]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
1da177e4 | 76 | |
96f951ed | 77 | #include <asm/switch_to.h> |
5517d86b | 78 | #include <asm/tlb.h> |
838225b4 | 79 | #include <asm/irq_regs.h> |
db7e527d | 80 | #include <asm/mutex.h> |
e6e6685a GC |
81 | #ifdef CONFIG_PARAVIRT |
82 | #include <asm/paravirt.h> | |
83 | #endif | |
1da177e4 | 84 | |
029632fb | 85 | #include "sched.h" |
ea138446 | 86 | #include "../workqueue_internal.h" |
29d5e047 | 87 | #include "../smpboot.h" |
6e0534f2 | 88 | |
a8d154b0 | 89 | #define CREATE_TRACE_POINTS |
ad8d75ff | 90 | #include <trace/events/sched.h> |
a8d154b0 | 91 | |
029632fb | 92 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 93 | { |
58088ad0 PT |
94 | unsigned long delta; |
95 | ktime_t soft, hard, now; | |
d0b27fa7 | 96 | |
58088ad0 PT |
97 | for (;;) { |
98 | if (hrtimer_active(period_timer)) | |
99 | break; | |
100 | ||
101 | now = hrtimer_cb_get_time(period_timer); | |
102 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 103 | |
58088ad0 PT |
104 | soft = hrtimer_get_softexpires(period_timer); |
105 | hard = hrtimer_get_expires(period_timer); | |
106 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
107 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
108 | HRTIMER_MODE_ABS_PINNED, 0); | |
109 | } | |
110 | } | |
111 | ||
029632fb PZ |
112 | DEFINE_MUTEX(sched_domains_mutex); |
113 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 114 | |
fe44d621 | 115 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 116 | |
029632fb | 117 | void update_rq_clock(struct rq *rq) |
3e51f33f | 118 | { |
fe44d621 | 119 | s64 delta; |
305e6835 | 120 | |
61eadef6 | 121 | if (rq->skip_clock_update > 0) |
f26f9aff | 122 | return; |
aa483808 | 123 | |
fe44d621 PZ |
124 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
125 | rq->clock += delta; | |
126 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
127 | } |
128 | ||
bf5c91ba IM |
129 | /* |
130 | * Debugging: various feature bits | |
131 | */ | |
f00b45c1 | 132 | |
f00b45c1 PZ |
133 | #define SCHED_FEAT(name, enabled) \ |
134 | (1UL << __SCHED_FEAT_##name) * enabled | | |
135 | ||
bf5c91ba | 136 | const_debug unsigned int sysctl_sched_features = |
391e43da | 137 | #include "features.h" |
f00b45c1 PZ |
138 | 0; |
139 | ||
140 | #undef SCHED_FEAT | |
141 | ||
142 | #ifdef CONFIG_SCHED_DEBUG | |
143 | #define SCHED_FEAT(name, enabled) \ | |
144 | #name , | |
145 | ||
1292531f | 146 | static const char * const sched_feat_names[] = { |
391e43da | 147 | #include "features.h" |
f00b45c1 PZ |
148 | }; |
149 | ||
150 | #undef SCHED_FEAT | |
151 | ||
34f3a814 | 152 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 153 | { |
f00b45c1 PZ |
154 | int i; |
155 | ||
f8b6d1cc | 156 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
157 | if (!(sysctl_sched_features & (1UL << i))) |
158 | seq_puts(m, "NO_"); | |
159 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 160 | } |
34f3a814 | 161 | seq_puts(m, "\n"); |
f00b45c1 | 162 | |
34f3a814 | 163 | return 0; |
f00b45c1 PZ |
164 | } |
165 | ||
f8b6d1cc PZ |
166 | #ifdef HAVE_JUMP_LABEL |
167 | ||
c5905afb IM |
168 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
169 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
170 | |
171 | #define SCHED_FEAT(name, enabled) \ | |
172 | jump_label_key__##enabled , | |
173 | ||
c5905afb | 174 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
175 | #include "features.h" |
176 | }; | |
177 | ||
178 | #undef SCHED_FEAT | |
179 | ||
180 | static void sched_feat_disable(int i) | |
181 | { | |
c5905afb IM |
182 | if (static_key_enabled(&sched_feat_keys[i])) |
183 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
184 | } |
185 | ||
186 | static void sched_feat_enable(int i) | |
187 | { | |
c5905afb IM |
188 | if (!static_key_enabled(&sched_feat_keys[i])) |
189 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
190 | } |
191 | #else | |
192 | static void sched_feat_disable(int i) { }; | |
193 | static void sched_feat_enable(int i) { }; | |
194 | #endif /* HAVE_JUMP_LABEL */ | |
195 | ||
1a687c2e | 196 | static int sched_feat_set(char *cmp) |
f00b45c1 | 197 | { |
f00b45c1 | 198 | int i; |
1a687c2e | 199 | int neg = 0; |
f00b45c1 | 200 | |
524429c3 | 201 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
202 | neg = 1; |
203 | cmp += 3; | |
204 | } | |
205 | ||
f8b6d1cc | 206 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 207 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 208 | if (neg) { |
f00b45c1 | 209 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
210 | sched_feat_disable(i); |
211 | } else { | |
f00b45c1 | 212 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
213 | sched_feat_enable(i); |
214 | } | |
f00b45c1 PZ |
215 | break; |
216 | } | |
217 | } | |
218 | ||
1a687c2e MG |
219 | return i; |
220 | } | |
221 | ||
222 | static ssize_t | |
223 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
224 | size_t cnt, loff_t *ppos) | |
225 | { | |
226 | char buf[64]; | |
227 | char *cmp; | |
228 | int i; | |
229 | ||
230 | if (cnt > 63) | |
231 | cnt = 63; | |
232 | ||
233 | if (copy_from_user(&buf, ubuf, cnt)) | |
234 | return -EFAULT; | |
235 | ||
236 | buf[cnt] = 0; | |
237 | cmp = strstrip(buf); | |
238 | ||
239 | i = sched_feat_set(cmp); | |
f8b6d1cc | 240 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
241 | return -EINVAL; |
242 | ||
42994724 | 243 | *ppos += cnt; |
f00b45c1 PZ |
244 | |
245 | return cnt; | |
246 | } | |
247 | ||
34f3a814 LZ |
248 | static int sched_feat_open(struct inode *inode, struct file *filp) |
249 | { | |
250 | return single_open(filp, sched_feat_show, NULL); | |
251 | } | |
252 | ||
828c0950 | 253 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
254 | .open = sched_feat_open, |
255 | .write = sched_feat_write, | |
256 | .read = seq_read, | |
257 | .llseek = seq_lseek, | |
258 | .release = single_release, | |
f00b45c1 PZ |
259 | }; |
260 | ||
261 | static __init int sched_init_debug(void) | |
262 | { | |
f00b45c1 PZ |
263 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
264 | &sched_feat_fops); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | late_initcall(sched_init_debug); | |
f8b6d1cc | 269 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 270 | |
b82d9fdd PZ |
271 | /* |
272 | * Number of tasks to iterate in a single balance run. | |
273 | * Limited because this is done with IRQs disabled. | |
274 | */ | |
275 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
276 | ||
e9e9250b PZ |
277 | /* |
278 | * period over which we average the RT time consumption, measured | |
279 | * in ms. | |
280 | * | |
281 | * default: 1s | |
282 | */ | |
283 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
284 | ||
fa85ae24 | 285 | /* |
9f0c1e56 | 286 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
287 | * default: 1s |
288 | */ | |
9f0c1e56 | 289 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 290 | |
029632fb | 291 | __read_mostly int scheduler_running; |
6892b75e | 292 | |
9f0c1e56 PZ |
293 | /* |
294 | * part of the period that we allow rt tasks to run in us. | |
295 | * default: 0.95s | |
296 | */ | |
297 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 298 | |
fa85ae24 | 299 | |
1da177e4 | 300 | |
0970d299 | 301 | /* |
0122ec5b | 302 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 303 | */ |
70b97a7f | 304 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
305 | __acquires(rq->lock) |
306 | { | |
0970d299 PZ |
307 | struct rq *rq; |
308 | ||
0122ec5b PZ |
309 | lockdep_assert_held(&p->pi_lock); |
310 | ||
3a5c359a | 311 | for (;;) { |
0970d299 | 312 | rq = task_rq(p); |
05fa785c | 313 | raw_spin_lock(&rq->lock); |
65cc8e48 | 314 | if (likely(rq == task_rq(p))) |
3a5c359a | 315 | return rq; |
05fa785c | 316 | raw_spin_unlock(&rq->lock); |
b29739f9 | 317 | } |
b29739f9 IM |
318 | } |
319 | ||
1da177e4 | 320 | /* |
0122ec5b | 321 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 322 | */ |
70b97a7f | 323 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 324 | __acquires(p->pi_lock) |
1da177e4 LT |
325 | __acquires(rq->lock) |
326 | { | |
70b97a7f | 327 | struct rq *rq; |
1da177e4 | 328 | |
3a5c359a | 329 | for (;;) { |
0122ec5b | 330 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 331 | rq = task_rq(p); |
05fa785c | 332 | raw_spin_lock(&rq->lock); |
65cc8e48 | 333 | if (likely(rq == task_rq(p))) |
3a5c359a | 334 | return rq; |
0122ec5b PZ |
335 | raw_spin_unlock(&rq->lock); |
336 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 337 | } |
1da177e4 LT |
338 | } |
339 | ||
a9957449 | 340 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
341 | __releases(rq->lock) |
342 | { | |
05fa785c | 343 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
344 | } |
345 | ||
0122ec5b PZ |
346 | static inline void |
347 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 348 | __releases(rq->lock) |
0122ec5b | 349 | __releases(p->pi_lock) |
1da177e4 | 350 | { |
0122ec5b PZ |
351 | raw_spin_unlock(&rq->lock); |
352 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
353 | } |
354 | ||
1da177e4 | 355 | /* |
cc2a73b5 | 356 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 357 | */ |
a9957449 | 358 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
359 | __acquires(rq->lock) |
360 | { | |
70b97a7f | 361 | struct rq *rq; |
1da177e4 LT |
362 | |
363 | local_irq_disable(); | |
364 | rq = this_rq(); | |
05fa785c | 365 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
366 | |
367 | return rq; | |
368 | } | |
369 | ||
8f4d37ec PZ |
370 | #ifdef CONFIG_SCHED_HRTICK |
371 | /* | |
372 | * Use HR-timers to deliver accurate preemption points. | |
373 | * | |
374 | * Its all a bit involved since we cannot program an hrt while holding the | |
375 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
376 | * reschedule event. | |
377 | * | |
378 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
379 | * rq->lock. | |
380 | */ | |
8f4d37ec | 381 | |
8f4d37ec PZ |
382 | static void hrtick_clear(struct rq *rq) |
383 | { | |
384 | if (hrtimer_active(&rq->hrtick_timer)) | |
385 | hrtimer_cancel(&rq->hrtick_timer); | |
386 | } | |
387 | ||
8f4d37ec PZ |
388 | /* |
389 | * High-resolution timer tick. | |
390 | * Runs from hardirq context with interrupts disabled. | |
391 | */ | |
392 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
393 | { | |
394 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
395 | ||
396 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
397 | ||
05fa785c | 398 | raw_spin_lock(&rq->lock); |
3e51f33f | 399 | update_rq_clock(rq); |
8f4d37ec | 400 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 401 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
402 | |
403 | return HRTIMER_NORESTART; | |
404 | } | |
405 | ||
95e904c7 | 406 | #ifdef CONFIG_SMP |
31656519 PZ |
407 | /* |
408 | * called from hardirq (IPI) context | |
409 | */ | |
410 | static void __hrtick_start(void *arg) | |
b328ca18 | 411 | { |
31656519 | 412 | struct rq *rq = arg; |
b328ca18 | 413 | |
05fa785c | 414 | raw_spin_lock(&rq->lock); |
31656519 PZ |
415 | hrtimer_restart(&rq->hrtick_timer); |
416 | rq->hrtick_csd_pending = 0; | |
05fa785c | 417 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
418 | } |
419 | ||
31656519 PZ |
420 | /* |
421 | * Called to set the hrtick timer state. | |
422 | * | |
423 | * called with rq->lock held and irqs disabled | |
424 | */ | |
029632fb | 425 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 426 | { |
31656519 PZ |
427 | struct hrtimer *timer = &rq->hrtick_timer; |
428 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 429 | |
cc584b21 | 430 | hrtimer_set_expires(timer, time); |
31656519 PZ |
431 | |
432 | if (rq == this_rq()) { | |
433 | hrtimer_restart(timer); | |
434 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 435 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
436 | rq->hrtick_csd_pending = 1; |
437 | } | |
b328ca18 PZ |
438 | } |
439 | ||
440 | static int | |
441 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
442 | { | |
443 | int cpu = (int)(long)hcpu; | |
444 | ||
445 | switch (action) { | |
446 | case CPU_UP_CANCELED: | |
447 | case CPU_UP_CANCELED_FROZEN: | |
448 | case CPU_DOWN_PREPARE: | |
449 | case CPU_DOWN_PREPARE_FROZEN: | |
450 | case CPU_DEAD: | |
451 | case CPU_DEAD_FROZEN: | |
31656519 | 452 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
453 | return NOTIFY_OK; |
454 | } | |
455 | ||
456 | return NOTIFY_DONE; | |
457 | } | |
458 | ||
fa748203 | 459 | static __init void init_hrtick(void) |
b328ca18 PZ |
460 | { |
461 | hotcpu_notifier(hotplug_hrtick, 0); | |
462 | } | |
31656519 PZ |
463 | #else |
464 | /* | |
465 | * Called to set the hrtick timer state. | |
466 | * | |
467 | * called with rq->lock held and irqs disabled | |
468 | */ | |
029632fb | 469 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 470 | { |
7f1e2ca9 | 471 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 472 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 473 | } |
b328ca18 | 474 | |
006c75f1 | 475 | static inline void init_hrtick(void) |
8f4d37ec | 476 | { |
8f4d37ec | 477 | } |
31656519 | 478 | #endif /* CONFIG_SMP */ |
8f4d37ec | 479 | |
31656519 | 480 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 481 | { |
31656519 PZ |
482 | #ifdef CONFIG_SMP |
483 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 484 | |
31656519 PZ |
485 | rq->hrtick_csd.flags = 0; |
486 | rq->hrtick_csd.func = __hrtick_start; | |
487 | rq->hrtick_csd.info = rq; | |
488 | #endif | |
8f4d37ec | 489 | |
31656519 PZ |
490 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
491 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 492 | } |
006c75f1 | 493 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
494 | static inline void hrtick_clear(struct rq *rq) |
495 | { | |
496 | } | |
497 | ||
8f4d37ec PZ |
498 | static inline void init_rq_hrtick(struct rq *rq) |
499 | { | |
500 | } | |
501 | ||
b328ca18 PZ |
502 | static inline void init_hrtick(void) |
503 | { | |
504 | } | |
006c75f1 | 505 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 506 | |
c24d20db IM |
507 | /* |
508 | * resched_task - mark a task 'to be rescheduled now'. | |
509 | * | |
510 | * On UP this means the setting of the need_resched flag, on SMP it | |
511 | * might also involve a cross-CPU call to trigger the scheduler on | |
512 | * the target CPU. | |
513 | */ | |
514 | #ifdef CONFIG_SMP | |
515 | ||
516 | #ifndef tsk_is_polling | |
16a80163 | 517 | #define tsk_is_polling(t) 0 |
c24d20db IM |
518 | #endif |
519 | ||
029632fb | 520 | void resched_task(struct task_struct *p) |
c24d20db IM |
521 | { |
522 | int cpu; | |
523 | ||
05fa785c | 524 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 525 | |
5ed0cec0 | 526 | if (test_tsk_need_resched(p)) |
c24d20db IM |
527 | return; |
528 | ||
5ed0cec0 | 529 | set_tsk_need_resched(p); |
c24d20db IM |
530 | |
531 | cpu = task_cpu(p); | |
532 | if (cpu == smp_processor_id()) | |
533 | return; | |
534 | ||
535 | /* NEED_RESCHED must be visible before we test polling */ | |
536 | smp_mb(); | |
537 | if (!tsk_is_polling(p)) | |
538 | smp_send_reschedule(cpu); | |
539 | } | |
540 | ||
029632fb | 541 | void resched_cpu(int cpu) |
c24d20db IM |
542 | { |
543 | struct rq *rq = cpu_rq(cpu); | |
544 | unsigned long flags; | |
545 | ||
05fa785c | 546 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
547 | return; |
548 | resched_task(cpu_curr(cpu)); | |
05fa785c | 549 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 550 | } |
06d8308c TG |
551 | |
552 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
553 | /* |
554 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
555 | * from an idle cpu. This is good for power-savings. | |
556 | * | |
557 | * We don't do similar optimization for completely idle system, as | |
558 | * selecting an idle cpu will add more delays to the timers than intended | |
559 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
560 | */ | |
561 | int get_nohz_timer_target(void) | |
562 | { | |
563 | int cpu = smp_processor_id(); | |
564 | int i; | |
565 | struct sched_domain *sd; | |
566 | ||
057f3fad | 567 | rcu_read_lock(); |
83cd4fe2 | 568 | for_each_domain(cpu, sd) { |
057f3fad PZ |
569 | for_each_cpu(i, sched_domain_span(sd)) { |
570 | if (!idle_cpu(i)) { | |
571 | cpu = i; | |
572 | goto unlock; | |
573 | } | |
574 | } | |
83cd4fe2 | 575 | } |
057f3fad PZ |
576 | unlock: |
577 | rcu_read_unlock(); | |
83cd4fe2 VP |
578 | return cpu; |
579 | } | |
06d8308c TG |
580 | /* |
581 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
582 | * idle CPU then this timer might expire before the next timer event | |
583 | * which is scheduled to wake up that CPU. In case of a completely | |
584 | * idle system the next event might even be infinite time into the | |
585 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
586 | * leaves the inner idle loop so the newly added timer is taken into | |
587 | * account when the CPU goes back to idle and evaluates the timer | |
588 | * wheel for the next timer event. | |
589 | */ | |
590 | void wake_up_idle_cpu(int cpu) | |
591 | { | |
592 | struct rq *rq = cpu_rq(cpu); | |
593 | ||
594 | if (cpu == smp_processor_id()) | |
595 | return; | |
596 | ||
597 | /* | |
598 | * This is safe, as this function is called with the timer | |
599 | * wheel base lock of (cpu) held. When the CPU is on the way | |
600 | * to idle and has not yet set rq->curr to idle then it will | |
601 | * be serialized on the timer wheel base lock and take the new | |
602 | * timer into account automatically. | |
603 | */ | |
604 | if (rq->curr != rq->idle) | |
605 | return; | |
45bf76df | 606 | |
45bf76df | 607 | /* |
06d8308c TG |
608 | * We can set TIF_RESCHED on the idle task of the other CPU |
609 | * lockless. The worst case is that the other CPU runs the | |
610 | * idle task through an additional NOOP schedule() | |
45bf76df | 611 | */ |
5ed0cec0 | 612 | set_tsk_need_resched(rq->idle); |
45bf76df | 613 | |
06d8308c TG |
614 | /* NEED_RESCHED must be visible before we test polling */ |
615 | smp_mb(); | |
616 | if (!tsk_is_polling(rq->idle)) | |
617 | smp_send_reschedule(cpu); | |
45bf76df IM |
618 | } |
619 | ||
ca38062e | 620 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 621 | { |
1c792db7 SS |
622 | int cpu = smp_processor_id(); |
623 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
624 | } |
625 | ||
ca38062e | 626 | #else /* CONFIG_NO_HZ */ |
45bf76df | 627 | |
ca38062e | 628 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 629 | { |
ca38062e | 630 | return false; |
2069dd75 PZ |
631 | } |
632 | ||
6d6bc0ad | 633 | #endif /* CONFIG_NO_HZ */ |
d842de87 | 634 | |
029632fb | 635 | void sched_avg_update(struct rq *rq) |
18d95a28 | 636 | { |
e9e9250b PZ |
637 | s64 period = sched_avg_period(); |
638 | ||
639 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
640 | /* |
641 | * Inline assembly required to prevent the compiler | |
642 | * optimising this loop into a divmod call. | |
643 | * See __iter_div_u64_rem() for another example of this. | |
644 | */ | |
645 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
646 | rq->age_stamp += period; |
647 | rq->rt_avg /= 2; | |
648 | } | |
18d95a28 PZ |
649 | } |
650 | ||
6d6bc0ad | 651 | #else /* !CONFIG_SMP */ |
029632fb | 652 | void resched_task(struct task_struct *p) |
18d95a28 | 653 | { |
05fa785c | 654 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 655 | set_tsk_need_resched(p); |
18d95a28 | 656 | } |
6d6bc0ad | 657 | #endif /* CONFIG_SMP */ |
18d95a28 | 658 | |
a790de99 PT |
659 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
660 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 661 | /* |
8277434e PT |
662 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
663 | * node and @up when leaving it for the final time. | |
664 | * | |
665 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 666 | */ |
029632fb | 667 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 668 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
669 | { |
670 | struct task_group *parent, *child; | |
eb755805 | 671 | int ret; |
c09595f6 | 672 | |
8277434e PT |
673 | parent = from; |
674 | ||
c09595f6 | 675 | down: |
eb755805 PZ |
676 | ret = (*down)(parent, data); |
677 | if (ret) | |
8277434e | 678 | goto out; |
c09595f6 PZ |
679 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
680 | parent = child; | |
681 | goto down; | |
682 | ||
683 | up: | |
684 | continue; | |
685 | } | |
eb755805 | 686 | ret = (*up)(parent, data); |
8277434e PT |
687 | if (ret || parent == from) |
688 | goto out; | |
c09595f6 PZ |
689 | |
690 | child = parent; | |
691 | parent = parent->parent; | |
692 | if (parent) | |
693 | goto up; | |
8277434e | 694 | out: |
eb755805 | 695 | return ret; |
c09595f6 PZ |
696 | } |
697 | ||
029632fb | 698 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 699 | { |
e2b245f8 | 700 | return 0; |
eb755805 | 701 | } |
18d95a28 PZ |
702 | #endif |
703 | ||
45bf76df IM |
704 | static void set_load_weight(struct task_struct *p) |
705 | { | |
f05998d4 NR |
706 | int prio = p->static_prio - MAX_RT_PRIO; |
707 | struct load_weight *load = &p->se.load; | |
708 | ||
dd41f596 IM |
709 | /* |
710 | * SCHED_IDLE tasks get minimal weight: | |
711 | */ | |
712 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 713 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 714 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
715 | return; |
716 | } | |
71f8bd46 | 717 | |
c8b28116 | 718 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 719 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
720 | } |
721 | ||
371fd7e7 | 722 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 723 | { |
a64692a3 | 724 | update_rq_clock(rq); |
dd41f596 | 725 | sched_info_queued(p); |
371fd7e7 | 726 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
727 | } |
728 | ||
371fd7e7 | 729 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 730 | { |
a64692a3 | 731 | update_rq_clock(rq); |
46ac22ba | 732 | sched_info_dequeued(p); |
371fd7e7 | 733 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
734 | } |
735 | ||
029632fb | 736 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
737 | { |
738 | if (task_contributes_to_load(p)) | |
739 | rq->nr_uninterruptible--; | |
740 | ||
371fd7e7 | 741 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
742 | } |
743 | ||
029632fb | 744 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
745 | { |
746 | if (task_contributes_to_load(p)) | |
747 | rq->nr_uninterruptible++; | |
748 | ||
371fd7e7 | 749 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
750 | } |
751 | ||
fe44d621 | 752 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 753 | { |
095c0aa8 GC |
754 | /* |
755 | * In theory, the compile should just see 0 here, and optimize out the call | |
756 | * to sched_rt_avg_update. But I don't trust it... | |
757 | */ | |
758 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
759 | s64 steal = 0, irq_delta = 0; | |
760 | #endif | |
761 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 762 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
763 | |
764 | /* | |
765 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
766 | * this case when a previous update_rq_clock() happened inside a | |
767 | * {soft,}irq region. | |
768 | * | |
769 | * When this happens, we stop ->clock_task and only update the | |
770 | * prev_irq_time stamp to account for the part that fit, so that a next | |
771 | * update will consume the rest. This ensures ->clock_task is | |
772 | * monotonic. | |
773 | * | |
774 | * It does however cause some slight miss-attribution of {soft,}irq | |
775 | * time, a more accurate solution would be to update the irq_time using | |
776 | * the current rq->clock timestamp, except that would require using | |
777 | * atomic ops. | |
778 | */ | |
779 | if (irq_delta > delta) | |
780 | irq_delta = delta; | |
781 | ||
782 | rq->prev_irq_time += irq_delta; | |
783 | delta -= irq_delta; | |
095c0aa8 GC |
784 | #endif |
785 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 786 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
787 | u64 st; |
788 | ||
789 | steal = paravirt_steal_clock(cpu_of(rq)); | |
790 | steal -= rq->prev_steal_time_rq; | |
791 | ||
792 | if (unlikely(steal > delta)) | |
793 | steal = delta; | |
794 | ||
795 | st = steal_ticks(steal); | |
796 | steal = st * TICK_NSEC; | |
797 | ||
798 | rq->prev_steal_time_rq += steal; | |
799 | ||
800 | delta -= steal; | |
801 | } | |
802 | #endif | |
803 | ||
fe44d621 PZ |
804 | rq->clock_task += delta; |
805 | ||
095c0aa8 GC |
806 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
807 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
808 | sched_rt_avg_update(rq, irq_delta + steal); | |
809 | #endif | |
aa483808 VP |
810 | } |
811 | ||
34f971f6 PZ |
812 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
813 | { | |
814 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
815 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
816 | ||
817 | if (stop) { | |
818 | /* | |
819 | * Make it appear like a SCHED_FIFO task, its something | |
820 | * userspace knows about and won't get confused about. | |
821 | * | |
822 | * Also, it will make PI more or less work without too | |
823 | * much confusion -- but then, stop work should not | |
824 | * rely on PI working anyway. | |
825 | */ | |
826 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
827 | ||
828 | stop->sched_class = &stop_sched_class; | |
829 | } | |
830 | ||
831 | cpu_rq(cpu)->stop = stop; | |
832 | ||
833 | if (old_stop) { | |
834 | /* | |
835 | * Reset it back to a normal scheduling class so that | |
836 | * it can die in pieces. | |
837 | */ | |
838 | old_stop->sched_class = &rt_sched_class; | |
839 | } | |
840 | } | |
841 | ||
14531189 | 842 | /* |
dd41f596 | 843 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 844 | */ |
14531189 IM |
845 | static inline int __normal_prio(struct task_struct *p) |
846 | { | |
dd41f596 | 847 | return p->static_prio; |
14531189 IM |
848 | } |
849 | ||
b29739f9 IM |
850 | /* |
851 | * Calculate the expected normal priority: i.e. priority | |
852 | * without taking RT-inheritance into account. Might be | |
853 | * boosted by interactivity modifiers. Changes upon fork, | |
854 | * setprio syscalls, and whenever the interactivity | |
855 | * estimator recalculates. | |
856 | */ | |
36c8b586 | 857 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
858 | { |
859 | int prio; | |
860 | ||
e05606d3 | 861 | if (task_has_rt_policy(p)) |
b29739f9 IM |
862 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
863 | else | |
864 | prio = __normal_prio(p); | |
865 | return prio; | |
866 | } | |
867 | ||
868 | /* | |
869 | * Calculate the current priority, i.e. the priority | |
870 | * taken into account by the scheduler. This value might | |
871 | * be boosted by RT tasks, or might be boosted by | |
872 | * interactivity modifiers. Will be RT if the task got | |
873 | * RT-boosted. If not then it returns p->normal_prio. | |
874 | */ | |
36c8b586 | 875 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
876 | { |
877 | p->normal_prio = normal_prio(p); | |
878 | /* | |
879 | * If we are RT tasks or we were boosted to RT priority, | |
880 | * keep the priority unchanged. Otherwise, update priority | |
881 | * to the normal priority: | |
882 | */ | |
883 | if (!rt_prio(p->prio)) | |
884 | return p->normal_prio; | |
885 | return p->prio; | |
886 | } | |
887 | ||
1da177e4 LT |
888 | /** |
889 | * task_curr - is this task currently executing on a CPU? | |
890 | * @p: the task in question. | |
891 | */ | |
36c8b586 | 892 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
893 | { |
894 | return cpu_curr(task_cpu(p)) == p; | |
895 | } | |
896 | ||
cb469845 SR |
897 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
898 | const struct sched_class *prev_class, | |
da7a735e | 899 | int oldprio) |
cb469845 SR |
900 | { |
901 | if (prev_class != p->sched_class) { | |
902 | if (prev_class->switched_from) | |
da7a735e PZ |
903 | prev_class->switched_from(rq, p); |
904 | p->sched_class->switched_to(rq, p); | |
905 | } else if (oldprio != p->prio) | |
906 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
907 | } |
908 | ||
029632fb | 909 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
910 | { |
911 | const struct sched_class *class; | |
912 | ||
913 | if (p->sched_class == rq->curr->sched_class) { | |
914 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
915 | } else { | |
916 | for_each_class(class) { | |
917 | if (class == rq->curr->sched_class) | |
918 | break; | |
919 | if (class == p->sched_class) { | |
920 | resched_task(rq->curr); | |
921 | break; | |
922 | } | |
923 | } | |
924 | } | |
925 | ||
926 | /* | |
927 | * A queue event has occurred, and we're going to schedule. In | |
928 | * this case, we can save a useless back to back clock update. | |
929 | */ | |
fd2f4419 | 930 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
931 | rq->skip_clock_update = 1; |
932 | } | |
933 | ||
582b336e MT |
934 | static ATOMIC_NOTIFIER_HEAD(task_migration_notifier); |
935 | ||
936 | void register_task_migration_notifier(struct notifier_block *n) | |
937 | { | |
938 | atomic_notifier_chain_register(&task_migration_notifier, n); | |
939 | } | |
940 | ||
1da177e4 | 941 | #ifdef CONFIG_SMP |
dd41f596 | 942 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 943 | { |
e2912009 PZ |
944 | #ifdef CONFIG_SCHED_DEBUG |
945 | /* | |
946 | * We should never call set_task_cpu() on a blocked task, | |
947 | * ttwu() will sort out the placement. | |
948 | */ | |
077614ee PZ |
949 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
950 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
951 | |
952 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
953 | /* |
954 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
955 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
956 | * | |
957 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 958 | * see task_group(). |
6c6c54e1 PZ |
959 | * |
960 | * Furthermore, all task_rq users should acquire both locks, see | |
961 | * task_rq_lock(). | |
962 | */ | |
0122ec5b PZ |
963 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
964 | lockdep_is_held(&task_rq(p)->lock))); | |
965 | #endif | |
e2912009 PZ |
966 | #endif |
967 | ||
de1d7286 | 968 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 969 | |
0c69774e | 970 | if (task_cpu(p) != new_cpu) { |
582b336e MT |
971 | struct task_migration_notifier tmn; |
972 | ||
0a74bef8 PT |
973 | if (p->sched_class->migrate_task_rq) |
974 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 975 | p->se.nr_migrations++; |
a8b0ca17 | 976 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
582b336e MT |
977 | |
978 | tmn.task = p; | |
979 | tmn.from_cpu = task_cpu(p); | |
980 | tmn.to_cpu = new_cpu; | |
981 | ||
982 | atomic_notifier_call_chain(&task_migration_notifier, 0, &tmn); | |
0c69774e | 983 | } |
dd41f596 IM |
984 | |
985 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
986 | } |
987 | ||
969c7921 | 988 | struct migration_arg { |
36c8b586 | 989 | struct task_struct *task; |
1da177e4 | 990 | int dest_cpu; |
70b97a7f | 991 | }; |
1da177e4 | 992 | |
969c7921 TH |
993 | static int migration_cpu_stop(void *data); |
994 | ||
1da177e4 LT |
995 | /* |
996 | * wait_task_inactive - wait for a thread to unschedule. | |
997 | * | |
85ba2d86 RM |
998 | * If @match_state is nonzero, it's the @p->state value just checked and |
999 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1000 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1001 | * we return a positive number (its total switch count). If a second call | |
1002 | * a short while later returns the same number, the caller can be sure that | |
1003 | * @p has remained unscheduled the whole time. | |
1004 | * | |
1da177e4 LT |
1005 | * The caller must ensure that the task *will* unschedule sometime soon, |
1006 | * else this function might spin for a *long* time. This function can't | |
1007 | * be called with interrupts off, or it may introduce deadlock with | |
1008 | * smp_call_function() if an IPI is sent by the same process we are | |
1009 | * waiting to become inactive. | |
1010 | */ | |
85ba2d86 | 1011 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1012 | { |
1013 | unsigned long flags; | |
dd41f596 | 1014 | int running, on_rq; |
85ba2d86 | 1015 | unsigned long ncsw; |
70b97a7f | 1016 | struct rq *rq; |
1da177e4 | 1017 | |
3a5c359a AK |
1018 | for (;;) { |
1019 | /* | |
1020 | * We do the initial early heuristics without holding | |
1021 | * any task-queue locks at all. We'll only try to get | |
1022 | * the runqueue lock when things look like they will | |
1023 | * work out! | |
1024 | */ | |
1025 | rq = task_rq(p); | |
fa490cfd | 1026 | |
3a5c359a AK |
1027 | /* |
1028 | * If the task is actively running on another CPU | |
1029 | * still, just relax and busy-wait without holding | |
1030 | * any locks. | |
1031 | * | |
1032 | * NOTE! Since we don't hold any locks, it's not | |
1033 | * even sure that "rq" stays as the right runqueue! | |
1034 | * But we don't care, since "task_running()" will | |
1035 | * return false if the runqueue has changed and p | |
1036 | * is actually now running somewhere else! | |
1037 | */ | |
85ba2d86 RM |
1038 | while (task_running(rq, p)) { |
1039 | if (match_state && unlikely(p->state != match_state)) | |
1040 | return 0; | |
3a5c359a | 1041 | cpu_relax(); |
85ba2d86 | 1042 | } |
fa490cfd | 1043 | |
3a5c359a AK |
1044 | /* |
1045 | * Ok, time to look more closely! We need the rq | |
1046 | * lock now, to be *sure*. If we're wrong, we'll | |
1047 | * just go back and repeat. | |
1048 | */ | |
1049 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1050 | trace_sched_wait_task(p); |
3a5c359a | 1051 | running = task_running(rq, p); |
fd2f4419 | 1052 | on_rq = p->on_rq; |
85ba2d86 | 1053 | ncsw = 0; |
f31e11d8 | 1054 | if (!match_state || p->state == match_state) |
93dcf55f | 1055 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1056 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1057 | |
85ba2d86 RM |
1058 | /* |
1059 | * If it changed from the expected state, bail out now. | |
1060 | */ | |
1061 | if (unlikely(!ncsw)) | |
1062 | break; | |
1063 | ||
3a5c359a AK |
1064 | /* |
1065 | * Was it really running after all now that we | |
1066 | * checked with the proper locks actually held? | |
1067 | * | |
1068 | * Oops. Go back and try again.. | |
1069 | */ | |
1070 | if (unlikely(running)) { | |
1071 | cpu_relax(); | |
1072 | continue; | |
1073 | } | |
fa490cfd | 1074 | |
3a5c359a AK |
1075 | /* |
1076 | * It's not enough that it's not actively running, | |
1077 | * it must be off the runqueue _entirely_, and not | |
1078 | * preempted! | |
1079 | * | |
80dd99b3 | 1080 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1081 | * running right now), it's preempted, and we should |
1082 | * yield - it could be a while. | |
1083 | */ | |
1084 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1085 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1086 | ||
1087 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1088 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1089 | continue; |
1090 | } | |
fa490cfd | 1091 | |
3a5c359a AK |
1092 | /* |
1093 | * Ahh, all good. It wasn't running, and it wasn't | |
1094 | * runnable, which means that it will never become | |
1095 | * running in the future either. We're all done! | |
1096 | */ | |
1097 | break; | |
1098 | } | |
85ba2d86 RM |
1099 | |
1100 | return ncsw; | |
1da177e4 LT |
1101 | } |
1102 | ||
1103 | /*** | |
1104 | * kick_process - kick a running thread to enter/exit the kernel | |
1105 | * @p: the to-be-kicked thread | |
1106 | * | |
1107 | * Cause a process which is running on another CPU to enter | |
1108 | * kernel-mode, without any delay. (to get signals handled.) | |
1109 | * | |
25985edc | 1110 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1111 | * because all it wants to ensure is that the remote task enters |
1112 | * the kernel. If the IPI races and the task has been migrated | |
1113 | * to another CPU then no harm is done and the purpose has been | |
1114 | * achieved as well. | |
1115 | */ | |
36c8b586 | 1116 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1117 | { |
1118 | int cpu; | |
1119 | ||
1120 | preempt_disable(); | |
1121 | cpu = task_cpu(p); | |
1122 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1123 | smp_send_reschedule(cpu); | |
1124 | preempt_enable(); | |
1125 | } | |
b43e3521 | 1126 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1127 | #endif /* CONFIG_SMP */ |
1da177e4 | 1128 | |
970b13ba | 1129 | #ifdef CONFIG_SMP |
30da688e | 1130 | /* |
013fdb80 | 1131 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1132 | */ |
5da9a0fb PZ |
1133 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1134 | { | |
aa00d89c TC |
1135 | int nid = cpu_to_node(cpu); |
1136 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1137 | enum { cpuset, possible, fail } state = cpuset; |
1138 | int dest_cpu; | |
5da9a0fb | 1139 | |
aa00d89c TC |
1140 | /* |
1141 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1142 | * will return -1. There is no cpu on the node, and we should | |
1143 | * select the cpu on the other node. | |
1144 | */ | |
1145 | if (nid != -1) { | |
1146 | nodemask = cpumask_of_node(nid); | |
1147 | ||
1148 | /* Look for allowed, online CPU in same node. */ | |
1149 | for_each_cpu(dest_cpu, nodemask) { | |
1150 | if (!cpu_online(dest_cpu)) | |
1151 | continue; | |
1152 | if (!cpu_active(dest_cpu)) | |
1153 | continue; | |
1154 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1155 | return dest_cpu; | |
1156 | } | |
2baab4e9 | 1157 | } |
5da9a0fb | 1158 | |
2baab4e9 PZ |
1159 | for (;;) { |
1160 | /* Any allowed, online CPU? */ | |
e3831edd | 1161 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1162 | if (!cpu_online(dest_cpu)) |
1163 | continue; | |
1164 | if (!cpu_active(dest_cpu)) | |
1165 | continue; | |
1166 | goto out; | |
1167 | } | |
5da9a0fb | 1168 | |
2baab4e9 PZ |
1169 | switch (state) { |
1170 | case cpuset: | |
1171 | /* No more Mr. Nice Guy. */ | |
1172 | cpuset_cpus_allowed_fallback(p); | |
1173 | state = possible; | |
1174 | break; | |
1175 | ||
1176 | case possible: | |
1177 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1178 | state = fail; | |
1179 | break; | |
1180 | ||
1181 | case fail: | |
1182 | BUG(); | |
1183 | break; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | out: | |
1188 | if (state != cpuset) { | |
1189 | /* | |
1190 | * Don't tell them about moving exiting tasks or | |
1191 | * kernel threads (both mm NULL), since they never | |
1192 | * leave kernel. | |
1193 | */ | |
1194 | if (p->mm && printk_ratelimit()) { | |
1195 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1196 | task_pid_nr(p), p->comm, cpu); | |
1197 | } | |
5da9a0fb PZ |
1198 | } |
1199 | ||
1200 | return dest_cpu; | |
1201 | } | |
1202 | ||
e2912009 | 1203 | /* |
013fdb80 | 1204 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1205 | */ |
970b13ba | 1206 | static inline |
7608dec2 | 1207 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1208 | { |
7608dec2 | 1209 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1210 | |
1211 | /* | |
1212 | * In order not to call set_task_cpu() on a blocking task we need | |
1213 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1214 | * cpu. | |
1215 | * | |
1216 | * Since this is common to all placement strategies, this lives here. | |
1217 | * | |
1218 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1219 | * not worry about this generic constraint ] | |
1220 | */ | |
fa17b507 | 1221 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1222 | !cpu_online(cpu))) |
5da9a0fb | 1223 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1224 | |
1225 | return cpu; | |
970b13ba | 1226 | } |
09a40af5 MG |
1227 | |
1228 | static void update_avg(u64 *avg, u64 sample) | |
1229 | { | |
1230 | s64 diff = sample - *avg; | |
1231 | *avg += diff >> 3; | |
1232 | } | |
970b13ba PZ |
1233 | #endif |
1234 | ||
d7c01d27 | 1235 | static void |
b84cb5df | 1236 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1237 | { |
d7c01d27 | 1238 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1239 | struct rq *rq = this_rq(); |
1240 | ||
d7c01d27 PZ |
1241 | #ifdef CONFIG_SMP |
1242 | int this_cpu = smp_processor_id(); | |
1243 | ||
1244 | if (cpu == this_cpu) { | |
1245 | schedstat_inc(rq, ttwu_local); | |
1246 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1247 | } else { | |
1248 | struct sched_domain *sd; | |
1249 | ||
1250 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1251 | rcu_read_lock(); |
d7c01d27 PZ |
1252 | for_each_domain(this_cpu, sd) { |
1253 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1254 | schedstat_inc(sd, ttwu_wake_remote); | |
1255 | break; | |
1256 | } | |
1257 | } | |
057f3fad | 1258 | rcu_read_unlock(); |
d7c01d27 | 1259 | } |
f339b9dc PZ |
1260 | |
1261 | if (wake_flags & WF_MIGRATED) | |
1262 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1263 | ||
d7c01d27 PZ |
1264 | #endif /* CONFIG_SMP */ |
1265 | ||
1266 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1267 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1268 | |
1269 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1270 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1271 | |
d7c01d27 PZ |
1272 | #endif /* CONFIG_SCHEDSTATS */ |
1273 | } | |
1274 | ||
1275 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1276 | { | |
9ed3811a | 1277 | activate_task(rq, p, en_flags); |
fd2f4419 | 1278 | p->on_rq = 1; |
c2f7115e PZ |
1279 | |
1280 | /* if a worker is waking up, notify workqueue */ | |
1281 | if (p->flags & PF_WQ_WORKER) | |
1282 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1283 | } |
1284 | ||
23f41eeb PZ |
1285 | /* |
1286 | * Mark the task runnable and perform wakeup-preemption. | |
1287 | */ | |
89363381 | 1288 | static void |
23f41eeb | 1289 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1290 | { |
89363381 | 1291 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1292 | check_preempt_curr(rq, p, wake_flags); |
1293 | ||
1294 | p->state = TASK_RUNNING; | |
1295 | #ifdef CONFIG_SMP | |
1296 | if (p->sched_class->task_woken) | |
1297 | p->sched_class->task_woken(rq, p); | |
1298 | ||
e69c6341 | 1299 | if (rq->idle_stamp) { |
9ed3811a TH |
1300 | u64 delta = rq->clock - rq->idle_stamp; |
1301 | u64 max = 2*sysctl_sched_migration_cost; | |
1302 | ||
1303 | if (delta > max) | |
1304 | rq->avg_idle = max; | |
1305 | else | |
1306 | update_avg(&rq->avg_idle, delta); | |
1307 | rq->idle_stamp = 0; | |
1308 | } | |
1309 | #endif | |
1310 | } | |
1311 | ||
c05fbafb PZ |
1312 | static void |
1313 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1314 | { | |
1315 | #ifdef CONFIG_SMP | |
1316 | if (p->sched_contributes_to_load) | |
1317 | rq->nr_uninterruptible--; | |
1318 | #endif | |
1319 | ||
1320 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1321 | ttwu_do_wakeup(rq, p, wake_flags); | |
1322 | } | |
1323 | ||
1324 | /* | |
1325 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1326 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1327 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1328 | * the task is still ->on_rq. | |
1329 | */ | |
1330 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1331 | { | |
1332 | struct rq *rq; | |
1333 | int ret = 0; | |
1334 | ||
1335 | rq = __task_rq_lock(p); | |
1336 | if (p->on_rq) { | |
1337 | ttwu_do_wakeup(rq, p, wake_flags); | |
1338 | ret = 1; | |
1339 | } | |
1340 | __task_rq_unlock(rq); | |
1341 | ||
1342 | return ret; | |
1343 | } | |
1344 | ||
317f3941 | 1345 | #ifdef CONFIG_SMP |
fa14ff4a | 1346 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1347 | { |
1348 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1349 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1350 | struct task_struct *p; | |
317f3941 PZ |
1351 | |
1352 | raw_spin_lock(&rq->lock); | |
1353 | ||
fa14ff4a PZ |
1354 | while (llist) { |
1355 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1356 | llist = llist_next(llist); | |
317f3941 PZ |
1357 | ttwu_do_activate(rq, p, 0); |
1358 | } | |
1359 | ||
1360 | raw_spin_unlock(&rq->lock); | |
1361 | } | |
1362 | ||
1363 | void scheduler_ipi(void) | |
1364 | { | |
ca38062e | 1365 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1366 | return; |
1367 | ||
1368 | /* | |
1369 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1370 | * traditionally all their work was done from the interrupt return | |
1371 | * path. Now that we actually do some work, we need to make sure | |
1372 | * we do call them. | |
1373 | * | |
1374 | * Some archs already do call them, luckily irq_enter/exit nest | |
1375 | * properly. | |
1376 | * | |
1377 | * Arguably we should visit all archs and update all handlers, | |
1378 | * however a fair share of IPIs are still resched only so this would | |
1379 | * somewhat pessimize the simple resched case. | |
1380 | */ | |
1381 | irq_enter(); | |
fa14ff4a | 1382 | sched_ttwu_pending(); |
ca38062e SS |
1383 | |
1384 | /* | |
1385 | * Check if someone kicked us for doing the nohz idle load balance. | |
1386 | */ | |
6eb57e0d SS |
1387 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1388 | this_rq()->idle_balance = 1; | |
ca38062e | 1389 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1390 | } |
c5d753a5 | 1391 | irq_exit(); |
317f3941 PZ |
1392 | } |
1393 | ||
1394 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1395 | { | |
fa14ff4a | 1396 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1397 | smp_send_reschedule(cpu); |
1398 | } | |
d6aa8f85 | 1399 | |
39be3501 | 1400 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1401 | { |
1402 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1403 | } | |
d6aa8f85 | 1404 | #endif /* CONFIG_SMP */ |
317f3941 | 1405 | |
c05fbafb PZ |
1406 | static void ttwu_queue(struct task_struct *p, int cpu) |
1407 | { | |
1408 | struct rq *rq = cpu_rq(cpu); | |
1409 | ||
17d9f311 | 1410 | #if defined(CONFIG_SMP) |
39be3501 | 1411 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1412 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1413 | ttwu_queue_remote(p, cpu); |
1414 | return; | |
1415 | } | |
1416 | #endif | |
1417 | ||
c05fbafb PZ |
1418 | raw_spin_lock(&rq->lock); |
1419 | ttwu_do_activate(rq, p, 0); | |
1420 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1421 | } |
1422 | ||
1423 | /** | |
1da177e4 | 1424 | * try_to_wake_up - wake up a thread |
9ed3811a | 1425 | * @p: the thread to be awakened |
1da177e4 | 1426 | * @state: the mask of task states that can be woken |
9ed3811a | 1427 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1428 | * |
1429 | * Put it on the run-queue if it's not already there. The "current" | |
1430 | * thread is always on the run-queue (except when the actual | |
1431 | * re-schedule is in progress), and as such you're allowed to do | |
1432 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1433 | * runnable without the overhead of this. | |
1434 | * | |
9ed3811a TH |
1435 | * Returns %true if @p was woken up, %false if it was already running |
1436 | * or @state didn't match @p's state. | |
1da177e4 | 1437 | */ |
e4a52bcb PZ |
1438 | static int |
1439 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1440 | { |
1da177e4 | 1441 | unsigned long flags; |
c05fbafb | 1442 | int cpu, success = 0; |
2398f2c6 | 1443 | |
04e2f174 | 1444 | smp_wmb(); |
013fdb80 | 1445 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1446 | if (!(p->state & state)) |
1da177e4 LT |
1447 | goto out; |
1448 | ||
c05fbafb | 1449 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1450 | cpu = task_cpu(p); |
1da177e4 | 1451 | |
c05fbafb PZ |
1452 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1453 | goto stat; | |
1da177e4 | 1454 | |
1da177e4 | 1455 | #ifdef CONFIG_SMP |
e9c84311 | 1456 | /* |
c05fbafb PZ |
1457 | * If the owning (remote) cpu is still in the middle of schedule() with |
1458 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1459 | */ |
f3e94786 | 1460 | while (p->on_cpu) |
e4a52bcb | 1461 | cpu_relax(); |
0970d299 | 1462 | /* |
e4a52bcb | 1463 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1464 | */ |
e4a52bcb | 1465 | smp_rmb(); |
1da177e4 | 1466 | |
a8e4f2ea | 1467 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1468 | p->state = TASK_WAKING; |
e7693a36 | 1469 | |
e4a52bcb | 1470 | if (p->sched_class->task_waking) |
74f8e4b2 | 1471 | p->sched_class->task_waking(p); |
efbbd05a | 1472 | |
7608dec2 | 1473 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1474 | if (task_cpu(p) != cpu) { |
1475 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1476 | set_task_cpu(p, cpu); |
f339b9dc | 1477 | } |
1da177e4 | 1478 | #endif /* CONFIG_SMP */ |
1da177e4 | 1479 | |
c05fbafb PZ |
1480 | ttwu_queue(p, cpu); |
1481 | stat: | |
b84cb5df | 1482 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1483 | out: |
013fdb80 | 1484 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1485 | |
1486 | return success; | |
1487 | } | |
1488 | ||
21aa9af0 TH |
1489 | /** |
1490 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1491 | * @p: the thread to be awakened | |
1492 | * | |
2acca55e | 1493 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1494 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1495 | * the current task. |
21aa9af0 TH |
1496 | */ |
1497 | static void try_to_wake_up_local(struct task_struct *p) | |
1498 | { | |
1499 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1500 | |
1501 | BUG_ON(rq != this_rq()); | |
1502 | BUG_ON(p == current); | |
1503 | lockdep_assert_held(&rq->lock); | |
1504 | ||
2acca55e PZ |
1505 | if (!raw_spin_trylock(&p->pi_lock)) { |
1506 | raw_spin_unlock(&rq->lock); | |
1507 | raw_spin_lock(&p->pi_lock); | |
1508 | raw_spin_lock(&rq->lock); | |
1509 | } | |
1510 | ||
21aa9af0 | 1511 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1512 | goto out; |
21aa9af0 | 1513 | |
fd2f4419 | 1514 | if (!p->on_rq) |
d7c01d27 PZ |
1515 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1516 | ||
23f41eeb | 1517 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1518 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1519 | out: |
1520 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1521 | } |
1522 | ||
50fa610a DH |
1523 | /** |
1524 | * wake_up_process - Wake up a specific process | |
1525 | * @p: The process to be woken up. | |
1526 | * | |
1527 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1528 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1529 | * running. | |
1530 | * | |
1531 | * It may be assumed that this function implies a write memory barrier before | |
1532 | * changing the task state if and only if any tasks are woken up. | |
1533 | */ | |
7ad5b3a5 | 1534 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1535 | { |
9067ac85 ON |
1536 | WARN_ON(task_is_stopped_or_traced(p)); |
1537 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1538 | } |
1da177e4 LT |
1539 | EXPORT_SYMBOL(wake_up_process); |
1540 | ||
7ad5b3a5 | 1541 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1542 | { |
1543 | return try_to_wake_up(p, state, 0); | |
1544 | } | |
1545 | ||
1da177e4 LT |
1546 | /* |
1547 | * Perform scheduler related setup for a newly forked process p. | |
1548 | * p is forked by current. | |
dd41f596 IM |
1549 | * |
1550 | * __sched_fork() is basic setup used by init_idle() too: | |
1551 | */ | |
1552 | static void __sched_fork(struct task_struct *p) | |
1553 | { | |
fd2f4419 PZ |
1554 | p->on_rq = 0; |
1555 | ||
1556 | p->se.on_rq = 0; | |
dd41f596 IM |
1557 | p->se.exec_start = 0; |
1558 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1559 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1560 | p->se.nr_migrations = 0; |
da7a735e | 1561 | p->se.vruntime = 0; |
fd2f4419 | 1562 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 1563 | |
f4e26b12 PT |
1564 | /* |
1565 | * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be | |
1566 | * removed when useful for applications beyond shares distribution (e.g. | |
1567 | * load-balance). | |
1568 | */ | |
1569 | #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED) | |
9d85f21c PT |
1570 | p->se.avg.runnable_avg_period = 0; |
1571 | p->se.avg.runnable_avg_sum = 0; | |
1572 | #endif | |
6cfb0d5d | 1573 | #ifdef CONFIG_SCHEDSTATS |
41acab88 | 1574 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1575 | #endif |
476d139c | 1576 | |
fa717060 | 1577 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1578 | |
e107be36 AK |
1579 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1580 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1581 | #endif | |
cbee9f88 PZ |
1582 | |
1583 | #ifdef CONFIG_NUMA_BALANCING | |
1584 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
1585 | p->mm->numa_next_scan = jiffies; | |
b8593bfd | 1586 | p->mm->numa_next_reset = jiffies; |
cbee9f88 PZ |
1587 | p->mm->numa_scan_seq = 0; |
1588 | } | |
1589 | ||
1590 | p->node_stamp = 0ULL; | |
1591 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
1592 | p->numa_migrate_seq = p->mm ? p->mm->numa_scan_seq - 1 : 0; | |
4b96a29b | 1593 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 PZ |
1594 | p->numa_work.next = &p->numa_work; |
1595 | #endif /* CONFIG_NUMA_BALANCING */ | |
dd41f596 IM |
1596 | } |
1597 | ||
1a687c2e | 1598 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1599 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1600 | void set_numabalancing_state(bool enabled) |
1601 | { | |
1602 | if (enabled) | |
1603 | sched_feat_set("NUMA"); | |
1604 | else | |
1605 | sched_feat_set("NO_NUMA"); | |
1606 | } | |
3105b86a MG |
1607 | #else |
1608 | __read_mostly bool numabalancing_enabled; | |
1609 | ||
1610 | void set_numabalancing_state(bool enabled) | |
1611 | { | |
1612 | numabalancing_enabled = enabled; | |
dd41f596 | 1613 | } |
3105b86a | 1614 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1615 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1616 | |
1617 | /* | |
1618 | * fork()/clone()-time setup: | |
1619 | */ | |
3e51e3ed | 1620 | void sched_fork(struct task_struct *p) |
dd41f596 | 1621 | { |
0122ec5b | 1622 | unsigned long flags; |
dd41f596 IM |
1623 | int cpu = get_cpu(); |
1624 | ||
1625 | __sched_fork(p); | |
06b83b5f | 1626 | /* |
0017d735 | 1627 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1628 | * nobody will actually run it, and a signal or other external |
1629 | * event cannot wake it up and insert it on the runqueue either. | |
1630 | */ | |
0017d735 | 1631 | p->state = TASK_RUNNING; |
dd41f596 | 1632 | |
c350a04e MG |
1633 | /* |
1634 | * Make sure we do not leak PI boosting priority to the child. | |
1635 | */ | |
1636 | p->prio = current->normal_prio; | |
1637 | ||
b9dc29e7 MG |
1638 | /* |
1639 | * Revert to default priority/policy on fork if requested. | |
1640 | */ | |
1641 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1642 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1643 | p->policy = SCHED_NORMAL; |
6c697bdf | 1644 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1645 | p->rt_priority = 0; |
1646 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1647 | p->static_prio = NICE_TO_PRIO(0); | |
1648 | ||
1649 | p->prio = p->normal_prio = __normal_prio(p); | |
1650 | set_load_weight(p); | |
6c697bdf | 1651 | |
b9dc29e7 MG |
1652 | /* |
1653 | * We don't need the reset flag anymore after the fork. It has | |
1654 | * fulfilled its duty: | |
1655 | */ | |
1656 | p->sched_reset_on_fork = 0; | |
1657 | } | |
ca94c442 | 1658 | |
2ddbf952 HS |
1659 | if (!rt_prio(p->prio)) |
1660 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1661 | |
cd29fe6f PZ |
1662 | if (p->sched_class->task_fork) |
1663 | p->sched_class->task_fork(p); | |
1664 | ||
86951599 PZ |
1665 | /* |
1666 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1667 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1668 | * is ran before sched_fork(). | |
1669 | * | |
1670 | * Silence PROVE_RCU. | |
1671 | */ | |
0122ec5b | 1672 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1673 | set_task_cpu(p, cpu); |
0122ec5b | 1674 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1675 | |
52f17b6c | 1676 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1677 | if (likely(sched_info_on())) |
52f17b6c | 1678 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1679 | #endif |
3ca7a440 PZ |
1680 | #if defined(CONFIG_SMP) |
1681 | p->on_cpu = 0; | |
4866cde0 | 1682 | #endif |
bdd4e85d | 1683 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1684 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1685 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1686 | #endif |
806c09a7 | 1687 | #ifdef CONFIG_SMP |
917b627d | 1688 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1689 | #endif |
917b627d | 1690 | |
476d139c | 1691 | put_cpu(); |
1da177e4 LT |
1692 | } |
1693 | ||
1694 | /* | |
1695 | * wake_up_new_task - wake up a newly created task for the first time. | |
1696 | * | |
1697 | * This function will do some initial scheduler statistics housekeeping | |
1698 | * that must be done for every newly created context, then puts the task | |
1699 | * on the runqueue and wakes it. | |
1700 | */ | |
3e51e3ed | 1701 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1702 | { |
1703 | unsigned long flags; | |
dd41f596 | 1704 | struct rq *rq; |
fabf318e | 1705 | |
ab2515c4 | 1706 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1707 | #ifdef CONFIG_SMP |
1708 | /* | |
1709 | * Fork balancing, do it here and not earlier because: | |
1710 | * - cpus_allowed can change in the fork path | |
1711 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1712 | */ |
ab2515c4 | 1713 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1714 | #endif |
1715 | ||
ab2515c4 | 1716 | rq = __task_rq_lock(p); |
cd29fe6f | 1717 | activate_task(rq, p, 0); |
fd2f4419 | 1718 | p->on_rq = 1; |
89363381 | 1719 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1720 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1721 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1722 | if (p->sched_class->task_woken) |
1723 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1724 | #endif |
0122ec5b | 1725 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1726 | } |
1727 | ||
e107be36 AK |
1728 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1729 | ||
1730 | /** | |
80dd99b3 | 1731 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1732 | * @notifier: notifier struct to register |
e107be36 AK |
1733 | */ |
1734 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1735 | { | |
1736 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1737 | } | |
1738 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1739 | ||
1740 | /** | |
1741 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1742 | * @notifier: notifier struct to unregister |
e107be36 AK |
1743 | * |
1744 | * This is safe to call from within a preemption notifier. | |
1745 | */ | |
1746 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1747 | { | |
1748 | hlist_del(¬ifier->link); | |
1749 | } | |
1750 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1751 | ||
1752 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1753 | { | |
1754 | struct preempt_notifier *notifier; | |
1755 | struct hlist_node *node; | |
1756 | ||
1757 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1758 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1759 | } | |
1760 | ||
1761 | static void | |
1762 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1763 | struct task_struct *next) | |
1764 | { | |
1765 | struct preempt_notifier *notifier; | |
1766 | struct hlist_node *node; | |
1767 | ||
1768 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1769 | notifier->ops->sched_out(notifier, next); | |
1770 | } | |
1771 | ||
6d6bc0ad | 1772 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1773 | |
1774 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1775 | { | |
1776 | } | |
1777 | ||
1778 | static void | |
1779 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1780 | struct task_struct *next) | |
1781 | { | |
1782 | } | |
1783 | ||
6d6bc0ad | 1784 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1785 | |
4866cde0 NP |
1786 | /** |
1787 | * prepare_task_switch - prepare to switch tasks | |
1788 | * @rq: the runqueue preparing to switch | |
421cee29 | 1789 | * @prev: the current task that is being switched out |
4866cde0 NP |
1790 | * @next: the task we are going to switch to. |
1791 | * | |
1792 | * This is called with the rq lock held and interrupts off. It must | |
1793 | * be paired with a subsequent finish_task_switch after the context | |
1794 | * switch. | |
1795 | * | |
1796 | * prepare_task_switch sets up locking and calls architecture specific | |
1797 | * hooks. | |
1798 | */ | |
e107be36 AK |
1799 | static inline void |
1800 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1801 | struct task_struct *next) | |
4866cde0 | 1802 | { |
895dd92c | 1803 | trace_sched_switch(prev, next); |
fe4b04fa PZ |
1804 | sched_info_switch(prev, next); |
1805 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1806 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1807 | prepare_lock_switch(rq, next); |
1808 | prepare_arch_switch(next); | |
1809 | } | |
1810 | ||
1da177e4 LT |
1811 | /** |
1812 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1813 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1814 | * @prev: the thread we just switched away from. |
1815 | * | |
4866cde0 NP |
1816 | * finish_task_switch must be called after the context switch, paired |
1817 | * with a prepare_task_switch call before the context switch. | |
1818 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1819 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1820 | * |
1821 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1822 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1823 | * with the lock held can cause deadlocks; see schedule() for |
1824 | * details.) | |
1825 | */ | |
a9957449 | 1826 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1827 | __releases(rq->lock) |
1828 | { | |
1da177e4 | 1829 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1830 | long prev_state; |
1da177e4 LT |
1831 | |
1832 | rq->prev_mm = NULL; | |
1833 | ||
1834 | /* | |
1835 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1836 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1837 | * schedule one last time. The schedule call will never return, and |
1838 | * the scheduled task must drop that reference. | |
c394cc9f | 1839 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1840 | * still held, otherwise prev could be scheduled on another cpu, die |
1841 | * there before we look at prev->state, and then the reference would | |
1842 | * be dropped twice. | |
1843 | * Manfred Spraul <manfred@colorfullife.com> | |
1844 | */ | |
55a101f8 | 1845 | prev_state = prev->state; |
bf9fae9f | 1846 | vtime_task_switch(prev); |
4866cde0 | 1847 | finish_arch_switch(prev); |
a8d757ef | 1848 | perf_event_task_sched_in(prev, current); |
4866cde0 | 1849 | finish_lock_switch(rq, prev); |
01f23e16 | 1850 | finish_arch_post_lock_switch(); |
e8fa1362 | 1851 | |
e107be36 | 1852 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1853 | if (mm) |
1854 | mmdrop(mm); | |
c394cc9f | 1855 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1856 | /* |
1857 | * Remove function-return probe instances associated with this | |
1858 | * task and put them back on the free list. | |
9761eea8 | 1859 | */ |
c6fd91f0 | 1860 | kprobe_flush_task(prev); |
1da177e4 | 1861 | put_task_struct(prev); |
c6fd91f0 | 1862 | } |
1da177e4 LT |
1863 | } |
1864 | ||
3f029d3c GH |
1865 | #ifdef CONFIG_SMP |
1866 | ||
1867 | /* assumes rq->lock is held */ | |
1868 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1869 | { | |
1870 | if (prev->sched_class->pre_schedule) | |
1871 | prev->sched_class->pre_schedule(rq, prev); | |
1872 | } | |
1873 | ||
1874 | /* rq->lock is NOT held, but preemption is disabled */ | |
1875 | static inline void post_schedule(struct rq *rq) | |
1876 | { | |
1877 | if (rq->post_schedule) { | |
1878 | unsigned long flags; | |
1879 | ||
05fa785c | 1880 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1881 | if (rq->curr->sched_class->post_schedule) |
1882 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1883 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1884 | |
1885 | rq->post_schedule = 0; | |
1886 | } | |
1887 | } | |
1888 | ||
1889 | #else | |
da19ab51 | 1890 | |
3f029d3c GH |
1891 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1892 | { | |
1893 | } | |
1894 | ||
1895 | static inline void post_schedule(struct rq *rq) | |
1896 | { | |
1da177e4 LT |
1897 | } |
1898 | ||
3f029d3c GH |
1899 | #endif |
1900 | ||
1da177e4 LT |
1901 | /** |
1902 | * schedule_tail - first thing a freshly forked thread must call. | |
1903 | * @prev: the thread we just switched away from. | |
1904 | */ | |
36c8b586 | 1905 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1906 | __releases(rq->lock) |
1907 | { | |
70b97a7f IM |
1908 | struct rq *rq = this_rq(); |
1909 | ||
4866cde0 | 1910 | finish_task_switch(rq, prev); |
da19ab51 | 1911 | |
3f029d3c GH |
1912 | /* |
1913 | * FIXME: do we need to worry about rq being invalidated by the | |
1914 | * task_switch? | |
1915 | */ | |
1916 | post_schedule(rq); | |
70b97a7f | 1917 | |
4866cde0 NP |
1918 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
1919 | /* In this case, finish_task_switch does not reenable preemption */ | |
1920 | preempt_enable(); | |
1921 | #endif | |
1da177e4 | 1922 | if (current->set_child_tid) |
b488893a | 1923 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1924 | } |
1925 | ||
1926 | /* | |
1927 | * context_switch - switch to the new MM and the new | |
1928 | * thread's register state. | |
1929 | */ | |
dd41f596 | 1930 | static inline void |
70b97a7f | 1931 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1932 | struct task_struct *next) |
1da177e4 | 1933 | { |
dd41f596 | 1934 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1935 | |
e107be36 | 1936 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 1937 | |
dd41f596 IM |
1938 | mm = next->mm; |
1939 | oldmm = prev->active_mm; | |
9226d125 ZA |
1940 | /* |
1941 | * For paravirt, this is coupled with an exit in switch_to to | |
1942 | * combine the page table reload and the switch backend into | |
1943 | * one hypercall. | |
1944 | */ | |
224101ed | 1945 | arch_start_context_switch(prev); |
9226d125 | 1946 | |
31915ab4 | 1947 | if (!mm) { |
1da177e4 LT |
1948 | next->active_mm = oldmm; |
1949 | atomic_inc(&oldmm->mm_count); | |
1950 | enter_lazy_tlb(oldmm, next); | |
1951 | } else | |
1952 | switch_mm(oldmm, mm, next); | |
1953 | ||
31915ab4 | 1954 | if (!prev->mm) { |
1da177e4 | 1955 | prev->active_mm = NULL; |
1da177e4 LT |
1956 | rq->prev_mm = oldmm; |
1957 | } | |
3a5f5e48 IM |
1958 | /* |
1959 | * Since the runqueue lock will be released by the next | |
1960 | * task (which is an invalid locking op but in the case | |
1961 | * of the scheduler it's an obvious special-case), so we | |
1962 | * do an early lockdep release here: | |
1963 | */ | |
1964 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1965 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1966 | #endif |
1da177e4 | 1967 | |
91d1aa43 | 1968 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
1969 | /* Here we just switch the register state and the stack. */ |
1970 | switch_to(prev, next, prev); | |
1971 | ||
dd41f596 IM |
1972 | barrier(); |
1973 | /* | |
1974 | * this_rq must be evaluated again because prev may have moved | |
1975 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1976 | * frame will be invalid. | |
1977 | */ | |
1978 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1979 | } |
1980 | ||
1981 | /* | |
1c3e8264 | 1982 | * nr_running and nr_context_switches: |
1da177e4 LT |
1983 | * |
1984 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 1985 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
1986 | */ |
1987 | unsigned long nr_running(void) | |
1988 | { | |
1989 | unsigned long i, sum = 0; | |
1990 | ||
1991 | for_each_online_cpu(i) | |
1992 | sum += cpu_rq(i)->nr_running; | |
1993 | ||
1994 | return sum; | |
f711f609 | 1995 | } |
1da177e4 | 1996 | |
1da177e4 | 1997 | unsigned long long nr_context_switches(void) |
46cb4b7c | 1998 | { |
cc94abfc SR |
1999 | int i; |
2000 | unsigned long long sum = 0; | |
46cb4b7c | 2001 | |
0a945022 | 2002 | for_each_possible_cpu(i) |
1da177e4 | 2003 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2004 | |
1da177e4 LT |
2005 | return sum; |
2006 | } | |
483b4ee6 | 2007 | |
1da177e4 LT |
2008 | unsigned long nr_iowait(void) |
2009 | { | |
2010 | unsigned long i, sum = 0; | |
483b4ee6 | 2011 | |
0a945022 | 2012 | for_each_possible_cpu(i) |
1da177e4 | 2013 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2014 | |
1da177e4 LT |
2015 | return sum; |
2016 | } | |
483b4ee6 | 2017 | |
8c215bd3 | 2018 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2019 | { |
8c215bd3 | 2020 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2021 | return atomic_read(&this->nr_iowait); |
2022 | } | |
46cb4b7c | 2023 | |
69d25870 AV |
2024 | unsigned long this_cpu_load(void) |
2025 | { | |
2026 | struct rq *this = this_rq(); | |
2027 | return this->cpu_load[0]; | |
2028 | } | |
e790fb0b | 2029 | |
46cb4b7c | 2030 | |
5167e8d5 PZ |
2031 | /* |
2032 | * Global load-average calculations | |
2033 | * | |
2034 | * We take a distributed and async approach to calculating the global load-avg | |
2035 | * in order to minimize overhead. | |
2036 | * | |
2037 | * The global load average is an exponentially decaying average of nr_running + | |
2038 | * nr_uninterruptible. | |
2039 | * | |
2040 | * Once every LOAD_FREQ: | |
2041 | * | |
2042 | * nr_active = 0; | |
2043 | * for_each_possible_cpu(cpu) | |
2044 | * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; | |
2045 | * | |
2046 | * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) | |
2047 | * | |
2048 | * Due to a number of reasons the above turns in the mess below: | |
2049 | * | |
2050 | * - for_each_possible_cpu() is prohibitively expensive on machines with | |
2051 | * serious number of cpus, therefore we need to take a distributed approach | |
2052 | * to calculating nr_active. | |
2053 | * | |
2054 | * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 | |
2055 | * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } | |
2056 | * | |
2057 | * So assuming nr_active := 0 when we start out -- true per definition, we | |
2058 | * can simply take per-cpu deltas and fold those into a global accumulate | |
2059 | * to obtain the same result. See calc_load_fold_active(). | |
2060 | * | |
2061 | * Furthermore, in order to avoid synchronizing all per-cpu delta folding | |
2062 | * across the machine, we assume 10 ticks is sufficient time for every | |
2063 | * cpu to have completed this task. | |
2064 | * | |
2065 | * This places an upper-bound on the IRQ-off latency of the machine. Then | |
2066 | * again, being late doesn't loose the delta, just wrecks the sample. | |
2067 | * | |
2068 | * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because | |
2069 | * this would add another cross-cpu cacheline miss and atomic operation | |
2070 | * to the wakeup path. Instead we increment on whatever cpu the task ran | |
2071 | * when it went into uninterruptible state and decrement on whatever cpu | |
2072 | * did the wakeup. This means that only the sum of nr_uninterruptible over | |
2073 | * all cpus yields the correct result. | |
2074 | * | |
2075 | * This covers the NO_HZ=n code, for extra head-aches, see the comment below. | |
2076 | */ | |
2077 | ||
dce48a84 TG |
2078 | /* Variables and functions for calc_load */ |
2079 | static atomic_long_t calc_load_tasks; | |
2080 | static unsigned long calc_load_update; | |
2081 | unsigned long avenrun[3]; | |
5167e8d5 PZ |
2082 | EXPORT_SYMBOL(avenrun); /* should be removed */ |
2083 | ||
2084 | /** | |
2085 | * get_avenrun - get the load average array | |
2086 | * @loads: pointer to dest load array | |
2087 | * @offset: offset to add | |
2088 | * @shift: shift count to shift the result left | |
2089 | * | |
2090 | * These values are estimates at best, so no need for locking. | |
2091 | */ | |
2092 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2093 | { | |
2094 | loads[0] = (avenrun[0] + offset) << shift; | |
2095 | loads[1] = (avenrun[1] + offset) << shift; | |
2096 | loads[2] = (avenrun[2] + offset) << shift; | |
2097 | } | |
46cb4b7c | 2098 | |
74f5187a PZ |
2099 | static long calc_load_fold_active(struct rq *this_rq) |
2100 | { | |
2101 | long nr_active, delta = 0; | |
2102 | ||
2103 | nr_active = this_rq->nr_running; | |
2104 | nr_active += (long) this_rq->nr_uninterruptible; | |
2105 | ||
2106 | if (nr_active != this_rq->calc_load_active) { | |
2107 | delta = nr_active - this_rq->calc_load_active; | |
2108 | this_rq->calc_load_active = nr_active; | |
2109 | } | |
2110 | ||
2111 | return delta; | |
2112 | } | |
2113 | ||
5167e8d5 PZ |
2114 | /* |
2115 | * a1 = a0 * e + a * (1 - e) | |
2116 | */ | |
0f004f5a PZ |
2117 | static unsigned long |
2118 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2119 | { | |
2120 | load *= exp; | |
2121 | load += active * (FIXED_1 - exp); | |
2122 | load += 1UL << (FSHIFT - 1); | |
2123 | return load >> FSHIFT; | |
2124 | } | |
2125 | ||
74f5187a PZ |
2126 | #ifdef CONFIG_NO_HZ |
2127 | /* | |
5167e8d5 PZ |
2128 | * Handle NO_HZ for the global load-average. |
2129 | * | |
2130 | * Since the above described distributed algorithm to compute the global | |
2131 | * load-average relies on per-cpu sampling from the tick, it is affected by | |
2132 | * NO_HZ. | |
2133 | * | |
2134 | * The basic idea is to fold the nr_active delta into a global idle-delta upon | |
2135 | * entering NO_HZ state such that we can include this as an 'extra' cpu delta | |
2136 | * when we read the global state. | |
2137 | * | |
2138 | * Obviously reality has to ruin such a delightfully simple scheme: | |
2139 | * | |
2140 | * - When we go NO_HZ idle during the window, we can negate our sample | |
2141 | * contribution, causing under-accounting. | |
2142 | * | |
2143 | * We avoid this by keeping two idle-delta counters and flipping them | |
2144 | * when the window starts, thus separating old and new NO_HZ load. | |
2145 | * | |
2146 | * The only trick is the slight shift in index flip for read vs write. | |
2147 | * | |
2148 | * 0s 5s 10s 15s | |
2149 | * +10 +10 +10 +10 | |
2150 | * |-|-----------|-|-----------|-|-----------|-| | |
2151 | * r:0 0 1 1 0 0 1 1 0 | |
2152 | * w:0 1 1 0 0 1 1 0 0 | |
2153 | * | |
2154 | * This ensures we'll fold the old idle contribution in this window while | |
2155 | * accumlating the new one. | |
2156 | * | |
2157 | * - When we wake up from NO_HZ idle during the window, we push up our | |
2158 | * contribution, since we effectively move our sample point to a known | |
2159 | * busy state. | |
2160 | * | |
2161 | * This is solved by pushing the window forward, and thus skipping the | |
2162 | * sample, for this cpu (effectively using the idle-delta for this cpu which | |
2163 | * was in effect at the time the window opened). This also solves the issue | |
2164 | * of having to deal with a cpu having been in NOHZ idle for multiple | |
2165 | * LOAD_FREQ intervals. | |
74f5187a PZ |
2166 | * |
2167 | * When making the ILB scale, we should try to pull this in as well. | |
2168 | */ | |
5167e8d5 PZ |
2169 | static atomic_long_t calc_load_idle[2]; |
2170 | static int calc_load_idx; | |
74f5187a | 2171 | |
5167e8d5 | 2172 | static inline int calc_load_write_idx(void) |
74f5187a | 2173 | { |
5167e8d5 PZ |
2174 | int idx = calc_load_idx; |
2175 | ||
2176 | /* | |
2177 | * See calc_global_nohz(), if we observe the new index, we also | |
2178 | * need to observe the new update time. | |
2179 | */ | |
2180 | smp_rmb(); | |
2181 | ||
2182 | /* | |
2183 | * If the folding window started, make sure we start writing in the | |
2184 | * next idle-delta. | |
2185 | */ | |
2186 | if (!time_before(jiffies, calc_load_update)) | |
2187 | idx++; | |
2188 | ||
2189 | return idx & 1; | |
2190 | } | |
2191 | ||
2192 | static inline int calc_load_read_idx(void) | |
2193 | { | |
2194 | return calc_load_idx & 1; | |
2195 | } | |
2196 | ||
2197 | void calc_load_enter_idle(void) | |
2198 | { | |
2199 | struct rq *this_rq = this_rq(); | |
74f5187a PZ |
2200 | long delta; |
2201 | ||
5167e8d5 PZ |
2202 | /* |
2203 | * We're going into NOHZ mode, if there's any pending delta, fold it | |
2204 | * into the pending idle delta. | |
2205 | */ | |
74f5187a | 2206 | delta = calc_load_fold_active(this_rq); |
5167e8d5 PZ |
2207 | if (delta) { |
2208 | int idx = calc_load_write_idx(); | |
2209 | atomic_long_add(delta, &calc_load_idle[idx]); | |
2210 | } | |
74f5187a PZ |
2211 | } |
2212 | ||
5167e8d5 | 2213 | void calc_load_exit_idle(void) |
74f5187a | 2214 | { |
5167e8d5 PZ |
2215 | struct rq *this_rq = this_rq(); |
2216 | ||
2217 | /* | |
2218 | * If we're still before the sample window, we're done. | |
2219 | */ | |
2220 | if (time_before(jiffies, this_rq->calc_load_update)) | |
2221 | return; | |
74f5187a PZ |
2222 | |
2223 | /* | |
5167e8d5 PZ |
2224 | * We woke inside or after the sample window, this means we're already |
2225 | * accounted through the nohz accounting, so skip the entire deal and | |
2226 | * sync up for the next window. | |
74f5187a | 2227 | */ |
5167e8d5 PZ |
2228 | this_rq->calc_load_update = calc_load_update; |
2229 | if (time_before(jiffies, this_rq->calc_load_update + 10)) | |
2230 | this_rq->calc_load_update += LOAD_FREQ; | |
2231 | } | |
2232 | ||
2233 | static long calc_load_fold_idle(void) | |
2234 | { | |
2235 | int idx = calc_load_read_idx(); | |
2236 | long delta = 0; | |
2237 | ||
2238 | if (atomic_long_read(&calc_load_idle[idx])) | |
2239 | delta = atomic_long_xchg(&calc_load_idle[idx], 0); | |
74f5187a PZ |
2240 | |
2241 | return delta; | |
2242 | } | |
0f004f5a PZ |
2243 | |
2244 | /** | |
2245 | * fixed_power_int - compute: x^n, in O(log n) time | |
2246 | * | |
2247 | * @x: base of the power | |
2248 | * @frac_bits: fractional bits of @x | |
2249 | * @n: power to raise @x to. | |
2250 | * | |
2251 | * By exploiting the relation between the definition of the natural power | |
2252 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2253 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2254 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2255 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2256 | * of course trivially computable in O(log_2 n), the length of our binary | |
2257 | * vector. | |
2258 | */ | |
2259 | static unsigned long | |
2260 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2261 | { | |
2262 | unsigned long result = 1UL << frac_bits; | |
2263 | ||
2264 | if (n) for (;;) { | |
2265 | if (n & 1) { | |
2266 | result *= x; | |
2267 | result += 1UL << (frac_bits - 1); | |
2268 | result >>= frac_bits; | |
2269 | } | |
2270 | n >>= 1; | |
2271 | if (!n) | |
2272 | break; | |
2273 | x *= x; | |
2274 | x += 1UL << (frac_bits - 1); | |
2275 | x >>= frac_bits; | |
2276 | } | |
2277 | ||
2278 | return result; | |
2279 | } | |
2280 | ||
2281 | /* | |
2282 | * a1 = a0 * e + a * (1 - e) | |
2283 | * | |
2284 | * a2 = a1 * e + a * (1 - e) | |
2285 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2286 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2287 | * | |
2288 | * a3 = a2 * e + a * (1 - e) | |
2289 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2290 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2291 | * | |
2292 | * ... | |
2293 | * | |
2294 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2295 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2296 | * = a0 * e^n + a * (1 - e^n) | |
2297 | * | |
2298 | * [1] application of the geometric series: | |
2299 | * | |
2300 | * n 1 - x^(n+1) | |
2301 | * S_n := \Sum x^i = ------------- | |
2302 | * i=0 1 - x | |
2303 | */ | |
2304 | static unsigned long | |
2305 | calc_load_n(unsigned long load, unsigned long exp, | |
2306 | unsigned long active, unsigned int n) | |
2307 | { | |
2308 | ||
2309 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2310 | } | |
2311 | ||
2312 | /* | |
2313 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2314 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2315 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2316 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2317 | * | |
2318 | * Once we've updated the global active value, we need to apply the exponential | |
2319 | * weights adjusted to the number of cycles missed. | |
2320 | */ | |
c308b56b | 2321 | static void calc_global_nohz(void) |
0f004f5a PZ |
2322 | { |
2323 | long delta, active, n; | |
2324 | ||
5167e8d5 PZ |
2325 | if (!time_before(jiffies, calc_load_update + 10)) { |
2326 | /* | |
2327 | * Catch-up, fold however many we are behind still | |
2328 | */ | |
2329 | delta = jiffies - calc_load_update - 10; | |
2330 | n = 1 + (delta / LOAD_FREQ); | |
0f004f5a | 2331 | |
5167e8d5 PZ |
2332 | active = atomic_long_read(&calc_load_tasks); |
2333 | active = active > 0 ? active * FIXED_1 : 0; | |
0f004f5a | 2334 | |
5167e8d5 PZ |
2335 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
2336 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2337 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
0f004f5a | 2338 | |
5167e8d5 PZ |
2339 | calc_load_update += n * LOAD_FREQ; |
2340 | } | |
74f5187a | 2341 | |
5167e8d5 PZ |
2342 | /* |
2343 | * Flip the idle index... | |
2344 | * | |
2345 | * Make sure we first write the new time then flip the index, so that | |
2346 | * calc_load_write_idx() will see the new time when it reads the new | |
2347 | * index, this avoids a double flip messing things up. | |
2348 | */ | |
2349 | smp_wmb(); | |
2350 | calc_load_idx++; | |
74f5187a | 2351 | } |
5167e8d5 | 2352 | #else /* !CONFIG_NO_HZ */ |
0f004f5a | 2353 | |
5167e8d5 PZ |
2354 | static inline long calc_load_fold_idle(void) { return 0; } |
2355 | static inline void calc_global_nohz(void) { } | |
74f5187a | 2356 | |
5167e8d5 | 2357 | #endif /* CONFIG_NO_HZ */ |
46cb4b7c | 2358 | |
46cb4b7c | 2359 | /* |
dce48a84 TG |
2360 | * calc_load - update the avenrun load estimates 10 ticks after the |
2361 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2362 | */ |
0f004f5a | 2363 | void calc_global_load(unsigned long ticks) |
7835b98b | 2364 | { |
5167e8d5 | 2365 | long active, delta; |
1da177e4 | 2366 | |
0f004f5a | 2367 | if (time_before(jiffies, calc_load_update + 10)) |
dce48a84 | 2368 | return; |
1da177e4 | 2369 | |
5167e8d5 PZ |
2370 | /* |
2371 | * Fold the 'old' idle-delta to include all NO_HZ cpus. | |
2372 | */ | |
2373 | delta = calc_load_fold_idle(); | |
2374 | if (delta) | |
2375 | atomic_long_add(delta, &calc_load_tasks); | |
2376 | ||
dce48a84 TG |
2377 | active = atomic_long_read(&calc_load_tasks); |
2378 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2379 | |
dce48a84 TG |
2380 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2381 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2382 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2383 | |
dce48a84 | 2384 | calc_load_update += LOAD_FREQ; |
c308b56b PZ |
2385 | |
2386 | /* | |
5167e8d5 | 2387 | * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. |
c308b56b PZ |
2388 | */ |
2389 | calc_global_nohz(); | |
dce48a84 | 2390 | } |
1da177e4 | 2391 | |
dce48a84 | 2392 | /* |
74f5187a PZ |
2393 | * Called from update_cpu_load() to periodically update this CPU's |
2394 | * active count. | |
dce48a84 TG |
2395 | */ |
2396 | static void calc_load_account_active(struct rq *this_rq) | |
2397 | { | |
74f5187a | 2398 | long delta; |
08c183f3 | 2399 | |
74f5187a PZ |
2400 | if (time_before(jiffies, this_rq->calc_load_update)) |
2401 | return; | |
783609c6 | 2402 | |
74f5187a | 2403 | delta = calc_load_fold_active(this_rq); |
74f5187a | 2404 | if (delta) |
dce48a84 | 2405 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2406 | |
2407 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2408 | } |
2409 | ||
5167e8d5 PZ |
2410 | /* |
2411 | * End of global load-average stuff | |
2412 | */ | |
2413 | ||
fdf3e95d VP |
2414 | /* |
2415 | * The exact cpuload at various idx values, calculated at every tick would be | |
2416 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2417 | * | |
2418 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2419 | * on nth tick when cpu may be busy, then we have: | |
2420 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2421 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2422 | * | |
2423 | * decay_load_missed() below does efficient calculation of | |
2424 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2425 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2426 | * | |
2427 | * The calculation is approximated on a 128 point scale. | |
2428 | * degrade_zero_ticks is the number of ticks after which load at any | |
2429 | * particular idx is approximated to be zero. | |
2430 | * degrade_factor is a precomputed table, a row for each load idx. | |
2431 | * Each column corresponds to degradation factor for a power of two ticks, | |
2432 | * based on 128 point scale. | |
2433 | * Example: | |
2434 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2435 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2436 | * | |
2437 | * With this power of 2 load factors, we can degrade the load n times | |
2438 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2439 | * n mult/shifts needed by the exact degradation. | |
2440 | */ | |
2441 | #define DEGRADE_SHIFT 7 | |
2442 | static const unsigned char | |
2443 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2444 | static const unsigned char | |
2445 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2446 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2447 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2448 | {96, 72, 40, 12, 1, 0, 0}, | |
2449 | {112, 98, 75, 43, 15, 1, 0}, | |
2450 | {120, 112, 98, 76, 45, 16, 2} }; | |
2451 | ||
2452 | /* | |
2453 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2454 | * would be when CPU is idle and so we just decay the old load without | |
2455 | * adding any new load. | |
2456 | */ | |
2457 | static unsigned long | |
2458 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2459 | { | |
2460 | int j = 0; | |
2461 | ||
2462 | if (!missed_updates) | |
2463 | return load; | |
2464 | ||
2465 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2466 | return 0; | |
2467 | ||
2468 | if (idx == 1) | |
2469 | return load >> missed_updates; | |
2470 | ||
2471 | while (missed_updates) { | |
2472 | if (missed_updates % 2) | |
2473 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2474 | ||
2475 | missed_updates >>= 1; | |
2476 | j++; | |
2477 | } | |
2478 | return load; | |
2479 | } | |
2480 | ||
46cb4b7c | 2481 | /* |
dd41f596 | 2482 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2483 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2484 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2485 | */ |
556061b0 PZ |
2486 | static void __update_cpu_load(struct rq *this_rq, unsigned long this_load, |
2487 | unsigned long pending_updates) | |
46cb4b7c | 2488 | { |
dd41f596 | 2489 | int i, scale; |
46cb4b7c | 2490 | |
dd41f596 | 2491 | this_rq->nr_load_updates++; |
46cb4b7c | 2492 | |
dd41f596 | 2493 | /* Update our load: */ |
fdf3e95d VP |
2494 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2495 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2496 | unsigned long old_load, new_load; |
7d1e6a9b | 2497 | |
dd41f596 | 2498 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2499 | |
dd41f596 | 2500 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2501 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2502 | new_load = this_load; |
a25707f3 IM |
2503 | /* |
2504 | * Round up the averaging division if load is increasing. This | |
2505 | * prevents us from getting stuck on 9 if the load is 10, for | |
2506 | * example. | |
2507 | */ | |
2508 | if (new_load > old_load) | |
fdf3e95d VP |
2509 | new_load += scale - 1; |
2510 | ||
2511 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2512 | } |
da2b71ed SS |
2513 | |
2514 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2515 | } |
2516 | ||
5aaa0b7a PZ |
2517 | #ifdef CONFIG_NO_HZ |
2518 | /* | |
2519 | * There is no sane way to deal with nohz on smp when using jiffies because the | |
2520 | * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading | |
2521 | * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}. | |
2522 | * | |
2523 | * Therefore we cannot use the delta approach from the regular tick since that | |
2524 | * would seriously skew the load calculation. However we'll make do for those | |
2525 | * updates happening while idle (nohz_idle_balance) or coming out of idle | |
2526 | * (tick_nohz_idle_exit). | |
2527 | * | |
2528 | * This means we might still be one tick off for nohz periods. | |
2529 | */ | |
2530 | ||
556061b0 PZ |
2531 | /* |
2532 | * Called from nohz_idle_balance() to update the load ratings before doing the | |
2533 | * idle balance. | |
2534 | */ | |
2535 | void update_idle_cpu_load(struct rq *this_rq) | |
2536 | { | |
5aaa0b7a | 2537 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
556061b0 PZ |
2538 | unsigned long load = this_rq->load.weight; |
2539 | unsigned long pending_updates; | |
2540 | ||
2541 | /* | |
5aaa0b7a | 2542 | * bail if there's load or we're actually up-to-date. |
556061b0 PZ |
2543 | */ |
2544 | if (load || curr_jiffies == this_rq->last_load_update_tick) | |
2545 | return; | |
2546 | ||
2547 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2548 | this_rq->last_load_update_tick = curr_jiffies; | |
2549 | ||
2550 | __update_cpu_load(this_rq, load, pending_updates); | |
2551 | } | |
2552 | ||
5aaa0b7a PZ |
2553 | /* |
2554 | * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed. | |
2555 | */ | |
2556 | void update_cpu_load_nohz(void) | |
2557 | { | |
2558 | struct rq *this_rq = this_rq(); | |
2559 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); | |
2560 | unsigned long pending_updates; | |
2561 | ||
2562 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2563 | return; | |
2564 | ||
2565 | raw_spin_lock(&this_rq->lock); | |
2566 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2567 | if (pending_updates) { | |
2568 | this_rq->last_load_update_tick = curr_jiffies; | |
2569 | /* | |
2570 | * We were idle, this means load 0, the current load might be | |
2571 | * !0 due to remote wakeups and the sort. | |
2572 | */ | |
2573 | __update_cpu_load(this_rq, 0, pending_updates); | |
2574 | } | |
2575 | raw_spin_unlock(&this_rq->lock); | |
2576 | } | |
2577 | #endif /* CONFIG_NO_HZ */ | |
2578 | ||
556061b0 PZ |
2579 | /* |
2580 | * Called from scheduler_tick() | |
2581 | */ | |
fdf3e95d VP |
2582 | static void update_cpu_load_active(struct rq *this_rq) |
2583 | { | |
556061b0 | 2584 | /* |
5aaa0b7a | 2585 | * See the mess around update_idle_cpu_load() / update_cpu_load_nohz(). |
556061b0 PZ |
2586 | */ |
2587 | this_rq->last_load_update_tick = jiffies; | |
2588 | __update_cpu_load(this_rq, this_rq->load.weight, 1); | |
46cb4b7c | 2589 | |
74f5187a | 2590 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2591 | } |
2592 | ||
dd41f596 | 2593 | #ifdef CONFIG_SMP |
8a0be9ef | 2594 | |
46cb4b7c | 2595 | /* |
38022906 PZ |
2596 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2597 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2598 | */ |
38022906 | 2599 | void sched_exec(void) |
46cb4b7c | 2600 | { |
38022906 | 2601 | struct task_struct *p = current; |
1da177e4 | 2602 | unsigned long flags; |
0017d735 | 2603 | int dest_cpu; |
46cb4b7c | 2604 | |
8f42ced9 | 2605 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2606 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2607 | if (dest_cpu == smp_processor_id()) |
2608 | goto unlock; | |
38022906 | 2609 | |
8f42ced9 | 2610 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2611 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2612 | |
8f42ced9 PZ |
2613 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2614 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2615 | return; |
2616 | } | |
0017d735 | 2617 | unlock: |
8f42ced9 | 2618 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2619 | } |
dd41f596 | 2620 | |
1da177e4 LT |
2621 | #endif |
2622 | ||
1da177e4 | 2623 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2624 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2625 | |
2626 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2627 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2628 | |
2629 | /* | |
c5f8d995 | 2630 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2631 | * @p in case that task is currently running. |
c5f8d995 HS |
2632 | * |
2633 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2634 | */ |
c5f8d995 HS |
2635 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2636 | { | |
2637 | u64 ns = 0; | |
2638 | ||
2639 | if (task_current(rq, p)) { | |
2640 | update_rq_clock(rq); | |
305e6835 | 2641 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2642 | if ((s64)ns < 0) |
2643 | ns = 0; | |
2644 | } | |
2645 | ||
2646 | return ns; | |
2647 | } | |
2648 | ||
bb34d92f | 2649 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2650 | { |
1da177e4 | 2651 | unsigned long flags; |
41b86e9c | 2652 | struct rq *rq; |
bb34d92f | 2653 | u64 ns = 0; |
48f24c4d | 2654 | |
41b86e9c | 2655 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2656 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2657 | task_rq_unlock(rq, p, &flags); |
1508487e | 2658 | |
c5f8d995 HS |
2659 | return ns; |
2660 | } | |
f06febc9 | 2661 | |
c5f8d995 HS |
2662 | /* |
2663 | * Return accounted runtime for the task. | |
2664 | * In case the task is currently running, return the runtime plus current's | |
2665 | * pending runtime that have not been accounted yet. | |
2666 | */ | |
2667 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2668 | { | |
2669 | unsigned long flags; | |
2670 | struct rq *rq; | |
2671 | u64 ns = 0; | |
2672 | ||
2673 | rq = task_rq_lock(p, &flags); | |
2674 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2675 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2676 | |
2677 | return ns; | |
2678 | } | |
48f24c4d | 2679 | |
7835b98b CL |
2680 | /* |
2681 | * This function gets called by the timer code, with HZ frequency. | |
2682 | * We call it with interrupts disabled. | |
7835b98b CL |
2683 | */ |
2684 | void scheduler_tick(void) | |
2685 | { | |
7835b98b CL |
2686 | int cpu = smp_processor_id(); |
2687 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2688 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2689 | |
2690 | sched_clock_tick(); | |
dd41f596 | 2691 | |
05fa785c | 2692 | raw_spin_lock(&rq->lock); |
3e51f33f | 2693 | update_rq_clock(rq); |
fdf3e95d | 2694 | update_cpu_load_active(rq); |
fa85ae24 | 2695 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2696 | raw_spin_unlock(&rq->lock); |
7835b98b | 2697 | |
e9d2b064 | 2698 | perf_event_task_tick(); |
e220d2dc | 2699 | |
e418e1c2 | 2700 | #ifdef CONFIG_SMP |
6eb57e0d | 2701 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2702 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2703 | #endif |
1da177e4 LT |
2704 | } |
2705 | ||
132380a0 | 2706 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2707 | { |
2708 | if (in_lock_functions(addr)) { | |
2709 | addr = CALLER_ADDR2; | |
2710 | if (in_lock_functions(addr)) | |
2711 | addr = CALLER_ADDR3; | |
2712 | } | |
2713 | return addr; | |
2714 | } | |
1da177e4 | 2715 | |
7e49fcce SR |
2716 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2717 | defined(CONFIG_PREEMPT_TRACER)) | |
2718 | ||
43627582 | 2719 | void __kprobes add_preempt_count(int val) |
1da177e4 | 2720 | { |
6cd8a4bb | 2721 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2722 | /* |
2723 | * Underflow? | |
2724 | */ | |
9a11b49a IM |
2725 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2726 | return; | |
6cd8a4bb | 2727 | #endif |
1da177e4 | 2728 | preempt_count() += val; |
6cd8a4bb | 2729 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2730 | /* |
2731 | * Spinlock count overflowing soon? | |
2732 | */ | |
33859f7f MOS |
2733 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2734 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2735 | #endif |
2736 | if (preempt_count() == val) | |
2737 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2738 | } |
2739 | EXPORT_SYMBOL(add_preempt_count); | |
2740 | ||
43627582 | 2741 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 2742 | { |
6cd8a4bb | 2743 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2744 | /* |
2745 | * Underflow? | |
2746 | */ | |
01e3eb82 | 2747 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2748 | return; |
1da177e4 LT |
2749 | /* |
2750 | * Is the spinlock portion underflowing? | |
2751 | */ | |
9a11b49a IM |
2752 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2753 | !(preempt_count() & PREEMPT_MASK))) | |
2754 | return; | |
6cd8a4bb | 2755 | #endif |
9a11b49a | 2756 | |
6cd8a4bb SR |
2757 | if (preempt_count() == val) |
2758 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2759 | preempt_count() -= val; |
2760 | } | |
2761 | EXPORT_SYMBOL(sub_preempt_count); | |
2762 | ||
2763 | #endif | |
2764 | ||
2765 | /* | |
dd41f596 | 2766 | * Print scheduling while atomic bug: |
1da177e4 | 2767 | */ |
dd41f596 | 2768 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2769 | { |
664dfa65 DJ |
2770 | if (oops_in_progress) |
2771 | return; | |
2772 | ||
3df0fc5b PZ |
2773 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2774 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2775 | |
dd41f596 | 2776 | debug_show_held_locks(prev); |
e21f5b15 | 2777 | print_modules(); |
dd41f596 IM |
2778 | if (irqs_disabled()) |
2779 | print_irqtrace_events(prev); | |
6135fc1e | 2780 | dump_stack(); |
373d4d09 | 2781 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2782 | } |
1da177e4 | 2783 | |
dd41f596 IM |
2784 | /* |
2785 | * Various schedule()-time debugging checks and statistics: | |
2786 | */ | |
2787 | static inline void schedule_debug(struct task_struct *prev) | |
2788 | { | |
1da177e4 | 2789 | /* |
41a2d6cf | 2790 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
2791 | * schedule() atomically, we ignore that path for now. |
2792 | * Otherwise, whine if we are scheduling when we should not be. | |
2793 | */ | |
3f33a7ce | 2794 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 2795 | __schedule_bug(prev); |
b3fbab05 | 2796 | rcu_sleep_check(); |
dd41f596 | 2797 | |
1da177e4 LT |
2798 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2799 | ||
2d72376b | 2800 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2801 | } |
2802 | ||
6cecd084 | 2803 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2804 | { |
61eadef6 | 2805 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2806 | update_rq_clock(rq); |
6cecd084 | 2807 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2808 | } |
2809 | ||
dd41f596 IM |
2810 | /* |
2811 | * Pick up the highest-prio task: | |
2812 | */ | |
2813 | static inline struct task_struct * | |
b67802ea | 2814 | pick_next_task(struct rq *rq) |
dd41f596 | 2815 | { |
5522d5d5 | 2816 | const struct sched_class *class; |
dd41f596 | 2817 | struct task_struct *p; |
1da177e4 LT |
2818 | |
2819 | /* | |
dd41f596 IM |
2820 | * Optimization: we know that if all tasks are in |
2821 | * the fair class we can call that function directly: | |
1da177e4 | 2822 | */ |
953bfcd1 | 2823 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2824 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2825 | if (likely(p)) |
2826 | return p; | |
1da177e4 LT |
2827 | } |
2828 | ||
34f971f6 | 2829 | for_each_class(class) { |
fb8d4724 | 2830 | p = class->pick_next_task(rq); |
dd41f596 IM |
2831 | if (p) |
2832 | return p; | |
dd41f596 | 2833 | } |
34f971f6 PZ |
2834 | |
2835 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2836 | } |
1da177e4 | 2837 | |
dd41f596 | 2838 | /* |
c259e01a | 2839 | * __schedule() is the main scheduler function. |
edde96ea PE |
2840 | * |
2841 | * The main means of driving the scheduler and thus entering this function are: | |
2842 | * | |
2843 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2844 | * | |
2845 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2846 | * paths. For example, see arch/x86/entry_64.S. | |
2847 | * | |
2848 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2849 | * interrupt handler scheduler_tick(). | |
2850 | * | |
2851 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2852 | * task to the run-queue and that's it. | |
2853 | * | |
2854 | * Now, if the new task added to the run-queue preempts the current | |
2855 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2856 | * called on the nearest possible occasion: | |
2857 | * | |
2858 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2859 | * | |
2860 | * - in syscall or exception context, at the next outmost | |
2861 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2862 | * spin_unlock()!) | |
2863 | * | |
2864 | * - in IRQ context, return from interrupt-handler to | |
2865 | * preemptible context | |
2866 | * | |
2867 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2868 | * then at the next: | |
2869 | * | |
2870 | * - cond_resched() call | |
2871 | * - explicit schedule() call | |
2872 | * - return from syscall or exception to user-space | |
2873 | * - return from interrupt-handler to user-space | |
dd41f596 | 2874 | */ |
c259e01a | 2875 | static void __sched __schedule(void) |
dd41f596 IM |
2876 | { |
2877 | struct task_struct *prev, *next; | |
67ca7bde | 2878 | unsigned long *switch_count; |
dd41f596 | 2879 | struct rq *rq; |
31656519 | 2880 | int cpu; |
dd41f596 | 2881 | |
ff743345 PZ |
2882 | need_resched: |
2883 | preempt_disable(); | |
dd41f596 IM |
2884 | cpu = smp_processor_id(); |
2885 | rq = cpu_rq(cpu); | |
25502a6c | 2886 | rcu_note_context_switch(cpu); |
dd41f596 | 2887 | prev = rq->curr; |
dd41f596 | 2888 | |
dd41f596 | 2889 | schedule_debug(prev); |
1da177e4 | 2890 | |
31656519 | 2891 | if (sched_feat(HRTICK)) |
f333fdc9 | 2892 | hrtick_clear(rq); |
8f4d37ec | 2893 | |
05fa785c | 2894 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2895 | |
246d86b5 | 2896 | switch_count = &prev->nivcsw; |
1da177e4 | 2897 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2898 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2899 | prev->state = TASK_RUNNING; |
21aa9af0 | 2900 | } else { |
2acca55e PZ |
2901 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2902 | prev->on_rq = 0; | |
2903 | ||
21aa9af0 | 2904 | /* |
2acca55e PZ |
2905 | * If a worker went to sleep, notify and ask workqueue |
2906 | * whether it wants to wake up a task to maintain | |
2907 | * concurrency. | |
21aa9af0 TH |
2908 | */ |
2909 | if (prev->flags & PF_WQ_WORKER) { | |
2910 | struct task_struct *to_wakeup; | |
2911 | ||
2912 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2913 | if (to_wakeup) | |
2914 | try_to_wake_up_local(to_wakeup); | |
2915 | } | |
21aa9af0 | 2916 | } |
dd41f596 | 2917 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2918 | } |
2919 | ||
3f029d3c | 2920 | pre_schedule(rq, prev); |
f65eda4f | 2921 | |
dd41f596 | 2922 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2923 | idle_balance(cpu, rq); |
1da177e4 | 2924 | |
df1c99d4 | 2925 | put_prev_task(rq, prev); |
b67802ea | 2926 | next = pick_next_task(rq); |
f26f9aff MG |
2927 | clear_tsk_need_resched(prev); |
2928 | rq->skip_clock_update = 0; | |
1da177e4 | 2929 | |
1da177e4 | 2930 | if (likely(prev != next)) { |
1da177e4 LT |
2931 | rq->nr_switches++; |
2932 | rq->curr = next; | |
2933 | ++*switch_count; | |
2934 | ||
dd41f596 | 2935 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2936 | /* |
246d86b5 ON |
2937 | * The context switch have flipped the stack from under us |
2938 | * and restored the local variables which were saved when | |
2939 | * this task called schedule() in the past. prev == current | |
2940 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2941 | */ |
2942 | cpu = smp_processor_id(); | |
2943 | rq = cpu_rq(cpu); | |
1da177e4 | 2944 | } else |
05fa785c | 2945 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2946 | |
3f029d3c | 2947 | post_schedule(rq); |
1da177e4 | 2948 | |
ba74c144 | 2949 | sched_preempt_enable_no_resched(); |
ff743345 | 2950 | if (need_resched()) |
1da177e4 LT |
2951 | goto need_resched; |
2952 | } | |
c259e01a | 2953 | |
9c40cef2 TG |
2954 | static inline void sched_submit_work(struct task_struct *tsk) |
2955 | { | |
3c7d5184 | 2956 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2957 | return; |
2958 | /* | |
2959 | * If we are going to sleep and we have plugged IO queued, | |
2960 | * make sure to submit it to avoid deadlocks. | |
2961 | */ | |
2962 | if (blk_needs_flush_plug(tsk)) | |
2963 | blk_schedule_flush_plug(tsk); | |
2964 | } | |
2965 | ||
6ebbe7a0 | 2966 | asmlinkage void __sched schedule(void) |
c259e01a | 2967 | { |
9c40cef2 TG |
2968 | struct task_struct *tsk = current; |
2969 | ||
2970 | sched_submit_work(tsk); | |
c259e01a TG |
2971 | __schedule(); |
2972 | } | |
1da177e4 LT |
2973 | EXPORT_SYMBOL(schedule); |
2974 | ||
91d1aa43 | 2975 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2976 | asmlinkage void __sched schedule_user(void) |
2977 | { | |
2978 | /* | |
2979 | * If we come here after a random call to set_need_resched(), | |
2980 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2981 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2982 | * we find a better solution. | |
2983 | */ | |
91d1aa43 | 2984 | user_exit(); |
20ab65e3 | 2985 | schedule(); |
91d1aa43 | 2986 | user_enter(); |
20ab65e3 FW |
2987 | } |
2988 | #endif | |
2989 | ||
c5491ea7 TG |
2990 | /** |
2991 | * schedule_preempt_disabled - called with preemption disabled | |
2992 | * | |
2993 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2994 | */ | |
2995 | void __sched schedule_preempt_disabled(void) | |
2996 | { | |
ba74c144 | 2997 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2998 | schedule(); |
2999 | preempt_disable(); | |
3000 | } | |
3001 | ||
c08f7829 | 3002 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3003 | |
c6eb3dda PZ |
3004 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3005 | { | |
c6eb3dda | 3006 | if (lock->owner != owner) |
307bf980 | 3007 | return false; |
0d66bf6d PZ |
3008 | |
3009 | /* | |
c6eb3dda PZ |
3010 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3011 | * lock->owner still matches owner, if that fails, owner might | |
3012 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3013 | * ensures the memory stays valid. | |
0d66bf6d | 3014 | */ |
c6eb3dda | 3015 | barrier(); |
0d66bf6d | 3016 | |
307bf980 | 3017 | return owner->on_cpu; |
c6eb3dda | 3018 | } |
0d66bf6d | 3019 | |
c6eb3dda PZ |
3020 | /* |
3021 | * Look out! "owner" is an entirely speculative pointer | |
3022 | * access and not reliable. | |
3023 | */ | |
3024 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3025 | { | |
3026 | if (!sched_feat(OWNER_SPIN)) | |
3027 | return 0; | |
0d66bf6d | 3028 | |
307bf980 | 3029 | rcu_read_lock(); |
c6eb3dda PZ |
3030 | while (owner_running(lock, owner)) { |
3031 | if (need_resched()) | |
307bf980 | 3032 | break; |
0d66bf6d | 3033 | |
335d7afb | 3034 | arch_mutex_cpu_relax(); |
0d66bf6d | 3035 | } |
307bf980 | 3036 | rcu_read_unlock(); |
4b402210 | 3037 | |
c6eb3dda | 3038 | /* |
307bf980 TG |
3039 | * We break out the loop above on need_resched() and when the |
3040 | * owner changed, which is a sign for heavy contention. Return | |
3041 | * success only when lock->owner is NULL. | |
c6eb3dda | 3042 | */ |
307bf980 | 3043 | return lock->owner == NULL; |
0d66bf6d PZ |
3044 | } |
3045 | #endif | |
3046 | ||
1da177e4 LT |
3047 | #ifdef CONFIG_PREEMPT |
3048 | /* | |
2ed6e34f | 3049 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3050 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3051 | * occur there and call schedule directly. |
3052 | */ | |
d1f74e20 | 3053 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3054 | { |
3055 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3056 | |
1da177e4 LT |
3057 | /* |
3058 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3059 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3060 | */ |
beed33a8 | 3061 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3062 | return; |
3063 | ||
3a5c359a | 3064 | do { |
d1f74e20 | 3065 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3066 | __schedule(); |
d1f74e20 | 3067 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3068 | |
3a5c359a AK |
3069 | /* |
3070 | * Check again in case we missed a preemption opportunity | |
3071 | * between schedule and now. | |
3072 | */ | |
3073 | barrier(); | |
5ed0cec0 | 3074 | } while (need_resched()); |
1da177e4 | 3075 | } |
1da177e4 LT |
3076 | EXPORT_SYMBOL(preempt_schedule); |
3077 | ||
3078 | /* | |
2ed6e34f | 3079 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3080 | * off of irq context. |
3081 | * Note, that this is called and return with irqs disabled. This will | |
3082 | * protect us against recursive calling from irq. | |
3083 | */ | |
3084 | asmlinkage void __sched preempt_schedule_irq(void) | |
3085 | { | |
3086 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3087 | |
2ed6e34f | 3088 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3089 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3090 | ||
91d1aa43 | 3091 | user_exit(); |
3a5c359a AK |
3092 | do { |
3093 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3094 | local_irq_enable(); |
c259e01a | 3095 | __schedule(); |
3a5c359a | 3096 | local_irq_disable(); |
3a5c359a | 3097 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3098 | |
3a5c359a AK |
3099 | /* |
3100 | * Check again in case we missed a preemption opportunity | |
3101 | * between schedule and now. | |
3102 | */ | |
3103 | barrier(); | |
5ed0cec0 | 3104 | } while (need_resched()); |
1da177e4 LT |
3105 | } |
3106 | ||
3107 | #endif /* CONFIG_PREEMPT */ | |
3108 | ||
63859d4f | 3109 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3110 | void *key) |
1da177e4 | 3111 | { |
63859d4f | 3112 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3113 | } |
1da177e4 LT |
3114 | EXPORT_SYMBOL(default_wake_function); |
3115 | ||
3116 | /* | |
41a2d6cf IM |
3117 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3118 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3119 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3120 | * | |
3121 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3122 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3123 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3124 | */ | |
78ddb08f | 3125 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3126 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3127 | { |
2e45874c | 3128 | wait_queue_t *curr, *next; |
1da177e4 | 3129 | |
2e45874c | 3130 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3131 | unsigned flags = curr->flags; |
3132 | ||
63859d4f | 3133 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3134 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3135 | break; |
3136 | } | |
3137 | } | |
3138 | ||
3139 | /** | |
3140 | * __wake_up - wake up threads blocked on a waitqueue. | |
3141 | * @q: the waitqueue | |
3142 | * @mode: which threads | |
3143 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3144 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3145 | * |
3146 | * It may be assumed that this function implies a write memory barrier before | |
3147 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3148 | */ |
7ad5b3a5 | 3149 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3150 | int nr_exclusive, void *key) |
1da177e4 LT |
3151 | { |
3152 | unsigned long flags; | |
3153 | ||
3154 | spin_lock_irqsave(&q->lock, flags); | |
3155 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3156 | spin_unlock_irqrestore(&q->lock, flags); | |
3157 | } | |
1da177e4 LT |
3158 | EXPORT_SYMBOL(__wake_up); |
3159 | ||
3160 | /* | |
3161 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3162 | */ | |
63b20011 | 3163 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3164 | { |
63b20011 | 3165 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3166 | } |
22c43c81 | 3167 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3168 | |
4ede816a DL |
3169 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3170 | { | |
3171 | __wake_up_common(q, mode, 1, 0, key); | |
3172 | } | |
bf294b41 | 3173 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3174 | |
1da177e4 | 3175 | /** |
4ede816a | 3176 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3177 | * @q: the waitqueue |
3178 | * @mode: which threads | |
3179 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3180 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3181 | * |
3182 | * The sync wakeup differs that the waker knows that it will schedule | |
3183 | * away soon, so while the target thread will be woken up, it will not | |
3184 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3185 | * with each other. This can prevent needless bouncing between CPUs. | |
3186 | * | |
3187 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3188 | * |
3189 | * It may be assumed that this function implies a write memory barrier before | |
3190 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3191 | */ |
4ede816a DL |
3192 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3193 | int nr_exclusive, void *key) | |
1da177e4 LT |
3194 | { |
3195 | unsigned long flags; | |
7d478721 | 3196 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3197 | |
3198 | if (unlikely(!q)) | |
3199 | return; | |
3200 | ||
3201 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3202 | wake_flags = 0; |
1da177e4 LT |
3203 | |
3204 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3205 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3206 | spin_unlock_irqrestore(&q->lock, flags); |
3207 | } | |
4ede816a DL |
3208 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3209 | ||
3210 | /* | |
3211 | * __wake_up_sync - see __wake_up_sync_key() | |
3212 | */ | |
3213 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3214 | { | |
3215 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3216 | } | |
1da177e4 LT |
3217 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3218 | ||
65eb3dc6 KD |
3219 | /** |
3220 | * complete: - signals a single thread waiting on this completion | |
3221 | * @x: holds the state of this particular completion | |
3222 | * | |
3223 | * This will wake up a single thread waiting on this completion. Threads will be | |
3224 | * awakened in the same order in which they were queued. | |
3225 | * | |
3226 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3227 | * |
3228 | * It may be assumed that this function implies a write memory barrier before | |
3229 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3230 | */ |
b15136e9 | 3231 | void complete(struct completion *x) |
1da177e4 LT |
3232 | { |
3233 | unsigned long flags; | |
3234 | ||
3235 | spin_lock_irqsave(&x->wait.lock, flags); | |
3236 | x->done++; | |
d9514f6c | 3237 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3238 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3239 | } | |
3240 | EXPORT_SYMBOL(complete); | |
3241 | ||
65eb3dc6 KD |
3242 | /** |
3243 | * complete_all: - signals all threads waiting on this completion | |
3244 | * @x: holds the state of this particular completion | |
3245 | * | |
3246 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3247 | * |
3248 | * It may be assumed that this function implies a write memory barrier before | |
3249 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3250 | */ |
b15136e9 | 3251 | void complete_all(struct completion *x) |
1da177e4 LT |
3252 | { |
3253 | unsigned long flags; | |
3254 | ||
3255 | spin_lock_irqsave(&x->wait.lock, flags); | |
3256 | x->done += UINT_MAX/2; | |
d9514f6c | 3257 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3258 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3259 | } | |
3260 | EXPORT_SYMBOL(complete_all); | |
3261 | ||
8cbbe86d AK |
3262 | static inline long __sched |
3263 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3264 | { |
1da177e4 LT |
3265 | if (!x->done) { |
3266 | DECLARE_WAITQUEUE(wait, current); | |
3267 | ||
a93d2f17 | 3268 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3269 | do { |
94d3d824 | 3270 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3271 | timeout = -ERESTARTSYS; |
3272 | break; | |
8cbbe86d AK |
3273 | } |
3274 | __set_current_state(state); | |
1da177e4 LT |
3275 | spin_unlock_irq(&x->wait.lock); |
3276 | timeout = schedule_timeout(timeout); | |
3277 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3278 | } while (!x->done && timeout); |
1da177e4 | 3279 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3280 | if (!x->done) |
3281 | return timeout; | |
1da177e4 LT |
3282 | } |
3283 | x->done--; | |
ea71a546 | 3284 | return timeout ?: 1; |
1da177e4 | 3285 | } |
1da177e4 | 3286 | |
8cbbe86d AK |
3287 | static long __sched |
3288 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3289 | { |
1da177e4 LT |
3290 | might_sleep(); |
3291 | ||
3292 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3293 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3294 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3295 | return timeout; |
3296 | } | |
1da177e4 | 3297 | |
65eb3dc6 KD |
3298 | /** |
3299 | * wait_for_completion: - waits for completion of a task | |
3300 | * @x: holds the state of this particular completion | |
3301 | * | |
3302 | * This waits to be signaled for completion of a specific task. It is NOT | |
3303 | * interruptible and there is no timeout. | |
3304 | * | |
3305 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3306 | * and interrupt capability. Also see complete(). | |
3307 | */ | |
b15136e9 | 3308 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3309 | { |
3310 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3311 | } |
8cbbe86d | 3312 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3313 | |
65eb3dc6 KD |
3314 | /** |
3315 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3316 | * @x: holds the state of this particular completion | |
3317 | * @timeout: timeout value in jiffies | |
3318 | * | |
3319 | * This waits for either a completion of a specific task to be signaled or for a | |
3320 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3321 | * interruptible. | |
c6dc7f05 BF |
3322 | * |
3323 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3324 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3325 | */ |
b15136e9 | 3326 | unsigned long __sched |
8cbbe86d | 3327 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3328 | { |
8cbbe86d | 3329 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3330 | } |
8cbbe86d | 3331 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3332 | |
65eb3dc6 KD |
3333 | /** |
3334 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3335 | * @x: holds the state of this particular completion | |
3336 | * | |
3337 | * This waits for completion of a specific task to be signaled. It is | |
3338 | * interruptible. | |
c6dc7f05 BF |
3339 | * |
3340 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3341 | */ |
8cbbe86d | 3342 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3343 | { |
51e97990 AK |
3344 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3345 | if (t == -ERESTARTSYS) | |
3346 | return t; | |
3347 | return 0; | |
0fec171c | 3348 | } |
8cbbe86d | 3349 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3350 | |
65eb3dc6 KD |
3351 | /** |
3352 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3353 | * @x: holds the state of this particular completion | |
3354 | * @timeout: timeout value in jiffies | |
3355 | * | |
3356 | * This waits for either a completion of a specific task to be signaled or for a | |
3357 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3358 | * |
3359 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3360 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3361 | */ |
6bf41237 | 3362 | long __sched |
8cbbe86d AK |
3363 | wait_for_completion_interruptible_timeout(struct completion *x, |
3364 | unsigned long timeout) | |
0fec171c | 3365 | { |
8cbbe86d | 3366 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3367 | } |
8cbbe86d | 3368 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3369 | |
65eb3dc6 KD |
3370 | /** |
3371 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3372 | * @x: holds the state of this particular completion | |
3373 | * | |
3374 | * This waits to be signaled for completion of a specific task. It can be | |
3375 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3376 | * |
3377 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3378 | */ |
009e577e MW |
3379 | int __sched wait_for_completion_killable(struct completion *x) |
3380 | { | |
3381 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3382 | if (t == -ERESTARTSYS) | |
3383 | return t; | |
3384 | return 0; | |
3385 | } | |
3386 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3387 | ||
0aa12fb4 SW |
3388 | /** |
3389 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3390 | * @x: holds the state of this particular completion | |
3391 | * @timeout: timeout value in jiffies | |
3392 | * | |
3393 | * This waits for either a completion of a specific task to be | |
3394 | * signaled or for a specified timeout to expire. It can be | |
3395 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3396 | * |
3397 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3398 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3399 | */ |
6bf41237 | 3400 | long __sched |
0aa12fb4 SW |
3401 | wait_for_completion_killable_timeout(struct completion *x, |
3402 | unsigned long timeout) | |
3403 | { | |
3404 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3405 | } | |
3406 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3407 | ||
be4de352 DC |
3408 | /** |
3409 | * try_wait_for_completion - try to decrement a completion without blocking | |
3410 | * @x: completion structure | |
3411 | * | |
3412 | * Returns: 0 if a decrement cannot be done without blocking | |
3413 | * 1 if a decrement succeeded. | |
3414 | * | |
3415 | * If a completion is being used as a counting completion, | |
3416 | * attempt to decrement the counter without blocking. This | |
3417 | * enables us to avoid waiting if the resource the completion | |
3418 | * is protecting is not available. | |
3419 | */ | |
3420 | bool try_wait_for_completion(struct completion *x) | |
3421 | { | |
7539a3b3 | 3422 | unsigned long flags; |
be4de352 DC |
3423 | int ret = 1; |
3424 | ||
7539a3b3 | 3425 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3426 | if (!x->done) |
3427 | ret = 0; | |
3428 | else | |
3429 | x->done--; | |
7539a3b3 | 3430 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3431 | return ret; |
3432 | } | |
3433 | EXPORT_SYMBOL(try_wait_for_completion); | |
3434 | ||
3435 | /** | |
3436 | * completion_done - Test to see if a completion has any waiters | |
3437 | * @x: completion structure | |
3438 | * | |
3439 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3440 | * 1 if there are no waiters. | |
3441 | * | |
3442 | */ | |
3443 | bool completion_done(struct completion *x) | |
3444 | { | |
7539a3b3 | 3445 | unsigned long flags; |
be4de352 DC |
3446 | int ret = 1; |
3447 | ||
7539a3b3 | 3448 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3449 | if (!x->done) |
3450 | ret = 0; | |
7539a3b3 | 3451 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3452 | return ret; |
3453 | } | |
3454 | EXPORT_SYMBOL(completion_done); | |
3455 | ||
8cbbe86d AK |
3456 | static long __sched |
3457 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3458 | { |
0fec171c IM |
3459 | unsigned long flags; |
3460 | wait_queue_t wait; | |
3461 | ||
3462 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3463 | |
8cbbe86d | 3464 | __set_current_state(state); |
1da177e4 | 3465 | |
8cbbe86d AK |
3466 | spin_lock_irqsave(&q->lock, flags); |
3467 | __add_wait_queue(q, &wait); | |
3468 | spin_unlock(&q->lock); | |
3469 | timeout = schedule_timeout(timeout); | |
3470 | spin_lock_irq(&q->lock); | |
3471 | __remove_wait_queue(q, &wait); | |
3472 | spin_unlock_irqrestore(&q->lock, flags); | |
3473 | ||
3474 | return timeout; | |
3475 | } | |
3476 | ||
3477 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3478 | { | |
3479 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3480 | } |
1da177e4 LT |
3481 | EXPORT_SYMBOL(interruptible_sleep_on); |
3482 | ||
0fec171c | 3483 | long __sched |
95cdf3b7 | 3484 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3485 | { |
8cbbe86d | 3486 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3487 | } |
1da177e4 LT |
3488 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3489 | ||
0fec171c | 3490 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3491 | { |
8cbbe86d | 3492 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3493 | } |
1da177e4 LT |
3494 | EXPORT_SYMBOL(sleep_on); |
3495 | ||
0fec171c | 3496 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3497 | { |
8cbbe86d | 3498 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3499 | } |
1da177e4 LT |
3500 | EXPORT_SYMBOL(sleep_on_timeout); |
3501 | ||
b29739f9 IM |
3502 | #ifdef CONFIG_RT_MUTEXES |
3503 | ||
3504 | /* | |
3505 | * rt_mutex_setprio - set the current priority of a task | |
3506 | * @p: task | |
3507 | * @prio: prio value (kernel-internal form) | |
3508 | * | |
3509 | * This function changes the 'effective' priority of a task. It does | |
3510 | * not touch ->normal_prio like __setscheduler(). | |
3511 | * | |
3512 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3513 | */ | |
36c8b586 | 3514 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3515 | { |
83b699ed | 3516 | int oldprio, on_rq, running; |
70b97a7f | 3517 | struct rq *rq; |
83ab0aa0 | 3518 | const struct sched_class *prev_class; |
b29739f9 IM |
3519 | |
3520 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3521 | ||
0122ec5b | 3522 | rq = __task_rq_lock(p); |
b29739f9 | 3523 | |
1c4dd99b TG |
3524 | /* |
3525 | * Idle task boosting is a nono in general. There is one | |
3526 | * exception, when PREEMPT_RT and NOHZ is active: | |
3527 | * | |
3528 | * The idle task calls get_next_timer_interrupt() and holds | |
3529 | * the timer wheel base->lock on the CPU and another CPU wants | |
3530 | * to access the timer (probably to cancel it). We can safely | |
3531 | * ignore the boosting request, as the idle CPU runs this code | |
3532 | * with interrupts disabled and will complete the lock | |
3533 | * protected section without being interrupted. So there is no | |
3534 | * real need to boost. | |
3535 | */ | |
3536 | if (unlikely(p == rq->idle)) { | |
3537 | WARN_ON(p != rq->curr); | |
3538 | WARN_ON(p->pi_blocked_on); | |
3539 | goto out_unlock; | |
3540 | } | |
3541 | ||
a8027073 | 3542 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3543 | oldprio = p->prio; |
83ab0aa0 | 3544 | prev_class = p->sched_class; |
fd2f4419 | 3545 | on_rq = p->on_rq; |
051a1d1a | 3546 | running = task_current(rq, p); |
0e1f3483 | 3547 | if (on_rq) |
69be72c1 | 3548 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3549 | if (running) |
3550 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3551 | |
3552 | if (rt_prio(prio)) | |
3553 | p->sched_class = &rt_sched_class; | |
3554 | else | |
3555 | p->sched_class = &fair_sched_class; | |
3556 | ||
b29739f9 IM |
3557 | p->prio = prio; |
3558 | ||
0e1f3483 HS |
3559 | if (running) |
3560 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3561 | if (on_rq) |
371fd7e7 | 3562 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3563 | |
da7a735e | 3564 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3565 | out_unlock: |
0122ec5b | 3566 | __task_rq_unlock(rq); |
b29739f9 | 3567 | } |
b29739f9 | 3568 | #endif |
36c8b586 | 3569 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3570 | { |
dd41f596 | 3571 | int old_prio, delta, on_rq; |
1da177e4 | 3572 | unsigned long flags; |
70b97a7f | 3573 | struct rq *rq; |
1da177e4 LT |
3574 | |
3575 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3576 | return; | |
3577 | /* | |
3578 | * We have to be careful, if called from sys_setpriority(), | |
3579 | * the task might be in the middle of scheduling on another CPU. | |
3580 | */ | |
3581 | rq = task_rq_lock(p, &flags); | |
3582 | /* | |
3583 | * The RT priorities are set via sched_setscheduler(), but we still | |
3584 | * allow the 'normal' nice value to be set - but as expected | |
3585 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3586 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3587 | */ |
e05606d3 | 3588 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3589 | p->static_prio = NICE_TO_PRIO(nice); |
3590 | goto out_unlock; | |
3591 | } | |
fd2f4419 | 3592 | on_rq = p->on_rq; |
c09595f6 | 3593 | if (on_rq) |
69be72c1 | 3594 | dequeue_task(rq, p, 0); |
1da177e4 | 3595 | |
1da177e4 | 3596 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3597 | set_load_weight(p); |
b29739f9 IM |
3598 | old_prio = p->prio; |
3599 | p->prio = effective_prio(p); | |
3600 | delta = p->prio - old_prio; | |
1da177e4 | 3601 | |
dd41f596 | 3602 | if (on_rq) { |
371fd7e7 | 3603 | enqueue_task(rq, p, 0); |
1da177e4 | 3604 | /* |
d5f9f942 AM |
3605 | * If the task increased its priority or is running and |
3606 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3607 | */ |
d5f9f942 | 3608 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3609 | resched_task(rq->curr); |
3610 | } | |
3611 | out_unlock: | |
0122ec5b | 3612 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3613 | } |
1da177e4 LT |
3614 | EXPORT_SYMBOL(set_user_nice); |
3615 | ||
e43379f1 MM |
3616 | /* |
3617 | * can_nice - check if a task can reduce its nice value | |
3618 | * @p: task | |
3619 | * @nice: nice value | |
3620 | */ | |
36c8b586 | 3621 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3622 | { |
024f4747 MM |
3623 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3624 | int nice_rlim = 20 - nice; | |
48f24c4d | 3625 | |
78d7d407 | 3626 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3627 | capable(CAP_SYS_NICE)); |
3628 | } | |
3629 | ||
1da177e4 LT |
3630 | #ifdef __ARCH_WANT_SYS_NICE |
3631 | ||
3632 | /* | |
3633 | * sys_nice - change the priority of the current process. | |
3634 | * @increment: priority increment | |
3635 | * | |
3636 | * sys_setpriority is a more generic, but much slower function that | |
3637 | * does similar things. | |
3638 | */ | |
5add95d4 | 3639 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3640 | { |
48f24c4d | 3641 | long nice, retval; |
1da177e4 LT |
3642 | |
3643 | /* | |
3644 | * Setpriority might change our priority at the same moment. | |
3645 | * We don't have to worry. Conceptually one call occurs first | |
3646 | * and we have a single winner. | |
3647 | */ | |
e43379f1 MM |
3648 | if (increment < -40) |
3649 | increment = -40; | |
1da177e4 LT |
3650 | if (increment > 40) |
3651 | increment = 40; | |
3652 | ||
2b8f836f | 3653 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3654 | if (nice < -20) |
3655 | nice = -20; | |
3656 | if (nice > 19) | |
3657 | nice = 19; | |
3658 | ||
e43379f1 MM |
3659 | if (increment < 0 && !can_nice(current, nice)) |
3660 | return -EPERM; | |
3661 | ||
1da177e4 LT |
3662 | retval = security_task_setnice(current, nice); |
3663 | if (retval) | |
3664 | return retval; | |
3665 | ||
3666 | set_user_nice(current, nice); | |
3667 | return 0; | |
3668 | } | |
3669 | ||
3670 | #endif | |
3671 | ||
3672 | /** | |
3673 | * task_prio - return the priority value of a given task. | |
3674 | * @p: the task in question. | |
3675 | * | |
3676 | * This is the priority value as seen by users in /proc. | |
3677 | * RT tasks are offset by -200. Normal tasks are centered | |
3678 | * around 0, value goes from -16 to +15. | |
3679 | */ | |
36c8b586 | 3680 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3681 | { |
3682 | return p->prio - MAX_RT_PRIO; | |
3683 | } | |
3684 | ||
3685 | /** | |
3686 | * task_nice - return the nice value of a given task. | |
3687 | * @p: the task in question. | |
3688 | */ | |
36c8b586 | 3689 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3690 | { |
3691 | return TASK_NICE(p); | |
3692 | } | |
150d8bed | 3693 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3694 | |
3695 | /** | |
3696 | * idle_cpu - is a given cpu idle currently? | |
3697 | * @cpu: the processor in question. | |
3698 | */ | |
3699 | int idle_cpu(int cpu) | |
3700 | { | |
908a3283 TG |
3701 | struct rq *rq = cpu_rq(cpu); |
3702 | ||
3703 | if (rq->curr != rq->idle) | |
3704 | return 0; | |
3705 | ||
3706 | if (rq->nr_running) | |
3707 | return 0; | |
3708 | ||
3709 | #ifdef CONFIG_SMP | |
3710 | if (!llist_empty(&rq->wake_list)) | |
3711 | return 0; | |
3712 | #endif | |
3713 | ||
3714 | return 1; | |
1da177e4 LT |
3715 | } |
3716 | ||
1da177e4 LT |
3717 | /** |
3718 | * idle_task - return the idle task for a given cpu. | |
3719 | * @cpu: the processor in question. | |
3720 | */ | |
36c8b586 | 3721 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3722 | { |
3723 | return cpu_rq(cpu)->idle; | |
3724 | } | |
3725 | ||
3726 | /** | |
3727 | * find_process_by_pid - find a process with a matching PID value. | |
3728 | * @pid: the pid in question. | |
3729 | */ | |
a9957449 | 3730 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3731 | { |
228ebcbe | 3732 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3733 | } |
3734 | ||
3735 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3736 | static void |
3737 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3738 | { |
1da177e4 LT |
3739 | p->policy = policy; |
3740 | p->rt_priority = prio; | |
b29739f9 IM |
3741 | p->normal_prio = normal_prio(p); |
3742 | /* we are holding p->pi_lock already */ | |
3743 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3744 | if (rt_prio(p->prio)) |
3745 | p->sched_class = &rt_sched_class; | |
3746 | else | |
3747 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3748 | set_load_weight(p); |
1da177e4 LT |
3749 | } |
3750 | ||
c69e8d9c DH |
3751 | /* |
3752 | * check the target process has a UID that matches the current process's | |
3753 | */ | |
3754 | static bool check_same_owner(struct task_struct *p) | |
3755 | { | |
3756 | const struct cred *cred = current_cred(), *pcred; | |
3757 | bool match; | |
3758 | ||
3759 | rcu_read_lock(); | |
3760 | pcred = __task_cred(p); | |
9c806aa0 EB |
3761 | match = (uid_eq(cred->euid, pcred->euid) || |
3762 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3763 | rcu_read_unlock(); |
3764 | return match; | |
3765 | } | |
3766 | ||
961ccddd | 3767 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3768 | const struct sched_param *param, bool user) |
1da177e4 | 3769 | { |
83b699ed | 3770 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3771 | unsigned long flags; |
83ab0aa0 | 3772 | const struct sched_class *prev_class; |
70b97a7f | 3773 | struct rq *rq; |
ca94c442 | 3774 | int reset_on_fork; |
1da177e4 | 3775 | |
66e5393a SR |
3776 | /* may grab non-irq protected spin_locks */ |
3777 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3778 | recheck: |
3779 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3780 | if (policy < 0) { |
3781 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3782 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3783 | } else { |
3784 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3785 | policy &= ~SCHED_RESET_ON_FORK; | |
3786 | ||
3787 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3788 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3789 | policy != SCHED_IDLE) | |
3790 | return -EINVAL; | |
3791 | } | |
3792 | ||
1da177e4 LT |
3793 | /* |
3794 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3795 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3796 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3797 | */ |
3798 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3799 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3800 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3801 | return -EINVAL; |
e05606d3 | 3802 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3803 | return -EINVAL; |
3804 | ||
37e4ab3f OC |
3805 | /* |
3806 | * Allow unprivileged RT tasks to decrease priority: | |
3807 | */ | |
961ccddd | 3808 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3809 | if (rt_policy(policy)) { |
a44702e8 ON |
3810 | unsigned long rlim_rtprio = |
3811 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3812 | |
3813 | /* can't set/change the rt policy */ | |
3814 | if (policy != p->policy && !rlim_rtprio) | |
3815 | return -EPERM; | |
3816 | ||
3817 | /* can't increase priority */ | |
3818 | if (param->sched_priority > p->rt_priority && | |
3819 | param->sched_priority > rlim_rtprio) | |
3820 | return -EPERM; | |
3821 | } | |
c02aa73b | 3822 | |
dd41f596 | 3823 | /* |
c02aa73b DH |
3824 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3825 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3826 | */ |
c02aa73b DH |
3827 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3828 | if (!can_nice(p, TASK_NICE(p))) | |
3829 | return -EPERM; | |
3830 | } | |
5fe1d75f | 3831 | |
37e4ab3f | 3832 | /* can't change other user's priorities */ |
c69e8d9c | 3833 | if (!check_same_owner(p)) |
37e4ab3f | 3834 | return -EPERM; |
ca94c442 LP |
3835 | |
3836 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3837 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3838 | return -EPERM; | |
37e4ab3f | 3839 | } |
1da177e4 | 3840 | |
725aad24 | 3841 | if (user) { |
b0ae1981 | 3842 | retval = security_task_setscheduler(p); |
725aad24 JF |
3843 | if (retval) |
3844 | return retval; | |
3845 | } | |
3846 | ||
b29739f9 IM |
3847 | /* |
3848 | * make sure no PI-waiters arrive (or leave) while we are | |
3849 | * changing the priority of the task: | |
0122ec5b | 3850 | * |
25985edc | 3851 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3852 | * runqueue lock must be held. |
3853 | */ | |
0122ec5b | 3854 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3855 | |
34f971f6 PZ |
3856 | /* |
3857 | * Changing the policy of the stop threads its a very bad idea | |
3858 | */ | |
3859 | if (p == rq->stop) { | |
0122ec5b | 3860 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3861 | return -EINVAL; |
3862 | } | |
3863 | ||
a51e9198 DF |
3864 | /* |
3865 | * If not changing anything there's no need to proceed further: | |
3866 | */ | |
3867 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
3868 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 3869 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3870 | return 0; |
3871 | } | |
3872 | ||
dc61b1d6 PZ |
3873 | #ifdef CONFIG_RT_GROUP_SCHED |
3874 | if (user) { | |
3875 | /* | |
3876 | * Do not allow realtime tasks into groups that have no runtime | |
3877 | * assigned. | |
3878 | */ | |
3879 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3880 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3881 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3882 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3883 | return -EPERM; |
3884 | } | |
3885 | } | |
3886 | #endif | |
3887 | ||
1da177e4 LT |
3888 | /* recheck policy now with rq lock held */ |
3889 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3890 | policy = oldpolicy = -1; | |
0122ec5b | 3891 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3892 | goto recheck; |
3893 | } | |
fd2f4419 | 3894 | on_rq = p->on_rq; |
051a1d1a | 3895 | running = task_current(rq, p); |
0e1f3483 | 3896 | if (on_rq) |
4ca9b72b | 3897 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3898 | if (running) |
3899 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3900 | |
ca94c442 LP |
3901 | p->sched_reset_on_fork = reset_on_fork; |
3902 | ||
1da177e4 | 3903 | oldprio = p->prio; |
83ab0aa0 | 3904 | prev_class = p->sched_class; |
dd41f596 | 3905 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 3906 | |
0e1f3483 HS |
3907 | if (running) |
3908 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3909 | if (on_rq) |
4ca9b72b | 3910 | enqueue_task(rq, p, 0); |
cb469845 | 3911 | |
da7a735e | 3912 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3913 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3914 | |
95e02ca9 TG |
3915 | rt_mutex_adjust_pi(p); |
3916 | ||
1da177e4 LT |
3917 | return 0; |
3918 | } | |
961ccddd RR |
3919 | |
3920 | /** | |
3921 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3922 | * @p: the task in question. | |
3923 | * @policy: new policy. | |
3924 | * @param: structure containing the new RT priority. | |
3925 | * | |
3926 | * NOTE that the task may be already dead. | |
3927 | */ | |
3928 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3929 | const struct sched_param *param) |
961ccddd RR |
3930 | { |
3931 | return __sched_setscheduler(p, policy, param, true); | |
3932 | } | |
1da177e4 LT |
3933 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3934 | ||
961ccddd RR |
3935 | /** |
3936 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3937 | * @p: the task in question. | |
3938 | * @policy: new policy. | |
3939 | * @param: structure containing the new RT priority. | |
3940 | * | |
3941 | * Just like sched_setscheduler, only don't bother checking if the | |
3942 | * current context has permission. For example, this is needed in | |
3943 | * stop_machine(): we create temporary high priority worker threads, | |
3944 | * but our caller might not have that capability. | |
3945 | */ | |
3946 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3947 | const struct sched_param *param) |
961ccddd RR |
3948 | { |
3949 | return __sched_setscheduler(p, policy, param, false); | |
3950 | } | |
3951 | ||
95cdf3b7 IM |
3952 | static int |
3953 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3954 | { |
1da177e4 LT |
3955 | struct sched_param lparam; |
3956 | struct task_struct *p; | |
36c8b586 | 3957 | int retval; |
1da177e4 LT |
3958 | |
3959 | if (!param || pid < 0) | |
3960 | return -EINVAL; | |
3961 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3962 | return -EFAULT; | |
5fe1d75f ON |
3963 | |
3964 | rcu_read_lock(); | |
3965 | retval = -ESRCH; | |
1da177e4 | 3966 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3967 | if (p != NULL) |
3968 | retval = sched_setscheduler(p, policy, &lparam); | |
3969 | rcu_read_unlock(); | |
36c8b586 | 3970 | |
1da177e4 LT |
3971 | return retval; |
3972 | } | |
3973 | ||
3974 | /** | |
3975 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3976 | * @pid: the pid in question. | |
3977 | * @policy: new policy. | |
3978 | * @param: structure containing the new RT priority. | |
3979 | */ | |
5add95d4 HC |
3980 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3981 | struct sched_param __user *, param) | |
1da177e4 | 3982 | { |
c21761f1 JB |
3983 | /* negative values for policy are not valid */ |
3984 | if (policy < 0) | |
3985 | return -EINVAL; | |
3986 | ||
1da177e4 LT |
3987 | return do_sched_setscheduler(pid, policy, param); |
3988 | } | |
3989 | ||
3990 | /** | |
3991 | * sys_sched_setparam - set/change the RT priority of a thread | |
3992 | * @pid: the pid in question. | |
3993 | * @param: structure containing the new RT priority. | |
3994 | */ | |
5add95d4 | 3995 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3996 | { |
3997 | return do_sched_setscheduler(pid, -1, param); | |
3998 | } | |
3999 | ||
4000 | /** | |
4001 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4002 | * @pid: the pid in question. | |
4003 | */ | |
5add95d4 | 4004 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4005 | { |
36c8b586 | 4006 | struct task_struct *p; |
3a5c359a | 4007 | int retval; |
1da177e4 LT |
4008 | |
4009 | if (pid < 0) | |
3a5c359a | 4010 | return -EINVAL; |
1da177e4 LT |
4011 | |
4012 | retval = -ESRCH; | |
5fe85be0 | 4013 | rcu_read_lock(); |
1da177e4 LT |
4014 | p = find_process_by_pid(pid); |
4015 | if (p) { | |
4016 | retval = security_task_getscheduler(p); | |
4017 | if (!retval) | |
ca94c442 LP |
4018 | retval = p->policy |
4019 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4020 | } |
5fe85be0 | 4021 | rcu_read_unlock(); |
1da177e4 LT |
4022 | return retval; |
4023 | } | |
4024 | ||
4025 | /** | |
ca94c442 | 4026 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4027 | * @pid: the pid in question. |
4028 | * @param: structure containing the RT priority. | |
4029 | */ | |
5add95d4 | 4030 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4031 | { |
4032 | struct sched_param lp; | |
36c8b586 | 4033 | struct task_struct *p; |
3a5c359a | 4034 | int retval; |
1da177e4 LT |
4035 | |
4036 | if (!param || pid < 0) | |
3a5c359a | 4037 | return -EINVAL; |
1da177e4 | 4038 | |
5fe85be0 | 4039 | rcu_read_lock(); |
1da177e4 LT |
4040 | p = find_process_by_pid(pid); |
4041 | retval = -ESRCH; | |
4042 | if (!p) | |
4043 | goto out_unlock; | |
4044 | ||
4045 | retval = security_task_getscheduler(p); | |
4046 | if (retval) | |
4047 | goto out_unlock; | |
4048 | ||
4049 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4050 | rcu_read_unlock(); |
1da177e4 LT |
4051 | |
4052 | /* | |
4053 | * This one might sleep, we cannot do it with a spinlock held ... | |
4054 | */ | |
4055 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4056 | ||
1da177e4 LT |
4057 | return retval; |
4058 | ||
4059 | out_unlock: | |
5fe85be0 | 4060 | rcu_read_unlock(); |
1da177e4 LT |
4061 | return retval; |
4062 | } | |
4063 | ||
96f874e2 | 4064 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4065 | { |
5a16f3d3 | 4066 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4067 | struct task_struct *p; |
4068 | int retval; | |
1da177e4 | 4069 | |
95402b38 | 4070 | get_online_cpus(); |
23f5d142 | 4071 | rcu_read_lock(); |
1da177e4 LT |
4072 | |
4073 | p = find_process_by_pid(pid); | |
4074 | if (!p) { | |
23f5d142 | 4075 | rcu_read_unlock(); |
95402b38 | 4076 | put_online_cpus(); |
1da177e4 LT |
4077 | return -ESRCH; |
4078 | } | |
4079 | ||
23f5d142 | 4080 | /* Prevent p going away */ |
1da177e4 | 4081 | get_task_struct(p); |
23f5d142 | 4082 | rcu_read_unlock(); |
1da177e4 | 4083 | |
5a16f3d3 RR |
4084 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4085 | retval = -ENOMEM; | |
4086 | goto out_put_task; | |
4087 | } | |
4088 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4089 | retval = -ENOMEM; | |
4090 | goto out_free_cpus_allowed; | |
4091 | } | |
1da177e4 | 4092 | retval = -EPERM; |
4c44aaaf EB |
4093 | if (!check_same_owner(p)) { |
4094 | rcu_read_lock(); | |
4095 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4096 | rcu_read_unlock(); | |
4097 | goto out_unlock; | |
4098 | } | |
4099 | rcu_read_unlock(); | |
4100 | } | |
1da177e4 | 4101 | |
b0ae1981 | 4102 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4103 | if (retval) |
4104 | goto out_unlock; | |
4105 | ||
5a16f3d3 RR |
4106 | cpuset_cpus_allowed(p, cpus_allowed); |
4107 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4108 | again: |
5a16f3d3 | 4109 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4110 | |
8707d8b8 | 4111 | if (!retval) { |
5a16f3d3 RR |
4112 | cpuset_cpus_allowed(p, cpus_allowed); |
4113 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4114 | /* |
4115 | * We must have raced with a concurrent cpuset | |
4116 | * update. Just reset the cpus_allowed to the | |
4117 | * cpuset's cpus_allowed | |
4118 | */ | |
5a16f3d3 | 4119 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4120 | goto again; |
4121 | } | |
4122 | } | |
1da177e4 | 4123 | out_unlock: |
5a16f3d3 RR |
4124 | free_cpumask_var(new_mask); |
4125 | out_free_cpus_allowed: | |
4126 | free_cpumask_var(cpus_allowed); | |
4127 | out_put_task: | |
1da177e4 | 4128 | put_task_struct(p); |
95402b38 | 4129 | put_online_cpus(); |
1da177e4 LT |
4130 | return retval; |
4131 | } | |
4132 | ||
4133 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4134 | struct cpumask *new_mask) |
1da177e4 | 4135 | { |
96f874e2 RR |
4136 | if (len < cpumask_size()) |
4137 | cpumask_clear(new_mask); | |
4138 | else if (len > cpumask_size()) | |
4139 | len = cpumask_size(); | |
4140 | ||
1da177e4 LT |
4141 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4142 | } | |
4143 | ||
4144 | /** | |
4145 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4146 | * @pid: pid of the process | |
4147 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4148 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4149 | */ | |
5add95d4 HC |
4150 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4151 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4152 | { |
5a16f3d3 | 4153 | cpumask_var_t new_mask; |
1da177e4 LT |
4154 | int retval; |
4155 | ||
5a16f3d3 RR |
4156 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4157 | return -ENOMEM; | |
1da177e4 | 4158 | |
5a16f3d3 RR |
4159 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4160 | if (retval == 0) | |
4161 | retval = sched_setaffinity(pid, new_mask); | |
4162 | free_cpumask_var(new_mask); | |
4163 | return retval; | |
1da177e4 LT |
4164 | } |
4165 | ||
96f874e2 | 4166 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4167 | { |
36c8b586 | 4168 | struct task_struct *p; |
31605683 | 4169 | unsigned long flags; |
1da177e4 | 4170 | int retval; |
1da177e4 | 4171 | |
95402b38 | 4172 | get_online_cpus(); |
23f5d142 | 4173 | rcu_read_lock(); |
1da177e4 LT |
4174 | |
4175 | retval = -ESRCH; | |
4176 | p = find_process_by_pid(pid); | |
4177 | if (!p) | |
4178 | goto out_unlock; | |
4179 | ||
e7834f8f DQ |
4180 | retval = security_task_getscheduler(p); |
4181 | if (retval) | |
4182 | goto out_unlock; | |
4183 | ||
013fdb80 | 4184 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4185 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4186 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4187 | |
4188 | out_unlock: | |
23f5d142 | 4189 | rcu_read_unlock(); |
95402b38 | 4190 | put_online_cpus(); |
1da177e4 | 4191 | |
9531b62f | 4192 | return retval; |
1da177e4 LT |
4193 | } |
4194 | ||
4195 | /** | |
4196 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4197 | * @pid: pid of the process | |
4198 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4199 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4200 | */ | |
5add95d4 HC |
4201 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4202 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4203 | { |
4204 | int ret; | |
f17c8607 | 4205 | cpumask_var_t mask; |
1da177e4 | 4206 | |
84fba5ec | 4207 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4208 | return -EINVAL; |
4209 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4210 | return -EINVAL; |
4211 | ||
f17c8607 RR |
4212 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4213 | return -ENOMEM; | |
1da177e4 | 4214 | |
f17c8607 RR |
4215 | ret = sched_getaffinity(pid, mask); |
4216 | if (ret == 0) { | |
8bc037fb | 4217 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4218 | |
4219 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4220 | ret = -EFAULT; |
4221 | else | |
cd3d8031 | 4222 | ret = retlen; |
f17c8607 RR |
4223 | } |
4224 | free_cpumask_var(mask); | |
1da177e4 | 4225 | |
f17c8607 | 4226 | return ret; |
1da177e4 LT |
4227 | } |
4228 | ||
4229 | /** | |
4230 | * sys_sched_yield - yield the current processor to other threads. | |
4231 | * | |
dd41f596 IM |
4232 | * This function yields the current CPU to other tasks. If there are no |
4233 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4234 | */ |
5add95d4 | 4235 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4236 | { |
70b97a7f | 4237 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4238 | |
2d72376b | 4239 | schedstat_inc(rq, yld_count); |
4530d7ab | 4240 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4241 | |
4242 | /* | |
4243 | * Since we are going to call schedule() anyway, there's | |
4244 | * no need to preempt or enable interrupts: | |
4245 | */ | |
4246 | __release(rq->lock); | |
8a25d5de | 4247 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4248 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4249 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4250 | |
4251 | schedule(); | |
4252 | ||
4253 | return 0; | |
4254 | } | |
4255 | ||
d86ee480 PZ |
4256 | static inline int should_resched(void) |
4257 | { | |
4258 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4259 | } | |
4260 | ||
e7b38404 | 4261 | static void __cond_resched(void) |
1da177e4 | 4262 | { |
e7aaaa69 | 4263 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4264 | __schedule(); |
e7aaaa69 | 4265 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4266 | } |
4267 | ||
02b67cc3 | 4268 | int __sched _cond_resched(void) |
1da177e4 | 4269 | { |
d86ee480 | 4270 | if (should_resched()) { |
1da177e4 LT |
4271 | __cond_resched(); |
4272 | return 1; | |
4273 | } | |
4274 | return 0; | |
4275 | } | |
02b67cc3 | 4276 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4277 | |
4278 | /* | |
613afbf8 | 4279 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4280 | * call schedule, and on return reacquire the lock. |
4281 | * | |
41a2d6cf | 4282 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4283 | * operations here to prevent schedule() from being called twice (once via |
4284 | * spin_unlock(), once by hand). | |
4285 | */ | |
613afbf8 | 4286 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4287 | { |
d86ee480 | 4288 | int resched = should_resched(); |
6df3cecb JK |
4289 | int ret = 0; |
4290 | ||
f607c668 PZ |
4291 | lockdep_assert_held(lock); |
4292 | ||
95c354fe | 4293 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4294 | spin_unlock(lock); |
d86ee480 | 4295 | if (resched) |
95c354fe NP |
4296 | __cond_resched(); |
4297 | else | |
4298 | cpu_relax(); | |
6df3cecb | 4299 | ret = 1; |
1da177e4 | 4300 | spin_lock(lock); |
1da177e4 | 4301 | } |
6df3cecb | 4302 | return ret; |
1da177e4 | 4303 | } |
613afbf8 | 4304 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4305 | |
613afbf8 | 4306 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4307 | { |
4308 | BUG_ON(!in_softirq()); | |
4309 | ||
d86ee480 | 4310 | if (should_resched()) { |
98d82567 | 4311 | local_bh_enable(); |
1da177e4 LT |
4312 | __cond_resched(); |
4313 | local_bh_disable(); | |
4314 | return 1; | |
4315 | } | |
4316 | return 0; | |
4317 | } | |
613afbf8 | 4318 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4319 | |
1da177e4 LT |
4320 | /** |
4321 | * yield - yield the current processor to other threads. | |
4322 | * | |
8e3fabfd PZ |
4323 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4324 | * | |
4325 | * The scheduler is at all times free to pick the calling task as the most | |
4326 | * eligible task to run, if removing the yield() call from your code breaks | |
4327 | * it, its already broken. | |
4328 | * | |
4329 | * Typical broken usage is: | |
4330 | * | |
4331 | * while (!event) | |
4332 | * yield(); | |
4333 | * | |
4334 | * where one assumes that yield() will let 'the other' process run that will | |
4335 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4336 | * happen. Never use yield() as a progress guarantee!! | |
4337 | * | |
4338 | * If you want to use yield() to wait for something, use wait_event(). | |
4339 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4340 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4341 | */ |
4342 | void __sched yield(void) | |
4343 | { | |
4344 | set_current_state(TASK_RUNNING); | |
4345 | sys_sched_yield(); | |
4346 | } | |
1da177e4 LT |
4347 | EXPORT_SYMBOL(yield); |
4348 | ||
d95f4122 MG |
4349 | /** |
4350 | * yield_to - yield the current processor to another thread in | |
4351 | * your thread group, or accelerate that thread toward the | |
4352 | * processor it's on. | |
16addf95 RD |
4353 | * @p: target task |
4354 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4355 | * |
4356 | * It's the caller's job to ensure that the target task struct | |
4357 | * can't go away on us before we can do any checks. | |
4358 | * | |
7b270f60 PZ |
4359 | * Returns: |
4360 | * true (>0) if we indeed boosted the target task. | |
4361 | * false (0) if we failed to boost the target. | |
4362 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4363 | */ |
4364 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4365 | { | |
4366 | struct task_struct *curr = current; | |
4367 | struct rq *rq, *p_rq; | |
4368 | unsigned long flags; | |
c3c18640 | 4369 | int yielded = 0; |
d95f4122 MG |
4370 | |
4371 | local_irq_save(flags); | |
4372 | rq = this_rq(); | |
4373 | ||
4374 | again: | |
4375 | p_rq = task_rq(p); | |
7b270f60 PZ |
4376 | /* |
4377 | * If we're the only runnable task on the rq and target rq also | |
4378 | * has only one task, there's absolutely no point in yielding. | |
4379 | */ | |
4380 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4381 | yielded = -ESRCH; | |
4382 | goto out_irq; | |
4383 | } | |
4384 | ||
d95f4122 MG |
4385 | double_rq_lock(rq, p_rq); |
4386 | while (task_rq(p) != p_rq) { | |
4387 | double_rq_unlock(rq, p_rq); | |
4388 | goto again; | |
4389 | } | |
4390 | ||
4391 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4392 | goto out_unlock; |
d95f4122 MG |
4393 | |
4394 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4395 | goto out_unlock; |
d95f4122 MG |
4396 | |
4397 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4398 | goto out_unlock; |
d95f4122 MG |
4399 | |
4400 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4401 | if (yielded) { |
d95f4122 | 4402 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4403 | /* |
4404 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4405 | * fairness. | |
4406 | */ | |
4407 | if (preempt && rq != p_rq) | |
4408 | resched_task(p_rq->curr); | |
4409 | } | |
d95f4122 | 4410 | |
7b270f60 | 4411 | out_unlock: |
d95f4122 | 4412 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4413 | out_irq: |
d95f4122 MG |
4414 | local_irq_restore(flags); |
4415 | ||
7b270f60 | 4416 | if (yielded > 0) |
d95f4122 MG |
4417 | schedule(); |
4418 | ||
4419 | return yielded; | |
4420 | } | |
4421 | EXPORT_SYMBOL_GPL(yield_to); | |
4422 | ||
1da177e4 | 4423 | /* |
41a2d6cf | 4424 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4425 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4426 | */ |
4427 | void __sched io_schedule(void) | |
4428 | { | |
54d35f29 | 4429 | struct rq *rq = raw_rq(); |
1da177e4 | 4430 | |
0ff92245 | 4431 | delayacct_blkio_start(); |
1da177e4 | 4432 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4433 | blk_flush_plug(current); |
8f0dfc34 | 4434 | current->in_iowait = 1; |
1da177e4 | 4435 | schedule(); |
8f0dfc34 | 4436 | current->in_iowait = 0; |
1da177e4 | 4437 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4438 | delayacct_blkio_end(); |
1da177e4 | 4439 | } |
1da177e4 LT |
4440 | EXPORT_SYMBOL(io_schedule); |
4441 | ||
4442 | long __sched io_schedule_timeout(long timeout) | |
4443 | { | |
54d35f29 | 4444 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4445 | long ret; |
4446 | ||
0ff92245 | 4447 | delayacct_blkio_start(); |
1da177e4 | 4448 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4449 | blk_flush_plug(current); |
8f0dfc34 | 4450 | current->in_iowait = 1; |
1da177e4 | 4451 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4452 | current->in_iowait = 0; |
1da177e4 | 4453 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4454 | delayacct_blkio_end(); |
1da177e4 LT |
4455 | return ret; |
4456 | } | |
4457 | ||
4458 | /** | |
4459 | * sys_sched_get_priority_max - return maximum RT priority. | |
4460 | * @policy: scheduling class. | |
4461 | * | |
4462 | * this syscall returns the maximum rt_priority that can be used | |
4463 | * by a given scheduling class. | |
4464 | */ | |
5add95d4 | 4465 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4466 | { |
4467 | int ret = -EINVAL; | |
4468 | ||
4469 | switch (policy) { | |
4470 | case SCHED_FIFO: | |
4471 | case SCHED_RR: | |
4472 | ret = MAX_USER_RT_PRIO-1; | |
4473 | break; | |
4474 | case SCHED_NORMAL: | |
b0a9499c | 4475 | case SCHED_BATCH: |
dd41f596 | 4476 | case SCHED_IDLE: |
1da177e4 LT |
4477 | ret = 0; |
4478 | break; | |
4479 | } | |
4480 | return ret; | |
4481 | } | |
4482 | ||
4483 | /** | |
4484 | * sys_sched_get_priority_min - return minimum RT priority. | |
4485 | * @policy: scheduling class. | |
4486 | * | |
4487 | * this syscall returns the minimum rt_priority that can be used | |
4488 | * by a given scheduling class. | |
4489 | */ | |
5add95d4 | 4490 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4491 | { |
4492 | int ret = -EINVAL; | |
4493 | ||
4494 | switch (policy) { | |
4495 | case SCHED_FIFO: | |
4496 | case SCHED_RR: | |
4497 | ret = 1; | |
4498 | break; | |
4499 | case SCHED_NORMAL: | |
b0a9499c | 4500 | case SCHED_BATCH: |
dd41f596 | 4501 | case SCHED_IDLE: |
1da177e4 LT |
4502 | ret = 0; |
4503 | } | |
4504 | return ret; | |
4505 | } | |
4506 | ||
4507 | /** | |
4508 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4509 | * @pid: pid of the process. | |
4510 | * @interval: userspace pointer to the timeslice value. | |
4511 | * | |
4512 | * this syscall writes the default timeslice value of a given process | |
4513 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4514 | */ | |
17da2bd9 | 4515 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4516 | struct timespec __user *, interval) |
1da177e4 | 4517 | { |
36c8b586 | 4518 | struct task_struct *p; |
a4ec24b4 | 4519 | unsigned int time_slice; |
dba091b9 TG |
4520 | unsigned long flags; |
4521 | struct rq *rq; | |
3a5c359a | 4522 | int retval; |
1da177e4 | 4523 | struct timespec t; |
1da177e4 LT |
4524 | |
4525 | if (pid < 0) | |
3a5c359a | 4526 | return -EINVAL; |
1da177e4 LT |
4527 | |
4528 | retval = -ESRCH; | |
1a551ae7 | 4529 | rcu_read_lock(); |
1da177e4 LT |
4530 | p = find_process_by_pid(pid); |
4531 | if (!p) | |
4532 | goto out_unlock; | |
4533 | ||
4534 | retval = security_task_getscheduler(p); | |
4535 | if (retval) | |
4536 | goto out_unlock; | |
4537 | ||
dba091b9 TG |
4538 | rq = task_rq_lock(p, &flags); |
4539 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4540 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4541 | |
1a551ae7 | 4542 | rcu_read_unlock(); |
a4ec24b4 | 4543 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4544 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4545 | return retval; |
3a5c359a | 4546 | |
1da177e4 | 4547 | out_unlock: |
1a551ae7 | 4548 | rcu_read_unlock(); |
1da177e4 LT |
4549 | return retval; |
4550 | } | |
4551 | ||
7c731e0a | 4552 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4553 | |
82a1fcb9 | 4554 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4555 | { |
1da177e4 | 4556 | unsigned long free = 0; |
4e79752c | 4557 | int ppid; |
36c8b586 | 4558 | unsigned state; |
1da177e4 | 4559 | |
1da177e4 | 4560 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4561 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4562 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4563 | #if BITS_PER_LONG == 32 |
1da177e4 | 4564 | if (state == TASK_RUNNING) |
3df0fc5b | 4565 | printk(KERN_CONT " running "); |
1da177e4 | 4566 | else |
3df0fc5b | 4567 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4568 | #else |
4569 | if (state == TASK_RUNNING) | |
3df0fc5b | 4570 | printk(KERN_CONT " running task "); |
1da177e4 | 4571 | else |
3df0fc5b | 4572 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4573 | #endif |
4574 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4575 | free = stack_not_used(p); |
1da177e4 | 4576 | #endif |
4e79752c PM |
4577 | rcu_read_lock(); |
4578 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4579 | rcu_read_unlock(); | |
3df0fc5b | 4580 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4581 | task_pid_nr(p), ppid, |
aa47b7e0 | 4582 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4583 | |
5fb5e6de | 4584 | show_stack(p, NULL); |
1da177e4 LT |
4585 | } |
4586 | ||
e59e2ae2 | 4587 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4588 | { |
36c8b586 | 4589 | struct task_struct *g, *p; |
1da177e4 | 4590 | |
4bd77321 | 4591 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4592 | printk(KERN_INFO |
4593 | " task PC stack pid father\n"); | |
1da177e4 | 4594 | #else |
3df0fc5b PZ |
4595 | printk(KERN_INFO |
4596 | " task PC stack pid father\n"); | |
1da177e4 | 4597 | #endif |
510f5acc | 4598 | rcu_read_lock(); |
1da177e4 LT |
4599 | do_each_thread(g, p) { |
4600 | /* | |
4601 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4602 | * console might take a lot of time: |
1da177e4 LT |
4603 | */ |
4604 | touch_nmi_watchdog(); | |
39bc89fd | 4605 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4606 | sched_show_task(p); |
1da177e4 LT |
4607 | } while_each_thread(g, p); |
4608 | ||
04c9167f JF |
4609 | touch_all_softlockup_watchdogs(); |
4610 | ||
dd41f596 IM |
4611 | #ifdef CONFIG_SCHED_DEBUG |
4612 | sysrq_sched_debug_show(); | |
4613 | #endif | |
510f5acc | 4614 | rcu_read_unlock(); |
e59e2ae2 IM |
4615 | /* |
4616 | * Only show locks if all tasks are dumped: | |
4617 | */ | |
93335a21 | 4618 | if (!state_filter) |
e59e2ae2 | 4619 | debug_show_all_locks(); |
1da177e4 LT |
4620 | } |
4621 | ||
1df21055 IM |
4622 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4623 | { | |
dd41f596 | 4624 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4625 | } |
4626 | ||
f340c0d1 IM |
4627 | /** |
4628 | * init_idle - set up an idle thread for a given CPU | |
4629 | * @idle: task in question | |
4630 | * @cpu: cpu the idle task belongs to | |
4631 | * | |
4632 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4633 | * flag, to make booting more robust. | |
4634 | */ | |
5c1e1767 | 4635 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4636 | { |
70b97a7f | 4637 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4638 | unsigned long flags; |
4639 | ||
05fa785c | 4640 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4641 | |
dd41f596 | 4642 | __sched_fork(idle); |
06b83b5f | 4643 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4644 | idle->se.exec_start = sched_clock(); |
4645 | ||
1e1b6c51 | 4646 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4647 | /* |
4648 | * We're having a chicken and egg problem, even though we are | |
4649 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4650 | * lockdep check in task_group() will fail. | |
4651 | * | |
4652 | * Similar case to sched_fork(). / Alternatively we could | |
4653 | * use task_rq_lock() here and obtain the other rq->lock. | |
4654 | * | |
4655 | * Silence PROVE_RCU | |
4656 | */ | |
4657 | rcu_read_lock(); | |
dd41f596 | 4658 | __set_task_cpu(idle, cpu); |
6506cf6c | 4659 | rcu_read_unlock(); |
1da177e4 | 4660 | |
1da177e4 | 4661 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4662 | #if defined(CONFIG_SMP) |
4663 | idle->on_cpu = 1; | |
4866cde0 | 4664 | #endif |
05fa785c | 4665 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4666 | |
4667 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4668 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4669 | |
dd41f596 IM |
4670 | /* |
4671 | * The idle tasks have their own, simple scheduling class: | |
4672 | */ | |
4673 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4674 | ftrace_graph_init_idle_task(idle, cpu); |
6a61671b | 4675 | vtime_init_idle(idle); |
f1c6f1a7 CE |
4676 | #if defined(CONFIG_SMP) |
4677 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4678 | #endif | |
19978ca6 IM |
4679 | } |
4680 | ||
1da177e4 | 4681 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4682 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4683 | { | |
4684 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4685 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4686 | |
4687 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4688 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4689 | } |
4690 | ||
1da177e4 LT |
4691 | /* |
4692 | * This is how migration works: | |
4693 | * | |
969c7921 TH |
4694 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4695 | * stop_one_cpu(). | |
4696 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4697 | * off the CPU) | |
4698 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4699 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4700 | * it and puts it into the right queue. |
969c7921 TH |
4701 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4702 | * is done. | |
1da177e4 LT |
4703 | */ |
4704 | ||
4705 | /* | |
4706 | * Change a given task's CPU affinity. Migrate the thread to a | |
4707 | * proper CPU and schedule it away if the CPU it's executing on | |
4708 | * is removed from the allowed bitmask. | |
4709 | * | |
4710 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4711 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4712 | * call is not atomic; no spinlocks may be held. |
4713 | */ | |
96f874e2 | 4714 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4715 | { |
4716 | unsigned long flags; | |
70b97a7f | 4717 | struct rq *rq; |
969c7921 | 4718 | unsigned int dest_cpu; |
48f24c4d | 4719 | int ret = 0; |
1da177e4 LT |
4720 | |
4721 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4722 | |
db44fc01 YZ |
4723 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4724 | goto out; | |
4725 | ||
6ad4c188 | 4726 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4727 | ret = -EINVAL; |
4728 | goto out; | |
4729 | } | |
4730 | ||
db44fc01 | 4731 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4732 | ret = -EINVAL; |
4733 | goto out; | |
4734 | } | |
4735 | ||
1e1b6c51 | 4736 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4737 | |
1da177e4 | 4738 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4739 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4740 | goto out; |
4741 | ||
969c7921 | 4742 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4743 | if (p->on_rq) { |
969c7921 | 4744 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4745 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4746 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4747 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4748 | tlb_migrate_finish(p->mm); |
4749 | return 0; | |
4750 | } | |
4751 | out: | |
0122ec5b | 4752 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4753 | |
1da177e4 LT |
4754 | return ret; |
4755 | } | |
cd8ba7cd | 4756 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4757 | |
4758 | /* | |
41a2d6cf | 4759 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4760 | * this because either it can't run here any more (set_cpus_allowed() |
4761 | * away from this CPU, or CPU going down), or because we're | |
4762 | * attempting to rebalance this task on exec (sched_exec). | |
4763 | * | |
4764 | * So we race with normal scheduler movements, but that's OK, as long | |
4765 | * as the task is no longer on this CPU. | |
efc30814 KK |
4766 | * |
4767 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4768 | */ |
efc30814 | 4769 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4770 | { |
70b97a7f | 4771 | struct rq *rq_dest, *rq_src; |
e2912009 | 4772 | int ret = 0; |
1da177e4 | 4773 | |
e761b772 | 4774 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4775 | return ret; |
1da177e4 LT |
4776 | |
4777 | rq_src = cpu_rq(src_cpu); | |
4778 | rq_dest = cpu_rq(dest_cpu); | |
4779 | ||
0122ec5b | 4780 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4781 | double_rq_lock(rq_src, rq_dest); |
4782 | /* Already moved. */ | |
4783 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4784 | goto done; |
1da177e4 | 4785 | /* Affinity changed (again). */ |
fa17b507 | 4786 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4787 | goto fail; |
1da177e4 | 4788 | |
e2912009 PZ |
4789 | /* |
4790 | * If we're not on a rq, the next wake-up will ensure we're | |
4791 | * placed properly. | |
4792 | */ | |
fd2f4419 | 4793 | if (p->on_rq) { |
4ca9b72b | 4794 | dequeue_task(rq_src, p, 0); |
e2912009 | 4795 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4796 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4797 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4798 | } |
b1e38734 | 4799 | done: |
efc30814 | 4800 | ret = 1; |
b1e38734 | 4801 | fail: |
1da177e4 | 4802 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4803 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4804 | return ret; |
1da177e4 LT |
4805 | } |
4806 | ||
4807 | /* | |
969c7921 TH |
4808 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4809 | * and performs thread migration by bumping thread off CPU then | |
4810 | * 'pushing' onto another runqueue. | |
1da177e4 | 4811 | */ |
969c7921 | 4812 | static int migration_cpu_stop(void *data) |
1da177e4 | 4813 | { |
969c7921 | 4814 | struct migration_arg *arg = data; |
f7b4cddc | 4815 | |
969c7921 TH |
4816 | /* |
4817 | * The original target cpu might have gone down and we might | |
4818 | * be on another cpu but it doesn't matter. | |
4819 | */ | |
f7b4cddc | 4820 | local_irq_disable(); |
969c7921 | 4821 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4822 | local_irq_enable(); |
1da177e4 | 4823 | return 0; |
f7b4cddc ON |
4824 | } |
4825 | ||
1da177e4 | 4826 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4827 | |
054b9108 | 4828 | /* |
48c5ccae PZ |
4829 | * Ensures that the idle task is using init_mm right before its cpu goes |
4830 | * offline. | |
054b9108 | 4831 | */ |
48c5ccae | 4832 | void idle_task_exit(void) |
1da177e4 | 4833 | { |
48c5ccae | 4834 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4835 | |
48c5ccae | 4836 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4837 | |
48c5ccae PZ |
4838 | if (mm != &init_mm) |
4839 | switch_mm(mm, &init_mm, current); | |
4840 | mmdrop(mm); | |
1da177e4 LT |
4841 | } |
4842 | ||
4843 | /* | |
5d180232 PZ |
4844 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4845 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4846 | * nr_active count is stable. | |
4847 | * | |
4848 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4849 | */ |
5d180232 | 4850 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4851 | { |
5d180232 PZ |
4852 | long delta = calc_load_fold_active(rq); |
4853 | if (delta) | |
4854 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4855 | } |
4856 | ||
48f24c4d | 4857 | /* |
48c5ccae PZ |
4858 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4859 | * try_to_wake_up()->select_task_rq(). | |
4860 | * | |
4861 | * Called with rq->lock held even though we'er in stop_machine() and | |
4862 | * there's no concurrency possible, we hold the required locks anyway | |
4863 | * because of lock validation efforts. | |
1da177e4 | 4864 | */ |
48c5ccae | 4865 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4866 | { |
70b97a7f | 4867 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4868 | struct task_struct *next, *stop = rq->stop; |
4869 | int dest_cpu; | |
1da177e4 LT |
4870 | |
4871 | /* | |
48c5ccae PZ |
4872 | * Fudge the rq selection such that the below task selection loop |
4873 | * doesn't get stuck on the currently eligible stop task. | |
4874 | * | |
4875 | * We're currently inside stop_machine() and the rq is either stuck | |
4876 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4877 | * either way we should never end up calling schedule() until we're | |
4878 | * done here. | |
1da177e4 | 4879 | */ |
48c5ccae | 4880 | rq->stop = NULL; |
48f24c4d | 4881 | |
dd41f596 | 4882 | for ( ; ; ) { |
48c5ccae PZ |
4883 | /* |
4884 | * There's this thread running, bail when that's the only | |
4885 | * remaining thread. | |
4886 | */ | |
4887 | if (rq->nr_running == 1) | |
dd41f596 | 4888 | break; |
48c5ccae | 4889 | |
b67802ea | 4890 | next = pick_next_task(rq); |
48c5ccae | 4891 | BUG_ON(!next); |
79c53799 | 4892 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4893 | |
48c5ccae PZ |
4894 | /* Find suitable destination for @next, with force if needed. */ |
4895 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4896 | raw_spin_unlock(&rq->lock); | |
4897 | ||
4898 | __migrate_task(next, dead_cpu, dest_cpu); | |
4899 | ||
4900 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4901 | } |
dce48a84 | 4902 | |
48c5ccae | 4903 | rq->stop = stop; |
dce48a84 | 4904 | } |
48c5ccae | 4905 | |
1da177e4 LT |
4906 | #endif /* CONFIG_HOTPLUG_CPU */ |
4907 | ||
e692ab53 NP |
4908 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4909 | ||
4910 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4911 | { |
4912 | .procname = "sched_domain", | |
c57baf1e | 4913 | .mode = 0555, |
e0361851 | 4914 | }, |
56992309 | 4915 | {} |
e692ab53 NP |
4916 | }; |
4917 | ||
4918 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4919 | { |
4920 | .procname = "kernel", | |
c57baf1e | 4921 | .mode = 0555, |
e0361851 AD |
4922 | .child = sd_ctl_dir, |
4923 | }, | |
56992309 | 4924 | {} |
e692ab53 NP |
4925 | }; |
4926 | ||
4927 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4928 | { | |
4929 | struct ctl_table *entry = | |
5cf9f062 | 4930 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4931 | |
e692ab53 NP |
4932 | return entry; |
4933 | } | |
4934 | ||
6382bc90 MM |
4935 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4936 | { | |
cd790076 | 4937 | struct ctl_table *entry; |
6382bc90 | 4938 | |
cd790076 MM |
4939 | /* |
4940 | * In the intermediate directories, both the child directory and | |
4941 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4942 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4943 | * static strings and all have proc handlers. |
4944 | */ | |
4945 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4946 | if (entry->child) |
4947 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4948 | if (entry->proc_handler == NULL) |
4949 | kfree(entry->procname); | |
4950 | } | |
6382bc90 MM |
4951 | |
4952 | kfree(*tablep); | |
4953 | *tablep = NULL; | |
4954 | } | |
4955 | ||
201c373e NK |
4956 | static int min_load_idx = 0; |
4957 | static int max_load_idx = CPU_LOAD_IDX_MAX; | |
4958 | ||
e692ab53 | 4959 | static void |
e0361851 | 4960 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4961 | const char *procname, void *data, int maxlen, |
201c373e NK |
4962 | umode_t mode, proc_handler *proc_handler, |
4963 | bool load_idx) | |
e692ab53 | 4964 | { |
e692ab53 NP |
4965 | entry->procname = procname; |
4966 | entry->data = data; | |
4967 | entry->maxlen = maxlen; | |
4968 | entry->mode = mode; | |
4969 | entry->proc_handler = proc_handler; | |
201c373e NK |
4970 | |
4971 | if (load_idx) { | |
4972 | entry->extra1 = &min_load_idx; | |
4973 | entry->extra2 = &max_load_idx; | |
4974 | } | |
e692ab53 NP |
4975 | } |
4976 | ||
4977 | static struct ctl_table * | |
4978 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4979 | { | |
a5d8c348 | 4980 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 4981 | |
ad1cdc1d MM |
4982 | if (table == NULL) |
4983 | return NULL; | |
4984 | ||
e0361851 | 4985 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4986 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4987 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4988 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4989 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4990 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4991 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4992 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4993 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4994 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4995 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4996 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4997 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4998 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4999 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 5000 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 5001 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 5002 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5003 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 5004 | &sd->cache_nice_tries, |
201c373e | 5005 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 5006 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 5007 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 5008 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 5009 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 5010 | /* &table[12] is terminator */ |
e692ab53 NP |
5011 | |
5012 | return table; | |
5013 | } | |
5014 | ||
9a4e7159 | 5015 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5016 | { |
5017 | struct ctl_table *entry, *table; | |
5018 | struct sched_domain *sd; | |
5019 | int domain_num = 0, i; | |
5020 | char buf[32]; | |
5021 | ||
5022 | for_each_domain(cpu, sd) | |
5023 | domain_num++; | |
5024 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5025 | if (table == NULL) |
5026 | return NULL; | |
e692ab53 NP |
5027 | |
5028 | i = 0; | |
5029 | for_each_domain(cpu, sd) { | |
5030 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5031 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5032 | entry->mode = 0555; |
e692ab53 NP |
5033 | entry->child = sd_alloc_ctl_domain_table(sd); |
5034 | entry++; | |
5035 | i++; | |
5036 | } | |
5037 | return table; | |
5038 | } | |
5039 | ||
5040 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5041 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5042 | { |
6ad4c188 | 5043 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5044 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5045 | char buf[32]; | |
5046 | ||
7378547f MM |
5047 | WARN_ON(sd_ctl_dir[0].child); |
5048 | sd_ctl_dir[0].child = entry; | |
5049 | ||
ad1cdc1d MM |
5050 | if (entry == NULL) |
5051 | return; | |
5052 | ||
6ad4c188 | 5053 | for_each_possible_cpu(i) { |
e692ab53 | 5054 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5055 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5056 | entry->mode = 0555; |
e692ab53 | 5057 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5058 | entry++; |
e692ab53 | 5059 | } |
7378547f MM |
5060 | |
5061 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5062 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5063 | } | |
6382bc90 | 5064 | |
7378547f | 5065 | /* may be called multiple times per register */ |
6382bc90 MM |
5066 | static void unregister_sched_domain_sysctl(void) |
5067 | { | |
7378547f MM |
5068 | if (sd_sysctl_header) |
5069 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5070 | sd_sysctl_header = NULL; |
7378547f MM |
5071 | if (sd_ctl_dir[0].child) |
5072 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5073 | } |
e692ab53 | 5074 | #else |
6382bc90 MM |
5075 | static void register_sched_domain_sysctl(void) |
5076 | { | |
5077 | } | |
5078 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5079 | { |
5080 | } | |
5081 | #endif | |
5082 | ||
1f11eb6a GH |
5083 | static void set_rq_online(struct rq *rq) |
5084 | { | |
5085 | if (!rq->online) { | |
5086 | const struct sched_class *class; | |
5087 | ||
c6c4927b | 5088 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5089 | rq->online = 1; |
5090 | ||
5091 | for_each_class(class) { | |
5092 | if (class->rq_online) | |
5093 | class->rq_online(rq); | |
5094 | } | |
5095 | } | |
5096 | } | |
5097 | ||
5098 | static void set_rq_offline(struct rq *rq) | |
5099 | { | |
5100 | if (rq->online) { | |
5101 | const struct sched_class *class; | |
5102 | ||
5103 | for_each_class(class) { | |
5104 | if (class->rq_offline) | |
5105 | class->rq_offline(rq); | |
5106 | } | |
5107 | ||
c6c4927b | 5108 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5109 | rq->online = 0; |
5110 | } | |
5111 | } | |
5112 | ||
1da177e4 LT |
5113 | /* |
5114 | * migration_call - callback that gets triggered when a CPU is added. | |
5115 | * Here we can start up the necessary migration thread for the new CPU. | |
5116 | */ | |
48f24c4d IM |
5117 | static int __cpuinit |
5118 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5119 | { |
48f24c4d | 5120 | int cpu = (long)hcpu; |
1da177e4 | 5121 | unsigned long flags; |
969c7921 | 5122 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5123 | |
48c5ccae | 5124 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5125 | |
1da177e4 | 5126 | case CPU_UP_PREPARE: |
a468d389 | 5127 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5128 | break; |
48f24c4d | 5129 | |
1da177e4 | 5130 | case CPU_ONLINE: |
1f94ef59 | 5131 | /* Update our root-domain */ |
05fa785c | 5132 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5133 | if (rq->rd) { |
c6c4927b | 5134 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5135 | |
5136 | set_rq_online(rq); | |
1f94ef59 | 5137 | } |
05fa785c | 5138 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5139 | break; |
48f24c4d | 5140 | |
1da177e4 | 5141 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5142 | case CPU_DYING: |
317f3941 | 5143 | sched_ttwu_pending(); |
57d885fe | 5144 | /* Update our root-domain */ |
05fa785c | 5145 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5146 | if (rq->rd) { |
c6c4927b | 5147 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5148 | set_rq_offline(rq); |
57d885fe | 5149 | } |
48c5ccae PZ |
5150 | migrate_tasks(cpu); |
5151 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5152 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5153 | break; |
48c5ccae | 5154 | |
5d180232 | 5155 | case CPU_DEAD: |
f319da0c | 5156 | calc_load_migrate(rq); |
57d885fe | 5157 | break; |
1da177e4 LT |
5158 | #endif |
5159 | } | |
49c022e6 PZ |
5160 | |
5161 | update_max_interval(); | |
5162 | ||
1da177e4 LT |
5163 | return NOTIFY_OK; |
5164 | } | |
5165 | ||
f38b0820 PM |
5166 | /* |
5167 | * Register at high priority so that task migration (migrate_all_tasks) | |
5168 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5169 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5170 | */ |
26c2143b | 5171 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5172 | .notifier_call = migration_call, |
50a323b7 | 5173 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5174 | }; |
5175 | ||
3a101d05 TH |
5176 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5177 | unsigned long action, void *hcpu) | |
5178 | { | |
5179 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5180 | case CPU_STARTING: |
3a101d05 TH |
5181 | case CPU_DOWN_FAILED: |
5182 | set_cpu_active((long)hcpu, true); | |
5183 | return NOTIFY_OK; | |
5184 | default: | |
5185 | return NOTIFY_DONE; | |
5186 | } | |
5187 | } | |
5188 | ||
5189 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5190 | unsigned long action, void *hcpu) | |
5191 | { | |
5192 | switch (action & ~CPU_TASKS_FROZEN) { | |
5193 | case CPU_DOWN_PREPARE: | |
5194 | set_cpu_active((long)hcpu, false); | |
5195 | return NOTIFY_OK; | |
5196 | default: | |
5197 | return NOTIFY_DONE; | |
5198 | } | |
5199 | } | |
5200 | ||
7babe8db | 5201 | static int __init migration_init(void) |
1da177e4 LT |
5202 | { |
5203 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5204 | int err; |
48f24c4d | 5205 | |
3a101d05 | 5206 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5207 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5208 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5209 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5210 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5211 | |
3a101d05 TH |
5212 | /* Register cpu active notifiers */ |
5213 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5214 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5215 | ||
a004cd42 | 5216 | return 0; |
1da177e4 | 5217 | } |
7babe8db | 5218 | early_initcall(migration_init); |
1da177e4 LT |
5219 | #endif |
5220 | ||
5221 | #ifdef CONFIG_SMP | |
476f3534 | 5222 | |
4cb98839 PZ |
5223 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5224 | ||
3e9830dc | 5225 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5226 | |
d039ac60 | 5227 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5228 | |
d039ac60 | 5229 | static int __init sched_debug_setup(char *str) |
f6630114 | 5230 | { |
d039ac60 | 5231 | sched_debug_enabled = 1; |
f6630114 MT |
5232 | |
5233 | return 0; | |
5234 | } | |
d039ac60 PZ |
5235 | early_param("sched_debug", sched_debug_setup); |
5236 | ||
5237 | static inline bool sched_debug(void) | |
5238 | { | |
5239 | return sched_debug_enabled; | |
5240 | } | |
f6630114 | 5241 | |
7c16ec58 | 5242 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5243 | struct cpumask *groupmask) |
1da177e4 | 5244 | { |
4dcf6aff | 5245 | struct sched_group *group = sd->groups; |
434d53b0 | 5246 | char str[256]; |
1da177e4 | 5247 | |
968ea6d8 | 5248 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5249 | cpumask_clear(groupmask); |
4dcf6aff IM |
5250 | |
5251 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5252 | ||
5253 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5254 | printk("does not load-balance\n"); |
4dcf6aff | 5255 | if (sd->parent) |
3df0fc5b PZ |
5256 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5257 | " has parent"); | |
4dcf6aff | 5258 | return -1; |
41c7ce9a NP |
5259 | } |
5260 | ||
3df0fc5b | 5261 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5262 | |
758b2cdc | 5263 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5264 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5265 | "CPU%d\n", cpu); | |
4dcf6aff | 5266 | } |
758b2cdc | 5267 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5268 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5269 | " CPU%d\n", cpu); | |
4dcf6aff | 5270 | } |
1da177e4 | 5271 | |
4dcf6aff | 5272 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5273 | do { |
4dcf6aff | 5274 | if (!group) { |
3df0fc5b PZ |
5275 | printk("\n"); |
5276 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5277 | break; |
5278 | } | |
5279 | ||
c3decf0d PZ |
5280 | /* |
5281 | * Even though we initialize ->power to something semi-sane, | |
5282 | * we leave power_orig unset. This allows us to detect if | |
5283 | * domain iteration is still funny without causing /0 traps. | |
5284 | */ | |
5285 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5286 | printk(KERN_CONT "\n"); |
5287 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5288 | "set\n"); | |
4dcf6aff IM |
5289 | break; |
5290 | } | |
1da177e4 | 5291 | |
758b2cdc | 5292 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5293 | printk(KERN_CONT "\n"); |
5294 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5295 | break; |
5296 | } | |
1da177e4 | 5297 | |
cb83b629 PZ |
5298 | if (!(sd->flags & SD_OVERLAP) && |
5299 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5300 | printk(KERN_CONT "\n"); |
5301 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5302 | break; |
5303 | } | |
1da177e4 | 5304 | |
758b2cdc | 5305 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5306 | |
968ea6d8 | 5307 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5308 | |
3df0fc5b | 5309 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5310 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5311 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5312 | group->sgp->power); |
381512cf | 5313 | } |
1da177e4 | 5314 | |
4dcf6aff IM |
5315 | group = group->next; |
5316 | } while (group != sd->groups); | |
3df0fc5b | 5317 | printk(KERN_CONT "\n"); |
1da177e4 | 5318 | |
758b2cdc | 5319 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5320 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5321 | |
758b2cdc RR |
5322 | if (sd->parent && |
5323 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5324 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5325 | "of domain->span\n"); | |
4dcf6aff IM |
5326 | return 0; |
5327 | } | |
1da177e4 | 5328 | |
4dcf6aff IM |
5329 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5330 | { | |
5331 | int level = 0; | |
1da177e4 | 5332 | |
d039ac60 | 5333 | if (!sched_debug_enabled) |
f6630114 MT |
5334 | return; |
5335 | ||
4dcf6aff IM |
5336 | if (!sd) { |
5337 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5338 | return; | |
5339 | } | |
1da177e4 | 5340 | |
4dcf6aff IM |
5341 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5342 | ||
5343 | for (;;) { | |
4cb98839 | 5344 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5345 | break; |
1da177e4 LT |
5346 | level++; |
5347 | sd = sd->parent; | |
33859f7f | 5348 | if (!sd) |
4dcf6aff IM |
5349 | break; |
5350 | } | |
1da177e4 | 5351 | } |
6d6bc0ad | 5352 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5353 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5354 | static inline bool sched_debug(void) |
5355 | { | |
5356 | return false; | |
5357 | } | |
6d6bc0ad | 5358 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5359 | |
1a20ff27 | 5360 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5361 | { |
758b2cdc | 5362 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5363 | return 1; |
5364 | ||
5365 | /* Following flags need at least 2 groups */ | |
5366 | if (sd->flags & (SD_LOAD_BALANCE | | |
5367 | SD_BALANCE_NEWIDLE | | |
5368 | SD_BALANCE_FORK | | |
89c4710e SS |
5369 | SD_BALANCE_EXEC | |
5370 | SD_SHARE_CPUPOWER | | |
5371 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5372 | if (sd->groups != sd->groups->next) |
5373 | return 0; | |
5374 | } | |
5375 | ||
5376 | /* Following flags don't use groups */ | |
c88d5910 | 5377 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5378 | return 0; |
5379 | ||
5380 | return 1; | |
5381 | } | |
5382 | ||
48f24c4d IM |
5383 | static int |
5384 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5385 | { |
5386 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5387 | ||
5388 | if (sd_degenerate(parent)) | |
5389 | return 1; | |
5390 | ||
758b2cdc | 5391 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5392 | return 0; |
5393 | ||
245af2c7 SS |
5394 | /* Flags needing groups don't count if only 1 group in parent */ |
5395 | if (parent->groups == parent->groups->next) { | |
5396 | pflags &= ~(SD_LOAD_BALANCE | | |
5397 | SD_BALANCE_NEWIDLE | | |
5398 | SD_BALANCE_FORK | | |
89c4710e SS |
5399 | SD_BALANCE_EXEC | |
5400 | SD_SHARE_CPUPOWER | | |
5401 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5402 | if (nr_node_ids == 1) |
5403 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5404 | } |
5405 | if (~cflags & pflags) | |
5406 | return 0; | |
5407 | ||
5408 | return 1; | |
5409 | } | |
5410 | ||
dce840a0 | 5411 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5412 | { |
dce840a0 | 5413 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5414 | |
68e74568 | 5415 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5416 | free_cpumask_var(rd->rto_mask); |
5417 | free_cpumask_var(rd->online); | |
5418 | free_cpumask_var(rd->span); | |
5419 | kfree(rd); | |
5420 | } | |
5421 | ||
57d885fe GH |
5422 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5423 | { | |
a0490fa3 | 5424 | struct root_domain *old_rd = NULL; |
57d885fe | 5425 | unsigned long flags; |
57d885fe | 5426 | |
05fa785c | 5427 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5428 | |
5429 | if (rq->rd) { | |
a0490fa3 | 5430 | old_rd = rq->rd; |
57d885fe | 5431 | |
c6c4927b | 5432 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5433 | set_rq_offline(rq); |
57d885fe | 5434 | |
c6c4927b | 5435 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5436 | |
a0490fa3 IM |
5437 | /* |
5438 | * If we dont want to free the old_rt yet then | |
5439 | * set old_rd to NULL to skip the freeing later | |
5440 | * in this function: | |
5441 | */ | |
5442 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5443 | old_rd = NULL; | |
57d885fe GH |
5444 | } |
5445 | ||
5446 | atomic_inc(&rd->refcount); | |
5447 | rq->rd = rd; | |
5448 | ||
c6c4927b | 5449 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5450 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5451 | set_rq_online(rq); |
57d885fe | 5452 | |
05fa785c | 5453 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5454 | |
5455 | if (old_rd) | |
dce840a0 | 5456 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5457 | } |
5458 | ||
68c38fc3 | 5459 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5460 | { |
5461 | memset(rd, 0, sizeof(*rd)); | |
5462 | ||
68c38fc3 | 5463 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5464 | goto out; |
68c38fc3 | 5465 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5466 | goto free_span; |
68c38fc3 | 5467 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5468 | goto free_online; |
6e0534f2 | 5469 | |
68c38fc3 | 5470 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5471 | goto free_rto_mask; |
c6c4927b | 5472 | return 0; |
6e0534f2 | 5473 | |
68e74568 RR |
5474 | free_rto_mask: |
5475 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5476 | free_online: |
5477 | free_cpumask_var(rd->online); | |
5478 | free_span: | |
5479 | free_cpumask_var(rd->span); | |
0c910d28 | 5480 | out: |
c6c4927b | 5481 | return -ENOMEM; |
57d885fe GH |
5482 | } |
5483 | ||
029632fb PZ |
5484 | /* |
5485 | * By default the system creates a single root-domain with all cpus as | |
5486 | * members (mimicking the global state we have today). | |
5487 | */ | |
5488 | struct root_domain def_root_domain; | |
5489 | ||
57d885fe GH |
5490 | static void init_defrootdomain(void) |
5491 | { | |
68c38fc3 | 5492 | init_rootdomain(&def_root_domain); |
c6c4927b | 5493 | |
57d885fe GH |
5494 | atomic_set(&def_root_domain.refcount, 1); |
5495 | } | |
5496 | ||
dc938520 | 5497 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5498 | { |
5499 | struct root_domain *rd; | |
5500 | ||
5501 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5502 | if (!rd) | |
5503 | return NULL; | |
5504 | ||
68c38fc3 | 5505 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5506 | kfree(rd); |
5507 | return NULL; | |
5508 | } | |
57d885fe GH |
5509 | |
5510 | return rd; | |
5511 | } | |
5512 | ||
e3589f6c PZ |
5513 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5514 | { | |
5515 | struct sched_group *tmp, *first; | |
5516 | ||
5517 | if (!sg) | |
5518 | return; | |
5519 | ||
5520 | first = sg; | |
5521 | do { | |
5522 | tmp = sg->next; | |
5523 | ||
5524 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5525 | kfree(sg->sgp); | |
5526 | ||
5527 | kfree(sg); | |
5528 | sg = tmp; | |
5529 | } while (sg != first); | |
5530 | } | |
5531 | ||
dce840a0 PZ |
5532 | static void free_sched_domain(struct rcu_head *rcu) |
5533 | { | |
5534 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5535 | |
5536 | /* | |
5537 | * If its an overlapping domain it has private groups, iterate and | |
5538 | * nuke them all. | |
5539 | */ | |
5540 | if (sd->flags & SD_OVERLAP) { | |
5541 | free_sched_groups(sd->groups, 1); | |
5542 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5543 | kfree(sd->groups->sgp); |
dce840a0 | 5544 | kfree(sd->groups); |
9c3f75cb | 5545 | } |
dce840a0 PZ |
5546 | kfree(sd); |
5547 | } | |
5548 | ||
5549 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5550 | { | |
5551 | call_rcu(&sd->rcu, free_sched_domain); | |
5552 | } | |
5553 | ||
5554 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5555 | { | |
5556 | for (; sd; sd = sd->parent) | |
5557 | destroy_sched_domain(sd, cpu); | |
5558 | } | |
5559 | ||
518cd623 PZ |
5560 | /* |
5561 | * Keep a special pointer to the highest sched_domain that has | |
5562 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5563 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5564 | * | |
5565 | * Also keep a unique ID per domain (we use the first cpu number in | |
5566 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5567 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5568 | */ |
5569 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5570 | DEFINE_PER_CPU(int, sd_llc_id); | |
5571 | ||
5572 | static void update_top_cache_domain(int cpu) | |
5573 | { | |
5574 | struct sched_domain *sd; | |
5575 | int id = cpu; | |
5576 | ||
5577 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
37407ea7 | 5578 | if (sd) |
518cd623 PZ |
5579 | id = cpumask_first(sched_domain_span(sd)); |
5580 | ||
5581 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5582 | per_cpu(sd_llc_id, cpu) = id; | |
5583 | } | |
5584 | ||
1da177e4 | 5585 | /* |
0eab9146 | 5586 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5587 | * hold the hotplug lock. |
5588 | */ | |
0eab9146 IM |
5589 | static void |
5590 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5591 | { |
70b97a7f | 5592 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5593 | struct sched_domain *tmp; |
5594 | ||
5595 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5596 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5597 | struct sched_domain *parent = tmp->parent; |
5598 | if (!parent) | |
5599 | break; | |
f29c9b1c | 5600 | |
1a848870 | 5601 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5602 | tmp->parent = parent->parent; |
1a848870 SS |
5603 | if (parent->parent) |
5604 | parent->parent->child = tmp; | |
dce840a0 | 5605 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5606 | } else |
5607 | tmp = tmp->parent; | |
245af2c7 SS |
5608 | } |
5609 | ||
1a848870 | 5610 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5611 | tmp = sd; |
245af2c7 | 5612 | sd = sd->parent; |
dce840a0 | 5613 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5614 | if (sd) |
5615 | sd->child = NULL; | |
5616 | } | |
1da177e4 | 5617 | |
4cb98839 | 5618 | sched_domain_debug(sd, cpu); |
1da177e4 | 5619 | |
57d885fe | 5620 | rq_attach_root(rq, rd); |
dce840a0 | 5621 | tmp = rq->sd; |
674311d5 | 5622 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5623 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5624 | |
5625 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5626 | } |
5627 | ||
5628 | /* cpus with isolated domains */ | |
dcc30a35 | 5629 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5630 | |
5631 | /* Setup the mask of cpus configured for isolated domains */ | |
5632 | static int __init isolated_cpu_setup(char *str) | |
5633 | { | |
bdddd296 | 5634 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5635 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5636 | return 1; |
5637 | } | |
5638 | ||
8927f494 | 5639 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5640 | |
d3081f52 PZ |
5641 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5642 | { | |
5643 | return cpumask_of_node(cpu_to_node(cpu)); | |
5644 | } | |
5645 | ||
dce840a0 PZ |
5646 | struct sd_data { |
5647 | struct sched_domain **__percpu sd; | |
5648 | struct sched_group **__percpu sg; | |
9c3f75cb | 5649 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5650 | }; |
5651 | ||
49a02c51 | 5652 | struct s_data { |
21d42ccf | 5653 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5654 | struct root_domain *rd; |
5655 | }; | |
5656 | ||
2109b99e | 5657 | enum s_alloc { |
2109b99e | 5658 | sa_rootdomain, |
21d42ccf | 5659 | sa_sd, |
dce840a0 | 5660 | sa_sd_storage, |
2109b99e AH |
5661 | sa_none, |
5662 | }; | |
5663 | ||
54ab4ff4 PZ |
5664 | struct sched_domain_topology_level; |
5665 | ||
5666 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5667 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5668 | ||
e3589f6c PZ |
5669 | #define SDTL_OVERLAP 0x01 |
5670 | ||
eb7a74e6 | 5671 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5672 | sched_domain_init_f init; |
5673 | sched_domain_mask_f mask; | |
e3589f6c | 5674 | int flags; |
cb83b629 | 5675 | int numa_level; |
54ab4ff4 | 5676 | struct sd_data data; |
eb7a74e6 PZ |
5677 | }; |
5678 | ||
c1174876 PZ |
5679 | /* |
5680 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5681 | * domain traversal. | |
5682 | * | |
5683 | * Asymmetric node setups can result in situations where the domain tree is of | |
5684 | * unequal depth, make sure to skip domains that already cover the entire | |
5685 | * range. | |
5686 | * | |
5687 | * In that case build_sched_domains() will have terminated the iteration early | |
5688 | * and our sibling sd spans will be empty. Domains should always include the | |
5689 | * cpu they're built on, so check that. | |
5690 | * | |
5691 | */ | |
5692 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5693 | { | |
5694 | const struct cpumask *span = sched_domain_span(sd); | |
5695 | struct sd_data *sdd = sd->private; | |
5696 | struct sched_domain *sibling; | |
5697 | int i; | |
5698 | ||
5699 | for_each_cpu(i, span) { | |
5700 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5701 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5702 | continue; | |
5703 | ||
5704 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5705 | } | |
5706 | } | |
5707 | ||
5708 | /* | |
5709 | * Return the canonical balance cpu for this group, this is the first cpu | |
5710 | * of this group that's also in the iteration mask. | |
5711 | */ | |
5712 | int group_balance_cpu(struct sched_group *sg) | |
5713 | { | |
5714 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5715 | } | |
5716 | ||
e3589f6c PZ |
5717 | static int |
5718 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5719 | { | |
5720 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5721 | const struct cpumask *span = sched_domain_span(sd); | |
5722 | struct cpumask *covered = sched_domains_tmpmask; | |
5723 | struct sd_data *sdd = sd->private; | |
5724 | struct sched_domain *child; | |
5725 | int i; | |
5726 | ||
5727 | cpumask_clear(covered); | |
5728 | ||
5729 | for_each_cpu(i, span) { | |
5730 | struct cpumask *sg_span; | |
5731 | ||
5732 | if (cpumask_test_cpu(i, covered)) | |
5733 | continue; | |
5734 | ||
c1174876 PZ |
5735 | child = *per_cpu_ptr(sdd->sd, i); |
5736 | ||
5737 | /* See the comment near build_group_mask(). */ | |
5738 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5739 | continue; | |
5740 | ||
e3589f6c | 5741 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5742 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5743 | |
5744 | if (!sg) | |
5745 | goto fail; | |
5746 | ||
5747 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5748 | if (child->child) { |
5749 | child = child->child; | |
5750 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5751 | } else | |
5752 | cpumask_set_cpu(i, sg_span); | |
5753 | ||
5754 | cpumask_or(covered, covered, sg_span); | |
5755 | ||
74a5ce20 | 5756 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5757 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5758 | build_group_mask(sd, sg); | |
5759 | ||
c3decf0d PZ |
5760 | /* |
5761 | * Initialize sgp->power such that even if we mess up the | |
5762 | * domains and no possible iteration will get us here, we won't | |
5763 | * die on a /0 trap. | |
5764 | */ | |
5765 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 5766 | |
c1174876 PZ |
5767 | /* |
5768 | * Make sure the first group of this domain contains the | |
5769 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5770 | * breaks. See update_sg_lb_stats(). | |
5771 | */ | |
74a5ce20 | 5772 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5773 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5774 | groups = sg; |
5775 | ||
5776 | if (!first) | |
5777 | first = sg; | |
5778 | if (last) | |
5779 | last->next = sg; | |
5780 | last = sg; | |
5781 | last->next = first; | |
5782 | } | |
5783 | sd->groups = groups; | |
5784 | ||
5785 | return 0; | |
5786 | ||
5787 | fail: | |
5788 | free_sched_groups(first, 0); | |
5789 | ||
5790 | return -ENOMEM; | |
5791 | } | |
5792 | ||
dce840a0 | 5793 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5794 | { |
dce840a0 PZ |
5795 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5796 | struct sched_domain *child = sd->child; | |
1da177e4 | 5797 | |
dce840a0 PZ |
5798 | if (child) |
5799 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5800 | |
9c3f75cb | 5801 | if (sg) { |
dce840a0 | 5802 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5803 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5804 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5805 | } |
dce840a0 PZ |
5806 | |
5807 | return cpu; | |
1e9f28fa | 5808 | } |
1e9f28fa | 5809 | |
01a08546 | 5810 | /* |
dce840a0 PZ |
5811 | * build_sched_groups will build a circular linked list of the groups |
5812 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5813 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5814 | * |
5815 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5816 | */ |
e3589f6c PZ |
5817 | static int |
5818 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5819 | { |
dce840a0 PZ |
5820 | struct sched_group *first = NULL, *last = NULL; |
5821 | struct sd_data *sdd = sd->private; | |
5822 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5823 | struct cpumask *covered; |
dce840a0 | 5824 | int i; |
9c1cfda2 | 5825 | |
e3589f6c PZ |
5826 | get_group(cpu, sdd, &sd->groups); |
5827 | atomic_inc(&sd->groups->ref); | |
5828 | ||
5829 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
5830 | return 0; | |
5831 | ||
f96225fd PZ |
5832 | lockdep_assert_held(&sched_domains_mutex); |
5833 | covered = sched_domains_tmpmask; | |
5834 | ||
dce840a0 | 5835 | cpumask_clear(covered); |
6711cab4 | 5836 | |
dce840a0 PZ |
5837 | for_each_cpu(i, span) { |
5838 | struct sched_group *sg; | |
5839 | int group = get_group(i, sdd, &sg); | |
5840 | int j; | |
6711cab4 | 5841 | |
dce840a0 PZ |
5842 | if (cpumask_test_cpu(i, covered)) |
5843 | continue; | |
6711cab4 | 5844 | |
dce840a0 | 5845 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5846 | sg->sgp->power = 0; |
c1174876 | 5847 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5848 | |
dce840a0 PZ |
5849 | for_each_cpu(j, span) { |
5850 | if (get_group(j, sdd, NULL) != group) | |
5851 | continue; | |
0601a88d | 5852 | |
dce840a0 PZ |
5853 | cpumask_set_cpu(j, covered); |
5854 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5855 | } | |
0601a88d | 5856 | |
dce840a0 PZ |
5857 | if (!first) |
5858 | first = sg; | |
5859 | if (last) | |
5860 | last->next = sg; | |
5861 | last = sg; | |
5862 | } | |
5863 | last->next = first; | |
e3589f6c PZ |
5864 | |
5865 | return 0; | |
0601a88d | 5866 | } |
51888ca2 | 5867 | |
89c4710e SS |
5868 | /* |
5869 | * Initialize sched groups cpu_power. | |
5870 | * | |
5871 | * cpu_power indicates the capacity of sched group, which is used while | |
5872 | * distributing the load between different sched groups in a sched domain. | |
5873 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5874 | * there are asymmetries in the topology. If there are asymmetries, group | |
5875 | * having more cpu_power will pickup more load compared to the group having | |
5876 | * less cpu_power. | |
89c4710e SS |
5877 | */ |
5878 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5879 | { | |
e3589f6c | 5880 | struct sched_group *sg = sd->groups; |
89c4710e | 5881 | |
e3589f6c PZ |
5882 | WARN_ON(!sd || !sg); |
5883 | ||
5884 | do { | |
5885 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5886 | sg = sg->next; | |
5887 | } while (sg != sd->groups); | |
89c4710e | 5888 | |
c1174876 | 5889 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5890 | return; |
aae6d3dd | 5891 | |
d274cb30 | 5892 | update_group_power(sd, cpu); |
69e1e811 | 5893 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5894 | } |
5895 | ||
029632fb PZ |
5896 | int __weak arch_sd_sibling_asym_packing(void) |
5897 | { | |
5898 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5899 | } |
5900 | ||
7c16ec58 MT |
5901 | /* |
5902 | * Initializers for schedule domains | |
5903 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5904 | */ | |
5905 | ||
a5d8c348 IM |
5906 | #ifdef CONFIG_SCHED_DEBUG |
5907 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5908 | #else | |
5909 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5910 | #endif | |
5911 | ||
54ab4ff4 PZ |
5912 | #define SD_INIT_FUNC(type) \ |
5913 | static noinline struct sched_domain * \ | |
5914 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5915 | { \ | |
5916 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5917 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5918 | SD_INIT_NAME(sd, type); \ |
5919 | sd->private = &tl->data; \ | |
5920 | return sd; \ | |
7c16ec58 MT |
5921 | } |
5922 | ||
5923 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5924 | #ifdef CONFIG_SCHED_SMT |
5925 | SD_INIT_FUNC(SIBLING) | |
5926 | #endif | |
5927 | #ifdef CONFIG_SCHED_MC | |
5928 | SD_INIT_FUNC(MC) | |
5929 | #endif | |
01a08546 HC |
5930 | #ifdef CONFIG_SCHED_BOOK |
5931 | SD_INIT_FUNC(BOOK) | |
5932 | #endif | |
7c16ec58 | 5933 | |
1d3504fc | 5934 | static int default_relax_domain_level = -1; |
60495e77 | 5935 | int sched_domain_level_max; |
1d3504fc HS |
5936 | |
5937 | static int __init setup_relax_domain_level(char *str) | |
5938 | { | |
a841f8ce DS |
5939 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5940 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5941 | |
1d3504fc HS |
5942 | return 1; |
5943 | } | |
5944 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5945 | ||
5946 | static void set_domain_attribute(struct sched_domain *sd, | |
5947 | struct sched_domain_attr *attr) | |
5948 | { | |
5949 | int request; | |
5950 | ||
5951 | if (!attr || attr->relax_domain_level < 0) { | |
5952 | if (default_relax_domain_level < 0) | |
5953 | return; | |
5954 | else | |
5955 | request = default_relax_domain_level; | |
5956 | } else | |
5957 | request = attr->relax_domain_level; | |
5958 | if (request < sd->level) { | |
5959 | /* turn off idle balance on this domain */ | |
c88d5910 | 5960 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5961 | } else { |
5962 | /* turn on idle balance on this domain */ | |
c88d5910 | 5963 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5964 | } |
5965 | } | |
5966 | ||
54ab4ff4 PZ |
5967 | static void __sdt_free(const struct cpumask *cpu_map); |
5968 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5969 | ||
2109b99e AH |
5970 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5971 | const struct cpumask *cpu_map) | |
5972 | { | |
5973 | switch (what) { | |
2109b99e | 5974 | case sa_rootdomain: |
822ff793 PZ |
5975 | if (!atomic_read(&d->rd->refcount)) |
5976 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5977 | case sa_sd: |
5978 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5979 | case sa_sd_storage: |
54ab4ff4 | 5980 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5981 | case sa_none: |
5982 | break; | |
5983 | } | |
5984 | } | |
3404c8d9 | 5985 | |
2109b99e AH |
5986 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5987 | const struct cpumask *cpu_map) | |
5988 | { | |
dce840a0 PZ |
5989 | memset(d, 0, sizeof(*d)); |
5990 | ||
54ab4ff4 PZ |
5991 | if (__sdt_alloc(cpu_map)) |
5992 | return sa_sd_storage; | |
dce840a0 PZ |
5993 | d->sd = alloc_percpu(struct sched_domain *); |
5994 | if (!d->sd) | |
5995 | return sa_sd_storage; | |
2109b99e | 5996 | d->rd = alloc_rootdomain(); |
dce840a0 | 5997 | if (!d->rd) |
21d42ccf | 5998 | return sa_sd; |
2109b99e AH |
5999 | return sa_rootdomain; |
6000 | } | |
57d885fe | 6001 | |
dce840a0 PZ |
6002 | /* |
6003 | * NULL the sd_data elements we've used to build the sched_domain and | |
6004 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6005 | * will not free the data we're using. | |
6006 | */ | |
6007 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6008 | { | |
6009 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6010 | |
6011 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6012 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6013 | ||
e3589f6c | 6014 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6015 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6016 | |
6017 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6018 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6019 | } |
6020 | ||
2c402dc3 PZ |
6021 | #ifdef CONFIG_SCHED_SMT |
6022 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6023 | { |
2c402dc3 | 6024 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6025 | } |
2c402dc3 | 6026 | #endif |
7f4588f3 | 6027 | |
d069b916 PZ |
6028 | /* |
6029 | * Topology list, bottom-up. | |
6030 | */ | |
2c402dc3 | 6031 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6032 | #ifdef CONFIG_SCHED_SMT |
6033 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6034 | #endif |
1e9f28fa | 6035 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6036 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6037 | #endif |
d069b916 PZ |
6038 | #ifdef CONFIG_SCHED_BOOK |
6039 | { sd_init_BOOK, cpu_book_mask, }, | |
6040 | #endif | |
6041 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
6042 | { NULL, }, |
6043 | }; | |
6044 | ||
6045 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6046 | ||
cb83b629 PZ |
6047 | #ifdef CONFIG_NUMA |
6048 | ||
6049 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
6050 | static int *sched_domains_numa_distance; |
6051 | static struct cpumask ***sched_domains_numa_masks; | |
6052 | static int sched_domains_curr_level; | |
6053 | ||
cb83b629 PZ |
6054 | static inline int sd_local_flags(int level) |
6055 | { | |
10717dcd | 6056 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
6057 | return 0; |
6058 | ||
6059 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6060 | } | |
6061 | ||
6062 | static struct sched_domain * | |
6063 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6064 | { | |
6065 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6066 | int level = tl->numa_level; | |
6067 | int sd_weight = cpumask_weight( | |
6068 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6069 | ||
6070 | *sd = (struct sched_domain){ | |
6071 | .min_interval = sd_weight, | |
6072 | .max_interval = 2*sd_weight, | |
6073 | .busy_factor = 32, | |
870a0bb5 | 6074 | .imbalance_pct = 125, |
cb83b629 PZ |
6075 | .cache_nice_tries = 2, |
6076 | .busy_idx = 3, | |
6077 | .idle_idx = 2, | |
6078 | .newidle_idx = 0, | |
6079 | .wake_idx = 0, | |
6080 | .forkexec_idx = 0, | |
6081 | ||
6082 | .flags = 1*SD_LOAD_BALANCE | |
6083 | | 1*SD_BALANCE_NEWIDLE | |
6084 | | 0*SD_BALANCE_EXEC | |
6085 | | 0*SD_BALANCE_FORK | |
6086 | | 0*SD_BALANCE_WAKE | |
6087 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 6088 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
6089 | | 0*SD_SHARE_PKG_RESOURCES |
6090 | | 1*SD_SERIALIZE | |
6091 | | 0*SD_PREFER_SIBLING | |
6092 | | sd_local_flags(level) | |
6093 | , | |
6094 | .last_balance = jiffies, | |
6095 | .balance_interval = sd_weight, | |
6096 | }; | |
6097 | SD_INIT_NAME(sd, NUMA); | |
6098 | sd->private = &tl->data; | |
6099 | ||
6100 | /* | |
6101 | * Ugly hack to pass state to sd_numa_mask()... | |
6102 | */ | |
6103 | sched_domains_curr_level = tl->numa_level; | |
6104 | ||
6105 | return sd; | |
6106 | } | |
6107 | ||
6108 | static const struct cpumask *sd_numa_mask(int cpu) | |
6109 | { | |
6110 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6111 | } | |
6112 | ||
d039ac60 PZ |
6113 | static void sched_numa_warn(const char *str) |
6114 | { | |
6115 | static int done = false; | |
6116 | int i,j; | |
6117 | ||
6118 | if (done) | |
6119 | return; | |
6120 | ||
6121 | done = true; | |
6122 | ||
6123 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6124 | ||
6125 | for (i = 0; i < nr_node_ids; i++) { | |
6126 | printk(KERN_WARNING " "); | |
6127 | for (j = 0; j < nr_node_ids; j++) | |
6128 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6129 | printk(KERN_CONT "\n"); | |
6130 | } | |
6131 | printk(KERN_WARNING "\n"); | |
6132 | } | |
6133 | ||
6134 | static bool find_numa_distance(int distance) | |
6135 | { | |
6136 | int i; | |
6137 | ||
6138 | if (distance == node_distance(0, 0)) | |
6139 | return true; | |
6140 | ||
6141 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6142 | if (sched_domains_numa_distance[i] == distance) | |
6143 | return true; | |
6144 | } | |
6145 | ||
6146 | return false; | |
6147 | } | |
6148 | ||
cb83b629 PZ |
6149 | static void sched_init_numa(void) |
6150 | { | |
6151 | int next_distance, curr_distance = node_distance(0, 0); | |
6152 | struct sched_domain_topology_level *tl; | |
6153 | int level = 0; | |
6154 | int i, j, k; | |
6155 | ||
cb83b629 PZ |
6156 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6157 | if (!sched_domains_numa_distance) | |
6158 | return; | |
6159 | ||
6160 | /* | |
6161 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6162 | * unique distances in the node_distance() table. | |
6163 | * | |
6164 | * Assumes node_distance(0,j) includes all distances in | |
6165 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6166 | */ |
6167 | next_distance = curr_distance; | |
6168 | for (i = 0; i < nr_node_ids; i++) { | |
6169 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6170 | for (k = 0; k < nr_node_ids; k++) { |
6171 | int distance = node_distance(i, k); | |
6172 | ||
6173 | if (distance > curr_distance && | |
6174 | (distance < next_distance || | |
6175 | next_distance == curr_distance)) | |
6176 | next_distance = distance; | |
6177 | ||
6178 | /* | |
6179 | * While not a strong assumption it would be nice to know | |
6180 | * about cases where if node A is connected to B, B is not | |
6181 | * equally connected to A. | |
6182 | */ | |
6183 | if (sched_debug() && node_distance(k, i) != distance) | |
6184 | sched_numa_warn("Node-distance not symmetric"); | |
6185 | ||
6186 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6187 | sched_numa_warn("Node-0 not representative"); | |
6188 | } | |
6189 | if (next_distance != curr_distance) { | |
6190 | sched_domains_numa_distance[level++] = next_distance; | |
6191 | sched_domains_numa_levels = level; | |
6192 | curr_distance = next_distance; | |
6193 | } else break; | |
cb83b629 | 6194 | } |
d039ac60 PZ |
6195 | |
6196 | /* | |
6197 | * In case of sched_debug() we verify the above assumption. | |
6198 | */ | |
6199 | if (!sched_debug()) | |
6200 | break; | |
cb83b629 PZ |
6201 | } |
6202 | /* | |
6203 | * 'level' contains the number of unique distances, excluding the | |
6204 | * identity distance node_distance(i,i). | |
6205 | * | |
6206 | * The sched_domains_nume_distance[] array includes the actual distance | |
6207 | * numbers. | |
6208 | */ | |
6209 | ||
5f7865f3 TC |
6210 | /* |
6211 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6212 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6213 | * the array will contain less then 'level' members. This could be | |
6214 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6215 | * in other functions. | |
6216 | * | |
6217 | * We reset it to 'level' at the end of this function. | |
6218 | */ | |
6219 | sched_domains_numa_levels = 0; | |
6220 | ||
cb83b629 PZ |
6221 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6222 | if (!sched_domains_numa_masks) | |
6223 | return; | |
6224 | ||
6225 | /* | |
6226 | * Now for each level, construct a mask per node which contains all | |
6227 | * cpus of nodes that are that many hops away from us. | |
6228 | */ | |
6229 | for (i = 0; i < level; i++) { | |
6230 | sched_domains_numa_masks[i] = | |
6231 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6232 | if (!sched_domains_numa_masks[i]) | |
6233 | return; | |
6234 | ||
6235 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6236 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6237 | if (!mask) |
6238 | return; | |
6239 | ||
6240 | sched_domains_numa_masks[i][j] = mask; | |
6241 | ||
6242 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6243 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6244 | continue; |
6245 | ||
6246 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6247 | } | |
6248 | } | |
6249 | } | |
6250 | ||
6251 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6252 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6253 | if (!tl) | |
6254 | return; | |
6255 | ||
6256 | /* | |
6257 | * Copy the default topology bits.. | |
6258 | */ | |
6259 | for (i = 0; default_topology[i].init; i++) | |
6260 | tl[i] = default_topology[i]; | |
6261 | ||
6262 | /* | |
6263 | * .. and append 'j' levels of NUMA goodness. | |
6264 | */ | |
6265 | for (j = 0; j < level; i++, j++) { | |
6266 | tl[i] = (struct sched_domain_topology_level){ | |
6267 | .init = sd_numa_init, | |
6268 | .mask = sd_numa_mask, | |
6269 | .flags = SDTL_OVERLAP, | |
6270 | .numa_level = j, | |
6271 | }; | |
6272 | } | |
6273 | ||
6274 | sched_domain_topology = tl; | |
5f7865f3 TC |
6275 | |
6276 | sched_domains_numa_levels = level; | |
cb83b629 | 6277 | } |
301a5cba TC |
6278 | |
6279 | static void sched_domains_numa_masks_set(int cpu) | |
6280 | { | |
6281 | int i, j; | |
6282 | int node = cpu_to_node(cpu); | |
6283 | ||
6284 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6285 | for (j = 0; j < nr_node_ids; j++) { | |
6286 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6287 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6288 | } | |
6289 | } | |
6290 | } | |
6291 | ||
6292 | static void sched_domains_numa_masks_clear(int cpu) | |
6293 | { | |
6294 | int i, j; | |
6295 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6296 | for (j = 0; j < nr_node_ids; j++) | |
6297 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6298 | } | |
6299 | } | |
6300 | ||
6301 | /* | |
6302 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6303 | * are onlined. | |
6304 | */ | |
6305 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6306 | unsigned long action, | |
6307 | void *hcpu) | |
6308 | { | |
6309 | int cpu = (long)hcpu; | |
6310 | ||
6311 | switch (action & ~CPU_TASKS_FROZEN) { | |
6312 | case CPU_ONLINE: | |
6313 | sched_domains_numa_masks_set(cpu); | |
6314 | break; | |
6315 | ||
6316 | case CPU_DEAD: | |
6317 | sched_domains_numa_masks_clear(cpu); | |
6318 | break; | |
6319 | ||
6320 | default: | |
6321 | return NOTIFY_DONE; | |
6322 | } | |
6323 | ||
6324 | return NOTIFY_OK; | |
cb83b629 PZ |
6325 | } |
6326 | #else | |
6327 | static inline void sched_init_numa(void) | |
6328 | { | |
6329 | } | |
301a5cba TC |
6330 | |
6331 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6332 | unsigned long action, | |
6333 | void *hcpu) | |
6334 | { | |
6335 | return 0; | |
6336 | } | |
cb83b629 PZ |
6337 | #endif /* CONFIG_NUMA */ |
6338 | ||
54ab4ff4 PZ |
6339 | static int __sdt_alloc(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 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6348 | if (!sdd->sd) | |
6349 | return -ENOMEM; | |
6350 | ||
6351 | sdd->sg = alloc_percpu(struct sched_group *); | |
6352 | if (!sdd->sg) | |
6353 | return -ENOMEM; | |
6354 | ||
9c3f75cb PZ |
6355 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6356 | if (!sdd->sgp) | |
6357 | return -ENOMEM; | |
6358 | ||
54ab4ff4 PZ |
6359 | for_each_cpu(j, cpu_map) { |
6360 | struct sched_domain *sd; | |
6361 | struct sched_group *sg; | |
9c3f75cb | 6362 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6363 | |
6364 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6365 | GFP_KERNEL, cpu_to_node(j)); | |
6366 | if (!sd) | |
6367 | return -ENOMEM; | |
6368 | ||
6369 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6370 | ||
6371 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6372 | GFP_KERNEL, cpu_to_node(j)); | |
6373 | if (!sg) | |
6374 | return -ENOMEM; | |
6375 | ||
30b4e9eb IM |
6376 | sg->next = sg; |
6377 | ||
54ab4ff4 | 6378 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6379 | |
c1174876 | 6380 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6381 | GFP_KERNEL, cpu_to_node(j)); |
6382 | if (!sgp) | |
6383 | return -ENOMEM; | |
6384 | ||
6385 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6386 | } |
6387 | } | |
6388 | ||
6389 | return 0; | |
6390 | } | |
6391 | ||
6392 | static void __sdt_free(const struct cpumask *cpu_map) | |
6393 | { | |
6394 | struct sched_domain_topology_level *tl; | |
6395 | int j; | |
6396 | ||
6397 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6398 | struct sd_data *sdd = &tl->data; | |
6399 | ||
6400 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6401 | struct sched_domain *sd; |
6402 | ||
6403 | if (sdd->sd) { | |
6404 | sd = *per_cpu_ptr(sdd->sd, j); | |
6405 | if (sd && (sd->flags & SD_OVERLAP)) | |
6406 | free_sched_groups(sd->groups, 0); | |
6407 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6408 | } | |
6409 | ||
6410 | if (sdd->sg) | |
6411 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6412 | if (sdd->sgp) | |
6413 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6414 | } |
6415 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6416 | sdd->sd = NULL; |
54ab4ff4 | 6417 | free_percpu(sdd->sg); |
fb2cf2c6 | 6418 | sdd->sg = NULL; |
9c3f75cb | 6419 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6420 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6421 | } |
6422 | } | |
6423 | ||
2c402dc3 PZ |
6424 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6425 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6426 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6427 | int cpu) |
6428 | { | |
54ab4ff4 | 6429 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6430 | if (!sd) |
d069b916 | 6431 | return child; |
2c402dc3 | 6432 | |
2c402dc3 | 6433 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6434 | if (child) { |
6435 | sd->level = child->level + 1; | |
6436 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6437 | child->parent = sd; |
60495e77 | 6438 | } |
d069b916 | 6439 | sd->child = child; |
a841f8ce | 6440 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6441 | |
6442 | return sd; | |
6443 | } | |
6444 | ||
2109b99e AH |
6445 | /* |
6446 | * Build sched domains for a given set of cpus and attach the sched domains | |
6447 | * to the individual cpus | |
6448 | */ | |
dce840a0 PZ |
6449 | static int build_sched_domains(const struct cpumask *cpu_map, |
6450 | struct sched_domain_attr *attr) | |
2109b99e AH |
6451 | { |
6452 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6453 | struct sched_domain *sd; |
2109b99e | 6454 | struct s_data d; |
822ff793 | 6455 | int i, ret = -ENOMEM; |
9c1cfda2 | 6456 | |
2109b99e AH |
6457 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6458 | if (alloc_state != sa_rootdomain) | |
6459 | goto error; | |
9c1cfda2 | 6460 | |
dce840a0 | 6461 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6462 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6463 | struct sched_domain_topology_level *tl; |
6464 | ||
3bd65a80 | 6465 | sd = NULL; |
e3589f6c | 6466 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6467 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6468 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6469 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6470 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6471 | break; | |
e3589f6c | 6472 | } |
d274cb30 | 6473 | |
d069b916 PZ |
6474 | while (sd->child) |
6475 | sd = sd->child; | |
6476 | ||
21d42ccf | 6477 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6478 | } |
6479 | ||
6480 | /* Build the groups for the domains */ | |
6481 | for_each_cpu(i, cpu_map) { | |
6482 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6483 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6484 | if (sd->flags & SD_OVERLAP) { |
6485 | if (build_overlap_sched_groups(sd, i)) | |
6486 | goto error; | |
6487 | } else { | |
6488 | if (build_sched_groups(sd, i)) | |
6489 | goto error; | |
6490 | } | |
1cf51902 | 6491 | } |
a06dadbe | 6492 | } |
9c1cfda2 | 6493 | |
1da177e4 | 6494 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6495 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6496 | if (!cpumask_test_cpu(i, cpu_map)) | |
6497 | continue; | |
9c1cfda2 | 6498 | |
dce840a0 PZ |
6499 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6500 | claim_allocations(i, sd); | |
cd4ea6ae | 6501 | init_sched_groups_power(i, sd); |
dce840a0 | 6502 | } |
f712c0c7 | 6503 | } |
9c1cfda2 | 6504 | |
1da177e4 | 6505 | /* Attach the domains */ |
dce840a0 | 6506 | rcu_read_lock(); |
abcd083a | 6507 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6508 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6509 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6510 | } |
dce840a0 | 6511 | rcu_read_unlock(); |
51888ca2 | 6512 | |
822ff793 | 6513 | ret = 0; |
51888ca2 | 6514 | error: |
2109b99e | 6515 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6516 | return ret; |
1da177e4 | 6517 | } |
029190c5 | 6518 | |
acc3f5d7 | 6519 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6520 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6521 | static struct sched_domain_attr *dattr_cur; |
6522 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6523 | |
6524 | /* | |
6525 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6526 | * cpumask) fails, then fallback to a single sched domain, |
6527 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6528 | */ |
4212823f | 6529 | static cpumask_var_t fallback_doms; |
029190c5 | 6530 | |
ee79d1bd HC |
6531 | /* |
6532 | * arch_update_cpu_topology lets virtualized architectures update the | |
6533 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6534 | * or 0 if it stayed the same. | |
6535 | */ | |
6536 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6537 | { |
ee79d1bd | 6538 | return 0; |
22e52b07 HC |
6539 | } |
6540 | ||
acc3f5d7 RR |
6541 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6542 | { | |
6543 | int i; | |
6544 | cpumask_var_t *doms; | |
6545 | ||
6546 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6547 | if (!doms) | |
6548 | return NULL; | |
6549 | for (i = 0; i < ndoms; i++) { | |
6550 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6551 | free_sched_domains(doms, i); | |
6552 | return NULL; | |
6553 | } | |
6554 | } | |
6555 | return doms; | |
6556 | } | |
6557 | ||
6558 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6559 | { | |
6560 | unsigned int i; | |
6561 | for (i = 0; i < ndoms; i++) | |
6562 | free_cpumask_var(doms[i]); | |
6563 | kfree(doms); | |
6564 | } | |
6565 | ||
1a20ff27 | 6566 | /* |
41a2d6cf | 6567 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6568 | * For now this just excludes isolated cpus, but could be used to |
6569 | * exclude other special cases in the future. | |
1a20ff27 | 6570 | */ |
c4a8849a | 6571 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6572 | { |
7378547f MM |
6573 | int err; |
6574 | ||
22e52b07 | 6575 | arch_update_cpu_topology(); |
029190c5 | 6576 | ndoms_cur = 1; |
acc3f5d7 | 6577 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6578 | if (!doms_cur) |
acc3f5d7 RR |
6579 | doms_cur = &fallback_doms; |
6580 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6581 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6582 | register_sched_domain_sysctl(); |
7378547f MM |
6583 | |
6584 | return err; | |
1a20ff27 DG |
6585 | } |
6586 | ||
1a20ff27 DG |
6587 | /* |
6588 | * Detach sched domains from a group of cpus specified in cpu_map | |
6589 | * These cpus will now be attached to the NULL domain | |
6590 | */ | |
96f874e2 | 6591 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6592 | { |
6593 | int i; | |
6594 | ||
dce840a0 | 6595 | rcu_read_lock(); |
abcd083a | 6596 | for_each_cpu(i, cpu_map) |
57d885fe | 6597 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6598 | rcu_read_unlock(); |
1a20ff27 DG |
6599 | } |
6600 | ||
1d3504fc HS |
6601 | /* handle null as "default" */ |
6602 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6603 | struct sched_domain_attr *new, int idx_new) | |
6604 | { | |
6605 | struct sched_domain_attr tmp; | |
6606 | ||
6607 | /* fast path */ | |
6608 | if (!new && !cur) | |
6609 | return 1; | |
6610 | ||
6611 | tmp = SD_ATTR_INIT; | |
6612 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6613 | new ? (new + idx_new) : &tmp, | |
6614 | sizeof(struct sched_domain_attr)); | |
6615 | } | |
6616 | ||
029190c5 PJ |
6617 | /* |
6618 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6619 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6620 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6621 | * It destroys each deleted domain and builds each new domain. | |
6622 | * | |
acc3f5d7 | 6623 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6624 | * The masks don't intersect (don't overlap.) We should setup one |
6625 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6626 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6627 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6628 | * it as it is. | |
6629 | * | |
acc3f5d7 RR |
6630 | * The passed in 'doms_new' should be allocated using |
6631 | * alloc_sched_domains. This routine takes ownership of it and will | |
6632 | * free_sched_domains it when done with it. If the caller failed the | |
6633 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6634 | * and partition_sched_domains() will fallback to the single partition | |
6635 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6636 | * |
96f874e2 | 6637 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6638 | * ndoms_new == 0 is a special case for destroying existing domains, |
6639 | * and it will not create the default domain. | |
dfb512ec | 6640 | * |
029190c5 PJ |
6641 | * Call with hotplug lock held |
6642 | */ | |
acc3f5d7 | 6643 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6644 | struct sched_domain_attr *dattr_new) |
029190c5 | 6645 | { |
dfb512ec | 6646 | int i, j, n; |
d65bd5ec | 6647 | int new_topology; |
029190c5 | 6648 | |
712555ee | 6649 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6650 | |
7378547f MM |
6651 | /* always unregister in case we don't destroy any domains */ |
6652 | unregister_sched_domain_sysctl(); | |
6653 | ||
d65bd5ec HC |
6654 | /* Let architecture update cpu core mappings. */ |
6655 | new_topology = arch_update_cpu_topology(); | |
6656 | ||
dfb512ec | 6657 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6658 | |
6659 | /* Destroy deleted domains */ | |
6660 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6661 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6662 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6663 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6664 | goto match1; |
6665 | } | |
6666 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6667 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6668 | match1: |
6669 | ; | |
6670 | } | |
6671 | ||
e761b772 MK |
6672 | if (doms_new == NULL) { |
6673 | ndoms_cur = 0; | |
acc3f5d7 | 6674 | doms_new = &fallback_doms; |
6ad4c188 | 6675 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6676 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6677 | } |
6678 | ||
029190c5 PJ |
6679 | /* Build new domains */ |
6680 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6681 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6682 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6683 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6684 | goto match2; |
6685 | } | |
6686 | /* no match - add a new doms_new */ | |
dce840a0 | 6687 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6688 | match2: |
6689 | ; | |
6690 | } | |
6691 | ||
6692 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6693 | if (doms_cur != &fallback_doms) |
6694 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6695 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6696 | doms_cur = doms_new; |
1d3504fc | 6697 | dattr_cur = dattr_new; |
029190c5 | 6698 | ndoms_cur = ndoms_new; |
7378547f MM |
6699 | |
6700 | register_sched_domain_sysctl(); | |
a1835615 | 6701 | |
712555ee | 6702 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6703 | } |
6704 | ||
d35be8ba SB |
6705 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6706 | ||
1da177e4 | 6707 | /* |
3a101d05 TH |
6708 | * Update cpusets according to cpu_active mask. If cpusets are |
6709 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6710 | * around partition_sched_domains(). | |
d35be8ba SB |
6711 | * |
6712 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6713 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6714 | */ |
0b2e918a TH |
6715 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6716 | void *hcpu) | |
e761b772 | 6717 | { |
d35be8ba SB |
6718 | switch (action) { |
6719 | case CPU_ONLINE_FROZEN: | |
6720 | case CPU_DOWN_FAILED_FROZEN: | |
6721 | ||
6722 | /* | |
6723 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6724 | * resume sequence. As long as this is not the last online | |
6725 | * operation in the resume sequence, just build a single sched | |
6726 | * domain, ignoring cpusets. | |
6727 | */ | |
6728 | num_cpus_frozen--; | |
6729 | if (likely(num_cpus_frozen)) { | |
6730 | partition_sched_domains(1, NULL, NULL); | |
6731 | break; | |
6732 | } | |
6733 | ||
6734 | /* | |
6735 | * This is the last CPU online operation. So fall through and | |
6736 | * restore the original sched domains by considering the | |
6737 | * cpuset configurations. | |
6738 | */ | |
6739 | ||
e761b772 | 6740 | case CPU_ONLINE: |
6ad4c188 | 6741 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6742 | cpuset_update_active_cpus(true); |
d35be8ba | 6743 | break; |
3a101d05 TH |
6744 | default: |
6745 | return NOTIFY_DONE; | |
6746 | } | |
d35be8ba | 6747 | return NOTIFY_OK; |
3a101d05 | 6748 | } |
e761b772 | 6749 | |
0b2e918a TH |
6750 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6751 | void *hcpu) | |
3a101d05 | 6752 | { |
d35be8ba | 6753 | switch (action) { |
3a101d05 | 6754 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6755 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6756 | break; |
6757 | case CPU_DOWN_PREPARE_FROZEN: | |
6758 | num_cpus_frozen++; | |
6759 | partition_sched_domains(1, NULL, NULL); | |
6760 | break; | |
e761b772 MK |
6761 | default: |
6762 | return NOTIFY_DONE; | |
6763 | } | |
d35be8ba | 6764 | return NOTIFY_OK; |
e761b772 | 6765 | } |
e761b772 | 6766 | |
1da177e4 LT |
6767 | void __init sched_init_smp(void) |
6768 | { | |
dcc30a35 RR |
6769 | cpumask_var_t non_isolated_cpus; |
6770 | ||
6771 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6772 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6773 | |
cb83b629 PZ |
6774 | sched_init_numa(); |
6775 | ||
95402b38 | 6776 | get_online_cpus(); |
712555ee | 6777 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6778 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6779 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6780 | if (cpumask_empty(non_isolated_cpus)) | |
6781 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6782 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6783 | put_online_cpus(); |
e761b772 | 6784 | |
301a5cba | 6785 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6786 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6787 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6788 | |
6789 | /* RT runtime code needs to handle some hotplug events */ | |
6790 | hotcpu_notifier(update_runtime, 0); | |
6791 | ||
b328ca18 | 6792 | init_hrtick(); |
5c1e1767 NP |
6793 | |
6794 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6795 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6796 | BUG(); |
19978ca6 | 6797 | sched_init_granularity(); |
dcc30a35 | 6798 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6799 | |
0e3900e6 | 6800 | init_sched_rt_class(); |
1da177e4 LT |
6801 | } |
6802 | #else | |
6803 | void __init sched_init_smp(void) | |
6804 | { | |
19978ca6 | 6805 | sched_init_granularity(); |
1da177e4 LT |
6806 | } |
6807 | #endif /* CONFIG_SMP */ | |
6808 | ||
cd1bb94b AB |
6809 | const_debug unsigned int sysctl_timer_migration = 1; |
6810 | ||
1da177e4 LT |
6811 | int in_sched_functions(unsigned long addr) |
6812 | { | |
1da177e4 LT |
6813 | return in_lock_functions(addr) || |
6814 | (addr >= (unsigned long)__sched_text_start | |
6815 | && addr < (unsigned long)__sched_text_end); | |
6816 | } | |
6817 | ||
029632fb PZ |
6818 | #ifdef CONFIG_CGROUP_SCHED |
6819 | struct task_group root_task_group; | |
35cf4e50 | 6820 | LIST_HEAD(task_groups); |
052f1dc7 | 6821 | #endif |
6f505b16 | 6822 | |
029632fb | 6823 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6824 | |
1da177e4 LT |
6825 | void __init sched_init(void) |
6826 | { | |
dd41f596 | 6827 | int i, j; |
434d53b0 MT |
6828 | unsigned long alloc_size = 0, ptr; |
6829 | ||
6830 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6831 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6832 | #endif | |
6833 | #ifdef CONFIG_RT_GROUP_SCHED | |
6834 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6835 | #endif |
df7c8e84 | 6836 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6837 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6838 | #endif |
434d53b0 | 6839 | if (alloc_size) { |
36b7b6d4 | 6840 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6841 | |
6842 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6843 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6844 | ptr += nr_cpu_ids * sizeof(void **); |
6845 | ||
07e06b01 | 6846 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6847 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6848 | |
6d6bc0ad | 6849 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6850 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6851 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6852 | ptr += nr_cpu_ids * sizeof(void **); |
6853 | ||
07e06b01 | 6854 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6855 | ptr += nr_cpu_ids * sizeof(void **); |
6856 | ||
6d6bc0ad | 6857 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6858 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6859 | for_each_possible_cpu(i) { | |
6860 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6861 | ptr += cpumask_size(); | |
6862 | } | |
6863 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6864 | } |
dd41f596 | 6865 | |
57d885fe GH |
6866 | #ifdef CONFIG_SMP |
6867 | init_defrootdomain(); | |
6868 | #endif | |
6869 | ||
d0b27fa7 PZ |
6870 | init_rt_bandwidth(&def_rt_bandwidth, |
6871 | global_rt_period(), global_rt_runtime()); | |
6872 | ||
6873 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6874 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6875 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6876 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6877 | |
7c941438 | 6878 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6879 | list_add(&root_task_group.list, &task_groups); |
6880 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6881 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6882 | autogroup_init(&init_task); |
54c707e9 | 6883 | |
7c941438 | 6884 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6885 | |
54c707e9 GC |
6886 | #ifdef CONFIG_CGROUP_CPUACCT |
6887 | root_cpuacct.cpustat = &kernel_cpustat; | |
6888 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
6889 | /* Too early, not expected to fail */ | |
6890 | BUG_ON(!root_cpuacct.cpuusage); | |
6891 | #endif | |
0a945022 | 6892 | for_each_possible_cpu(i) { |
70b97a7f | 6893 | struct rq *rq; |
1da177e4 LT |
6894 | |
6895 | rq = cpu_rq(i); | |
05fa785c | 6896 | raw_spin_lock_init(&rq->lock); |
7897986b | 6897 | rq->nr_running = 0; |
dce48a84 TG |
6898 | rq->calc_load_active = 0; |
6899 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6900 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6901 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6902 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6903 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6904 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6905 | /* |
07e06b01 | 6906 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6907 | * |
6908 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6909 | * gets 100% of the cpu resources in the system. This overall | |
6910 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6911 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6912 | * based on each entity's (task or task-group's) weight |
6913 | * (se->load.weight). | |
6914 | * | |
07e06b01 | 6915 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6916 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6917 | * then A0's share of the cpu resource is: | |
6918 | * | |
0d905bca | 6919 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6920 | * |
07e06b01 YZ |
6921 | * We achieve this by letting root_task_group's tasks sit |
6922 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6923 | */ |
ab84d31e | 6924 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6925 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6926 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6927 | ||
6928 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6929 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6930 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6931 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6932 | #endif |
1da177e4 | 6933 | |
dd41f596 IM |
6934 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6935 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6936 | |
6937 | rq->last_load_update_tick = jiffies; | |
6938 | ||
1da177e4 | 6939 | #ifdef CONFIG_SMP |
41c7ce9a | 6940 | rq->sd = NULL; |
57d885fe | 6941 | rq->rd = NULL; |
1399fa78 | 6942 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6943 | rq->post_schedule = 0; |
1da177e4 | 6944 | rq->active_balance = 0; |
dd41f596 | 6945 | rq->next_balance = jiffies; |
1da177e4 | 6946 | rq->push_cpu = 0; |
0a2966b4 | 6947 | rq->cpu = i; |
1f11eb6a | 6948 | rq->online = 0; |
eae0c9df MG |
6949 | rq->idle_stamp = 0; |
6950 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
6951 | |
6952 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6953 | ||
dc938520 | 6954 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 6955 | #ifdef CONFIG_NO_HZ |
1c792db7 | 6956 | rq->nohz_flags = 0; |
83cd4fe2 | 6957 | #endif |
1da177e4 | 6958 | #endif |
8f4d37ec | 6959 | init_rq_hrtick(rq); |
1da177e4 | 6960 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6961 | } |
6962 | ||
2dd73a4f | 6963 | set_load_weight(&init_task); |
b50f60ce | 6964 | |
e107be36 AK |
6965 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6966 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6967 | #endif | |
6968 | ||
b50f60ce | 6969 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6970 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6971 | #endif |
6972 | ||
1da177e4 LT |
6973 | /* |
6974 | * The boot idle thread does lazy MMU switching as well: | |
6975 | */ | |
6976 | atomic_inc(&init_mm.mm_count); | |
6977 | enter_lazy_tlb(&init_mm, current); | |
6978 | ||
6979 | /* | |
6980 | * Make us the idle thread. Technically, schedule() should not be | |
6981 | * called from this thread, however somewhere below it might be, | |
6982 | * but because we are the idle thread, we just pick up running again | |
6983 | * when this runqueue becomes "idle". | |
6984 | */ | |
6985 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6986 | |
6987 | calc_load_update = jiffies + LOAD_FREQ; | |
6988 | ||
dd41f596 IM |
6989 | /* |
6990 | * During early bootup we pretend to be a normal task: | |
6991 | */ | |
6992 | current->sched_class = &fair_sched_class; | |
6892b75e | 6993 | |
bf4d83f6 | 6994 | #ifdef CONFIG_SMP |
4cb98839 | 6995 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6996 | /* May be allocated at isolcpus cmdline parse time */ |
6997 | if (cpu_isolated_map == NULL) | |
6998 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6999 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7000 | #endif |
7001 | init_sched_fair_class(); | |
6a7b3dc3 | 7002 | |
6892b75e | 7003 | scheduler_running = 1; |
1da177e4 LT |
7004 | } |
7005 | ||
d902db1e | 7006 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7007 | static inline int preempt_count_equals(int preempt_offset) |
7008 | { | |
234da7bc | 7009 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7010 | |
4ba8216c | 7011 | return (nested == preempt_offset); |
e4aafea2 FW |
7012 | } |
7013 | ||
d894837f | 7014 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7015 | { |
1da177e4 LT |
7016 | static unsigned long prev_jiffy; /* ratelimiting */ |
7017 | ||
b3fbab05 | 7018 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7019 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7020 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7021 | return; |
7022 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7023 | return; | |
7024 | prev_jiffy = jiffies; | |
7025 | ||
3df0fc5b PZ |
7026 | printk(KERN_ERR |
7027 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7028 | file, line); | |
7029 | printk(KERN_ERR | |
7030 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7031 | in_atomic(), irqs_disabled(), | |
7032 | current->pid, current->comm); | |
aef745fc IM |
7033 | |
7034 | debug_show_held_locks(current); | |
7035 | if (irqs_disabled()) | |
7036 | print_irqtrace_events(current); | |
7037 | dump_stack(); | |
1da177e4 LT |
7038 | } |
7039 | EXPORT_SYMBOL(__might_sleep); | |
7040 | #endif | |
7041 | ||
7042 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7043 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7044 | { | |
da7a735e PZ |
7045 | const struct sched_class *prev_class = p->sched_class; |
7046 | int old_prio = p->prio; | |
3a5e4dc1 | 7047 | int on_rq; |
3e51f33f | 7048 | |
fd2f4419 | 7049 | on_rq = p->on_rq; |
3a5e4dc1 | 7050 | if (on_rq) |
4ca9b72b | 7051 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7052 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7053 | if (on_rq) { | |
4ca9b72b | 7054 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7055 | resched_task(rq->curr); |
7056 | } | |
da7a735e PZ |
7057 | |
7058 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7059 | } |
7060 | ||
1da177e4 LT |
7061 | void normalize_rt_tasks(void) |
7062 | { | |
a0f98a1c | 7063 | struct task_struct *g, *p; |
1da177e4 | 7064 | unsigned long flags; |
70b97a7f | 7065 | struct rq *rq; |
1da177e4 | 7066 | |
4cf5d77a | 7067 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7068 | do_each_thread(g, p) { |
178be793 IM |
7069 | /* |
7070 | * Only normalize user tasks: | |
7071 | */ | |
7072 | if (!p->mm) | |
7073 | continue; | |
7074 | ||
6cfb0d5d | 7075 | p->se.exec_start = 0; |
6cfb0d5d | 7076 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7077 | p->se.statistics.wait_start = 0; |
7078 | p->se.statistics.sleep_start = 0; | |
7079 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7080 | #endif |
dd41f596 IM |
7081 | |
7082 | if (!rt_task(p)) { | |
7083 | /* | |
7084 | * Renice negative nice level userspace | |
7085 | * tasks back to 0: | |
7086 | */ | |
7087 | if (TASK_NICE(p) < 0 && p->mm) | |
7088 | set_user_nice(p, 0); | |
1da177e4 | 7089 | continue; |
dd41f596 | 7090 | } |
1da177e4 | 7091 | |
1d615482 | 7092 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7093 | rq = __task_rq_lock(p); |
1da177e4 | 7094 | |
178be793 | 7095 | normalize_task(rq, p); |
3a5e4dc1 | 7096 | |
b29739f9 | 7097 | __task_rq_unlock(rq); |
1d615482 | 7098 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7099 | } while_each_thread(g, p); |
7100 | ||
4cf5d77a | 7101 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7102 | } |
7103 | ||
7104 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7105 | |
67fc4e0c | 7106 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7107 | /* |
67fc4e0c | 7108 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7109 | * |
7110 | * They can only be called when the whole system has been | |
7111 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7112 | * activity can take place. Using them for anything else would | |
7113 | * be a serious bug, and as a result, they aren't even visible | |
7114 | * under any other configuration. | |
7115 | */ | |
7116 | ||
7117 | /** | |
7118 | * curr_task - return the current task for a given cpu. | |
7119 | * @cpu: the processor in question. | |
7120 | * | |
7121 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7122 | */ | |
36c8b586 | 7123 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7124 | { |
7125 | return cpu_curr(cpu); | |
7126 | } | |
7127 | ||
67fc4e0c JW |
7128 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7129 | ||
7130 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7131 | /** |
7132 | * set_curr_task - set the current task for a given cpu. | |
7133 | * @cpu: the processor in question. | |
7134 | * @p: the task pointer to set. | |
7135 | * | |
7136 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7137 | * are serviced on a separate stack. It allows the architecture to switch the |
7138 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7139 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7140 | * and caller must save the original value of the current task (see | |
7141 | * curr_task() above) and restore that value before reenabling interrupts and | |
7142 | * re-starting the system. | |
7143 | * | |
7144 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7145 | */ | |
36c8b586 | 7146 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7147 | { |
7148 | cpu_curr(cpu) = p; | |
7149 | } | |
7150 | ||
7151 | #endif | |
29f59db3 | 7152 | |
7c941438 | 7153 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7154 | /* task_group_lock serializes the addition/removal of task groups */ |
7155 | static DEFINE_SPINLOCK(task_group_lock); | |
7156 | ||
bccbe08a PZ |
7157 | static void free_sched_group(struct task_group *tg) |
7158 | { | |
7159 | free_fair_sched_group(tg); | |
7160 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7161 | autogroup_free(tg); |
bccbe08a PZ |
7162 | kfree(tg); |
7163 | } | |
7164 | ||
7165 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7166 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7167 | { |
7168 | struct task_group *tg; | |
bccbe08a PZ |
7169 | |
7170 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7171 | if (!tg) | |
7172 | return ERR_PTR(-ENOMEM); | |
7173 | ||
ec7dc8ac | 7174 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7175 | goto err; |
7176 | ||
ec7dc8ac | 7177 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7178 | goto err; |
7179 | ||
ace783b9 LZ |
7180 | return tg; |
7181 | ||
7182 | err: | |
7183 | free_sched_group(tg); | |
7184 | return ERR_PTR(-ENOMEM); | |
7185 | } | |
7186 | ||
7187 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7188 | { | |
7189 | unsigned long flags; | |
7190 | ||
8ed36996 | 7191 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7192 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7193 | |
7194 | WARN_ON(!parent); /* root should already exist */ | |
7195 | ||
7196 | tg->parent = parent; | |
f473aa5e | 7197 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7198 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7199 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7200 | } |
7201 | ||
9b5b7751 | 7202 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7203 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7204 | { |
29f59db3 | 7205 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7206 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7207 | } |
7208 | ||
9b5b7751 | 7209 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7210 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7211 | { |
7212 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7213 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7214 | } | |
7215 | ||
7216 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7217 | { |
8ed36996 | 7218 | unsigned long flags; |
9b5b7751 | 7219 | int i; |
29f59db3 | 7220 | |
3d4b47b4 PZ |
7221 | /* end participation in shares distribution */ |
7222 | for_each_possible_cpu(i) | |
bccbe08a | 7223 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7224 | |
7225 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7226 | list_del_rcu(&tg->list); |
f473aa5e | 7227 | list_del_rcu(&tg->siblings); |
8ed36996 | 7228 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7229 | } |
7230 | ||
9b5b7751 | 7231 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7232 | * The caller of this function should have put the task in its new group |
7233 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7234 | * reflect its new group. | |
9b5b7751 SV |
7235 | */ |
7236 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7237 | { |
8323f26c | 7238 | struct task_group *tg; |
29f59db3 SV |
7239 | int on_rq, running; |
7240 | unsigned long flags; | |
7241 | struct rq *rq; | |
7242 | ||
7243 | rq = task_rq_lock(tsk, &flags); | |
7244 | ||
051a1d1a | 7245 | running = task_current(rq, tsk); |
fd2f4419 | 7246 | on_rq = tsk->on_rq; |
29f59db3 | 7247 | |
0e1f3483 | 7248 | if (on_rq) |
29f59db3 | 7249 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7250 | if (unlikely(running)) |
7251 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7252 | |
8323f26c PZ |
7253 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
7254 | lockdep_is_held(&tsk->sighand->siglock)), | |
7255 | struct task_group, css); | |
7256 | tg = autogroup_task_group(tsk, tg); | |
7257 | tsk->sched_task_group = tg; | |
7258 | ||
810b3817 | 7259 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7260 | if (tsk->sched_class->task_move_group) |
7261 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7262 | else | |
810b3817 | 7263 | #endif |
b2b5ce02 | 7264 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7265 | |
0e1f3483 HS |
7266 | if (unlikely(running)) |
7267 | tsk->sched_class->set_curr_task(rq); | |
7268 | if (on_rq) | |
371fd7e7 | 7269 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7270 | |
0122ec5b | 7271 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7272 | } |
7c941438 | 7273 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7274 | |
a790de99 | 7275 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7276 | static unsigned long to_ratio(u64 period, u64 runtime) |
7277 | { | |
7278 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7279 | return 1ULL << 20; |
9f0c1e56 | 7280 | |
9a7e0b18 | 7281 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7282 | } |
a790de99 PT |
7283 | #endif |
7284 | ||
7285 | #ifdef CONFIG_RT_GROUP_SCHED | |
7286 | /* | |
7287 | * Ensure that the real time constraints are schedulable. | |
7288 | */ | |
7289 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7290 | |
9a7e0b18 PZ |
7291 | /* Must be called with tasklist_lock held */ |
7292 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7293 | { |
9a7e0b18 | 7294 | struct task_struct *g, *p; |
b40b2e8e | 7295 | |
9a7e0b18 | 7296 | do_each_thread(g, p) { |
029632fb | 7297 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7298 | return 1; |
7299 | } while_each_thread(g, p); | |
b40b2e8e | 7300 | |
9a7e0b18 PZ |
7301 | return 0; |
7302 | } | |
b40b2e8e | 7303 | |
9a7e0b18 PZ |
7304 | struct rt_schedulable_data { |
7305 | struct task_group *tg; | |
7306 | u64 rt_period; | |
7307 | u64 rt_runtime; | |
7308 | }; | |
b40b2e8e | 7309 | |
a790de99 | 7310 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7311 | { |
7312 | struct rt_schedulable_data *d = data; | |
7313 | struct task_group *child; | |
7314 | unsigned long total, sum = 0; | |
7315 | u64 period, runtime; | |
b40b2e8e | 7316 | |
9a7e0b18 PZ |
7317 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7318 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7319 | |
9a7e0b18 PZ |
7320 | if (tg == d->tg) { |
7321 | period = d->rt_period; | |
7322 | runtime = d->rt_runtime; | |
b40b2e8e | 7323 | } |
b40b2e8e | 7324 | |
4653f803 PZ |
7325 | /* |
7326 | * Cannot have more runtime than the period. | |
7327 | */ | |
7328 | if (runtime > period && runtime != RUNTIME_INF) | |
7329 | return -EINVAL; | |
6f505b16 | 7330 | |
4653f803 PZ |
7331 | /* |
7332 | * Ensure we don't starve existing RT tasks. | |
7333 | */ | |
9a7e0b18 PZ |
7334 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7335 | return -EBUSY; | |
6f505b16 | 7336 | |
9a7e0b18 | 7337 | total = to_ratio(period, runtime); |
6f505b16 | 7338 | |
4653f803 PZ |
7339 | /* |
7340 | * Nobody can have more than the global setting allows. | |
7341 | */ | |
7342 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7343 | return -EINVAL; | |
6f505b16 | 7344 | |
4653f803 PZ |
7345 | /* |
7346 | * The sum of our children's runtime should not exceed our own. | |
7347 | */ | |
9a7e0b18 PZ |
7348 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7349 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7350 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7351 | |
9a7e0b18 PZ |
7352 | if (child == d->tg) { |
7353 | period = d->rt_period; | |
7354 | runtime = d->rt_runtime; | |
7355 | } | |
6f505b16 | 7356 | |
9a7e0b18 | 7357 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7358 | } |
6f505b16 | 7359 | |
9a7e0b18 PZ |
7360 | if (sum > total) |
7361 | return -EINVAL; | |
7362 | ||
7363 | return 0; | |
6f505b16 PZ |
7364 | } |
7365 | ||
9a7e0b18 | 7366 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7367 | { |
8277434e PT |
7368 | int ret; |
7369 | ||
9a7e0b18 PZ |
7370 | struct rt_schedulable_data data = { |
7371 | .tg = tg, | |
7372 | .rt_period = period, | |
7373 | .rt_runtime = runtime, | |
7374 | }; | |
7375 | ||
8277434e PT |
7376 | rcu_read_lock(); |
7377 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7378 | rcu_read_unlock(); | |
7379 | ||
7380 | return ret; | |
521f1a24 DG |
7381 | } |
7382 | ||
ab84d31e | 7383 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7384 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7385 | { |
ac086bc2 | 7386 | int i, err = 0; |
9f0c1e56 | 7387 | |
9f0c1e56 | 7388 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7389 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7390 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7391 | if (err) | |
9f0c1e56 | 7392 | goto unlock; |
ac086bc2 | 7393 | |
0986b11b | 7394 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7395 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7396 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7397 | |
7398 | for_each_possible_cpu(i) { | |
7399 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7400 | ||
0986b11b | 7401 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7402 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7403 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7404 | } |
0986b11b | 7405 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7406 | unlock: |
521f1a24 | 7407 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7408 | mutex_unlock(&rt_constraints_mutex); |
7409 | ||
7410 | return err; | |
6f505b16 PZ |
7411 | } |
7412 | ||
d0b27fa7 PZ |
7413 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7414 | { | |
7415 | u64 rt_runtime, rt_period; | |
7416 | ||
7417 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7418 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7419 | if (rt_runtime_us < 0) | |
7420 | rt_runtime = RUNTIME_INF; | |
7421 | ||
ab84d31e | 7422 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7423 | } |
7424 | ||
9f0c1e56 PZ |
7425 | long sched_group_rt_runtime(struct task_group *tg) |
7426 | { | |
7427 | u64 rt_runtime_us; | |
7428 | ||
d0b27fa7 | 7429 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7430 | return -1; |
7431 | ||
d0b27fa7 | 7432 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7433 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7434 | return rt_runtime_us; | |
7435 | } | |
d0b27fa7 PZ |
7436 | |
7437 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7438 | { | |
7439 | u64 rt_runtime, rt_period; | |
7440 | ||
7441 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7442 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7443 | ||
619b0488 R |
7444 | if (rt_period == 0) |
7445 | return -EINVAL; | |
7446 | ||
ab84d31e | 7447 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7448 | } |
7449 | ||
7450 | long sched_group_rt_period(struct task_group *tg) | |
7451 | { | |
7452 | u64 rt_period_us; | |
7453 | ||
7454 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7455 | do_div(rt_period_us, NSEC_PER_USEC); | |
7456 | return rt_period_us; | |
7457 | } | |
7458 | ||
7459 | static int sched_rt_global_constraints(void) | |
7460 | { | |
4653f803 | 7461 | u64 runtime, period; |
d0b27fa7 PZ |
7462 | int ret = 0; |
7463 | ||
ec5d4989 HS |
7464 | if (sysctl_sched_rt_period <= 0) |
7465 | return -EINVAL; | |
7466 | ||
4653f803 PZ |
7467 | runtime = global_rt_runtime(); |
7468 | period = global_rt_period(); | |
7469 | ||
7470 | /* | |
7471 | * Sanity check on the sysctl variables. | |
7472 | */ | |
7473 | if (runtime > period && runtime != RUNTIME_INF) | |
7474 | return -EINVAL; | |
10b612f4 | 7475 | |
d0b27fa7 | 7476 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7477 | read_lock(&tasklist_lock); |
4653f803 | 7478 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7479 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7480 | mutex_unlock(&rt_constraints_mutex); |
7481 | ||
7482 | return ret; | |
7483 | } | |
54e99124 DG |
7484 | |
7485 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7486 | { | |
7487 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7488 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7489 | return 0; | |
7490 | ||
7491 | return 1; | |
7492 | } | |
7493 | ||
6d6bc0ad | 7494 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7495 | static int sched_rt_global_constraints(void) |
7496 | { | |
ac086bc2 PZ |
7497 | unsigned long flags; |
7498 | int i; | |
7499 | ||
ec5d4989 HS |
7500 | if (sysctl_sched_rt_period <= 0) |
7501 | return -EINVAL; | |
7502 | ||
60aa605d PZ |
7503 | /* |
7504 | * There's always some RT tasks in the root group | |
7505 | * -- migration, kstopmachine etc.. | |
7506 | */ | |
7507 | if (sysctl_sched_rt_runtime == 0) | |
7508 | return -EBUSY; | |
7509 | ||
0986b11b | 7510 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7511 | for_each_possible_cpu(i) { |
7512 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7513 | ||
0986b11b | 7514 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7515 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7516 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7517 | } |
0986b11b | 7518 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7519 | |
d0b27fa7 PZ |
7520 | return 0; |
7521 | } | |
6d6bc0ad | 7522 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7523 | |
ce0dbbbb CW |
7524 | int sched_rr_handler(struct ctl_table *table, int write, |
7525 | void __user *buffer, size_t *lenp, | |
7526 | loff_t *ppos) | |
7527 | { | |
7528 | int ret; | |
7529 | static DEFINE_MUTEX(mutex); | |
7530 | ||
7531 | mutex_lock(&mutex); | |
7532 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7533 | /* make sure that internally we keep jiffies */ | |
7534 | /* also, writing zero resets timeslice to default */ | |
7535 | if (!ret && write) { | |
7536 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7537 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7538 | } | |
7539 | mutex_unlock(&mutex); | |
7540 | return ret; | |
7541 | } | |
7542 | ||
d0b27fa7 | 7543 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7544 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7545 | loff_t *ppos) |
7546 | { | |
7547 | int ret; | |
7548 | int old_period, old_runtime; | |
7549 | static DEFINE_MUTEX(mutex); | |
7550 | ||
7551 | mutex_lock(&mutex); | |
7552 | old_period = sysctl_sched_rt_period; | |
7553 | old_runtime = sysctl_sched_rt_runtime; | |
7554 | ||
8d65af78 | 7555 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7556 | |
7557 | if (!ret && write) { | |
7558 | ret = sched_rt_global_constraints(); | |
7559 | if (ret) { | |
7560 | sysctl_sched_rt_period = old_period; | |
7561 | sysctl_sched_rt_runtime = old_runtime; | |
7562 | } else { | |
7563 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7564 | def_rt_bandwidth.rt_period = | |
7565 | ns_to_ktime(global_rt_period()); | |
7566 | } | |
7567 | } | |
7568 | mutex_unlock(&mutex); | |
7569 | ||
7570 | return ret; | |
7571 | } | |
68318b8e | 7572 | |
052f1dc7 | 7573 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7574 | |
7575 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7576 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7577 | { |
2b01dfe3 PM |
7578 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7579 | struct task_group, css); | |
68318b8e SV |
7580 | } |
7581 | ||
92fb9748 | 7582 | static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp) |
68318b8e | 7583 | { |
ec7dc8ac | 7584 | struct task_group *tg, *parent; |
68318b8e | 7585 | |
2b01dfe3 | 7586 | if (!cgrp->parent) { |
68318b8e | 7587 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7588 | return &root_task_group.css; |
68318b8e SV |
7589 | } |
7590 | ||
ec7dc8ac DG |
7591 | parent = cgroup_tg(cgrp->parent); |
7592 | tg = sched_create_group(parent); | |
68318b8e SV |
7593 | if (IS_ERR(tg)) |
7594 | return ERR_PTR(-ENOMEM); | |
7595 | ||
68318b8e SV |
7596 | return &tg->css; |
7597 | } | |
7598 | ||
ace783b9 LZ |
7599 | static int cpu_cgroup_css_online(struct cgroup *cgrp) |
7600 | { | |
7601 | struct task_group *tg = cgroup_tg(cgrp); | |
7602 | struct task_group *parent; | |
7603 | ||
7604 | if (!cgrp->parent) | |
7605 | return 0; | |
7606 | ||
7607 | parent = cgroup_tg(cgrp->parent); | |
7608 | sched_online_group(tg, parent); | |
7609 | return 0; | |
7610 | } | |
7611 | ||
92fb9748 | 7612 | static void cpu_cgroup_css_free(struct cgroup *cgrp) |
68318b8e | 7613 | { |
2b01dfe3 | 7614 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7615 | |
7616 | sched_destroy_group(tg); | |
7617 | } | |
7618 | ||
ace783b9 LZ |
7619 | static void cpu_cgroup_css_offline(struct cgroup *cgrp) |
7620 | { | |
7621 | struct task_group *tg = cgroup_tg(cgrp); | |
7622 | ||
7623 | sched_offline_group(tg); | |
7624 | } | |
7625 | ||
761b3ef5 | 7626 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 7627 | struct cgroup_taskset *tset) |
68318b8e | 7628 | { |
bb9d97b6 TH |
7629 | struct task_struct *task; |
7630 | ||
7631 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7632 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7633 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7634 | return -EINVAL; | |
b68aa230 | 7635 | #else |
bb9d97b6 TH |
7636 | /* We don't support RT-tasks being in separate groups */ |
7637 | if (task->sched_class != &fair_sched_class) | |
7638 | return -EINVAL; | |
b68aa230 | 7639 | #endif |
bb9d97b6 | 7640 | } |
be367d09 BB |
7641 | return 0; |
7642 | } | |
68318b8e | 7643 | |
761b3ef5 | 7644 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 7645 | struct cgroup_taskset *tset) |
68318b8e | 7646 | { |
bb9d97b6 TH |
7647 | struct task_struct *task; |
7648 | ||
7649 | cgroup_taskset_for_each(task, cgrp, tset) | |
7650 | sched_move_task(task); | |
68318b8e SV |
7651 | } |
7652 | ||
068c5cc5 | 7653 | static void |
761b3ef5 LZ |
7654 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
7655 | struct task_struct *task) | |
068c5cc5 PZ |
7656 | { |
7657 | /* | |
7658 | * cgroup_exit() is called in the copy_process() failure path. | |
7659 | * Ignore this case since the task hasn't ran yet, this avoids | |
7660 | * trying to poke a half freed task state from generic code. | |
7661 | */ | |
7662 | if (!(task->flags & PF_EXITING)) | |
7663 | return; | |
7664 | ||
7665 | sched_move_task(task); | |
7666 | } | |
7667 | ||
052f1dc7 | 7668 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7669 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7670 | u64 shareval) |
68318b8e | 7671 | { |
c8b28116 | 7672 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7673 | } |
7674 | ||
f4c753b7 | 7675 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7676 | { |
2b01dfe3 | 7677 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7678 | |
c8b28116 | 7679 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7680 | } |
ab84d31e PT |
7681 | |
7682 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7683 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7684 | ||
ab84d31e PT |
7685 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7686 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7687 | ||
a790de99 PT |
7688 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7689 | ||
ab84d31e PT |
7690 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7691 | { | |
56f570e5 | 7692 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7693 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7694 | |
7695 | if (tg == &root_task_group) | |
7696 | return -EINVAL; | |
7697 | ||
7698 | /* | |
7699 | * Ensure we have at some amount of bandwidth every period. This is | |
7700 | * to prevent reaching a state of large arrears when throttled via | |
7701 | * entity_tick() resulting in prolonged exit starvation. | |
7702 | */ | |
7703 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7704 | return -EINVAL; | |
7705 | ||
7706 | /* | |
7707 | * Likewise, bound things on the otherside by preventing insane quota | |
7708 | * periods. This also allows us to normalize in computing quota | |
7709 | * feasibility. | |
7710 | */ | |
7711 | if (period > max_cfs_quota_period) | |
7712 | return -EINVAL; | |
7713 | ||
a790de99 PT |
7714 | mutex_lock(&cfs_constraints_mutex); |
7715 | ret = __cfs_schedulable(tg, period, quota); | |
7716 | if (ret) | |
7717 | goto out_unlock; | |
7718 | ||
58088ad0 | 7719 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7720 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7721 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7722 | raw_spin_lock_irq(&cfs_b->lock); |
7723 | cfs_b->period = ns_to_ktime(period); | |
7724 | cfs_b->quota = quota; | |
58088ad0 | 7725 | |
a9cf55b2 | 7726 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7727 | /* restart the period timer (if active) to handle new period expiry */ |
7728 | if (runtime_enabled && cfs_b->timer_active) { | |
7729 | /* force a reprogram */ | |
7730 | cfs_b->timer_active = 0; | |
7731 | __start_cfs_bandwidth(cfs_b); | |
7732 | } | |
ab84d31e PT |
7733 | raw_spin_unlock_irq(&cfs_b->lock); |
7734 | ||
7735 | for_each_possible_cpu(i) { | |
7736 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7737 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7738 | |
7739 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7740 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7741 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7742 | |
029632fb | 7743 | if (cfs_rq->throttled) |
671fd9da | 7744 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7745 | raw_spin_unlock_irq(&rq->lock); |
7746 | } | |
a790de99 PT |
7747 | out_unlock: |
7748 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7749 | |
a790de99 | 7750 | return ret; |
ab84d31e PT |
7751 | } |
7752 | ||
7753 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7754 | { | |
7755 | u64 quota, period; | |
7756 | ||
029632fb | 7757 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7758 | if (cfs_quota_us < 0) |
7759 | quota = RUNTIME_INF; | |
7760 | else | |
7761 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7762 | ||
7763 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7764 | } | |
7765 | ||
7766 | long tg_get_cfs_quota(struct task_group *tg) | |
7767 | { | |
7768 | u64 quota_us; | |
7769 | ||
029632fb | 7770 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7771 | return -1; |
7772 | ||
029632fb | 7773 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7774 | do_div(quota_us, NSEC_PER_USEC); |
7775 | ||
7776 | return quota_us; | |
7777 | } | |
7778 | ||
7779 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7780 | { | |
7781 | u64 quota, period; | |
7782 | ||
7783 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7784 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7785 | |
ab84d31e PT |
7786 | return tg_set_cfs_bandwidth(tg, period, quota); |
7787 | } | |
7788 | ||
7789 | long tg_get_cfs_period(struct task_group *tg) | |
7790 | { | |
7791 | u64 cfs_period_us; | |
7792 | ||
029632fb | 7793 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7794 | do_div(cfs_period_us, NSEC_PER_USEC); |
7795 | ||
7796 | return cfs_period_us; | |
7797 | } | |
7798 | ||
7799 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7800 | { | |
7801 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7802 | } | |
7803 | ||
7804 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7805 | s64 cfs_quota_us) | |
7806 | { | |
7807 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7808 | } | |
7809 | ||
7810 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7811 | { | |
7812 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7813 | } | |
7814 | ||
7815 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7816 | u64 cfs_period_us) | |
7817 | { | |
7818 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7819 | } | |
7820 | ||
a790de99 PT |
7821 | struct cfs_schedulable_data { |
7822 | struct task_group *tg; | |
7823 | u64 period, quota; | |
7824 | }; | |
7825 | ||
7826 | /* | |
7827 | * normalize group quota/period to be quota/max_period | |
7828 | * note: units are usecs | |
7829 | */ | |
7830 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7831 | struct cfs_schedulable_data *d) | |
7832 | { | |
7833 | u64 quota, period; | |
7834 | ||
7835 | if (tg == d->tg) { | |
7836 | period = d->period; | |
7837 | quota = d->quota; | |
7838 | } else { | |
7839 | period = tg_get_cfs_period(tg); | |
7840 | quota = tg_get_cfs_quota(tg); | |
7841 | } | |
7842 | ||
7843 | /* note: these should typically be equivalent */ | |
7844 | if (quota == RUNTIME_INF || quota == -1) | |
7845 | return RUNTIME_INF; | |
7846 | ||
7847 | return to_ratio(period, quota); | |
7848 | } | |
7849 | ||
7850 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7851 | { | |
7852 | struct cfs_schedulable_data *d = data; | |
029632fb | 7853 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7854 | s64 quota = 0, parent_quota = -1; |
7855 | ||
7856 | if (!tg->parent) { | |
7857 | quota = RUNTIME_INF; | |
7858 | } else { | |
029632fb | 7859 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7860 | |
7861 | quota = normalize_cfs_quota(tg, d); | |
7862 | parent_quota = parent_b->hierarchal_quota; | |
7863 | ||
7864 | /* | |
7865 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7866 | * limit is set | |
7867 | */ | |
7868 | if (quota == RUNTIME_INF) | |
7869 | quota = parent_quota; | |
7870 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7871 | return -EINVAL; | |
7872 | } | |
7873 | cfs_b->hierarchal_quota = quota; | |
7874 | ||
7875 | return 0; | |
7876 | } | |
7877 | ||
7878 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7879 | { | |
8277434e | 7880 | int ret; |
a790de99 PT |
7881 | struct cfs_schedulable_data data = { |
7882 | .tg = tg, | |
7883 | .period = period, | |
7884 | .quota = quota, | |
7885 | }; | |
7886 | ||
7887 | if (quota != RUNTIME_INF) { | |
7888 | do_div(data.period, NSEC_PER_USEC); | |
7889 | do_div(data.quota, NSEC_PER_USEC); | |
7890 | } | |
7891 | ||
8277434e PT |
7892 | rcu_read_lock(); |
7893 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7894 | rcu_read_unlock(); | |
7895 | ||
7896 | return ret; | |
a790de99 | 7897 | } |
e8da1b18 NR |
7898 | |
7899 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7900 | struct cgroup_map_cb *cb) | |
7901 | { | |
7902 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7903 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7904 | |
7905 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7906 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7907 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7908 | ||
7909 | return 0; | |
7910 | } | |
ab84d31e | 7911 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7912 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7913 | |
052f1dc7 | 7914 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7915 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7916 | s64 val) |
6f505b16 | 7917 | { |
06ecb27c | 7918 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7919 | } |
7920 | ||
06ecb27c | 7921 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7922 | { |
06ecb27c | 7923 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7924 | } |
d0b27fa7 PZ |
7925 | |
7926 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7927 | u64 rt_period_us) | |
7928 | { | |
7929 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7930 | } | |
7931 | ||
7932 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7933 | { | |
7934 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7935 | } | |
6d6bc0ad | 7936 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7937 | |
fe5c7cc2 | 7938 | static struct cftype cpu_files[] = { |
052f1dc7 | 7939 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7940 | { |
7941 | .name = "shares", | |
f4c753b7 PM |
7942 | .read_u64 = cpu_shares_read_u64, |
7943 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7944 | }, |
052f1dc7 | 7945 | #endif |
ab84d31e PT |
7946 | #ifdef CONFIG_CFS_BANDWIDTH |
7947 | { | |
7948 | .name = "cfs_quota_us", | |
7949 | .read_s64 = cpu_cfs_quota_read_s64, | |
7950 | .write_s64 = cpu_cfs_quota_write_s64, | |
7951 | }, | |
7952 | { | |
7953 | .name = "cfs_period_us", | |
7954 | .read_u64 = cpu_cfs_period_read_u64, | |
7955 | .write_u64 = cpu_cfs_period_write_u64, | |
7956 | }, | |
e8da1b18 NR |
7957 | { |
7958 | .name = "stat", | |
7959 | .read_map = cpu_stats_show, | |
7960 | }, | |
ab84d31e | 7961 | #endif |
052f1dc7 | 7962 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7963 | { |
9f0c1e56 | 7964 | .name = "rt_runtime_us", |
06ecb27c PM |
7965 | .read_s64 = cpu_rt_runtime_read, |
7966 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7967 | }, |
d0b27fa7 PZ |
7968 | { |
7969 | .name = "rt_period_us", | |
f4c753b7 PM |
7970 | .read_u64 = cpu_rt_period_read_uint, |
7971 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7972 | }, |
052f1dc7 | 7973 | #endif |
4baf6e33 | 7974 | { } /* terminate */ |
68318b8e SV |
7975 | }; |
7976 | ||
68318b8e | 7977 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 7978 | .name = "cpu", |
92fb9748 TH |
7979 | .css_alloc = cpu_cgroup_css_alloc, |
7980 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
7981 | .css_online = cpu_cgroup_css_online, |
7982 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
7983 | .can_attach = cpu_cgroup_can_attach, |
7984 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7985 | .exit = cpu_cgroup_exit, |
38605cae | 7986 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 7987 | .base_cftypes = cpu_files, |
68318b8e SV |
7988 | .early_init = 1, |
7989 | }; | |
7990 | ||
052f1dc7 | 7991 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7992 | |
7993 | #ifdef CONFIG_CGROUP_CPUACCT | |
7994 | ||
7995 | /* | |
7996 | * CPU accounting code for task groups. | |
7997 | * | |
7998 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7999 | * (balbir@in.ibm.com). | |
8000 | */ | |
8001 | ||
73fbec60 FW |
8002 | struct cpuacct root_cpuacct; |
8003 | ||
d842de87 | 8004 | /* create a new cpu accounting group */ |
92fb9748 | 8005 | static struct cgroup_subsys_state *cpuacct_css_alloc(struct cgroup *cgrp) |
d842de87 | 8006 | { |
54c707e9 | 8007 | struct cpuacct *ca; |
d842de87 | 8008 | |
54c707e9 GC |
8009 | if (!cgrp->parent) |
8010 | return &root_cpuacct.css; | |
8011 | ||
8012 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 8013 | if (!ca) |
ef12fefa | 8014 | goto out; |
d842de87 SV |
8015 | |
8016 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8017 | if (!ca->cpuusage) |
8018 | goto out_free_ca; | |
8019 | ||
54c707e9 GC |
8020 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
8021 | if (!ca->cpustat) | |
8022 | goto out_free_cpuusage; | |
934352f2 | 8023 | |
d842de87 | 8024 | return &ca->css; |
ef12fefa | 8025 | |
54c707e9 | 8026 | out_free_cpuusage: |
ef12fefa BR |
8027 | free_percpu(ca->cpuusage); |
8028 | out_free_ca: | |
8029 | kfree(ca); | |
8030 | out: | |
8031 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8032 | } |
8033 | ||
8034 | /* destroy an existing cpu accounting group */ | |
92fb9748 | 8035 | static void cpuacct_css_free(struct cgroup *cgrp) |
d842de87 | 8036 | { |
32cd756a | 8037 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 8038 | |
54c707e9 | 8039 | free_percpu(ca->cpustat); |
d842de87 SV |
8040 | free_percpu(ca->cpuusage); |
8041 | kfree(ca); | |
8042 | } | |
8043 | ||
720f5498 KC |
8044 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8045 | { | |
b36128c8 | 8046 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8047 | u64 data; |
8048 | ||
8049 | #ifndef CONFIG_64BIT | |
8050 | /* | |
8051 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8052 | */ | |
05fa785c | 8053 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8054 | data = *cpuusage; |
05fa785c | 8055 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8056 | #else |
8057 | data = *cpuusage; | |
8058 | #endif | |
8059 | ||
8060 | return data; | |
8061 | } | |
8062 | ||
8063 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8064 | { | |
b36128c8 | 8065 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8066 | |
8067 | #ifndef CONFIG_64BIT | |
8068 | /* | |
8069 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8070 | */ | |
05fa785c | 8071 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8072 | *cpuusage = val; |
05fa785c | 8073 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8074 | #else |
8075 | *cpuusage = val; | |
8076 | #endif | |
8077 | } | |
8078 | ||
d842de87 | 8079 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8080 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8081 | { |
32cd756a | 8082 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8083 | u64 totalcpuusage = 0; |
8084 | int i; | |
8085 | ||
720f5498 KC |
8086 | for_each_present_cpu(i) |
8087 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8088 | |
8089 | return totalcpuusage; | |
8090 | } | |
8091 | ||
0297b803 DG |
8092 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8093 | u64 reset) | |
8094 | { | |
8095 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8096 | int err = 0; | |
8097 | int i; | |
8098 | ||
8099 | if (reset) { | |
8100 | err = -EINVAL; | |
8101 | goto out; | |
8102 | } | |
8103 | ||
720f5498 KC |
8104 | for_each_present_cpu(i) |
8105 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8106 | |
0297b803 DG |
8107 | out: |
8108 | return err; | |
8109 | } | |
8110 | ||
e9515c3c KC |
8111 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8112 | struct seq_file *m) | |
8113 | { | |
8114 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8115 | u64 percpu; | |
8116 | int i; | |
8117 | ||
8118 | for_each_present_cpu(i) { | |
8119 | percpu = cpuacct_cpuusage_read(ca, i); | |
8120 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8121 | } | |
8122 | seq_printf(m, "\n"); | |
8123 | return 0; | |
8124 | } | |
8125 | ||
ef12fefa BR |
8126 | static const char *cpuacct_stat_desc[] = { |
8127 | [CPUACCT_STAT_USER] = "user", | |
8128 | [CPUACCT_STAT_SYSTEM] = "system", | |
8129 | }; | |
8130 | ||
8131 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8132 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8133 | { |
8134 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8135 | int cpu; |
8136 | s64 val = 0; | |
ef12fefa | 8137 | |
54c707e9 GC |
8138 | for_each_online_cpu(cpu) { |
8139 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8140 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8141 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8142 | } |
54c707e9 GC |
8143 | val = cputime64_to_clock_t(val); |
8144 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8145 | |
54c707e9 GC |
8146 | val = 0; |
8147 | for_each_online_cpu(cpu) { | |
8148 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8149 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8150 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8151 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8152 | } |
54c707e9 GC |
8153 | |
8154 | val = cputime64_to_clock_t(val); | |
8155 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8156 | ||
ef12fefa BR |
8157 | return 0; |
8158 | } | |
8159 | ||
d842de87 SV |
8160 | static struct cftype files[] = { |
8161 | { | |
8162 | .name = "usage", | |
f4c753b7 PM |
8163 | .read_u64 = cpuusage_read, |
8164 | .write_u64 = cpuusage_write, | |
d842de87 | 8165 | }, |
e9515c3c KC |
8166 | { |
8167 | .name = "usage_percpu", | |
8168 | .read_seq_string = cpuacct_percpu_seq_read, | |
8169 | }, | |
ef12fefa BR |
8170 | { |
8171 | .name = "stat", | |
8172 | .read_map = cpuacct_stats_show, | |
8173 | }, | |
4baf6e33 | 8174 | { } /* terminate */ |
d842de87 SV |
8175 | }; |
8176 | ||
d842de87 SV |
8177 | /* |
8178 | * charge this task's execution time to its accounting group. | |
8179 | * | |
8180 | * called with rq->lock held. | |
8181 | */ | |
029632fb | 8182 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8183 | { |
8184 | struct cpuacct *ca; | |
934352f2 | 8185 | int cpu; |
d842de87 | 8186 | |
c40c6f85 | 8187 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8188 | return; |
8189 | ||
934352f2 | 8190 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8191 | |
8192 | rcu_read_lock(); | |
8193 | ||
d842de87 | 8194 | ca = task_ca(tsk); |
d842de87 | 8195 | |
44252e42 | 8196 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8197 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8198 | *cpuusage += cputime; |
8199 | } | |
a18b83b7 BR |
8200 | |
8201 | rcu_read_unlock(); | |
d842de87 SV |
8202 | } |
8203 | ||
8204 | struct cgroup_subsys cpuacct_subsys = { | |
8205 | .name = "cpuacct", | |
92fb9748 TH |
8206 | .css_alloc = cpuacct_css_alloc, |
8207 | .css_free = cpuacct_css_free, | |
d842de87 | 8208 | .subsys_id = cpuacct_subsys_id, |
4baf6e33 | 8209 | .base_cftypes = files, |
d842de87 SV |
8210 | }; |
8211 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
b637a328 PM |
8212 | |
8213 | void dump_cpu_task(int cpu) | |
8214 | { | |
8215 | pr_info("Task dump for CPU %d:\n", cpu); | |
8216 | sched_show_task(cpu_curr(cpu)); | |
8217 | } |