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. | |
8f4d37ec | 373 | */ |
8f4d37ec | 374 | |
8f4d37ec PZ |
375 | static void hrtick_clear(struct rq *rq) |
376 | { | |
377 | if (hrtimer_active(&rq->hrtick_timer)) | |
378 | hrtimer_cancel(&rq->hrtick_timer); | |
379 | } | |
380 | ||
8f4d37ec PZ |
381 | /* |
382 | * High-resolution timer tick. | |
383 | * Runs from hardirq context with interrupts disabled. | |
384 | */ | |
385 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
386 | { | |
387 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
388 | ||
389 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
390 | ||
05fa785c | 391 | raw_spin_lock(&rq->lock); |
3e51f33f | 392 | update_rq_clock(rq); |
8f4d37ec | 393 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 394 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
395 | |
396 | return HRTIMER_NORESTART; | |
397 | } | |
398 | ||
95e904c7 | 399 | #ifdef CONFIG_SMP |
971ee28c PZ |
400 | |
401 | static int __hrtick_restart(struct rq *rq) | |
402 | { | |
403 | struct hrtimer *timer = &rq->hrtick_timer; | |
404 | ktime_t time = hrtimer_get_softexpires(timer); | |
405 | ||
406 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
407 | } | |
408 | ||
31656519 PZ |
409 | /* |
410 | * called from hardirq (IPI) context | |
411 | */ | |
412 | static void __hrtick_start(void *arg) | |
b328ca18 | 413 | { |
31656519 | 414 | struct rq *rq = arg; |
b328ca18 | 415 | |
05fa785c | 416 | raw_spin_lock(&rq->lock); |
971ee28c | 417 | __hrtick_restart(rq); |
31656519 | 418 | rq->hrtick_csd_pending = 0; |
05fa785c | 419 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
420 | } |
421 | ||
31656519 PZ |
422 | /* |
423 | * Called to set the hrtick timer state. | |
424 | * | |
425 | * called with rq->lock held and irqs disabled | |
426 | */ | |
029632fb | 427 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 428 | { |
31656519 PZ |
429 | struct hrtimer *timer = &rq->hrtick_timer; |
430 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 431 | |
cc584b21 | 432 | hrtimer_set_expires(timer, time); |
31656519 PZ |
433 | |
434 | if (rq == this_rq()) { | |
971ee28c | 435 | __hrtick_restart(rq); |
31656519 | 436 | } else if (!rq->hrtick_csd_pending) { |
6e275637 | 437 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
438 | rq->hrtick_csd_pending = 1; |
439 | } | |
b328ca18 PZ |
440 | } |
441 | ||
442 | static int | |
443 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
444 | { | |
445 | int cpu = (int)(long)hcpu; | |
446 | ||
447 | switch (action) { | |
448 | case CPU_UP_CANCELED: | |
449 | case CPU_UP_CANCELED_FROZEN: | |
450 | case CPU_DOWN_PREPARE: | |
451 | case CPU_DOWN_PREPARE_FROZEN: | |
452 | case CPU_DEAD: | |
453 | case CPU_DEAD_FROZEN: | |
31656519 | 454 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
455 | return NOTIFY_OK; |
456 | } | |
457 | ||
458 | return NOTIFY_DONE; | |
459 | } | |
460 | ||
fa748203 | 461 | static __init void init_hrtick(void) |
b328ca18 PZ |
462 | { |
463 | hotcpu_notifier(hotplug_hrtick, 0); | |
464 | } | |
31656519 PZ |
465 | #else |
466 | /* | |
467 | * Called to set the hrtick timer state. | |
468 | * | |
469 | * called with rq->lock held and irqs disabled | |
470 | */ | |
029632fb | 471 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 472 | { |
7f1e2ca9 | 473 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 474 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 475 | } |
b328ca18 | 476 | |
006c75f1 | 477 | static inline void init_hrtick(void) |
8f4d37ec | 478 | { |
8f4d37ec | 479 | } |
31656519 | 480 | #endif /* CONFIG_SMP */ |
8f4d37ec | 481 | |
31656519 | 482 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 483 | { |
31656519 PZ |
484 | #ifdef CONFIG_SMP |
485 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 486 | |
31656519 PZ |
487 | rq->hrtick_csd.flags = 0; |
488 | rq->hrtick_csd.func = __hrtick_start; | |
489 | rq->hrtick_csd.info = rq; | |
490 | #endif | |
8f4d37ec | 491 | |
31656519 PZ |
492 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
493 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 494 | } |
006c75f1 | 495 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
496 | static inline void hrtick_clear(struct rq *rq) |
497 | { | |
498 | } | |
499 | ||
8f4d37ec PZ |
500 | static inline void init_rq_hrtick(struct rq *rq) |
501 | { | |
502 | } | |
503 | ||
b328ca18 PZ |
504 | static inline void init_hrtick(void) |
505 | { | |
506 | } | |
006c75f1 | 507 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 508 | |
c24d20db IM |
509 | /* |
510 | * resched_task - mark a task 'to be rescheduled now'. | |
511 | * | |
512 | * On UP this means the setting of the need_resched flag, on SMP it | |
513 | * might also involve a cross-CPU call to trigger the scheduler on | |
514 | * the target CPU. | |
515 | */ | |
029632fb | 516 | void resched_task(struct task_struct *p) |
c24d20db IM |
517 | { |
518 | int cpu; | |
519 | ||
b021fe3e | 520 | lockdep_assert_held(&task_rq(p)->lock); |
c24d20db | 521 | |
5ed0cec0 | 522 | if (test_tsk_need_resched(p)) |
c24d20db IM |
523 | return; |
524 | ||
5ed0cec0 | 525 | set_tsk_need_resched(p); |
c24d20db IM |
526 | |
527 | cpu = task_cpu(p); | |
f27dde8d PZ |
528 | if (cpu == smp_processor_id()) { |
529 | set_preempt_need_resched(); | |
c24d20db | 530 | return; |
f27dde8d | 531 | } |
c24d20db IM |
532 | |
533 | /* NEED_RESCHED must be visible before we test polling */ | |
534 | smp_mb(); | |
535 | if (!tsk_is_polling(p)) | |
536 | smp_send_reschedule(cpu); | |
537 | } | |
538 | ||
029632fb | 539 | void resched_cpu(int cpu) |
c24d20db IM |
540 | { |
541 | struct rq *rq = cpu_rq(cpu); | |
542 | unsigned long flags; | |
543 | ||
05fa785c | 544 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
545 | return; |
546 | resched_task(cpu_curr(cpu)); | |
05fa785c | 547 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 548 | } |
06d8308c | 549 | |
b021fe3e | 550 | #ifdef CONFIG_SMP |
3451d024 | 551 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
552 | /* |
553 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
554 | * from an idle cpu. This is good for power-savings. | |
555 | * | |
556 | * We don't do similar optimization for completely idle system, as | |
557 | * selecting an idle cpu will add more delays to the timers than intended | |
558 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
559 | */ | |
560 | int get_nohz_timer_target(void) | |
561 | { | |
562 | int cpu = smp_processor_id(); | |
563 | int i; | |
564 | struct sched_domain *sd; | |
565 | ||
057f3fad | 566 | rcu_read_lock(); |
83cd4fe2 | 567 | for_each_domain(cpu, sd) { |
057f3fad PZ |
568 | for_each_cpu(i, sched_domain_span(sd)) { |
569 | if (!idle_cpu(i)) { | |
570 | cpu = i; | |
571 | goto unlock; | |
572 | } | |
573 | } | |
83cd4fe2 | 574 | } |
057f3fad PZ |
575 | unlock: |
576 | rcu_read_unlock(); | |
83cd4fe2 VP |
577 | return cpu; |
578 | } | |
06d8308c TG |
579 | /* |
580 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
581 | * idle CPU then this timer might expire before the next timer event | |
582 | * which is scheduled to wake up that CPU. In case of a completely | |
583 | * idle system the next event might even be infinite time into the | |
584 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
585 | * leaves the inner idle loop so the newly added timer is taken into | |
586 | * account when the CPU goes back to idle and evaluates the timer | |
587 | * wheel for the next timer event. | |
588 | */ | |
1c20091e | 589 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
590 | { |
591 | struct rq *rq = cpu_rq(cpu); | |
592 | ||
593 | if (cpu == smp_processor_id()) | |
594 | return; | |
595 | ||
596 | /* | |
597 | * This is safe, as this function is called with the timer | |
598 | * wheel base lock of (cpu) held. When the CPU is on the way | |
599 | * to idle and has not yet set rq->curr to idle then it will | |
600 | * be serialized on the timer wheel base lock and take the new | |
601 | * timer into account automatically. | |
602 | */ | |
603 | if (rq->curr != rq->idle) | |
604 | return; | |
45bf76df | 605 | |
45bf76df | 606 | /* |
06d8308c TG |
607 | * We can set TIF_RESCHED on the idle task of the other CPU |
608 | * lockless. The worst case is that the other CPU runs the | |
609 | * idle task through an additional NOOP schedule() | |
45bf76df | 610 | */ |
5ed0cec0 | 611 | set_tsk_need_resched(rq->idle); |
45bf76df | 612 | |
06d8308c TG |
613 | /* NEED_RESCHED must be visible before we test polling */ |
614 | smp_mb(); | |
615 | if (!tsk_is_polling(rq->idle)) | |
616 | smp_send_reschedule(cpu); | |
45bf76df IM |
617 | } |
618 | ||
c5bfece2 | 619 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 620 | { |
c5bfece2 | 621 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
622 | if (cpu != smp_processor_id() || |
623 | tick_nohz_tick_stopped()) | |
624 | smp_send_reschedule(cpu); | |
625 | return true; | |
626 | } | |
627 | ||
628 | return false; | |
629 | } | |
630 | ||
631 | void wake_up_nohz_cpu(int cpu) | |
632 | { | |
c5bfece2 | 633 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
634 | wake_up_idle_cpu(cpu); |
635 | } | |
636 | ||
ca38062e | 637 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 638 | { |
1c792db7 | 639 | int cpu = smp_processor_id(); |
873b4c65 VG |
640 | |
641 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
642 | return false; | |
643 | ||
644 | if (idle_cpu(cpu) && !need_resched()) | |
645 | return true; | |
646 | ||
647 | /* | |
648 | * We can't run Idle Load Balance on this CPU for this time so we | |
649 | * cancel it and clear NOHZ_BALANCE_KICK | |
650 | */ | |
651 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
652 | return false; | |
45bf76df IM |
653 | } |
654 | ||
3451d024 | 655 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 656 | |
ca38062e | 657 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 658 | { |
ca38062e | 659 | return false; |
2069dd75 PZ |
660 | } |
661 | ||
3451d024 | 662 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 663 | |
ce831b38 FW |
664 | #ifdef CONFIG_NO_HZ_FULL |
665 | bool sched_can_stop_tick(void) | |
666 | { | |
667 | struct rq *rq; | |
668 | ||
669 | rq = this_rq(); | |
670 | ||
671 | /* Make sure rq->nr_running update is visible after the IPI */ | |
672 | smp_rmb(); | |
673 | ||
674 | /* More than one running task need preemption */ | |
675 | if (rq->nr_running > 1) | |
676 | return false; | |
677 | ||
678 | return true; | |
679 | } | |
680 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 681 | |
029632fb | 682 | void sched_avg_update(struct rq *rq) |
18d95a28 | 683 | { |
e9e9250b PZ |
684 | s64 period = sched_avg_period(); |
685 | ||
78becc27 | 686 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
687 | /* |
688 | * Inline assembly required to prevent the compiler | |
689 | * optimising this loop into a divmod call. | |
690 | * See __iter_div_u64_rem() for another example of this. | |
691 | */ | |
692 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
693 | rq->age_stamp += period; |
694 | rq->rt_avg /= 2; | |
695 | } | |
18d95a28 PZ |
696 | } |
697 | ||
6d6bc0ad | 698 | #endif /* CONFIG_SMP */ |
18d95a28 | 699 | |
a790de99 PT |
700 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
701 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 702 | /* |
8277434e PT |
703 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
704 | * node and @up when leaving it for the final time. | |
705 | * | |
706 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 707 | */ |
029632fb | 708 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 709 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
710 | { |
711 | struct task_group *parent, *child; | |
eb755805 | 712 | int ret; |
c09595f6 | 713 | |
8277434e PT |
714 | parent = from; |
715 | ||
c09595f6 | 716 | down: |
eb755805 PZ |
717 | ret = (*down)(parent, data); |
718 | if (ret) | |
8277434e | 719 | goto out; |
c09595f6 PZ |
720 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
721 | parent = child; | |
722 | goto down; | |
723 | ||
724 | up: | |
725 | continue; | |
726 | } | |
eb755805 | 727 | ret = (*up)(parent, data); |
8277434e PT |
728 | if (ret || parent == from) |
729 | goto out; | |
c09595f6 PZ |
730 | |
731 | child = parent; | |
732 | parent = parent->parent; | |
733 | if (parent) | |
734 | goto up; | |
8277434e | 735 | out: |
eb755805 | 736 | return ret; |
c09595f6 PZ |
737 | } |
738 | ||
029632fb | 739 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 740 | { |
e2b245f8 | 741 | return 0; |
eb755805 | 742 | } |
18d95a28 PZ |
743 | #endif |
744 | ||
45bf76df IM |
745 | static void set_load_weight(struct task_struct *p) |
746 | { | |
f05998d4 NR |
747 | int prio = p->static_prio - MAX_RT_PRIO; |
748 | struct load_weight *load = &p->se.load; | |
749 | ||
dd41f596 IM |
750 | /* |
751 | * SCHED_IDLE tasks get minimal weight: | |
752 | */ | |
753 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 754 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 755 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
756 | return; |
757 | } | |
71f8bd46 | 758 | |
c8b28116 | 759 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 760 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
761 | } |
762 | ||
371fd7e7 | 763 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 764 | { |
a64692a3 | 765 | update_rq_clock(rq); |
43148951 | 766 | sched_info_queued(rq, p); |
371fd7e7 | 767 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
768 | } |
769 | ||
371fd7e7 | 770 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 771 | { |
a64692a3 | 772 | update_rq_clock(rq); |
43148951 | 773 | sched_info_dequeued(rq, p); |
371fd7e7 | 774 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
775 | } |
776 | ||
029632fb | 777 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
778 | { |
779 | if (task_contributes_to_load(p)) | |
780 | rq->nr_uninterruptible--; | |
781 | ||
371fd7e7 | 782 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
783 | } |
784 | ||
029632fb | 785 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
786 | { |
787 | if (task_contributes_to_load(p)) | |
788 | rq->nr_uninterruptible++; | |
789 | ||
371fd7e7 | 790 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
791 | } |
792 | ||
fe44d621 | 793 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 794 | { |
095c0aa8 GC |
795 | /* |
796 | * In theory, the compile should just see 0 here, and optimize out the call | |
797 | * to sched_rt_avg_update. But I don't trust it... | |
798 | */ | |
799 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
800 | s64 steal = 0, irq_delta = 0; | |
801 | #endif | |
802 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 803 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
804 | |
805 | /* | |
806 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
807 | * this case when a previous update_rq_clock() happened inside a | |
808 | * {soft,}irq region. | |
809 | * | |
810 | * When this happens, we stop ->clock_task and only update the | |
811 | * prev_irq_time stamp to account for the part that fit, so that a next | |
812 | * update will consume the rest. This ensures ->clock_task is | |
813 | * monotonic. | |
814 | * | |
815 | * It does however cause some slight miss-attribution of {soft,}irq | |
816 | * time, a more accurate solution would be to update the irq_time using | |
817 | * the current rq->clock timestamp, except that would require using | |
818 | * atomic ops. | |
819 | */ | |
820 | if (irq_delta > delta) | |
821 | irq_delta = delta; | |
822 | ||
823 | rq->prev_irq_time += irq_delta; | |
824 | delta -= irq_delta; | |
095c0aa8 GC |
825 | #endif |
826 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 827 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
828 | u64 st; |
829 | ||
830 | steal = paravirt_steal_clock(cpu_of(rq)); | |
831 | steal -= rq->prev_steal_time_rq; | |
832 | ||
833 | if (unlikely(steal > delta)) | |
834 | steal = delta; | |
835 | ||
836 | st = steal_ticks(steal); | |
837 | steal = st * TICK_NSEC; | |
838 | ||
839 | rq->prev_steal_time_rq += steal; | |
840 | ||
841 | delta -= steal; | |
842 | } | |
843 | #endif | |
844 | ||
fe44d621 PZ |
845 | rq->clock_task += delta; |
846 | ||
095c0aa8 GC |
847 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
848 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
849 | sched_rt_avg_update(rq, irq_delta + steal); | |
850 | #endif | |
aa483808 VP |
851 | } |
852 | ||
34f971f6 PZ |
853 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
854 | { | |
855 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
856 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
857 | ||
858 | if (stop) { | |
859 | /* | |
860 | * Make it appear like a SCHED_FIFO task, its something | |
861 | * userspace knows about and won't get confused about. | |
862 | * | |
863 | * Also, it will make PI more or less work without too | |
864 | * much confusion -- but then, stop work should not | |
865 | * rely on PI working anyway. | |
866 | */ | |
867 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
868 | ||
869 | stop->sched_class = &stop_sched_class; | |
870 | } | |
871 | ||
872 | cpu_rq(cpu)->stop = stop; | |
873 | ||
874 | if (old_stop) { | |
875 | /* | |
876 | * Reset it back to a normal scheduling class so that | |
877 | * it can die in pieces. | |
878 | */ | |
879 | old_stop->sched_class = &rt_sched_class; | |
880 | } | |
881 | } | |
882 | ||
14531189 | 883 | /* |
dd41f596 | 884 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 885 | */ |
14531189 IM |
886 | static inline int __normal_prio(struct task_struct *p) |
887 | { | |
dd41f596 | 888 | return p->static_prio; |
14531189 IM |
889 | } |
890 | ||
b29739f9 IM |
891 | /* |
892 | * Calculate the expected normal priority: i.e. priority | |
893 | * without taking RT-inheritance into account. Might be | |
894 | * boosted by interactivity modifiers. Changes upon fork, | |
895 | * setprio syscalls, and whenever the interactivity | |
896 | * estimator recalculates. | |
897 | */ | |
36c8b586 | 898 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
899 | { |
900 | int prio; | |
901 | ||
e05606d3 | 902 | if (task_has_rt_policy(p)) |
b29739f9 IM |
903 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
904 | else | |
905 | prio = __normal_prio(p); | |
906 | return prio; | |
907 | } | |
908 | ||
909 | /* | |
910 | * Calculate the current priority, i.e. the priority | |
911 | * taken into account by the scheduler. This value might | |
912 | * be boosted by RT tasks, or might be boosted by | |
913 | * interactivity modifiers. Will be RT if the task got | |
914 | * RT-boosted. If not then it returns p->normal_prio. | |
915 | */ | |
36c8b586 | 916 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
917 | { |
918 | p->normal_prio = normal_prio(p); | |
919 | /* | |
920 | * If we are RT tasks or we were boosted to RT priority, | |
921 | * keep the priority unchanged. Otherwise, update priority | |
922 | * to the normal priority: | |
923 | */ | |
924 | if (!rt_prio(p->prio)) | |
925 | return p->normal_prio; | |
926 | return p->prio; | |
927 | } | |
928 | ||
1da177e4 LT |
929 | /** |
930 | * task_curr - is this task currently executing on a CPU? | |
931 | * @p: the task in question. | |
e69f6186 YB |
932 | * |
933 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 934 | */ |
36c8b586 | 935 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
936 | { |
937 | return cpu_curr(task_cpu(p)) == p; | |
938 | } | |
939 | ||
cb469845 SR |
940 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
941 | const struct sched_class *prev_class, | |
da7a735e | 942 | int oldprio) |
cb469845 SR |
943 | { |
944 | if (prev_class != p->sched_class) { | |
945 | if (prev_class->switched_from) | |
da7a735e PZ |
946 | prev_class->switched_from(rq, p); |
947 | p->sched_class->switched_to(rq, p); | |
948 | } else if (oldprio != p->prio) | |
949 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
950 | } |
951 | ||
029632fb | 952 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
953 | { |
954 | const struct sched_class *class; | |
955 | ||
956 | if (p->sched_class == rq->curr->sched_class) { | |
957 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
958 | } else { | |
959 | for_each_class(class) { | |
960 | if (class == rq->curr->sched_class) | |
961 | break; | |
962 | if (class == p->sched_class) { | |
963 | resched_task(rq->curr); | |
964 | break; | |
965 | } | |
966 | } | |
967 | } | |
968 | ||
969 | /* | |
970 | * A queue event has occurred, and we're going to schedule. In | |
971 | * this case, we can save a useless back to back clock update. | |
972 | */ | |
fd2f4419 | 973 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
974 | rq->skip_clock_update = 1; |
975 | } | |
976 | ||
1da177e4 | 977 | #ifdef CONFIG_SMP |
dd41f596 | 978 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 979 | { |
e2912009 PZ |
980 | #ifdef CONFIG_SCHED_DEBUG |
981 | /* | |
982 | * We should never call set_task_cpu() on a blocked task, | |
983 | * ttwu() will sort out the placement. | |
984 | */ | |
077614ee | 985 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 986 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
987 | |
988 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
989 | /* |
990 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
991 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
992 | * | |
993 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 994 | * see task_group(). |
6c6c54e1 PZ |
995 | * |
996 | * Furthermore, all task_rq users should acquire both locks, see | |
997 | * task_rq_lock(). | |
998 | */ | |
0122ec5b PZ |
999 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1000 | lockdep_is_held(&task_rq(p)->lock))); | |
1001 | #endif | |
e2912009 PZ |
1002 | #endif |
1003 | ||
de1d7286 | 1004 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1005 | |
0c69774e | 1006 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1007 | if (p->sched_class->migrate_task_rq) |
1008 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1009 | p->se.nr_migrations++; |
a8b0ca17 | 1010 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1011 | } |
dd41f596 IM |
1012 | |
1013 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1014 | } |
1015 | ||
ac66f547 PZ |
1016 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1017 | { | |
1018 | if (p->on_rq) { | |
1019 | struct rq *src_rq, *dst_rq; | |
1020 | ||
1021 | src_rq = task_rq(p); | |
1022 | dst_rq = cpu_rq(cpu); | |
1023 | ||
1024 | deactivate_task(src_rq, p, 0); | |
1025 | set_task_cpu(p, cpu); | |
1026 | activate_task(dst_rq, p, 0); | |
1027 | check_preempt_curr(dst_rq, p, 0); | |
1028 | } else { | |
1029 | /* | |
1030 | * Task isn't running anymore; make it appear like we migrated | |
1031 | * it before it went to sleep. This means on wakeup we make the | |
1032 | * previous cpu our targer instead of where it really is. | |
1033 | */ | |
1034 | p->wake_cpu = cpu; | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | struct migration_swap_arg { | |
1039 | struct task_struct *src_task, *dst_task; | |
1040 | int src_cpu, dst_cpu; | |
1041 | }; | |
1042 | ||
1043 | static int migrate_swap_stop(void *data) | |
1044 | { | |
1045 | struct migration_swap_arg *arg = data; | |
1046 | struct rq *src_rq, *dst_rq; | |
1047 | int ret = -EAGAIN; | |
1048 | ||
1049 | src_rq = cpu_rq(arg->src_cpu); | |
1050 | dst_rq = cpu_rq(arg->dst_cpu); | |
1051 | ||
1052 | double_rq_lock(src_rq, dst_rq); | |
1053 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1054 | goto unlock; | |
1055 | ||
1056 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1057 | goto unlock; | |
1058 | ||
1059 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1060 | goto unlock; | |
1061 | ||
1062 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1063 | goto unlock; | |
1064 | ||
1065 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1066 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1067 | ||
1068 | ret = 0; | |
1069 | ||
1070 | unlock: | |
1071 | double_rq_unlock(src_rq, dst_rq); | |
1072 | ||
1073 | return ret; | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Cross migrate two tasks | |
1078 | */ | |
1079 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1080 | { | |
1081 | struct migration_swap_arg arg; | |
1082 | int ret = -EINVAL; | |
1083 | ||
1084 | get_online_cpus(); | |
1085 | ||
1086 | arg = (struct migration_swap_arg){ | |
1087 | .src_task = cur, | |
1088 | .src_cpu = task_cpu(cur), | |
1089 | .dst_task = p, | |
1090 | .dst_cpu = task_cpu(p), | |
1091 | }; | |
1092 | ||
1093 | if (arg.src_cpu == arg.dst_cpu) | |
1094 | goto out; | |
1095 | ||
1096 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) | |
1097 | goto out; | |
1098 | ||
1099 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1100 | goto out; | |
1101 | ||
1102 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1103 | goto out; | |
1104 | ||
1105 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); | |
1106 | ||
1107 | out: | |
1108 | put_online_cpus(); | |
1109 | return ret; | |
1110 | } | |
1111 | ||
969c7921 | 1112 | struct migration_arg { |
36c8b586 | 1113 | struct task_struct *task; |
1da177e4 | 1114 | int dest_cpu; |
70b97a7f | 1115 | }; |
1da177e4 | 1116 | |
969c7921 TH |
1117 | static int migration_cpu_stop(void *data); |
1118 | ||
1da177e4 LT |
1119 | /* |
1120 | * wait_task_inactive - wait for a thread to unschedule. | |
1121 | * | |
85ba2d86 RM |
1122 | * If @match_state is nonzero, it's the @p->state value just checked and |
1123 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1124 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1125 | * we return a positive number (its total switch count). If a second call | |
1126 | * a short while later returns the same number, the caller can be sure that | |
1127 | * @p has remained unscheduled the whole time. | |
1128 | * | |
1da177e4 LT |
1129 | * The caller must ensure that the task *will* unschedule sometime soon, |
1130 | * else this function might spin for a *long* time. This function can't | |
1131 | * be called with interrupts off, or it may introduce deadlock with | |
1132 | * smp_call_function() if an IPI is sent by the same process we are | |
1133 | * waiting to become inactive. | |
1134 | */ | |
85ba2d86 | 1135 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1136 | { |
1137 | unsigned long flags; | |
dd41f596 | 1138 | int running, on_rq; |
85ba2d86 | 1139 | unsigned long ncsw; |
70b97a7f | 1140 | struct rq *rq; |
1da177e4 | 1141 | |
3a5c359a AK |
1142 | for (;;) { |
1143 | /* | |
1144 | * We do the initial early heuristics without holding | |
1145 | * any task-queue locks at all. We'll only try to get | |
1146 | * the runqueue lock when things look like they will | |
1147 | * work out! | |
1148 | */ | |
1149 | rq = task_rq(p); | |
fa490cfd | 1150 | |
3a5c359a AK |
1151 | /* |
1152 | * If the task is actively running on another CPU | |
1153 | * still, just relax and busy-wait without holding | |
1154 | * any locks. | |
1155 | * | |
1156 | * NOTE! Since we don't hold any locks, it's not | |
1157 | * even sure that "rq" stays as the right runqueue! | |
1158 | * But we don't care, since "task_running()" will | |
1159 | * return false if the runqueue has changed and p | |
1160 | * is actually now running somewhere else! | |
1161 | */ | |
85ba2d86 RM |
1162 | while (task_running(rq, p)) { |
1163 | if (match_state && unlikely(p->state != match_state)) | |
1164 | return 0; | |
3a5c359a | 1165 | cpu_relax(); |
85ba2d86 | 1166 | } |
fa490cfd | 1167 | |
3a5c359a AK |
1168 | /* |
1169 | * Ok, time to look more closely! We need the rq | |
1170 | * lock now, to be *sure*. If we're wrong, we'll | |
1171 | * just go back and repeat. | |
1172 | */ | |
1173 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1174 | trace_sched_wait_task(p); |
3a5c359a | 1175 | running = task_running(rq, p); |
fd2f4419 | 1176 | on_rq = p->on_rq; |
85ba2d86 | 1177 | ncsw = 0; |
f31e11d8 | 1178 | if (!match_state || p->state == match_state) |
93dcf55f | 1179 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1180 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1181 | |
85ba2d86 RM |
1182 | /* |
1183 | * If it changed from the expected state, bail out now. | |
1184 | */ | |
1185 | if (unlikely(!ncsw)) | |
1186 | break; | |
1187 | ||
3a5c359a AK |
1188 | /* |
1189 | * Was it really running after all now that we | |
1190 | * checked with the proper locks actually held? | |
1191 | * | |
1192 | * Oops. Go back and try again.. | |
1193 | */ | |
1194 | if (unlikely(running)) { | |
1195 | cpu_relax(); | |
1196 | continue; | |
1197 | } | |
fa490cfd | 1198 | |
3a5c359a AK |
1199 | /* |
1200 | * It's not enough that it's not actively running, | |
1201 | * it must be off the runqueue _entirely_, and not | |
1202 | * preempted! | |
1203 | * | |
80dd99b3 | 1204 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1205 | * running right now), it's preempted, and we should |
1206 | * yield - it could be a while. | |
1207 | */ | |
1208 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1209 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1210 | ||
1211 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1212 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1213 | continue; |
1214 | } | |
fa490cfd | 1215 | |
3a5c359a AK |
1216 | /* |
1217 | * Ahh, all good. It wasn't running, and it wasn't | |
1218 | * runnable, which means that it will never become | |
1219 | * running in the future either. We're all done! | |
1220 | */ | |
1221 | break; | |
1222 | } | |
85ba2d86 RM |
1223 | |
1224 | return ncsw; | |
1da177e4 LT |
1225 | } |
1226 | ||
1227 | /*** | |
1228 | * kick_process - kick a running thread to enter/exit the kernel | |
1229 | * @p: the to-be-kicked thread | |
1230 | * | |
1231 | * Cause a process which is running on another CPU to enter | |
1232 | * kernel-mode, without any delay. (to get signals handled.) | |
1233 | * | |
25985edc | 1234 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1235 | * because all it wants to ensure is that the remote task enters |
1236 | * the kernel. If the IPI races and the task has been migrated | |
1237 | * to another CPU then no harm is done and the purpose has been | |
1238 | * achieved as well. | |
1239 | */ | |
36c8b586 | 1240 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1241 | { |
1242 | int cpu; | |
1243 | ||
1244 | preempt_disable(); | |
1245 | cpu = task_cpu(p); | |
1246 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1247 | smp_send_reschedule(cpu); | |
1248 | preempt_enable(); | |
1249 | } | |
b43e3521 | 1250 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1251 | #endif /* CONFIG_SMP */ |
1da177e4 | 1252 | |
970b13ba | 1253 | #ifdef CONFIG_SMP |
30da688e | 1254 | /* |
013fdb80 | 1255 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1256 | */ |
5da9a0fb PZ |
1257 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1258 | { | |
aa00d89c TC |
1259 | int nid = cpu_to_node(cpu); |
1260 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1261 | enum { cpuset, possible, fail } state = cpuset; |
1262 | int dest_cpu; | |
5da9a0fb | 1263 | |
aa00d89c TC |
1264 | /* |
1265 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1266 | * will return -1. There is no cpu on the node, and we should | |
1267 | * select the cpu on the other node. | |
1268 | */ | |
1269 | if (nid != -1) { | |
1270 | nodemask = cpumask_of_node(nid); | |
1271 | ||
1272 | /* Look for allowed, online CPU in same node. */ | |
1273 | for_each_cpu(dest_cpu, nodemask) { | |
1274 | if (!cpu_online(dest_cpu)) | |
1275 | continue; | |
1276 | if (!cpu_active(dest_cpu)) | |
1277 | continue; | |
1278 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1279 | return dest_cpu; | |
1280 | } | |
2baab4e9 | 1281 | } |
5da9a0fb | 1282 | |
2baab4e9 PZ |
1283 | for (;;) { |
1284 | /* Any allowed, online CPU? */ | |
e3831edd | 1285 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1286 | if (!cpu_online(dest_cpu)) |
1287 | continue; | |
1288 | if (!cpu_active(dest_cpu)) | |
1289 | continue; | |
1290 | goto out; | |
1291 | } | |
5da9a0fb | 1292 | |
2baab4e9 PZ |
1293 | switch (state) { |
1294 | case cpuset: | |
1295 | /* No more Mr. Nice Guy. */ | |
1296 | cpuset_cpus_allowed_fallback(p); | |
1297 | state = possible; | |
1298 | break; | |
1299 | ||
1300 | case possible: | |
1301 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1302 | state = fail; | |
1303 | break; | |
1304 | ||
1305 | case fail: | |
1306 | BUG(); | |
1307 | break; | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | out: | |
1312 | if (state != cpuset) { | |
1313 | /* | |
1314 | * Don't tell them about moving exiting tasks or | |
1315 | * kernel threads (both mm NULL), since they never | |
1316 | * leave kernel. | |
1317 | */ | |
1318 | if (p->mm && printk_ratelimit()) { | |
1319 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1320 | task_pid_nr(p), p->comm, cpu); | |
1321 | } | |
5da9a0fb PZ |
1322 | } |
1323 | ||
1324 | return dest_cpu; | |
1325 | } | |
1326 | ||
e2912009 | 1327 | /* |
013fdb80 | 1328 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1329 | */ |
970b13ba | 1330 | static inline |
ac66f547 | 1331 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1332 | { |
ac66f547 | 1333 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1334 | |
1335 | /* | |
1336 | * In order not to call set_task_cpu() on a blocking task we need | |
1337 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1338 | * cpu. | |
1339 | * | |
1340 | * Since this is common to all placement strategies, this lives here. | |
1341 | * | |
1342 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1343 | * not worry about this generic constraint ] | |
1344 | */ | |
fa17b507 | 1345 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1346 | !cpu_online(cpu))) |
5da9a0fb | 1347 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1348 | |
1349 | return cpu; | |
970b13ba | 1350 | } |
09a40af5 MG |
1351 | |
1352 | static void update_avg(u64 *avg, u64 sample) | |
1353 | { | |
1354 | s64 diff = sample - *avg; | |
1355 | *avg += diff >> 3; | |
1356 | } | |
970b13ba PZ |
1357 | #endif |
1358 | ||
d7c01d27 | 1359 | static void |
b84cb5df | 1360 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1361 | { |
d7c01d27 | 1362 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1363 | struct rq *rq = this_rq(); |
1364 | ||
d7c01d27 PZ |
1365 | #ifdef CONFIG_SMP |
1366 | int this_cpu = smp_processor_id(); | |
1367 | ||
1368 | if (cpu == this_cpu) { | |
1369 | schedstat_inc(rq, ttwu_local); | |
1370 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1371 | } else { | |
1372 | struct sched_domain *sd; | |
1373 | ||
1374 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1375 | rcu_read_lock(); |
d7c01d27 PZ |
1376 | for_each_domain(this_cpu, sd) { |
1377 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1378 | schedstat_inc(sd, ttwu_wake_remote); | |
1379 | break; | |
1380 | } | |
1381 | } | |
057f3fad | 1382 | rcu_read_unlock(); |
d7c01d27 | 1383 | } |
f339b9dc PZ |
1384 | |
1385 | if (wake_flags & WF_MIGRATED) | |
1386 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1387 | ||
d7c01d27 PZ |
1388 | #endif /* CONFIG_SMP */ |
1389 | ||
1390 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1391 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1392 | |
1393 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1394 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1395 | |
d7c01d27 PZ |
1396 | #endif /* CONFIG_SCHEDSTATS */ |
1397 | } | |
1398 | ||
1399 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1400 | { | |
9ed3811a | 1401 | activate_task(rq, p, en_flags); |
fd2f4419 | 1402 | p->on_rq = 1; |
c2f7115e PZ |
1403 | |
1404 | /* if a worker is waking up, notify workqueue */ | |
1405 | if (p->flags & PF_WQ_WORKER) | |
1406 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1407 | } |
1408 | ||
23f41eeb PZ |
1409 | /* |
1410 | * Mark the task runnable and perform wakeup-preemption. | |
1411 | */ | |
89363381 | 1412 | static void |
23f41eeb | 1413 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1414 | { |
9ed3811a | 1415 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1416 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1417 | |
1418 | p->state = TASK_RUNNING; | |
1419 | #ifdef CONFIG_SMP | |
1420 | if (p->sched_class->task_woken) | |
1421 | p->sched_class->task_woken(rq, p); | |
1422 | ||
e69c6341 | 1423 | if (rq->idle_stamp) { |
78becc27 | 1424 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1425 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1426 | |
abfafa54 JL |
1427 | update_avg(&rq->avg_idle, delta); |
1428 | ||
1429 | if (rq->avg_idle > max) | |
9ed3811a | 1430 | rq->avg_idle = max; |
abfafa54 | 1431 | |
9ed3811a TH |
1432 | rq->idle_stamp = 0; |
1433 | } | |
1434 | #endif | |
1435 | } | |
1436 | ||
c05fbafb PZ |
1437 | static void |
1438 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1439 | { | |
1440 | #ifdef CONFIG_SMP | |
1441 | if (p->sched_contributes_to_load) | |
1442 | rq->nr_uninterruptible--; | |
1443 | #endif | |
1444 | ||
1445 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1446 | ttwu_do_wakeup(rq, p, wake_flags); | |
1447 | } | |
1448 | ||
1449 | /* | |
1450 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1451 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1452 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1453 | * the task is still ->on_rq. | |
1454 | */ | |
1455 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1456 | { | |
1457 | struct rq *rq; | |
1458 | int ret = 0; | |
1459 | ||
1460 | rq = __task_rq_lock(p); | |
1461 | if (p->on_rq) { | |
1ad4ec0d FW |
1462 | /* check_preempt_curr() may use rq clock */ |
1463 | update_rq_clock(rq); | |
c05fbafb PZ |
1464 | ttwu_do_wakeup(rq, p, wake_flags); |
1465 | ret = 1; | |
1466 | } | |
1467 | __task_rq_unlock(rq); | |
1468 | ||
1469 | return ret; | |
1470 | } | |
1471 | ||
317f3941 | 1472 | #ifdef CONFIG_SMP |
fa14ff4a | 1473 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1474 | { |
1475 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1476 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1477 | struct task_struct *p; | |
317f3941 PZ |
1478 | |
1479 | raw_spin_lock(&rq->lock); | |
1480 | ||
fa14ff4a PZ |
1481 | while (llist) { |
1482 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1483 | llist = llist_next(llist); | |
317f3941 PZ |
1484 | ttwu_do_activate(rq, p, 0); |
1485 | } | |
1486 | ||
1487 | raw_spin_unlock(&rq->lock); | |
1488 | } | |
1489 | ||
1490 | void scheduler_ipi(void) | |
1491 | { | |
f27dde8d PZ |
1492 | /* |
1493 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1494 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1495 | * this IPI. | |
1496 | */ | |
1497 | if (tif_need_resched()) | |
1498 | set_preempt_need_resched(); | |
1499 | ||
873b4c65 VG |
1500 | if (llist_empty(&this_rq()->wake_list) |
1501 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1502 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1503 | return; |
1504 | ||
1505 | /* | |
1506 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1507 | * traditionally all their work was done from the interrupt return | |
1508 | * path. Now that we actually do some work, we need to make sure | |
1509 | * we do call them. | |
1510 | * | |
1511 | * Some archs already do call them, luckily irq_enter/exit nest | |
1512 | * properly. | |
1513 | * | |
1514 | * Arguably we should visit all archs and update all handlers, | |
1515 | * however a fair share of IPIs are still resched only so this would | |
1516 | * somewhat pessimize the simple resched case. | |
1517 | */ | |
1518 | irq_enter(); | |
ff442c51 | 1519 | tick_nohz_full_check(); |
fa14ff4a | 1520 | sched_ttwu_pending(); |
ca38062e SS |
1521 | |
1522 | /* | |
1523 | * Check if someone kicked us for doing the nohz idle load balance. | |
1524 | */ | |
873b4c65 | 1525 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1526 | this_rq()->idle_balance = 1; |
ca38062e | 1527 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1528 | } |
c5d753a5 | 1529 | irq_exit(); |
317f3941 PZ |
1530 | } |
1531 | ||
1532 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1533 | { | |
fa14ff4a | 1534 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1535 | smp_send_reschedule(cpu); |
1536 | } | |
d6aa8f85 | 1537 | |
39be3501 | 1538 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1539 | { |
1540 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1541 | } | |
d6aa8f85 | 1542 | #endif /* CONFIG_SMP */ |
317f3941 | 1543 | |
c05fbafb PZ |
1544 | static void ttwu_queue(struct task_struct *p, int cpu) |
1545 | { | |
1546 | struct rq *rq = cpu_rq(cpu); | |
1547 | ||
17d9f311 | 1548 | #if defined(CONFIG_SMP) |
39be3501 | 1549 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1550 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1551 | ttwu_queue_remote(p, cpu); |
1552 | return; | |
1553 | } | |
1554 | #endif | |
1555 | ||
c05fbafb PZ |
1556 | raw_spin_lock(&rq->lock); |
1557 | ttwu_do_activate(rq, p, 0); | |
1558 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1559 | } |
1560 | ||
1561 | /** | |
1da177e4 | 1562 | * try_to_wake_up - wake up a thread |
9ed3811a | 1563 | * @p: the thread to be awakened |
1da177e4 | 1564 | * @state: the mask of task states that can be woken |
9ed3811a | 1565 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1566 | * |
1567 | * Put it on the run-queue if it's not already there. The "current" | |
1568 | * thread is always on the run-queue (except when the actual | |
1569 | * re-schedule is in progress), and as such you're allowed to do | |
1570 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1571 | * runnable without the overhead of this. | |
1572 | * | |
e69f6186 | 1573 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1574 | * or @state didn't match @p's state. |
1da177e4 | 1575 | */ |
e4a52bcb PZ |
1576 | static int |
1577 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1578 | { |
1da177e4 | 1579 | unsigned long flags; |
c05fbafb | 1580 | int cpu, success = 0; |
2398f2c6 | 1581 | |
e0acd0a6 ON |
1582 | /* |
1583 | * If we are going to wake up a thread waiting for CONDITION we | |
1584 | * need to ensure that CONDITION=1 done by the caller can not be | |
1585 | * reordered with p->state check below. This pairs with mb() in | |
1586 | * set_current_state() the waiting thread does. | |
1587 | */ | |
1588 | smp_mb__before_spinlock(); | |
013fdb80 | 1589 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1590 | if (!(p->state & state)) |
1da177e4 LT |
1591 | goto out; |
1592 | ||
c05fbafb | 1593 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1594 | cpu = task_cpu(p); |
1da177e4 | 1595 | |
c05fbafb PZ |
1596 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1597 | goto stat; | |
1da177e4 | 1598 | |
1da177e4 | 1599 | #ifdef CONFIG_SMP |
e9c84311 | 1600 | /* |
c05fbafb PZ |
1601 | * If the owning (remote) cpu is still in the middle of schedule() with |
1602 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1603 | */ |
f3e94786 | 1604 | while (p->on_cpu) |
e4a52bcb | 1605 | cpu_relax(); |
0970d299 | 1606 | /* |
e4a52bcb | 1607 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1608 | */ |
e4a52bcb | 1609 | smp_rmb(); |
1da177e4 | 1610 | |
a8e4f2ea | 1611 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1612 | p->state = TASK_WAKING; |
e7693a36 | 1613 | |
e4a52bcb | 1614 | if (p->sched_class->task_waking) |
74f8e4b2 | 1615 | p->sched_class->task_waking(p); |
efbbd05a | 1616 | |
ac66f547 | 1617 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1618 | if (task_cpu(p) != cpu) { |
1619 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1620 | set_task_cpu(p, cpu); |
f339b9dc | 1621 | } |
1da177e4 | 1622 | #endif /* CONFIG_SMP */ |
1da177e4 | 1623 | |
c05fbafb PZ |
1624 | ttwu_queue(p, cpu); |
1625 | stat: | |
b84cb5df | 1626 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1627 | out: |
013fdb80 | 1628 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1629 | |
1630 | return success; | |
1631 | } | |
1632 | ||
21aa9af0 TH |
1633 | /** |
1634 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1635 | * @p: the thread to be awakened | |
1636 | * | |
2acca55e | 1637 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1638 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1639 | * the current task. |
21aa9af0 TH |
1640 | */ |
1641 | static void try_to_wake_up_local(struct task_struct *p) | |
1642 | { | |
1643 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1644 | |
383efcd0 TH |
1645 | if (WARN_ON_ONCE(rq != this_rq()) || |
1646 | WARN_ON_ONCE(p == current)) | |
1647 | return; | |
1648 | ||
21aa9af0 TH |
1649 | lockdep_assert_held(&rq->lock); |
1650 | ||
2acca55e PZ |
1651 | if (!raw_spin_trylock(&p->pi_lock)) { |
1652 | raw_spin_unlock(&rq->lock); | |
1653 | raw_spin_lock(&p->pi_lock); | |
1654 | raw_spin_lock(&rq->lock); | |
1655 | } | |
1656 | ||
21aa9af0 | 1657 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1658 | goto out; |
21aa9af0 | 1659 | |
fd2f4419 | 1660 | if (!p->on_rq) |
d7c01d27 PZ |
1661 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1662 | ||
23f41eeb | 1663 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1664 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1665 | out: |
1666 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1667 | } |
1668 | ||
50fa610a DH |
1669 | /** |
1670 | * wake_up_process - Wake up a specific process | |
1671 | * @p: The process to be woken up. | |
1672 | * | |
1673 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1674 | * processes. |
1675 | * | |
1676 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1677 | * |
1678 | * It may be assumed that this function implies a write memory barrier before | |
1679 | * changing the task state if and only if any tasks are woken up. | |
1680 | */ | |
7ad5b3a5 | 1681 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1682 | { |
9067ac85 ON |
1683 | WARN_ON(task_is_stopped_or_traced(p)); |
1684 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1685 | } |
1da177e4 LT |
1686 | EXPORT_SYMBOL(wake_up_process); |
1687 | ||
7ad5b3a5 | 1688 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1689 | { |
1690 | return try_to_wake_up(p, state, 0); | |
1691 | } | |
1692 | ||
1da177e4 LT |
1693 | /* |
1694 | * Perform scheduler related setup for a newly forked process p. | |
1695 | * p is forked by current. | |
dd41f596 IM |
1696 | * |
1697 | * __sched_fork() is basic setup used by init_idle() too: | |
1698 | */ | |
1699 | static void __sched_fork(struct task_struct *p) | |
1700 | { | |
fd2f4419 PZ |
1701 | p->on_rq = 0; |
1702 | ||
1703 | p->se.on_rq = 0; | |
dd41f596 IM |
1704 | p->se.exec_start = 0; |
1705 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1706 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1707 | p->se.nr_migrations = 0; |
da7a735e | 1708 | p->se.vruntime = 0; |
fd2f4419 | 1709 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1710 | |
1711 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1712 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1713 | #endif |
476d139c | 1714 | |
fa717060 | 1715 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1716 | |
e107be36 AK |
1717 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1718 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1719 | #endif | |
cbee9f88 PZ |
1720 | |
1721 | #ifdef CONFIG_NUMA_BALANCING | |
1722 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 MG |
1723 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
1724 | p->mm->numa_next_reset = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_period_reset); | |
cbee9f88 PZ |
1725 | p->mm->numa_scan_seq = 0; |
1726 | } | |
1727 | ||
1728 | p->node_stamp = 0ULL; | |
1729 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
6fe6b2d6 | 1730 | p->numa_migrate_seq = 1; |
4b96a29b | 1731 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
688b7585 | 1732 | p->numa_preferred_nid = -1; |
cbee9f88 | 1733 | p->numa_work.next = &p->numa_work; |
f809ca9a | 1734 | p->numa_faults = NULL; |
745d6147 | 1735 | p->numa_faults_buffer = NULL; |
cbee9f88 | 1736 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1737 | } |
1738 | ||
1a687c2e | 1739 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1740 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1741 | void set_numabalancing_state(bool enabled) |
1742 | { | |
1743 | if (enabled) | |
1744 | sched_feat_set("NUMA"); | |
1745 | else | |
1746 | sched_feat_set("NO_NUMA"); | |
1747 | } | |
3105b86a MG |
1748 | #else |
1749 | __read_mostly bool numabalancing_enabled; | |
1750 | ||
1751 | void set_numabalancing_state(bool enabled) | |
1752 | { | |
1753 | numabalancing_enabled = enabled; | |
dd41f596 | 1754 | } |
3105b86a | 1755 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1756 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1757 | |
1758 | /* | |
1759 | * fork()/clone()-time setup: | |
1760 | */ | |
3e51e3ed | 1761 | void sched_fork(struct task_struct *p) |
dd41f596 | 1762 | { |
0122ec5b | 1763 | unsigned long flags; |
dd41f596 IM |
1764 | int cpu = get_cpu(); |
1765 | ||
1766 | __sched_fork(p); | |
06b83b5f | 1767 | /* |
0017d735 | 1768 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1769 | * nobody will actually run it, and a signal or other external |
1770 | * event cannot wake it up and insert it on the runqueue either. | |
1771 | */ | |
0017d735 | 1772 | p->state = TASK_RUNNING; |
dd41f596 | 1773 | |
c350a04e MG |
1774 | /* |
1775 | * Make sure we do not leak PI boosting priority to the child. | |
1776 | */ | |
1777 | p->prio = current->normal_prio; | |
1778 | ||
b9dc29e7 MG |
1779 | /* |
1780 | * Revert to default priority/policy on fork if requested. | |
1781 | */ | |
1782 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1783 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1784 | p->policy = SCHED_NORMAL; |
6c697bdf | 1785 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1786 | p->rt_priority = 0; |
1787 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1788 | p->static_prio = NICE_TO_PRIO(0); | |
1789 | ||
1790 | p->prio = p->normal_prio = __normal_prio(p); | |
1791 | set_load_weight(p); | |
6c697bdf | 1792 | |
b9dc29e7 MG |
1793 | /* |
1794 | * We don't need the reset flag anymore after the fork. It has | |
1795 | * fulfilled its duty: | |
1796 | */ | |
1797 | p->sched_reset_on_fork = 0; | |
1798 | } | |
ca94c442 | 1799 | |
2ddbf952 HS |
1800 | if (!rt_prio(p->prio)) |
1801 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1802 | |
cd29fe6f PZ |
1803 | if (p->sched_class->task_fork) |
1804 | p->sched_class->task_fork(p); | |
1805 | ||
86951599 PZ |
1806 | /* |
1807 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1808 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1809 | * is ran before sched_fork(). | |
1810 | * | |
1811 | * Silence PROVE_RCU. | |
1812 | */ | |
0122ec5b | 1813 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1814 | set_task_cpu(p, cpu); |
0122ec5b | 1815 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1816 | |
52f17b6c | 1817 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1818 | if (likely(sched_info_on())) |
52f17b6c | 1819 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1820 | #endif |
3ca7a440 PZ |
1821 | #if defined(CONFIG_SMP) |
1822 | p->on_cpu = 0; | |
4866cde0 | 1823 | #endif |
01028747 | 1824 | init_task_preempt_count(p); |
806c09a7 | 1825 | #ifdef CONFIG_SMP |
917b627d | 1826 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1827 | #endif |
917b627d | 1828 | |
476d139c | 1829 | put_cpu(); |
1da177e4 LT |
1830 | } |
1831 | ||
1832 | /* | |
1833 | * wake_up_new_task - wake up a newly created task for the first time. | |
1834 | * | |
1835 | * This function will do some initial scheduler statistics housekeeping | |
1836 | * that must be done for every newly created context, then puts the task | |
1837 | * on the runqueue and wakes it. | |
1838 | */ | |
3e51e3ed | 1839 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1840 | { |
1841 | unsigned long flags; | |
dd41f596 | 1842 | struct rq *rq; |
fabf318e | 1843 | |
ab2515c4 | 1844 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1845 | #ifdef CONFIG_SMP |
1846 | /* | |
1847 | * Fork balancing, do it here and not earlier because: | |
1848 | * - cpus_allowed can change in the fork path | |
1849 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1850 | */ |
ac66f547 | 1851 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1852 | #endif |
1853 | ||
a75cdaa9 AS |
1854 | /* Initialize new task's runnable average */ |
1855 | init_task_runnable_average(p); | |
ab2515c4 | 1856 | rq = __task_rq_lock(p); |
cd29fe6f | 1857 | activate_task(rq, p, 0); |
fd2f4419 | 1858 | p->on_rq = 1; |
89363381 | 1859 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1860 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1861 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1862 | if (p->sched_class->task_woken) |
1863 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1864 | #endif |
0122ec5b | 1865 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1866 | } |
1867 | ||
e107be36 AK |
1868 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1869 | ||
1870 | /** | |
80dd99b3 | 1871 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1872 | * @notifier: notifier struct to register |
e107be36 AK |
1873 | */ |
1874 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1875 | { | |
1876 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1877 | } | |
1878 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1879 | ||
1880 | /** | |
1881 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1882 | * @notifier: notifier struct to unregister |
e107be36 AK |
1883 | * |
1884 | * This is safe to call from within a preemption notifier. | |
1885 | */ | |
1886 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1887 | { | |
1888 | hlist_del(¬ifier->link); | |
1889 | } | |
1890 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1891 | ||
1892 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1893 | { | |
1894 | struct preempt_notifier *notifier; | |
e107be36 | 1895 | |
b67bfe0d | 1896 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1897 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
1898 | } | |
1899 | ||
1900 | static void | |
1901 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1902 | struct task_struct *next) | |
1903 | { | |
1904 | struct preempt_notifier *notifier; | |
e107be36 | 1905 | |
b67bfe0d | 1906 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1907 | notifier->ops->sched_out(notifier, next); |
1908 | } | |
1909 | ||
6d6bc0ad | 1910 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1911 | |
1912 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1913 | { | |
1914 | } | |
1915 | ||
1916 | static void | |
1917 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1918 | struct task_struct *next) | |
1919 | { | |
1920 | } | |
1921 | ||
6d6bc0ad | 1922 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1923 | |
4866cde0 NP |
1924 | /** |
1925 | * prepare_task_switch - prepare to switch tasks | |
1926 | * @rq: the runqueue preparing to switch | |
421cee29 | 1927 | * @prev: the current task that is being switched out |
4866cde0 NP |
1928 | * @next: the task we are going to switch to. |
1929 | * | |
1930 | * This is called with the rq lock held and interrupts off. It must | |
1931 | * be paired with a subsequent finish_task_switch after the context | |
1932 | * switch. | |
1933 | * | |
1934 | * prepare_task_switch sets up locking and calls architecture specific | |
1935 | * hooks. | |
1936 | */ | |
e107be36 AK |
1937 | static inline void |
1938 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1939 | struct task_struct *next) | |
4866cde0 | 1940 | { |
895dd92c | 1941 | trace_sched_switch(prev, next); |
43148951 | 1942 | sched_info_switch(rq, prev, next); |
fe4b04fa | 1943 | perf_event_task_sched_out(prev, next); |
e107be36 | 1944 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1945 | prepare_lock_switch(rq, next); |
1946 | prepare_arch_switch(next); | |
1947 | } | |
1948 | ||
1da177e4 LT |
1949 | /** |
1950 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1951 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1952 | * @prev: the thread we just switched away from. |
1953 | * | |
4866cde0 NP |
1954 | * finish_task_switch must be called after the context switch, paired |
1955 | * with a prepare_task_switch call before the context switch. | |
1956 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1957 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1958 | * |
1959 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1960 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1961 | * with the lock held can cause deadlocks; see schedule() for |
1962 | * details.) | |
1963 | */ | |
a9957449 | 1964 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1965 | __releases(rq->lock) |
1966 | { | |
1da177e4 | 1967 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1968 | long prev_state; |
1da177e4 LT |
1969 | |
1970 | rq->prev_mm = NULL; | |
1971 | ||
1972 | /* | |
1973 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1974 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1975 | * schedule one last time. The schedule call will never return, and |
1976 | * the scheduled task must drop that reference. | |
c394cc9f | 1977 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1978 | * still held, otherwise prev could be scheduled on another cpu, die |
1979 | * there before we look at prev->state, and then the reference would | |
1980 | * be dropped twice. | |
1981 | * Manfred Spraul <manfred@colorfullife.com> | |
1982 | */ | |
55a101f8 | 1983 | prev_state = prev->state; |
bf9fae9f | 1984 | vtime_task_switch(prev); |
4866cde0 | 1985 | finish_arch_switch(prev); |
a8d757ef | 1986 | perf_event_task_sched_in(prev, current); |
4866cde0 | 1987 | finish_lock_switch(rq, prev); |
01f23e16 | 1988 | finish_arch_post_lock_switch(); |
e8fa1362 | 1989 | |
e107be36 | 1990 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1991 | if (mm) |
1992 | mmdrop(mm); | |
c394cc9f | 1993 | if (unlikely(prev_state == TASK_DEAD)) { |
f809ca9a MG |
1994 | task_numa_free(prev); |
1995 | ||
c6fd91f0 | 1996 | /* |
1997 | * Remove function-return probe instances associated with this | |
1998 | * task and put them back on the free list. | |
9761eea8 | 1999 | */ |
c6fd91f0 | 2000 | kprobe_flush_task(prev); |
1da177e4 | 2001 | put_task_struct(prev); |
c6fd91f0 | 2002 | } |
99e5ada9 FW |
2003 | |
2004 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2005 | } |
2006 | ||
3f029d3c GH |
2007 | #ifdef CONFIG_SMP |
2008 | ||
2009 | /* assumes rq->lock is held */ | |
2010 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2011 | { | |
2012 | if (prev->sched_class->pre_schedule) | |
2013 | prev->sched_class->pre_schedule(rq, prev); | |
2014 | } | |
2015 | ||
2016 | /* rq->lock is NOT held, but preemption is disabled */ | |
2017 | static inline void post_schedule(struct rq *rq) | |
2018 | { | |
2019 | if (rq->post_schedule) { | |
2020 | unsigned long flags; | |
2021 | ||
05fa785c | 2022 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2023 | if (rq->curr->sched_class->post_schedule) |
2024 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2025 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2026 | |
2027 | rq->post_schedule = 0; | |
2028 | } | |
2029 | } | |
2030 | ||
2031 | #else | |
da19ab51 | 2032 | |
3f029d3c GH |
2033 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2034 | { | |
2035 | } | |
2036 | ||
2037 | static inline void post_schedule(struct rq *rq) | |
2038 | { | |
1da177e4 LT |
2039 | } |
2040 | ||
3f029d3c GH |
2041 | #endif |
2042 | ||
1da177e4 LT |
2043 | /** |
2044 | * schedule_tail - first thing a freshly forked thread must call. | |
2045 | * @prev: the thread we just switched away from. | |
2046 | */ | |
36c8b586 | 2047 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2048 | __releases(rq->lock) |
2049 | { | |
70b97a7f IM |
2050 | struct rq *rq = this_rq(); |
2051 | ||
4866cde0 | 2052 | finish_task_switch(rq, prev); |
da19ab51 | 2053 | |
3f029d3c GH |
2054 | /* |
2055 | * FIXME: do we need to worry about rq being invalidated by the | |
2056 | * task_switch? | |
2057 | */ | |
2058 | post_schedule(rq); | |
70b97a7f | 2059 | |
4866cde0 NP |
2060 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2061 | /* In this case, finish_task_switch does not reenable preemption */ | |
2062 | preempt_enable(); | |
2063 | #endif | |
1da177e4 | 2064 | if (current->set_child_tid) |
b488893a | 2065 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2066 | } |
2067 | ||
2068 | /* | |
2069 | * context_switch - switch to the new MM and the new | |
2070 | * thread's register state. | |
2071 | */ | |
dd41f596 | 2072 | static inline void |
70b97a7f | 2073 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2074 | struct task_struct *next) |
1da177e4 | 2075 | { |
dd41f596 | 2076 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2077 | |
e107be36 | 2078 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2079 | |
dd41f596 IM |
2080 | mm = next->mm; |
2081 | oldmm = prev->active_mm; | |
9226d125 ZA |
2082 | /* |
2083 | * For paravirt, this is coupled with an exit in switch_to to | |
2084 | * combine the page table reload and the switch backend into | |
2085 | * one hypercall. | |
2086 | */ | |
224101ed | 2087 | arch_start_context_switch(prev); |
9226d125 | 2088 | |
31915ab4 | 2089 | if (!mm) { |
1da177e4 LT |
2090 | next->active_mm = oldmm; |
2091 | atomic_inc(&oldmm->mm_count); | |
2092 | enter_lazy_tlb(oldmm, next); | |
2093 | } else | |
2094 | switch_mm(oldmm, mm, next); | |
2095 | ||
31915ab4 | 2096 | if (!prev->mm) { |
1da177e4 | 2097 | prev->active_mm = NULL; |
1da177e4 LT |
2098 | rq->prev_mm = oldmm; |
2099 | } | |
3a5f5e48 IM |
2100 | /* |
2101 | * Since the runqueue lock will be released by the next | |
2102 | * task (which is an invalid locking op but in the case | |
2103 | * of the scheduler it's an obvious special-case), so we | |
2104 | * do an early lockdep release here: | |
2105 | */ | |
2106 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2107 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2108 | #endif |
1da177e4 | 2109 | |
91d1aa43 | 2110 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2111 | /* Here we just switch the register state and the stack. */ |
2112 | switch_to(prev, next, prev); | |
2113 | ||
dd41f596 IM |
2114 | barrier(); |
2115 | /* | |
2116 | * this_rq must be evaluated again because prev may have moved | |
2117 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2118 | * frame will be invalid. | |
2119 | */ | |
2120 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2121 | } |
2122 | ||
2123 | /* | |
1c3e8264 | 2124 | * nr_running and nr_context_switches: |
1da177e4 LT |
2125 | * |
2126 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2127 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2128 | */ |
2129 | unsigned long nr_running(void) | |
2130 | { | |
2131 | unsigned long i, sum = 0; | |
2132 | ||
2133 | for_each_online_cpu(i) | |
2134 | sum += cpu_rq(i)->nr_running; | |
2135 | ||
2136 | return sum; | |
f711f609 | 2137 | } |
1da177e4 | 2138 | |
1da177e4 | 2139 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2140 | { |
cc94abfc SR |
2141 | int i; |
2142 | unsigned long long sum = 0; | |
46cb4b7c | 2143 | |
0a945022 | 2144 | for_each_possible_cpu(i) |
1da177e4 | 2145 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2146 | |
1da177e4 LT |
2147 | return sum; |
2148 | } | |
483b4ee6 | 2149 | |
1da177e4 LT |
2150 | unsigned long nr_iowait(void) |
2151 | { | |
2152 | unsigned long i, sum = 0; | |
483b4ee6 | 2153 | |
0a945022 | 2154 | for_each_possible_cpu(i) |
1da177e4 | 2155 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2156 | |
1da177e4 LT |
2157 | return sum; |
2158 | } | |
483b4ee6 | 2159 | |
8c215bd3 | 2160 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2161 | { |
8c215bd3 | 2162 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2163 | return atomic_read(&this->nr_iowait); |
2164 | } | |
46cb4b7c | 2165 | |
dd41f596 | 2166 | #ifdef CONFIG_SMP |
8a0be9ef | 2167 | |
46cb4b7c | 2168 | /* |
38022906 PZ |
2169 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2170 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2171 | */ |
38022906 | 2172 | void sched_exec(void) |
46cb4b7c | 2173 | { |
38022906 | 2174 | struct task_struct *p = current; |
1da177e4 | 2175 | unsigned long flags; |
0017d735 | 2176 | int dest_cpu; |
46cb4b7c | 2177 | |
8f42ced9 | 2178 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2179 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2180 | if (dest_cpu == smp_processor_id()) |
2181 | goto unlock; | |
38022906 | 2182 | |
8f42ced9 | 2183 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2184 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2185 | |
8f42ced9 PZ |
2186 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2187 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2188 | return; |
2189 | } | |
0017d735 | 2190 | unlock: |
8f42ced9 | 2191 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2192 | } |
dd41f596 | 2193 | |
1da177e4 LT |
2194 | #endif |
2195 | ||
1da177e4 | 2196 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2197 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2198 | |
2199 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2200 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2201 | |
2202 | /* | |
c5f8d995 | 2203 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2204 | * @p in case that task is currently running. |
c5f8d995 HS |
2205 | * |
2206 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2207 | */ |
c5f8d995 HS |
2208 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2209 | { | |
2210 | u64 ns = 0; | |
2211 | ||
2212 | if (task_current(rq, p)) { | |
2213 | update_rq_clock(rq); | |
78becc27 | 2214 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2215 | if ((s64)ns < 0) |
2216 | ns = 0; | |
2217 | } | |
2218 | ||
2219 | return ns; | |
2220 | } | |
2221 | ||
bb34d92f | 2222 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2223 | { |
1da177e4 | 2224 | unsigned long flags; |
41b86e9c | 2225 | struct rq *rq; |
bb34d92f | 2226 | u64 ns = 0; |
48f24c4d | 2227 | |
41b86e9c | 2228 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2229 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2230 | task_rq_unlock(rq, p, &flags); |
1508487e | 2231 | |
c5f8d995 HS |
2232 | return ns; |
2233 | } | |
f06febc9 | 2234 | |
c5f8d995 HS |
2235 | /* |
2236 | * Return accounted runtime for the task. | |
2237 | * In case the task is currently running, return the runtime plus current's | |
2238 | * pending runtime that have not been accounted yet. | |
2239 | */ | |
2240 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2241 | { | |
2242 | unsigned long flags; | |
2243 | struct rq *rq; | |
2244 | u64 ns = 0; | |
2245 | ||
2246 | rq = task_rq_lock(p, &flags); | |
2247 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2248 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2249 | |
2250 | return ns; | |
2251 | } | |
48f24c4d | 2252 | |
7835b98b CL |
2253 | /* |
2254 | * This function gets called by the timer code, with HZ frequency. | |
2255 | * We call it with interrupts disabled. | |
7835b98b CL |
2256 | */ |
2257 | void scheduler_tick(void) | |
2258 | { | |
7835b98b CL |
2259 | int cpu = smp_processor_id(); |
2260 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2261 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2262 | |
2263 | sched_clock_tick(); | |
dd41f596 | 2264 | |
05fa785c | 2265 | raw_spin_lock(&rq->lock); |
3e51f33f | 2266 | update_rq_clock(rq); |
fa85ae24 | 2267 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2268 | update_cpu_load_active(rq); |
05fa785c | 2269 | raw_spin_unlock(&rq->lock); |
7835b98b | 2270 | |
e9d2b064 | 2271 | perf_event_task_tick(); |
e220d2dc | 2272 | |
e418e1c2 | 2273 | #ifdef CONFIG_SMP |
6eb57e0d | 2274 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2275 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2276 | #endif |
265f22a9 | 2277 | rq_last_tick_reset(rq); |
1da177e4 LT |
2278 | } |
2279 | ||
265f22a9 FW |
2280 | #ifdef CONFIG_NO_HZ_FULL |
2281 | /** | |
2282 | * scheduler_tick_max_deferment | |
2283 | * | |
2284 | * Keep at least one tick per second when a single | |
2285 | * active task is running because the scheduler doesn't | |
2286 | * yet completely support full dynticks environment. | |
2287 | * | |
2288 | * This makes sure that uptime, CFS vruntime, load | |
2289 | * balancing, etc... continue to move forward, even | |
2290 | * with a very low granularity. | |
e69f6186 YB |
2291 | * |
2292 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2293 | */ |
2294 | u64 scheduler_tick_max_deferment(void) | |
2295 | { | |
2296 | struct rq *rq = this_rq(); | |
2297 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2298 | ||
2299 | next = rq->last_sched_tick + HZ; | |
2300 | ||
2301 | if (time_before_eq(next, now)) | |
2302 | return 0; | |
2303 | ||
2304 | return jiffies_to_usecs(next - now) * NSEC_PER_USEC; | |
1da177e4 | 2305 | } |
265f22a9 | 2306 | #endif |
1da177e4 | 2307 | |
132380a0 | 2308 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2309 | { |
2310 | if (in_lock_functions(addr)) { | |
2311 | addr = CALLER_ADDR2; | |
2312 | if (in_lock_functions(addr)) | |
2313 | addr = CALLER_ADDR3; | |
2314 | } | |
2315 | return addr; | |
2316 | } | |
1da177e4 | 2317 | |
7e49fcce SR |
2318 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2319 | defined(CONFIG_PREEMPT_TRACER)) | |
2320 | ||
bdb43806 | 2321 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2322 | { |
6cd8a4bb | 2323 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2324 | /* |
2325 | * Underflow? | |
2326 | */ | |
9a11b49a IM |
2327 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2328 | return; | |
6cd8a4bb | 2329 | #endif |
bdb43806 | 2330 | __preempt_count_add(val); |
6cd8a4bb | 2331 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2332 | /* |
2333 | * Spinlock count overflowing soon? | |
2334 | */ | |
33859f7f MOS |
2335 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2336 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2337 | #endif |
2338 | if (preempt_count() == val) | |
2339 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 | 2340 | } |
bdb43806 | 2341 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2342 | |
bdb43806 | 2343 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2344 | { |
6cd8a4bb | 2345 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2346 | /* |
2347 | * Underflow? | |
2348 | */ | |
01e3eb82 | 2349 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2350 | return; |
1da177e4 LT |
2351 | /* |
2352 | * Is the spinlock portion underflowing? | |
2353 | */ | |
9a11b49a IM |
2354 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2355 | !(preempt_count() & PREEMPT_MASK))) | |
2356 | return; | |
6cd8a4bb | 2357 | #endif |
9a11b49a | 2358 | |
6cd8a4bb SR |
2359 | if (preempt_count() == val) |
2360 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2361 | __preempt_count_sub(val); |
1da177e4 | 2362 | } |
bdb43806 | 2363 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2364 | |
2365 | #endif | |
2366 | ||
2367 | /* | |
dd41f596 | 2368 | * Print scheduling while atomic bug: |
1da177e4 | 2369 | */ |
dd41f596 | 2370 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2371 | { |
664dfa65 DJ |
2372 | if (oops_in_progress) |
2373 | return; | |
2374 | ||
3df0fc5b PZ |
2375 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2376 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2377 | |
dd41f596 | 2378 | debug_show_held_locks(prev); |
e21f5b15 | 2379 | print_modules(); |
dd41f596 IM |
2380 | if (irqs_disabled()) |
2381 | print_irqtrace_events(prev); | |
6135fc1e | 2382 | dump_stack(); |
373d4d09 | 2383 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2384 | } |
1da177e4 | 2385 | |
dd41f596 IM |
2386 | /* |
2387 | * Various schedule()-time debugging checks and statistics: | |
2388 | */ | |
2389 | static inline void schedule_debug(struct task_struct *prev) | |
2390 | { | |
1da177e4 | 2391 | /* |
41a2d6cf | 2392 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
2393 | * schedule() atomically, we ignore that path for now. |
2394 | * Otherwise, whine if we are scheduling when we should not be. | |
2395 | */ | |
3f33a7ce | 2396 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 2397 | __schedule_bug(prev); |
b3fbab05 | 2398 | rcu_sleep_check(); |
dd41f596 | 2399 | |
1da177e4 LT |
2400 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2401 | ||
2d72376b | 2402 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2403 | } |
2404 | ||
6cecd084 | 2405 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2406 | { |
61eadef6 | 2407 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2408 | update_rq_clock(rq); |
6cecd084 | 2409 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2410 | } |
2411 | ||
dd41f596 IM |
2412 | /* |
2413 | * Pick up the highest-prio task: | |
2414 | */ | |
2415 | static inline struct task_struct * | |
b67802ea | 2416 | pick_next_task(struct rq *rq) |
dd41f596 | 2417 | { |
5522d5d5 | 2418 | const struct sched_class *class; |
dd41f596 | 2419 | struct task_struct *p; |
1da177e4 LT |
2420 | |
2421 | /* | |
dd41f596 IM |
2422 | * Optimization: we know that if all tasks are in |
2423 | * the fair class we can call that function directly: | |
1da177e4 | 2424 | */ |
953bfcd1 | 2425 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2426 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2427 | if (likely(p)) |
2428 | return p; | |
1da177e4 LT |
2429 | } |
2430 | ||
34f971f6 | 2431 | for_each_class(class) { |
fb8d4724 | 2432 | p = class->pick_next_task(rq); |
dd41f596 IM |
2433 | if (p) |
2434 | return p; | |
dd41f596 | 2435 | } |
34f971f6 PZ |
2436 | |
2437 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2438 | } |
1da177e4 | 2439 | |
dd41f596 | 2440 | /* |
c259e01a | 2441 | * __schedule() is the main scheduler function. |
edde96ea PE |
2442 | * |
2443 | * The main means of driving the scheduler and thus entering this function are: | |
2444 | * | |
2445 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2446 | * | |
2447 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2448 | * paths. For example, see arch/x86/entry_64.S. | |
2449 | * | |
2450 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2451 | * interrupt handler scheduler_tick(). | |
2452 | * | |
2453 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2454 | * task to the run-queue and that's it. | |
2455 | * | |
2456 | * Now, if the new task added to the run-queue preempts the current | |
2457 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2458 | * called on the nearest possible occasion: | |
2459 | * | |
2460 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2461 | * | |
2462 | * - in syscall or exception context, at the next outmost | |
2463 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2464 | * spin_unlock()!) | |
2465 | * | |
2466 | * - in IRQ context, return from interrupt-handler to | |
2467 | * preemptible context | |
2468 | * | |
2469 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2470 | * then at the next: | |
2471 | * | |
2472 | * - cond_resched() call | |
2473 | * - explicit schedule() call | |
2474 | * - return from syscall or exception to user-space | |
2475 | * - return from interrupt-handler to user-space | |
dd41f596 | 2476 | */ |
c259e01a | 2477 | static void __sched __schedule(void) |
dd41f596 IM |
2478 | { |
2479 | struct task_struct *prev, *next; | |
67ca7bde | 2480 | unsigned long *switch_count; |
dd41f596 | 2481 | struct rq *rq; |
31656519 | 2482 | int cpu; |
dd41f596 | 2483 | |
ff743345 PZ |
2484 | need_resched: |
2485 | preempt_disable(); | |
dd41f596 IM |
2486 | cpu = smp_processor_id(); |
2487 | rq = cpu_rq(cpu); | |
25502a6c | 2488 | rcu_note_context_switch(cpu); |
dd41f596 | 2489 | prev = rq->curr; |
dd41f596 | 2490 | |
dd41f596 | 2491 | schedule_debug(prev); |
1da177e4 | 2492 | |
31656519 | 2493 | if (sched_feat(HRTICK)) |
f333fdc9 | 2494 | hrtick_clear(rq); |
8f4d37ec | 2495 | |
e0acd0a6 ON |
2496 | /* |
2497 | * Make sure that signal_pending_state()->signal_pending() below | |
2498 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2499 | * done by the caller to avoid the race with signal_wake_up(). | |
2500 | */ | |
2501 | smp_mb__before_spinlock(); | |
05fa785c | 2502 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2503 | |
246d86b5 | 2504 | switch_count = &prev->nivcsw; |
1da177e4 | 2505 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2506 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2507 | prev->state = TASK_RUNNING; |
21aa9af0 | 2508 | } else { |
2acca55e PZ |
2509 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2510 | prev->on_rq = 0; | |
2511 | ||
21aa9af0 | 2512 | /* |
2acca55e PZ |
2513 | * If a worker went to sleep, notify and ask workqueue |
2514 | * whether it wants to wake up a task to maintain | |
2515 | * concurrency. | |
21aa9af0 TH |
2516 | */ |
2517 | if (prev->flags & PF_WQ_WORKER) { | |
2518 | struct task_struct *to_wakeup; | |
2519 | ||
2520 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2521 | if (to_wakeup) | |
2522 | try_to_wake_up_local(to_wakeup); | |
2523 | } | |
21aa9af0 | 2524 | } |
dd41f596 | 2525 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2526 | } |
2527 | ||
3f029d3c | 2528 | pre_schedule(rq, prev); |
f65eda4f | 2529 | |
dd41f596 | 2530 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2531 | idle_balance(cpu, rq); |
1da177e4 | 2532 | |
df1c99d4 | 2533 | put_prev_task(rq, prev); |
b67802ea | 2534 | next = pick_next_task(rq); |
f26f9aff | 2535 | clear_tsk_need_resched(prev); |
f27dde8d | 2536 | clear_preempt_need_resched(); |
f26f9aff | 2537 | rq->skip_clock_update = 0; |
1da177e4 | 2538 | |
1da177e4 | 2539 | if (likely(prev != next)) { |
1da177e4 LT |
2540 | rq->nr_switches++; |
2541 | rq->curr = next; | |
2542 | ++*switch_count; | |
2543 | ||
dd41f596 | 2544 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2545 | /* |
246d86b5 ON |
2546 | * The context switch have flipped the stack from under us |
2547 | * and restored the local variables which were saved when | |
2548 | * this task called schedule() in the past. prev == current | |
2549 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2550 | */ |
2551 | cpu = smp_processor_id(); | |
2552 | rq = cpu_rq(cpu); | |
1da177e4 | 2553 | } else |
05fa785c | 2554 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2555 | |
3f029d3c | 2556 | post_schedule(rq); |
1da177e4 | 2557 | |
ba74c144 | 2558 | sched_preempt_enable_no_resched(); |
ff743345 | 2559 | if (need_resched()) |
1da177e4 LT |
2560 | goto need_resched; |
2561 | } | |
c259e01a | 2562 | |
9c40cef2 TG |
2563 | static inline void sched_submit_work(struct task_struct *tsk) |
2564 | { | |
3c7d5184 | 2565 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2566 | return; |
2567 | /* | |
2568 | * If we are going to sleep and we have plugged IO queued, | |
2569 | * make sure to submit it to avoid deadlocks. | |
2570 | */ | |
2571 | if (blk_needs_flush_plug(tsk)) | |
2572 | blk_schedule_flush_plug(tsk); | |
2573 | } | |
2574 | ||
6ebbe7a0 | 2575 | asmlinkage void __sched schedule(void) |
c259e01a | 2576 | { |
9c40cef2 TG |
2577 | struct task_struct *tsk = current; |
2578 | ||
2579 | sched_submit_work(tsk); | |
c259e01a TG |
2580 | __schedule(); |
2581 | } | |
1da177e4 LT |
2582 | EXPORT_SYMBOL(schedule); |
2583 | ||
91d1aa43 | 2584 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2585 | asmlinkage void __sched schedule_user(void) |
2586 | { | |
2587 | /* | |
2588 | * If we come here after a random call to set_need_resched(), | |
2589 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2590 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2591 | * we find a better solution. | |
2592 | */ | |
91d1aa43 | 2593 | user_exit(); |
20ab65e3 | 2594 | schedule(); |
91d1aa43 | 2595 | user_enter(); |
20ab65e3 FW |
2596 | } |
2597 | #endif | |
2598 | ||
c5491ea7 TG |
2599 | /** |
2600 | * schedule_preempt_disabled - called with preemption disabled | |
2601 | * | |
2602 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2603 | */ | |
2604 | void __sched schedule_preempt_disabled(void) | |
2605 | { | |
ba74c144 | 2606 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2607 | schedule(); |
2608 | preempt_disable(); | |
2609 | } | |
2610 | ||
1da177e4 LT |
2611 | #ifdef CONFIG_PREEMPT |
2612 | /* | |
2ed6e34f | 2613 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2614 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2615 | * occur there and call schedule directly. |
2616 | */ | |
d1f74e20 | 2617 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 | 2618 | { |
1da177e4 LT |
2619 | /* |
2620 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2621 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2622 | */ |
fbb00b56 | 2623 | if (likely(!preemptible())) |
1da177e4 LT |
2624 | return; |
2625 | ||
3a5c359a | 2626 | do { |
bdb43806 | 2627 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2628 | __schedule(); |
bdb43806 | 2629 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2630 | |
3a5c359a AK |
2631 | /* |
2632 | * Check again in case we missed a preemption opportunity | |
2633 | * between schedule and now. | |
2634 | */ | |
2635 | barrier(); | |
5ed0cec0 | 2636 | } while (need_resched()); |
1da177e4 | 2637 | } |
1da177e4 LT |
2638 | EXPORT_SYMBOL(preempt_schedule); |
2639 | ||
2640 | /* | |
2ed6e34f | 2641 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2642 | * off of irq context. |
2643 | * Note, that this is called and return with irqs disabled. This will | |
2644 | * protect us against recursive calling from irq. | |
2645 | */ | |
2646 | asmlinkage void __sched preempt_schedule_irq(void) | |
2647 | { | |
b22366cd | 2648 | enum ctx_state prev_state; |
6478d880 | 2649 | |
2ed6e34f | 2650 | /* Catch callers which need to be fixed */ |
f27dde8d | 2651 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2652 | |
b22366cd FW |
2653 | prev_state = exception_enter(); |
2654 | ||
3a5c359a | 2655 | do { |
bdb43806 | 2656 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2657 | local_irq_enable(); |
c259e01a | 2658 | __schedule(); |
3a5c359a | 2659 | local_irq_disable(); |
bdb43806 | 2660 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2661 | |
3a5c359a AK |
2662 | /* |
2663 | * Check again in case we missed a preemption opportunity | |
2664 | * between schedule and now. | |
2665 | */ | |
2666 | barrier(); | |
5ed0cec0 | 2667 | } while (need_resched()); |
b22366cd FW |
2668 | |
2669 | exception_exit(prev_state); | |
1da177e4 LT |
2670 | } |
2671 | ||
2672 | #endif /* CONFIG_PREEMPT */ | |
2673 | ||
63859d4f | 2674 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2675 | void *key) |
1da177e4 | 2676 | { |
63859d4f | 2677 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2678 | } |
1da177e4 LT |
2679 | EXPORT_SYMBOL(default_wake_function); |
2680 | ||
2681 | /* | |
41a2d6cf IM |
2682 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
2683 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
2684 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
2685 | * | |
2686 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 2687 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
2688 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
2689 | */ | |
78ddb08f | 2690 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 2691 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 2692 | { |
2e45874c | 2693 | wait_queue_t *curr, *next; |
1da177e4 | 2694 | |
2e45874c | 2695 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
2696 | unsigned flags = curr->flags; |
2697 | ||
63859d4f | 2698 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 2699 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
2700 | break; |
2701 | } | |
2702 | } | |
2703 | ||
2704 | /** | |
2705 | * __wake_up - wake up threads blocked on a waitqueue. | |
2706 | * @q: the waitqueue | |
2707 | * @mode: which threads | |
2708 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 2709 | * @key: is directly passed to the wakeup function |
50fa610a DH |
2710 | * |
2711 | * It may be assumed that this function implies a write memory barrier before | |
2712 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 2713 | */ |
7ad5b3a5 | 2714 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 2715 | int nr_exclusive, void *key) |
1da177e4 LT |
2716 | { |
2717 | unsigned long flags; | |
2718 | ||
2719 | spin_lock_irqsave(&q->lock, flags); | |
2720 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
2721 | spin_unlock_irqrestore(&q->lock, flags); | |
2722 | } | |
1da177e4 LT |
2723 | EXPORT_SYMBOL(__wake_up); |
2724 | ||
2725 | /* | |
2726 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
2727 | */ | |
63b20011 | 2728 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 2729 | { |
63b20011 | 2730 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 2731 | } |
22c43c81 | 2732 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 2733 | |
4ede816a DL |
2734 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
2735 | { | |
2736 | __wake_up_common(q, mode, 1, 0, key); | |
2737 | } | |
bf294b41 | 2738 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 2739 | |
1da177e4 | 2740 | /** |
4ede816a | 2741 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
2742 | * @q: the waitqueue |
2743 | * @mode: which threads | |
2744 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 2745 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
2746 | * |
2747 | * The sync wakeup differs that the waker knows that it will schedule | |
2748 | * away soon, so while the target thread will be woken up, it will not | |
2749 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
2750 | * with each other. This can prevent needless bouncing between CPUs. | |
2751 | * | |
2752 | * On UP it can prevent extra preemption. | |
50fa610a DH |
2753 | * |
2754 | * It may be assumed that this function implies a write memory barrier before | |
2755 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 2756 | */ |
4ede816a DL |
2757 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
2758 | int nr_exclusive, void *key) | |
1da177e4 LT |
2759 | { |
2760 | unsigned long flags; | |
7d478721 | 2761 | int wake_flags = WF_SYNC; |
1da177e4 LT |
2762 | |
2763 | if (unlikely(!q)) | |
2764 | return; | |
2765 | ||
cedce3e7 | 2766 | if (unlikely(nr_exclusive != 1)) |
7d478721 | 2767 | wake_flags = 0; |
1da177e4 LT |
2768 | |
2769 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 2770 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
2771 | spin_unlock_irqrestore(&q->lock, flags); |
2772 | } | |
4ede816a DL |
2773 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
2774 | ||
2775 | /* | |
2776 | * __wake_up_sync - see __wake_up_sync_key() | |
2777 | */ | |
2778 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
2779 | { | |
2780 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
2781 | } | |
1da177e4 LT |
2782 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
2783 | ||
65eb3dc6 KD |
2784 | /** |
2785 | * complete: - signals a single thread waiting on this completion | |
2786 | * @x: holds the state of this particular completion | |
2787 | * | |
2788 | * This will wake up a single thread waiting on this completion. Threads will be | |
2789 | * awakened in the same order in which they were queued. | |
2790 | * | |
2791 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
2792 | * |
2793 | * It may be assumed that this function implies a write memory barrier before | |
2794 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 2795 | */ |
b15136e9 | 2796 | void complete(struct completion *x) |
1da177e4 LT |
2797 | { |
2798 | unsigned long flags; | |
2799 | ||
2800 | spin_lock_irqsave(&x->wait.lock, flags); | |
2801 | x->done++; | |
d9514f6c | 2802 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
2803 | spin_unlock_irqrestore(&x->wait.lock, flags); |
2804 | } | |
2805 | EXPORT_SYMBOL(complete); | |
2806 | ||
65eb3dc6 KD |
2807 | /** |
2808 | * complete_all: - signals all threads waiting on this completion | |
2809 | * @x: holds the state of this particular completion | |
2810 | * | |
2811 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
2812 | * |
2813 | * It may be assumed that this function implies a write memory barrier before | |
2814 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 2815 | */ |
b15136e9 | 2816 | void complete_all(struct completion *x) |
1da177e4 LT |
2817 | { |
2818 | unsigned long flags; | |
2819 | ||
2820 | spin_lock_irqsave(&x->wait.lock, flags); | |
2821 | x->done += UINT_MAX/2; | |
d9514f6c | 2822 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
2823 | spin_unlock_irqrestore(&x->wait.lock, flags); |
2824 | } | |
2825 | EXPORT_SYMBOL(complete_all); | |
2826 | ||
8cbbe86d | 2827 | static inline long __sched |
686855f5 VD |
2828 | do_wait_for_common(struct completion *x, |
2829 | long (*action)(long), long timeout, int state) | |
1da177e4 | 2830 | { |
1da177e4 LT |
2831 | if (!x->done) { |
2832 | DECLARE_WAITQUEUE(wait, current); | |
2833 | ||
a93d2f17 | 2834 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 2835 | do { |
94d3d824 | 2836 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
2837 | timeout = -ERESTARTSYS; |
2838 | break; | |
8cbbe86d AK |
2839 | } |
2840 | __set_current_state(state); | |
1da177e4 | 2841 | spin_unlock_irq(&x->wait.lock); |
686855f5 | 2842 | timeout = action(timeout); |
1da177e4 | 2843 | spin_lock_irq(&x->wait.lock); |
ea71a546 | 2844 | } while (!x->done && timeout); |
1da177e4 | 2845 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
2846 | if (!x->done) |
2847 | return timeout; | |
1da177e4 LT |
2848 | } |
2849 | x->done--; | |
ea71a546 | 2850 | return timeout ?: 1; |
1da177e4 | 2851 | } |
1da177e4 | 2852 | |
686855f5 VD |
2853 | static inline long __sched |
2854 | __wait_for_common(struct completion *x, | |
2855 | long (*action)(long), long timeout, int state) | |
1da177e4 | 2856 | { |
1da177e4 LT |
2857 | might_sleep(); |
2858 | ||
2859 | spin_lock_irq(&x->wait.lock); | |
686855f5 | 2860 | timeout = do_wait_for_common(x, action, timeout, state); |
1da177e4 | 2861 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
2862 | return timeout; |
2863 | } | |
1da177e4 | 2864 | |
686855f5 VD |
2865 | static long __sched |
2866 | wait_for_common(struct completion *x, long timeout, int state) | |
2867 | { | |
2868 | return __wait_for_common(x, schedule_timeout, timeout, state); | |
2869 | } | |
2870 | ||
2871 | static long __sched | |
2872 | wait_for_common_io(struct completion *x, long timeout, int state) | |
2873 | { | |
2874 | return __wait_for_common(x, io_schedule_timeout, timeout, state); | |
2875 | } | |
2876 | ||
65eb3dc6 KD |
2877 | /** |
2878 | * wait_for_completion: - waits for completion of a task | |
2879 | * @x: holds the state of this particular completion | |
2880 | * | |
2881 | * This waits to be signaled for completion of a specific task. It is NOT | |
2882 | * interruptible and there is no timeout. | |
2883 | * | |
2884 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
2885 | * and interrupt capability. Also see complete(). | |
2886 | */ | |
b15136e9 | 2887 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
2888 | { |
2889 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 2890 | } |
8cbbe86d | 2891 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 2892 | |
65eb3dc6 KD |
2893 | /** |
2894 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
2895 | * @x: holds the state of this particular completion | |
2896 | * @timeout: timeout value in jiffies | |
2897 | * | |
2898 | * This waits for either a completion of a specific task to be signaled or for a | |
2899 | * specified timeout to expire. The timeout is in jiffies. It is not | |
2900 | * interruptible. | |
c6dc7f05 | 2901 | * |
e69f6186 YB |
2902 | * Return: 0 if timed out, and positive (at least 1, or number of jiffies left |
2903 | * till timeout) if completed. | |
65eb3dc6 | 2904 | */ |
b15136e9 | 2905 | unsigned long __sched |
8cbbe86d | 2906 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 2907 | { |
8cbbe86d | 2908 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 2909 | } |
8cbbe86d | 2910 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 2911 | |
686855f5 VD |
2912 | /** |
2913 | * wait_for_completion_io: - waits for completion of a task | |
2914 | * @x: holds the state of this particular completion | |
2915 | * | |
2916 | * This waits to be signaled for completion of a specific task. It is NOT | |
2917 | * interruptible and there is no timeout. The caller is accounted as waiting | |
2918 | * for IO. | |
2919 | */ | |
2920 | void __sched wait_for_completion_io(struct completion *x) | |
2921 | { | |
2922 | wait_for_common_io(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
2923 | } | |
2924 | EXPORT_SYMBOL(wait_for_completion_io); | |
2925 | ||
2926 | /** | |
2927 | * wait_for_completion_io_timeout: - waits for completion of a task (w/timeout) | |
2928 | * @x: holds the state of this particular completion | |
2929 | * @timeout: timeout value in jiffies | |
2930 | * | |
2931 | * This waits for either a completion of a specific task to be signaled or for a | |
2932 | * specified timeout to expire. The timeout is in jiffies. It is not | |
2933 | * interruptible. The caller is accounted as waiting for IO. | |
2934 | * | |
e69f6186 YB |
2935 | * Return: 0 if timed out, and positive (at least 1, or number of jiffies left |
2936 | * till timeout) if completed. | |
686855f5 VD |
2937 | */ |
2938 | unsigned long __sched | |
2939 | wait_for_completion_io_timeout(struct completion *x, unsigned long timeout) | |
2940 | { | |
2941 | return wait_for_common_io(x, timeout, TASK_UNINTERRUPTIBLE); | |
2942 | } | |
2943 | EXPORT_SYMBOL(wait_for_completion_io_timeout); | |
2944 | ||
65eb3dc6 KD |
2945 | /** |
2946 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
2947 | * @x: holds the state of this particular completion | |
2948 | * | |
2949 | * This waits for completion of a specific task to be signaled. It is | |
2950 | * interruptible. | |
c6dc7f05 | 2951 | * |
e69f6186 | 2952 | * Return: -ERESTARTSYS if interrupted, 0 if completed. |
65eb3dc6 | 2953 | */ |
8cbbe86d | 2954 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 2955 | { |
51e97990 AK |
2956 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
2957 | if (t == -ERESTARTSYS) | |
2958 | return t; | |
2959 | return 0; | |
0fec171c | 2960 | } |
8cbbe86d | 2961 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 2962 | |
65eb3dc6 KD |
2963 | /** |
2964 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
2965 | * @x: holds the state of this particular completion | |
2966 | * @timeout: timeout value in jiffies | |
2967 | * | |
2968 | * This waits for either a completion of a specific task to be signaled or for a | |
2969 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 | 2970 | * |
e69f6186 YB |
2971 | * Return: -ERESTARTSYS if interrupted, 0 if timed out, positive (at least 1, |
2972 | * or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 2973 | */ |
6bf41237 | 2974 | long __sched |
8cbbe86d AK |
2975 | wait_for_completion_interruptible_timeout(struct completion *x, |
2976 | unsigned long timeout) | |
0fec171c | 2977 | { |
8cbbe86d | 2978 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 2979 | } |
8cbbe86d | 2980 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 2981 | |
65eb3dc6 KD |
2982 | /** |
2983 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
2984 | * @x: holds the state of this particular completion | |
2985 | * | |
2986 | * This waits to be signaled for completion of a specific task. It can be | |
2987 | * interrupted by a kill signal. | |
c6dc7f05 | 2988 | * |
e69f6186 | 2989 | * Return: -ERESTARTSYS if interrupted, 0 if completed. |
65eb3dc6 | 2990 | */ |
009e577e MW |
2991 | int __sched wait_for_completion_killable(struct completion *x) |
2992 | { | |
2993 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
2994 | if (t == -ERESTARTSYS) | |
2995 | return t; | |
2996 | return 0; | |
2997 | } | |
2998 | EXPORT_SYMBOL(wait_for_completion_killable); | |
2999 | ||
0aa12fb4 SW |
3000 | /** |
3001 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3002 | * @x: holds the state of this particular completion | |
3003 | * @timeout: timeout value in jiffies | |
3004 | * | |
3005 | * This waits for either a completion of a specific task to be | |
3006 | * signaled or for a specified timeout to expire. It can be | |
3007 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 | 3008 | * |
e69f6186 YB |
3009 | * Return: -ERESTARTSYS if interrupted, 0 if timed out, positive (at least 1, |
3010 | * or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3011 | */ |
6bf41237 | 3012 | long __sched |
0aa12fb4 SW |
3013 | wait_for_completion_killable_timeout(struct completion *x, |
3014 | unsigned long timeout) | |
3015 | { | |
3016 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3017 | } | |
3018 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3019 | ||
be4de352 DC |
3020 | /** |
3021 | * try_wait_for_completion - try to decrement a completion without blocking | |
3022 | * @x: completion structure | |
3023 | * | |
e69f6186 | 3024 | * Return: 0 if a decrement cannot be done without blocking |
be4de352 DC |
3025 | * 1 if a decrement succeeded. |
3026 | * | |
3027 | * If a completion is being used as a counting completion, | |
3028 | * attempt to decrement the counter without blocking. This | |
3029 | * enables us to avoid waiting if the resource the completion | |
3030 | * is protecting is not available. | |
3031 | */ | |
3032 | bool try_wait_for_completion(struct completion *x) | |
3033 | { | |
7539a3b3 | 3034 | unsigned long flags; |
be4de352 DC |
3035 | int ret = 1; |
3036 | ||
7539a3b3 | 3037 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3038 | if (!x->done) |
3039 | ret = 0; | |
3040 | else | |
3041 | x->done--; | |
7539a3b3 | 3042 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3043 | return ret; |
3044 | } | |
3045 | EXPORT_SYMBOL(try_wait_for_completion); | |
3046 | ||
3047 | /** | |
3048 | * completion_done - Test to see if a completion has any waiters | |
3049 | * @x: completion structure | |
3050 | * | |
e69f6186 | 3051 | * Return: 0 if there are waiters (wait_for_completion() in progress) |
be4de352 DC |
3052 | * 1 if there are no waiters. |
3053 | * | |
3054 | */ | |
3055 | bool completion_done(struct completion *x) | |
3056 | { | |
7539a3b3 | 3057 | unsigned long flags; |
be4de352 DC |
3058 | int ret = 1; |
3059 | ||
7539a3b3 | 3060 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3061 | if (!x->done) |
3062 | ret = 0; | |
7539a3b3 | 3063 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3064 | return ret; |
3065 | } | |
3066 | EXPORT_SYMBOL(completion_done); | |
3067 | ||
8cbbe86d AK |
3068 | static long __sched |
3069 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3070 | { |
0fec171c IM |
3071 | unsigned long flags; |
3072 | wait_queue_t wait; | |
3073 | ||
3074 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3075 | |
8cbbe86d | 3076 | __set_current_state(state); |
1da177e4 | 3077 | |
8cbbe86d AK |
3078 | spin_lock_irqsave(&q->lock, flags); |
3079 | __add_wait_queue(q, &wait); | |
3080 | spin_unlock(&q->lock); | |
3081 | timeout = schedule_timeout(timeout); | |
3082 | spin_lock_irq(&q->lock); | |
3083 | __remove_wait_queue(q, &wait); | |
3084 | spin_unlock_irqrestore(&q->lock, flags); | |
3085 | ||
3086 | return timeout; | |
3087 | } | |
3088 | ||
3089 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3090 | { | |
3091 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3092 | } |
1da177e4 LT |
3093 | EXPORT_SYMBOL(interruptible_sleep_on); |
3094 | ||
0fec171c | 3095 | long __sched |
95cdf3b7 | 3096 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3097 | { |
8cbbe86d | 3098 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3099 | } |
1da177e4 LT |
3100 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3101 | ||
0fec171c | 3102 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3103 | { |
8cbbe86d | 3104 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3105 | } |
1da177e4 LT |
3106 | EXPORT_SYMBOL(sleep_on); |
3107 | ||
0fec171c | 3108 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3109 | { |
8cbbe86d | 3110 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3111 | } |
1da177e4 LT |
3112 | EXPORT_SYMBOL(sleep_on_timeout); |
3113 | ||
b29739f9 IM |
3114 | #ifdef CONFIG_RT_MUTEXES |
3115 | ||
3116 | /* | |
3117 | * rt_mutex_setprio - set the current priority of a task | |
3118 | * @p: task | |
3119 | * @prio: prio value (kernel-internal form) | |
3120 | * | |
3121 | * This function changes the 'effective' priority of a task. It does | |
3122 | * not touch ->normal_prio like __setscheduler(). | |
3123 | * | |
3124 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3125 | */ | |
36c8b586 | 3126 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3127 | { |
83b699ed | 3128 | int oldprio, on_rq, running; |
70b97a7f | 3129 | struct rq *rq; |
83ab0aa0 | 3130 | const struct sched_class *prev_class; |
b29739f9 IM |
3131 | |
3132 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3133 | ||
0122ec5b | 3134 | rq = __task_rq_lock(p); |
b29739f9 | 3135 | |
1c4dd99b TG |
3136 | /* |
3137 | * Idle task boosting is a nono in general. There is one | |
3138 | * exception, when PREEMPT_RT and NOHZ is active: | |
3139 | * | |
3140 | * The idle task calls get_next_timer_interrupt() and holds | |
3141 | * the timer wheel base->lock on the CPU and another CPU wants | |
3142 | * to access the timer (probably to cancel it). We can safely | |
3143 | * ignore the boosting request, as the idle CPU runs this code | |
3144 | * with interrupts disabled and will complete the lock | |
3145 | * protected section without being interrupted. So there is no | |
3146 | * real need to boost. | |
3147 | */ | |
3148 | if (unlikely(p == rq->idle)) { | |
3149 | WARN_ON(p != rq->curr); | |
3150 | WARN_ON(p->pi_blocked_on); | |
3151 | goto out_unlock; | |
3152 | } | |
3153 | ||
a8027073 | 3154 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3155 | oldprio = p->prio; |
83ab0aa0 | 3156 | prev_class = p->sched_class; |
fd2f4419 | 3157 | on_rq = p->on_rq; |
051a1d1a | 3158 | running = task_current(rq, p); |
0e1f3483 | 3159 | if (on_rq) |
69be72c1 | 3160 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3161 | if (running) |
3162 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3163 | |
3164 | if (rt_prio(prio)) | |
3165 | p->sched_class = &rt_sched_class; | |
3166 | else | |
3167 | p->sched_class = &fair_sched_class; | |
3168 | ||
b29739f9 IM |
3169 | p->prio = prio; |
3170 | ||
0e1f3483 HS |
3171 | if (running) |
3172 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3173 | if (on_rq) |
371fd7e7 | 3174 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3175 | |
da7a735e | 3176 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3177 | out_unlock: |
0122ec5b | 3178 | __task_rq_unlock(rq); |
b29739f9 | 3179 | } |
b29739f9 | 3180 | #endif |
36c8b586 | 3181 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3182 | { |
dd41f596 | 3183 | int old_prio, delta, on_rq; |
1da177e4 | 3184 | unsigned long flags; |
70b97a7f | 3185 | struct rq *rq; |
1da177e4 LT |
3186 | |
3187 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3188 | return; | |
3189 | /* | |
3190 | * We have to be careful, if called from sys_setpriority(), | |
3191 | * the task might be in the middle of scheduling on another CPU. | |
3192 | */ | |
3193 | rq = task_rq_lock(p, &flags); | |
3194 | /* | |
3195 | * The RT priorities are set via sched_setscheduler(), but we still | |
3196 | * allow the 'normal' nice value to be set - but as expected | |
3197 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3198 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3199 | */ |
e05606d3 | 3200 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3201 | p->static_prio = NICE_TO_PRIO(nice); |
3202 | goto out_unlock; | |
3203 | } | |
fd2f4419 | 3204 | on_rq = p->on_rq; |
c09595f6 | 3205 | if (on_rq) |
69be72c1 | 3206 | dequeue_task(rq, p, 0); |
1da177e4 | 3207 | |
1da177e4 | 3208 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3209 | set_load_weight(p); |
b29739f9 IM |
3210 | old_prio = p->prio; |
3211 | p->prio = effective_prio(p); | |
3212 | delta = p->prio - old_prio; | |
1da177e4 | 3213 | |
dd41f596 | 3214 | if (on_rq) { |
371fd7e7 | 3215 | enqueue_task(rq, p, 0); |
1da177e4 | 3216 | /* |
d5f9f942 AM |
3217 | * If the task increased its priority or is running and |
3218 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3219 | */ |
d5f9f942 | 3220 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3221 | resched_task(rq->curr); |
3222 | } | |
3223 | out_unlock: | |
0122ec5b | 3224 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3225 | } |
1da177e4 LT |
3226 | EXPORT_SYMBOL(set_user_nice); |
3227 | ||
e43379f1 MM |
3228 | /* |
3229 | * can_nice - check if a task can reduce its nice value | |
3230 | * @p: task | |
3231 | * @nice: nice value | |
3232 | */ | |
36c8b586 | 3233 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3234 | { |
024f4747 MM |
3235 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3236 | int nice_rlim = 20 - nice; | |
48f24c4d | 3237 | |
78d7d407 | 3238 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3239 | capable(CAP_SYS_NICE)); |
3240 | } | |
3241 | ||
1da177e4 LT |
3242 | #ifdef __ARCH_WANT_SYS_NICE |
3243 | ||
3244 | /* | |
3245 | * sys_nice - change the priority of the current process. | |
3246 | * @increment: priority increment | |
3247 | * | |
3248 | * sys_setpriority is a more generic, but much slower function that | |
3249 | * does similar things. | |
3250 | */ | |
5add95d4 | 3251 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3252 | { |
48f24c4d | 3253 | long nice, retval; |
1da177e4 LT |
3254 | |
3255 | /* | |
3256 | * Setpriority might change our priority at the same moment. | |
3257 | * We don't have to worry. Conceptually one call occurs first | |
3258 | * and we have a single winner. | |
3259 | */ | |
e43379f1 MM |
3260 | if (increment < -40) |
3261 | increment = -40; | |
1da177e4 LT |
3262 | if (increment > 40) |
3263 | increment = 40; | |
3264 | ||
2b8f836f | 3265 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3266 | if (nice < -20) |
3267 | nice = -20; | |
3268 | if (nice > 19) | |
3269 | nice = 19; | |
3270 | ||
e43379f1 MM |
3271 | if (increment < 0 && !can_nice(current, nice)) |
3272 | return -EPERM; | |
3273 | ||
1da177e4 LT |
3274 | retval = security_task_setnice(current, nice); |
3275 | if (retval) | |
3276 | return retval; | |
3277 | ||
3278 | set_user_nice(current, nice); | |
3279 | return 0; | |
3280 | } | |
3281 | ||
3282 | #endif | |
3283 | ||
3284 | /** | |
3285 | * task_prio - return the priority value of a given task. | |
3286 | * @p: the task in question. | |
3287 | * | |
e69f6186 | 3288 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3289 | * RT tasks are offset by -200. Normal tasks are centered |
3290 | * around 0, value goes from -16 to +15. | |
3291 | */ | |
36c8b586 | 3292 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3293 | { |
3294 | return p->prio - MAX_RT_PRIO; | |
3295 | } | |
3296 | ||
3297 | /** | |
3298 | * task_nice - return the nice value of a given task. | |
3299 | * @p: the task in question. | |
e69f6186 YB |
3300 | * |
3301 | * Return: The nice value [ -20 ... 0 ... 19 ]. | |
1da177e4 | 3302 | */ |
36c8b586 | 3303 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3304 | { |
3305 | return TASK_NICE(p); | |
3306 | } | |
150d8bed | 3307 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3308 | |
3309 | /** | |
3310 | * idle_cpu - is a given cpu idle currently? | |
3311 | * @cpu: the processor in question. | |
e69f6186 YB |
3312 | * |
3313 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3314 | */ |
3315 | int idle_cpu(int cpu) | |
3316 | { | |
908a3283 TG |
3317 | struct rq *rq = cpu_rq(cpu); |
3318 | ||
3319 | if (rq->curr != rq->idle) | |
3320 | return 0; | |
3321 | ||
3322 | if (rq->nr_running) | |
3323 | return 0; | |
3324 | ||
3325 | #ifdef CONFIG_SMP | |
3326 | if (!llist_empty(&rq->wake_list)) | |
3327 | return 0; | |
3328 | #endif | |
3329 | ||
3330 | return 1; | |
1da177e4 LT |
3331 | } |
3332 | ||
1da177e4 LT |
3333 | /** |
3334 | * idle_task - return the idle task for a given cpu. | |
3335 | * @cpu: the processor in question. | |
e69f6186 YB |
3336 | * |
3337 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3338 | */ |
36c8b586 | 3339 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3340 | { |
3341 | return cpu_rq(cpu)->idle; | |
3342 | } | |
3343 | ||
3344 | /** | |
3345 | * find_process_by_pid - find a process with a matching PID value. | |
3346 | * @pid: the pid in question. | |
e69f6186 YB |
3347 | * |
3348 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3349 | */ |
a9957449 | 3350 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3351 | { |
228ebcbe | 3352 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3353 | } |
3354 | ||
3355 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3356 | static void |
3357 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3358 | { |
1da177e4 LT |
3359 | p->policy = policy; |
3360 | p->rt_priority = prio; | |
b29739f9 IM |
3361 | p->normal_prio = normal_prio(p); |
3362 | /* we are holding p->pi_lock already */ | |
3363 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3364 | if (rt_prio(p->prio)) |
3365 | p->sched_class = &rt_sched_class; | |
3366 | else | |
3367 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3368 | set_load_weight(p); |
1da177e4 LT |
3369 | } |
3370 | ||
c69e8d9c DH |
3371 | /* |
3372 | * check the target process has a UID that matches the current process's | |
3373 | */ | |
3374 | static bool check_same_owner(struct task_struct *p) | |
3375 | { | |
3376 | const struct cred *cred = current_cred(), *pcred; | |
3377 | bool match; | |
3378 | ||
3379 | rcu_read_lock(); | |
3380 | pcred = __task_cred(p); | |
9c806aa0 EB |
3381 | match = (uid_eq(cred->euid, pcred->euid) || |
3382 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3383 | rcu_read_unlock(); |
3384 | return match; | |
3385 | } | |
3386 | ||
961ccddd | 3387 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3388 | const struct sched_param *param, bool user) |
1da177e4 | 3389 | { |
83b699ed | 3390 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3391 | unsigned long flags; |
83ab0aa0 | 3392 | const struct sched_class *prev_class; |
70b97a7f | 3393 | struct rq *rq; |
ca94c442 | 3394 | int reset_on_fork; |
1da177e4 | 3395 | |
66e5393a SR |
3396 | /* may grab non-irq protected spin_locks */ |
3397 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3398 | recheck: |
3399 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3400 | if (policy < 0) { |
3401 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3402 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3403 | } else { |
3404 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3405 | policy &= ~SCHED_RESET_ON_FORK; | |
3406 | ||
3407 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3408 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3409 | policy != SCHED_IDLE) | |
3410 | return -EINVAL; | |
3411 | } | |
3412 | ||
1da177e4 LT |
3413 | /* |
3414 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3415 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3416 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3417 | */ |
3418 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3419 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3420 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3421 | return -EINVAL; |
e05606d3 | 3422 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3423 | return -EINVAL; |
3424 | ||
37e4ab3f OC |
3425 | /* |
3426 | * Allow unprivileged RT tasks to decrease priority: | |
3427 | */ | |
961ccddd | 3428 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3429 | if (rt_policy(policy)) { |
a44702e8 ON |
3430 | unsigned long rlim_rtprio = |
3431 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3432 | |
3433 | /* can't set/change the rt policy */ | |
3434 | if (policy != p->policy && !rlim_rtprio) | |
3435 | return -EPERM; | |
3436 | ||
3437 | /* can't increase priority */ | |
3438 | if (param->sched_priority > p->rt_priority && | |
3439 | param->sched_priority > rlim_rtprio) | |
3440 | return -EPERM; | |
3441 | } | |
c02aa73b | 3442 | |
dd41f596 | 3443 | /* |
c02aa73b DH |
3444 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3445 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3446 | */ |
c02aa73b DH |
3447 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3448 | if (!can_nice(p, TASK_NICE(p))) | |
3449 | return -EPERM; | |
3450 | } | |
5fe1d75f | 3451 | |
37e4ab3f | 3452 | /* can't change other user's priorities */ |
c69e8d9c | 3453 | if (!check_same_owner(p)) |
37e4ab3f | 3454 | return -EPERM; |
ca94c442 LP |
3455 | |
3456 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3457 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3458 | return -EPERM; | |
37e4ab3f | 3459 | } |
1da177e4 | 3460 | |
725aad24 | 3461 | if (user) { |
b0ae1981 | 3462 | retval = security_task_setscheduler(p); |
725aad24 JF |
3463 | if (retval) |
3464 | return retval; | |
3465 | } | |
3466 | ||
b29739f9 IM |
3467 | /* |
3468 | * make sure no PI-waiters arrive (or leave) while we are | |
3469 | * changing the priority of the task: | |
0122ec5b | 3470 | * |
25985edc | 3471 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3472 | * runqueue lock must be held. |
3473 | */ | |
0122ec5b | 3474 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3475 | |
34f971f6 PZ |
3476 | /* |
3477 | * Changing the policy of the stop threads its a very bad idea | |
3478 | */ | |
3479 | if (p == rq->stop) { | |
0122ec5b | 3480 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3481 | return -EINVAL; |
3482 | } | |
3483 | ||
a51e9198 DF |
3484 | /* |
3485 | * If not changing anything there's no need to proceed further: | |
3486 | */ | |
3487 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
3488 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 3489 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3490 | return 0; |
3491 | } | |
3492 | ||
dc61b1d6 PZ |
3493 | #ifdef CONFIG_RT_GROUP_SCHED |
3494 | if (user) { | |
3495 | /* | |
3496 | * Do not allow realtime tasks into groups that have no runtime | |
3497 | * assigned. | |
3498 | */ | |
3499 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3500 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3501 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3502 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3503 | return -EPERM; |
3504 | } | |
3505 | } | |
3506 | #endif | |
3507 | ||
1da177e4 LT |
3508 | /* recheck policy now with rq lock held */ |
3509 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3510 | policy = oldpolicy = -1; | |
0122ec5b | 3511 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3512 | goto recheck; |
3513 | } | |
fd2f4419 | 3514 | on_rq = p->on_rq; |
051a1d1a | 3515 | running = task_current(rq, p); |
0e1f3483 | 3516 | if (on_rq) |
4ca9b72b | 3517 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3518 | if (running) |
3519 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3520 | |
ca94c442 LP |
3521 | p->sched_reset_on_fork = reset_on_fork; |
3522 | ||
1da177e4 | 3523 | oldprio = p->prio; |
83ab0aa0 | 3524 | prev_class = p->sched_class; |
dd41f596 | 3525 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 3526 | |
0e1f3483 HS |
3527 | if (running) |
3528 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3529 | if (on_rq) |
4ca9b72b | 3530 | enqueue_task(rq, p, 0); |
cb469845 | 3531 | |
da7a735e | 3532 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3533 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3534 | |
95e02ca9 TG |
3535 | rt_mutex_adjust_pi(p); |
3536 | ||
1da177e4 LT |
3537 | return 0; |
3538 | } | |
961ccddd RR |
3539 | |
3540 | /** | |
3541 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3542 | * @p: the task in question. | |
3543 | * @policy: new policy. | |
3544 | * @param: structure containing the new RT priority. | |
3545 | * | |
e69f6186 YB |
3546 | * Return: 0 on success. An error code otherwise. |
3547 | * | |
961ccddd RR |
3548 | * NOTE that the task may be already dead. |
3549 | */ | |
3550 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3551 | const struct sched_param *param) |
961ccddd RR |
3552 | { |
3553 | return __sched_setscheduler(p, policy, param, true); | |
3554 | } | |
1da177e4 LT |
3555 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3556 | ||
961ccddd RR |
3557 | /** |
3558 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3559 | * @p: the task in question. | |
3560 | * @policy: new policy. | |
3561 | * @param: structure containing the new RT priority. | |
3562 | * | |
3563 | * Just like sched_setscheduler, only don't bother checking if the | |
3564 | * current context has permission. For example, this is needed in | |
3565 | * stop_machine(): we create temporary high priority worker threads, | |
3566 | * but our caller might not have that capability. | |
e69f6186 YB |
3567 | * |
3568 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3569 | */ |
3570 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3571 | const struct sched_param *param) |
961ccddd RR |
3572 | { |
3573 | return __sched_setscheduler(p, policy, param, false); | |
3574 | } | |
3575 | ||
95cdf3b7 IM |
3576 | static int |
3577 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3578 | { |
1da177e4 LT |
3579 | struct sched_param lparam; |
3580 | struct task_struct *p; | |
36c8b586 | 3581 | int retval; |
1da177e4 LT |
3582 | |
3583 | if (!param || pid < 0) | |
3584 | return -EINVAL; | |
3585 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3586 | return -EFAULT; | |
5fe1d75f ON |
3587 | |
3588 | rcu_read_lock(); | |
3589 | retval = -ESRCH; | |
1da177e4 | 3590 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3591 | if (p != NULL) |
3592 | retval = sched_setscheduler(p, policy, &lparam); | |
3593 | rcu_read_unlock(); | |
36c8b586 | 3594 | |
1da177e4 LT |
3595 | return retval; |
3596 | } | |
3597 | ||
3598 | /** | |
3599 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3600 | * @pid: the pid in question. | |
3601 | * @policy: new policy. | |
3602 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3603 | * |
3604 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3605 | */ |
5add95d4 HC |
3606 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3607 | struct sched_param __user *, param) | |
1da177e4 | 3608 | { |
c21761f1 JB |
3609 | /* negative values for policy are not valid */ |
3610 | if (policy < 0) | |
3611 | return -EINVAL; | |
3612 | ||
1da177e4 LT |
3613 | return do_sched_setscheduler(pid, policy, param); |
3614 | } | |
3615 | ||
3616 | /** | |
3617 | * sys_sched_setparam - set/change the RT priority of a thread | |
3618 | * @pid: the pid in question. | |
3619 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3620 | * |
3621 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3622 | */ |
5add95d4 | 3623 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3624 | { |
3625 | return do_sched_setscheduler(pid, -1, param); | |
3626 | } | |
3627 | ||
3628 | /** | |
3629 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3630 | * @pid: the pid in question. | |
e69f6186 YB |
3631 | * |
3632 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3633 | * code. | |
1da177e4 | 3634 | */ |
5add95d4 | 3635 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3636 | { |
36c8b586 | 3637 | struct task_struct *p; |
3a5c359a | 3638 | int retval; |
1da177e4 LT |
3639 | |
3640 | if (pid < 0) | |
3a5c359a | 3641 | return -EINVAL; |
1da177e4 LT |
3642 | |
3643 | retval = -ESRCH; | |
5fe85be0 | 3644 | rcu_read_lock(); |
1da177e4 LT |
3645 | p = find_process_by_pid(pid); |
3646 | if (p) { | |
3647 | retval = security_task_getscheduler(p); | |
3648 | if (!retval) | |
ca94c442 LP |
3649 | retval = p->policy |
3650 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3651 | } |
5fe85be0 | 3652 | rcu_read_unlock(); |
1da177e4 LT |
3653 | return retval; |
3654 | } | |
3655 | ||
3656 | /** | |
ca94c442 | 3657 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3658 | * @pid: the pid in question. |
3659 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3660 | * |
3661 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3662 | * code. | |
1da177e4 | 3663 | */ |
5add95d4 | 3664 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3665 | { |
3666 | struct sched_param lp; | |
36c8b586 | 3667 | struct task_struct *p; |
3a5c359a | 3668 | int retval; |
1da177e4 LT |
3669 | |
3670 | if (!param || pid < 0) | |
3a5c359a | 3671 | return -EINVAL; |
1da177e4 | 3672 | |
5fe85be0 | 3673 | rcu_read_lock(); |
1da177e4 LT |
3674 | p = find_process_by_pid(pid); |
3675 | retval = -ESRCH; | |
3676 | if (!p) | |
3677 | goto out_unlock; | |
3678 | ||
3679 | retval = security_task_getscheduler(p); | |
3680 | if (retval) | |
3681 | goto out_unlock; | |
3682 | ||
3683 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3684 | rcu_read_unlock(); |
1da177e4 LT |
3685 | |
3686 | /* | |
3687 | * This one might sleep, we cannot do it with a spinlock held ... | |
3688 | */ | |
3689 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3690 | ||
1da177e4 LT |
3691 | return retval; |
3692 | ||
3693 | out_unlock: | |
5fe85be0 | 3694 | rcu_read_unlock(); |
1da177e4 LT |
3695 | return retval; |
3696 | } | |
3697 | ||
96f874e2 | 3698 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3699 | { |
5a16f3d3 | 3700 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3701 | struct task_struct *p; |
3702 | int retval; | |
1da177e4 | 3703 | |
95402b38 | 3704 | get_online_cpus(); |
23f5d142 | 3705 | rcu_read_lock(); |
1da177e4 LT |
3706 | |
3707 | p = find_process_by_pid(pid); | |
3708 | if (!p) { | |
23f5d142 | 3709 | rcu_read_unlock(); |
95402b38 | 3710 | put_online_cpus(); |
1da177e4 LT |
3711 | return -ESRCH; |
3712 | } | |
3713 | ||
23f5d142 | 3714 | /* Prevent p going away */ |
1da177e4 | 3715 | get_task_struct(p); |
23f5d142 | 3716 | rcu_read_unlock(); |
1da177e4 | 3717 | |
14a40ffc TH |
3718 | if (p->flags & PF_NO_SETAFFINITY) { |
3719 | retval = -EINVAL; | |
3720 | goto out_put_task; | |
3721 | } | |
5a16f3d3 RR |
3722 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3723 | retval = -ENOMEM; | |
3724 | goto out_put_task; | |
3725 | } | |
3726 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3727 | retval = -ENOMEM; | |
3728 | goto out_free_cpus_allowed; | |
3729 | } | |
1da177e4 | 3730 | retval = -EPERM; |
4c44aaaf EB |
3731 | if (!check_same_owner(p)) { |
3732 | rcu_read_lock(); | |
3733 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3734 | rcu_read_unlock(); | |
3735 | goto out_unlock; | |
3736 | } | |
3737 | rcu_read_unlock(); | |
3738 | } | |
1da177e4 | 3739 | |
b0ae1981 | 3740 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3741 | if (retval) |
3742 | goto out_unlock; | |
3743 | ||
5a16f3d3 RR |
3744 | cpuset_cpus_allowed(p, cpus_allowed); |
3745 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 3746 | again: |
5a16f3d3 | 3747 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3748 | |
8707d8b8 | 3749 | if (!retval) { |
5a16f3d3 RR |
3750 | cpuset_cpus_allowed(p, cpus_allowed); |
3751 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3752 | /* |
3753 | * We must have raced with a concurrent cpuset | |
3754 | * update. Just reset the cpus_allowed to the | |
3755 | * cpuset's cpus_allowed | |
3756 | */ | |
5a16f3d3 | 3757 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3758 | goto again; |
3759 | } | |
3760 | } | |
1da177e4 | 3761 | out_unlock: |
5a16f3d3 RR |
3762 | free_cpumask_var(new_mask); |
3763 | out_free_cpus_allowed: | |
3764 | free_cpumask_var(cpus_allowed); | |
3765 | out_put_task: | |
1da177e4 | 3766 | put_task_struct(p); |
95402b38 | 3767 | put_online_cpus(); |
1da177e4 LT |
3768 | return retval; |
3769 | } | |
3770 | ||
3771 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3772 | struct cpumask *new_mask) |
1da177e4 | 3773 | { |
96f874e2 RR |
3774 | if (len < cpumask_size()) |
3775 | cpumask_clear(new_mask); | |
3776 | else if (len > cpumask_size()) | |
3777 | len = cpumask_size(); | |
3778 | ||
1da177e4 LT |
3779 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3780 | } | |
3781 | ||
3782 | /** | |
3783 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3784 | * @pid: pid of the process | |
3785 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3786 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3787 | * |
3788 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3789 | */ |
5add95d4 HC |
3790 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3791 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3792 | { |
5a16f3d3 | 3793 | cpumask_var_t new_mask; |
1da177e4 LT |
3794 | int retval; |
3795 | ||
5a16f3d3 RR |
3796 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3797 | return -ENOMEM; | |
1da177e4 | 3798 | |
5a16f3d3 RR |
3799 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3800 | if (retval == 0) | |
3801 | retval = sched_setaffinity(pid, new_mask); | |
3802 | free_cpumask_var(new_mask); | |
3803 | return retval; | |
1da177e4 LT |
3804 | } |
3805 | ||
96f874e2 | 3806 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 3807 | { |
36c8b586 | 3808 | struct task_struct *p; |
31605683 | 3809 | unsigned long flags; |
1da177e4 | 3810 | int retval; |
1da177e4 | 3811 | |
95402b38 | 3812 | get_online_cpus(); |
23f5d142 | 3813 | rcu_read_lock(); |
1da177e4 LT |
3814 | |
3815 | retval = -ESRCH; | |
3816 | p = find_process_by_pid(pid); | |
3817 | if (!p) | |
3818 | goto out_unlock; | |
3819 | ||
e7834f8f DQ |
3820 | retval = security_task_getscheduler(p); |
3821 | if (retval) | |
3822 | goto out_unlock; | |
3823 | ||
013fdb80 | 3824 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 3825 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 3826 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
3827 | |
3828 | out_unlock: | |
23f5d142 | 3829 | rcu_read_unlock(); |
95402b38 | 3830 | put_online_cpus(); |
1da177e4 | 3831 | |
9531b62f | 3832 | return retval; |
1da177e4 LT |
3833 | } |
3834 | ||
3835 | /** | |
3836 | * sys_sched_getaffinity - get the cpu affinity of a process | |
3837 | * @pid: pid of the process | |
3838 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3839 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
3840 | * |
3841 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3842 | */ |
5add95d4 HC |
3843 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
3844 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
3845 | { |
3846 | int ret; | |
f17c8607 | 3847 | cpumask_var_t mask; |
1da177e4 | 3848 | |
84fba5ec | 3849 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
3850 | return -EINVAL; |
3851 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
3852 | return -EINVAL; |
3853 | ||
f17c8607 RR |
3854 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
3855 | return -ENOMEM; | |
1da177e4 | 3856 | |
f17c8607 RR |
3857 | ret = sched_getaffinity(pid, mask); |
3858 | if (ret == 0) { | |
8bc037fb | 3859 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
3860 | |
3861 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
3862 | ret = -EFAULT; |
3863 | else | |
cd3d8031 | 3864 | ret = retlen; |
f17c8607 RR |
3865 | } |
3866 | free_cpumask_var(mask); | |
1da177e4 | 3867 | |
f17c8607 | 3868 | return ret; |
1da177e4 LT |
3869 | } |
3870 | ||
3871 | /** | |
3872 | * sys_sched_yield - yield the current processor to other threads. | |
3873 | * | |
dd41f596 IM |
3874 | * This function yields the current CPU to other tasks. If there are no |
3875 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
3876 | * |
3877 | * Return: 0. | |
1da177e4 | 3878 | */ |
5add95d4 | 3879 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 3880 | { |
70b97a7f | 3881 | struct rq *rq = this_rq_lock(); |
1da177e4 | 3882 | |
2d72376b | 3883 | schedstat_inc(rq, yld_count); |
4530d7ab | 3884 | current->sched_class->yield_task(rq); |
1da177e4 LT |
3885 | |
3886 | /* | |
3887 | * Since we are going to call schedule() anyway, there's | |
3888 | * no need to preempt or enable interrupts: | |
3889 | */ | |
3890 | __release(rq->lock); | |
8a25d5de | 3891 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 3892 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 3893 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
3894 | |
3895 | schedule(); | |
3896 | ||
3897 | return 0; | |
3898 | } | |
3899 | ||
e7b38404 | 3900 | static void __cond_resched(void) |
1da177e4 | 3901 | { |
bdb43806 | 3902 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 3903 | __schedule(); |
bdb43806 | 3904 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
3905 | } |
3906 | ||
02b67cc3 | 3907 | int __sched _cond_resched(void) |
1da177e4 | 3908 | { |
d86ee480 | 3909 | if (should_resched()) { |
1da177e4 LT |
3910 | __cond_resched(); |
3911 | return 1; | |
3912 | } | |
3913 | return 0; | |
3914 | } | |
02b67cc3 | 3915 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
3916 | |
3917 | /* | |
613afbf8 | 3918 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
3919 | * call schedule, and on return reacquire the lock. |
3920 | * | |
41a2d6cf | 3921 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
3922 | * operations here to prevent schedule() from being called twice (once via |
3923 | * spin_unlock(), once by hand). | |
3924 | */ | |
613afbf8 | 3925 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 3926 | { |
d86ee480 | 3927 | int resched = should_resched(); |
6df3cecb JK |
3928 | int ret = 0; |
3929 | ||
f607c668 PZ |
3930 | lockdep_assert_held(lock); |
3931 | ||
95c354fe | 3932 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 3933 | spin_unlock(lock); |
d86ee480 | 3934 | if (resched) |
95c354fe NP |
3935 | __cond_resched(); |
3936 | else | |
3937 | cpu_relax(); | |
6df3cecb | 3938 | ret = 1; |
1da177e4 | 3939 | spin_lock(lock); |
1da177e4 | 3940 | } |
6df3cecb | 3941 | return ret; |
1da177e4 | 3942 | } |
613afbf8 | 3943 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 3944 | |
613afbf8 | 3945 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
3946 | { |
3947 | BUG_ON(!in_softirq()); | |
3948 | ||
d86ee480 | 3949 | if (should_resched()) { |
98d82567 | 3950 | local_bh_enable(); |
1da177e4 LT |
3951 | __cond_resched(); |
3952 | local_bh_disable(); | |
3953 | return 1; | |
3954 | } | |
3955 | return 0; | |
3956 | } | |
613afbf8 | 3957 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 3958 | |
1da177e4 LT |
3959 | /** |
3960 | * yield - yield the current processor to other threads. | |
3961 | * | |
8e3fabfd PZ |
3962 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
3963 | * | |
3964 | * The scheduler is at all times free to pick the calling task as the most | |
3965 | * eligible task to run, if removing the yield() call from your code breaks | |
3966 | * it, its already broken. | |
3967 | * | |
3968 | * Typical broken usage is: | |
3969 | * | |
3970 | * while (!event) | |
3971 | * yield(); | |
3972 | * | |
3973 | * where one assumes that yield() will let 'the other' process run that will | |
3974 | * make event true. If the current task is a SCHED_FIFO task that will never | |
3975 | * happen. Never use yield() as a progress guarantee!! | |
3976 | * | |
3977 | * If you want to use yield() to wait for something, use wait_event(). | |
3978 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
3979 | * If you still want to use yield(), do not! | |
1da177e4 LT |
3980 | */ |
3981 | void __sched yield(void) | |
3982 | { | |
3983 | set_current_state(TASK_RUNNING); | |
3984 | sys_sched_yield(); | |
3985 | } | |
1da177e4 LT |
3986 | EXPORT_SYMBOL(yield); |
3987 | ||
d95f4122 MG |
3988 | /** |
3989 | * yield_to - yield the current processor to another thread in | |
3990 | * your thread group, or accelerate that thread toward the | |
3991 | * processor it's on. | |
16addf95 RD |
3992 | * @p: target task |
3993 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
3994 | * |
3995 | * It's the caller's job to ensure that the target task struct | |
3996 | * can't go away on us before we can do any checks. | |
3997 | * | |
e69f6186 | 3998 | * Return: |
7b270f60 PZ |
3999 | * true (>0) if we indeed boosted the target task. |
4000 | * false (0) if we failed to boost the target. | |
4001 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4002 | */ |
4003 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4004 | { | |
4005 | struct task_struct *curr = current; | |
4006 | struct rq *rq, *p_rq; | |
4007 | unsigned long flags; | |
c3c18640 | 4008 | int yielded = 0; |
d95f4122 MG |
4009 | |
4010 | local_irq_save(flags); | |
4011 | rq = this_rq(); | |
4012 | ||
4013 | again: | |
4014 | p_rq = task_rq(p); | |
7b270f60 PZ |
4015 | /* |
4016 | * If we're the only runnable task on the rq and target rq also | |
4017 | * has only one task, there's absolutely no point in yielding. | |
4018 | */ | |
4019 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4020 | yielded = -ESRCH; | |
4021 | goto out_irq; | |
4022 | } | |
4023 | ||
d95f4122 MG |
4024 | double_rq_lock(rq, p_rq); |
4025 | while (task_rq(p) != p_rq) { | |
4026 | double_rq_unlock(rq, p_rq); | |
4027 | goto again; | |
4028 | } | |
4029 | ||
4030 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4031 | goto out_unlock; |
d95f4122 MG |
4032 | |
4033 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4034 | goto out_unlock; |
d95f4122 MG |
4035 | |
4036 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4037 | goto out_unlock; |
d95f4122 MG |
4038 | |
4039 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4040 | if (yielded) { |
d95f4122 | 4041 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4042 | /* |
4043 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4044 | * fairness. | |
4045 | */ | |
4046 | if (preempt && rq != p_rq) | |
4047 | resched_task(p_rq->curr); | |
4048 | } | |
d95f4122 | 4049 | |
7b270f60 | 4050 | out_unlock: |
d95f4122 | 4051 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4052 | out_irq: |
d95f4122 MG |
4053 | local_irq_restore(flags); |
4054 | ||
7b270f60 | 4055 | if (yielded > 0) |
d95f4122 MG |
4056 | schedule(); |
4057 | ||
4058 | return yielded; | |
4059 | } | |
4060 | EXPORT_SYMBOL_GPL(yield_to); | |
4061 | ||
1da177e4 | 4062 | /* |
41a2d6cf | 4063 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4064 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4065 | */ |
4066 | void __sched io_schedule(void) | |
4067 | { | |
54d35f29 | 4068 | struct rq *rq = raw_rq(); |
1da177e4 | 4069 | |
0ff92245 | 4070 | delayacct_blkio_start(); |
1da177e4 | 4071 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4072 | blk_flush_plug(current); |
8f0dfc34 | 4073 | current->in_iowait = 1; |
1da177e4 | 4074 | schedule(); |
8f0dfc34 | 4075 | current->in_iowait = 0; |
1da177e4 | 4076 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4077 | delayacct_blkio_end(); |
1da177e4 | 4078 | } |
1da177e4 LT |
4079 | EXPORT_SYMBOL(io_schedule); |
4080 | ||
4081 | long __sched io_schedule_timeout(long timeout) | |
4082 | { | |
54d35f29 | 4083 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4084 | long ret; |
4085 | ||
0ff92245 | 4086 | delayacct_blkio_start(); |
1da177e4 | 4087 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4088 | blk_flush_plug(current); |
8f0dfc34 | 4089 | current->in_iowait = 1; |
1da177e4 | 4090 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4091 | current->in_iowait = 0; |
1da177e4 | 4092 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4093 | delayacct_blkio_end(); |
1da177e4 LT |
4094 | return ret; |
4095 | } | |
4096 | ||
4097 | /** | |
4098 | * sys_sched_get_priority_max - return maximum RT priority. | |
4099 | * @policy: scheduling class. | |
4100 | * | |
e69f6186 YB |
4101 | * Return: On success, this syscall returns the maximum |
4102 | * rt_priority that can be used by a given scheduling class. | |
4103 | * On failure, a negative error code is returned. | |
1da177e4 | 4104 | */ |
5add95d4 | 4105 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4106 | { |
4107 | int ret = -EINVAL; | |
4108 | ||
4109 | switch (policy) { | |
4110 | case SCHED_FIFO: | |
4111 | case SCHED_RR: | |
4112 | ret = MAX_USER_RT_PRIO-1; | |
4113 | break; | |
4114 | case SCHED_NORMAL: | |
b0a9499c | 4115 | case SCHED_BATCH: |
dd41f596 | 4116 | case SCHED_IDLE: |
1da177e4 LT |
4117 | ret = 0; |
4118 | break; | |
4119 | } | |
4120 | return ret; | |
4121 | } | |
4122 | ||
4123 | /** | |
4124 | * sys_sched_get_priority_min - return minimum RT priority. | |
4125 | * @policy: scheduling class. | |
4126 | * | |
e69f6186 YB |
4127 | * Return: On success, this syscall returns the minimum |
4128 | * rt_priority that can be used by a given scheduling class. | |
4129 | * On failure, a negative error code is returned. | |
1da177e4 | 4130 | */ |
5add95d4 | 4131 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4132 | { |
4133 | int ret = -EINVAL; | |
4134 | ||
4135 | switch (policy) { | |
4136 | case SCHED_FIFO: | |
4137 | case SCHED_RR: | |
4138 | ret = 1; | |
4139 | break; | |
4140 | case SCHED_NORMAL: | |
b0a9499c | 4141 | case SCHED_BATCH: |
dd41f596 | 4142 | case SCHED_IDLE: |
1da177e4 LT |
4143 | ret = 0; |
4144 | } | |
4145 | return ret; | |
4146 | } | |
4147 | ||
4148 | /** | |
4149 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4150 | * @pid: pid of the process. | |
4151 | * @interval: userspace pointer to the timeslice value. | |
4152 | * | |
4153 | * this syscall writes the default timeslice value of a given process | |
4154 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4155 | * |
4156 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4157 | * an error code. | |
1da177e4 | 4158 | */ |
17da2bd9 | 4159 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4160 | struct timespec __user *, interval) |
1da177e4 | 4161 | { |
36c8b586 | 4162 | struct task_struct *p; |
a4ec24b4 | 4163 | unsigned int time_slice; |
dba091b9 TG |
4164 | unsigned long flags; |
4165 | struct rq *rq; | |
3a5c359a | 4166 | int retval; |
1da177e4 | 4167 | struct timespec t; |
1da177e4 LT |
4168 | |
4169 | if (pid < 0) | |
3a5c359a | 4170 | return -EINVAL; |
1da177e4 LT |
4171 | |
4172 | retval = -ESRCH; | |
1a551ae7 | 4173 | rcu_read_lock(); |
1da177e4 LT |
4174 | p = find_process_by_pid(pid); |
4175 | if (!p) | |
4176 | goto out_unlock; | |
4177 | ||
4178 | retval = security_task_getscheduler(p); | |
4179 | if (retval) | |
4180 | goto out_unlock; | |
4181 | ||
dba091b9 TG |
4182 | rq = task_rq_lock(p, &flags); |
4183 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4184 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4185 | |
1a551ae7 | 4186 | rcu_read_unlock(); |
a4ec24b4 | 4187 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4188 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4189 | return retval; |
3a5c359a | 4190 | |
1da177e4 | 4191 | out_unlock: |
1a551ae7 | 4192 | rcu_read_unlock(); |
1da177e4 LT |
4193 | return retval; |
4194 | } | |
4195 | ||
7c731e0a | 4196 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4197 | |
82a1fcb9 | 4198 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4199 | { |
1da177e4 | 4200 | unsigned long free = 0; |
4e79752c | 4201 | int ppid; |
36c8b586 | 4202 | unsigned state; |
1da177e4 | 4203 | |
1da177e4 | 4204 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4205 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4206 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4207 | #if BITS_PER_LONG == 32 |
1da177e4 | 4208 | if (state == TASK_RUNNING) |
3df0fc5b | 4209 | printk(KERN_CONT " running "); |
1da177e4 | 4210 | else |
3df0fc5b | 4211 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4212 | #else |
4213 | if (state == TASK_RUNNING) | |
3df0fc5b | 4214 | printk(KERN_CONT " running task "); |
1da177e4 | 4215 | else |
3df0fc5b | 4216 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4217 | #endif |
4218 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4219 | free = stack_not_used(p); |
1da177e4 | 4220 | #endif |
4e79752c PM |
4221 | rcu_read_lock(); |
4222 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4223 | rcu_read_unlock(); | |
3df0fc5b | 4224 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4225 | task_pid_nr(p), ppid, |
aa47b7e0 | 4226 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4227 | |
3d1cb205 | 4228 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4229 | show_stack(p, NULL); |
1da177e4 LT |
4230 | } |
4231 | ||
e59e2ae2 | 4232 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4233 | { |
36c8b586 | 4234 | struct task_struct *g, *p; |
1da177e4 | 4235 | |
4bd77321 | 4236 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4237 | printk(KERN_INFO |
4238 | " task PC stack pid father\n"); | |
1da177e4 | 4239 | #else |
3df0fc5b PZ |
4240 | printk(KERN_INFO |
4241 | " task PC stack pid father\n"); | |
1da177e4 | 4242 | #endif |
510f5acc | 4243 | rcu_read_lock(); |
1da177e4 LT |
4244 | do_each_thread(g, p) { |
4245 | /* | |
4246 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4247 | * console might take a lot of time: |
1da177e4 LT |
4248 | */ |
4249 | touch_nmi_watchdog(); | |
39bc89fd | 4250 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4251 | sched_show_task(p); |
1da177e4 LT |
4252 | } while_each_thread(g, p); |
4253 | ||
04c9167f JF |
4254 | touch_all_softlockup_watchdogs(); |
4255 | ||
dd41f596 IM |
4256 | #ifdef CONFIG_SCHED_DEBUG |
4257 | sysrq_sched_debug_show(); | |
4258 | #endif | |
510f5acc | 4259 | rcu_read_unlock(); |
e59e2ae2 IM |
4260 | /* |
4261 | * Only show locks if all tasks are dumped: | |
4262 | */ | |
93335a21 | 4263 | if (!state_filter) |
e59e2ae2 | 4264 | debug_show_all_locks(); |
1da177e4 LT |
4265 | } |
4266 | ||
0db0628d | 4267 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4268 | { |
dd41f596 | 4269 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4270 | } |
4271 | ||
f340c0d1 IM |
4272 | /** |
4273 | * init_idle - set up an idle thread for a given CPU | |
4274 | * @idle: task in question | |
4275 | * @cpu: cpu the idle task belongs to | |
4276 | * | |
4277 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4278 | * flag, to make booting more robust. | |
4279 | */ | |
0db0628d | 4280 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4281 | { |
70b97a7f | 4282 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4283 | unsigned long flags; |
4284 | ||
05fa785c | 4285 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4286 | |
dd41f596 | 4287 | __sched_fork(idle); |
06b83b5f | 4288 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4289 | idle->se.exec_start = sched_clock(); |
4290 | ||
1e1b6c51 | 4291 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4292 | /* |
4293 | * We're having a chicken and egg problem, even though we are | |
4294 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4295 | * lockdep check in task_group() will fail. | |
4296 | * | |
4297 | * Similar case to sched_fork(). / Alternatively we could | |
4298 | * use task_rq_lock() here and obtain the other rq->lock. | |
4299 | * | |
4300 | * Silence PROVE_RCU | |
4301 | */ | |
4302 | rcu_read_lock(); | |
dd41f596 | 4303 | __set_task_cpu(idle, cpu); |
6506cf6c | 4304 | rcu_read_unlock(); |
1da177e4 | 4305 | |
1da177e4 | 4306 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4307 | #if defined(CONFIG_SMP) |
4308 | idle->on_cpu = 1; | |
4866cde0 | 4309 | #endif |
05fa785c | 4310 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4311 | |
4312 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4313 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4314 | |
dd41f596 IM |
4315 | /* |
4316 | * The idle tasks have their own, simple scheduling class: | |
4317 | */ | |
4318 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4319 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4320 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4321 | #if defined(CONFIG_SMP) |
4322 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4323 | #endif | |
19978ca6 IM |
4324 | } |
4325 | ||
1da177e4 | 4326 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4327 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4328 | { | |
4329 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4330 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4331 | |
4332 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4333 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4334 | } |
4335 | ||
1da177e4 LT |
4336 | /* |
4337 | * This is how migration works: | |
4338 | * | |
969c7921 TH |
4339 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4340 | * stop_one_cpu(). | |
4341 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4342 | * off the CPU) | |
4343 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4344 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4345 | * it and puts it into the right queue. |
969c7921 TH |
4346 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4347 | * is done. | |
1da177e4 LT |
4348 | */ |
4349 | ||
4350 | /* | |
4351 | * Change a given task's CPU affinity. Migrate the thread to a | |
4352 | * proper CPU and schedule it away if the CPU it's executing on | |
4353 | * is removed from the allowed bitmask. | |
4354 | * | |
4355 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4356 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4357 | * call is not atomic; no spinlocks may be held. |
4358 | */ | |
96f874e2 | 4359 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4360 | { |
4361 | unsigned long flags; | |
70b97a7f | 4362 | struct rq *rq; |
969c7921 | 4363 | unsigned int dest_cpu; |
48f24c4d | 4364 | int ret = 0; |
1da177e4 LT |
4365 | |
4366 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4367 | |
db44fc01 YZ |
4368 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4369 | goto out; | |
4370 | ||
6ad4c188 | 4371 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4372 | ret = -EINVAL; |
4373 | goto out; | |
4374 | } | |
4375 | ||
1e1b6c51 | 4376 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4377 | |
1da177e4 | 4378 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4379 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4380 | goto out; |
4381 | ||
969c7921 | 4382 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4383 | if (p->on_rq) { |
969c7921 | 4384 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4385 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4386 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4387 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4388 | tlb_migrate_finish(p->mm); |
4389 | return 0; | |
4390 | } | |
4391 | out: | |
0122ec5b | 4392 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4393 | |
1da177e4 LT |
4394 | return ret; |
4395 | } | |
cd8ba7cd | 4396 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4397 | |
4398 | /* | |
41a2d6cf | 4399 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4400 | * this because either it can't run here any more (set_cpus_allowed() |
4401 | * away from this CPU, or CPU going down), or because we're | |
4402 | * attempting to rebalance this task on exec (sched_exec). | |
4403 | * | |
4404 | * So we race with normal scheduler movements, but that's OK, as long | |
4405 | * as the task is no longer on this CPU. | |
efc30814 KK |
4406 | * |
4407 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4408 | */ |
efc30814 | 4409 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4410 | { |
70b97a7f | 4411 | struct rq *rq_dest, *rq_src; |
e2912009 | 4412 | int ret = 0; |
1da177e4 | 4413 | |
e761b772 | 4414 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4415 | return ret; |
1da177e4 LT |
4416 | |
4417 | rq_src = cpu_rq(src_cpu); | |
4418 | rq_dest = cpu_rq(dest_cpu); | |
4419 | ||
0122ec5b | 4420 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4421 | double_rq_lock(rq_src, rq_dest); |
4422 | /* Already moved. */ | |
4423 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4424 | goto done; |
1da177e4 | 4425 | /* Affinity changed (again). */ |
fa17b507 | 4426 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4427 | goto fail; |
1da177e4 | 4428 | |
e2912009 PZ |
4429 | /* |
4430 | * If we're not on a rq, the next wake-up will ensure we're | |
4431 | * placed properly. | |
4432 | */ | |
fd2f4419 | 4433 | if (p->on_rq) { |
4ca9b72b | 4434 | dequeue_task(rq_src, p, 0); |
e2912009 | 4435 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4436 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4437 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4438 | } |
b1e38734 | 4439 | done: |
efc30814 | 4440 | ret = 1; |
b1e38734 | 4441 | fail: |
1da177e4 | 4442 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4443 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4444 | return ret; |
1da177e4 LT |
4445 | } |
4446 | ||
e6628d5b MG |
4447 | #ifdef CONFIG_NUMA_BALANCING |
4448 | /* Migrate current task p to target_cpu */ | |
4449 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4450 | { | |
4451 | struct migration_arg arg = { p, target_cpu }; | |
4452 | int curr_cpu = task_cpu(p); | |
4453 | ||
4454 | if (curr_cpu == target_cpu) | |
4455 | return 0; | |
4456 | ||
4457 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4458 | return -EINVAL; | |
4459 | ||
4460 | /* TODO: This is not properly updating schedstats */ | |
4461 | ||
4462 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); | |
4463 | } | |
4464 | #endif | |
4465 | ||
1da177e4 | 4466 | /* |
969c7921 TH |
4467 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4468 | * and performs thread migration by bumping thread off CPU then | |
4469 | * 'pushing' onto another runqueue. | |
1da177e4 | 4470 | */ |
969c7921 | 4471 | static int migration_cpu_stop(void *data) |
1da177e4 | 4472 | { |
969c7921 | 4473 | struct migration_arg *arg = data; |
f7b4cddc | 4474 | |
969c7921 TH |
4475 | /* |
4476 | * The original target cpu might have gone down and we might | |
4477 | * be on another cpu but it doesn't matter. | |
4478 | */ | |
f7b4cddc | 4479 | local_irq_disable(); |
969c7921 | 4480 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4481 | local_irq_enable(); |
1da177e4 | 4482 | return 0; |
f7b4cddc ON |
4483 | } |
4484 | ||
1da177e4 | 4485 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4486 | |
054b9108 | 4487 | /* |
48c5ccae PZ |
4488 | * Ensures that the idle task is using init_mm right before its cpu goes |
4489 | * offline. | |
054b9108 | 4490 | */ |
48c5ccae | 4491 | void idle_task_exit(void) |
1da177e4 | 4492 | { |
48c5ccae | 4493 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4494 | |
48c5ccae | 4495 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4496 | |
48c5ccae PZ |
4497 | if (mm != &init_mm) |
4498 | switch_mm(mm, &init_mm, current); | |
4499 | mmdrop(mm); | |
1da177e4 LT |
4500 | } |
4501 | ||
4502 | /* | |
5d180232 PZ |
4503 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4504 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4505 | * nr_active count is stable. | |
4506 | * | |
4507 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4508 | */ |
5d180232 | 4509 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4510 | { |
5d180232 PZ |
4511 | long delta = calc_load_fold_active(rq); |
4512 | if (delta) | |
4513 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4514 | } |
4515 | ||
48f24c4d | 4516 | /* |
48c5ccae PZ |
4517 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4518 | * try_to_wake_up()->select_task_rq(). | |
4519 | * | |
4520 | * Called with rq->lock held even though we'er in stop_machine() and | |
4521 | * there's no concurrency possible, we hold the required locks anyway | |
4522 | * because of lock validation efforts. | |
1da177e4 | 4523 | */ |
48c5ccae | 4524 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4525 | { |
70b97a7f | 4526 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4527 | struct task_struct *next, *stop = rq->stop; |
4528 | int dest_cpu; | |
1da177e4 LT |
4529 | |
4530 | /* | |
48c5ccae PZ |
4531 | * Fudge the rq selection such that the below task selection loop |
4532 | * doesn't get stuck on the currently eligible stop task. | |
4533 | * | |
4534 | * We're currently inside stop_machine() and the rq is either stuck | |
4535 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4536 | * either way we should never end up calling schedule() until we're | |
4537 | * done here. | |
1da177e4 | 4538 | */ |
48c5ccae | 4539 | rq->stop = NULL; |
48f24c4d | 4540 | |
77bd3970 FW |
4541 | /* |
4542 | * put_prev_task() and pick_next_task() sched | |
4543 | * class method both need to have an up-to-date | |
4544 | * value of rq->clock[_task] | |
4545 | */ | |
4546 | update_rq_clock(rq); | |
4547 | ||
dd41f596 | 4548 | for ( ; ; ) { |
48c5ccae PZ |
4549 | /* |
4550 | * There's this thread running, bail when that's the only | |
4551 | * remaining thread. | |
4552 | */ | |
4553 | if (rq->nr_running == 1) | |
dd41f596 | 4554 | break; |
48c5ccae | 4555 | |
b67802ea | 4556 | next = pick_next_task(rq); |
48c5ccae | 4557 | BUG_ON(!next); |
79c53799 | 4558 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4559 | |
48c5ccae PZ |
4560 | /* Find suitable destination for @next, with force if needed. */ |
4561 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4562 | raw_spin_unlock(&rq->lock); | |
4563 | ||
4564 | __migrate_task(next, dead_cpu, dest_cpu); | |
4565 | ||
4566 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4567 | } |
dce48a84 | 4568 | |
48c5ccae | 4569 | rq->stop = stop; |
dce48a84 | 4570 | } |
48c5ccae | 4571 | |
1da177e4 LT |
4572 | #endif /* CONFIG_HOTPLUG_CPU */ |
4573 | ||
e692ab53 NP |
4574 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4575 | ||
4576 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4577 | { |
4578 | .procname = "sched_domain", | |
c57baf1e | 4579 | .mode = 0555, |
e0361851 | 4580 | }, |
56992309 | 4581 | {} |
e692ab53 NP |
4582 | }; |
4583 | ||
4584 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4585 | { |
4586 | .procname = "kernel", | |
c57baf1e | 4587 | .mode = 0555, |
e0361851 AD |
4588 | .child = sd_ctl_dir, |
4589 | }, | |
56992309 | 4590 | {} |
e692ab53 NP |
4591 | }; |
4592 | ||
4593 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4594 | { | |
4595 | struct ctl_table *entry = | |
5cf9f062 | 4596 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4597 | |
e692ab53 NP |
4598 | return entry; |
4599 | } | |
4600 | ||
6382bc90 MM |
4601 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4602 | { | |
cd790076 | 4603 | struct ctl_table *entry; |
6382bc90 | 4604 | |
cd790076 MM |
4605 | /* |
4606 | * In the intermediate directories, both the child directory and | |
4607 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4608 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4609 | * static strings and all have proc handlers. |
4610 | */ | |
4611 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4612 | if (entry->child) |
4613 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4614 | if (entry->proc_handler == NULL) |
4615 | kfree(entry->procname); | |
4616 | } | |
6382bc90 MM |
4617 | |
4618 | kfree(*tablep); | |
4619 | *tablep = NULL; | |
4620 | } | |
4621 | ||
201c373e | 4622 | static int min_load_idx = 0; |
fd9b86d3 | 4623 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4624 | |
e692ab53 | 4625 | static void |
e0361851 | 4626 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4627 | const char *procname, void *data, int maxlen, |
201c373e NK |
4628 | umode_t mode, proc_handler *proc_handler, |
4629 | bool load_idx) | |
e692ab53 | 4630 | { |
e692ab53 NP |
4631 | entry->procname = procname; |
4632 | entry->data = data; | |
4633 | entry->maxlen = maxlen; | |
4634 | entry->mode = mode; | |
4635 | entry->proc_handler = proc_handler; | |
201c373e NK |
4636 | |
4637 | if (load_idx) { | |
4638 | entry->extra1 = &min_load_idx; | |
4639 | entry->extra2 = &max_load_idx; | |
4640 | } | |
e692ab53 NP |
4641 | } |
4642 | ||
4643 | static struct ctl_table * | |
4644 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4645 | { | |
a5d8c348 | 4646 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 4647 | |
ad1cdc1d MM |
4648 | if (table == NULL) |
4649 | return NULL; | |
4650 | ||
e0361851 | 4651 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4652 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4653 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4654 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4655 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4656 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4657 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4658 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4659 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4660 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4661 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4662 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4663 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4664 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4665 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4666 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4667 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4668 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4669 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4670 | &sd->cache_nice_tries, |
201c373e | 4671 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4672 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4673 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 4674 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 4675 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 4676 | /* &table[12] is terminator */ |
e692ab53 NP |
4677 | |
4678 | return table; | |
4679 | } | |
4680 | ||
be7002e6 | 4681 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4682 | { |
4683 | struct ctl_table *entry, *table; | |
4684 | struct sched_domain *sd; | |
4685 | int domain_num = 0, i; | |
4686 | char buf[32]; | |
4687 | ||
4688 | for_each_domain(cpu, sd) | |
4689 | domain_num++; | |
4690 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4691 | if (table == NULL) |
4692 | return NULL; | |
e692ab53 NP |
4693 | |
4694 | i = 0; | |
4695 | for_each_domain(cpu, sd) { | |
4696 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4697 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4698 | entry->mode = 0555; |
e692ab53 NP |
4699 | entry->child = sd_alloc_ctl_domain_table(sd); |
4700 | entry++; | |
4701 | i++; | |
4702 | } | |
4703 | return table; | |
4704 | } | |
4705 | ||
4706 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4707 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4708 | { |
6ad4c188 | 4709 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4710 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4711 | char buf[32]; | |
4712 | ||
7378547f MM |
4713 | WARN_ON(sd_ctl_dir[0].child); |
4714 | sd_ctl_dir[0].child = entry; | |
4715 | ||
ad1cdc1d MM |
4716 | if (entry == NULL) |
4717 | return; | |
4718 | ||
6ad4c188 | 4719 | for_each_possible_cpu(i) { |
e692ab53 | 4720 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4721 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4722 | entry->mode = 0555; |
e692ab53 | 4723 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4724 | entry++; |
e692ab53 | 4725 | } |
7378547f MM |
4726 | |
4727 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4728 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4729 | } | |
6382bc90 | 4730 | |
7378547f | 4731 | /* may be called multiple times per register */ |
6382bc90 MM |
4732 | static void unregister_sched_domain_sysctl(void) |
4733 | { | |
7378547f MM |
4734 | if (sd_sysctl_header) |
4735 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4736 | sd_sysctl_header = NULL; |
7378547f MM |
4737 | if (sd_ctl_dir[0].child) |
4738 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4739 | } |
e692ab53 | 4740 | #else |
6382bc90 MM |
4741 | static void register_sched_domain_sysctl(void) |
4742 | { | |
4743 | } | |
4744 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4745 | { |
4746 | } | |
4747 | #endif | |
4748 | ||
1f11eb6a GH |
4749 | static void set_rq_online(struct rq *rq) |
4750 | { | |
4751 | if (!rq->online) { | |
4752 | const struct sched_class *class; | |
4753 | ||
c6c4927b | 4754 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4755 | rq->online = 1; |
4756 | ||
4757 | for_each_class(class) { | |
4758 | if (class->rq_online) | |
4759 | class->rq_online(rq); | |
4760 | } | |
4761 | } | |
4762 | } | |
4763 | ||
4764 | static void set_rq_offline(struct rq *rq) | |
4765 | { | |
4766 | if (rq->online) { | |
4767 | const struct sched_class *class; | |
4768 | ||
4769 | for_each_class(class) { | |
4770 | if (class->rq_offline) | |
4771 | class->rq_offline(rq); | |
4772 | } | |
4773 | ||
c6c4927b | 4774 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4775 | rq->online = 0; |
4776 | } | |
4777 | } | |
4778 | ||
1da177e4 LT |
4779 | /* |
4780 | * migration_call - callback that gets triggered when a CPU is added. | |
4781 | * Here we can start up the necessary migration thread for the new CPU. | |
4782 | */ | |
0db0628d | 4783 | static int |
48f24c4d | 4784 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 4785 | { |
48f24c4d | 4786 | int cpu = (long)hcpu; |
1da177e4 | 4787 | unsigned long flags; |
969c7921 | 4788 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 4789 | |
48c5ccae | 4790 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 4791 | |
1da177e4 | 4792 | case CPU_UP_PREPARE: |
a468d389 | 4793 | rq->calc_load_update = calc_load_update; |
1da177e4 | 4794 | break; |
48f24c4d | 4795 | |
1da177e4 | 4796 | case CPU_ONLINE: |
1f94ef59 | 4797 | /* Update our root-domain */ |
05fa785c | 4798 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 4799 | if (rq->rd) { |
c6c4927b | 4800 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
4801 | |
4802 | set_rq_online(rq); | |
1f94ef59 | 4803 | } |
05fa785c | 4804 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 4805 | break; |
48f24c4d | 4806 | |
1da177e4 | 4807 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 4808 | case CPU_DYING: |
317f3941 | 4809 | sched_ttwu_pending(); |
57d885fe | 4810 | /* Update our root-domain */ |
05fa785c | 4811 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 4812 | if (rq->rd) { |
c6c4927b | 4813 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 4814 | set_rq_offline(rq); |
57d885fe | 4815 | } |
48c5ccae PZ |
4816 | migrate_tasks(cpu); |
4817 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 4818 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 4819 | break; |
48c5ccae | 4820 | |
5d180232 | 4821 | case CPU_DEAD: |
f319da0c | 4822 | calc_load_migrate(rq); |
57d885fe | 4823 | break; |
1da177e4 LT |
4824 | #endif |
4825 | } | |
49c022e6 PZ |
4826 | |
4827 | update_max_interval(); | |
4828 | ||
1da177e4 LT |
4829 | return NOTIFY_OK; |
4830 | } | |
4831 | ||
f38b0820 PM |
4832 | /* |
4833 | * Register at high priority so that task migration (migrate_all_tasks) | |
4834 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 4835 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 4836 | */ |
0db0628d | 4837 | static struct notifier_block migration_notifier = { |
1da177e4 | 4838 | .notifier_call = migration_call, |
50a323b7 | 4839 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
4840 | }; |
4841 | ||
0db0628d | 4842 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
4843 | unsigned long action, void *hcpu) |
4844 | { | |
4845 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 4846 | case CPU_STARTING: |
3a101d05 TH |
4847 | case CPU_DOWN_FAILED: |
4848 | set_cpu_active((long)hcpu, true); | |
4849 | return NOTIFY_OK; | |
4850 | default: | |
4851 | return NOTIFY_DONE; | |
4852 | } | |
4853 | } | |
4854 | ||
0db0628d | 4855 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
4856 | unsigned long action, void *hcpu) |
4857 | { | |
4858 | switch (action & ~CPU_TASKS_FROZEN) { | |
4859 | case CPU_DOWN_PREPARE: | |
4860 | set_cpu_active((long)hcpu, false); | |
4861 | return NOTIFY_OK; | |
4862 | default: | |
4863 | return NOTIFY_DONE; | |
4864 | } | |
4865 | } | |
4866 | ||
7babe8db | 4867 | static int __init migration_init(void) |
1da177e4 LT |
4868 | { |
4869 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 4870 | int err; |
48f24c4d | 4871 | |
3a101d05 | 4872 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
4873 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
4874 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
4875 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
4876 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 4877 | |
3a101d05 TH |
4878 | /* Register cpu active notifiers */ |
4879 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
4880 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
4881 | ||
a004cd42 | 4882 | return 0; |
1da177e4 | 4883 | } |
7babe8db | 4884 | early_initcall(migration_init); |
1da177e4 LT |
4885 | #endif |
4886 | ||
4887 | #ifdef CONFIG_SMP | |
476f3534 | 4888 | |
4cb98839 PZ |
4889 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
4890 | ||
3e9830dc | 4891 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 4892 | |
d039ac60 | 4893 | static __read_mostly int sched_debug_enabled; |
f6630114 | 4894 | |
d039ac60 | 4895 | static int __init sched_debug_setup(char *str) |
f6630114 | 4896 | { |
d039ac60 | 4897 | sched_debug_enabled = 1; |
f6630114 MT |
4898 | |
4899 | return 0; | |
4900 | } | |
d039ac60 PZ |
4901 | early_param("sched_debug", sched_debug_setup); |
4902 | ||
4903 | static inline bool sched_debug(void) | |
4904 | { | |
4905 | return sched_debug_enabled; | |
4906 | } | |
f6630114 | 4907 | |
7c16ec58 | 4908 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 4909 | struct cpumask *groupmask) |
1da177e4 | 4910 | { |
4dcf6aff | 4911 | struct sched_group *group = sd->groups; |
434d53b0 | 4912 | char str[256]; |
1da177e4 | 4913 | |
968ea6d8 | 4914 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 4915 | cpumask_clear(groupmask); |
4dcf6aff IM |
4916 | |
4917 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
4918 | ||
4919 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 4920 | printk("does not load-balance\n"); |
4dcf6aff | 4921 | if (sd->parent) |
3df0fc5b PZ |
4922 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
4923 | " has parent"); | |
4dcf6aff | 4924 | return -1; |
41c7ce9a NP |
4925 | } |
4926 | ||
3df0fc5b | 4927 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 4928 | |
758b2cdc | 4929 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
4930 | printk(KERN_ERR "ERROR: domain->span does not contain " |
4931 | "CPU%d\n", cpu); | |
4dcf6aff | 4932 | } |
758b2cdc | 4933 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
4934 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
4935 | " CPU%d\n", cpu); | |
4dcf6aff | 4936 | } |
1da177e4 | 4937 | |
4dcf6aff | 4938 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 4939 | do { |
4dcf6aff | 4940 | if (!group) { |
3df0fc5b PZ |
4941 | printk("\n"); |
4942 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
4943 | break; |
4944 | } | |
4945 | ||
c3decf0d PZ |
4946 | /* |
4947 | * Even though we initialize ->power to something semi-sane, | |
4948 | * we leave power_orig unset. This allows us to detect if | |
4949 | * domain iteration is still funny without causing /0 traps. | |
4950 | */ | |
4951 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
4952 | printk(KERN_CONT "\n"); |
4953 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
4954 | "set\n"); | |
4dcf6aff IM |
4955 | break; |
4956 | } | |
1da177e4 | 4957 | |
758b2cdc | 4958 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
4959 | printk(KERN_CONT "\n"); |
4960 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
4961 | break; |
4962 | } | |
1da177e4 | 4963 | |
cb83b629 PZ |
4964 | if (!(sd->flags & SD_OVERLAP) && |
4965 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
4966 | printk(KERN_CONT "\n"); |
4967 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
4968 | break; |
4969 | } | |
1da177e4 | 4970 | |
758b2cdc | 4971 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 4972 | |
968ea6d8 | 4973 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 4974 | |
3df0fc5b | 4975 | printk(KERN_CONT " %s", str); |
9c3f75cb | 4976 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 4977 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 4978 | group->sgp->power); |
381512cf | 4979 | } |
1da177e4 | 4980 | |
4dcf6aff IM |
4981 | group = group->next; |
4982 | } while (group != sd->groups); | |
3df0fc5b | 4983 | printk(KERN_CONT "\n"); |
1da177e4 | 4984 | |
758b2cdc | 4985 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 4986 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 4987 | |
758b2cdc RR |
4988 | if (sd->parent && |
4989 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
4990 | printk(KERN_ERR "ERROR: parent span is not a superset " |
4991 | "of domain->span\n"); | |
4dcf6aff IM |
4992 | return 0; |
4993 | } | |
1da177e4 | 4994 | |
4dcf6aff IM |
4995 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
4996 | { | |
4997 | int level = 0; | |
1da177e4 | 4998 | |
d039ac60 | 4999 | if (!sched_debug_enabled) |
f6630114 MT |
5000 | return; |
5001 | ||
4dcf6aff IM |
5002 | if (!sd) { |
5003 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5004 | return; | |
5005 | } | |
1da177e4 | 5006 | |
4dcf6aff IM |
5007 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5008 | ||
5009 | for (;;) { | |
4cb98839 | 5010 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5011 | break; |
1da177e4 LT |
5012 | level++; |
5013 | sd = sd->parent; | |
33859f7f | 5014 | if (!sd) |
4dcf6aff IM |
5015 | break; |
5016 | } | |
1da177e4 | 5017 | } |
6d6bc0ad | 5018 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5019 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5020 | static inline bool sched_debug(void) |
5021 | { | |
5022 | return false; | |
5023 | } | |
6d6bc0ad | 5024 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5025 | |
1a20ff27 | 5026 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5027 | { |
758b2cdc | 5028 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5029 | return 1; |
5030 | ||
5031 | /* Following flags need at least 2 groups */ | |
5032 | if (sd->flags & (SD_LOAD_BALANCE | | |
5033 | SD_BALANCE_NEWIDLE | | |
5034 | SD_BALANCE_FORK | | |
89c4710e SS |
5035 | SD_BALANCE_EXEC | |
5036 | SD_SHARE_CPUPOWER | | |
5037 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5038 | if (sd->groups != sd->groups->next) |
5039 | return 0; | |
5040 | } | |
5041 | ||
5042 | /* Following flags don't use groups */ | |
c88d5910 | 5043 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5044 | return 0; |
5045 | ||
5046 | return 1; | |
5047 | } | |
5048 | ||
48f24c4d IM |
5049 | static int |
5050 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5051 | { |
5052 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5053 | ||
5054 | if (sd_degenerate(parent)) | |
5055 | return 1; | |
5056 | ||
758b2cdc | 5057 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5058 | return 0; |
5059 | ||
245af2c7 SS |
5060 | /* Flags needing groups don't count if only 1 group in parent */ |
5061 | if (parent->groups == parent->groups->next) { | |
5062 | pflags &= ~(SD_LOAD_BALANCE | | |
5063 | SD_BALANCE_NEWIDLE | | |
5064 | SD_BALANCE_FORK | | |
89c4710e SS |
5065 | SD_BALANCE_EXEC | |
5066 | SD_SHARE_CPUPOWER | | |
10866e62 PZ |
5067 | SD_SHARE_PKG_RESOURCES | |
5068 | SD_PREFER_SIBLING); | |
5436499e KC |
5069 | if (nr_node_ids == 1) |
5070 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5071 | } |
5072 | if (~cflags & pflags) | |
5073 | return 0; | |
5074 | ||
5075 | return 1; | |
5076 | } | |
5077 | ||
dce840a0 | 5078 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5079 | { |
dce840a0 | 5080 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5081 | |
68e74568 | 5082 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5083 | free_cpumask_var(rd->rto_mask); |
5084 | free_cpumask_var(rd->online); | |
5085 | free_cpumask_var(rd->span); | |
5086 | kfree(rd); | |
5087 | } | |
5088 | ||
57d885fe GH |
5089 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5090 | { | |
a0490fa3 | 5091 | struct root_domain *old_rd = NULL; |
57d885fe | 5092 | unsigned long flags; |
57d885fe | 5093 | |
05fa785c | 5094 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5095 | |
5096 | if (rq->rd) { | |
a0490fa3 | 5097 | old_rd = rq->rd; |
57d885fe | 5098 | |
c6c4927b | 5099 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5100 | set_rq_offline(rq); |
57d885fe | 5101 | |
c6c4927b | 5102 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5103 | |
a0490fa3 IM |
5104 | /* |
5105 | * If we dont want to free the old_rt yet then | |
5106 | * set old_rd to NULL to skip the freeing later | |
5107 | * in this function: | |
5108 | */ | |
5109 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5110 | old_rd = NULL; | |
57d885fe GH |
5111 | } |
5112 | ||
5113 | atomic_inc(&rd->refcount); | |
5114 | rq->rd = rd; | |
5115 | ||
c6c4927b | 5116 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5117 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5118 | set_rq_online(rq); |
57d885fe | 5119 | |
05fa785c | 5120 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5121 | |
5122 | if (old_rd) | |
dce840a0 | 5123 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5124 | } |
5125 | ||
68c38fc3 | 5126 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5127 | { |
5128 | memset(rd, 0, sizeof(*rd)); | |
5129 | ||
68c38fc3 | 5130 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5131 | goto out; |
68c38fc3 | 5132 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5133 | goto free_span; |
68c38fc3 | 5134 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5135 | goto free_online; |
6e0534f2 | 5136 | |
68c38fc3 | 5137 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5138 | goto free_rto_mask; |
c6c4927b | 5139 | return 0; |
6e0534f2 | 5140 | |
68e74568 RR |
5141 | free_rto_mask: |
5142 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5143 | free_online: |
5144 | free_cpumask_var(rd->online); | |
5145 | free_span: | |
5146 | free_cpumask_var(rd->span); | |
0c910d28 | 5147 | out: |
c6c4927b | 5148 | return -ENOMEM; |
57d885fe GH |
5149 | } |
5150 | ||
029632fb PZ |
5151 | /* |
5152 | * By default the system creates a single root-domain with all cpus as | |
5153 | * members (mimicking the global state we have today). | |
5154 | */ | |
5155 | struct root_domain def_root_domain; | |
5156 | ||
57d885fe GH |
5157 | static void init_defrootdomain(void) |
5158 | { | |
68c38fc3 | 5159 | init_rootdomain(&def_root_domain); |
c6c4927b | 5160 | |
57d885fe GH |
5161 | atomic_set(&def_root_domain.refcount, 1); |
5162 | } | |
5163 | ||
dc938520 | 5164 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5165 | { |
5166 | struct root_domain *rd; | |
5167 | ||
5168 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5169 | if (!rd) | |
5170 | return NULL; | |
5171 | ||
68c38fc3 | 5172 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5173 | kfree(rd); |
5174 | return NULL; | |
5175 | } | |
57d885fe GH |
5176 | |
5177 | return rd; | |
5178 | } | |
5179 | ||
e3589f6c PZ |
5180 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5181 | { | |
5182 | struct sched_group *tmp, *first; | |
5183 | ||
5184 | if (!sg) | |
5185 | return; | |
5186 | ||
5187 | first = sg; | |
5188 | do { | |
5189 | tmp = sg->next; | |
5190 | ||
5191 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5192 | kfree(sg->sgp); | |
5193 | ||
5194 | kfree(sg); | |
5195 | sg = tmp; | |
5196 | } while (sg != first); | |
5197 | } | |
5198 | ||
dce840a0 PZ |
5199 | static void free_sched_domain(struct rcu_head *rcu) |
5200 | { | |
5201 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5202 | |
5203 | /* | |
5204 | * If its an overlapping domain it has private groups, iterate and | |
5205 | * nuke them all. | |
5206 | */ | |
5207 | if (sd->flags & SD_OVERLAP) { | |
5208 | free_sched_groups(sd->groups, 1); | |
5209 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5210 | kfree(sd->groups->sgp); |
dce840a0 | 5211 | kfree(sd->groups); |
9c3f75cb | 5212 | } |
dce840a0 PZ |
5213 | kfree(sd); |
5214 | } | |
5215 | ||
5216 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5217 | { | |
5218 | call_rcu(&sd->rcu, free_sched_domain); | |
5219 | } | |
5220 | ||
5221 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5222 | { | |
5223 | for (; sd; sd = sd->parent) | |
5224 | destroy_sched_domain(sd, cpu); | |
5225 | } | |
5226 | ||
518cd623 PZ |
5227 | /* |
5228 | * Keep a special pointer to the highest sched_domain that has | |
5229 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5230 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5231 | * | |
5232 | * Also keep a unique ID per domain (we use the first cpu number in | |
5233 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5234 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5235 | */ |
5236 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5237 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5238 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5239 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
518cd623 PZ |
5240 | |
5241 | static void update_top_cache_domain(int cpu) | |
5242 | { | |
5243 | struct sched_domain *sd; | |
5244 | int id = cpu; | |
7d9ffa89 | 5245 | int size = 1; |
518cd623 PZ |
5246 | |
5247 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5248 | if (sd) { |
518cd623 | 5249 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 PZ |
5250 | size = cpumask_weight(sched_domain_span(sd)); |
5251 | } | |
518cd623 PZ |
5252 | |
5253 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5254 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5255 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5256 | |
5257 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5258 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
518cd623 PZ |
5259 | } |
5260 | ||
1da177e4 | 5261 | /* |
0eab9146 | 5262 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5263 | * hold the hotplug lock. |
5264 | */ | |
0eab9146 IM |
5265 | static void |
5266 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5267 | { |
70b97a7f | 5268 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5269 | struct sched_domain *tmp; |
5270 | ||
5271 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5272 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5273 | struct sched_domain *parent = tmp->parent; |
5274 | if (!parent) | |
5275 | break; | |
f29c9b1c | 5276 | |
1a848870 | 5277 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5278 | tmp->parent = parent->parent; |
1a848870 SS |
5279 | if (parent->parent) |
5280 | parent->parent->child = tmp; | |
10866e62 PZ |
5281 | /* |
5282 | * Transfer SD_PREFER_SIBLING down in case of a | |
5283 | * degenerate parent; the spans match for this | |
5284 | * so the property transfers. | |
5285 | */ | |
5286 | if (parent->flags & SD_PREFER_SIBLING) | |
5287 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5288 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5289 | } else |
5290 | tmp = tmp->parent; | |
245af2c7 SS |
5291 | } |
5292 | ||
1a848870 | 5293 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5294 | tmp = sd; |
245af2c7 | 5295 | sd = sd->parent; |
dce840a0 | 5296 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5297 | if (sd) |
5298 | sd->child = NULL; | |
5299 | } | |
1da177e4 | 5300 | |
4cb98839 | 5301 | sched_domain_debug(sd, cpu); |
1da177e4 | 5302 | |
57d885fe | 5303 | rq_attach_root(rq, rd); |
dce840a0 | 5304 | tmp = rq->sd; |
674311d5 | 5305 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5306 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5307 | |
5308 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5309 | } |
5310 | ||
5311 | /* cpus with isolated domains */ | |
dcc30a35 | 5312 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5313 | |
5314 | /* Setup the mask of cpus configured for isolated domains */ | |
5315 | static int __init isolated_cpu_setup(char *str) | |
5316 | { | |
bdddd296 | 5317 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5318 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5319 | return 1; |
5320 | } | |
5321 | ||
8927f494 | 5322 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5323 | |
d3081f52 PZ |
5324 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5325 | { | |
5326 | return cpumask_of_node(cpu_to_node(cpu)); | |
5327 | } | |
5328 | ||
dce840a0 PZ |
5329 | struct sd_data { |
5330 | struct sched_domain **__percpu sd; | |
5331 | struct sched_group **__percpu sg; | |
9c3f75cb | 5332 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5333 | }; |
5334 | ||
49a02c51 | 5335 | struct s_data { |
21d42ccf | 5336 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5337 | struct root_domain *rd; |
5338 | }; | |
5339 | ||
2109b99e | 5340 | enum s_alloc { |
2109b99e | 5341 | sa_rootdomain, |
21d42ccf | 5342 | sa_sd, |
dce840a0 | 5343 | sa_sd_storage, |
2109b99e AH |
5344 | sa_none, |
5345 | }; | |
5346 | ||
54ab4ff4 PZ |
5347 | struct sched_domain_topology_level; |
5348 | ||
5349 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5350 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5351 | ||
e3589f6c PZ |
5352 | #define SDTL_OVERLAP 0x01 |
5353 | ||
eb7a74e6 | 5354 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5355 | sched_domain_init_f init; |
5356 | sched_domain_mask_f mask; | |
e3589f6c | 5357 | int flags; |
cb83b629 | 5358 | int numa_level; |
54ab4ff4 | 5359 | struct sd_data data; |
eb7a74e6 PZ |
5360 | }; |
5361 | ||
c1174876 PZ |
5362 | /* |
5363 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5364 | * domain traversal. | |
5365 | * | |
5366 | * Asymmetric node setups can result in situations where the domain tree is of | |
5367 | * unequal depth, make sure to skip domains that already cover the entire | |
5368 | * range. | |
5369 | * | |
5370 | * In that case build_sched_domains() will have terminated the iteration early | |
5371 | * and our sibling sd spans will be empty. Domains should always include the | |
5372 | * cpu they're built on, so check that. | |
5373 | * | |
5374 | */ | |
5375 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5376 | { | |
5377 | const struct cpumask *span = sched_domain_span(sd); | |
5378 | struct sd_data *sdd = sd->private; | |
5379 | struct sched_domain *sibling; | |
5380 | int i; | |
5381 | ||
5382 | for_each_cpu(i, span) { | |
5383 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5384 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5385 | continue; | |
5386 | ||
5387 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5388 | } | |
5389 | } | |
5390 | ||
5391 | /* | |
5392 | * Return the canonical balance cpu for this group, this is the first cpu | |
5393 | * of this group that's also in the iteration mask. | |
5394 | */ | |
5395 | int group_balance_cpu(struct sched_group *sg) | |
5396 | { | |
5397 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5398 | } | |
5399 | ||
e3589f6c PZ |
5400 | static int |
5401 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5402 | { | |
5403 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5404 | const struct cpumask *span = sched_domain_span(sd); | |
5405 | struct cpumask *covered = sched_domains_tmpmask; | |
5406 | struct sd_data *sdd = sd->private; | |
5407 | struct sched_domain *child; | |
5408 | int i; | |
5409 | ||
5410 | cpumask_clear(covered); | |
5411 | ||
5412 | for_each_cpu(i, span) { | |
5413 | struct cpumask *sg_span; | |
5414 | ||
5415 | if (cpumask_test_cpu(i, covered)) | |
5416 | continue; | |
5417 | ||
c1174876 PZ |
5418 | child = *per_cpu_ptr(sdd->sd, i); |
5419 | ||
5420 | /* See the comment near build_group_mask(). */ | |
5421 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5422 | continue; | |
5423 | ||
e3589f6c | 5424 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5425 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5426 | |
5427 | if (!sg) | |
5428 | goto fail; | |
5429 | ||
5430 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5431 | if (child->child) { |
5432 | child = child->child; | |
5433 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5434 | } else | |
5435 | cpumask_set_cpu(i, sg_span); | |
5436 | ||
5437 | cpumask_or(covered, covered, sg_span); | |
5438 | ||
74a5ce20 | 5439 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5440 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5441 | build_group_mask(sd, sg); | |
5442 | ||
c3decf0d PZ |
5443 | /* |
5444 | * Initialize sgp->power such that even if we mess up the | |
5445 | * domains and no possible iteration will get us here, we won't | |
5446 | * die on a /0 trap. | |
5447 | */ | |
5448 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 5449 | |
c1174876 PZ |
5450 | /* |
5451 | * Make sure the first group of this domain contains the | |
5452 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5453 | * breaks. See update_sg_lb_stats(). | |
5454 | */ | |
74a5ce20 | 5455 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5456 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5457 | groups = sg; |
5458 | ||
5459 | if (!first) | |
5460 | first = sg; | |
5461 | if (last) | |
5462 | last->next = sg; | |
5463 | last = sg; | |
5464 | last->next = first; | |
5465 | } | |
5466 | sd->groups = groups; | |
5467 | ||
5468 | return 0; | |
5469 | ||
5470 | fail: | |
5471 | free_sched_groups(first, 0); | |
5472 | ||
5473 | return -ENOMEM; | |
5474 | } | |
5475 | ||
dce840a0 | 5476 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5477 | { |
dce840a0 PZ |
5478 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5479 | struct sched_domain *child = sd->child; | |
1da177e4 | 5480 | |
dce840a0 PZ |
5481 | if (child) |
5482 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5483 | |
9c3f75cb | 5484 | if (sg) { |
dce840a0 | 5485 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5486 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5487 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5488 | } |
dce840a0 PZ |
5489 | |
5490 | return cpu; | |
1e9f28fa | 5491 | } |
1e9f28fa | 5492 | |
01a08546 | 5493 | /* |
dce840a0 PZ |
5494 | * build_sched_groups will build a circular linked list of the groups |
5495 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5496 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5497 | * |
5498 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5499 | */ |
e3589f6c PZ |
5500 | static int |
5501 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5502 | { |
dce840a0 PZ |
5503 | struct sched_group *first = NULL, *last = NULL; |
5504 | struct sd_data *sdd = sd->private; | |
5505 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5506 | struct cpumask *covered; |
dce840a0 | 5507 | int i; |
9c1cfda2 | 5508 | |
e3589f6c PZ |
5509 | get_group(cpu, sdd, &sd->groups); |
5510 | atomic_inc(&sd->groups->ref); | |
5511 | ||
0936629f | 5512 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5513 | return 0; |
5514 | ||
f96225fd PZ |
5515 | lockdep_assert_held(&sched_domains_mutex); |
5516 | covered = sched_domains_tmpmask; | |
5517 | ||
dce840a0 | 5518 | cpumask_clear(covered); |
6711cab4 | 5519 | |
dce840a0 PZ |
5520 | for_each_cpu(i, span) { |
5521 | struct sched_group *sg; | |
cd08e923 | 5522 | int group, j; |
6711cab4 | 5523 | |
dce840a0 PZ |
5524 | if (cpumask_test_cpu(i, covered)) |
5525 | continue; | |
6711cab4 | 5526 | |
cd08e923 | 5527 | group = get_group(i, sdd, &sg); |
dce840a0 | 5528 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5529 | sg->sgp->power = 0; |
c1174876 | 5530 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5531 | |
dce840a0 PZ |
5532 | for_each_cpu(j, span) { |
5533 | if (get_group(j, sdd, NULL) != group) | |
5534 | continue; | |
0601a88d | 5535 | |
dce840a0 PZ |
5536 | cpumask_set_cpu(j, covered); |
5537 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5538 | } | |
0601a88d | 5539 | |
dce840a0 PZ |
5540 | if (!first) |
5541 | first = sg; | |
5542 | if (last) | |
5543 | last->next = sg; | |
5544 | last = sg; | |
5545 | } | |
5546 | last->next = first; | |
e3589f6c PZ |
5547 | |
5548 | return 0; | |
0601a88d | 5549 | } |
51888ca2 | 5550 | |
89c4710e SS |
5551 | /* |
5552 | * Initialize sched groups cpu_power. | |
5553 | * | |
5554 | * cpu_power indicates the capacity of sched group, which is used while | |
5555 | * distributing the load between different sched groups in a sched domain. | |
5556 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5557 | * there are asymmetries in the topology. If there are asymmetries, group | |
5558 | * having more cpu_power will pickup more load compared to the group having | |
5559 | * less cpu_power. | |
89c4710e SS |
5560 | */ |
5561 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5562 | { | |
e3589f6c | 5563 | struct sched_group *sg = sd->groups; |
89c4710e | 5564 | |
94c95ba6 | 5565 | WARN_ON(!sg); |
e3589f6c PZ |
5566 | |
5567 | do { | |
5568 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5569 | sg = sg->next; | |
5570 | } while (sg != sd->groups); | |
89c4710e | 5571 | |
c1174876 | 5572 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5573 | return; |
aae6d3dd | 5574 | |
d274cb30 | 5575 | update_group_power(sd, cpu); |
69e1e811 | 5576 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5577 | } |
5578 | ||
029632fb PZ |
5579 | int __weak arch_sd_sibling_asym_packing(void) |
5580 | { | |
5581 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5582 | } |
5583 | ||
7c16ec58 MT |
5584 | /* |
5585 | * Initializers for schedule domains | |
5586 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5587 | */ | |
5588 | ||
a5d8c348 IM |
5589 | #ifdef CONFIG_SCHED_DEBUG |
5590 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5591 | #else | |
5592 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5593 | #endif | |
5594 | ||
54ab4ff4 PZ |
5595 | #define SD_INIT_FUNC(type) \ |
5596 | static noinline struct sched_domain * \ | |
5597 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5598 | { \ | |
5599 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5600 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5601 | SD_INIT_NAME(sd, type); \ |
5602 | sd->private = &tl->data; \ | |
5603 | return sd; \ | |
7c16ec58 MT |
5604 | } |
5605 | ||
5606 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5607 | #ifdef CONFIG_SCHED_SMT |
5608 | SD_INIT_FUNC(SIBLING) | |
5609 | #endif | |
5610 | #ifdef CONFIG_SCHED_MC | |
5611 | SD_INIT_FUNC(MC) | |
5612 | #endif | |
01a08546 HC |
5613 | #ifdef CONFIG_SCHED_BOOK |
5614 | SD_INIT_FUNC(BOOK) | |
5615 | #endif | |
7c16ec58 | 5616 | |
1d3504fc | 5617 | static int default_relax_domain_level = -1; |
60495e77 | 5618 | int sched_domain_level_max; |
1d3504fc HS |
5619 | |
5620 | static int __init setup_relax_domain_level(char *str) | |
5621 | { | |
a841f8ce DS |
5622 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5623 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5624 | |
1d3504fc HS |
5625 | return 1; |
5626 | } | |
5627 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5628 | ||
5629 | static void set_domain_attribute(struct sched_domain *sd, | |
5630 | struct sched_domain_attr *attr) | |
5631 | { | |
5632 | int request; | |
5633 | ||
5634 | if (!attr || attr->relax_domain_level < 0) { | |
5635 | if (default_relax_domain_level < 0) | |
5636 | return; | |
5637 | else | |
5638 | request = default_relax_domain_level; | |
5639 | } else | |
5640 | request = attr->relax_domain_level; | |
5641 | if (request < sd->level) { | |
5642 | /* turn off idle balance on this domain */ | |
c88d5910 | 5643 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5644 | } else { |
5645 | /* turn on idle balance on this domain */ | |
c88d5910 | 5646 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5647 | } |
5648 | } | |
5649 | ||
54ab4ff4 PZ |
5650 | static void __sdt_free(const struct cpumask *cpu_map); |
5651 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5652 | ||
2109b99e AH |
5653 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5654 | const struct cpumask *cpu_map) | |
5655 | { | |
5656 | switch (what) { | |
2109b99e | 5657 | case sa_rootdomain: |
822ff793 PZ |
5658 | if (!atomic_read(&d->rd->refcount)) |
5659 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5660 | case sa_sd: |
5661 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5662 | case sa_sd_storage: |
54ab4ff4 | 5663 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5664 | case sa_none: |
5665 | break; | |
5666 | } | |
5667 | } | |
3404c8d9 | 5668 | |
2109b99e AH |
5669 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5670 | const struct cpumask *cpu_map) | |
5671 | { | |
dce840a0 PZ |
5672 | memset(d, 0, sizeof(*d)); |
5673 | ||
54ab4ff4 PZ |
5674 | if (__sdt_alloc(cpu_map)) |
5675 | return sa_sd_storage; | |
dce840a0 PZ |
5676 | d->sd = alloc_percpu(struct sched_domain *); |
5677 | if (!d->sd) | |
5678 | return sa_sd_storage; | |
2109b99e | 5679 | d->rd = alloc_rootdomain(); |
dce840a0 | 5680 | if (!d->rd) |
21d42ccf | 5681 | return sa_sd; |
2109b99e AH |
5682 | return sa_rootdomain; |
5683 | } | |
57d885fe | 5684 | |
dce840a0 PZ |
5685 | /* |
5686 | * NULL the sd_data elements we've used to build the sched_domain and | |
5687 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5688 | * will not free the data we're using. | |
5689 | */ | |
5690 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5691 | { | |
5692 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5693 | |
5694 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5695 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5696 | ||
e3589f6c | 5697 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5698 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5699 | |
5700 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5701 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5702 | } |
5703 | ||
2c402dc3 PZ |
5704 | #ifdef CONFIG_SCHED_SMT |
5705 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 5706 | { |
2c402dc3 | 5707 | return topology_thread_cpumask(cpu); |
3bd65a80 | 5708 | } |
2c402dc3 | 5709 | #endif |
7f4588f3 | 5710 | |
d069b916 PZ |
5711 | /* |
5712 | * Topology list, bottom-up. | |
5713 | */ | |
2c402dc3 | 5714 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
5715 | #ifdef CONFIG_SCHED_SMT |
5716 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 5717 | #endif |
1e9f28fa | 5718 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 5719 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 5720 | #endif |
d069b916 PZ |
5721 | #ifdef CONFIG_SCHED_BOOK |
5722 | { sd_init_BOOK, cpu_book_mask, }, | |
5723 | #endif | |
5724 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
5725 | { NULL, }, |
5726 | }; | |
5727 | ||
5728 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
5729 | ||
27723a68 VK |
5730 | #define for_each_sd_topology(tl) \ |
5731 | for (tl = sched_domain_topology; tl->init; tl++) | |
5732 | ||
cb83b629 PZ |
5733 | #ifdef CONFIG_NUMA |
5734 | ||
5735 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
5736 | static int *sched_domains_numa_distance; |
5737 | static struct cpumask ***sched_domains_numa_masks; | |
5738 | static int sched_domains_curr_level; | |
5739 | ||
cb83b629 PZ |
5740 | static inline int sd_local_flags(int level) |
5741 | { | |
10717dcd | 5742 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
5743 | return 0; |
5744 | ||
5745 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
5746 | } | |
5747 | ||
5748 | static struct sched_domain * | |
5749 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
5750 | { | |
5751 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
5752 | int level = tl->numa_level; | |
5753 | int sd_weight = cpumask_weight( | |
5754 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
5755 | ||
5756 | *sd = (struct sched_domain){ | |
5757 | .min_interval = sd_weight, | |
5758 | .max_interval = 2*sd_weight, | |
5759 | .busy_factor = 32, | |
870a0bb5 | 5760 | .imbalance_pct = 125, |
cb83b629 PZ |
5761 | .cache_nice_tries = 2, |
5762 | .busy_idx = 3, | |
5763 | .idle_idx = 2, | |
5764 | .newidle_idx = 0, | |
5765 | .wake_idx = 0, | |
5766 | .forkexec_idx = 0, | |
5767 | ||
5768 | .flags = 1*SD_LOAD_BALANCE | |
5769 | | 1*SD_BALANCE_NEWIDLE | |
5770 | | 0*SD_BALANCE_EXEC | |
5771 | | 0*SD_BALANCE_FORK | |
5772 | | 0*SD_BALANCE_WAKE | |
5773 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 5774 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
5775 | | 0*SD_SHARE_PKG_RESOURCES |
5776 | | 1*SD_SERIALIZE | |
5777 | | 0*SD_PREFER_SIBLING | |
3a7053b3 | 5778 | | 1*SD_NUMA |
cb83b629 PZ |
5779 | | sd_local_flags(level) |
5780 | , | |
5781 | .last_balance = jiffies, | |
5782 | .balance_interval = sd_weight, | |
5783 | }; | |
5784 | SD_INIT_NAME(sd, NUMA); | |
5785 | sd->private = &tl->data; | |
5786 | ||
5787 | /* | |
5788 | * Ugly hack to pass state to sd_numa_mask()... | |
5789 | */ | |
5790 | sched_domains_curr_level = tl->numa_level; | |
5791 | ||
5792 | return sd; | |
5793 | } | |
5794 | ||
5795 | static const struct cpumask *sd_numa_mask(int cpu) | |
5796 | { | |
5797 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
5798 | } | |
5799 | ||
d039ac60 PZ |
5800 | static void sched_numa_warn(const char *str) |
5801 | { | |
5802 | static int done = false; | |
5803 | int i,j; | |
5804 | ||
5805 | if (done) | |
5806 | return; | |
5807 | ||
5808 | done = true; | |
5809 | ||
5810 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
5811 | ||
5812 | for (i = 0; i < nr_node_ids; i++) { | |
5813 | printk(KERN_WARNING " "); | |
5814 | for (j = 0; j < nr_node_ids; j++) | |
5815 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
5816 | printk(KERN_CONT "\n"); | |
5817 | } | |
5818 | printk(KERN_WARNING "\n"); | |
5819 | } | |
5820 | ||
5821 | static bool find_numa_distance(int distance) | |
5822 | { | |
5823 | int i; | |
5824 | ||
5825 | if (distance == node_distance(0, 0)) | |
5826 | return true; | |
5827 | ||
5828 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5829 | if (sched_domains_numa_distance[i] == distance) | |
5830 | return true; | |
5831 | } | |
5832 | ||
5833 | return false; | |
5834 | } | |
5835 | ||
cb83b629 PZ |
5836 | static void sched_init_numa(void) |
5837 | { | |
5838 | int next_distance, curr_distance = node_distance(0, 0); | |
5839 | struct sched_domain_topology_level *tl; | |
5840 | int level = 0; | |
5841 | int i, j, k; | |
5842 | ||
cb83b629 PZ |
5843 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
5844 | if (!sched_domains_numa_distance) | |
5845 | return; | |
5846 | ||
5847 | /* | |
5848 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
5849 | * unique distances in the node_distance() table. | |
5850 | * | |
5851 | * Assumes node_distance(0,j) includes all distances in | |
5852 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
5853 | */ |
5854 | next_distance = curr_distance; | |
5855 | for (i = 0; i < nr_node_ids; i++) { | |
5856 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
5857 | for (k = 0; k < nr_node_ids; k++) { |
5858 | int distance = node_distance(i, k); | |
5859 | ||
5860 | if (distance > curr_distance && | |
5861 | (distance < next_distance || | |
5862 | next_distance == curr_distance)) | |
5863 | next_distance = distance; | |
5864 | ||
5865 | /* | |
5866 | * While not a strong assumption it would be nice to know | |
5867 | * about cases where if node A is connected to B, B is not | |
5868 | * equally connected to A. | |
5869 | */ | |
5870 | if (sched_debug() && node_distance(k, i) != distance) | |
5871 | sched_numa_warn("Node-distance not symmetric"); | |
5872 | ||
5873 | if (sched_debug() && i && !find_numa_distance(distance)) | |
5874 | sched_numa_warn("Node-0 not representative"); | |
5875 | } | |
5876 | if (next_distance != curr_distance) { | |
5877 | sched_domains_numa_distance[level++] = next_distance; | |
5878 | sched_domains_numa_levels = level; | |
5879 | curr_distance = next_distance; | |
5880 | } else break; | |
cb83b629 | 5881 | } |
d039ac60 PZ |
5882 | |
5883 | /* | |
5884 | * In case of sched_debug() we verify the above assumption. | |
5885 | */ | |
5886 | if (!sched_debug()) | |
5887 | break; | |
cb83b629 PZ |
5888 | } |
5889 | /* | |
5890 | * 'level' contains the number of unique distances, excluding the | |
5891 | * identity distance node_distance(i,i). | |
5892 | * | |
28b4a521 | 5893 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
5894 | * numbers. |
5895 | */ | |
5896 | ||
5f7865f3 TC |
5897 | /* |
5898 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
5899 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
5900 | * the array will contain less then 'level' members. This could be | |
5901 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
5902 | * in other functions. | |
5903 | * | |
5904 | * We reset it to 'level' at the end of this function. | |
5905 | */ | |
5906 | sched_domains_numa_levels = 0; | |
5907 | ||
cb83b629 PZ |
5908 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
5909 | if (!sched_domains_numa_masks) | |
5910 | return; | |
5911 | ||
5912 | /* | |
5913 | * Now for each level, construct a mask per node which contains all | |
5914 | * cpus of nodes that are that many hops away from us. | |
5915 | */ | |
5916 | for (i = 0; i < level; i++) { | |
5917 | sched_domains_numa_masks[i] = | |
5918 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
5919 | if (!sched_domains_numa_masks[i]) | |
5920 | return; | |
5921 | ||
5922 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 5923 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
5924 | if (!mask) |
5925 | return; | |
5926 | ||
5927 | sched_domains_numa_masks[i][j] = mask; | |
5928 | ||
5929 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 5930 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
5931 | continue; |
5932 | ||
5933 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
5934 | } | |
5935 | } | |
5936 | } | |
5937 | ||
5938 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
5939 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
5940 | if (!tl) | |
5941 | return; | |
5942 | ||
5943 | /* | |
5944 | * Copy the default topology bits.. | |
5945 | */ | |
5946 | for (i = 0; default_topology[i].init; i++) | |
5947 | tl[i] = default_topology[i]; | |
5948 | ||
5949 | /* | |
5950 | * .. and append 'j' levels of NUMA goodness. | |
5951 | */ | |
5952 | for (j = 0; j < level; i++, j++) { | |
5953 | tl[i] = (struct sched_domain_topology_level){ | |
5954 | .init = sd_numa_init, | |
5955 | .mask = sd_numa_mask, | |
5956 | .flags = SDTL_OVERLAP, | |
5957 | .numa_level = j, | |
5958 | }; | |
5959 | } | |
5960 | ||
5961 | sched_domain_topology = tl; | |
5f7865f3 TC |
5962 | |
5963 | sched_domains_numa_levels = level; | |
cb83b629 | 5964 | } |
301a5cba TC |
5965 | |
5966 | static void sched_domains_numa_masks_set(int cpu) | |
5967 | { | |
5968 | int i, j; | |
5969 | int node = cpu_to_node(cpu); | |
5970 | ||
5971 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5972 | for (j = 0; j < nr_node_ids; j++) { | |
5973 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
5974 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
5975 | } | |
5976 | } | |
5977 | } | |
5978 | ||
5979 | static void sched_domains_numa_masks_clear(int cpu) | |
5980 | { | |
5981 | int i, j; | |
5982 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5983 | for (j = 0; j < nr_node_ids; j++) | |
5984 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
5985 | } | |
5986 | } | |
5987 | ||
5988 | /* | |
5989 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
5990 | * are onlined. | |
5991 | */ | |
5992 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
5993 | unsigned long action, | |
5994 | void *hcpu) | |
5995 | { | |
5996 | int cpu = (long)hcpu; | |
5997 | ||
5998 | switch (action & ~CPU_TASKS_FROZEN) { | |
5999 | case CPU_ONLINE: | |
6000 | sched_domains_numa_masks_set(cpu); | |
6001 | break; | |
6002 | ||
6003 | case CPU_DEAD: | |
6004 | sched_domains_numa_masks_clear(cpu); | |
6005 | break; | |
6006 | ||
6007 | default: | |
6008 | return NOTIFY_DONE; | |
6009 | } | |
6010 | ||
6011 | return NOTIFY_OK; | |
cb83b629 PZ |
6012 | } |
6013 | #else | |
6014 | static inline void sched_init_numa(void) | |
6015 | { | |
6016 | } | |
301a5cba TC |
6017 | |
6018 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6019 | unsigned long action, | |
6020 | void *hcpu) | |
6021 | { | |
6022 | return 0; | |
6023 | } | |
cb83b629 PZ |
6024 | #endif /* CONFIG_NUMA */ |
6025 | ||
54ab4ff4 PZ |
6026 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6027 | { | |
6028 | struct sched_domain_topology_level *tl; | |
6029 | int j; | |
6030 | ||
27723a68 | 6031 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6032 | struct sd_data *sdd = &tl->data; |
6033 | ||
6034 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6035 | if (!sdd->sd) | |
6036 | return -ENOMEM; | |
6037 | ||
6038 | sdd->sg = alloc_percpu(struct sched_group *); | |
6039 | if (!sdd->sg) | |
6040 | return -ENOMEM; | |
6041 | ||
9c3f75cb PZ |
6042 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6043 | if (!sdd->sgp) | |
6044 | return -ENOMEM; | |
6045 | ||
54ab4ff4 PZ |
6046 | for_each_cpu(j, cpu_map) { |
6047 | struct sched_domain *sd; | |
6048 | struct sched_group *sg; | |
9c3f75cb | 6049 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6050 | |
6051 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6052 | GFP_KERNEL, cpu_to_node(j)); | |
6053 | if (!sd) | |
6054 | return -ENOMEM; | |
6055 | ||
6056 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6057 | ||
6058 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6059 | GFP_KERNEL, cpu_to_node(j)); | |
6060 | if (!sg) | |
6061 | return -ENOMEM; | |
6062 | ||
30b4e9eb IM |
6063 | sg->next = sg; |
6064 | ||
54ab4ff4 | 6065 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6066 | |
c1174876 | 6067 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6068 | GFP_KERNEL, cpu_to_node(j)); |
6069 | if (!sgp) | |
6070 | return -ENOMEM; | |
6071 | ||
6072 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6073 | } |
6074 | } | |
6075 | ||
6076 | return 0; | |
6077 | } | |
6078 | ||
6079 | static void __sdt_free(const struct cpumask *cpu_map) | |
6080 | { | |
6081 | struct sched_domain_topology_level *tl; | |
6082 | int j; | |
6083 | ||
27723a68 | 6084 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6085 | struct sd_data *sdd = &tl->data; |
6086 | ||
6087 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6088 | struct sched_domain *sd; |
6089 | ||
6090 | if (sdd->sd) { | |
6091 | sd = *per_cpu_ptr(sdd->sd, j); | |
6092 | if (sd && (sd->flags & SD_OVERLAP)) | |
6093 | free_sched_groups(sd->groups, 0); | |
6094 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6095 | } | |
6096 | ||
6097 | if (sdd->sg) | |
6098 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6099 | if (sdd->sgp) | |
6100 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6101 | } |
6102 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6103 | sdd->sd = NULL; |
54ab4ff4 | 6104 | free_percpu(sdd->sg); |
fb2cf2c6 | 6105 | sdd->sg = NULL; |
9c3f75cb | 6106 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6107 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6108 | } |
6109 | } | |
6110 | ||
2c402dc3 | 6111 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6112 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6113 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6114 | { |
54ab4ff4 | 6115 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6116 | if (!sd) |
d069b916 | 6117 | return child; |
2c402dc3 | 6118 | |
2c402dc3 | 6119 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6120 | if (child) { |
6121 | sd->level = child->level + 1; | |
6122 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6123 | child->parent = sd; |
c75e0128 | 6124 | sd->child = child; |
60495e77 | 6125 | } |
a841f8ce | 6126 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6127 | |
6128 | return sd; | |
6129 | } | |
6130 | ||
2109b99e AH |
6131 | /* |
6132 | * Build sched domains for a given set of cpus and attach the sched domains | |
6133 | * to the individual cpus | |
6134 | */ | |
dce840a0 PZ |
6135 | static int build_sched_domains(const struct cpumask *cpu_map, |
6136 | struct sched_domain_attr *attr) | |
2109b99e | 6137 | { |
1c632169 | 6138 | enum s_alloc alloc_state; |
dce840a0 | 6139 | struct sched_domain *sd; |
2109b99e | 6140 | struct s_data d; |
822ff793 | 6141 | int i, ret = -ENOMEM; |
9c1cfda2 | 6142 | |
2109b99e AH |
6143 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6144 | if (alloc_state != sa_rootdomain) | |
6145 | goto error; | |
9c1cfda2 | 6146 | |
dce840a0 | 6147 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6148 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6149 | struct sched_domain_topology_level *tl; |
6150 | ||
3bd65a80 | 6151 | sd = NULL; |
27723a68 | 6152 | for_each_sd_topology(tl) { |
4a850cbe | 6153 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6154 | if (tl == sched_domain_topology) |
6155 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6156 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6157 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6158 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6159 | break; | |
e3589f6c | 6160 | } |
dce840a0 PZ |
6161 | } |
6162 | ||
6163 | /* Build the groups for the domains */ | |
6164 | for_each_cpu(i, cpu_map) { | |
6165 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6166 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6167 | if (sd->flags & SD_OVERLAP) { |
6168 | if (build_overlap_sched_groups(sd, i)) | |
6169 | goto error; | |
6170 | } else { | |
6171 | if (build_sched_groups(sd, i)) | |
6172 | goto error; | |
6173 | } | |
1cf51902 | 6174 | } |
a06dadbe | 6175 | } |
9c1cfda2 | 6176 | |
1da177e4 | 6177 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6178 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6179 | if (!cpumask_test_cpu(i, cpu_map)) | |
6180 | continue; | |
9c1cfda2 | 6181 | |
dce840a0 PZ |
6182 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6183 | claim_allocations(i, sd); | |
cd4ea6ae | 6184 | init_sched_groups_power(i, sd); |
dce840a0 | 6185 | } |
f712c0c7 | 6186 | } |
9c1cfda2 | 6187 | |
1da177e4 | 6188 | /* Attach the domains */ |
dce840a0 | 6189 | rcu_read_lock(); |
abcd083a | 6190 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6191 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6192 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6193 | } |
dce840a0 | 6194 | rcu_read_unlock(); |
51888ca2 | 6195 | |
822ff793 | 6196 | ret = 0; |
51888ca2 | 6197 | error: |
2109b99e | 6198 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6199 | return ret; |
1da177e4 | 6200 | } |
029190c5 | 6201 | |
acc3f5d7 | 6202 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6203 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6204 | static struct sched_domain_attr *dattr_cur; |
6205 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6206 | |
6207 | /* | |
6208 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6209 | * cpumask) fails, then fallback to a single sched domain, |
6210 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6211 | */ |
4212823f | 6212 | static cpumask_var_t fallback_doms; |
029190c5 | 6213 | |
ee79d1bd HC |
6214 | /* |
6215 | * arch_update_cpu_topology lets virtualized architectures update the | |
6216 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6217 | * or 0 if it stayed the same. | |
6218 | */ | |
6219 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6220 | { |
ee79d1bd | 6221 | return 0; |
22e52b07 HC |
6222 | } |
6223 | ||
acc3f5d7 RR |
6224 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6225 | { | |
6226 | int i; | |
6227 | cpumask_var_t *doms; | |
6228 | ||
6229 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6230 | if (!doms) | |
6231 | return NULL; | |
6232 | for (i = 0; i < ndoms; i++) { | |
6233 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6234 | free_sched_domains(doms, i); | |
6235 | return NULL; | |
6236 | } | |
6237 | } | |
6238 | return doms; | |
6239 | } | |
6240 | ||
6241 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6242 | { | |
6243 | unsigned int i; | |
6244 | for (i = 0; i < ndoms; i++) | |
6245 | free_cpumask_var(doms[i]); | |
6246 | kfree(doms); | |
6247 | } | |
6248 | ||
1a20ff27 | 6249 | /* |
41a2d6cf | 6250 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6251 | * For now this just excludes isolated cpus, but could be used to |
6252 | * exclude other special cases in the future. | |
1a20ff27 | 6253 | */ |
c4a8849a | 6254 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6255 | { |
7378547f MM |
6256 | int err; |
6257 | ||
22e52b07 | 6258 | arch_update_cpu_topology(); |
029190c5 | 6259 | ndoms_cur = 1; |
acc3f5d7 | 6260 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6261 | if (!doms_cur) |
acc3f5d7 RR |
6262 | doms_cur = &fallback_doms; |
6263 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6264 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6265 | register_sched_domain_sysctl(); |
7378547f MM |
6266 | |
6267 | return err; | |
1a20ff27 DG |
6268 | } |
6269 | ||
1a20ff27 DG |
6270 | /* |
6271 | * Detach sched domains from a group of cpus specified in cpu_map | |
6272 | * These cpus will now be attached to the NULL domain | |
6273 | */ | |
96f874e2 | 6274 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6275 | { |
6276 | int i; | |
6277 | ||
dce840a0 | 6278 | rcu_read_lock(); |
abcd083a | 6279 | for_each_cpu(i, cpu_map) |
57d885fe | 6280 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6281 | rcu_read_unlock(); |
1a20ff27 DG |
6282 | } |
6283 | ||
1d3504fc HS |
6284 | /* handle null as "default" */ |
6285 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6286 | struct sched_domain_attr *new, int idx_new) | |
6287 | { | |
6288 | struct sched_domain_attr tmp; | |
6289 | ||
6290 | /* fast path */ | |
6291 | if (!new && !cur) | |
6292 | return 1; | |
6293 | ||
6294 | tmp = SD_ATTR_INIT; | |
6295 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6296 | new ? (new + idx_new) : &tmp, | |
6297 | sizeof(struct sched_domain_attr)); | |
6298 | } | |
6299 | ||
029190c5 PJ |
6300 | /* |
6301 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6302 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6303 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6304 | * It destroys each deleted domain and builds each new domain. | |
6305 | * | |
acc3f5d7 | 6306 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6307 | * The masks don't intersect (don't overlap.) We should setup one |
6308 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6309 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6310 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6311 | * it as it is. | |
6312 | * | |
acc3f5d7 RR |
6313 | * The passed in 'doms_new' should be allocated using |
6314 | * alloc_sched_domains. This routine takes ownership of it and will | |
6315 | * free_sched_domains it when done with it. If the caller failed the | |
6316 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6317 | * and partition_sched_domains() will fallback to the single partition | |
6318 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6319 | * |
96f874e2 | 6320 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6321 | * ndoms_new == 0 is a special case for destroying existing domains, |
6322 | * and it will not create the default domain. | |
dfb512ec | 6323 | * |
029190c5 PJ |
6324 | * Call with hotplug lock held |
6325 | */ | |
acc3f5d7 | 6326 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6327 | struct sched_domain_attr *dattr_new) |
029190c5 | 6328 | { |
dfb512ec | 6329 | int i, j, n; |
d65bd5ec | 6330 | int new_topology; |
029190c5 | 6331 | |
712555ee | 6332 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6333 | |
7378547f MM |
6334 | /* always unregister in case we don't destroy any domains */ |
6335 | unregister_sched_domain_sysctl(); | |
6336 | ||
d65bd5ec HC |
6337 | /* Let architecture update cpu core mappings. */ |
6338 | new_topology = arch_update_cpu_topology(); | |
6339 | ||
dfb512ec | 6340 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6341 | |
6342 | /* Destroy deleted domains */ | |
6343 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6344 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6345 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6346 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6347 | goto match1; |
6348 | } | |
6349 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6350 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6351 | match1: |
6352 | ; | |
6353 | } | |
6354 | ||
c8d2d47a | 6355 | n = ndoms_cur; |
e761b772 | 6356 | if (doms_new == NULL) { |
c8d2d47a | 6357 | n = 0; |
acc3f5d7 | 6358 | doms_new = &fallback_doms; |
6ad4c188 | 6359 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6360 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6361 | } |
6362 | ||
029190c5 PJ |
6363 | /* Build new domains */ |
6364 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6365 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6366 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6367 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6368 | goto match2; |
6369 | } | |
6370 | /* no match - add a new doms_new */ | |
dce840a0 | 6371 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6372 | match2: |
6373 | ; | |
6374 | } | |
6375 | ||
6376 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6377 | if (doms_cur != &fallback_doms) |
6378 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6379 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6380 | doms_cur = doms_new; |
1d3504fc | 6381 | dattr_cur = dattr_new; |
029190c5 | 6382 | ndoms_cur = ndoms_new; |
7378547f MM |
6383 | |
6384 | register_sched_domain_sysctl(); | |
a1835615 | 6385 | |
712555ee | 6386 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6387 | } |
6388 | ||
d35be8ba SB |
6389 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6390 | ||
1da177e4 | 6391 | /* |
3a101d05 TH |
6392 | * Update cpusets according to cpu_active mask. If cpusets are |
6393 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6394 | * around partition_sched_domains(). | |
d35be8ba SB |
6395 | * |
6396 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6397 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6398 | */ |
0b2e918a TH |
6399 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6400 | void *hcpu) | |
e761b772 | 6401 | { |
d35be8ba SB |
6402 | switch (action) { |
6403 | case CPU_ONLINE_FROZEN: | |
6404 | case CPU_DOWN_FAILED_FROZEN: | |
6405 | ||
6406 | /* | |
6407 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6408 | * resume sequence. As long as this is not the last online | |
6409 | * operation in the resume sequence, just build a single sched | |
6410 | * domain, ignoring cpusets. | |
6411 | */ | |
6412 | num_cpus_frozen--; | |
6413 | if (likely(num_cpus_frozen)) { | |
6414 | partition_sched_domains(1, NULL, NULL); | |
6415 | break; | |
6416 | } | |
6417 | ||
6418 | /* | |
6419 | * This is the last CPU online operation. So fall through and | |
6420 | * restore the original sched domains by considering the | |
6421 | * cpuset configurations. | |
6422 | */ | |
6423 | ||
e761b772 | 6424 | case CPU_ONLINE: |
6ad4c188 | 6425 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6426 | cpuset_update_active_cpus(true); |
d35be8ba | 6427 | break; |
3a101d05 TH |
6428 | default: |
6429 | return NOTIFY_DONE; | |
6430 | } | |
d35be8ba | 6431 | return NOTIFY_OK; |
3a101d05 | 6432 | } |
e761b772 | 6433 | |
0b2e918a TH |
6434 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6435 | void *hcpu) | |
3a101d05 | 6436 | { |
d35be8ba | 6437 | switch (action) { |
3a101d05 | 6438 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6439 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6440 | break; |
6441 | case CPU_DOWN_PREPARE_FROZEN: | |
6442 | num_cpus_frozen++; | |
6443 | partition_sched_domains(1, NULL, NULL); | |
6444 | break; | |
e761b772 MK |
6445 | default: |
6446 | return NOTIFY_DONE; | |
6447 | } | |
d35be8ba | 6448 | return NOTIFY_OK; |
e761b772 | 6449 | } |
e761b772 | 6450 | |
1da177e4 LT |
6451 | void __init sched_init_smp(void) |
6452 | { | |
dcc30a35 RR |
6453 | cpumask_var_t non_isolated_cpus; |
6454 | ||
6455 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6456 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6457 | |
cb83b629 PZ |
6458 | sched_init_numa(); |
6459 | ||
95402b38 | 6460 | get_online_cpus(); |
712555ee | 6461 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6462 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6463 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6464 | if (cpumask_empty(non_isolated_cpus)) | |
6465 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6466 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6467 | put_online_cpus(); |
e761b772 | 6468 | |
301a5cba | 6469 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6470 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6471 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6472 | |
b328ca18 | 6473 | init_hrtick(); |
5c1e1767 NP |
6474 | |
6475 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6476 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6477 | BUG(); |
19978ca6 | 6478 | sched_init_granularity(); |
dcc30a35 | 6479 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6480 | |
0e3900e6 | 6481 | init_sched_rt_class(); |
1da177e4 LT |
6482 | } |
6483 | #else | |
6484 | void __init sched_init_smp(void) | |
6485 | { | |
19978ca6 | 6486 | sched_init_granularity(); |
1da177e4 LT |
6487 | } |
6488 | #endif /* CONFIG_SMP */ | |
6489 | ||
cd1bb94b AB |
6490 | const_debug unsigned int sysctl_timer_migration = 1; |
6491 | ||
1da177e4 LT |
6492 | int in_sched_functions(unsigned long addr) |
6493 | { | |
1da177e4 LT |
6494 | return in_lock_functions(addr) || |
6495 | (addr >= (unsigned long)__sched_text_start | |
6496 | && addr < (unsigned long)__sched_text_end); | |
6497 | } | |
6498 | ||
029632fb | 6499 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6500 | /* |
6501 | * Default task group. | |
6502 | * Every task in system belongs to this group at bootup. | |
6503 | */ | |
029632fb | 6504 | struct task_group root_task_group; |
35cf4e50 | 6505 | LIST_HEAD(task_groups); |
052f1dc7 | 6506 | #endif |
6f505b16 | 6507 | |
e6252c3e | 6508 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6509 | |
1da177e4 LT |
6510 | void __init sched_init(void) |
6511 | { | |
dd41f596 | 6512 | int i, j; |
434d53b0 MT |
6513 | unsigned long alloc_size = 0, ptr; |
6514 | ||
6515 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6516 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6517 | #endif | |
6518 | #ifdef CONFIG_RT_GROUP_SCHED | |
6519 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6520 | #endif |
df7c8e84 | 6521 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6522 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6523 | #endif |
434d53b0 | 6524 | if (alloc_size) { |
36b7b6d4 | 6525 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6526 | |
6527 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6528 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6529 | ptr += nr_cpu_ids * sizeof(void **); |
6530 | ||
07e06b01 | 6531 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6532 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6533 | |
6d6bc0ad | 6534 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6535 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6536 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6537 | ptr += nr_cpu_ids * sizeof(void **); |
6538 | ||
07e06b01 | 6539 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6540 | ptr += nr_cpu_ids * sizeof(void **); |
6541 | ||
6d6bc0ad | 6542 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6543 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6544 | for_each_possible_cpu(i) { | |
e6252c3e | 6545 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6546 | ptr += cpumask_size(); |
6547 | } | |
6548 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6549 | } |
dd41f596 | 6550 | |
57d885fe GH |
6551 | #ifdef CONFIG_SMP |
6552 | init_defrootdomain(); | |
6553 | #endif | |
6554 | ||
d0b27fa7 PZ |
6555 | init_rt_bandwidth(&def_rt_bandwidth, |
6556 | global_rt_period(), global_rt_runtime()); | |
6557 | ||
6558 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6559 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6560 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6561 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6562 | |
7c941438 | 6563 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6564 | list_add(&root_task_group.list, &task_groups); |
6565 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6566 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6567 | autogroup_init(&init_task); |
54c707e9 | 6568 | |
7c941438 | 6569 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6570 | |
0a945022 | 6571 | for_each_possible_cpu(i) { |
70b97a7f | 6572 | struct rq *rq; |
1da177e4 LT |
6573 | |
6574 | rq = cpu_rq(i); | |
05fa785c | 6575 | raw_spin_lock_init(&rq->lock); |
7897986b | 6576 | rq->nr_running = 0; |
dce48a84 TG |
6577 | rq->calc_load_active = 0; |
6578 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6579 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6580 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6581 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6582 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6583 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6584 | /* |
07e06b01 | 6585 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6586 | * |
6587 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6588 | * gets 100% of the cpu resources in the system. This overall | |
6589 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6590 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6591 | * based on each entity's (task or task-group's) weight |
6592 | * (se->load.weight). | |
6593 | * | |
07e06b01 | 6594 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6595 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6596 | * then A0's share of the cpu resource is: | |
6597 | * | |
0d905bca | 6598 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6599 | * |
07e06b01 YZ |
6600 | * We achieve this by letting root_task_group's tasks sit |
6601 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6602 | */ |
ab84d31e | 6603 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6604 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6605 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6606 | ||
6607 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6608 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6609 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6610 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6611 | #endif |
1da177e4 | 6612 | |
dd41f596 IM |
6613 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6614 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6615 | |
6616 | rq->last_load_update_tick = jiffies; | |
6617 | ||
1da177e4 | 6618 | #ifdef CONFIG_SMP |
41c7ce9a | 6619 | rq->sd = NULL; |
57d885fe | 6620 | rq->rd = NULL; |
1399fa78 | 6621 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6622 | rq->post_schedule = 0; |
1da177e4 | 6623 | rq->active_balance = 0; |
dd41f596 | 6624 | rq->next_balance = jiffies; |
1da177e4 | 6625 | rq->push_cpu = 0; |
0a2966b4 | 6626 | rq->cpu = i; |
1f11eb6a | 6627 | rq->online = 0; |
eae0c9df MG |
6628 | rq->idle_stamp = 0; |
6629 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6630 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6631 | |
6632 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6633 | ||
dc938520 | 6634 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6635 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6636 | rq->nohz_flags = 0; |
83cd4fe2 | 6637 | #endif |
265f22a9 FW |
6638 | #ifdef CONFIG_NO_HZ_FULL |
6639 | rq->last_sched_tick = 0; | |
6640 | #endif | |
1da177e4 | 6641 | #endif |
8f4d37ec | 6642 | init_rq_hrtick(rq); |
1da177e4 | 6643 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6644 | } |
6645 | ||
2dd73a4f | 6646 | set_load_weight(&init_task); |
b50f60ce | 6647 | |
e107be36 AK |
6648 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6649 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6650 | #endif | |
6651 | ||
b50f60ce | 6652 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6653 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6654 | #endif |
6655 | ||
1da177e4 LT |
6656 | /* |
6657 | * The boot idle thread does lazy MMU switching as well: | |
6658 | */ | |
6659 | atomic_inc(&init_mm.mm_count); | |
6660 | enter_lazy_tlb(&init_mm, current); | |
6661 | ||
6662 | /* | |
6663 | * Make us the idle thread. Technically, schedule() should not be | |
6664 | * called from this thread, however somewhere below it might be, | |
6665 | * but because we are the idle thread, we just pick up running again | |
6666 | * when this runqueue becomes "idle". | |
6667 | */ | |
6668 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6669 | |
6670 | calc_load_update = jiffies + LOAD_FREQ; | |
6671 | ||
dd41f596 IM |
6672 | /* |
6673 | * During early bootup we pretend to be a normal task: | |
6674 | */ | |
6675 | current->sched_class = &fair_sched_class; | |
6892b75e | 6676 | |
bf4d83f6 | 6677 | #ifdef CONFIG_SMP |
4cb98839 | 6678 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6679 | /* May be allocated at isolcpus cmdline parse time */ |
6680 | if (cpu_isolated_map == NULL) | |
6681 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6682 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6683 | #endif |
6684 | init_sched_fair_class(); | |
6a7b3dc3 | 6685 | |
6892b75e | 6686 | scheduler_running = 1; |
1da177e4 LT |
6687 | } |
6688 | ||
d902db1e | 6689 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6690 | static inline int preempt_count_equals(int preempt_offset) |
6691 | { | |
234da7bc | 6692 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6693 | |
4ba8216c | 6694 | return (nested == preempt_offset); |
e4aafea2 FW |
6695 | } |
6696 | ||
d894837f | 6697 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6698 | { |
1da177e4 LT |
6699 | static unsigned long prev_jiffy; /* ratelimiting */ |
6700 | ||
b3fbab05 | 6701 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6702 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
6703 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
6704 | return; |
6705 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6706 | return; | |
6707 | prev_jiffy = jiffies; | |
6708 | ||
3df0fc5b PZ |
6709 | printk(KERN_ERR |
6710 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6711 | file, line); | |
6712 | printk(KERN_ERR | |
6713 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6714 | in_atomic(), irqs_disabled(), | |
6715 | current->pid, current->comm); | |
aef745fc IM |
6716 | |
6717 | debug_show_held_locks(current); | |
6718 | if (irqs_disabled()) | |
6719 | print_irqtrace_events(current); | |
6720 | dump_stack(); | |
1da177e4 LT |
6721 | } |
6722 | EXPORT_SYMBOL(__might_sleep); | |
6723 | #endif | |
6724 | ||
6725 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6726 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6727 | { | |
da7a735e PZ |
6728 | const struct sched_class *prev_class = p->sched_class; |
6729 | int old_prio = p->prio; | |
3a5e4dc1 | 6730 | int on_rq; |
3e51f33f | 6731 | |
fd2f4419 | 6732 | on_rq = p->on_rq; |
3a5e4dc1 | 6733 | if (on_rq) |
4ca9b72b | 6734 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
6735 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
6736 | if (on_rq) { | |
4ca9b72b | 6737 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
6738 | resched_task(rq->curr); |
6739 | } | |
da7a735e PZ |
6740 | |
6741 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
6742 | } |
6743 | ||
1da177e4 LT |
6744 | void normalize_rt_tasks(void) |
6745 | { | |
a0f98a1c | 6746 | struct task_struct *g, *p; |
1da177e4 | 6747 | unsigned long flags; |
70b97a7f | 6748 | struct rq *rq; |
1da177e4 | 6749 | |
4cf5d77a | 6750 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 6751 | do_each_thread(g, p) { |
178be793 IM |
6752 | /* |
6753 | * Only normalize user tasks: | |
6754 | */ | |
6755 | if (!p->mm) | |
6756 | continue; | |
6757 | ||
6cfb0d5d | 6758 | p->se.exec_start = 0; |
6cfb0d5d | 6759 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
6760 | p->se.statistics.wait_start = 0; |
6761 | p->se.statistics.sleep_start = 0; | |
6762 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 6763 | #endif |
dd41f596 IM |
6764 | |
6765 | if (!rt_task(p)) { | |
6766 | /* | |
6767 | * Renice negative nice level userspace | |
6768 | * tasks back to 0: | |
6769 | */ | |
6770 | if (TASK_NICE(p) < 0 && p->mm) | |
6771 | set_user_nice(p, 0); | |
1da177e4 | 6772 | continue; |
dd41f596 | 6773 | } |
1da177e4 | 6774 | |
1d615482 | 6775 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 6776 | rq = __task_rq_lock(p); |
1da177e4 | 6777 | |
178be793 | 6778 | normalize_task(rq, p); |
3a5e4dc1 | 6779 | |
b29739f9 | 6780 | __task_rq_unlock(rq); |
1d615482 | 6781 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
6782 | } while_each_thread(g, p); |
6783 | ||
4cf5d77a | 6784 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
6785 | } |
6786 | ||
6787 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6788 | |
67fc4e0c | 6789 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6790 | /* |
67fc4e0c | 6791 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6792 | * |
6793 | * They can only be called when the whole system has been | |
6794 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6795 | * activity can take place. Using them for anything else would | |
6796 | * be a serious bug, and as a result, they aren't even visible | |
6797 | * under any other configuration. | |
6798 | */ | |
6799 | ||
6800 | /** | |
6801 | * curr_task - return the current task for a given cpu. | |
6802 | * @cpu: the processor in question. | |
6803 | * | |
6804 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6805 | * |
6806 | * Return: The current task for @cpu. | |
1df5c10a | 6807 | */ |
36c8b586 | 6808 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6809 | { |
6810 | return cpu_curr(cpu); | |
6811 | } | |
6812 | ||
67fc4e0c JW |
6813 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6814 | ||
6815 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
6816 | /** |
6817 | * set_curr_task - set the current task for a given cpu. | |
6818 | * @cpu: the processor in question. | |
6819 | * @p: the task pointer to set. | |
6820 | * | |
6821 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
6822 | * are serviced on a separate stack. It allows the architecture to switch the |
6823 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
6824 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6825 | * and caller must save the original value of the current task (see | |
6826 | * curr_task() above) and restore that value before reenabling interrupts and | |
6827 | * re-starting the system. | |
6828 | * | |
6829 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6830 | */ | |
36c8b586 | 6831 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6832 | { |
6833 | cpu_curr(cpu) = p; | |
6834 | } | |
6835 | ||
6836 | #endif | |
29f59db3 | 6837 | |
7c941438 | 6838 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6839 | /* task_group_lock serializes the addition/removal of task groups */ |
6840 | static DEFINE_SPINLOCK(task_group_lock); | |
6841 | ||
bccbe08a PZ |
6842 | static void free_sched_group(struct task_group *tg) |
6843 | { | |
6844 | free_fair_sched_group(tg); | |
6845 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6846 | autogroup_free(tg); |
bccbe08a PZ |
6847 | kfree(tg); |
6848 | } | |
6849 | ||
6850 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6851 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6852 | { |
6853 | struct task_group *tg; | |
bccbe08a PZ |
6854 | |
6855 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
6856 | if (!tg) | |
6857 | return ERR_PTR(-ENOMEM); | |
6858 | ||
ec7dc8ac | 6859 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6860 | goto err; |
6861 | ||
ec7dc8ac | 6862 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6863 | goto err; |
6864 | ||
ace783b9 LZ |
6865 | return tg; |
6866 | ||
6867 | err: | |
6868 | free_sched_group(tg); | |
6869 | return ERR_PTR(-ENOMEM); | |
6870 | } | |
6871 | ||
6872 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6873 | { | |
6874 | unsigned long flags; | |
6875 | ||
8ed36996 | 6876 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6877 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
6878 | |
6879 | WARN_ON(!parent); /* root should already exist */ | |
6880 | ||
6881 | tg->parent = parent; | |
f473aa5e | 6882 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6883 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6884 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6885 | } |
6886 | ||
9b5b7751 | 6887 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 6888 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6889 | { |
29f59db3 | 6890 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 6891 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6892 | } |
6893 | ||
9b5b7751 | 6894 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6895 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
6896 | { |
6897 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6898 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
6899 | } | |
6900 | ||
6901 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6902 | { |
8ed36996 | 6903 | unsigned long flags; |
9b5b7751 | 6904 | int i; |
29f59db3 | 6905 | |
3d4b47b4 PZ |
6906 | /* end participation in shares distribution */ |
6907 | for_each_possible_cpu(i) | |
bccbe08a | 6908 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
6909 | |
6910 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6911 | list_del_rcu(&tg->list); |
f473aa5e | 6912 | list_del_rcu(&tg->siblings); |
8ed36996 | 6913 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6914 | } |
6915 | ||
9b5b7751 | 6916 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6917 | * The caller of this function should have put the task in its new group |
6918 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6919 | * reflect its new group. | |
9b5b7751 SV |
6920 | */ |
6921 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 6922 | { |
8323f26c | 6923 | struct task_group *tg; |
29f59db3 SV |
6924 | int on_rq, running; |
6925 | unsigned long flags; | |
6926 | struct rq *rq; | |
6927 | ||
6928 | rq = task_rq_lock(tsk, &flags); | |
6929 | ||
051a1d1a | 6930 | running = task_current(rq, tsk); |
fd2f4419 | 6931 | on_rq = tsk->on_rq; |
29f59db3 | 6932 | |
0e1f3483 | 6933 | if (on_rq) |
29f59db3 | 6934 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
6935 | if (unlikely(running)) |
6936 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 6937 | |
8af01f56 | 6938 | tg = container_of(task_css_check(tsk, cpu_cgroup_subsys_id, |
8323f26c PZ |
6939 | lockdep_is_held(&tsk->sighand->siglock)), |
6940 | struct task_group, css); | |
6941 | tg = autogroup_task_group(tsk, tg); | |
6942 | tsk->sched_task_group = tg; | |
6943 | ||
810b3817 | 6944 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
6945 | if (tsk->sched_class->task_move_group) |
6946 | tsk->sched_class->task_move_group(tsk, on_rq); | |
6947 | else | |
810b3817 | 6948 | #endif |
b2b5ce02 | 6949 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 6950 | |
0e1f3483 HS |
6951 | if (unlikely(running)) |
6952 | tsk->sched_class->set_curr_task(rq); | |
6953 | if (on_rq) | |
371fd7e7 | 6954 | enqueue_task(rq, tsk, 0); |
29f59db3 | 6955 | |
0122ec5b | 6956 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 6957 | } |
7c941438 | 6958 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 6959 | |
a790de99 | 6960 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
6961 | static unsigned long to_ratio(u64 period, u64 runtime) |
6962 | { | |
6963 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 6964 | return 1ULL << 20; |
9f0c1e56 | 6965 | |
9a7e0b18 | 6966 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 6967 | } |
a790de99 PT |
6968 | #endif |
6969 | ||
6970 | #ifdef CONFIG_RT_GROUP_SCHED | |
6971 | /* | |
6972 | * Ensure that the real time constraints are schedulable. | |
6973 | */ | |
6974 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 6975 | |
9a7e0b18 PZ |
6976 | /* Must be called with tasklist_lock held */ |
6977 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 6978 | { |
9a7e0b18 | 6979 | struct task_struct *g, *p; |
b40b2e8e | 6980 | |
9a7e0b18 | 6981 | do_each_thread(g, p) { |
029632fb | 6982 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
6983 | return 1; |
6984 | } while_each_thread(g, p); | |
b40b2e8e | 6985 | |
9a7e0b18 PZ |
6986 | return 0; |
6987 | } | |
b40b2e8e | 6988 | |
9a7e0b18 PZ |
6989 | struct rt_schedulable_data { |
6990 | struct task_group *tg; | |
6991 | u64 rt_period; | |
6992 | u64 rt_runtime; | |
6993 | }; | |
b40b2e8e | 6994 | |
a790de99 | 6995 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
6996 | { |
6997 | struct rt_schedulable_data *d = data; | |
6998 | struct task_group *child; | |
6999 | unsigned long total, sum = 0; | |
7000 | u64 period, runtime; | |
b40b2e8e | 7001 | |
9a7e0b18 PZ |
7002 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7003 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7004 | |
9a7e0b18 PZ |
7005 | if (tg == d->tg) { |
7006 | period = d->rt_period; | |
7007 | runtime = d->rt_runtime; | |
b40b2e8e | 7008 | } |
b40b2e8e | 7009 | |
4653f803 PZ |
7010 | /* |
7011 | * Cannot have more runtime than the period. | |
7012 | */ | |
7013 | if (runtime > period && runtime != RUNTIME_INF) | |
7014 | return -EINVAL; | |
6f505b16 | 7015 | |
4653f803 PZ |
7016 | /* |
7017 | * Ensure we don't starve existing RT tasks. | |
7018 | */ | |
9a7e0b18 PZ |
7019 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7020 | return -EBUSY; | |
6f505b16 | 7021 | |
9a7e0b18 | 7022 | total = to_ratio(period, runtime); |
6f505b16 | 7023 | |
4653f803 PZ |
7024 | /* |
7025 | * Nobody can have more than the global setting allows. | |
7026 | */ | |
7027 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7028 | return -EINVAL; | |
6f505b16 | 7029 | |
4653f803 PZ |
7030 | /* |
7031 | * The sum of our children's runtime should not exceed our own. | |
7032 | */ | |
9a7e0b18 PZ |
7033 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7034 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7035 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7036 | |
9a7e0b18 PZ |
7037 | if (child == d->tg) { |
7038 | period = d->rt_period; | |
7039 | runtime = d->rt_runtime; | |
7040 | } | |
6f505b16 | 7041 | |
9a7e0b18 | 7042 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7043 | } |
6f505b16 | 7044 | |
9a7e0b18 PZ |
7045 | if (sum > total) |
7046 | return -EINVAL; | |
7047 | ||
7048 | return 0; | |
6f505b16 PZ |
7049 | } |
7050 | ||
9a7e0b18 | 7051 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7052 | { |
8277434e PT |
7053 | int ret; |
7054 | ||
9a7e0b18 PZ |
7055 | struct rt_schedulable_data data = { |
7056 | .tg = tg, | |
7057 | .rt_period = period, | |
7058 | .rt_runtime = runtime, | |
7059 | }; | |
7060 | ||
8277434e PT |
7061 | rcu_read_lock(); |
7062 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7063 | rcu_read_unlock(); | |
7064 | ||
7065 | return ret; | |
521f1a24 DG |
7066 | } |
7067 | ||
ab84d31e | 7068 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7069 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7070 | { |
ac086bc2 | 7071 | int i, err = 0; |
9f0c1e56 | 7072 | |
9f0c1e56 | 7073 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7074 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7075 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7076 | if (err) | |
9f0c1e56 | 7077 | goto unlock; |
ac086bc2 | 7078 | |
0986b11b | 7079 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7080 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7081 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7082 | |
7083 | for_each_possible_cpu(i) { | |
7084 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7085 | ||
0986b11b | 7086 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7087 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7088 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7089 | } |
0986b11b | 7090 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7091 | unlock: |
521f1a24 | 7092 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7093 | mutex_unlock(&rt_constraints_mutex); |
7094 | ||
7095 | return err; | |
6f505b16 PZ |
7096 | } |
7097 | ||
25cc7da7 | 7098 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7099 | { |
7100 | u64 rt_runtime, rt_period; | |
7101 | ||
7102 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7103 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7104 | if (rt_runtime_us < 0) | |
7105 | rt_runtime = RUNTIME_INF; | |
7106 | ||
ab84d31e | 7107 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7108 | } |
7109 | ||
25cc7da7 | 7110 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7111 | { |
7112 | u64 rt_runtime_us; | |
7113 | ||
d0b27fa7 | 7114 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7115 | return -1; |
7116 | ||
d0b27fa7 | 7117 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7118 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7119 | return rt_runtime_us; | |
7120 | } | |
d0b27fa7 | 7121 | |
25cc7da7 | 7122 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7123 | { |
7124 | u64 rt_runtime, rt_period; | |
7125 | ||
7126 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7127 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7128 | ||
619b0488 R |
7129 | if (rt_period == 0) |
7130 | return -EINVAL; | |
7131 | ||
ab84d31e | 7132 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7133 | } |
7134 | ||
25cc7da7 | 7135 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7136 | { |
7137 | u64 rt_period_us; | |
7138 | ||
7139 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7140 | do_div(rt_period_us, NSEC_PER_USEC); | |
7141 | return rt_period_us; | |
7142 | } | |
7143 | ||
7144 | static int sched_rt_global_constraints(void) | |
7145 | { | |
4653f803 | 7146 | u64 runtime, period; |
d0b27fa7 PZ |
7147 | int ret = 0; |
7148 | ||
ec5d4989 HS |
7149 | if (sysctl_sched_rt_period <= 0) |
7150 | return -EINVAL; | |
7151 | ||
4653f803 PZ |
7152 | runtime = global_rt_runtime(); |
7153 | period = global_rt_period(); | |
7154 | ||
7155 | /* | |
7156 | * Sanity check on the sysctl variables. | |
7157 | */ | |
7158 | if (runtime > period && runtime != RUNTIME_INF) | |
7159 | return -EINVAL; | |
10b612f4 | 7160 | |
d0b27fa7 | 7161 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7162 | read_lock(&tasklist_lock); |
4653f803 | 7163 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7164 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7165 | mutex_unlock(&rt_constraints_mutex); |
7166 | ||
7167 | return ret; | |
7168 | } | |
54e99124 | 7169 | |
25cc7da7 | 7170 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7171 | { |
7172 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7173 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7174 | return 0; | |
7175 | ||
7176 | return 1; | |
7177 | } | |
7178 | ||
6d6bc0ad | 7179 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7180 | static int sched_rt_global_constraints(void) |
7181 | { | |
ac086bc2 PZ |
7182 | unsigned long flags; |
7183 | int i; | |
7184 | ||
ec5d4989 HS |
7185 | if (sysctl_sched_rt_period <= 0) |
7186 | return -EINVAL; | |
7187 | ||
60aa605d PZ |
7188 | /* |
7189 | * There's always some RT tasks in the root group | |
7190 | * -- migration, kstopmachine etc.. | |
7191 | */ | |
7192 | if (sysctl_sched_rt_runtime == 0) | |
7193 | return -EBUSY; | |
7194 | ||
0986b11b | 7195 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7196 | for_each_possible_cpu(i) { |
7197 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7198 | ||
0986b11b | 7199 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7200 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7201 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7202 | } |
0986b11b | 7203 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7204 | |
d0b27fa7 PZ |
7205 | return 0; |
7206 | } | |
6d6bc0ad | 7207 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7208 | |
ce0dbbbb CW |
7209 | int sched_rr_handler(struct ctl_table *table, int write, |
7210 | void __user *buffer, size_t *lenp, | |
7211 | loff_t *ppos) | |
7212 | { | |
7213 | int ret; | |
7214 | static DEFINE_MUTEX(mutex); | |
7215 | ||
7216 | mutex_lock(&mutex); | |
7217 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7218 | /* make sure that internally we keep jiffies */ | |
7219 | /* also, writing zero resets timeslice to default */ | |
7220 | if (!ret && write) { | |
7221 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7222 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7223 | } | |
7224 | mutex_unlock(&mutex); | |
7225 | return ret; | |
7226 | } | |
7227 | ||
d0b27fa7 | 7228 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7229 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7230 | loff_t *ppos) |
7231 | { | |
7232 | int ret; | |
7233 | int old_period, old_runtime; | |
7234 | static DEFINE_MUTEX(mutex); | |
7235 | ||
7236 | mutex_lock(&mutex); | |
7237 | old_period = sysctl_sched_rt_period; | |
7238 | old_runtime = sysctl_sched_rt_runtime; | |
7239 | ||
8d65af78 | 7240 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7241 | |
7242 | if (!ret && write) { | |
7243 | ret = sched_rt_global_constraints(); | |
7244 | if (ret) { | |
7245 | sysctl_sched_rt_period = old_period; | |
7246 | sysctl_sched_rt_runtime = old_runtime; | |
7247 | } else { | |
7248 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7249 | def_rt_bandwidth.rt_period = | |
7250 | ns_to_ktime(global_rt_period()); | |
7251 | } | |
7252 | } | |
7253 | mutex_unlock(&mutex); | |
7254 | ||
7255 | return ret; | |
7256 | } | |
68318b8e | 7257 | |
052f1dc7 | 7258 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7259 | |
a7c6d554 | 7260 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7261 | { |
a7c6d554 | 7262 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7263 | } |
7264 | ||
eb95419b TH |
7265 | static struct cgroup_subsys_state * |
7266 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7267 | { |
eb95419b TH |
7268 | struct task_group *parent = css_tg(parent_css); |
7269 | struct task_group *tg; | |
68318b8e | 7270 | |
eb95419b | 7271 | if (!parent) { |
68318b8e | 7272 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7273 | return &root_task_group.css; |
68318b8e SV |
7274 | } |
7275 | ||
ec7dc8ac | 7276 | tg = sched_create_group(parent); |
68318b8e SV |
7277 | if (IS_ERR(tg)) |
7278 | return ERR_PTR(-ENOMEM); | |
7279 | ||
68318b8e SV |
7280 | return &tg->css; |
7281 | } | |
7282 | ||
eb95419b | 7283 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7284 | { |
eb95419b TH |
7285 | struct task_group *tg = css_tg(css); |
7286 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7287 | |
63876986 TH |
7288 | if (parent) |
7289 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7290 | return 0; |
7291 | } | |
7292 | ||
eb95419b | 7293 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7294 | { |
eb95419b | 7295 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7296 | |
7297 | sched_destroy_group(tg); | |
7298 | } | |
7299 | ||
eb95419b | 7300 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7301 | { |
eb95419b | 7302 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7303 | |
7304 | sched_offline_group(tg); | |
7305 | } | |
7306 | ||
eb95419b | 7307 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7308 | struct cgroup_taskset *tset) |
68318b8e | 7309 | { |
bb9d97b6 TH |
7310 | struct task_struct *task; |
7311 | ||
d99c8727 | 7312 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7313 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7314 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7315 | return -EINVAL; |
b68aa230 | 7316 | #else |
bb9d97b6 TH |
7317 | /* We don't support RT-tasks being in separate groups */ |
7318 | if (task->sched_class != &fair_sched_class) | |
7319 | return -EINVAL; | |
b68aa230 | 7320 | #endif |
bb9d97b6 | 7321 | } |
be367d09 BB |
7322 | return 0; |
7323 | } | |
68318b8e | 7324 | |
eb95419b | 7325 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7326 | struct cgroup_taskset *tset) |
68318b8e | 7327 | { |
bb9d97b6 TH |
7328 | struct task_struct *task; |
7329 | ||
d99c8727 | 7330 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7331 | sched_move_task(task); |
68318b8e SV |
7332 | } |
7333 | ||
eb95419b TH |
7334 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7335 | struct cgroup_subsys_state *old_css, | |
7336 | struct task_struct *task) | |
068c5cc5 PZ |
7337 | { |
7338 | /* | |
7339 | * cgroup_exit() is called in the copy_process() failure path. | |
7340 | * Ignore this case since the task hasn't ran yet, this avoids | |
7341 | * trying to poke a half freed task state from generic code. | |
7342 | */ | |
7343 | if (!(task->flags & PF_EXITING)) | |
7344 | return; | |
7345 | ||
7346 | sched_move_task(task); | |
7347 | } | |
7348 | ||
052f1dc7 | 7349 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7350 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7351 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7352 | { |
182446d0 | 7353 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7354 | } |
7355 | ||
182446d0 TH |
7356 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7357 | struct cftype *cft) | |
68318b8e | 7358 | { |
182446d0 | 7359 | struct task_group *tg = css_tg(css); |
68318b8e | 7360 | |
c8b28116 | 7361 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7362 | } |
ab84d31e PT |
7363 | |
7364 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7365 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7366 | ||
ab84d31e PT |
7367 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7368 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7369 | ||
a790de99 PT |
7370 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7371 | ||
ab84d31e PT |
7372 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7373 | { | |
56f570e5 | 7374 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7375 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7376 | |
7377 | if (tg == &root_task_group) | |
7378 | return -EINVAL; | |
7379 | ||
7380 | /* | |
7381 | * Ensure we have at some amount of bandwidth every period. This is | |
7382 | * to prevent reaching a state of large arrears when throttled via | |
7383 | * entity_tick() resulting in prolonged exit starvation. | |
7384 | */ | |
7385 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7386 | return -EINVAL; | |
7387 | ||
7388 | /* | |
7389 | * Likewise, bound things on the otherside by preventing insane quota | |
7390 | * periods. This also allows us to normalize in computing quota | |
7391 | * feasibility. | |
7392 | */ | |
7393 | if (period > max_cfs_quota_period) | |
7394 | return -EINVAL; | |
7395 | ||
a790de99 PT |
7396 | mutex_lock(&cfs_constraints_mutex); |
7397 | ret = __cfs_schedulable(tg, period, quota); | |
7398 | if (ret) | |
7399 | goto out_unlock; | |
7400 | ||
58088ad0 | 7401 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7402 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7403 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7404 | raw_spin_lock_irq(&cfs_b->lock); |
7405 | cfs_b->period = ns_to_ktime(period); | |
7406 | cfs_b->quota = quota; | |
58088ad0 | 7407 | |
a9cf55b2 | 7408 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7409 | /* restart the period timer (if active) to handle new period expiry */ |
7410 | if (runtime_enabled && cfs_b->timer_active) { | |
7411 | /* force a reprogram */ | |
7412 | cfs_b->timer_active = 0; | |
7413 | __start_cfs_bandwidth(cfs_b); | |
7414 | } | |
ab84d31e PT |
7415 | raw_spin_unlock_irq(&cfs_b->lock); |
7416 | ||
7417 | for_each_possible_cpu(i) { | |
7418 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7419 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7420 | |
7421 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7422 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7423 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7424 | |
029632fb | 7425 | if (cfs_rq->throttled) |
671fd9da | 7426 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7427 | raw_spin_unlock_irq(&rq->lock); |
7428 | } | |
a790de99 PT |
7429 | out_unlock: |
7430 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7431 | |
a790de99 | 7432 | return ret; |
ab84d31e PT |
7433 | } |
7434 | ||
7435 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7436 | { | |
7437 | u64 quota, period; | |
7438 | ||
029632fb | 7439 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7440 | if (cfs_quota_us < 0) |
7441 | quota = RUNTIME_INF; | |
7442 | else | |
7443 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7444 | ||
7445 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7446 | } | |
7447 | ||
7448 | long tg_get_cfs_quota(struct task_group *tg) | |
7449 | { | |
7450 | u64 quota_us; | |
7451 | ||
029632fb | 7452 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7453 | return -1; |
7454 | ||
029632fb | 7455 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7456 | do_div(quota_us, NSEC_PER_USEC); |
7457 | ||
7458 | return quota_us; | |
7459 | } | |
7460 | ||
7461 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7462 | { | |
7463 | u64 quota, period; | |
7464 | ||
7465 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7466 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7467 | |
ab84d31e PT |
7468 | return tg_set_cfs_bandwidth(tg, period, quota); |
7469 | } | |
7470 | ||
7471 | long tg_get_cfs_period(struct task_group *tg) | |
7472 | { | |
7473 | u64 cfs_period_us; | |
7474 | ||
029632fb | 7475 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7476 | do_div(cfs_period_us, NSEC_PER_USEC); |
7477 | ||
7478 | return cfs_period_us; | |
7479 | } | |
7480 | ||
182446d0 TH |
7481 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7482 | struct cftype *cft) | |
ab84d31e | 7483 | { |
182446d0 | 7484 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7485 | } |
7486 | ||
182446d0 TH |
7487 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7488 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7489 | { |
182446d0 | 7490 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7491 | } |
7492 | ||
182446d0 TH |
7493 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7494 | struct cftype *cft) | |
ab84d31e | 7495 | { |
182446d0 | 7496 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7497 | } |
7498 | ||
182446d0 TH |
7499 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7500 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7501 | { |
182446d0 | 7502 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7503 | } |
7504 | ||
a790de99 PT |
7505 | struct cfs_schedulable_data { |
7506 | struct task_group *tg; | |
7507 | u64 period, quota; | |
7508 | }; | |
7509 | ||
7510 | /* | |
7511 | * normalize group quota/period to be quota/max_period | |
7512 | * note: units are usecs | |
7513 | */ | |
7514 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7515 | struct cfs_schedulable_data *d) | |
7516 | { | |
7517 | u64 quota, period; | |
7518 | ||
7519 | if (tg == d->tg) { | |
7520 | period = d->period; | |
7521 | quota = d->quota; | |
7522 | } else { | |
7523 | period = tg_get_cfs_period(tg); | |
7524 | quota = tg_get_cfs_quota(tg); | |
7525 | } | |
7526 | ||
7527 | /* note: these should typically be equivalent */ | |
7528 | if (quota == RUNTIME_INF || quota == -1) | |
7529 | return RUNTIME_INF; | |
7530 | ||
7531 | return to_ratio(period, quota); | |
7532 | } | |
7533 | ||
7534 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7535 | { | |
7536 | struct cfs_schedulable_data *d = data; | |
029632fb | 7537 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7538 | s64 quota = 0, parent_quota = -1; |
7539 | ||
7540 | if (!tg->parent) { | |
7541 | quota = RUNTIME_INF; | |
7542 | } else { | |
029632fb | 7543 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7544 | |
7545 | quota = normalize_cfs_quota(tg, d); | |
7546 | parent_quota = parent_b->hierarchal_quota; | |
7547 | ||
7548 | /* | |
7549 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7550 | * limit is set | |
7551 | */ | |
7552 | if (quota == RUNTIME_INF) | |
7553 | quota = parent_quota; | |
7554 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7555 | return -EINVAL; | |
7556 | } | |
7557 | cfs_b->hierarchal_quota = quota; | |
7558 | ||
7559 | return 0; | |
7560 | } | |
7561 | ||
7562 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7563 | { | |
8277434e | 7564 | int ret; |
a790de99 PT |
7565 | struct cfs_schedulable_data data = { |
7566 | .tg = tg, | |
7567 | .period = period, | |
7568 | .quota = quota, | |
7569 | }; | |
7570 | ||
7571 | if (quota != RUNTIME_INF) { | |
7572 | do_div(data.period, NSEC_PER_USEC); | |
7573 | do_div(data.quota, NSEC_PER_USEC); | |
7574 | } | |
7575 | ||
8277434e PT |
7576 | rcu_read_lock(); |
7577 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7578 | rcu_read_unlock(); | |
7579 | ||
7580 | return ret; | |
a790de99 | 7581 | } |
e8da1b18 | 7582 | |
182446d0 | 7583 | static int cpu_stats_show(struct cgroup_subsys_state *css, struct cftype *cft, |
e8da1b18 NR |
7584 | struct cgroup_map_cb *cb) |
7585 | { | |
182446d0 | 7586 | struct task_group *tg = css_tg(css); |
029632fb | 7587 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7588 | |
7589 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7590 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7591 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7592 | ||
7593 | return 0; | |
7594 | } | |
ab84d31e | 7595 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7596 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7597 | |
052f1dc7 | 7598 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7599 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7600 | struct cftype *cft, s64 val) | |
6f505b16 | 7601 | { |
182446d0 | 7602 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7603 | } |
7604 | ||
182446d0 TH |
7605 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7606 | struct cftype *cft) | |
6f505b16 | 7607 | { |
182446d0 | 7608 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7609 | } |
d0b27fa7 | 7610 | |
182446d0 TH |
7611 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7612 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7613 | { |
182446d0 | 7614 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7615 | } |
7616 | ||
182446d0 TH |
7617 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7618 | struct cftype *cft) | |
d0b27fa7 | 7619 | { |
182446d0 | 7620 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7621 | } |
6d6bc0ad | 7622 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7623 | |
fe5c7cc2 | 7624 | static struct cftype cpu_files[] = { |
052f1dc7 | 7625 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7626 | { |
7627 | .name = "shares", | |
f4c753b7 PM |
7628 | .read_u64 = cpu_shares_read_u64, |
7629 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7630 | }, |
052f1dc7 | 7631 | #endif |
ab84d31e PT |
7632 | #ifdef CONFIG_CFS_BANDWIDTH |
7633 | { | |
7634 | .name = "cfs_quota_us", | |
7635 | .read_s64 = cpu_cfs_quota_read_s64, | |
7636 | .write_s64 = cpu_cfs_quota_write_s64, | |
7637 | }, | |
7638 | { | |
7639 | .name = "cfs_period_us", | |
7640 | .read_u64 = cpu_cfs_period_read_u64, | |
7641 | .write_u64 = cpu_cfs_period_write_u64, | |
7642 | }, | |
e8da1b18 NR |
7643 | { |
7644 | .name = "stat", | |
7645 | .read_map = cpu_stats_show, | |
7646 | }, | |
ab84d31e | 7647 | #endif |
052f1dc7 | 7648 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7649 | { |
9f0c1e56 | 7650 | .name = "rt_runtime_us", |
06ecb27c PM |
7651 | .read_s64 = cpu_rt_runtime_read, |
7652 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7653 | }, |
d0b27fa7 PZ |
7654 | { |
7655 | .name = "rt_period_us", | |
f4c753b7 PM |
7656 | .read_u64 = cpu_rt_period_read_uint, |
7657 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7658 | }, |
052f1dc7 | 7659 | #endif |
4baf6e33 | 7660 | { } /* terminate */ |
68318b8e SV |
7661 | }; |
7662 | ||
68318b8e | 7663 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 7664 | .name = "cpu", |
92fb9748 TH |
7665 | .css_alloc = cpu_cgroup_css_alloc, |
7666 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
7667 | .css_online = cpu_cgroup_css_online, |
7668 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
7669 | .can_attach = cpu_cgroup_can_attach, |
7670 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7671 | .exit = cpu_cgroup_exit, |
38605cae | 7672 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 7673 | .base_cftypes = cpu_files, |
68318b8e SV |
7674 | .early_init = 1, |
7675 | }; | |
7676 | ||
052f1dc7 | 7677 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7678 | |
b637a328 PM |
7679 | void dump_cpu_task(int cpu) |
7680 | { | |
7681 | pr_info("Task dump for CPU %d:\n", cpu); | |
7682 | sched_show_task(cpu_curr(cpu)); | |
7683 | } |