &fair_sched_class; | |
3527 | ||
b29739f9 IM |
3528 | p->prio = prio; |
3529 | ||
0e1f3483 HS |
3530 | if (running) |
3531 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3532 | if (on_rq) |
371fd7e7 | 3533 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3534 | |
da7a735e | 3535 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3536 | out_unlock: |
0122ec5b | 3537 | __task_rq_unlock(rq); |
b29739f9 | 3538 | } |
b29739f9 | 3539 | #endif |
36c8b586 | 3540 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3541 | { |
dd41f596 | 3542 | int old_prio, delta, on_rq; |
1da177e4 | 3543 | unsigned long flags; |
70b97a7f | 3544 | struct rq *rq; |
1da177e4 LT |
3545 | |
3546 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3547 | return; | |
3548 | /* | |
3549 | * We have to be careful, if called from sys_setpriority(), | |
3550 | * the task might be in the middle of scheduling on another CPU. | |
3551 | */ | |
3552 | rq = task_rq_lock(p, &flags); | |
3553 | /* | |
3554 | * The RT priorities are set via sched_setscheduler(), but we still | |
3555 | * allow the 'normal' nice value to be set - but as expected | |
3556 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3557 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3558 | */ |
e05606d3 | 3559 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3560 | p->static_prio = NICE_TO_PRIO(nice); |
3561 | goto out_unlock; | |
3562 | } | |
fd2f4419 | 3563 | on_rq = p->on_rq; |
c09595f6 | 3564 | if (on_rq) |
69be72c1 | 3565 | dequeue_task(rq, p, 0); |
1da177e4 | 3566 | |
1da177e4 | 3567 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3568 | set_load_weight(p); |
b29739f9 IM |
3569 | old_prio = p->prio; |
3570 | p->prio = effective_prio(p); | |
3571 | delta = p->prio - old_prio; | |
1da177e4 | 3572 | |
dd41f596 | 3573 | if (on_rq) { |
371fd7e7 | 3574 | enqueue_task(rq, p, 0); |
1da177e4 | 3575 | /* |
d5f9f942 AM |
3576 | * If the task increased its priority or is running and |
3577 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3578 | */ |
d5f9f942 | 3579 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3580 | resched_task(rq->curr); |
3581 | } | |
3582 | out_unlock: | |
0122ec5b | 3583 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3584 | } |
1da177e4 LT |
3585 | EXPORT_SYMBOL(set_user_nice); |
3586 | ||
e43379f1 MM |
3587 | /* |
3588 | * can_nice - check if a task can reduce its nice value | |
3589 | * @p: task | |
3590 | * @nice: nice value | |
3591 | */ | |
36c8b586 | 3592 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3593 | { |
024f4747 MM |
3594 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3595 | int nice_rlim = 20 - nice; | |
48f24c4d | 3596 | |
78d7d407 | 3597 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3598 | capable(CAP_SYS_NICE)); |
3599 | } | |
3600 | ||
1da177e4 LT |
3601 | #ifdef __ARCH_WANT_SYS_NICE |
3602 | ||
3603 | /* | |
3604 | * sys_nice - change the priority of the current process. | |
3605 | * @increment: priority increment | |
3606 | * | |
3607 | * sys_setpriority is a more generic, but much slower function that | |
3608 | * does similar things. | |
3609 | */ | |
5add95d4 | 3610 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3611 | { |
48f24c4d | 3612 | long nice, retval; |
1da177e4 LT |
3613 | |
3614 | /* | |
3615 | * Setpriority might change our priority at the same moment. | |
3616 | * We don't have to worry. Conceptually one call occurs first | |
3617 | * and we have a single winner. | |
3618 | */ | |
e43379f1 MM |
3619 | if (increment < -40) |
3620 | increment = -40; | |
1da177e4 LT |
3621 | if (increment > 40) |
3622 | increment = 40; | |
3623 | ||
2b8f836f | 3624 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3625 | if (nice < -20) |
3626 | nice = -20; | |
3627 | if (nice > 19) | |
3628 | nice = 19; | |
3629 | ||
e43379f1 MM |
3630 | if (increment < 0 && !can_nice(current, nice)) |
3631 | return -EPERM; | |
3632 | ||
1da177e4 LT |
3633 | retval = security_task_setnice(current, nice); |
3634 | if (retval) | |
3635 | return retval; | |
3636 | ||
3637 | set_user_nice(current, nice); | |
3638 | return 0; | |
3639 | } | |
3640 | ||
3641 | #endif | |
3642 | ||
3643 | /** | |
3644 | * task_prio - return the priority value of a given task. | |
3645 | * @p: the task in question. | |
3646 | * | |
3647 | * This is the priority value as seen by users in /proc. | |
3648 | * RT tasks are offset by -200. Normal tasks are centered | |
3649 | * around 0, value goes from -16 to +15. | |
3650 | */ | |
36c8b586 | 3651 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3652 | { |
3653 | return p->prio - MAX_RT_PRIO; | |
3654 | } | |
3655 | ||
3656 | /** | |
3657 | * task_nice - return the nice value of a given task. | |
3658 | * @p: the task in question. | |
3659 | */ | |
36c8b586 | 3660 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3661 | { |
3662 | return TASK_NICE(p); | |
3663 | } | |
150d8bed | 3664 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3665 | |
3666 | /** | |
3667 | * idle_cpu - is a given cpu idle currently? | |
3668 | * @cpu: the processor in question. | |
3669 | */ | |
3670 | int idle_cpu(int cpu) | |
3671 | { | |
908a3283 TG |
3672 | struct rq *rq = cpu_rq(cpu); |
3673 | ||
3674 | if (rq->curr != rq->idle) | |
3675 | return 0; | |
3676 | ||
3677 | if (rq->nr_running) | |
3678 | return 0; | |
3679 | ||
3680 | #ifdef CONFIG_SMP | |
3681 | if (!llist_empty(&rq->wake_list)) | |
3682 | return 0; | |
3683 | #endif | |
3684 | ||
3685 | return 1; | |
1da177e4 LT |
3686 | } |
3687 | ||
1da177e4 LT |
3688 | /** |
3689 | * idle_task - return the idle task for a given cpu. | |
3690 | * @cpu: the processor in question. | |
3691 | */ | |
36c8b586 | 3692 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3693 | { |
3694 | return cpu_rq(cpu)->idle; | |
3695 | } | |
3696 | ||
3697 | /** | |
3698 | * find_process_by_pid - find a process with a matching PID value. | |
3699 | * @pid: the pid in question. | |
3700 | */ | |
a9957449 | 3701 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3702 | { |
228ebcbe | 3703 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3704 | } |
3705 | ||
3706 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3707 | static void |
3708 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3709 | { |
1da177e4 LT |
3710 | p->policy = policy; |
3711 | p->rt_priority = prio; | |
b29739f9 IM |
3712 | p->normal_prio = normal_prio(p); |
3713 | /* we are holding p->pi_lock already */ | |
3714 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3715 | if (rt_prio(p->prio)) |
3716 | p->sched_class = &rt_sched_class; | |
3717 | else | |
3718 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3719 | set_load_weight(p); |
1da177e4 LT |
3720 | } |
3721 | ||
c69e8d9c DH |
3722 | /* |
3723 | * check the target process has a UID that matches the current process's | |
3724 | */ | |
3725 | static bool check_same_owner(struct task_struct *p) | |
3726 | { | |
3727 | const struct cred *cred = current_cred(), *pcred; | |
3728 | bool match; | |
3729 | ||
3730 | rcu_read_lock(); | |
3731 | pcred = __task_cred(p); | |
9c806aa0 EB |
3732 | match = (uid_eq(cred->euid, pcred->euid) || |
3733 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3734 | rcu_read_unlock(); |
3735 | return match; | |
3736 | } | |
3737 | ||
961ccddd | 3738 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3739 | const struct sched_param *param, bool user) |
1da177e4 | 3740 | { |
83b699ed | 3741 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3742 | unsigned long flags; |
83ab0aa0 | 3743 | const struct sched_class *prev_class; |
70b97a7f | 3744 | struct rq *rq; |
ca94c442 | 3745 | int reset_on_fork; |
1da177e4 | 3746 | |
66e5393a SR |
3747 | /* may grab non-irq protected spin_locks */ |
3748 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3749 | recheck: |
3750 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3751 | if (policy < 0) { |
3752 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3753 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3754 | } else { |
3755 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3756 | policy &= ~SCHED_RESET_ON_FORK; | |
3757 | ||
3758 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3759 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3760 | policy != SCHED_IDLE) | |
3761 | return -EINVAL; | |
3762 | } | |
3763 | ||
1da177e4 LT |
3764 | /* |
3765 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3766 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3767 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3768 | */ |
3769 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3770 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3771 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3772 | return -EINVAL; |
e05606d3 | 3773 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3774 | return -EINVAL; |
3775 | ||
37e4ab3f OC |
3776 | /* |
3777 | * Allow unprivileged RT tasks to decrease priority: | |
3778 | */ | |
961ccddd | 3779 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3780 | if (rt_policy(policy)) { |
a44702e8 ON |
3781 | unsigned long rlim_rtprio = |
3782 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3783 | |
3784 | /* can't set/change the rt policy */ | |
3785 | if (policy != p->policy && !rlim_rtprio) | |
3786 | return -EPERM; | |
3787 | ||
3788 | /* can't increase priority */ | |
3789 | if (param->sched_priority > p->rt_priority && | |
3790 | param->sched_priority > rlim_rtprio) | |
3791 | return -EPERM; | |
3792 | } | |
c02aa73b | 3793 | |
dd41f596 | 3794 | /* |
c02aa73b DH |
3795 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3796 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3797 | */ |
c02aa73b DH |
3798 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3799 | if (!can_nice(p, TASK_NICE(p))) | |
3800 | return -EPERM; | |
3801 | } | |
5fe1d75f | 3802 | |
37e4ab3f | 3803 | /* can't change other user's priorities */ |
c69e8d9c | 3804 | if (!check_same_owner(p)) |
37e4ab3f | 3805 | return -EPERM; |
ca94c442 LP |
3806 | |
3807 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3808 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3809 | return -EPERM; | |
37e4ab3f | 3810 | } |
1da177e4 | 3811 | |
725aad24 | 3812 | if (user) { |
b0ae1981 | 3813 | retval = security_task_setscheduler(p); |
725aad24 JF |
3814 | if (retval) |
3815 | return retval; | |
3816 | } | |
3817 | ||
b29739f9 IM |
3818 | /* |
3819 | * make sure no PI-waiters arrive (or leave) while we are | |
3820 | * changing the priority of the task: | |
0122ec5b | 3821 | * |
25985edc | 3822 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3823 | * runqueue lock must be held. |
3824 | */ | |
0122ec5b | 3825 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3826 | |
34f971f6 PZ |
3827 | /* |
3828 | * Changing the policy of the stop threads its a very bad idea | |
3829 | */ | |
3830 | if (p == rq->stop) { | |
0122ec5b | 3831 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3832 | return -EINVAL; |
3833 | } | |
3834 | ||
a51e9198 DF |
3835 | /* |
3836 | * If not changing anything there's no need to proceed further: | |
3837 | */ | |
3838 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
3839 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 3840 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3841 | return 0; |
3842 | } | |
3843 | ||
dc61b1d6 PZ |
3844 | #ifdef CONFIG_RT_GROUP_SCHED |
3845 | if (user) { | |
3846 | /* | |
3847 | * Do not allow realtime tasks into groups that have no runtime | |
3848 | * assigned. | |
3849 | */ | |
3850 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3851 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3852 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3853 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3854 | return -EPERM; |
3855 | } | |
3856 | } | |
3857 | #endif | |
3858 | ||
1da177e4 LT |
3859 | /* recheck policy now with rq lock held */ |
3860 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3861 | policy = oldpolicy = -1; | |
0122ec5b | 3862 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3863 | goto recheck; |
3864 | } | |
fd2f4419 | 3865 | on_rq = p->on_rq; |
051a1d1a | 3866 | running = task_current(rq, p); |
0e1f3483 | 3867 | if (on_rq) |
4ca9b72b | 3868 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3869 | if (running) |
3870 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3871 | |
ca94c442 LP |
3872 | p->sched_reset_on_fork = reset_on_fork; |
3873 | ||
1da177e4 | 3874 | oldprio = p->prio; |
83ab0aa0 | 3875 | prev_class = p->sched_class; |
dd41f596 | 3876 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 3877 | |
0e1f3483 HS |
3878 | if (running) |
3879 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3880 | if (on_rq) |
4ca9b72b | 3881 | enqueue_task(rq, p, 0); |
cb469845 | 3882 | |
da7a735e | 3883 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3884 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3885 | |
95e02ca9 TG |
3886 | rt_mutex_adjust_pi(p); |
3887 | ||
1da177e4 LT |
3888 | return 0; |
3889 | } | |
961ccddd RR |
3890 | |
3891 | /** | |
3892 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3893 | * @p: the task in question. | |
3894 | * @policy: new policy. | |
3895 | * @param: structure containing the new RT priority. | |
3896 | * | |
3897 | * NOTE that the task may be already dead. | |
3898 | */ | |
3899 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3900 | const struct sched_param *param) |
961ccddd RR |
3901 | { |
3902 | return __sched_setscheduler(p, policy, param, true); | |
3903 | } | |
1da177e4 LT |
3904 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3905 | ||
961ccddd RR |
3906 | /** |
3907 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3908 | * @p: the task in question. | |
3909 | * @policy: new policy. | |
3910 | * @param: structure containing the new RT priority. | |
3911 | * | |
3912 | * Just like sched_setscheduler, only don't bother checking if the | |
3913 | * current context has permission. For example, this is needed in | |
3914 | * stop_machine(): we create temporary high priority worker threads, | |
3915 | * but our caller might not have that capability. | |
3916 | */ | |
3917 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3918 | const struct sched_param *param) |
961ccddd RR |
3919 | { |
3920 | return __sched_setscheduler(p, policy, param, false); | |
3921 | } | |
3922 | ||
95cdf3b7 IM |
3923 | static int |
3924 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3925 | { |
1da177e4 LT |
3926 | struct sched_param lparam; |
3927 | struct task_struct *p; | |
36c8b586 | 3928 | int retval; |
1da177e4 LT |
3929 | |
3930 | if (!param || pid < 0) | |
3931 | return -EINVAL; | |
3932 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3933 | return -EFAULT; | |
5fe1d75f ON |
3934 | |
3935 | rcu_read_lock(); | |
3936 | retval = -ESRCH; | |
1da177e4 | 3937 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3938 | if (p != NULL) |
3939 | retval = sched_setscheduler(p, policy, &lparam); | |
3940 | rcu_read_unlock(); | |
36c8b586 | 3941 | |
1da177e4 LT |
3942 | return retval; |
3943 | } | |
3944 | ||
3945 | /** | |
3946 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3947 | * @pid: the pid in question. | |
3948 | * @policy: new policy. | |
3949 | * @param: structure containing the new RT priority. | |
3950 | */ | |
5add95d4 HC |
3951 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3952 | struct sched_param __user *, param) | |
1da177e4 | 3953 | { |
c21761f1 JB |
3954 | /* negative values for policy are not valid */ |
3955 | if (policy < 0) | |
3956 | return -EINVAL; | |
3957 | ||
1da177e4 LT |
3958 | return do_sched_setscheduler(pid, policy, param); |
3959 | } | |
3960 | ||
3961 | /** | |
3962 | * sys_sched_setparam - set/change the RT priority of a thread | |
3963 | * @pid: the pid in question. | |
3964 | * @param: structure containing the new RT priority. | |
3965 | */ | |
5add95d4 | 3966 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3967 | { |
3968 | return do_sched_setscheduler(pid, -1, param); | |
3969 | } | |
3970 | ||
3971 | /** | |
3972 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3973 | * @pid: the pid in question. | |
3974 | */ | |
5add95d4 | 3975 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3976 | { |
36c8b586 | 3977 | struct task_struct *p; |
3a5c359a | 3978 | int retval; |
1da177e4 LT |
3979 | |
3980 | if (pid < 0) | |
3a5c359a | 3981 | return -EINVAL; |
1da177e4 LT |
3982 | |
3983 | retval = -ESRCH; | |
5fe85be0 | 3984 | rcu_read_lock(); |
1da177e4 LT |
3985 | p = find_process_by_pid(pid); |
3986 | if (p) { | |
3987 | retval = security_task_getscheduler(p); | |
3988 | if (!retval) | |
ca94c442 LP |
3989 | retval = p->policy |
3990 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3991 | } |
5fe85be0 | 3992 | rcu_read_unlock(); |
1da177e4 LT |
3993 | return retval; |
3994 | } | |
3995 | ||
3996 | /** | |
ca94c442 | 3997 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3998 | * @pid: the pid in question. |
3999 | * @param: structure containing the RT priority. | |
4000 | */ | |
5add95d4 | 4001 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4002 | { |
4003 | struct sched_param lp; | |
36c8b586 | 4004 | struct task_struct *p; |
3a5c359a | 4005 | int retval; |
1da177e4 LT |
4006 | |
4007 | if (!param || pid < 0) | |
3a5c359a | 4008 | return -EINVAL; |
1da177e4 | 4009 | |
5fe85be0 | 4010 | rcu_read_lock(); |
1da177e4 LT |
4011 | p = find_process_by_pid(pid); |
4012 | retval = -ESRCH; | |
4013 | if (!p) | |
4014 | goto out_unlock; | |
4015 | ||
4016 | retval = security_task_getscheduler(p); | |
4017 | if (retval) | |
4018 | goto out_unlock; | |
4019 | ||
4020 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4021 | rcu_read_unlock(); |
1da177e4 LT |
4022 | |
4023 | /* | |
4024 | * This one might sleep, we cannot do it with a spinlock held ... | |
4025 | */ | |
4026 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4027 | ||
1da177e4 LT |
4028 | return retval; |
4029 | ||
4030 | out_unlock: | |
5fe85be0 | 4031 | rcu_read_unlock(); |
1da177e4 LT |
4032 | return retval; |
4033 | } | |
4034 | ||
96f874e2 | 4035 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4036 | { |
5a16f3d3 | 4037 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4038 | struct task_struct *p; |
4039 | int retval; | |
1da177e4 | 4040 | |
95402b38 | 4041 | get_online_cpus(); |
23f5d142 | 4042 | rcu_read_lock(); |
1da177e4 LT |
4043 | |
4044 | p = find_process_by_pid(pid); | |
4045 | if (!p) { | |
23f5d142 | 4046 | rcu_read_unlock(); |
95402b38 | 4047 | put_online_cpus(); |
1da177e4 LT |
4048 | return -ESRCH; |
4049 | } | |
4050 | ||
23f5d142 | 4051 | /* Prevent p going away */ |
1da177e4 | 4052 | get_task_struct(p); |
23f5d142 | 4053 | rcu_read_unlock(); |
1da177e4 | 4054 | |
5a16f3d3 RR |
4055 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4056 | retval = -ENOMEM; | |
4057 | goto out_put_task; | |
4058 | } | |
4059 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4060 | retval = -ENOMEM; | |
4061 | goto out_free_cpus_allowed; | |
4062 | } | |
1da177e4 | 4063 | retval = -EPERM; |
f1c84dae | 4064 | if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE)) |
1da177e4 LT |
4065 | goto out_unlock; |
4066 | ||
b0ae1981 | 4067 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4068 | if (retval) |
4069 | goto out_unlock; | |
4070 | ||
5a16f3d3 RR |
4071 | cpuset_cpus_allowed(p, cpus_allowed); |
4072 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4073 | again: |
5a16f3d3 | 4074 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4075 | |
8707d8b8 | 4076 | if (!retval) { |
5a16f3d3 RR |
4077 | cpuset_cpus_allowed(p, cpus_allowed); |
4078 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4079 | /* |
4080 | * We must have raced with a concurrent cpuset | |
4081 | * update. Just reset the cpus_allowed to the | |
4082 | * cpuset's cpus_allowed | |
4083 | */ | |
5a16f3d3 | 4084 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4085 | goto again; |
4086 | } | |
4087 | } | |
1da177e4 | 4088 | out_unlock: |
5a16f3d3 RR |
4089 | free_cpumask_var(new_mask); |
4090 | out_free_cpus_allowed: | |
4091 | free_cpumask_var(cpus_allowed); | |
4092 | out_put_task: | |
1da177e4 | 4093 | put_task_struct(p); |
95402b38 | 4094 | put_online_cpus(); |
1da177e4 LT |
4095 | return retval; |
4096 | } | |
4097 | ||
4098 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4099 | struct cpumask *new_mask) |
1da177e4 | 4100 | { |
96f874e2 RR |
4101 | if (len < cpumask_size()) |
4102 | cpumask_clear(new_mask); | |
4103 | else if (len > cpumask_size()) | |
4104 | len = cpumask_size(); | |
4105 | ||
1da177e4 LT |
4106 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4107 | } | |
4108 | ||
4109 | /** | |
4110 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4111 | * @pid: pid of the process | |
4112 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4113 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4114 | */ | |
5add95d4 HC |
4115 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4116 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4117 | { |
5a16f3d3 | 4118 | cpumask_var_t new_mask; |
1da177e4 LT |
4119 | int retval; |
4120 | ||
5a16f3d3 RR |
4121 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4122 | return -ENOMEM; | |
1da177e4 | 4123 | |
5a16f3d3 RR |
4124 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4125 | if (retval == 0) | |
4126 | retval = sched_setaffinity(pid, new_mask); | |
4127 | free_cpumask_var(new_mask); | |
4128 | return retval; | |
1da177e4 LT |
4129 | } |
4130 | ||
96f874e2 | 4131 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4132 | { |
36c8b586 | 4133 | struct task_struct *p; |
31605683 | 4134 | unsigned long flags; |
1da177e4 | 4135 | int retval; |
1da177e4 | 4136 | |
95402b38 | 4137 | get_online_cpus(); |
23f5d142 | 4138 | rcu_read_lock(); |
1da177e4 LT |
4139 | |
4140 | retval = -ESRCH; | |
4141 | p = find_process_by_pid(pid); | |
4142 | if (!p) | |
4143 | goto out_unlock; | |
4144 | ||
e7834f8f DQ |
4145 | retval = security_task_getscheduler(p); |
4146 | if (retval) | |
4147 | goto out_unlock; | |
4148 | ||
013fdb80 | 4149 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4150 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4151 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4152 | |
4153 | out_unlock: | |
23f5d142 | 4154 | rcu_read_unlock(); |
95402b38 | 4155 | put_online_cpus(); |
1da177e4 | 4156 | |
9531b62f | 4157 | return retval; |
1da177e4 LT |
4158 | } |
4159 | ||
4160 | /** | |
4161 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4162 | * @pid: pid of the process | |
4163 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4164 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4165 | */ | |
5add95d4 HC |
4166 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4167 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4168 | { |
4169 | int ret; | |
f17c8607 | 4170 | cpumask_var_t mask; |
1da177e4 | 4171 | |
84fba5ec | 4172 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4173 | return -EINVAL; |
4174 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4175 | return -EINVAL; |
4176 | ||
f17c8607 RR |
4177 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4178 | return -ENOMEM; | |
1da177e4 | 4179 | |
f17c8607 RR |
4180 | ret = sched_getaffinity(pid, mask); |
4181 | if (ret == 0) { | |
8bc037fb | 4182 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4183 | |
4184 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4185 | ret = -EFAULT; |
4186 | else | |
cd3d8031 | 4187 | ret = retlen; |
f17c8607 RR |
4188 | } |
4189 | free_cpumask_var(mask); | |
1da177e4 | 4190 | |
f17c8607 | 4191 | return ret; |
1da177e4 LT |
4192 | } |
4193 | ||
4194 | /** | |
4195 | * sys_sched_yield - yield the current processor to other threads. | |
4196 | * | |
dd41f596 IM |
4197 | * This function yields the current CPU to other tasks. If there are no |
4198 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4199 | */ |
5add95d4 | 4200 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4201 | { |
70b97a7f | 4202 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4203 | |
2d72376b | 4204 | schedstat_inc(rq, yld_count); |
4530d7ab | 4205 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4206 | |
4207 | /* | |
4208 | * Since we are going to call schedule() anyway, there's | |
4209 | * no need to preempt or enable interrupts: | |
4210 | */ | |
4211 | __release(rq->lock); | |
8a25d5de | 4212 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4213 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4214 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4215 | |
4216 | schedule(); | |
4217 | ||
4218 | return 0; | |
4219 | } | |
4220 | ||
d86ee480 PZ |
4221 | static inline int should_resched(void) |
4222 | { | |
4223 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4224 | } | |
4225 | ||
e7b38404 | 4226 | static void __cond_resched(void) |
1da177e4 | 4227 | { |
e7aaaa69 | 4228 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4229 | __schedule(); |
e7aaaa69 | 4230 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4231 | } |
4232 | ||
02b67cc3 | 4233 | int __sched _cond_resched(void) |
1da177e4 | 4234 | { |
d86ee480 | 4235 | if (should_resched()) { |
1da177e4 LT |
4236 | __cond_resched(); |
4237 | return 1; | |
4238 | } | |
4239 | return 0; | |
4240 | } | |
02b67cc3 | 4241 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4242 | |
4243 | /* | |
613afbf8 | 4244 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4245 | * call schedule, and on return reacquire the lock. |
4246 | * | |
41a2d6cf | 4247 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4248 | * operations here to prevent schedule() from being called twice (once via |
4249 | * spin_unlock(), once by hand). | |
4250 | */ | |
613afbf8 | 4251 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4252 | { |
d86ee480 | 4253 | int resched = should_resched(); |
6df3cecb JK |
4254 | int ret = 0; |
4255 | ||
f607c668 PZ |
4256 | lockdep_assert_held(lock); |
4257 | ||
95c354fe | 4258 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4259 | spin_unlock(lock); |
d86ee480 | 4260 | if (resched) |
95c354fe NP |
4261 | __cond_resched(); |
4262 | else | |
4263 | cpu_relax(); | |
6df3cecb | 4264 | ret = 1; |
1da177e4 | 4265 | spin_lock(lock); |
1da177e4 | 4266 | } |
6df3cecb | 4267 | return ret; |
1da177e4 | 4268 | } |
613afbf8 | 4269 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4270 | |
613afbf8 | 4271 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4272 | { |
4273 | BUG_ON(!in_softirq()); | |
4274 | ||
d86ee480 | 4275 | if (should_resched()) { |
98d82567 | 4276 | local_bh_enable(); |
1da177e4 LT |
4277 | __cond_resched(); |
4278 | local_bh_disable(); | |
4279 | return 1; | |
4280 | } | |
4281 | return 0; | |
4282 | } | |
613afbf8 | 4283 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4284 | |
1da177e4 LT |
4285 | /** |
4286 | * yield - yield the current processor to other threads. | |
4287 | * | |
8e3fabfd PZ |
4288 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4289 | * | |
4290 | * The scheduler is at all times free to pick the calling task as the most | |
4291 | * eligible task to run, if removing the yield() call from your code breaks | |
4292 | * it, its already broken. | |
4293 | * | |
4294 | * Typical broken usage is: | |
4295 | * | |
4296 | * while (!event) | |
4297 | * yield(); | |
4298 | * | |
4299 | * where one assumes that yield() will let 'the other' process run that will | |
4300 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4301 | * happen. Never use yield() as a progress guarantee!! | |
4302 | * | |
4303 | * If you want to use yield() to wait for something, use wait_event(). | |
4304 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4305 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4306 | */ |
4307 | void __sched yield(void) | |
4308 | { | |
4309 | set_current_state(TASK_RUNNING); | |
4310 | sys_sched_yield(); | |
4311 | } | |
1da177e4 LT |
4312 | EXPORT_SYMBOL(yield); |
4313 | ||
d95f4122 MG |
4314 | /** |
4315 | * yield_to - yield the current processor to another thread in | |
4316 | * your thread group, or accelerate that thread toward the | |
4317 | * processor it's on. | |
16addf95 RD |
4318 | * @p: target task |
4319 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4320 | * |
4321 | * It's the caller's job to ensure that the target task struct | |
4322 | * can't go away on us before we can do any checks. | |
4323 | * | |
4324 | * Returns true if we indeed boosted the target task. | |
4325 | */ | |
4326 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4327 | { | |
4328 | struct task_struct *curr = current; | |
4329 | struct rq *rq, *p_rq; | |
4330 | unsigned long flags; | |
4331 | bool yielded = 0; | |
4332 | ||
4333 | local_irq_save(flags); | |
4334 | rq = this_rq(); | |
4335 | ||
4336 | again: | |
4337 | p_rq = task_rq(p); | |
4338 | double_rq_lock(rq, p_rq); | |
4339 | while (task_rq(p) != p_rq) { | |
4340 | double_rq_unlock(rq, p_rq); | |
4341 | goto again; | |
4342 | } | |
4343 | ||
4344 | if (!curr->sched_class->yield_to_task) | |
4345 | goto out; | |
4346 | ||
4347 | if (curr->sched_class != p->sched_class) | |
4348 | goto out; | |
4349 | ||
4350 | if (task_running(p_rq, p) || p->state) | |
4351 | goto out; | |
4352 | ||
4353 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4354 | if (yielded) { |
d95f4122 | 4355 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4356 | /* |
4357 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4358 | * fairness. | |
4359 | */ | |
4360 | if (preempt && rq != p_rq) | |
4361 | resched_task(p_rq->curr); | |
4362 | } | |
d95f4122 MG |
4363 | |
4364 | out: | |
4365 | double_rq_unlock(rq, p_rq); | |
4366 | local_irq_restore(flags); | |
4367 | ||
4368 | if (yielded) | |
4369 | schedule(); | |
4370 | ||
4371 | return yielded; | |
4372 | } | |
4373 | EXPORT_SYMBOL_GPL(yield_to); | |
4374 | ||
1da177e4 | 4375 | /* |
41a2d6cf | 4376 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4377 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4378 | */ |
4379 | void __sched io_schedule(void) | |
4380 | { | |
54d35f29 | 4381 | struct rq *rq = raw_rq(); |
1da177e4 | 4382 | |
0ff92245 | 4383 | delayacct_blkio_start(); |
1da177e4 | 4384 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4385 | blk_flush_plug(current); |
8f0dfc34 | 4386 | current->in_iowait = 1; |
1da177e4 | 4387 | schedule(); |
8f0dfc34 | 4388 | current->in_iowait = 0; |
1da177e4 | 4389 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4390 | delayacct_blkio_end(); |
1da177e4 | 4391 | } |
1da177e4 LT |
4392 | EXPORT_SYMBOL(io_schedule); |
4393 | ||
4394 | long __sched io_schedule_timeout(long timeout) | |
4395 | { | |
54d35f29 | 4396 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4397 | long ret; |
4398 | ||
0ff92245 | 4399 | delayacct_blkio_start(); |
1da177e4 | 4400 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4401 | blk_flush_plug(current); |
8f0dfc34 | 4402 | current->in_iowait = 1; |
1da177e4 | 4403 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4404 | current->in_iowait = 0; |
1da177e4 | 4405 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4406 | delayacct_blkio_end(); |
1da177e4 LT |
4407 | return ret; |
4408 | } | |
4409 | ||
4410 | /** | |
4411 | * sys_sched_get_priority_max - return maximum RT priority. | |
4412 | * @policy: scheduling class. | |
4413 | * | |
4414 | * this syscall returns the maximum rt_priority that can be used | |
4415 | * by a given scheduling class. | |
4416 | */ | |
5add95d4 | 4417 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4418 | { |
4419 | int ret = -EINVAL; | |
4420 | ||
4421 | switch (policy) { | |
4422 | case SCHED_FIFO: | |
4423 | case SCHED_RR: | |
4424 | ret = MAX_USER_RT_PRIO-1; | |
4425 | break; | |
4426 | case SCHED_NORMAL: | |
b0a9499c | 4427 | case SCHED_BATCH: |
dd41f596 | 4428 | case SCHED_IDLE: |
1da177e4 LT |
4429 | ret = 0; |
4430 | break; | |
4431 | } | |
4432 | return ret; | |
4433 | } | |
4434 | ||
4435 | /** | |
4436 | * sys_sched_get_priority_min - return minimum RT priority. | |
4437 | * @policy: scheduling class. | |
4438 | * | |
4439 | * this syscall returns the minimum rt_priority that can be used | |
4440 | * by a given scheduling class. | |
4441 | */ | |
5add95d4 | 4442 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4443 | { |
4444 | int ret = -EINVAL; | |
4445 | ||
4446 | switch (policy) { | |
4447 | case SCHED_FIFO: | |
4448 | case SCHED_RR: | |
4449 | ret = 1; | |
4450 | break; | |
4451 | case SCHED_NORMAL: | |
b0a9499c | 4452 | case SCHED_BATCH: |
dd41f596 | 4453 | case SCHED_IDLE: |
1da177e4 LT |
4454 | ret = 0; |
4455 | } | |
4456 | return ret; | |
4457 | } | |
4458 | ||
4459 | /** | |
4460 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4461 | * @pid: pid of the process. | |
4462 | * @interval: userspace pointer to the timeslice value. | |
4463 | * | |
4464 | * this syscall writes the default timeslice value of a given process | |
4465 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4466 | */ | |
17da2bd9 | 4467 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4468 | struct timespec __user *, interval) |
1da177e4 | 4469 | { |
36c8b586 | 4470 | struct task_struct *p; |
a4ec24b4 | 4471 | unsigned int time_slice; |
dba091b9 TG |
4472 | unsigned long flags; |
4473 | struct rq *rq; | |
3a5c359a | 4474 | int retval; |
1da177e4 | 4475 | struct timespec t; |
1da177e4 LT |
4476 | |
4477 | if (pid < 0) | |
3a5c359a | 4478 | return -EINVAL; |
1da177e4 LT |
4479 | |
4480 | retval = -ESRCH; | |
1a551ae7 | 4481 | rcu_read_lock(); |
1da177e4 LT |
4482 | p = find_process_by_pid(pid); |
4483 | if (!p) | |
4484 | goto out_unlock; | |
4485 | ||
4486 | retval = security_task_getscheduler(p); | |
4487 | if (retval) | |
4488 | goto out_unlock; | |
4489 | ||
dba091b9 TG |
4490 | rq = task_rq_lock(p, &flags); |
4491 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4492 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4493 | |
1a551ae7 | 4494 | rcu_read_unlock(); |
a4ec24b4 | 4495 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4496 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4497 | return retval; |
3a5c359a | 4498 | |
1da177e4 | 4499 | out_unlock: |
1a551ae7 | 4500 | rcu_read_unlock(); |
1da177e4 LT |
4501 | return retval; |
4502 | } | |
4503 | ||
7c731e0a | 4504 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4505 | |
82a1fcb9 | 4506 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4507 | { |
1da177e4 | 4508 | unsigned long free = 0; |
36c8b586 | 4509 | unsigned state; |
1da177e4 | 4510 | |
1da177e4 | 4511 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4512 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4513 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4514 | #if BITS_PER_LONG == 32 |
1da177e4 | 4515 | if (state == TASK_RUNNING) |
3df0fc5b | 4516 | printk(KERN_CONT " running "); |
1da177e4 | 4517 | else |
3df0fc5b | 4518 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4519 | #else |
4520 | if (state == TASK_RUNNING) | |
3df0fc5b | 4521 | printk(KERN_CONT " running task "); |
1da177e4 | 4522 | else |
3df0fc5b | 4523 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4524 | #endif |
4525 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4526 | free = stack_not_used(p); |
1da177e4 | 4527 | #endif |
3df0fc5b | 4528 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
07cde260 | 4529 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), |
aa47b7e0 | 4530 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4531 | |
5fb5e6de | 4532 | show_stack(p, NULL); |
1da177e4 LT |
4533 | } |
4534 | ||
e59e2ae2 | 4535 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4536 | { |
36c8b586 | 4537 | struct task_struct *g, *p; |
1da177e4 | 4538 | |
4bd77321 | 4539 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4540 | printk(KERN_INFO |
4541 | " task PC stack pid father\n"); | |
1da177e4 | 4542 | #else |
3df0fc5b PZ |
4543 | printk(KERN_INFO |
4544 | " task PC stack pid father\n"); | |
1da177e4 | 4545 | #endif |
510f5acc | 4546 | rcu_read_lock(); |
1da177e4 LT |
4547 | do_each_thread(g, p) { |
4548 | /* | |
4549 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4550 | * console might take a lot of time: |
1da177e4 LT |
4551 | */ |
4552 | touch_nmi_watchdog(); | |
39bc89fd | 4553 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4554 | sched_show_task(p); |
1da177e4 LT |
4555 | } while_each_thread(g, p); |
4556 | ||
04c9167f JF |
4557 | touch_all_softlockup_watchdogs(); |
4558 | ||
dd41f596 IM |
4559 | #ifdef CONFIG_SCHED_DEBUG |
4560 | sysrq_sched_debug_show(); | |
4561 | #endif | |
510f5acc | 4562 | rcu_read_unlock(); |
e59e2ae2 IM |
4563 | /* |
4564 | * Only show locks if all tasks are dumped: | |
4565 | */ | |
93335a21 | 4566 | if (!state_filter) |
e59e2ae2 | 4567 | debug_show_all_locks(); |
1da177e4 LT |
4568 | } |
4569 | ||
1df21055 IM |
4570 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4571 | { | |
dd41f596 | 4572 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4573 | } |
4574 | ||
f340c0d1 IM |
4575 | /** |
4576 | * init_idle - set up an idle thread for a given CPU | |
4577 | * @idle: task in question | |
4578 | * @cpu: cpu the idle task belongs to | |
4579 | * | |
4580 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4581 | * flag, to make booting more robust. | |
4582 | */ | |
5c1e1767 | 4583 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4584 | { |
70b97a7f | 4585 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4586 | unsigned long flags; |
4587 | ||
05fa785c | 4588 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4589 | |
dd41f596 | 4590 | __sched_fork(idle); |
06b83b5f | 4591 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4592 | idle->se.exec_start = sched_clock(); |
4593 | ||
1e1b6c51 | 4594 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4595 | /* |
4596 | * We're having a chicken and egg problem, even though we are | |
4597 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4598 | * lockdep check in task_group() will fail. | |
4599 | * | |
4600 | * Similar case to sched_fork(). / Alternatively we could | |
4601 | * use task_rq_lock() here and obtain the other rq->lock. | |
4602 | * | |
4603 | * Silence PROVE_RCU | |
4604 | */ | |
4605 | rcu_read_lock(); | |
dd41f596 | 4606 | __set_task_cpu(idle, cpu); |
6506cf6c | 4607 | rcu_read_unlock(); |
1da177e4 | 4608 | |
1da177e4 | 4609 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4610 | #if defined(CONFIG_SMP) |
4611 | idle->on_cpu = 1; | |
4866cde0 | 4612 | #endif |
05fa785c | 4613 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4614 | |
4615 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4616 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4617 | |
dd41f596 IM |
4618 | /* |
4619 | * The idle tasks have their own, simple scheduling class: | |
4620 | */ | |
4621 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4622 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
4623 | #if defined(CONFIG_SMP) |
4624 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4625 | #endif | |
19978ca6 IM |
4626 | } |
4627 | ||
1da177e4 | 4628 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4629 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4630 | { | |
4631 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4632 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4633 | |
4634 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4635 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4636 | } |
4637 | ||
1da177e4 LT |
4638 | /* |
4639 | * This is how migration works: | |
4640 | * | |
969c7921 TH |
4641 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4642 | * stop_one_cpu(). | |
4643 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4644 | * off the CPU) | |
4645 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4646 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4647 | * it and puts it into the right queue. |
969c7921 TH |
4648 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4649 | * is done. | |
1da177e4 LT |
4650 | */ |
4651 | ||
4652 | /* | |
4653 | * Change a given task's CPU affinity. Migrate the thread to a | |
4654 | * proper CPU and schedule it away if the CPU it's executing on | |
4655 | * is removed from the allowed bitmask. | |
4656 | * | |
4657 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4658 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4659 | * call is not atomic; no spinlocks may be held. |
4660 | */ | |
96f874e2 | 4661 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4662 | { |
4663 | unsigned long flags; | |
70b97a7f | 4664 | struct rq *rq; |
969c7921 | 4665 | unsigned int dest_cpu; |
48f24c4d | 4666 | int ret = 0; |
1da177e4 LT |
4667 | |
4668 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4669 | |
db44fc01 YZ |
4670 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4671 | goto out; | |
4672 | ||
6ad4c188 | 4673 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4674 | ret = -EINVAL; |
4675 | goto out; | |
4676 | } | |
4677 | ||
db44fc01 | 4678 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4679 | ret = -EINVAL; |
4680 | goto out; | |
4681 | } | |
4682 | ||
1e1b6c51 | 4683 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4684 | |
1da177e4 | 4685 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4686 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4687 | goto out; |
4688 | ||
969c7921 | 4689 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4690 | if (p->on_rq) { |
969c7921 | 4691 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4692 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4693 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4694 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4695 | tlb_migrate_finish(p->mm); |
4696 | return 0; | |
4697 | } | |
4698 | out: | |
0122ec5b | 4699 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4700 | |
1da177e4 LT |
4701 | return ret; |
4702 | } | |
cd8ba7cd | 4703 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4704 | |
4705 | /* | |
41a2d6cf | 4706 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4707 | * this because either it can't run here any more (set_cpus_allowed() |
4708 | * away from this CPU, or CPU going down), or because we're | |
4709 | * attempting to rebalance this task on exec (sched_exec). | |
4710 | * | |
4711 | * So we race with normal scheduler movements, but that's OK, as long | |
4712 | * as the task is no longer on this CPU. | |
efc30814 KK |
4713 | * |
4714 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4715 | */ |
efc30814 | 4716 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4717 | { |
70b97a7f | 4718 | struct rq *rq_dest, *rq_src; |
e2912009 | 4719 | int ret = 0; |
1da177e4 | 4720 | |
e761b772 | 4721 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4722 | return ret; |
1da177e4 LT |
4723 | |
4724 | rq_src = cpu_rq(src_cpu); | |
4725 | rq_dest = cpu_rq(dest_cpu); | |
4726 | ||
0122ec5b | 4727 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4728 | double_rq_lock(rq_src, rq_dest); |
4729 | /* Already moved. */ | |
4730 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4731 | goto done; |
1da177e4 | 4732 | /* Affinity changed (again). */ |
fa17b507 | 4733 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4734 | goto fail; |
1da177e4 | 4735 | |
e2912009 PZ |
4736 | /* |
4737 | * If we're not on a rq, the next wake-up will ensure we're | |
4738 | * placed properly. | |
4739 | */ | |
fd2f4419 | 4740 | if (p->on_rq) { |
4ca9b72b | 4741 | dequeue_task(rq_src, p, 0); |
e2912009 | 4742 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4743 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4744 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4745 | } |
b1e38734 | 4746 | done: |
efc30814 | 4747 | ret = 1; |
b1e38734 | 4748 | fail: |
1da177e4 | 4749 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4750 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4751 | return ret; |
1da177e4 LT |
4752 | } |
4753 | ||
4754 | /* | |
969c7921 TH |
4755 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4756 | * and performs thread migration by bumping thread off CPU then | |
4757 | * 'pushing' onto another runqueue. | |
1da177e4 | 4758 | */ |
969c7921 | 4759 | static int migration_cpu_stop(void *data) |
1da177e4 | 4760 | { |
969c7921 | 4761 | struct migration_arg *arg = data; |
f7b4cddc | 4762 | |
969c7921 TH |
4763 | /* |
4764 | * The original target cpu might have gone down and we might | |
4765 | * be on another cpu but it doesn't matter. | |
4766 | */ | |
f7b4cddc | 4767 | local_irq_disable(); |
969c7921 | 4768 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4769 | local_irq_enable(); |
1da177e4 | 4770 | return 0; |
f7b4cddc ON |
4771 | } |
4772 | ||
1da177e4 | 4773 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4774 | |
054b9108 | 4775 | /* |
48c5ccae PZ |
4776 | * Ensures that the idle task is using init_mm right before its cpu goes |
4777 | * offline. | |
054b9108 | 4778 | */ |
48c5ccae | 4779 | void idle_task_exit(void) |
1da177e4 | 4780 | { |
48c5ccae | 4781 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4782 | |
48c5ccae | 4783 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4784 | |
48c5ccae PZ |
4785 | if (mm != &init_mm) |
4786 | switch_mm(mm, &init_mm, current); | |
4787 | mmdrop(mm); | |
1da177e4 LT |
4788 | } |
4789 | ||
4790 | /* | |
5d180232 PZ |
4791 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4792 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4793 | * nr_active count is stable. | |
4794 | * | |
4795 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4796 | */ |
5d180232 | 4797 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4798 | { |
5d180232 PZ |
4799 | long delta = calc_load_fold_active(rq); |
4800 | if (delta) | |
4801 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4802 | } |
4803 | ||
48f24c4d | 4804 | /* |
48c5ccae PZ |
4805 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4806 | * try_to_wake_up()->select_task_rq(). | |
4807 | * | |
4808 | * Called with rq->lock held even though we'er in stop_machine() and | |
4809 | * there's no concurrency possible, we hold the required locks anyway | |
4810 | * because of lock validation efforts. | |
1da177e4 | 4811 | */ |
48c5ccae | 4812 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4813 | { |
70b97a7f | 4814 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4815 | struct task_struct *next, *stop = rq->stop; |
4816 | int dest_cpu; | |
1da177e4 LT |
4817 | |
4818 | /* | |
48c5ccae PZ |
4819 | * Fudge the rq selection such that the below task selection loop |
4820 | * doesn't get stuck on the currently eligible stop task. | |
4821 | * | |
4822 | * We're currently inside stop_machine() and the rq is either stuck | |
4823 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4824 | * either way we should never end up calling schedule() until we're | |
4825 | * done here. | |
1da177e4 | 4826 | */ |
48c5ccae | 4827 | rq->stop = NULL; |
48f24c4d | 4828 | |
dd41f596 | 4829 | for ( ; ; ) { |
48c5ccae PZ |
4830 | /* |
4831 | * There's this thread running, bail when that's the only | |
4832 | * remaining thread. | |
4833 | */ | |
4834 | if (rq->nr_running == 1) | |
dd41f596 | 4835 | break; |
48c5ccae | 4836 | |
b67802ea | 4837 | next = pick_next_task(rq); |
48c5ccae | 4838 | BUG_ON(!next); |
79c53799 | 4839 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4840 | |
48c5ccae PZ |
4841 | /* Find suitable destination for @next, with force if needed. */ |
4842 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4843 | raw_spin_unlock(&rq->lock); | |
4844 | ||
4845 | __migrate_task(next, dead_cpu, dest_cpu); | |
4846 | ||
4847 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4848 | } |
dce48a84 | 4849 | |
48c5ccae | 4850 | rq->stop = stop; |
dce48a84 | 4851 | } |
48c5ccae | 4852 | |
1da177e4 LT |
4853 | #endif /* CONFIG_HOTPLUG_CPU */ |
4854 | ||
e692ab53 NP |
4855 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4856 | ||
4857 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4858 | { |
4859 | .procname = "sched_domain", | |
c57baf1e | 4860 | .mode = 0555, |
e0361851 | 4861 | }, |
56992309 | 4862 | {} |
e692ab53 NP |
4863 | }; |
4864 | ||
4865 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4866 | { |
4867 | .procname = "kernel", | |
c57baf1e | 4868 | .mode = 0555, |
e0361851 AD |
4869 | .child = sd_ctl_dir, |
4870 | }, | |
56992309 | 4871 | {} |
e692ab53 NP |
4872 | }; |
4873 | ||
4874 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4875 | { | |
4876 | struct ctl_table *entry = | |
5cf9f062 | 4877 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4878 | |
e692ab53 NP |
4879 | return entry; |
4880 | } | |
4881 | ||
6382bc90 MM |
4882 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4883 | { | |
cd790076 | 4884 | struct ctl_table *entry; |
6382bc90 | 4885 | |
cd790076 MM |
4886 | /* |
4887 | * In the intermediate directories, both the child directory and | |
4888 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4889 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4890 | * static strings and all have proc handlers. |
4891 | */ | |
4892 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4893 | if (entry->child) |
4894 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4895 | if (entry->proc_handler == NULL) |
4896 | kfree(entry->procname); | |
4897 | } | |
6382bc90 MM |
4898 | |
4899 | kfree(*tablep); | |
4900 | *tablep = NULL; | |
4901 | } | |
4902 | ||
201c373e NK |
4903 | static int min_load_idx = 0; |
4904 | static int max_load_idx = CPU_LOAD_IDX_MAX; | |
4905 | ||
e692ab53 | 4906 | static void |
e0361851 | 4907 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4908 | const char *procname, void *data, int maxlen, |
201c373e NK |
4909 | umode_t mode, proc_handler *proc_handler, |
4910 | bool load_idx) | |
e692ab53 | 4911 | { |
e692ab53 NP |
4912 | entry->procname = procname; |
4913 | entry->data = data; | |
4914 | entry->maxlen = maxlen; | |
4915 | entry->mode = mode; | |
4916 | entry->proc_handler = proc_handler; | |
201c373e NK |
4917 | |
4918 | if (load_idx) { | |
4919 | entry->extra1 = &min_load_idx; | |
4920 | entry->extra2 = &max_load_idx; | |
4921 | } | |
e692ab53 NP |
4922 | } |
4923 | ||
4924 | static struct ctl_table * | |
4925 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4926 | { | |
a5d8c348 | 4927 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 4928 | |
ad1cdc1d MM |
4929 | if (table == NULL) |
4930 | return NULL; | |
4931 | ||
e0361851 | 4932 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4933 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4934 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4935 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4936 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4937 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4938 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4939 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4940 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4941 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4942 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4943 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4944 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4945 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4946 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4947 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4948 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4949 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4950 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4951 | &sd->cache_nice_tries, |
201c373e | 4952 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4953 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4954 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 4955 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 4956 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 4957 | /* &table[12] is terminator */ |
e692ab53 NP |
4958 | |
4959 | return table; | |
4960 | } | |
4961 | ||
9a4e7159 | 4962 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4963 | { |
4964 | struct ctl_table *entry, *table; | |
4965 | struct sched_domain *sd; | |
4966 | int domain_num = 0, i; | |
4967 | char buf[32]; | |
4968 | ||
4969 | for_each_domain(cpu, sd) | |
4970 | domain_num++; | |
4971 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4972 | if (table == NULL) |
4973 | return NULL; | |
e692ab53 NP |
4974 | |
4975 | i = 0; | |
4976 | for_each_domain(cpu, sd) { | |
4977 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4978 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4979 | entry->mode = 0555; |
e692ab53 NP |
4980 | entry->child = sd_alloc_ctl_domain_table(sd); |
4981 | entry++; | |
4982 | i++; | |
4983 | } | |
4984 | return table; | |
4985 | } | |
4986 | ||
4987 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4988 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4989 | { |
6ad4c188 | 4990 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4991 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4992 | char buf[32]; | |
4993 | ||
7378547f MM |
4994 | WARN_ON(sd_ctl_dir[0].child); |
4995 | sd_ctl_dir[0].child = entry; | |
4996 | ||
ad1cdc1d MM |
4997 | if (entry == NULL) |
4998 | return; | |
4999 | ||
6ad4c188 | 5000 | for_each_possible_cpu(i) { |
e692ab53 | 5001 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5002 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5003 | entry->mode = 0555; |
e692ab53 | 5004 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5005 | entry++; |
e692ab53 | 5006 | } |
7378547f MM |
5007 | |
5008 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5009 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5010 | } | |
6382bc90 | 5011 | |
7378547f | 5012 | /* may be called multiple times per register */ |
6382bc90 MM |
5013 | static void unregister_sched_domain_sysctl(void) |
5014 | { | |
7378547f MM |
5015 | if (sd_sysctl_header) |
5016 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5017 | sd_sysctl_header = NULL; |
7378547f MM |
5018 | if (sd_ctl_dir[0].child) |
5019 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5020 | } |
e692ab53 | 5021 | #else |
6382bc90 MM |
5022 | static void register_sched_domain_sysctl(void) |
5023 | { | |
5024 | } | |
5025 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5026 | { |
5027 | } | |
5028 | #endif | |
5029 | ||
1f11eb6a GH |
5030 | static void set_rq_online(struct rq *rq) |
5031 | { | |
5032 | if (!rq->online) { | |
5033 | const struct sched_class *class; | |
5034 | ||
c6c4927b | 5035 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5036 | rq->online = 1; |
5037 | ||
5038 | for_each_class(class) { | |
5039 | if (class->rq_online) | |
5040 | class->rq_online(rq); | |
5041 | } | |
5042 | } | |
5043 | } | |
5044 | ||
5045 | static void set_rq_offline(struct rq *rq) | |
5046 | { | |
5047 | if (rq->online) { | |
5048 | const struct sched_class *class; | |
5049 | ||
5050 | for_each_class(class) { | |
5051 | if (class->rq_offline) | |
5052 | class->rq_offline(rq); | |
5053 | } | |
5054 | ||
c6c4927b | 5055 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5056 | rq->online = 0; |
5057 | } | |
5058 | } | |
5059 | ||
1da177e4 LT |
5060 | /* |
5061 | * migration_call - callback that gets triggered when a CPU is added. | |
5062 | * Here we can start up the necessary migration thread for the new CPU. | |
5063 | */ | |
48f24c4d IM |
5064 | static int __cpuinit |
5065 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5066 | { |
48f24c4d | 5067 | int cpu = (long)hcpu; |
1da177e4 | 5068 | unsigned long flags; |
969c7921 | 5069 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5070 | |
48c5ccae | 5071 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5072 | |
1da177e4 | 5073 | case CPU_UP_PREPARE: |
a468d389 | 5074 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5075 | break; |
48f24c4d | 5076 | |
1da177e4 | 5077 | case CPU_ONLINE: |
1f94ef59 | 5078 | /* Update our root-domain */ |
05fa785c | 5079 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5080 | if (rq->rd) { |
c6c4927b | 5081 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5082 | |
5083 | set_rq_online(rq); | |
1f94ef59 | 5084 | } |
05fa785c | 5085 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5086 | break; |
48f24c4d | 5087 | |
1da177e4 | 5088 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5089 | case CPU_DYING: |
317f3941 | 5090 | sched_ttwu_pending(); |
57d885fe | 5091 | /* Update our root-domain */ |
05fa785c | 5092 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5093 | if (rq->rd) { |
c6c4927b | 5094 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5095 | set_rq_offline(rq); |
57d885fe | 5096 | } |
48c5ccae PZ |
5097 | migrate_tasks(cpu); |
5098 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5099 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5100 | break; |
48c5ccae | 5101 | |
5d180232 | 5102 | case CPU_DEAD: |
f319da0c | 5103 | calc_load_migrate(rq); |
57d885fe | 5104 | break; |
1da177e4 LT |
5105 | #endif |
5106 | } | |
49c022e6 PZ |
5107 | |
5108 | update_max_interval(); | |
5109 | ||
1da177e4 LT |
5110 | return NOTIFY_OK; |
5111 | } | |
5112 | ||
f38b0820 PM |
5113 | /* |
5114 | * Register at high priority so that task migration (migrate_all_tasks) | |
5115 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5116 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5117 | */ |
26c2143b | 5118 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5119 | .notifier_call = migration_call, |
50a323b7 | 5120 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5121 | }; |
5122 | ||
3a101d05 TH |
5123 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5124 | unsigned long action, void *hcpu) | |
5125 | { | |
5126 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5127 | case CPU_STARTING: |
3a101d05 TH |
5128 | case CPU_DOWN_FAILED: |
5129 | set_cpu_active((long)hcpu, true); | |
5130 | return NOTIFY_OK; | |
5131 | default: | |
5132 | return NOTIFY_DONE; | |
5133 | } | |
5134 | } | |
5135 | ||
5136 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5137 | unsigned long action, void *hcpu) | |
5138 | { | |
5139 | switch (action & ~CPU_TASKS_FROZEN) { | |
5140 | case CPU_DOWN_PREPARE: | |
5141 | set_cpu_active((long)hcpu, false); | |
5142 | return NOTIFY_OK; | |
5143 | default: | |
5144 | return NOTIFY_DONE; | |
5145 | } | |
5146 | } | |
5147 | ||
7babe8db | 5148 | static int __init migration_init(void) |
1da177e4 LT |
5149 | { |
5150 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5151 | int err; |
48f24c4d | 5152 | |
3a101d05 | 5153 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5154 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5155 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5156 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5157 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5158 | |
3a101d05 TH |
5159 | /* Register cpu active notifiers */ |
5160 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5161 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5162 | ||
a004cd42 | 5163 | return 0; |
1da177e4 | 5164 | } |
7babe8db | 5165 | early_initcall(migration_init); |
1da177e4 LT |
5166 | #endif |
5167 | ||
5168 | #ifdef CONFIG_SMP | |
476f3534 | 5169 | |
4cb98839 PZ |
5170 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5171 | ||
3e9830dc | 5172 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5173 | |
d039ac60 | 5174 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5175 | |
d039ac60 | 5176 | static int __init sched_debug_setup(char *str) |
f6630114 | 5177 | { |
d039ac60 | 5178 | sched_debug_enabled = 1; |
f6630114 MT |
5179 | |
5180 | return 0; | |
5181 | } | |
d039ac60 PZ |
5182 | early_param("sched_debug", sched_debug_setup); |
5183 | ||
5184 | static inline bool sched_debug(void) | |
5185 | { | |
5186 | return sched_debug_enabled; | |
5187 | } | |
f6630114 | 5188 | |
7c16ec58 | 5189 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5190 | struct cpumask *groupmask) |
1da177e4 | 5191 | { |
4dcf6aff | 5192 | struct sched_group *group = sd->groups; |
434d53b0 | 5193 | char str[256]; |
1da177e4 | 5194 | |
968ea6d8 | 5195 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5196 | cpumask_clear(groupmask); |
4dcf6aff IM |
5197 | |
5198 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5199 | ||
5200 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5201 | printk("does not load-balance\n"); |
4dcf6aff | 5202 | if (sd->parent) |
3df0fc5b PZ |
5203 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5204 | " has parent"); | |
4dcf6aff | 5205 | return -1; |
41c7ce9a NP |
5206 | } |
5207 | ||
3df0fc5b | 5208 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5209 | |
758b2cdc | 5210 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5211 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5212 | "CPU%d\n", cpu); | |
4dcf6aff | 5213 | } |
758b2cdc | 5214 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5215 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5216 | " CPU%d\n", cpu); | |
4dcf6aff | 5217 | } |
1da177e4 | 5218 | |
4dcf6aff | 5219 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5220 | do { |
4dcf6aff | 5221 | if (!group) { |
3df0fc5b PZ |
5222 | printk("\n"); |
5223 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5224 | break; |
5225 | } | |
5226 | ||
c3decf0d PZ |
5227 | /* |
5228 | * Even though we initialize ->power to something semi-sane, | |
5229 | * we leave power_orig unset. This allows us to detect if | |
5230 | * domain iteration is still funny without causing /0 traps. | |
5231 | */ | |
5232 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5233 | printk(KERN_CONT "\n"); |
5234 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5235 | "set\n"); | |
4dcf6aff IM |
5236 | break; |
5237 | } | |
1da177e4 | 5238 | |
758b2cdc | 5239 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5240 | printk(KERN_CONT "\n"); |
5241 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5242 | break; |
5243 | } | |
1da177e4 | 5244 | |
cb83b629 PZ |
5245 | if (!(sd->flags & SD_OVERLAP) && |
5246 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5247 | printk(KERN_CONT "\n"); |
5248 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5249 | break; |
5250 | } | |
1da177e4 | 5251 | |
758b2cdc | 5252 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5253 | |
968ea6d8 | 5254 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5255 | |
3df0fc5b | 5256 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5257 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5258 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5259 | group->sgp->power); |
381512cf | 5260 | } |
1da177e4 | 5261 | |
4dcf6aff IM |
5262 | group = group->next; |
5263 | } while (group != sd->groups); | |
3df0fc5b | 5264 | printk(KERN_CONT "\n"); |
1da177e4 | 5265 | |
758b2cdc | 5266 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5267 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5268 | |
758b2cdc RR |
5269 | if (sd->parent && |
5270 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5271 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5272 | "of domain->span\n"); | |
4dcf6aff IM |
5273 | return 0; |
5274 | } | |
1da177e4 | 5275 | |
4dcf6aff IM |
5276 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5277 | { | |
5278 | int level = 0; | |
1da177e4 | 5279 | |
d039ac60 | 5280 | if (!sched_debug_enabled) |
f6630114 MT |
5281 | return; |
5282 | ||
4dcf6aff IM |
5283 | if (!sd) { |
5284 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5285 | return; | |
5286 | } | |
1da177e4 | 5287 | |
4dcf6aff IM |
5288 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5289 | ||
5290 | for (;;) { | |
4cb98839 | 5291 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5292 | break; |
1da177e4 LT |
5293 | level++; |
5294 | sd = sd->parent; | |
33859f7f | 5295 | if (!sd) |
4dcf6aff IM |
5296 | break; |
5297 | } | |
1da177e4 | 5298 | } |
6d6bc0ad | 5299 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5300 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5301 | static inline bool sched_debug(void) |
5302 | { | |
5303 | return false; | |
5304 | } | |
6d6bc0ad | 5305 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5306 | |
1a20ff27 | 5307 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5308 | { |
758b2cdc | 5309 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5310 | return 1; |
5311 | ||
5312 | /* Following flags need at least 2 groups */ | |
5313 | if (sd->flags & (SD_LOAD_BALANCE | | |
5314 | SD_BALANCE_NEWIDLE | | |
5315 | SD_BALANCE_FORK | | |
89c4710e SS |
5316 | SD_BALANCE_EXEC | |
5317 | SD_SHARE_CPUPOWER | | |
5318 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5319 | if (sd->groups != sd->groups->next) |
5320 | return 0; | |
5321 | } | |
5322 | ||
5323 | /* Following flags don't use groups */ | |
c88d5910 | 5324 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5325 | return 0; |
5326 | ||
5327 | return 1; | |
5328 | } | |
5329 | ||
48f24c4d IM |
5330 | static int |
5331 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5332 | { |
5333 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5334 | ||
5335 | if (sd_degenerate(parent)) | |
5336 | return 1; | |
5337 | ||
758b2cdc | 5338 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5339 | return 0; |
5340 | ||
245af2c7 SS |
5341 | /* Flags needing groups don't count if only 1 group in parent */ |
5342 | if (parent->groups == parent->groups->next) { | |
5343 | pflags &= ~(SD_LOAD_BALANCE | | |
5344 | SD_BALANCE_NEWIDLE | | |
5345 | SD_BALANCE_FORK | | |
89c4710e SS |
5346 | SD_BALANCE_EXEC | |
5347 | SD_SHARE_CPUPOWER | | |
5348 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5349 | if (nr_node_ids == 1) |
5350 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5351 | } |
5352 | if (~cflags & pflags) | |
5353 | return 0; | |
5354 | ||
5355 | return 1; | |
5356 | } | |
5357 | ||
dce840a0 | 5358 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5359 | { |
dce840a0 | 5360 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5361 | |
68e74568 | 5362 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5363 | free_cpumask_var(rd->rto_mask); |
5364 | free_cpumask_var(rd->online); | |
5365 | free_cpumask_var(rd->span); | |
5366 | kfree(rd); | |
5367 | } | |
5368 | ||
57d885fe GH |
5369 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5370 | { | |
a0490fa3 | 5371 | struct root_domain *old_rd = NULL; |
57d885fe | 5372 | unsigned long flags; |
57d885fe | 5373 | |
05fa785c | 5374 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5375 | |
5376 | if (rq->rd) { | |
a0490fa3 | 5377 | old_rd = rq->rd; |
57d885fe | 5378 | |
c6c4927b | 5379 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5380 | set_rq_offline(rq); |
57d885fe | 5381 | |
c6c4927b | 5382 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5383 | |
a0490fa3 IM |
5384 | /* |
5385 | * If we dont want to free the old_rt yet then | |
5386 | * set old_rd to NULL to skip the freeing later | |
5387 | * in this function: | |
5388 | */ | |
5389 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5390 | old_rd = NULL; | |
57d885fe GH |
5391 | } |
5392 | ||
5393 | atomic_inc(&rd->refcount); | |
5394 | rq->rd = rd; | |
5395 | ||
c6c4927b | 5396 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5397 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5398 | set_rq_online(rq); |
57d885fe | 5399 | |
05fa785c | 5400 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5401 | |
5402 | if (old_rd) | |
dce840a0 | 5403 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5404 | } |
5405 | ||
68c38fc3 | 5406 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5407 | { |
5408 | memset(rd, 0, sizeof(*rd)); | |
5409 | ||
68c38fc3 | 5410 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5411 | goto out; |
68c38fc3 | 5412 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5413 | goto free_span; |
68c38fc3 | 5414 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5415 | goto free_online; |
6e0534f2 | 5416 | |
68c38fc3 | 5417 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5418 | goto free_rto_mask; |
c6c4927b | 5419 | return 0; |
6e0534f2 | 5420 | |
68e74568 RR |
5421 | free_rto_mask: |
5422 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5423 | free_online: |
5424 | free_cpumask_var(rd->online); | |
5425 | free_span: | |
5426 | free_cpumask_var(rd->span); | |
0c910d28 | 5427 | out: |
c6c4927b | 5428 | return -ENOMEM; |
57d885fe GH |
5429 | } |
5430 | ||
029632fb PZ |
5431 | /* |
5432 | * By default the system creates a single root-domain with all cpus as | |
5433 | * members (mimicking the global state we have today). | |
5434 | */ | |
5435 | struct root_domain def_root_domain; | |
5436 | ||
57d885fe GH |
5437 | static void init_defrootdomain(void) |
5438 | { | |
68c38fc3 | 5439 | init_rootdomain(&def_root_domain); |
c6c4927b | 5440 | |
57d885fe GH |
5441 | atomic_set(&def_root_domain.refcount, 1); |
5442 | } | |
5443 | ||
dc938520 | 5444 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5445 | { |
5446 | struct root_domain *rd; | |
5447 | ||
5448 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5449 | if (!rd) | |
5450 | return NULL; | |
5451 | ||
68c38fc3 | 5452 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5453 | kfree(rd); |
5454 | return NULL; | |
5455 | } | |
57d885fe GH |
5456 | |
5457 | return rd; | |
5458 | } | |
5459 | ||
e3589f6c PZ |
5460 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5461 | { | |
5462 | struct sched_group *tmp, *first; | |
5463 | ||
5464 | if (!sg) | |
5465 | return; | |
5466 | ||
5467 | first = sg; | |
5468 | do { | |
5469 | tmp = sg->next; | |
5470 | ||
5471 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5472 | kfree(sg->sgp); | |
5473 | ||
5474 | kfree(sg); | |
5475 | sg = tmp; | |
5476 | } while (sg != first); | |
5477 | } | |
5478 | ||
dce840a0 PZ |
5479 | static void free_sched_domain(struct rcu_head *rcu) |
5480 | { | |
5481 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5482 | |
5483 | /* | |
5484 | * If its an overlapping domain it has private groups, iterate and | |
5485 | * nuke them all. | |
5486 | */ | |
5487 | if (sd->flags & SD_OVERLAP) { | |
5488 | free_sched_groups(sd->groups, 1); | |
5489 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5490 | kfree(sd->groups->sgp); |
dce840a0 | 5491 | kfree(sd->groups); |
9c3f75cb | 5492 | } |
dce840a0 PZ |
5493 | kfree(sd); |
5494 | } | |
5495 | ||
5496 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5497 | { | |
5498 | call_rcu(&sd->rcu, free_sched_domain); | |
5499 | } | |
5500 | ||
5501 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5502 | { | |
5503 | for (; sd; sd = sd->parent) | |
5504 | destroy_sched_domain(sd, cpu); | |
5505 | } | |
5506 | ||
518cd623 PZ |
5507 | /* |
5508 | * Keep a special pointer to the highest sched_domain that has | |
5509 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5510 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5511 | * | |
5512 | * Also keep a unique ID per domain (we use the first cpu number in | |
5513 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5514 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5515 | */ |
5516 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5517 | DEFINE_PER_CPU(int, sd_llc_id); | |
5518 | ||
5519 | static void update_top_cache_domain(int cpu) | |
5520 | { | |
5521 | struct sched_domain *sd; | |
5522 | int id = cpu; | |
5523 | ||
5524 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
37407ea7 | 5525 | if (sd) |
518cd623 PZ |
5526 | id = cpumask_first(sched_domain_span(sd)); |
5527 | ||
5528 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5529 | per_cpu(sd_llc_id, cpu) = id; | |
5530 | } | |
5531 | ||
1da177e4 | 5532 | /* |
0eab9146 | 5533 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5534 | * hold the hotplug lock. |
5535 | */ | |
0eab9146 IM |
5536 | static void |
5537 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5538 | { |
70b97a7f | 5539 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5540 | struct sched_domain *tmp; |
5541 | ||
5542 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5543 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5544 | struct sched_domain *parent = tmp->parent; |
5545 | if (!parent) | |
5546 | break; | |
f29c9b1c | 5547 | |
1a848870 | 5548 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5549 | tmp->parent = parent->parent; |
1a848870 SS |
5550 | if (parent->parent) |
5551 | parent->parent->child = tmp; | |
dce840a0 | 5552 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5553 | } else |
5554 | tmp = tmp->parent; | |
245af2c7 SS |
5555 | } |
5556 | ||
1a848870 | 5557 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5558 | tmp = sd; |
245af2c7 | 5559 | sd = sd->parent; |
dce840a0 | 5560 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5561 | if (sd) |
5562 | sd->child = NULL; | |
5563 | } | |
1da177e4 | 5564 | |
4cb98839 | 5565 | sched_domain_debug(sd, cpu); |
1da177e4 | 5566 | |
57d885fe | 5567 | rq_attach_root(rq, rd); |
dce840a0 | 5568 | tmp = rq->sd; |
674311d5 | 5569 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5570 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5571 | |
5572 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5573 | } |
5574 | ||
5575 | /* cpus with isolated domains */ | |
dcc30a35 | 5576 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5577 | |
5578 | /* Setup the mask of cpus configured for isolated domains */ | |
5579 | static int __init isolated_cpu_setup(char *str) | |
5580 | { | |
bdddd296 | 5581 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5582 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5583 | return 1; |
5584 | } | |
5585 | ||
8927f494 | 5586 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5587 | |
d3081f52 PZ |
5588 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5589 | { | |
5590 | return cpumask_of_node(cpu_to_node(cpu)); | |
5591 | } | |
5592 | ||
dce840a0 PZ |
5593 | struct sd_data { |
5594 | struct sched_domain **__percpu sd; | |
5595 | struct sched_group **__percpu sg; | |
9c3f75cb | 5596 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5597 | }; |
5598 | ||
49a02c51 | 5599 | struct s_data { |
21d42ccf | 5600 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5601 | struct root_domain *rd; |
5602 | }; | |
5603 | ||
2109b99e | 5604 | enum s_alloc { |
2109b99e | 5605 | sa_rootdomain, |
21d42ccf | 5606 | sa_sd, |
dce840a0 | 5607 | sa_sd_storage, |
2109b99e AH |
5608 | sa_none, |
5609 | }; | |
5610 | ||
54ab4ff4 PZ |
5611 | struct sched_domain_topology_level; |
5612 | ||
5613 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5614 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5615 | ||
e3589f6c PZ |
5616 | #define SDTL_OVERLAP 0x01 |
5617 | ||
eb7a74e6 | 5618 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5619 | sched_domain_init_f init; |
5620 | sched_domain_mask_f mask; | |
e3589f6c | 5621 | int flags; |
cb83b629 | 5622 | int numa_level; |
54ab4ff4 | 5623 | struct sd_data data; |
eb7a74e6 PZ |
5624 | }; |
5625 | ||
c1174876 PZ |
5626 | /* |
5627 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5628 | * domain traversal. | |
5629 | * | |
5630 | * Asymmetric node setups can result in situations where the domain tree is of | |
5631 | * unequal depth, make sure to skip domains that already cover the entire | |
5632 | * range. | |
5633 | * | |
5634 | * In that case build_sched_domains() will have terminated the iteration early | |
5635 | * and our sibling sd spans will be empty. Domains should always include the | |
5636 | * cpu they're built on, so check that. | |
5637 | * | |
5638 | */ | |
5639 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5640 | { | |
5641 | const struct cpumask *span = sched_domain_span(sd); | |
5642 | struct sd_data *sdd = sd->private; | |
5643 | struct sched_domain *sibling; | |
5644 | int i; | |
5645 | ||
5646 | for_each_cpu(i, span) { | |
5647 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5648 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5649 | continue; | |
5650 | ||
5651 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5652 | } | |
5653 | } | |
5654 | ||
5655 | /* | |
5656 | * Return the canonical balance cpu for this group, this is the first cpu | |
5657 | * of this group that's also in the iteration mask. | |
5658 | */ | |
5659 | int group_balance_cpu(struct sched_group *sg) | |
5660 | { | |
5661 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5662 | } | |
5663 | ||
e3589f6c PZ |
5664 | static int |
5665 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5666 | { | |
5667 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5668 | const struct cpumask *span = sched_domain_span(sd); | |
5669 | struct cpumask *covered = sched_domains_tmpmask; | |
5670 | struct sd_data *sdd = sd->private; | |
5671 | struct sched_domain *child; | |
5672 | int i; | |
5673 | ||
5674 | cpumask_clear(covered); | |
5675 | ||
5676 | for_each_cpu(i, span) { | |
5677 | struct cpumask *sg_span; | |
5678 | ||
5679 | if (cpumask_test_cpu(i, covered)) | |
5680 | continue; | |
5681 | ||
c1174876 PZ |
5682 | child = *per_cpu_ptr(sdd->sd, i); |
5683 | ||
5684 | /* See the comment near build_group_mask(). */ | |
5685 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5686 | continue; | |
5687 | ||
e3589f6c | 5688 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5689 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5690 | |
5691 | if (!sg) | |
5692 | goto fail; | |
5693 | ||
5694 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5695 | if (child->child) { |
5696 | child = child->child; | |
5697 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5698 | } else | |
5699 | cpumask_set_cpu(i, sg_span); | |
5700 | ||
5701 | cpumask_or(covered, covered, sg_span); | |
5702 | ||
74a5ce20 | 5703 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5704 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5705 | build_group_mask(sd, sg); | |
5706 | ||
c3decf0d PZ |
5707 | /* |
5708 | * Initialize sgp->power such that even if we mess up the | |
5709 | * domains and no possible iteration will get us here, we won't | |
5710 | * die on a /0 trap. | |
5711 | */ | |
5712 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 5713 | |
c1174876 PZ |
5714 | /* |
5715 | * Make sure the first group of this domain contains the | |
5716 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5717 | * breaks. See update_sg_lb_stats(). | |
5718 | */ | |
74a5ce20 | 5719 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5720 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5721 | groups = sg; |
5722 | ||
5723 | if (!first) | |
5724 | first = sg; | |
5725 | if (last) | |
5726 | last->next = sg; | |
5727 | last = sg; | |
5728 | last->next = first; | |
5729 | } | |
5730 | sd->groups = groups; | |
5731 | ||
5732 | return 0; | |
5733 | ||
5734 | fail: | |
5735 | free_sched_groups(first, 0); | |
5736 | ||
5737 | return -ENOMEM; | |
5738 | } | |
5739 | ||
dce840a0 | 5740 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5741 | { |
dce840a0 PZ |
5742 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5743 | struct sched_domain *child = sd->child; | |
1da177e4 | 5744 | |
dce840a0 PZ |
5745 | if (child) |
5746 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5747 | |
9c3f75cb | 5748 | if (sg) { |
dce840a0 | 5749 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5750 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5751 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5752 | } |
dce840a0 PZ |
5753 | |
5754 | return cpu; | |
1e9f28fa | 5755 | } |
1e9f28fa | 5756 | |
01a08546 | 5757 | /* |
dce840a0 PZ |
5758 | * build_sched_groups will build a circular linked list of the groups |
5759 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5760 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5761 | * |
5762 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5763 | */ |
e3589f6c PZ |
5764 | static int |
5765 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5766 | { |
dce840a0 PZ |
5767 | struct sched_group *first = NULL, *last = NULL; |
5768 | struct sd_data *sdd = sd->private; | |
5769 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5770 | struct cpumask *covered; |
dce840a0 | 5771 | int i; |
9c1cfda2 | 5772 | |
e3589f6c PZ |
5773 | get_group(cpu, sdd, &sd->groups); |
5774 | atomic_inc(&sd->groups->ref); | |
5775 | ||
5776 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
5777 | return 0; | |
5778 | ||
f96225fd PZ |
5779 | lockdep_assert_held(&sched_domains_mutex); |
5780 | covered = sched_domains_tmpmask; | |
5781 | ||
dce840a0 | 5782 | cpumask_clear(covered); |
6711cab4 | 5783 | |
dce840a0 PZ |
5784 | for_each_cpu(i, span) { |
5785 | struct sched_group *sg; | |
5786 | int group = get_group(i, sdd, &sg); | |
5787 | int j; | |
6711cab4 | 5788 | |
dce840a0 PZ |
5789 | if (cpumask_test_cpu(i, covered)) |
5790 | continue; | |
6711cab4 | 5791 | |
dce840a0 | 5792 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5793 | sg->sgp->power = 0; |
c1174876 | 5794 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5795 | |
dce840a0 PZ |
5796 | for_each_cpu(j, span) { |
5797 | if (get_group(j, sdd, NULL) != group) | |
5798 | continue; | |
0601a88d | 5799 | |
dce840a0 PZ |
5800 | cpumask_set_cpu(j, covered); |
5801 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5802 | } | |
0601a88d | 5803 | |
dce840a0 PZ |
5804 | if (!first) |
5805 | first = sg; | |
5806 | if (last) | |
5807 | last->next = sg; | |
5808 | last = sg; | |
5809 | } | |
5810 | last->next = first; | |
e3589f6c PZ |
5811 | |
5812 | return 0; | |
0601a88d | 5813 | } |
51888ca2 | 5814 | |
89c4710e SS |
5815 | /* |
5816 | * Initialize sched groups cpu_power. | |
5817 | * | |
5818 | * cpu_power indicates the capacity of sched group, which is used while | |
5819 | * distributing the load between different sched groups in a sched domain. | |
5820 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5821 | * there are asymmetries in the topology. If there are asymmetries, group | |
5822 | * having more cpu_power will pickup more load compared to the group having | |
5823 | * less cpu_power. | |
89c4710e SS |
5824 | */ |
5825 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5826 | { | |
e3589f6c | 5827 | struct sched_group *sg = sd->groups; |
89c4710e | 5828 | |
e3589f6c PZ |
5829 | WARN_ON(!sd || !sg); |
5830 | ||
5831 | do { | |
5832 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5833 | sg = sg->next; | |
5834 | } while (sg != sd->groups); | |
89c4710e | 5835 | |
c1174876 | 5836 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5837 | return; |
aae6d3dd | 5838 | |
d274cb30 | 5839 | update_group_power(sd, cpu); |
69e1e811 | 5840 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5841 | } |
5842 | ||
029632fb PZ |
5843 | int __weak arch_sd_sibling_asym_packing(void) |
5844 | { | |
5845 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5846 | } |
5847 | ||
7c16ec58 MT |
5848 | /* |
5849 | * Initializers for schedule domains | |
5850 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5851 | */ | |
5852 | ||
a5d8c348 IM |
5853 | #ifdef CONFIG_SCHED_DEBUG |
5854 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5855 | #else | |
5856 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5857 | #endif | |
5858 | ||
54ab4ff4 PZ |
5859 | #define SD_INIT_FUNC(type) \ |
5860 | static noinline struct sched_domain * \ | |
5861 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5862 | { \ | |
5863 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5864 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5865 | SD_INIT_NAME(sd, type); \ |
5866 | sd->private = &tl->data; \ | |
5867 | return sd; \ | |
7c16ec58 MT |
5868 | } |
5869 | ||
5870 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5871 | #ifdef CONFIG_SCHED_SMT |
5872 | SD_INIT_FUNC(SIBLING) | |
5873 | #endif | |
5874 | #ifdef CONFIG_SCHED_MC | |
5875 | SD_INIT_FUNC(MC) | |
5876 | #endif | |
01a08546 HC |
5877 | #ifdef CONFIG_SCHED_BOOK |
5878 | SD_INIT_FUNC(BOOK) | |
5879 | #endif | |
7c16ec58 | 5880 | |
1d3504fc | 5881 | static int default_relax_domain_level = -1; |
60495e77 | 5882 | int sched_domain_level_max; |
1d3504fc HS |
5883 | |
5884 | static int __init setup_relax_domain_level(char *str) | |
5885 | { | |
a841f8ce DS |
5886 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5887 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5888 | |
1d3504fc HS |
5889 | return 1; |
5890 | } | |
5891 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5892 | ||
5893 | static void set_domain_attribute(struct sched_domain *sd, | |
5894 | struct sched_domain_attr *attr) | |
5895 | { | |
5896 | int request; | |
5897 | ||
5898 | if (!attr || attr->relax_domain_level < 0) { | |
5899 | if (default_relax_domain_level < 0) | |
5900 | return; | |
5901 | else | |
5902 | request = default_relax_domain_level; | |
5903 | } else | |
5904 | request = attr->relax_domain_level; | |
5905 | if (request < sd->level) { | |
5906 | /* turn off idle balance on this domain */ | |
c88d5910 | 5907 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5908 | } else { |
5909 | /* turn on idle balance on this domain */ | |
c88d5910 | 5910 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5911 | } |
5912 | } | |
5913 | ||
54ab4ff4 PZ |
5914 | static void __sdt_free(const struct cpumask *cpu_map); |
5915 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5916 | ||
2109b99e AH |
5917 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5918 | const struct cpumask *cpu_map) | |
5919 | { | |
5920 | switch (what) { | |
2109b99e | 5921 | case sa_rootdomain: |
822ff793 PZ |
5922 | if (!atomic_read(&d->rd->refcount)) |
5923 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5924 | case sa_sd: |
5925 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5926 | case sa_sd_storage: |
54ab4ff4 | 5927 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5928 | case sa_none: |
5929 | break; | |
5930 | } | |
5931 | } | |
3404c8d9 | 5932 | |
2109b99e AH |
5933 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5934 | const struct cpumask *cpu_map) | |
5935 | { | |
dce840a0 PZ |
5936 | memset(d, 0, sizeof(*d)); |
5937 | ||
54ab4ff4 PZ |
5938 | if (__sdt_alloc(cpu_map)) |
5939 | return sa_sd_storage; | |
dce840a0 PZ |
5940 | d->sd = alloc_percpu(struct sched_domain *); |
5941 | if (!d->sd) | |
5942 | return sa_sd_storage; | |
2109b99e | 5943 | d->rd = alloc_rootdomain(); |
dce840a0 | 5944 | if (!d->rd) |
21d42ccf | 5945 | return sa_sd; |
2109b99e AH |
5946 | return sa_rootdomain; |
5947 | } | |
57d885fe | 5948 | |
dce840a0 PZ |
5949 | /* |
5950 | * NULL the sd_data elements we've used to build the sched_domain and | |
5951 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5952 | * will not free the data we're using. | |
5953 | */ | |
5954 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5955 | { | |
5956 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5957 | |
5958 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5959 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5960 | ||
e3589f6c | 5961 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5962 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5963 | |
5964 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5965 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5966 | } |
5967 | ||
2c402dc3 PZ |
5968 | #ifdef CONFIG_SCHED_SMT |
5969 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 5970 | { |
2c402dc3 | 5971 | return topology_thread_cpumask(cpu); |
3bd65a80 | 5972 | } |
2c402dc3 | 5973 | #endif |
7f4588f3 | 5974 | |
d069b916 PZ |
5975 | /* |
5976 | * Topology list, bottom-up. | |
5977 | */ | |
2c402dc3 | 5978 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
5979 | #ifdef CONFIG_SCHED_SMT |
5980 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 5981 | #endif |
1e9f28fa | 5982 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 5983 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 5984 | #endif |
d069b916 PZ |
5985 | #ifdef CONFIG_SCHED_BOOK |
5986 | { sd_init_BOOK, cpu_book_mask, }, | |
5987 | #endif | |
5988 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
5989 | { NULL, }, |
5990 | }; | |
5991 | ||
5992 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
5993 | ||
cb83b629 PZ |
5994 | #ifdef CONFIG_NUMA |
5995 | ||
5996 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
5997 | static int *sched_domains_numa_distance; |
5998 | static struct cpumask ***sched_domains_numa_masks; | |
5999 | static int sched_domains_curr_level; | |
6000 | ||
cb83b629 PZ |
6001 | static inline int sd_local_flags(int level) |
6002 | { | |
10717dcd | 6003 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
6004 | return 0; |
6005 | ||
6006 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6007 | } | |
6008 | ||
6009 | static struct sched_domain * | |
6010 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6011 | { | |
6012 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6013 | int level = tl->numa_level; | |
6014 | int sd_weight = cpumask_weight( | |
6015 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6016 | ||
6017 | *sd = (struct sched_domain){ | |
6018 | .min_interval = sd_weight, | |
6019 | .max_interval = 2*sd_weight, | |
6020 | .busy_factor = 32, | |
870a0bb5 | 6021 | .imbalance_pct = 125, |
cb83b629 PZ |
6022 | .cache_nice_tries = 2, |
6023 | .busy_idx = 3, | |
6024 | .idle_idx = 2, | |
6025 | .newidle_idx = 0, | |
6026 | .wake_idx = 0, | |
6027 | .forkexec_idx = 0, | |
6028 | ||
6029 | .flags = 1*SD_LOAD_BALANCE | |
6030 | | 1*SD_BALANCE_NEWIDLE | |
6031 | | 0*SD_BALANCE_EXEC | |
6032 | | 0*SD_BALANCE_FORK | |
6033 | | 0*SD_BALANCE_WAKE | |
6034 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 6035 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
6036 | | 0*SD_SHARE_PKG_RESOURCES |
6037 | | 1*SD_SERIALIZE | |
6038 | | 0*SD_PREFER_SIBLING | |
6039 | | sd_local_flags(level) | |
6040 | , | |
6041 | .last_balance = jiffies, | |
6042 | .balance_interval = sd_weight, | |
6043 | }; | |
6044 | SD_INIT_NAME(sd, NUMA); | |
6045 | sd->private = &tl->data; | |
6046 | ||
6047 | /* | |
6048 | * Ugly hack to pass state to sd_numa_mask()... | |
6049 | */ | |
6050 | sched_domains_curr_level = tl->numa_level; | |
6051 | ||
6052 | return sd; | |
6053 | } | |
6054 | ||
6055 | static const struct cpumask *sd_numa_mask(int cpu) | |
6056 | { | |
6057 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6058 | } | |
6059 | ||
d039ac60 PZ |
6060 | static void sched_numa_warn(const char *str) |
6061 | { | |
6062 | static int done = false; | |
6063 | int i,j; | |
6064 | ||
6065 | if (done) | |
6066 | return; | |
6067 | ||
6068 | done = true; | |
6069 | ||
6070 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6071 | ||
6072 | for (i = 0; i < nr_node_ids; i++) { | |
6073 | printk(KERN_WARNING " "); | |
6074 | for (j = 0; j < nr_node_ids; j++) | |
6075 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6076 | printk(KERN_CONT "\n"); | |
6077 | } | |
6078 | printk(KERN_WARNING "\n"); | |
6079 | } | |
6080 | ||
6081 | static bool find_numa_distance(int distance) | |
6082 | { | |
6083 | int i; | |
6084 | ||
6085 | if (distance == node_distance(0, 0)) | |
6086 | return true; | |
6087 | ||
6088 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6089 | if (sched_domains_numa_distance[i] == distance) | |
6090 | return true; | |
6091 | } | |
6092 | ||
6093 | return false; | |
6094 | } | |
6095 | ||
cb83b629 PZ |
6096 | static void sched_init_numa(void) |
6097 | { | |
6098 | int next_distance, curr_distance = node_distance(0, 0); | |
6099 | struct sched_domain_topology_level *tl; | |
6100 | int level = 0; | |
6101 | int i, j, k; | |
6102 | ||
cb83b629 PZ |
6103 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6104 | if (!sched_domains_numa_distance) | |
6105 | return; | |
6106 | ||
6107 | /* | |
6108 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6109 | * unique distances in the node_distance() table. | |
6110 | * | |
6111 | * Assumes node_distance(0,j) includes all distances in | |
6112 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6113 | */ |
6114 | next_distance = curr_distance; | |
6115 | for (i = 0; i < nr_node_ids; i++) { | |
6116 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6117 | for (k = 0; k < nr_node_ids; k++) { |
6118 | int distance = node_distance(i, k); | |
6119 | ||
6120 | if (distance > curr_distance && | |
6121 | (distance < next_distance || | |
6122 | next_distance == curr_distance)) | |
6123 | next_distance = distance; | |
6124 | ||
6125 | /* | |
6126 | * While not a strong assumption it would be nice to know | |
6127 | * about cases where if node A is connected to B, B is not | |
6128 | * equally connected to A. | |
6129 | */ | |
6130 | if (sched_debug() && node_distance(k, i) != distance) | |
6131 | sched_numa_warn("Node-distance not symmetric"); | |
6132 | ||
6133 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6134 | sched_numa_warn("Node-0 not representative"); | |
6135 | } | |
6136 | if (next_distance != curr_distance) { | |
6137 | sched_domains_numa_distance[level++] = next_distance; | |
6138 | sched_domains_numa_levels = level; | |
6139 | curr_distance = next_distance; | |
6140 | } else break; | |
cb83b629 | 6141 | } |
d039ac60 PZ |
6142 | |
6143 | /* | |
6144 | * In case of sched_debug() we verify the above assumption. | |
6145 | */ | |
6146 | if (!sched_debug()) | |
6147 | break; | |
cb83b629 PZ |
6148 | } |
6149 | /* | |
6150 | * 'level' contains the number of unique distances, excluding the | |
6151 | * identity distance node_distance(i,i). | |
6152 | * | |
6153 | * The sched_domains_nume_distance[] array includes the actual distance | |
6154 | * numbers. | |
6155 | */ | |
6156 | ||
5f7865f3 TC |
6157 | /* |
6158 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6159 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6160 | * the array will contain less then 'level' members. This could be | |
6161 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6162 | * in other functions. | |
6163 | * | |
6164 | * We reset it to 'level' at the end of this function. | |
6165 | */ | |
6166 | sched_domains_numa_levels = 0; | |
6167 | ||
cb83b629 PZ |
6168 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6169 | if (!sched_domains_numa_masks) | |
6170 | return; | |
6171 | ||
6172 | /* | |
6173 | * Now for each level, construct a mask per node which contains all | |
6174 | * cpus of nodes that are that many hops away from us. | |
6175 | */ | |
6176 | for (i = 0; i < level; i++) { | |
6177 | sched_domains_numa_masks[i] = | |
6178 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6179 | if (!sched_domains_numa_masks[i]) | |
6180 | return; | |
6181 | ||
6182 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6183 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6184 | if (!mask) |
6185 | return; | |
6186 | ||
6187 | sched_domains_numa_masks[i][j] = mask; | |
6188 | ||
6189 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6190 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6191 | continue; |
6192 | ||
6193 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6194 | } | |
6195 | } | |
6196 | } | |
6197 | ||
6198 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6199 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6200 | if (!tl) | |
6201 | return; | |
6202 | ||
6203 | /* | |
6204 | * Copy the default topology bits.. | |
6205 | */ | |
6206 | for (i = 0; default_topology[i].init; i++) | |
6207 | tl[i] = default_topology[i]; | |
6208 | ||
6209 | /* | |
6210 | * .. and append 'j' levels of NUMA goodness. | |
6211 | */ | |
6212 | for (j = 0; j < level; i++, j++) { | |
6213 | tl[i] = (struct sched_domain_topology_level){ | |
6214 | .init = sd_numa_init, | |
6215 | .mask = sd_numa_mask, | |
6216 | .flags = SDTL_OVERLAP, | |
6217 | .numa_level = j, | |
6218 | }; | |
6219 | } | |
6220 | ||
6221 | sched_domain_topology = tl; | |
5f7865f3 TC |
6222 | |
6223 | sched_domains_numa_levels = level; | |
cb83b629 | 6224 | } |
301a5cba TC |
6225 | |
6226 | static void sched_domains_numa_masks_set(int cpu) | |
6227 | { | |
6228 | int i, j; | |
6229 | int node = cpu_to_node(cpu); | |
6230 | ||
6231 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6232 | for (j = 0; j < nr_node_ids; j++) { | |
6233 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6234 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6235 | } | |
6236 | } | |
6237 | } | |
6238 | ||
6239 | static void sched_domains_numa_masks_clear(int cpu) | |
6240 | { | |
6241 | int i, j; | |
6242 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6243 | for (j = 0; j < nr_node_ids; j++) | |
6244 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6245 | } | |
6246 | } | |
6247 | ||
6248 | /* | |
6249 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6250 | * are onlined. | |
6251 | */ | |
6252 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6253 | unsigned long action, | |
6254 | void *hcpu) | |
6255 | { | |
6256 | int cpu = (long)hcpu; | |
6257 | ||
6258 | switch (action & ~CPU_TASKS_FROZEN) { | |
6259 | case CPU_ONLINE: | |
6260 | sched_domains_numa_masks_set(cpu); | |
6261 | break; | |
6262 | ||
6263 | case CPU_DEAD: | |
6264 | sched_domains_numa_masks_clear(cpu); | |
6265 | break; | |
6266 | ||
6267 | default: | |
6268 | return NOTIFY_DONE; | |
6269 | } | |
6270 | ||
6271 | return NOTIFY_OK; | |
cb83b629 PZ |
6272 | } |
6273 | #else | |
6274 | static inline void sched_init_numa(void) | |
6275 | { | |
6276 | } | |
301a5cba TC |
6277 | |
6278 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6279 | unsigned long action, | |
6280 | void *hcpu) | |
6281 | { | |
6282 | return 0; | |
6283 | } | |
cb83b629 PZ |
6284 | #endif /* CONFIG_NUMA */ |
6285 | ||
54ab4ff4 PZ |
6286 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6287 | { | |
6288 | struct sched_domain_topology_level *tl; | |
6289 | int j; | |
6290 | ||
6291 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6292 | struct sd_data *sdd = &tl->data; | |
6293 | ||
6294 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6295 | if (!sdd->sd) | |
6296 | return -ENOMEM; | |
6297 | ||
6298 | sdd->sg = alloc_percpu(struct sched_group *); | |
6299 | if (!sdd->sg) | |
6300 | return -ENOMEM; | |
6301 | ||
9c3f75cb PZ |
6302 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6303 | if (!sdd->sgp) | |
6304 | return -ENOMEM; | |
6305 | ||
54ab4ff4 PZ |
6306 | for_each_cpu(j, cpu_map) { |
6307 | struct sched_domain *sd; | |
6308 | struct sched_group *sg; | |
9c3f75cb | 6309 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6310 | |
6311 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6312 | GFP_KERNEL, cpu_to_node(j)); | |
6313 | if (!sd) | |
6314 | return -ENOMEM; | |
6315 | ||
6316 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6317 | ||
6318 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6319 | GFP_KERNEL, cpu_to_node(j)); | |
6320 | if (!sg) | |
6321 | return -ENOMEM; | |
6322 | ||
30b4e9eb IM |
6323 | sg->next = sg; |
6324 | ||
54ab4ff4 | 6325 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6326 | |
c1174876 | 6327 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6328 | GFP_KERNEL, cpu_to_node(j)); |
6329 | if (!sgp) | |
6330 | return -ENOMEM; | |
6331 | ||
6332 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6333 | } |
6334 | } | |
6335 | ||
6336 | return 0; | |
6337 | } | |
6338 | ||
6339 | static void __sdt_free(const struct cpumask *cpu_map) | |
6340 | { | |
6341 | struct sched_domain_topology_level *tl; | |
6342 | int j; | |
6343 | ||
6344 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6345 | struct sd_data *sdd = &tl->data; | |
6346 | ||
6347 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6348 | struct sched_domain *sd; |
6349 | ||
6350 | if (sdd->sd) { | |
6351 | sd = *per_cpu_ptr(sdd->sd, j); | |
6352 | if (sd && (sd->flags & SD_OVERLAP)) | |
6353 | free_sched_groups(sd->groups, 0); | |
6354 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6355 | } | |
6356 | ||
6357 | if (sdd->sg) | |
6358 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6359 | if (sdd->sgp) | |
6360 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6361 | } |
6362 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6363 | sdd->sd = NULL; |
54ab4ff4 | 6364 | free_percpu(sdd->sg); |
fb2cf2c6 | 6365 | sdd->sg = NULL; |
9c3f75cb | 6366 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6367 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6368 | } |
6369 | } | |
6370 | ||
2c402dc3 PZ |
6371 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6372 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6373 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6374 | int cpu) |
6375 | { | |
54ab4ff4 | 6376 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6377 | if (!sd) |
d069b916 | 6378 | return child; |
2c402dc3 | 6379 | |
2c402dc3 | 6380 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6381 | if (child) { |
6382 | sd->level = child->level + 1; | |
6383 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6384 | child->parent = sd; |
60495e77 | 6385 | } |
d069b916 | 6386 | sd->child = child; |
a841f8ce | 6387 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6388 | |
6389 | return sd; | |
6390 | } | |
6391 | ||
2109b99e AH |
6392 | /* |
6393 | * Build sched domains for a given set of cpus and attach the sched domains | |
6394 | * to the individual cpus | |
6395 | */ | |
dce840a0 PZ |
6396 | static int build_sched_domains(const struct cpumask *cpu_map, |
6397 | struct sched_domain_attr *attr) | |
2109b99e AH |
6398 | { |
6399 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6400 | struct sched_domain *sd; |
2109b99e | 6401 | struct s_data d; |
822ff793 | 6402 | int i, ret = -ENOMEM; |
9c1cfda2 | 6403 | |
2109b99e AH |
6404 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6405 | if (alloc_state != sa_rootdomain) | |
6406 | goto error; | |
9c1cfda2 | 6407 | |
dce840a0 | 6408 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6409 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6410 | struct sched_domain_topology_level *tl; |
6411 | ||
3bd65a80 | 6412 | sd = NULL; |
e3589f6c | 6413 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6414 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6415 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6416 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6417 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6418 | break; | |
e3589f6c | 6419 | } |
d274cb30 | 6420 | |
d069b916 PZ |
6421 | while (sd->child) |
6422 | sd = sd->child; | |
6423 | ||
21d42ccf | 6424 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6425 | } |
6426 | ||
6427 | /* Build the groups for the domains */ | |
6428 | for_each_cpu(i, cpu_map) { | |
6429 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6430 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6431 | if (sd->flags & SD_OVERLAP) { |
6432 | if (build_overlap_sched_groups(sd, i)) | |
6433 | goto error; | |
6434 | } else { | |
6435 | if (build_sched_groups(sd, i)) | |
6436 | goto error; | |
6437 | } | |
1cf51902 | 6438 | } |
a06dadbe | 6439 | } |
9c1cfda2 | 6440 | |
1da177e4 | 6441 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6442 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6443 | if (!cpumask_test_cpu(i, cpu_map)) | |
6444 | continue; | |
9c1cfda2 | 6445 | |
dce840a0 PZ |
6446 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6447 | claim_allocations(i, sd); | |
cd4ea6ae | 6448 | init_sched_groups_power(i, sd); |
dce840a0 | 6449 | } |
f712c0c7 | 6450 | } |
9c1cfda2 | 6451 | |
1da177e4 | 6452 | /* Attach the domains */ |
dce840a0 | 6453 | rcu_read_lock(); |
abcd083a | 6454 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6455 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6456 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6457 | } |
dce840a0 | 6458 | rcu_read_unlock(); |
51888ca2 | 6459 | |
822ff793 | 6460 | ret = 0; |
51888ca2 | 6461 | error: |
2109b99e | 6462 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6463 | return ret; |
1da177e4 | 6464 | } |
029190c5 | 6465 | |
acc3f5d7 | 6466 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6467 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6468 | static struct sched_domain_attr *dattr_cur; |
6469 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6470 | |
6471 | /* | |
6472 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6473 | * cpumask) fails, then fallback to a single sched domain, |
6474 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6475 | */ |
4212823f | 6476 | static cpumask_var_t fallback_doms; |
029190c5 | 6477 | |
ee79d1bd HC |
6478 | /* |
6479 | * arch_update_cpu_topology lets virtualized architectures update the | |
6480 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6481 | * or 0 if it stayed the same. | |
6482 | */ | |
6483 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6484 | { |
ee79d1bd | 6485 | return 0; |
22e52b07 HC |
6486 | } |
6487 | ||
acc3f5d7 RR |
6488 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6489 | { | |
6490 | int i; | |
6491 | cpumask_var_t *doms; | |
6492 | ||
6493 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6494 | if (!doms) | |
6495 | return NULL; | |
6496 | for (i = 0; i < ndoms; i++) { | |
6497 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6498 | free_sched_domains(doms, i); | |
6499 | return NULL; | |
6500 | } | |
6501 | } | |
6502 | return doms; | |
6503 | } | |
6504 | ||
6505 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6506 | { | |
6507 | unsigned int i; | |
6508 | for (i = 0; i < ndoms; i++) | |
6509 | free_cpumask_var(doms[i]); | |
6510 | kfree(doms); | |
6511 | } | |
6512 | ||
1a20ff27 | 6513 | /* |
41a2d6cf | 6514 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6515 | * For now this just excludes isolated cpus, but could be used to |
6516 | * exclude other special cases in the future. | |
1a20ff27 | 6517 | */ |
c4a8849a | 6518 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6519 | { |
7378547f MM |
6520 | int err; |
6521 | ||
22e52b07 | 6522 | arch_update_cpu_topology(); |
029190c5 | 6523 | ndoms_cur = 1; |
acc3f5d7 | 6524 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6525 | if (!doms_cur) |
acc3f5d7 RR |
6526 | doms_cur = &fallback_doms; |
6527 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6528 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6529 | register_sched_domain_sysctl(); |
7378547f MM |
6530 | |
6531 | return err; | |
1a20ff27 DG |
6532 | } |
6533 | ||
1a20ff27 DG |
6534 | /* |
6535 | * Detach sched domains from a group of cpus specified in cpu_map | |
6536 | * These cpus will now be attached to the NULL domain | |
6537 | */ | |
96f874e2 | 6538 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6539 | { |
6540 | int i; | |
6541 | ||
dce840a0 | 6542 | rcu_read_lock(); |
abcd083a | 6543 | for_each_cpu(i, cpu_map) |
57d885fe | 6544 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6545 | rcu_read_unlock(); |
1a20ff27 DG |
6546 | } |
6547 | ||
1d3504fc HS |
6548 | /* handle null as "default" */ |
6549 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6550 | struct sched_domain_attr *new, int idx_new) | |
6551 | { | |
6552 | struct sched_domain_attr tmp; | |
6553 | ||
6554 | /* fast path */ | |
6555 | if (!new && !cur) | |
6556 | return 1; | |
6557 | ||
6558 | tmp = SD_ATTR_INIT; | |
6559 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6560 | new ? (new + idx_new) : &tmp, | |
6561 | sizeof(struct sched_domain_attr)); | |
6562 | } | |
6563 | ||
029190c5 PJ |
6564 | /* |
6565 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6566 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6567 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6568 | * It destroys each deleted domain and builds each new domain. | |
6569 | * | |
acc3f5d7 | 6570 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6571 | * The masks don't intersect (don't overlap.) We should setup one |
6572 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6573 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6574 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6575 | * it as it is. | |
6576 | * | |
acc3f5d7 RR |
6577 | * The passed in 'doms_new' should be allocated using |
6578 | * alloc_sched_domains. This routine takes ownership of it and will | |
6579 | * free_sched_domains it when done with it. If the caller failed the | |
6580 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6581 | * and partition_sched_domains() will fallback to the single partition | |
6582 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6583 | * |
96f874e2 | 6584 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6585 | * ndoms_new == 0 is a special case for destroying existing domains, |
6586 | * and it will not create the default domain. | |
dfb512ec | 6587 | * |
029190c5 PJ |
6588 | * Call with hotplug lock held |
6589 | */ | |
acc3f5d7 | 6590 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6591 | struct sched_domain_attr *dattr_new) |
029190c5 | 6592 | { |
dfb512ec | 6593 | int i, j, n; |
d65bd5ec | 6594 | int new_topology; |
029190c5 | 6595 | |
712555ee | 6596 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6597 | |
7378547f MM |
6598 | /* always unregister in case we don't destroy any domains */ |
6599 | unregister_sched_domain_sysctl(); | |
6600 | ||
d65bd5ec HC |
6601 | /* Let architecture update cpu core mappings. */ |
6602 | new_topology = arch_update_cpu_topology(); | |
6603 | ||
dfb512ec | 6604 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6605 | |
6606 | /* Destroy deleted domains */ | |
6607 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6608 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6609 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6610 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6611 | goto match1; |
6612 | } | |
6613 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6614 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6615 | match1: |
6616 | ; | |
6617 | } | |
6618 | ||
e761b772 MK |
6619 | if (doms_new == NULL) { |
6620 | ndoms_cur = 0; | |
acc3f5d7 | 6621 | doms_new = &fallback_doms; |
6ad4c188 | 6622 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6623 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6624 | } |
6625 | ||
029190c5 PJ |
6626 | /* Build new domains */ |
6627 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6628 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6629 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6630 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6631 | goto match2; |
6632 | } | |
6633 | /* no match - add a new doms_new */ | |
dce840a0 | 6634 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6635 | match2: |
6636 | ; | |
6637 | } | |
6638 | ||
6639 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6640 | if (doms_cur != &fallback_doms) |
6641 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6642 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6643 | doms_cur = doms_new; |
1d3504fc | 6644 | dattr_cur = dattr_new; |
029190c5 | 6645 | ndoms_cur = ndoms_new; |
7378547f MM |
6646 | |
6647 | register_sched_domain_sysctl(); | |
a1835615 | 6648 | |
712555ee | 6649 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6650 | } |
6651 | ||
d35be8ba SB |
6652 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6653 | ||
1da177e4 | 6654 | /* |
3a101d05 TH |
6655 | * Update cpusets according to cpu_active mask. If cpusets are |
6656 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6657 | * around partition_sched_domains(). | |
d35be8ba SB |
6658 | * |
6659 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6660 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6661 | */ |
0b2e918a TH |
6662 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6663 | void *hcpu) | |
e761b772 | 6664 | { |
d35be8ba SB |
6665 | switch (action) { |
6666 | case CPU_ONLINE_FROZEN: | |
6667 | case CPU_DOWN_FAILED_FROZEN: | |
6668 | ||
6669 | /* | |
6670 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6671 | * resume sequence. As long as this is not the last online | |
6672 | * operation in the resume sequence, just build a single sched | |
6673 | * domain, ignoring cpusets. | |
6674 | */ | |
6675 | num_cpus_frozen--; | |
6676 | if (likely(num_cpus_frozen)) { | |
6677 | partition_sched_domains(1, NULL, NULL); | |
6678 | break; | |
6679 | } | |
6680 | ||
6681 | /* | |
6682 | * This is the last CPU online operation. So fall through and | |
6683 | * restore the original sched domains by considering the | |
6684 | * cpuset configurations. | |
6685 | */ | |
6686 | ||
e761b772 | 6687 | case CPU_ONLINE: |
6ad4c188 | 6688 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6689 | cpuset_update_active_cpus(true); |
d35be8ba | 6690 | break; |
3a101d05 TH |
6691 | default: |
6692 | return NOTIFY_DONE; | |
6693 | } | |
d35be8ba | 6694 | return NOTIFY_OK; |
3a101d05 | 6695 | } |
e761b772 | 6696 | |
0b2e918a TH |
6697 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6698 | void *hcpu) | |
3a101d05 | 6699 | { |
d35be8ba | 6700 | switch (action) { |
3a101d05 | 6701 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6702 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6703 | break; |
6704 | case CPU_DOWN_PREPARE_FROZEN: | |
6705 | num_cpus_frozen++; | |
6706 | partition_sched_domains(1, NULL, NULL); | |
6707 | break; | |
e761b772 MK |
6708 | default: |
6709 | return NOTIFY_DONE; | |
6710 | } | |
d35be8ba | 6711 | return NOTIFY_OK; |
e761b772 | 6712 | } |
e761b772 | 6713 | |
1da177e4 LT |
6714 | void __init sched_init_smp(void) |
6715 | { | |
dcc30a35 RR |
6716 | cpumask_var_t non_isolated_cpus; |
6717 | ||
6718 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6719 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6720 | |
cb83b629 PZ |
6721 | sched_init_numa(); |
6722 | ||
95402b38 | 6723 | get_online_cpus(); |
712555ee | 6724 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6725 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6726 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6727 | if (cpumask_empty(non_isolated_cpus)) | |
6728 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6729 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6730 | put_online_cpus(); |
e761b772 | 6731 | |
301a5cba | 6732 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6733 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6734 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6735 | |
6736 | /* RT runtime code needs to handle some hotplug events */ | |
6737 | hotcpu_notifier(update_runtime, 0); | |
6738 | ||
b328ca18 | 6739 | init_hrtick(); |
5c1e1767 NP |
6740 | |
6741 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6742 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6743 | BUG(); |
19978ca6 | 6744 | sched_init_granularity(); |
dcc30a35 | 6745 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6746 | |
0e3900e6 | 6747 | init_sched_rt_class(); |
1da177e4 LT |
6748 | } |
6749 | #else | |
6750 | void __init sched_init_smp(void) | |
6751 | { | |
19978ca6 | 6752 | sched_init_granularity(); |
1da177e4 LT |
6753 | } |
6754 | #endif /* CONFIG_SMP */ | |
6755 | ||
cd1bb94b AB |
6756 | const_debug unsigned int sysctl_timer_migration = 1; |
6757 | ||
1da177e4 LT |
6758 | int in_sched_functions(unsigned long addr) |
6759 | { | |
1da177e4 LT |
6760 | return in_lock_functions(addr) || |
6761 | (addr >= (unsigned long)__sched_text_start | |
6762 | && addr < (unsigned long)__sched_text_end); | |
6763 | } | |
6764 | ||
029632fb PZ |
6765 | #ifdef CONFIG_CGROUP_SCHED |
6766 | struct task_group root_task_group; | |
35cf4e50 | 6767 | LIST_HEAD(task_groups); |
052f1dc7 | 6768 | #endif |
6f505b16 | 6769 | |
029632fb | 6770 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6771 | |
1da177e4 LT |
6772 | void __init sched_init(void) |
6773 | { | |
dd41f596 | 6774 | int i, j; |
434d53b0 MT |
6775 | unsigned long alloc_size = 0, ptr; |
6776 | ||
6777 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6778 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6779 | #endif | |
6780 | #ifdef CONFIG_RT_GROUP_SCHED | |
6781 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6782 | #endif |
df7c8e84 | 6783 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6784 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6785 | #endif |
434d53b0 | 6786 | if (alloc_size) { |
36b7b6d4 | 6787 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6788 | |
6789 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6790 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6791 | ptr += nr_cpu_ids * sizeof(void **); |
6792 | ||
07e06b01 | 6793 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6794 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6795 | |
6d6bc0ad | 6796 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6797 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6798 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6799 | ptr += nr_cpu_ids * sizeof(void **); |
6800 | ||
07e06b01 | 6801 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6802 | ptr += nr_cpu_ids * sizeof(void **); |
6803 | ||
6d6bc0ad | 6804 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6805 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6806 | for_each_possible_cpu(i) { | |
6807 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6808 | ptr += cpumask_size(); | |
6809 | } | |
6810 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6811 | } |
dd41f596 | 6812 | |
57d885fe GH |
6813 | #ifdef CONFIG_SMP |
6814 | init_defrootdomain(); | |
6815 | #endif | |
6816 | ||
d0b27fa7 PZ |
6817 | init_rt_bandwidth(&def_rt_bandwidth, |
6818 | global_rt_period(), global_rt_runtime()); | |
6819 | ||
6820 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6821 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6822 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6823 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6824 | |
7c941438 | 6825 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6826 | list_add(&root_task_group.list, &task_groups); |
6827 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6828 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6829 | autogroup_init(&init_task); |
54c707e9 | 6830 | |
7c941438 | 6831 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6832 | |
54c707e9 GC |
6833 | #ifdef CONFIG_CGROUP_CPUACCT |
6834 | root_cpuacct.cpustat = &kernel_cpustat; | |
6835 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
6836 | /* Too early, not expected to fail */ | |
6837 | BUG_ON(!root_cpuacct.cpuusage); | |
6838 | #endif | |
0a945022 | 6839 | for_each_possible_cpu(i) { |
70b97a7f | 6840 | struct rq *rq; |
1da177e4 LT |
6841 | |
6842 | rq = cpu_rq(i); | |
05fa785c | 6843 | raw_spin_lock_init(&rq->lock); |
7897986b | 6844 | rq->nr_running = 0; |
dce48a84 TG |
6845 | rq->calc_load_active = 0; |
6846 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6847 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6848 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6849 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6850 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6851 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6852 | /* |
07e06b01 | 6853 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6854 | * |
6855 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6856 | * gets 100% of the cpu resources in the system. This overall | |
6857 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6858 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6859 | * based on each entity's (task or task-group's) weight |
6860 | * (se->load.weight). | |
6861 | * | |
07e06b01 | 6862 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6863 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6864 | * then A0's share of the cpu resource is: | |
6865 | * | |
0d905bca | 6866 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6867 | * |
07e06b01 YZ |
6868 | * We achieve this by letting root_task_group's tasks sit |
6869 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6870 | */ |
ab84d31e | 6871 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6872 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6873 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6874 | ||
6875 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6876 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6877 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6878 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6879 | #endif |
1da177e4 | 6880 | |
dd41f596 IM |
6881 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6882 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6883 | |
6884 | rq->last_load_update_tick = jiffies; | |
6885 | ||
1da177e4 | 6886 | #ifdef CONFIG_SMP |
41c7ce9a | 6887 | rq->sd = NULL; |
57d885fe | 6888 | rq->rd = NULL; |
1399fa78 | 6889 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6890 | rq->post_schedule = 0; |
1da177e4 | 6891 | rq->active_balance = 0; |
dd41f596 | 6892 | rq->next_balance = jiffies; |
1da177e4 | 6893 | rq->push_cpu = 0; |
0a2966b4 | 6894 | rq->cpu = i; |
1f11eb6a | 6895 | rq->online = 0; |
eae0c9df MG |
6896 | rq->idle_stamp = 0; |
6897 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
6898 | |
6899 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6900 | ||
dc938520 | 6901 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 6902 | #ifdef CONFIG_NO_HZ |
1c792db7 | 6903 | rq->nohz_flags = 0; |
83cd4fe2 | 6904 | #endif |
1da177e4 | 6905 | #endif |
8f4d37ec | 6906 | init_rq_hrtick(rq); |
1da177e4 | 6907 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6908 | } |
6909 | ||
2dd73a4f | 6910 | set_load_weight(&init_task); |
b50f60ce | 6911 | |
e107be36 AK |
6912 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6913 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6914 | #endif | |
6915 | ||
b50f60ce | 6916 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6917 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6918 | #endif |
6919 | ||
1da177e4 LT |
6920 | /* |
6921 | * The boot idle thread does lazy MMU switching as well: | |
6922 | */ | |
6923 | atomic_inc(&init_mm.mm_count); | |
6924 | enter_lazy_tlb(&init_mm, current); | |
6925 | ||
6926 | /* | |
6927 | * Make us the idle thread. Technically, schedule() should not be | |
6928 | * called from this thread, however somewhere below it might be, | |
6929 | * but because we are the idle thread, we just pick up running again | |
6930 | * when this runqueue becomes "idle". | |
6931 | */ | |
6932 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6933 | |
6934 | calc_load_update = jiffies + LOAD_FREQ; | |
6935 | ||
dd41f596 IM |
6936 | /* |
6937 | * During early bootup we pretend to be a normal task: | |
6938 | */ | |
6939 | current->sched_class = &fair_sched_class; | |
6892b75e | 6940 | |
bf4d83f6 | 6941 | #ifdef CONFIG_SMP |
4cb98839 | 6942 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6943 | /* May be allocated at isolcpus cmdline parse time */ |
6944 | if (cpu_isolated_map == NULL) | |
6945 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6946 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6947 | #endif |
6948 | init_sched_fair_class(); | |
6a7b3dc3 | 6949 | |
6892b75e | 6950 | scheduler_running = 1; |
1da177e4 LT |
6951 | } |
6952 | ||
d902db1e | 6953 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6954 | static inline int preempt_count_equals(int preempt_offset) |
6955 | { | |
234da7bc | 6956 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6957 | |
4ba8216c | 6958 | return (nested == preempt_offset); |
e4aafea2 FW |
6959 | } |
6960 | ||
d894837f | 6961 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6962 | { |
1da177e4 LT |
6963 | static unsigned long prev_jiffy; /* ratelimiting */ |
6964 | ||
b3fbab05 | 6965 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6966 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
6967 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
6968 | return; |
6969 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6970 | return; | |
6971 | prev_jiffy = jiffies; | |
6972 | ||
3df0fc5b PZ |
6973 | printk(KERN_ERR |
6974 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6975 | file, line); | |
6976 | printk(KERN_ERR | |
6977 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6978 | in_atomic(), irqs_disabled(), | |
6979 | current->pid, current->comm); | |
aef745fc IM |
6980 | |
6981 | debug_show_held_locks(current); | |
6982 | if (irqs_disabled()) | |
6983 | print_irqtrace_events(current); | |
6984 | dump_stack(); | |
1da177e4 LT |
6985 | } |
6986 | EXPORT_SYMBOL(__might_sleep); | |
6987 | #endif | |
6988 | ||
6989 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6990 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6991 | { | |
da7a735e PZ |
6992 | const struct sched_class *prev_class = p->sched_class; |
6993 | int old_prio = p->prio; | |
3a5e4dc1 | 6994 | int on_rq; |
3e51f33f | 6995 | |
fd2f4419 | 6996 | on_rq = p->on_rq; |
3a5e4dc1 | 6997 | if (on_rq) |
4ca9b72b | 6998 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
6999 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7000 | if (on_rq) { | |
4ca9b72b | 7001 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7002 | resched_task(rq->curr); |
7003 | } | |
da7a735e PZ |
7004 | |
7005 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7006 | } |
7007 | ||
1da177e4 LT |
7008 | void normalize_rt_tasks(void) |
7009 | { | |
a0f98a1c | 7010 | struct task_struct *g, *p; |
1da177e4 | 7011 | unsigned long flags; |
70b97a7f | 7012 | struct rq *rq; |
1da177e4 | 7013 | |
4cf5d77a | 7014 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7015 | do_each_thread(g, p) { |
178be793 IM |
7016 | /* |
7017 | * Only normalize user tasks: | |
7018 | */ | |
7019 | if (!p->mm) | |
7020 | continue; | |
7021 | ||
6cfb0d5d | 7022 | p->se.exec_start = 0; |
6cfb0d5d | 7023 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7024 | p->se.statistics.wait_start = 0; |
7025 | p->se.statistics.sleep_start = 0; | |
7026 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7027 | #endif |
dd41f596 IM |
7028 | |
7029 | if (!rt_task(p)) { | |
7030 | /* | |
7031 | * Renice negative nice level userspace | |
7032 | * tasks back to 0: | |
7033 | */ | |
7034 | if (TASK_NICE(p) < 0 && p->mm) | |
7035 | set_user_nice(p, 0); | |
1da177e4 | 7036 | continue; |
dd41f596 | 7037 | } |
1da177e4 | 7038 | |
1d615482 | 7039 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7040 | rq = __task_rq_lock(p); |
1da177e4 | 7041 | |
178be793 | 7042 | normalize_task(rq, p); |
3a5e4dc1 | 7043 | |
b29739f9 | 7044 | __task_rq_unlock(rq); |
1d615482 | 7045 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7046 | } while_each_thread(g, p); |
7047 | ||
4cf5d77a | 7048 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7049 | } |
7050 | ||
7051 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7052 | |
67fc4e0c | 7053 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7054 | /* |
67fc4e0c | 7055 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7056 | * |
7057 | * They can only be called when the whole system has been | |
7058 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7059 | * activity can take place. Using them for anything else would | |
7060 | * be a serious bug, and as a result, they aren't even visible | |
7061 | * under any other configuration. | |
7062 | */ | |
7063 | ||
7064 | /** | |
7065 | * curr_task - return the current task for a given cpu. | |
7066 | * @cpu: the processor in question. | |
7067 | * | |
7068 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7069 | */ | |
36c8b586 | 7070 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7071 | { |
7072 | return cpu_curr(cpu); | |
7073 | } | |
7074 | ||
67fc4e0c JW |
7075 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7076 | ||
7077 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7078 | /** |
7079 | * set_curr_task - set the current task for a given cpu. | |
7080 | * @cpu: the processor in question. | |
7081 | * @p: the task pointer to set. | |
7082 | * | |
7083 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7084 | * are serviced on a separate stack. It allows the architecture to switch the |
7085 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7086 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7087 | * and caller must save the original value of the current task (see | |
7088 | * curr_task() above) and restore that value before reenabling interrupts and | |
7089 | * re-starting the system. | |
7090 | * | |
7091 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7092 | */ | |
36c8b586 | 7093 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7094 | { |
7095 | cpu_curr(cpu) = p; | |
7096 | } | |
7097 | ||
7098 | #endif | |
29f59db3 | 7099 | |
7c941438 | 7100 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7101 | /* task_group_lock serializes the addition/removal of task groups */ |
7102 | static DEFINE_SPINLOCK(task_group_lock); | |
7103 | ||
bccbe08a PZ |
7104 | static void free_sched_group(struct task_group *tg) |
7105 | { | |
7106 | free_fair_sched_group(tg); | |
7107 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7108 | autogroup_free(tg); |
bccbe08a PZ |
7109 | kfree(tg); |
7110 | } | |
7111 | ||
7112 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7113 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7114 | { |
7115 | struct task_group *tg; | |
7116 | unsigned long flags; | |
bccbe08a PZ |
7117 | |
7118 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7119 | if (!tg) | |
7120 | return ERR_PTR(-ENOMEM); | |
7121 | ||
ec7dc8ac | 7122 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7123 | goto err; |
7124 | ||
ec7dc8ac | 7125 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7126 | goto err; |
7127 | ||
8ed36996 | 7128 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7129 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7130 | |
7131 | WARN_ON(!parent); /* root should already exist */ | |
7132 | ||
7133 | tg->parent = parent; | |
f473aa5e | 7134 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7135 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7136 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7137 | |
9b5b7751 | 7138 | return tg; |
29f59db3 SV |
7139 | |
7140 | err: | |
6f505b16 | 7141 | free_sched_group(tg); |
29f59db3 SV |
7142 | return ERR_PTR(-ENOMEM); |
7143 | } | |
7144 | ||
9b5b7751 | 7145 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7146 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7147 | { |
29f59db3 | 7148 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7149 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7150 | } |
7151 | ||
9b5b7751 | 7152 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7153 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7154 | { |
8ed36996 | 7155 | unsigned long flags; |
9b5b7751 | 7156 | int i; |
29f59db3 | 7157 | |
3d4b47b4 PZ |
7158 | /* end participation in shares distribution */ |
7159 | for_each_possible_cpu(i) | |
bccbe08a | 7160 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7161 | |
7162 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7163 | list_del_rcu(&tg->list); |
f473aa5e | 7164 | list_del_rcu(&tg->siblings); |
8ed36996 | 7165 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7166 | |
9b5b7751 | 7167 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7168 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7169 | } |
7170 | ||
9b5b7751 | 7171 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7172 | * The caller of this function should have put the task in its new group |
7173 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7174 | * reflect its new group. | |
9b5b7751 SV |
7175 | */ |
7176 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7177 | { |
8323f26c | 7178 | struct task_group *tg; |
29f59db3 SV |
7179 | int on_rq, running; |
7180 | unsigned long flags; | |
7181 | struct rq *rq; | |
7182 | ||
7183 | rq = task_rq_lock(tsk, &flags); | |
7184 | ||
051a1d1a | 7185 | running = task_current(rq, tsk); |
fd2f4419 | 7186 | on_rq = tsk->on_rq; |
29f59db3 | 7187 | |
0e1f3483 | 7188 | if (on_rq) |
29f59db3 | 7189 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7190 | if (unlikely(running)) |
7191 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7192 | |
8323f26c PZ |
7193 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
7194 | lockdep_is_held(&tsk->sighand->siglock)), | |
7195 | struct task_group, css); | |
7196 | tg = autogroup_task_group(tsk, tg); | |
7197 | tsk->sched_task_group = tg; | |
7198 | ||
810b3817 | 7199 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7200 | if (tsk->sched_class->task_move_group) |
7201 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7202 | else | |
810b3817 | 7203 | #endif |
b2b5ce02 | 7204 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7205 | |
0e1f3483 HS |
7206 | if (unlikely(running)) |
7207 | tsk->sched_class->set_curr_task(rq); | |
7208 | if (on_rq) | |
371fd7e7 | 7209 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7210 | |
0122ec5b | 7211 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7212 | } |
7c941438 | 7213 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7214 | |
a790de99 | 7215 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7216 | static unsigned long to_ratio(u64 period, u64 runtime) |
7217 | { | |
7218 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7219 | return 1ULL << 20; |
9f0c1e56 | 7220 | |
9a7e0b18 | 7221 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7222 | } |
a790de99 PT |
7223 | #endif |
7224 | ||
7225 | #ifdef CONFIG_RT_GROUP_SCHED | |
7226 | /* | |
7227 | * Ensure that the real time constraints are schedulable. | |
7228 | */ | |
7229 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7230 | |
9a7e0b18 PZ |
7231 | /* Must be called with tasklist_lock held */ |
7232 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7233 | { |
9a7e0b18 | 7234 | struct task_struct *g, *p; |
b40b2e8e | 7235 | |
9a7e0b18 | 7236 | do_each_thread(g, p) { |
029632fb | 7237 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7238 | return 1; |
7239 | } while_each_thread(g, p); | |
b40b2e8e | 7240 | |
9a7e0b18 PZ |
7241 | return 0; |
7242 | } | |
b40b2e8e | 7243 | |
9a7e0b18 PZ |
7244 | struct rt_schedulable_data { |
7245 | struct task_group *tg; | |
7246 | u64 rt_period; | |
7247 | u64 rt_runtime; | |
7248 | }; | |
b40b2e8e | 7249 | |
a790de99 | 7250 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7251 | { |
7252 | struct rt_schedulable_data *d = data; | |
7253 | struct task_group *child; | |
7254 | unsigned long total, sum = 0; | |
7255 | u64 period, runtime; | |
b40b2e8e | 7256 | |
9a7e0b18 PZ |
7257 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7258 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7259 | |
9a7e0b18 PZ |
7260 | if (tg == d->tg) { |
7261 | period = d->rt_period; | |
7262 | runtime = d->rt_runtime; | |
b40b2e8e | 7263 | } |
b40b2e8e | 7264 | |
4653f803 PZ |
7265 | /* |
7266 | * Cannot have more runtime than the period. | |
7267 | */ | |
7268 | if (runtime > period && runtime != RUNTIME_INF) | |
7269 | return -EINVAL; | |
6f505b16 | 7270 | |
4653f803 PZ |
7271 | /* |
7272 | * Ensure we don't starve existing RT tasks. | |
7273 | */ | |
9a7e0b18 PZ |
7274 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7275 | return -EBUSY; | |
6f505b16 | 7276 | |
9a7e0b18 | 7277 | total = to_ratio(period, runtime); |
6f505b16 | 7278 | |
4653f803 PZ |
7279 | /* |
7280 | * Nobody can have more than the global setting allows. | |
7281 | */ | |
7282 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7283 | return -EINVAL; | |
6f505b16 | 7284 | |
4653f803 PZ |
7285 | /* |
7286 | * The sum of our children's runtime should not exceed our own. | |
7287 | */ | |
9a7e0b18 PZ |
7288 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7289 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7290 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7291 | |
9a7e0b18 PZ |
7292 | if (child == d->tg) { |
7293 | period = d->rt_period; | |
7294 | runtime = d->rt_runtime; | |
7295 | } | |
6f505b16 | 7296 | |
9a7e0b18 | 7297 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7298 | } |
6f505b16 | 7299 | |
9a7e0b18 PZ |
7300 | if (sum > total) |
7301 | return -EINVAL; | |
7302 | ||
7303 | return 0; | |
6f505b16 PZ |
7304 | } |
7305 | ||
9a7e0b18 | 7306 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7307 | { |
8277434e PT |
7308 | int ret; |
7309 | ||
9a7e0b18 PZ |
7310 | struct rt_schedulable_data data = { |
7311 | .tg = tg, | |
7312 | .rt_period = period, | |
7313 | .rt_runtime = runtime, | |
7314 | }; | |
7315 | ||
8277434e PT |
7316 | rcu_read_lock(); |
7317 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7318 | rcu_read_unlock(); | |
7319 | ||
7320 | return ret; | |
521f1a24 DG |
7321 | } |
7322 | ||
ab84d31e | 7323 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7324 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7325 | { |
ac086bc2 | 7326 | int i, err = 0; |
9f0c1e56 | 7327 | |
9f0c1e56 | 7328 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7329 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7330 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7331 | if (err) | |
9f0c1e56 | 7332 | goto unlock; |
ac086bc2 | 7333 | |
0986b11b | 7334 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7335 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7336 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7337 | |
7338 | for_each_possible_cpu(i) { | |
7339 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7340 | ||
0986b11b | 7341 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7342 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7343 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7344 | } |
0986b11b | 7345 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7346 | unlock: |
521f1a24 | 7347 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7348 | mutex_unlock(&rt_constraints_mutex); |
7349 | ||
7350 | return err; | |
6f505b16 PZ |
7351 | } |
7352 | ||
d0b27fa7 PZ |
7353 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7354 | { | |
7355 | u64 rt_runtime, rt_period; | |
7356 | ||
7357 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7358 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7359 | if (rt_runtime_us < 0) | |
7360 | rt_runtime = RUNTIME_INF; | |
7361 | ||
ab84d31e | 7362 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7363 | } |
7364 | ||
9f0c1e56 PZ |
7365 | long sched_group_rt_runtime(struct task_group *tg) |
7366 | { | |
7367 | u64 rt_runtime_us; | |
7368 | ||
d0b27fa7 | 7369 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7370 | return -1; |
7371 | ||
d0b27fa7 | 7372 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7373 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7374 | return rt_runtime_us; | |
7375 | } | |
d0b27fa7 PZ |
7376 | |
7377 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7378 | { | |
7379 | u64 rt_runtime, rt_period; | |
7380 | ||
7381 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7382 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7383 | ||
619b0488 R |
7384 | if (rt_period == 0) |
7385 | return -EINVAL; | |
7386 | ||
ab84d31e | 7387 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7388 | } |
7389 | ||
7390 | long sched_group_rt_period(struct task_group *tg) | |
7391 | { | |
7392 | u64 rt_period_us; | |
7393 | ||
7394 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7395 | do_div(rt_period_us, NSEC_PER_USEC); | |
7396 | return rt_period_us; | |
7397 | } | |
7398 | ||
7399 | static int sched_rt_global_constraints(void) | |
7400 | { | |
4653f803 | 7401 | u64 runtime, period; |
d0b27fa7 PZ |
7402 | int ret = 0; |
7403 | ||
ec5d4989 HS |
7404 | if (sysctl_sched_rt_period <= 0) |
7405 | return -EINVAL; | |
7406 | ||
4653f803 PZ |
7407 | runtime = global_rt_runtime(); |
7408 | period = global_rt_period(); | |
7409 | ||
7410 | /* | |
7411 | * Sanity check on the sysctl variables. | |
7412 | */ | |
7413 | if (runtime > period && runtime != RUNTIME_INF) | |
7414 | return -EINVAL; | |
10b612f4 | 7415 | |
d0b27fa7 | 7416 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7417 | read_lock(&tasklist_lock); |
4653f803 | 7418 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7419 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7420 | mutex_unlock(&rt_constraints_mutex); |
7421 | ||
7422 | return ret; | |
7423 | } | |
54e99124 DG |
7424 | |
7425 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7426 | { | |
7427 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7428 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7429 | return 0; | |
7430 | ||
7431 | return 1; | |
7432 | } | |
7433 | ||
6d6bc0ad | 7434 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7435 | static int sched_rt_global_constraints(void) |
7436 | { | |
ac086bc2 PZ |
7437 | unsigned long flags; |
7438 | int i; | |
7439 | ||
ec5d4989 HS |
7440 | if (sysctl_sched_rt_period <= 0) |
7441 | return -EINVAL; | |
7442 | ||
60aa605d PZ |
7443 | /* |
7444 | * There's always some RT tasks in the root group | |
7445 | * -- migration, kstopmachine etc.. | |
7446 | */ | |
7447 | if (sysctl_sched_rt_runtime == 0) | |
7448 | return -EBUSY; | |
7449 | ||
0986b11b | 7450 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7451 | for_each_possible_cpu(i) { |
7452 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7453 | ||
0986b11b | 7454 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7455 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7456 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7457 | } |
0986b11b | 7458 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7459 | |
d0b27fa7 PZ |
7460 | return 0; |
7461 | } | |
6d6bc0ad | 7462 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7463 | |
7464 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7465 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7466 | loff_t *ppos) |
7467 | { | |
7468 | int ret; | |
7469 | int old_period, old_runtime; | |
7470 | static DEFINE_MUTEX(mutex); | |
7471 | ||
7472 | mutex_lock(&mutex); | |
7473 | old_period = sysctl_sched_rt_period; | |
7474 | old_runtime = sysctl_sched_rt_runtime; | |
7475 | ||
8d65af78 | 7476 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7477 | |
7478 | if (!ret && write) { | |
7479 | ret = sched_rt_global_constraints(); | |
7480 | if (ret) { | |
7481 | sysctl_sched_rt_period = old_period; | |
7482 | sysctl_sched_rt_runtime = old_runtime; | |
7483 | } else { | |
7484 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7485 | def_rt_bandwidth.rt_period = | |
7486 | ns_to_ktime(global_rt_period()); | |
7487 | } | |
7488 | } | |
7489 | mutex_unlock(&mutex); | |
7490 | ||
7491 | return ret; | |
7492 | } | |
68318b8e | 7493 | |
052f1dc7 | 7494 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7495 | |
7496 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7497 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7498 | { |
2b01dfe3 PM |
7499 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7500 | struct task_group, css); | |
68318b8e SV |
7501 | } |
7502 | ||
761b3ef5 | 7503 | static struct cgroup_subsys_state *cpu_cgroup_create(struct cgroup *cgrp) |
68318b8e | 7504 | { |
ec7dc8ac | 7505 | struct task_group *tg, *parent; |
68318b8e | 7506 | |
2b01dfe3 | 7507 | if (!cgrp->parent) { |
68318b8e | 7508 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7509 | return &root_task_group.css; |
68318b8e SV |
7510 | } |
7511 | ||
ec7dc8ac DG |
7512 | parent = cgroup_tg(cgrp->parent); |
7513 | tg = sched_create_group(parent); | |
68318b8e SV |
7514 | if (IS_ERR(tg)) |
7515 | return ERR_PTR(-ENOMEM); | |
7516 | ||
68318b8e SV |
7517 | return &tg->css; |
7518 | } | |
7519 | ||
761b3ef5 | 7520 | static void cpu_cgroup_destroy(struct cgroup *cgrp) |
68318b8e | 7521 | { |
2b01dfe3 | 7522 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7523 | |
7524 | sched_destroy_group(tg); | |
7525 | } | |
7526 | ||
761b3ef5 | 7527 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 7528 | struct cgroup_taskset *tset) |
68318b8e | 7529 | { |
bb9d97b6 TH |
7530 | struct task_struct *task; |
7531 | ||
7532 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7533 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7534 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7535 | return -EINVAL; | |
b68aa230 | 7536 | #else |
bb9d97b6 TH |
7537 | /* We don't support RT-tasks being in separate groups */ |
7538 | if (task->sched_class != &fair_sched_class) | |
7539 | return -EINVAL; | |
b68aa230 | 7540 | #endif |
bb9d97b6 | 7541 | } |
be367d09 BB |
7542 | return 0; |
7543 | } | |
68318b8e | 7544 | |
761b3ef5 | 7545 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 7546 | struct cgroup_taskset *tset) |
68318b8e | 7547 | { |
bb9d97b6 TH |
7548 | struct task_struct *task; |
7549 | ||
7550 | cgroup_taskset_for_each(task, cgrp, tset) | |
7551 | sched_move_task(task); | |
68318b8e SV |
7552 | } |
7553 | ||
068c5cc5 | 7554 | static void |
761b3ef5 LZ |
7555 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
7556 | struct task_struct *task) | |
068c5cc5 PZ |
7557 | { |
7558 | /* | |
7559 | * cgroup_exit() is called in the copy_process() failure path. | |
7560 | * Ignore this case since the task hasn't ran yet, this avoids | |
7561 | * trying to poke a half freed task state from generic code. | |
7562 | */ | |
7563 | if (!(task->flags & PF_EXITING)) | |
7564 | return; | |
7565 | ||
7566 | sched_move_task(task); | |
7567 | } | |
7568 | ||
052f1dc7 | 7569 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7570 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7571 | u64 shareval) |
68318b8e | 7572 | { |
c8b28116 | 7573 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7574 | } |
7575 | ||
f4c753b7 | 7576 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7577 | { |
2b01dfe3 | 7578 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7579 | |
c8b28116 | 7580 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7581 | } |
ab84d31e PT |
7582 | |
7583 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7584 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7585 | ||
ab84d31e PT |
7586 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7587 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7588 | ||
a790de99 PT |
7589 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7590 | ||
ab84d31e PT |
7591 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7592 | { | |
56f570e5 | 7593 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7594 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7595 | |
7596 | if (tg == &root_task_group) | |
7597 | return -EINVAL; | |
7598 | ||
7599 | /* | |
7600 | * Ensure we have at some amount of bandwidth every period. This is | |
7601 | * to prevent reaching a state of large arrears when throttled via | |
7602 | * entity_tick() resulting in prolonged exit starvation. | |
7603 | */ | |
7604 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7605 | return -EINVAL; | |
7606 | ||
7607 | /* | |
7608 | * Likewise, bound things on the otherside by preventing insane quota | |
7609 | * periods. This also allows us to normalize in computing quota | |
7610 | * feasibility. | |
7611 | */ | |
7612 | if (period > max_cfs_quota_period) | |
7613 | return -EINVAL; | |
7614 | ||
a790de99 PT |
7615 | mutex_lock(&cfs_constraints_mutex); |
7616 | ret = __cfs_schedulable(tg, period, quota); | |
7617 | if (ret) | |
7618 | goto out_unlock; | |
7619 | ||
58088ad0 | 7620 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7621 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7622 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7623 | raw_spin_lock_irq(&cfs_b->lock); |
7624 | cfs_b->period = ns_to_ktime(period); | |
7625 | cfs_b->quota = quota; | |
58088ad0 | 7626 | |
a9cf55b2 | 7627 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7628 | /* restart the period timer (if active) to handle new period expiry */ |
7629 | if (runtime_enabled && cfs_b->timer_active) { | |
7630 | /* force a reprogram */ | |
7631 | cfs_b->timer_active = 0; | |
7632 | __start_cfs_bandwidth(cfs_b); | |
7633 | } | |
ab84d31e PT |
7634 | raw_spin_unlock_irq(&cfs_b->lock); |
7635 | ||
7636 | for_each_possible_cpu(i) { | |
7637 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7638 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7639 | |
7640 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7641 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7642 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7643 | |
029632fb | 7644 | if (cfs_rq->throttled) |
671fd9da | 7645 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7646 | raw_spin_unlock_irq(&rq->lock); |
7647 | } | |
a790de99 PT |
7648 | out_unlock: |
7649 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7650 | |
a790de99 | 7651 | return ret; |
ab84d31e PT |
7652 | } |
7653 | ||
7654 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7655 | { | |
7656 | u64 quota, period; | |
7657 | ||
029632fb | 7658 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7659 | if (cfs_quota_us < 0) |
7660 | quota = RUNTIME_INF; | |
7661 | else | |
7662 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7663 | ||
7664 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7665 | } | |
7666 | ||
7667 | long tg_get_cfs_quota(struct task_group *tg) | |
7668 | { | |
7669 | u64 quota_us; | |
7670 | ||
029632fb | 7671 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7672 | return -1; |
7673 | ||
029632fb | 7674 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7675 | do_div(quota_us, NSEC_PER_USEC); |
7676 | ||
7677 | return quota_us; | |
7678 | } | |
7679 | ||
7680 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7681 | { | |
7682 | u64 quota, period; | |
7683 | ||
7684 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7685 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7686 | |
ab84d31e PT |
7687 | return tg_set_cfs_bandwidth(tg, period, quota); |
7688 | } | |
7689 | ||
7690 | long tg_get_cfs_period(struct task_group *tg) | |
7691 | { | |
7692 | u64 cfs_period_us; | |
7693 | ||
029632fb | 7694 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7695 | do_div(cfs_period_us, NSEC_PER_USEC); |
7696 | ||
7697 | return cfs_period_us; | |
7698 | } | |
7699 | ||
7700 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7701 | { | |
7702 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7703 | } | |
7704 | ||
7705 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7706 | s64 cfs_quota_us) | |
7707 | { | |
7708 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7709 | } | |
7710 | ||
7711 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7712 | { | |
7713 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7714 | } | |
7715 | ||
7716 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7717 | u64 cfs_period_us) | |
7718 | { | |
7719 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7720 | } | |
7721 | ||
a790de99 PT |
7722 | struct cfs_schedulable_data { |
7723 | struct task_group *tg; | |
7724 | u64 period, quota; | |
7725 | }; | |
7726 | ||
7727 | /* | |
7728 | * normalize group quota/period to be quota/max_period | |
7729 | * note: units are usecs | |
7730 | */ | |
7731 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7732 | struct cfs_schedulable_data *d) | |
7733 | { | |
7734 | u64 quota, period; | |
7735 | ||
7736 | if (tg == d->tg) { | |
7737 | period = d->period; | |
7738 | quota = d->quota; | |
7739 | } else { | |
7740 | period = tg_get_cfs_period(tg); | |
7741 | quota = tg_get_cfs_quota(tg); | |
7742 | } | |
7743 | ||
7744 | /* note: these should typically be equivalent */ | |
7745 | if (quota == RUNTIME_INF || quota == -1) | |
7746 | return RUNTIME_INF; | |
7747 | ||
7748 | return to_ratio(period, quota); | |
7749 | } | |
7750 | ||
7751 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7752 | { | |
7753 | struct cfs_schedulable_data *d = data; | |
029632fb | 7754 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7755 | s64 quota = 0, parent_quota = -1; |
7756 | ||
7757 | if (!tg->parent) { | |
7758 | quota = RUNTIME_INF; | |
7759 | } else { | |
029632fb | 7760 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7761 | |
7762 | quota = normalize_cfs_quota(tg, d); | |
7763 | parent_quota = parent_b->hierarchal_quota; | |
7764 | ||
7765 | /* | |
7766 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7767 | * limit is set | |
7768 | */ | |
7769 | if (quota == RUNTIME_INF) | |
7770 | quota = parent_quota; | |
7771 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7772 | return -EINVAL; | |
7773 | } | |
7774 | cfs_b->hierarchal_quota = quota; | |
7775 | ||
7776 | return 0; | |
7777 | } | |
7778 | ||
7779 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7780 | { | |
8277434e | 7781 | int ret; |
a790de99 PT |
7782 | struct cfs_schedulable_data data = { |
7783 | .tg = tg, | |
7784 | .period = period, | |
7785 | .quota = quota, | |
7786 | }; | |
7787 | ||
7788 | if (quota != RUNTIME_INF) { | |
7789 | do_div(data.period, NSEC_PER_USEC); | |
7790 | do_div(data.quota, NSEC_PER_USEC); | |
7791 | } | |
7792 | ||
8277434e PT |
7793 | rcu_read_lock(); |
7794 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7795 | rcu_read_unlock(); | |
7796 | ||
7797 | return ret; | |
a790de99 | 7798 | } |
e8da1b18 NR |
7799 | |
7800 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7801 | struct cgroup_map_cb *cb) | |
7802 | { | |
7803 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7804 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7805 | |
7806 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7807 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7808 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7809 | ||
7810 | return 0; | |
7811 | } | |
ab84d31e | 7812 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7813 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7814 | |
052f1dc7 | 7815 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7816 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7817 | s64 val) |
6f505b16 | 7818 | { |
06ecb27c | 7819 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7820 | } |
7821 | ||
06ecb27c | 7822 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7823 | { |
06ecb27c | 7824 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7825 | } |
d0b27fa7 PZ |
7826 | |
7827 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7828 | u64 rt_period_us) | |
7829 | { | |
7830 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7831 | } | |
7832 | ||
7833 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7834 | { | |
7835 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7836 | } | |
6d6bc0ad | 7837 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7838 | |
fe5c7cc2 | 7839 | static struct cftype cpu_files[] = { |
052f1dc7 | 7840 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7841 | { |
7842 | .name = "shares", | |
f4c753b7 PM |
7843 | .read_u64 = cpu_shares_read_u64, |
7844 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7845 | }, |
052f1dc7 | 7846 | #endif |
ab84d31e PT |
7847 | #ifdef CONFIG_CFS_BANDWIDTH |
7848 | { | |
7849 | .name = "cfs_quota_us", | |
7850 | .read_s64 = cpu_cfs_quota_read_s64, | |
7851 | .write_s64 = cpu_cfs_quota_write_s64, | |
7852 | }, | |
7853 | { | |
7854 | .name = "cfs_period_us", | |
7855 | .read_u64 = cpu_cfs_period_read_u64, | |
7856 | .write_u64 = cpu_cfs_period_write_u64, | |
7857 | }, | |
e8da1b18 NR |
7858 | { |
7859 | .name = "stat", | |
7860 | .read_map = cpu_stats_show, | |
7861 | }, | |
ab84d31e | 7862 | #endif |
052f1dc7 | 7863 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7864 | { |
9f0c1e56 | 7865 | .name = "rt_runtime_us", |
06ecb27c PM |
7866 | .read_s64 = cpu_rt_runtime_read, |
7867 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7868 | }, |
d0b27fa7 PZ |
7869 | { |
7870 | .name = "rt_period_us", | |
f4c753b7 PM |
7871 | .read_u64 = cpu_rt_period_read_uint, |
7872 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7873 | }, |
052f1dc7 | 7874 | #endif |
4baf6e33 | 7875 | { } /* terminate */ |
68318b8e SV |
7876 | }; |
7877 | ||
68318b8e | 7878 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae IM |
7879 | .name = "cpu", |
7880 | .create = cpu_cgroup_create, | |
7881 | .destroy = cpu_cgroup_destroy, | |
bb9d97b6 TH |
7882 | .can_attach = cpu_cgroup_can_attach, |
7883 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7884 | .exit = cpu_cgroup_exit, |
38605cae | 7885 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 7886 | .base_cftypes = cpu_files, |
68318b8e SV |
7887 | .early_init = 1, |
7888 | }; | |
7889 | ||
052f1dc7 | 7890 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7891 | |
7892 | #ifdef CONFIG_CGROUP_CPUACCT | |
7893 | ||
7894 | /* | |
7895 | * CPU accounting code for task groups. | |
7896 | * | |
7897 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7898 | * (balbir@in.ibm.com). | |
7899 | */ | |
7900 | ||
73fbec60 FW |
7901 | struct cpuacct root_cpuacct; |
7902 | ||
d842de87 | 7903 | /* create a new cpu accounting group */ |
761b3ef5 | 7904 | static struct cgroup_subsys_state *cpuacct_create(struct cgroup *cgrp) |
d842de87 | 7905 | { |
54c707e9 | 7906 | struct cpuacct *ca; |
d842de87 | 7907 | |
54c707e9 GC |
7908 | if (!cgrp->parent) |
7909 | return &root_cpuacct.css; | |
7910 | ||
7911 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 7912 | if (!ca) |
ef12fefa | 7913 | goto out; |
d842de87 SV |
7914 | |
7915 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
7916 | if (!ca->cpuusage) |
7917 | goto out_free_ca; | |
7918 | ||
54c707e9 GC |
7919 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
7920 | if (!ca->cpustat) | |
7921 | goto out_free_cpuusage; | |
934352f2 | 7922 | |
d842de87 | 7923 | return &ca->css; |
ef12fefa | 7924 | |
54c707e9 | 7925 | out_free_cpuusage: |
ef12fefa BR |
7926 | free_percpu(ca->cpuusage); |
7927 | out_free_ca: | |
7928 | kfree(ca); | |
7929 | out: | |
7930 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
7931 | } |
7932 | ||
7933 | /* destroy an existing cpu accounting group */ | |
761b3ef5 | 7934 | static void cpuacct_destroy(struct cgroup *cgrp) |
d842de87 | 7935 | { |
32cd756a | 7936 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 7937 | |
54c707e9 | 7938 | free_percpu(ca->cpustat); |
d842de87 SV |
7939 | free_percpu(ca->cpuusage); |
7940 | kfree(ca); | |
7941 | } | |
7942 | ||
720f5498 KC |
7943 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
7944 | { | |
b36128c8 | 7945 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7946 | u64 data; |
7947 | ||
7948 | #ifndef CONFIG_64BIT | |
7949 | /* | |
7950 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
7951 | */ | |
05fa785c | 7952 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 7953 | data = *cpuusage; |
05fa785c | 7954 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
7955 | #else |
7956 | data = *cpuusage; | |
7957 | #endif | |
7958 | ||
7959 | return data; | |
7960 | } | |
7961 | ||
7962 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
7963 | { | |
b36128c8 | 7964 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7965 | |
7966 | #ifndef CONFIG_64BIT | |
7967 | /* | |
7968 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
7969 | */ | |
05fa785c | 7970 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 7971 | *cpuusage = val; |
05fa785c | 7972 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
7973 | #else |
7974 | *cpuusage = val; | |
7975 | #endif | |
7976 | } | |
7977 | ||
d842de87 | 7978 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 7979 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 7980 | { |
32cd756a | 7981 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
7982 | u64 totalcpuusage = 0; |
7983 | int i; | |
7984 | ||
720f5498 KC |
7985 | for_each_present_cpu(i) |
7986 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
7987 | |
7988 | return totalcpuusage; | |
7989 | } | |
7990 | ||
0297b803 DG |
7991 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
7992 | u64 reset) | |
7993 | { | |
7994 | struct cpuacct *ca = cgroup_ca(cgrp); | |
7995 | int err = 0; | |
7996 | int i; | |
7997 | ||
7998 | if (reset) { | |
7999 | err = -EINVAL; | |
8000 | goto out; | |
8001 | } | |
8002 | ||
720f5498 KC |
8003 | for_each_present_cpu(i) |
8004 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8005 | |
0297b803 DG |
8006 | out: |
8007 | return err; | |
8008 | } | |
8009 | ||
e9515c3c KC |
8010 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8011 | struct seq_file *m) | |
8012 | { | |
8013 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8014 | u64 percpu; | |
8015 | int i; | |
8016 | ||
8017 | for_each_present_cpu(i) { | |
8018 | percpu = cpuacct_cpuusage_read(ca, i); | |
8019 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8020 | } | |
8021 | seq_printf(m, "\n"); | |
8022 | return 0; | |
8023 | } | |
8024 | ||
ef12fefa BR |
8025 | static const char *cpuacct_stat_desc[] = { |
8026 | [CPUACCT_STAT_USER] = "user", | |
8027 | [CPUACCT_STAT_SYSTEM] = "system", | |
8028 | }; | |
8029 | ||
8030 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8031 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8032 | { |
8033 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8034 | int cpu; |
8035 | s64 val = 0; | |
ef12fefa | 8036 | |
54c707e9 GC |
8037 | for_each_online_cpu(cpu) { |
8038 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8039 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8040 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8041 | } |
54c707e9 GC |
8042 | val = cputime64_to_clock_t(val); |
8043 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8044 | |
54c707e9 GC |
8045 | val = 0; |
8046 | for_each_online_cpu(cpu) { | |
8047 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8048 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8049 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8050 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8051 | } |
54c707e9 GC |
8052 | |
8053 | val = cputime64_to_clock_t(val); | |
8054 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8055 | ||
ef12fefa BR |
8056 | return 0; |
8057 | } | |
8058 | ||
d842de87 SV |
8059 | static struct cftype files[] = { |
8060 | { | |
8061 | .name = "usage", | |
f4c753b7 PM |
8062 | .read_u64 = cpuusage_read, |
8063 | .write_u64 = cpuusage_write, | |
d842de87 | 8064 | }, |
e9515c3c KC |
8065 | { |
8066 | .name = "usage_percpu", | |
8067 | .read_seq_string = cpuacct_percpu_seq_read, | |
8068 | }, | |
ef12fefa BR |
8069 | { |
8070 | .name = "stat", | |
8071 | .read_map = cpuacct_stats_show, | |
8072 | }, | |
4baf6e33 | 8073 | { } /* terminate */ |
d842de87 SV |
8074 | }; |
8075 | ||
d842de87 SV |
8076 | /* |
8077 | * charge this task's execution time to its accounting group. | |
8078 | * | |
8079 | * called with rq->lock held. | |
8080 | */ | |
029632fb | 8081 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8082 | { |
8083 | struct cpuacct *ca; | |
934352f2 | 8084 | int cpu; |
d842de87 | 8085 | |
c40c6f85 | 8086 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8087 | return; |
8088 | ||
934352f2 | 8089 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8090 | |
8091 | rcu_read_lock(); | |
8092 | ||
d842de87 | 8093 | ca = task_ca(tsk); |
d842de87 | 8094 | |
44252e42 | 8095 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8096 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8097 | *cpuusage += cputime; |
8098 | } | |
a18b83b7 BR |
8099 | |
8100 | rcu_read_unlock(); | |
d842de87 SV |
8101 | } |
8102 | ||
8103 | struct cgroup_subsys cpuacct_subsys = { | |
8104 | .name = "cpuacct", | |
8105 | .create = cpuacct_create, | |
8106 | .destroy = cpuacct_destroy, | |
d842de87 | 8107 | .subsys_id = cpuacct_subsys_id, |
4baf6e33 | 8108 | .base_cftypes = files, |
d842de87 SV |
8109 | }; |
8110 | #endif /* CONFIG_CGROUP_CPUACCT */ |