Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
e6e6685a GC |
77 | #ifdef CONFIG_PARAVIRT |
78 | #include <asm/paravirt.h> | |
79 | #endif | |
1da177e4 | 80 | |
029632fb | 81 | #include "sched.h" |
391e43da | 82 | #include "../workqueue_sched.h" |
6e0534f2 | 83 | |
a8d154b0 | 84 | #define CREATE_TRACE_POINTS |
ad8d75ff | 85 | #include <trace/events/sched.h> |
a8d154b0 | 86 | |
029632fb | 87 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 88 | { |
58088ad0 PT |
89 | unsigned long delta; |
90 | ktime_t soft, hard, now; | |
d0b27fa7 | 91 | |
58088ad0 PT |
92 | for (;;) { |
93 | if (hrtimer_active(period_timer)) | |
94 | break; | |
95 | ||
96 | now = hrtimer_cb_get_time(period_timer); | |
97 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 98 | |
58088ad0 PT |
99 | soft = hrtimer_get_softexpires(period_timer); |
100 | hard = hrtimer_get_expires(period_timer); | |
101 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
102 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
103 | HRTIMER_MODE_ABS_PINNED, 0); | |
104 | } | |
105 | } | |
106 | ||
029632fb PZ |
107 | DEFINE_MUTEX(sched_domains_mutex); |
108 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 109 | |
fe44d621 | 110 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 111 | |
029632fb | 112 | void update_rq_clock(struct rq *rq) |
3e51f33f | 113 | { |
fe44d621 | 114 | s64 delta; |
305e6835 | 115 | |
61eadef6 | 116 | if (rq->skip_clock_update > 0) |
f26f9aff | 117 | return; |
aa483808 | 118 | |
fe44d621 PZ |
119 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
120 | rq->clock += delta; | |
121 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
122 | } |
123 | ||
bf5c91ba IM |
124 | /* |
125 | * Debugging: various feature bits | |
126 | */ | |
f00b45c1 | 127 | |
f00b45c1 PZ |
128 | #define SCHED_FEAT(name, enabled) \ |
129 | (1UL << __SCHED_FEAT_##name) * enabled | | |
130 | ||
bf5c91ba | 131 | const_debug unsigned int sysctl_sched_features = |
391e43da | 132 | #include "features.h" |
f00b45c1 PZ |
133 | 0; |
134 | ||
135 | #undef SCHED_FEAT | |
136 | ||
137 | #ifdef CONFIG_SCHED_DEBUG | |
138 | #define SCHED_FEAT(name, enabled) \ | |
139 | #name , | |
140 | ||
983ed7a6 | 141 | static __read_mostly char *sched_feat_names[] = { |
391e43da | 142 | #include "features.h" |
f00b45c1 PZ |
143 | NULL |
144 | }; | |
145 | ||
146 | #undef SCHED_FEAT | |
147 | ||
34f3a814 | 148 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 149 | { |
f00b45c1 PZ |
150 | int i; |
151 | ||
f8b6d1cc | 152 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
153 | if (!(sysctl_sched_features & (1UL << i))) |
154 | seq_puts(m, "NO_"); | |
155 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 156 | } |
34f3a814 | 157 | seq_puts(m, "\n"); |
f00b45c1 | 158 | |
34f3a814 | 159 | return 0; |
f00b45c1 PZ |
160 | } |
161 | ||
f8b6d1cc PZ |
162 | #ifdef HAVE_JUMP_LABEL |
163 | ||
164 | #define jump_label_key__true jump_label_key_enabled | |
165 | #define jump_label_key__false jump_label_key_disabled | |
166 | ||
167 | #define SCHED_FEAT(name, enabled) \ | |
168 | jump_label_key__##enabled , | |
169 | ||
170 | struct jump_label_key sched_feat_keys[__SCHED_FEAT_NR] = { | |
171 | #include "features.h" | |
172 | }; | |
173 | ||
174 | #undef SCHED_FEAT | |
175 | ||
176 | static void sched_feat_disable(int i) | |
177 | { | |
178 | if (jump_label_enabled(&sched_feat_keys[i])) | |
179 | jump_label_dec(&sched_feat_keys[i]); | |
180 | } | |
181 | ||
182 | static void sched_feat_enable(int i) | |
183 | { | |
184 | if (!jump_label_enabled(&sched_feat_keys[i])) | |
185 | jump_label_inc(&sched_feat_keys[i]); | |
186 | } | |
187 | #else | |
188 | static void sched_feat_disable(int i) { }; | |
189 | static void sched_feat_enable(int i) { }; | |
190 | #endif /* HAVE_JUMP_LABEL */ | |
191 | ||
f00b45c1 PZ |
192 | static ssize_t |
193 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
194 | size_t cnt, loff_t *ppos) | |
195 | { | |
196 | char buf[64]; | |
7740191c | 197 | char *cmp; |
f00b45c1 PZ |
198 | int neg = 0; |
199 | int i; | |
200 | ||
201 | if (cnt > 63) | |
202 | cnt = 63; | |
203 | ||
204 | if (copy_from_user(&buf, ubuf, cnt)) | |
205 | return -EFAULT; | |
206 | ||
207 | buf[cnt] = 0; | |
7740191c | 208 | cmp = strstrip(buf); |
f00b45c1 | 209 | |
524429c3 | 210 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
211 | neg = 1; |
212 | cmp += 3; | |
213 | } | |
214 | ||
f8b6d1cc | 215 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 216 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 217 | if (neg) { |
f00b45c1 | 218 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
219 | sched_feat_disable(i); |
220 | } else { | |
f00b45c1 | 221 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
222 | sched_feat_enable(i); |
223 | } | |
f00b45c1 PZ |
224 | break; |
225 | } | |
226 | } | |
227 | ||
f8b6d1cc | 228 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
229 | return -EINVAL; |
230 | ||
42994724 | 231 | *ppos += cnt; |
f00b45c1 PZ |
232 | |
233 | return cnt; | |
234 | } | |
235 | ||
34f3a814 LZ |
236 | static int sched_feat_open(struct inode *inode, struct file *filp) |
237 | { | |
238 | return single_open(filp, sched_feat_show, NULL); | |
239 | } | |
240 | ||
828c0950 | 241 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
242 | .open = sched_feat_open, |
243 | .write = sched_feat_write, | |
244 | .read = seq_read, | |
245 | .llseek = seq_lseek, | |
246 | .release = single_release, | |
f00b45c1 PZ |
247 | }; |
248 | ||
249 | static __init int sched_init_debug(void) | |
250 | { | |
f00b45c1 PZ |
251 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
252 | &sched_feat_fops); | |
253 | ||
254 | return 0; | |
255 | } | |
256 | late_initcall(sched_init_debug); | |
f8b6d1cc | 257 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 258 | |
b82d9fdd PZ |
259 | /* |
260 | * Number of tasks to iterate in a single balance run. | |
261 | * Limited because this is done with IRQs disabled. | |
262 | */ | |
263 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
264 | ||
e9e9250b PZ |
265 | /* |
266 | * period over which we average the RT time consumption, measured | |
267 | * in ms. | |
268 | * | |
269 | * default: 1s | |
270 | */ | |
271 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
272 | ||
fa85ae24 | 273 | /* |
9f0c1e56 | 274 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
275 | * default: 1s |
276 | */ | |
9f0c1e56 | 277 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 278 | |
029632fb | 279 | __read_mostly int scheduler_running; |
6892b75e | 280 | |
9f0c1e56 PZ |
281 | /* |
282 | * part of the period that we allow rt tasks to run in us. | |
283 | * default: 0.95s | |
284 | */ | |
285 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 286 | |
fa85ae24 | 287 | |
1da177e4 | 288 | |
0970d299 | 289 | /* |
0122ec5b | 290 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 291 | */ |
70b97a7f | 292 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
293 | __acquires(rq->lock) |
294 | { | |
0970d299 PZ |
295 | struct rq *rq; |
296 | ||
0122ec5b PZ |
297 | lockdep_assert_held(&p->pi_lock); |
298 | ||
3a5c359a | 299 | for (;;) { |
0970d299 | 300 | rq = task_rq(p); |
05fa785c | 301 | raw_spin_lock(&rq->lock); |
65cc8e48 | 302 | if (likely(rq == task_rq(p))) |
3a5c359a | 303 | return rq; |
05fa785c | 304 | raw_spin_unlock(&rq->lock); |
b29739f9 | 305 | } |
b29739f9 IM |
306 | } |
307 | ||
1da177e4 | 308 | /* |
0122ec5b | 309 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 310 | */ |
70b97a7f | 311 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 312 | __acquires(p->pi_lock) |
1da177e4 LT |
313 | __acquires(rq->lock) |
314 | { | |
70b97a7f | 315 | struct rq *rq; |
1da177e4 | 316 | |
3a5c359a | 317 | for (;;) { |
0122ec5b | 318 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 319 | rq = task_rq(p); |
05fa785c | 320 | raw_spin_lock(&rq->lock); |
65cc8e48 | 321 | if (likely(rq == task_rq(p))) |
3a5c359a | 322 | return rq; |
0122ec5b PZ |
323 | raw_spin_unlock(&rq->lock); |
324 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 325 | } |
1da177e4 LT |
326 | } |
327 | ||
a9957449 | 328 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
329 | __releases(rq->lock) |
330 | { | |
05fa785c | 331 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
332 | } |
333 | ||
0122ec5b PZ |
334 | static inline void |
335 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 336 | __releases(rq->lock) |
0122ec5b | 337 | __releases(p->pi_lock) |
1da177e4 | 338 | { |
0122ec5b PZ |
339 | raw_spin_unlock(&rq->lock); |
340 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
341 | } |
342 | ||
1da177e4 | 343 | /* |
cc2a73b5 | 344 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 345 | */ |
a9957449 | 346 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
347 | __acquires(rq->lock) |
348 | { | |
70b97a7f | 349 | struct rq *rq; |
1da177e4 LT |
350 | |
351 | local_irq_disable(); | |
352 | rq = this_rq(); | |
05fa785c | 353 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
354 | |
355 | return rq; | |
356 | } | |
357 | ||
8f4d37ec PZ |
358 | #ifdef CONFIG_SCHED_HRTICK |
359 | /* | |
360 | * Use HR-timers to deliver accurate preemption points. | |
361 | * | |
362 | * Its all a bit involved since we cannot program an hrt while holding the | |
363 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
364 | * reschedule event. | |
365 | * | |
366 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
367 | * rq->lock. | |
368 | */ | |
8f4d37ec | 369 | |
8f4d37ec PZ |
370 | static void hrtick_clear(struct rq *rq) |
371 | { | |
372 | if (hrtimer_active(&rq->hrtick_timer)) | |
373 | hrtimer_cancel(&rq->hrtick_timer); | |
374 | } | |
375 | ||
8f4d37ec PZ |
376 | /* |
377 | * High-resolution timer tick. | |
378 | * Runs from hardirq context with interrupts disabled. | |
379 | */ | |
380 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
381 | { | |
382 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
383 | ||
384 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
385 | ||
05fa785c | 386 | raw_spin_lock(&rq->lock); |
3e51f33f | 387 | update_rq_clock(rq); |
8f4d37ec | 388 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 389 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
390 | |
391 | return HRTIMER_NORESTART; | |
392 | } | |
393 | ||
95e904c7 | 394 | #ifdef CONFIG_SMP |
31656519 PZ |
395 | /* |
396 | * called from hardirq (IPI) context | |
397 | */ | |
398 | static void __hrtick_start(void *arg) | |
b328ca18 | 399 | { |
31656519 | 400 | struct rq *rq = arg; |
b328ca18 | 401 | |
05fa785c | 402 | raw_spin_lock(&rq->lock); |
31656519 PZ |
403 | hrtimer_restart(&rq->hrtick_timer); |
404 | rq->hrtick_csd_pending = 0; | |
05fa785c | 405 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
406 | } |
407 | ||
31656519 PZ |
408 | /* |
409 | * Called to set the hrtick timer state. | |
410 | * | |
411 | * called with rq->lock held and irqs disabled | |
412 | */ | |
029632fb | 413 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 414 | { |
31656519 PZ |
415 | struct hrtimer *timer = &rq->hrtick_timer; |
416 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 417 | |
cc584b21 | 418 | hrtimer_set_expires(timer, time); |
31656519 PZ |
419 | |
420 | if (rq == this_rq()) { | |
421 | hrtimer_restart(timer); | |
422 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 423 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
424 | rq->hrtick_csd_pending = 1; |
425 | } | |
b328ca18 PZ |
426 | } |
427 | ||
428 | static int | |
429 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
430 | { | |
431 | int cpu = (int)(long)hcpu; | |
432 | ||
433 | switch (action) { | |
434 | case CPU_UP_CANCELED: | |
435 | case CPU_UP_CANCELED_FROZEN: | |
436 | case CPU_DOWN_PREPARE: | |
437 | case CPU_DOWN_PREPARE_FROZEN: | |
438 | case CPU_DEAD: | |
439 | case CPU_DEAD_FROZEN: | |
31656519 | 440 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
441 | return NOTIFY_OK; |
442 | } | |
443 | ||
444 | return NOTIFY_DONE; | |
445 | } | |
446 | ||
fa748203 | 447 | static __init void init_hrtick(void) |
b328ca18 PZ |
448 | { |
449 | hotcpu_notifier(hotplug_hrtick, 0); | |
450 | } | |
31656519 PZ |
451 | #else |
452 | /* | |
453 | * Called to set the hrtick timer state. | |
454 | * | |
455 | * called with rq->lock held and irqs disabled | |
456 | */ | |
029632fb | 457 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 458 | { |
7f1e2ca9 | 459 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 460 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 461 | } |
b328ca18 | 462 | |
006c75f1 | 463 | static inline void init_hrtick(void) |
8f4d37ec | 464 | { |
8f4d37ec | 465 | } |
31656519 | 466 | #endif /* CONFIG_SMP */ |
8f4d37ec | 467 | |
31656519 | 468 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 469 | { |
31656519 PZ |
470 | #ifdef CONFIG_SMP |
471 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 472 | |
31656519 PZ |
473 | rq->hrtick_csd.flags = 0; |
474 | rq->hrtick_csd.func = __hrtick_start; | |
475 | rq->hrtick_csd.info = rq; | |
476 | #endif | |
8f4d37ec | 477 | |
31656519 PZ |
478 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
479 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 480 | } |
006c75f1 | 481 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
482 | static inline void hrtick_clear(struct rq *rq) |
483 | { | |
484 | } | |
485 | ||
8f4d37ec PZ |
486 | static inline void init_rq_hrtick(struct rq *rq) |
487 | { | |
488 | } | |
489 | ||
b328ca18 PZ |
490 | static inline void init_hrtick(void) |
491 | { | |
492 | } | |
006c75f1 | 493 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 494 | |
c24d20db IM |
495 | /* |
496 | * resched_task - mark a task 'to be rescheduled now'. | |
497 | * | |
498 | * On UP this means the setting of the need_resched flag, on SMP it | |
499 | * might also involve a cross-CPU call to trigger the scheduler on | |
500 | * the target CPU. | |
501 | */ | |
502 | #ifdef CONFIG_SMP | |
503 | ||
504 | #ifndef tsk_is_polling | |
505 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
506 | #endif | |
507 | ||
029632fb | 508 | void resched_task(struct task_struct *p) |
c24d20db IM |
509 | { |
510 | int cpu; | |
511 | ||
05fa785c | 512 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 513 | |
5ed0cec0 | 514 | if (test_tsk_need_resched(p)) |
c24d20db IM |
515 | return; |
516 | ||
5ed0cec0 | 517 | set_tsk_need_resched(p); |
c24d20db IM |
518 | |
519 | cpu = task_cpu(p); | |
520 | if (cpu == smp_processor_id()) | |
521 | return; | |
522 | ||
523 | /* NEED_RESCHED must be visible before we test polling */ | |
524 | smp_mb(); | |
525 | if (!tsk_is_polling(p)) | |
526 | smp_send_reschedule(cpu); | |
527 | } | |
528 | ||
029632fb | 529 | void resched_cpu(int cpu) |
c24d20db IM |
530 | { |
531 | struct rq *rq = cpu_rq(cpu); | |
532 | unsigned long flags; | |
533 | ||
05fa785c | 534 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
535 | return; |
536 | resched_task(cpu_curr(cpu)); | |
05fa785c | 537 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 538 | } |
06d8308c TG |
539 | |
540 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
541 | /* |
542 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
543 | * from an idle cpu. This is good for power-savings. | |
544 | * | |
545 | * We don't do similar optimization for completely idle system, as | |
546 | * selecting an idle cpu will add more delays to the timers than intended | |
547 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
548 | */ | |
549 | int get_nohz_timer_target(void) | |
550 | { | |
551 | int cpu = smp_processor_id(); | |
552 | int i; | |
553 | struct sched_domain *sd; | |
554 | ||
057f3fad | 555 | rcu_read_lock(); |
83cd4fe2 | 556 | for_each_domain(cpu, sd) { |
057f3fad PZ |
557 | for_each_cpu(i, sched_domain_span(sd)) { |
558 | if (!idle_cpu(i)) { | |
559 | cpu = i; | |
560 | goto unlock; | |
561 | } | |
562 | } | |
83cd4fe2 | 563 | } |
057f3fad PZ |
564 | unlock: |
565 | rcu_read_unlock(); | |
83cd4fe2 VP |
566 | return cpu; |
567 | } | |
06d8308c TG |
568 | /* |
569 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
570 | * idle CPU then this timer might expire before the next timer event | |
571 | * which is scheduled to wake up that CPU. In case of a completely | |
572 | * idle system the next event might even be infinite time into the | |
573 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
574 | * leaves the inner idle loop so the newly added timer is taken into | |
575 | * account when the CPU goes back to idle and evaluates the timer | |
576 | * wheel for the next timer event. | |
577 | */ | |
578 | void wake_up_idle_cpu(int cpu) | |
579 | { | |
580 | struct rq *rq = cpu_rq(cpu); | |
581 | ||
582 | if (cpu == smp_processor_id()) | |
583 | return; | |
584 | ||
585 | /* | |
586 | * This is safe, as this function is called with the timer | |
587 | * wheel base lock of (cpu) held. When the CPU is on the way | |
588 | * to idle and has not yet set rq->curr to idle then it will | |
589 | * be serialized on the timer wheel base lock and take the new | |
590 | * timer into account automatically. | |
591 | */ | |
592 | if (rq->curr != rq->idle) | |
593 | return; | |
45bf76df | 594 | |
45bf76df | 595 | /* |
06d8308c TG |
596 | * We can set TIF_RESCHED on the idle task of the other CPU |
597 | * lockless. The worst case is that the other CPU runs the | |
598 | * idle task through an additional NOOP schedule() | |
45bf76df | 599 | */ |
5ed0cec0 | 600 | set_tsk_need_resched(rq->idle); |
45bf76df | 601 | |
06d8308c TG |
602 | /* NEED_RESCHED must be visible before we test polling */ |
603 | smp_mb(); | |
604 | if (!tsk_is_polling(rq->idle)) | |
605 | smp_send_reschedule(cpu); | |
45bf76df IM |
606 | } |
607 | ||
ca38062e | 608 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 609 | { |
1c792db7 SS |
610 | int cpu = smp_processor_id(); |
611 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
612 | } |
613 | ||
ca38062e | 614 | #else /* CONFIG_NO_HZ */ |
45bf76df | 615 | |
ca38062e | 616 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 617 | { |
ca38062e | 618 | return false; |
2069dd75 PZ |
619 | } |
620 | ||
6d6bc0ad | 621 | #endif /* CONFIG_NO_HZ */ |
d842de87 | 622 | |
029632fb | 623 | void sched_avg_update(struct rq *rq) |
18d95a28 | 624 | { |
e9e9250b PZ |
625 | s64 period = sched_avg_period(); |
626 | ||
627 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
628 | /* |
629 | * Inline assembly required to prevent the compiler | |
630 | * optimising this loop into a divmod call. | |
631 | * See __iter_div_u64_rem() for another example of this. | |
632 | */ | |
633 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
634 | rq->age_stamp += period; |
635 | rq->rt_avg /= 2; | |
636 | } | |
18d95a28 PZ |
637 | } |
638 | ||
6d6bc0ad | 639 | #else /* !CONFIG_SMP */ |
029632fb | 640 | void resched_task(struct task_struct *p) |
18d95a28 | 641 | { |
05fa785c | 642 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 643 | set_tsk_need_resched(p); |
18d95a28 | 644 | } |
6d6bc0ad | 645 | #endif /* CONFIG_SMP */ |
18d95a28 | 646 | |
a790de99 PT |
647 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
648 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 649 | /* |
8277434e PT |
650 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
651 | * node and @up when leaving it for the final time. | |
652 | * | |
653 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 654 | */ |
029632fb | 655 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 656 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
657 | { |
658 | struct task_group *parent, *child; | |
eb755805 | 659 | int ret; |
c09595f6 | 660 | |
8277434e PT |
661 | parent = from; |
662 | ||
c09595f6 | 663 | down: |
eb755805 PZ |
664 | ret = (*down)(parent, data); |
665 | if (ret) | |
8277434e | 666 | goto out; |
c09595f6 PZ |
667 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
668 | parent = child; | |
669 | goto down; | |
670 | ||
671 | up: | |
672 | continue; | |
673 | } | |
eb755805 | 674 | ret = (*up)(parent, data); |
8277434e PT |
675 | if (ret || parent == from) |
676 | goto out; | |
c09595f6 PZ |
677 | |
678 | child = parent; | |
679 | parent = parent->parent; | |
680 | if (parent) | |
681 | goto up; | |
8277434e | 682 | out: |
eb755805 | 683 | return ret; |
c09595f6 PZ |
684 | } |
685 | ||
029632fb | 686 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 687 | { |
e2b245f8 | 688 | return 0; |
eb755805 | 689 | } |
18d95a28 PZ |
690 | #endif |
691 | ||
029632fb | 692 | void update_cpu_load(struct rq *this_rq); |
9c217245 | 693 | |
45bf76df IM |
694 | static void set_load_weight(struct task_struct *p) |
695 | { | |
f05998d4 NR |
696 | int prio = p->static_prio - MAX_RT_PRIO; |
697 | struct load_weight *load = &p->se.load; | |
698 | ||
dd41f596 IM |
699 | /* |
700 | * SCHED_IDLE tasks get minimal weight: | |
701 | */ | |
702 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 703 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 704 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
705 | return; |
706 | } | |
71f8bd46 | 707 | |
c8b28116 | 708 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 709 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
710 | } |
711 | ||
371fd7e7 | 712 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 713 | { |
a64692a3 | 714 | update_rq_clock(rq); |
dd41f596 | 715 | sched_info_queued(p); |
371fd7e7 | 716 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
717 | } |
718 | ||
371fd7e7 | 719 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 720 | { |
a64692a3 | 721 | update_rq_clock(rq); |
46ac22ba | 722 | sched_info_dequeued(p); |
371fd7e7 | 723 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
724 | } |
725 | ||
1e3c88bd PZ |
726 | /* |
727 | * activate_task - move a task to the runqueue. | |
728 | */ | |
029632fb | 729 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
730 | { |
731 | if (task_contributes_to_load(p)) | |
732 | rq->nr_uninterruptible--; | |
733 | ||
371fd7e7 | 734 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
735 | } |
736 | ||
737 | /* | |
738 | * deactivate_task - remove a task from the runqueue. | |
739 | */ | |
029632fb | 740 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
741 | { |
742 | if (task_contributes_to_load(p)) | |
743 | rq->nr_uninterruptible++; | |
744 | ||
371fd7e7 | 745 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
746 | } |
747 | ||
b52bfee4 VP |
748 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
749 | ||
305e6835 VP |
750 | /* |
751 | * There are no locks covering percpu hardirq/softirq time. | |
752 | * They are only modified in account_system_vtime, on corresponding CPU | |
753 | * with interrupts disabled. So, writes are safe. | |
754 | * They are read and saved off onto struct rq in update_rq_clock(). | |
755 | * This may result in other CPU reading this CPU's irq time and can | |
756 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
757 | * or new value with a side effect of accounting a slice of irq time to wrong |
758 | * task when irq is in progress while we read rq->clock. That is a worthy | |
759 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 760 | */ |
b52bfee4 VP |
761 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
762 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
763 | ||
764 | static DEFINE_PER_CPU(u64, irq_start_time); | |
765 | static int sched_clock_irqtime; | |
766 | ||
767 | void enable_sched_clock_irqtime(void) | |
768 | { | |
769 | sched_clock_irqtime = 1; | |
770 | } | |
771 | ||
772 | void disable_sched_clock_irqtime(void) | |
773 | { | |
774 | sched_clock_irqtime = 0; | |
775 | } | |
776 | ||
8e92c201 PZ |
777 | #ifndef CONFIG_64BIT |
778 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
779 | ||
780 | static inline void irq_time_write_begin(void) | |
781 | { | |
782 | __this_cpu_inc(irq_time_seq.sequence); | |
783 | smp_wmb(); | |
784 | } | |
785 | ||
786 | static inline void irq_time_write_end(void) | |
787 | { | |
788 | smp_wmb(); | |
789 | __this_cpu_inc(irq_time_seq.sequence); | |
790 | } | |
791 | ||
792 | static inline u64 irq_time_read(int cpu) | |
793 | { | |
794 | u64 irq_time; | |
795 | unsigned seq; | |
796 | ||
797 | do { | |
798 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
799 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
800 | per_cpu(cpu_hardirq_time, cpu); | |
801 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
802 | ||
803 | return irq_time; | |
804 | } | |
805 | #else /* CONFIG_64BIT */ | |
806 | static inline void irq_time_write_begin(void) | |
807 | { | |
808 | } | |
809 | ||
810 | static inline void irq_time_write_end(void) | |
811 | { | |
812 | } | |
813 | ||
814 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 815 | { |
305e6835 VP |
816 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
817 | } | |
8e92c201 | 818 | #endif /* CONFIG_64BIT */ |
305e6835 | 819 | |
fe44d621 PZ |
820 | /* |
821 | * Called before incrementing preempt_count on {soft,}irq_enter | |
822 | * and before decrementing preempt_count on {soft,}irq_exit. | |
823 | */ | |
b52bfee4 VP |
824 | void account_system_vtime(struct task_struct *curr) |
825 | { | |
826 | unsigned long flags; | |
fe44d621 | 827 | s64 delta; |
b52bfee4 | 828 | int cpu; |
b52bfee4 VP |
829 | |
830 | if (!sched_clock_irqtime) | |
831 | return; | |
832 | ||
833 | local_irq_save(flags); | |
834 | ||
b52bfee4 | 835 | cpu = smp_processor_id(); |
fe44d621 PZ |
836 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
837 | __this_cpu_add(irq_start_time, delta); | |
838 | ||
8e92c201 | 839 | irq_time_write_begin(); |
b52bfee4 VP |
840 | /* |
841 | * We do not account for softirq time from ksoftirqd here. | |
842 | * We want to continue accounting softirq time to ksoftirqd thread | |
843 | * in that case, so as not to confuse scheduler with a special task | |
844 | * that do not consume any time, but still wants to run. | |
845 | */ | |
846 | if (hardirq_count()) | |
fe44d621 | 847 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 848 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 849 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 850 | |
8e92c201 | 851 | irq_time_write_end(); |
b52bfee4 VP |
852 | local_irq_restore(flags); |
853 | } | |
b7dadc38 | 854 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 855 | |
e6e6685a GC |
856 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
857 | ||
858 | #ifdef CONFIG_PARAVIRT | |
859 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 860 | { |
e6e6685a GC |
861 | if (unlikely(steal > NSEC_PER_SEC)) |
862 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 863 | |
e6e6685a GC |
864 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
865 | } | |
866 | #endif | |
867 | ||
fe44d621 | 868 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 869 | { |
095c0aa8 GC |
870 | /* |
871 | * In theory, the compile should just see 0 here, and optimize out the call | |
872 | * to sched_rt_avg_update. But I don't trust it... | |
873 | */ | |
874 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
875 | s64 steal = 0, irq_delta = 0; | |
876 | #endif | |
877 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 878 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
879 | |
880 | /* | |
881 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
882 | * this case when a previous update_rq_clock() happened inside a | |
883 | * {soft,}irq region. | |
884 | * | |
885 | * When this happens, we stop ->clock_task and only update the | |
886 | * prev_irq_time stamp to account for the part that fit, so that a next | |
887 | * update will consume the rest. This ensures ->clock_task is | |
888 | * monotonic. | |
889 | * | |
890 | * It does however cause some slight miss-attribution of {soft,}irq | |
891 | * time, a more accurate solution would be to update the irq_time using | |
892 | * the current rq->clock timestamp, except that would require using | |
893 | * atomic ops. | |
894 | */ | |
895 | if (irq_delta > delta) | |
896 | irq_delta = delta; | |
897 | ||
898 | rq->prev_irq_time += irq_delta; | |
899 | delta -= irq_delta; | |
095c0aa8 GC |
900 | #endif |
901 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
902 | if (static_branch((¶virt_steal_rq_enabled))) { | |
903 | u64 st; | |
904 | ||
905 | steal = paravirt_steal_clock(cpu_of(rq)); | |
906 | steal -= rq->prev_steal_time_rq; | |
907 | ||
908 | if (unlikely(steal > delta)) | |
909 | steal = delta; | |
910 | ||
911 | st = steal_ticks(steal); | |
912 | steal = st * TICK_NSEC; | |
913 | ||
914 | rq->prev_steal_time_rq += steal; | |
915 | ||
916 | delta -= steal; | |
917 | } | |
918 | #endif | |
919 | ||
fe44d621 PZ |
920 | rq->clock_task += delta; |
921 | ||
095c0aa8 GC |
922 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
923 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
924 | sched_rt_avg_update(rq, irq_delta + steal); | |
925 | #endif | |
aa483808 VP |
926 | } |
927 | ||
095c0aa8 | 928 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
929 | static int irqtime_account_hi_update(void) |
930 | { | |
3292beb3 | 931 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
932 | unsigned long flags; |
933 | u64 latest_ns; | |
934 | int ret = 0; | |
935 | ||
936 | local_irq_save(flags); | |
937 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
612ef28a | 938 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) |
abb74cef VP |
939 | ret = 1; |
940 | local_irq_restore(flags); | |
941 | return ret; | |
942 | } | |
943 | ||
944 | static int irqtime_account_si_update(void) | |
945 | { | |
3292beb3 | 946 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
947 | unsigned long flags; |
948 | u64 latest_ns; | |
949 | int ret = 0; | |
950 | ||
951 | local_irq_save(flags); | |
952 | latest_ns = this_cpu_read(cpu_softirq_time); | |
612ef28a | 953 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) |
abb74cef VP |
954 | ret = 1; |
955 | local_irq_restore(flags); | |
956 | return ret; | |
957 | } | |
958 | ||
fe44d621 | 959 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 960 | |
abb74cef VP |
961 | #define sched_clock_irqtime (0) |
962 | ||
095c0aa8 | 963 | #endif |
b52bfee4 | 964 | |
34f971f6 PZ |
965 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
966 | { | |
967 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
968 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
969 | ||
970 | if (stop) { | |
971 | /* | |
972 | * Make it appear like a SCHED_FIFO task, its something | |
973 | * userspace knows about and won't get confused about. | |
974 | * | |
975 | * Also, it will make PI more or less work without too | |
976 | * much confusion -- but then, stop work should not | |
977 | * rely on PI working anyway. | |
978 | */ | |
979 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
980 | ||
981 | stop->sched_class = &stop_sched_class; | |
982 | } | |
983 | ||
984 | cpu_rq(cpu)->stop = stop; | |
985 | ||
986 | if (old_stop) { | |
987 | /* | |
988 | * Reset it back to a normal scheduling class so that | |
989 | * it can die in pieces. | |
990 | */ | |
991 | old_stop->sched_class = &rt_sched_class; | |
992 | } | |
993 | } | |
994 | ||
14531189 | 995 | /* |
dd41f596 | 996 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 997 | */ |
14531189 IM |
998 | static inline int __normal_prio(struct task_struct *p) |
999 | { | |
dd41f596 | 1000 | return p->static_prio; |
14531189 IM |
1001 | } |
1002 | ||
b29739f9 IM |
1003 | /* |
1004 | * Calculate the expected normal priority: i.e. priority | |
1005 | * without taking RT-inheritance into account. Might be | |
1006 | * boosted by interactivity modifiers. Changes upon fork, | |
1007 | * setprio syscalls, and whenever the interactivity | |
1008 | * estimator recalculates. | |
1009 | */ | |
36c8b586 | 1010 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1011 | { |
1012 | int prio; | |
1013 | ||
e05606d3 | 1014 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1015 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1016 | else | |
1017 | prio = __normal_prio(p); | |
1018 | return prio; | |
1019 | } | |
1020 | ||
1021 | /* | |
1022 | * Calculate the current priority, i.e. the priority | |
1023 | * taken into account by the scheduler. This value might | |
1024 | * be boosted by RT tasks, or might be boosted by | |
1025 | * interactivity modifiers. Will be RT if the task got | |
1026 | * RT-boosted. If not then it returns p->normal_prio. | |
1027 | */ | |
36c8b586 | 1028 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1029 | { |
1030 | p->normal_prio = normal_prio(p); | |
1031 | /* | |
1032 | * If we are RT tasks or we were boosted to RT priority, | |
1033 | * keep the priority unchanged. Otherwise, update priority | |
1034 | * to the normal priority: | |
1035 | */ | |
1036 | if (!rt_prio(p->prio)) | |
1037 | return p->normal_prio; | |
1038 | return p->prio; | |
1039 | } | |
1040 | ||
1da177e4 LT |
1041 | /** |
1042 | * task_curr - is this task currently executing on a CPU? | |
1043 | * @p: the task in question. | |
1044 | */ | |
36c8b586 | 1045 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1046 | { |
1047 | return cpu_curr(task_cpu(p)) == p; | |
1048 | } | |
1049 | ||
cb469845 SR |
1050 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1051 | const struct sched_class *prev_class, | |
da7a735e | 1052 | int oldprio) |
cb469845 SR |
1053 | { |
1054 | if (prev_class != p->sched_class) { | |
1055 | if (prev_class->switched_from) | |
da7a735e PZ |
1056 | prev_class->switched_from(rq, p); |
1057 | p->sched_class->switched_to(rq, p); | |
1058 | } else if (oldprio != p->prio) | |
1059 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
1060 | } |
1061 | ||
029632fb | 1062 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1063 | { |
1064 | const struct sched_class *class; | |
1065 | ||
1066 | if (p->sched_class == rq->curr->sched_class) { | |
1067 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1068 | } else { | |
1069 | for_each_class(class) { | |
1070 | if (class == rq->curr->sched_class) | |
1071 | break; | |
1072 | if (class == p->sched_class) { | |
1073 | resched_task(rq->curr); | |
1074 | break; | |
1075 | } | |
1076 | } | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * A queue event has occurred, and we're going to schedule. In | |
1081 | * this case, we can save a useless back to back clock update. | |
1082 | */ | |
fd2f4419 | 1083 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1084 | rq->skip_clock_update = 1; |
1085 | } | |
1086 | ||
1da177e4 | 1087 | #ifdef CONFIG_SMP |
dd41f596 | 1088 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1089 | { |
e2912009 PZ |
1090 | #ifdef CONFIG_SCHED_DEBUG |
1091 | /* | |
1092 | * We should never call set_task_cpu() on a blocked task, | |
1093 | * ttwu() will sort out the placement. | |
1094 | */ | |
077614ee PZ |
1095 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
1096 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
1097 | |
1098 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1099 | /* |
1100 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1101 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1102 | * | |
1103 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
1104 | * see set_task_rq(). | |
1105 | * | |
1106 | * Furthermore, all task_rq users should acquire both locks, see | |
1107 | * task_rq_lock(). | |
1108 | */ | |
0122ec5b PZ |
1109 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1110 | lockdep_is_held(&task_rq(p)->lock))); | |
1111 | #endif | |
e2912009 PZ |
1112 | #endif |
1113 | ||
de1d7286 | 1114 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1115 | |
0c69774e PZ |
1116 | if (task_cpu(p) != new_cpu) { |
1117 | p->se.nr_migrations++; | |
a8b0ca17 | 1118 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1119 | } |
dd41f596 IM |
1120 | |
1121 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1122 | } |
1123 | ||
969c7921 | 1124 | struct migration_arg { |
36c8b586 | 1125 | struct task_struct *task; |
1da177e4 | 1126 | int dest_cpu; |
70b97a7f | 1127 | }; |
1da177e4 | 1128 | |
969c7921 TH |
1129 | static int migration_cpu_stop(void *data); |
1130 | ||
1da177e4 LT |
1131 | /* |
1132 | * wait_task_inactive - wait for a thread to unschedule. | |
1133 | * | |
85ba2d86 RM |
1134 | * If @match_state is nonzero, it's the @p->state value just checked and |
1135 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1136 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1137 | * we return a positive number (its total switch count). If a second call | |
1138 | * a short while later returns the same number, the caller can be sure that | |
1139 | * @p has remained unscheduled the whole time. | |
1140 | * | |
1da177e4 LT |
1141 | * The caller must ensure that the task *will* unschedule sometime soon, |
1142 | * else this function might spin for a *long* time. This function can't | |
1143 | * be called with interrupts off, or it may introduce deadlock with | |
1144 | * smp_call_function() if an IPI is sent by the same process we are | |
1145 | * waiting to become inactive. | |
1146 | */ | |
85ba2d86 | 1147 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1148 | { |
1149 | unsigned long flags; | |
dd41f596 | 1150 | int running, on_rq; |
85ba2d86 | 1151 | unsigned long ncsw; |
70b97a7f | 1152 | struct rq *rq; |
1da177e4 | 1153 | |
3a5c359a AK |
1154 | for (;;) { |
1155 | /* | |
1156 | * We do the initial early heuristics without holding | |
1157 | * any task-queue locks at all. We'll only try to get | |
1158 | * the runqueue lock when things look like they will | |
1159 | * work out! | |
1160 | */ | |
1161 | rq = task_rq(p); | |
fa490cfd | 1162 | |
3a5c359a AK |
1163 | /* |
1164 | * If the task is actively running on another CPU | |
1165 | * still, just relax and busy-wait without holding | |
1166 | * any locks. | |
1167 | * | |
1168 | * NOTE! Since we don't hold any locks, it's not | |
1169 | * even sure that "rq" stays as the right runqueue! | |
1170 | * But we don't care, since "task_running()" will | |
1171 | * return false if the runqueue has changed and p | |
1172 | * is actually now running somewhere else! | |
1173 | */ | |
85ba2d86 RM |
1174 | while (task_running(rq, p)) { |
1175 | if (match_state && unlikely(p->state != match_state)) | |
1176 | return 0; | |
3a5c359a | 1177 | cpu_relax(); |
85ba2d86 | 1178 | } |
fa490cfd | 1179 | |
3a5c359a AK |
1180 | /* |
1181 | * Ok, time to look more closely! We need the rq | |
1182 | * lock now, to be *sure*. If we're wrong, we'll | |
1183 | * just go back and repeat. | |
1184 | */ | |
1185 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1186 | trace_sched_wait_task(p); |
3a5c359a | 1187 | running = task_running(rq, p); |
fd2f4419 | 1188 | on_rq = p->on_rq; |
85ba2d86 | 1189 | ncsw = 0; |
f31e11d8 | 1190 | if (!match_state || p->state == match_state) |
93dcf55f | 1191 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1192 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1193 | |
85ba2d86 RM |
1194 | /* |
1195 | * If it changed from the expected state, bail out now. | |
1196 | */ | |
1197 | if (unlikely(!ncsw)) | |
1198 | break; | |
1199 | ||
3a5c359a AK |
1200 | /* |
1201 | * Was it really running after all now that we | |
1202 | * checked with the proper locks actually held? | |
1203 | * | |
1204 | * Oops. Go back and try again.. | |
1205 | */ | |
1206 | if (unlikely(running)) { | |
1207 | cpu_relax(); | |
1208 | continue; | |
1209 | } | |
fa490cfd | 1210 | |
3a5c359a AK |
1211 | /* |
1212 | * It's not enough that it's not actively running, | |
1213 | * it must be off the runqueue _entirely_, and not | |
1214 | * preempted! | |
1215 | * | |
80dd99b3 | 1216 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1217 | * running right now), it's preempted, and we should |
1218 | * yield - it could be a while. | |
1219 | */ | |
1220 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1221 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1222 | ||
1223 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1224 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1225 | continue; |
1226 | } | |
fa490cfd | 1227 | |
3a5c359a AK |
1228 | /* |
1229 | * Ahh, all good. It wasn't running, and it wasn't | |
1230 | * runnable, which means that it will never become | |
1231 | * running in the future either. We're all done! | |
1232 | */ | |
1233 | break; | |
1234 | } | |
85ba2d86 RM |
1235 | |
1236 | return ncsw; | |
1da177e4 LT |
1237 | } |
1238 | ||
1239 | /*** | |
1240 | * kick_process - kick a running thread to enter/exit the kernel | |
1241 | * @p: the to-be-kicked thread | |
1242 | * | |
1243 | * Cause a process which is running on another CPU to enter | |
1244 | * kernel-mode, without any delay. (to get signals handled.) | |
1245 | * | |
25985edc | 1246 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1247 | * because all it wants to ensure is that the remote task enters |
1248 | * the kernel. If the IPI races and the task has been migrated | |
1249 | * to another CPU then no harm is done and the purpose has been | |
1250 | * achieved as well. | |
1251 | */ | |
36c8b586 | 1252 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1253 | { |
1254 | int cpu; | |
1255 | ||
1256 | preempt_disable(); | |
1257 | cpu = task_cpu(p); | |
1258 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1259 | smp_send_reschedule(cpu); | |
1260 | preempt_enable(); | |
1261 | } | |
b43e3521 | 1262 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1263 | #endif /* CONFIG_SMP */ |
1da177e4 | 1264 | |
970b13ba | 1265 | #ifdef CONFIG_SMP |
30da688e | 1266 | /* |
013fdb80 | 1267 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1268 | */ |
5da9a0fb PZ |
1269 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1270 | { | |
1271 | int dest_cpu; | |
1272 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
1273 | ||
1274 | /* Look for allowed, online CPU in same node. */ | |
1275 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
fa17b507 | 1276 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb PZ |
1277 | return dest_cpu; |
1278 | ||
1279 | /* Any allowed, online CPU? */ | |
fa17b507 | 1280 | dest_cpu = cpumask_any_and(tsk_cpus_allowed(p), cpu_active_mask); |
5da9a0fb PZ |
1281 | if (dest_cpu < nr_cpu_ids) |
1282 | return dest_cpu; | |
1283 | ||
1284 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
1285 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
1286 | /* | |
1287 | * Don't tell them about moving exiting tasks or | |
1288 | * kernel threads (both mm NULL), since they never | |
1289 | * leave kernel. | |
1290 | */ | |
1291 | if (p->mm && printk_ratelimit()) { | |
1292 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
1293 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
1294 | } |
1295 | ||
1296 | return dest_cpu; | |
1297 | } | |
1298 | ||
e2912009 | 1299 | /* |
013fdb80 | 1300 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1301 | */ |
970b13ba | 1302 | static inline |
7608dec2 | 1303 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1304 | { |
7608dec2 | 1305 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1306 | |
1307 | /* | |
1308 | * In order not to call set_task_cpu() on a blocking task we need | |
1309 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1310 | * cpu. | |
1311 | * | |
1312 | * Since this is common to all placement strategies, this lives here. | |
1313 | * | |
1314 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1315 | * not worry about this generic constraint ] | |
1316 | */ | |
fa17b507 | 1317 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1318 | !cpu_online(cpu))) |
5da9a0fb | 1319 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1320 | |
1321 | return cpu; | |
970b13ba | 1322 | } |
09a40af5 MG |
1323 | |
1324 | static void update_avg(u64 *avg, u64 sample) | |
1325 | { | |
1326 | s64 diff = sample - *avg; | |
1327 | *avg += diff >> 3; | |
1328 | } | |
970b13ba PZ |
1329 | #endif |
1330 | ||
d7c01d27 | 1331 | static void |
b84cb5df | 1332 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1333 | { |
d7c01d27 | 1334 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1335 | struct rq *rq = this_rq(); |
1336 | ||
d7c01d27 PZ |
1337 | #ifdef CONFIG_SMP |
1338 | int this_cpu = smp_processor_id(); | |
1339 | ||
1340 | if (cpu == this_cpu) { | |
1341 | schedstat_inc(rq, ttwu_local); | |
1342 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1343 | } else { | |
1344 | struct sched_domain *sd; | |
1345 | ||
1346 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1347 | rcu_read_lock(); |
d7c01d27 PZ |
1348 | for_each_domain(this_cpu, sd) { |
1349 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1350 | schedstat_inc(sd, ttwu_wake_remote); | |
1351 | break; | |
1352 | } | |
1353 | } | |
057f3fad | 1354 | rcu_read_unlock(); |
d7c01d27 | 1355 | } |
f339b9dc PZ |
1356 | |
1357 | if (wake_flags & WF_MIGRATED) | |
1358 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1359 | ||
d7c01d27 PZ |
1360 | #endif /* CONFIG_SMP */ |
1361 | ||
1362 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1363 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1364 | |
1365 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1366 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1367 | |
d7c01d27 PZ |
1368 | #endif /* CONFIG_SCHEDSTATS */ |
1369 | } | |
1370 | ||
1371 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1372 | { | |
9ed3811a | 1373 | activate_task(rq, p, en_flags); |
fd2f4419 | 1374 | p->on_rq = 1; |
c2f7115e PZ |
1375 | |
1376 | /* if a worker is waking up, notify workqueue */ | |
1377 | if (p->flags & PF_WQ_WORKER) | |
1378 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1379 | } |
1380 | ||
23f41eeb PZ |
1381 | /* |
1382 | * Mark the task runnable and perform wakeup-preemption. | |
1383 | */ | |
89363381 | 1384 | static void |
23f41eeb | 1385 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1386 | { |
89363381 | 1387 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1388 | check_preempt_curr(rq, p, wake_flags); |
1389 | ||
1390 | p->state = TASK_RUNNING; | |
1391 | #ifdef CONFIG_SMP | |
1392 | if (p->sched_class->task_woken) | |
1393 | p->sched_class->task_woken(rq, p); | |
1394 | ||
e69c6341 | 1395 | if (rq->idle_stamp) { |
9ed3811a TH |
1396 | u64 delta = rq->clock - rq->idle_stamp; |
1397 | u64 max = 2*sysctl_sched_migration_cost; | |
1398 | ||
1399 | if (delta > max) | |
1400 | rq->avg_idle = max; | |
1401 | else | |
1402 | update_avg(&rq->avg_idle, delta); | |
1403 | rq->idle_stamp = 0; | |
1404 | } | |
1405 | #endif | |
1406 | } | |
1407 | ||
c05fbafb PZ |
1408 | static void |
1409 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1410 | { | |
1411 | #ifdef CONFIG_SMP | |
1412 | if (p->sched_contributes_to_load) | |
1413 | rq->nr_uninterruptible--; | |
1414 | #endif | |
1415 | ||
1416 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1417 | ttwu_do_wakeup(rq, p, wake_flags); | |
1418 | } | |
1419 | ||
1420 | /* | |
1421 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1422 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1423 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1424 | * the task is still ->on_rq. | |
1425 | */ | |
1426 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1427 | { | |
1428 | struct rq *rq; | |
1429 | int ret = 0; | |
1430 | ||
1431 | rq = __task_rq_lock(p); | |
1432 | if (p->on_rq) { | |
1433 | ttwu_do_wakeup(rq, p, wake_flags); | |
1434 | ret = 1; | |
1435 | } | |
1436 | __task_rq_unlock(rq); | |
1437 | ||
1438 | return ret; | |
1439 | } | |
1440 | ||
317f3941 | 1441 | #ifdef CONFIG_SMP |
fa14ff4a | 1442 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1443 | { |
1444 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1445 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1446 | struct task_struct *p; | |
317f3941 PZ |
1447 | |
1448 | raw_spin_lock(&rq->lock); | |
1449 | ||
fa14ff4a PZ |
1450 | while (llist) { |
1451 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1452 | llist = llist_next(llist); | |
317f3941 PZ |
1453 | ttwu_do_activate(rq, p, 0); |
1454 | } | |
1455 | ||
1456 | raw_spin_unlock(&rq->lock); | |
1457 | } | |
1458 | ||
1459 | void scheduler_ipi(void) | |
1460 | { | |
ca38062e | 1461 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1462 | return; |
1463 | ||
1464 | /* | |
1465 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1466 | * traditionally all their work was done from the interrupt return | |
1467 | * path. Now that we actually do some work, we need to make sure | |
1468 | * we do call them. | |
1469 | * | |
1470 | * Some archs already do call them, luckily irq_enter/exit nest | |
1471 | * properly. | |
1472 | * | |
1473 | * Arguably we should visit all archs and update all handlers, | |
1474 | * however a fair share of IPIs are still resched only so this would | |
1475 | * somewhat pessimize the simple resched case. | |
1476 | */ | |
1477 | irq_enter(); | |
fa14ff4a | 1478 | sched_ttwu_pending(); |
ca38062e SS |
1479 | |
1480 | /* | |
1481 | * Check if someone kicked us for doing the nohz idle load balance. | |
1482 | */ | |
6eb57e0d SS |
1483 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1484 | this_rq()->idle_balance = 1; | |
ca38062e | 1485 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1486 | } |
c5d753a5 | 1487 | irq_exit(); |
317f3941 PZ |
1488 | } |
1489 | ||
1490 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1491 | { | |
fa14ff4a | 1492 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1493 | smp_send_reschedule(cpu); |
1494 | } | |
d6aa8f85 PZ |
1495 | |
1496 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1497 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
1498 | { | |
1499 | struct rq *rq; | |
1500 | int ret = 0; | |
1501 | ||
1502 | rq = __task_rq_lock(p); | |
1503 | if (p->on_cpu) { | |
1504 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
1505 | ttwu_do_wakeup(rq, p, wake_flags); | |
1506 | ret = 1; | |
1507 | } | |
1508 | __task_rq_unlock(rq); | |
1509 | ||
1510 | return ret; | |
1511 | ||
1512 | } | |
1513 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
518cd623 PZ |
1514 | |
1515 | static inline int ttwu_share_cache(int this_cpu, int that_cpu) | |
1516 | { | |
1517 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1518 | } | |
d6aa8f85 | 1519 | #endif /* CONFIG_SMP */ |
317f3941 | 1520 | |
c05fbafb PZ |
1521 | static void ttwu_queue(struct task_struct *p, int cpu) |
1522 | { | |
1523 | struct rq *rq = cpu_rq(cpu); | |
1524 | ||
17d9f311 | 1525 | #if defined(CONFIG_SMP) |
518cd623 | 1526 | if (sched_feat(TTWU_QUEUE) && !ttwu_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1527 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1528 | ttwu_queue_remote(p, cpu); |
1529 | return; | |
1530 | } | |
1531 | #endif | |
1532 | ||
c05fbafb PZ |
1533 | raw_spin_lock(&rq->lock); |
1534 | ttwu_do_activate(rq, p, 0); | |
1535 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1536 | } |
1537 | ||
1538 | /** | |
1da177e4 | 1539 | * try_to_wake_up - wake up a thread |
9ed3811a | 1540 | * @p: the thread to be awakened |
1da177e4 | 1541 | * @state: the mask of task states that can be woken |
9ed3811a | 1542 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1543 | * |
1544 | * Put it on the run-queue if it's not already there. The "current" | |
1545 | * thread is always on the run-queue (except when the actual | |
1546 | * re-schedule is in progress), and as such you're allowed to do | |
1547 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1548 | * runnable without the overhead of this. | |
1549 | * | |
9ed3811a TH |
1550 | * Returns %true if @p was woken up, %false if it was already running |
1551 | * or @state didn't match @p's state. | |
1da177e4 | 1552 | */ |
e4a52bcb PZ |
1553 | static int |
1554 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1555 | { |
1da177e4 | 1556 | unsigned long flags; |
c05fbafb | 1557 | int cpu, success = 0; |
2398f2c6 | 1558 | |
04e2f174 | 1559 | smp_wmb(); |
013fdb80 | 1560 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1561 | if (!(p->state & state)) |
1da177e4 LT |
1562 | goto out; |
1563 | ||
c05fbafb | 1564 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1565 | cpu = task_cpu(p); |
1da177e4 | 1566 | |
c05fbafb PZ |
1567 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1568 | goto stat; | |
1da177e4 | 1569 | |
1da177e4 | 1570 | #ifdef CONFIG_SMP |
e9c84311 | 1571 | /* |
c05fbafb PZ |
1572 | * If the owning (remote) cpu is still in the middle of schedule() with |
1573 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1574 | */ |
e4a52bcb PZ |
1575 | while (p->on_cpu) { |
1576 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1577 | /* | |
d6aa8f85 PZ |
1578 | * In case the architecture enables interrupts in |
1579 | * context_switch(), we cannot busy wait, since that | |
1580 | * would lead to deadlocks when an interrupt hits and | |
1581 | * tries to wake up @prev. So bail and do a complete | |
1582 | * remote wakeup. | |
e4a52bcb | 1583 | */ |
d6aa8f85 | 1584 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 1585 | goto stat; |
d6aa8f85 | 1586 | #else |
e4a52bcb | 1587 | cpu_relax(); |
d6aa8f85 | 1588 | #endif |
371fd7e7 | 1589 | } |
0970d299 | 1590 | /* |
e4a52bcb | 1591 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1592 | */ |
e4a52bcb | 1593 | smp_rmb(); |
1da177e4 | 1594 | |
a8e4f2ea | 1595 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1596 | p->state = TASK_WAKING; |
e7693a36 | 1597 | |
e4a52bcb | 1598 | if (p->sched_class->task_waking) |
74f8e4b2 | 1599 | p->sched_class->task_waking(p); |
efbbd05a | 1600 | |
7608dec2 | 1601 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1602 | if (task_cpu(p) != cpu) { |
1603 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1604 | set_task_cpu(p, cpu); |
f339b9dc | 1605 | } |
1da177e4 | 1606 | #endif /* CONFIG_SMP */ |
1da177e4 | 1607 | |
c05fbafb PZ |
1608 | ttwu_queue(p, cpu); |
1609 | stat: | |
b84cb5df | 1610 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1611 | out: |
013fdb80 | 1612 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1613 | |
1614 | return success; | |
1615 | } | |
1616 | ||
21aa9af0 TH |
1617 | /** |
1618 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1619 | * @p: the thread to be awakened | |
1620 | * | |
2acca55e | 1621 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1622 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1623 | * the current task. |
21aa9af0 TH |
1624 | */ |
1625 | static void try_to_wake_up_local(struct task_struct *p) | |
1626 | { | |
1627 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1628 | |
1629 | BUG_ON(rq != this_rq()); | |
1630 | BUG_ON(p == current); | |
1631 | lockdep_assert_held(&rq->lock); | |
1632 | ||
2acca55e PZ |
1633 | if (!raw_spin_trylock(&p->pi_lock)) { |
1634 | raw_spin_unlock(&rq->lock); | |
1635 | raw_spin_lock(&p->pi_lock); | |
1636 | raw_spin_lock(&rq->lock); | |
1637 | } | |
1638 | ||
21aa9af0 | 1639 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1640 | goto out; |
21aa9af0 | 1641 | |
fd2f4419 | 1642 | if (!p->on_rq) |
d7c01d27 PZ |
1643 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1644 | ||
23f41eeb | 1645 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1646 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1647 | out: |
1648 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1649 | } |
1650 | ||
50fa610a DH |
1651 | /** |
1652 | * wake_up_process - Wake up a specific process | |
1653 | * @p: The process to be woken up. | |
1654 | * | |
1655 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1656 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1657 | * running. | |
1658 | * | |
1659 | * It may be assumed that this function implies a write memory barrier before | |
1660 | * changing the task state if and only if any tasks are woken up. | |
1661 | */ | |
7ad5b3a5 | 1662 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1663 | { |
d9514f6c | 1664 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1665 | } |
1da177e4 LT |
1666 | EXPORT_SYMBOL(wake_up_process); |
1667 | ||
7ad5b3a5 | 1668 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1669 | { |
1670 | return try_to_wake_up(p, state, 0); | |
1671 | } | |
1672 | ||
1da177e4 LT |
1673 | /* |
1674 | * Perform scheduler related setup for a newly forked process p. | |
1675 | * p is forked by current. | |
dd41f596 IM |
1676 | * |
1677 | * __sched_fork() is basic setup used by init_idle() too: | |
1678 | */ | |
1679 | static void __sched_fork(struct task_struct *p) | |
1680 | { | |
fd2f4419 PZ |
1681 | p->on_rq = 0; |
1682 | ||
1683 | p->se.on_rq = 0; | |
dd41f596 IM |
1684 | p->se.exec_start = 0; |
1685 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1686 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1687 | p->se.nr_migrations = 0; |
da7a735e | 1688 | p->se.vruntime = 0; |
fd2f4419 | 1689 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1690 | |
1691 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1692 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1693 | #endif |
476d139c | 1694 | |
fa717060 | 1695 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1696 | |
e107be36 AK |
1697 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1698 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1699 | #endif | |
dd41f596 IM |
1700 | } |
1701 | ||
1702 | /* | |
1703 | * fork()/clone()-time setup: | |
1704 | */ | |
3e51e3ed | 1705 | void sched_fork(struct task_struct *p) |
dd41f596 | 1706 | { |
0122ec5b | 1707 | unsigned long flags; |
dd41f596 IM |
1708 | int cpu = get_cpu(); |
1709 | ||
1710 | __sched_fork(p); | |
06b83b5f | 1711 | /* |
0017d735 | 1712 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1713 | * nobody will actually run it, and a signal or other external |
1714 | * event cannot wake it up and insert it on the runqueue either. | |
1715 | */ | |
0017d735 | 1716 | p->state = TASK_RUNNING; |
dd41f596 | 1717 | |
c350a04e MG |
1718 | /* |
1719 | * Make sure we do not leak PI boosting priority to the child. | |
1720 | */ | |
1721 | p->prio = current->normal_prio; | |
1722 | ||
b9dc29e7 MG |
1723 | /* |
1724 | * Revert to default priority/policy on fork if requested. | |
1725 | */ | |
1726 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1727 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1728 | p->policy = SCHED_NORMAL; |
6c697bdf | 1729 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1730 | p->rt_priority = 0; |
1731 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1732 | p->static_prio = NICE_TO_PRIO(0); | |
1733 | ||
1734 | p->prio = p->normal_prio = __normal_prio(p); | |
1735 | set_load_weight(p); | |
6c697bdf | 1736 | |
b9dc29e7 MG |
1737 | /* |
1738 | * We don't need the reset flag anymore after the fork. It has | |
1739 | * fulfilled its duty: | |
1740 | */ | |
1741 | p->sched_reset_on_fork = 0; | |
1742 | } | |
ca94c442 | 1743 | |
2ddbf952 HS |
1744 | if (!rt_prio(p->prio)) |
1745 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1746 | |
cd29fe6f PZ |
1747 | if (p->sched_class->task_fork) |
1748 | p->sched_class->task_fork(p); | |
1749 | ||
86951599 PZ |
1750 | /* |
1751 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1752 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1753 | * is ran before sched_fork(). | |
1754 | * | |
1755 | * Silence PROVE_RCU. | |
1756 | */ | |
0122ec5b | 1757 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1758 | set_task_cpu(p, cpu); |
0122ec5b | 1759 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1760 | |
52f17b6c | 1761 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1762 | if (likely(sched_info_on())) |
52f17b6c | 1763 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1764 | #endif |
3ca7a440 PZ |
1765 | #if defined(CONFIG_SMP) |
1766 | p->on_cpu = 0; | |
4866cde0 | 1767 | #endif |
bdd4e85d | 1768 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1769 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1770 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1771 | #endif |
806c09a7 | 1772 | #ifdef CONFIG_SMP |
917b627d | 1773 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1774 | #endif |
917b627d | 1775 | |
476d139c | 1776 | put_cpu(); |
1da177e4 LT |
1777 | } |
1778 | ||
1779 | /* | |
1780 | * wake_up_new_task - wake up a newly created task for the first time. | |
1781 | * | |
1782 | * This function will do some initial scheduler statistics housekeeping | |
1783 | * that must be done for every newly created context, then puts the task | |
1784 | * on the runqueue and wakes it. | |
1785 | */ | |
3e51e3ed | 1786 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1787 | { |
1788 | unsigned long flags; | |
dd41f596 | 1789 | struct rq *rq; |
fabf318e | 1790 | |
ab2515c4 | 1791 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1792 | #ifdef CONFIG_SMP |
1793 | /* | |
1794 | * Fork balancing, do it here and not earlier because: | |
1795 | * - cpus_allowed can change in the fork path | |
1796 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1797 | */ |
ab2515c4 | 1798 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1799 | #endif |
1800 | ||
ab2515c4 | 1801 | rq = __task_rq_lock(p); |
cd29fe6f | 1802 | activate_task(rq, p, 0); |
fd2f4419 | 1803 | p->on_rq = 1; |
89363381 | 1804 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1805 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1806 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1807 | if (p->sched_class->task_woken) |
1808 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1809 | #endif |
0122ec5b | 1810 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1811 | } |
1812 | ||
e107be36 AK |
1813 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1814 | ||
1815 | /** | |
80dd99b3 | 1816 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1817 | * @notifier: notifier struct to register |
e107be36 AK |
1818 | */ |
1819 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1820 | { | |
1821 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1822 | } | |
1823 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1824 | ||
1825 | /** | |
1826 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1827 | * @notifier: notifier struct to unregister |
e107be36 AK |
1828 | * |
1829 | * This is safe to call from within a preemption notifier. | |
1830 | */ | |
1831 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1832 | { | |
1833 | hlist_del(¬ifier->link); | |
1834 | } | |
1835 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1836 | ||
1837 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1838 | { | |
1839 | struct preempt_notifier *notifier; | |
1840 | struct hlist_node *node; | |
1841 | ||
1842 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1843 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1844 | } | |
1845 | ||
1846 | static void | |
1847 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1848 | struct task_struct *next) | |
1849 | { | |
1850 | struct preempt_notifier *notifier; | |
1851 | struct hlist_node *node; | |
1852 | ||
1853 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1854 | notifier->ops->sched_out(notifier, next); | |
1855 | } | |
1856 | ||
6d6bc0ad | 1857 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1858 | |
1859 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1860 | { | |
1861 | } | |
1862 | ||
1863 | static void | |
1864 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1865 | struct task_struct *next) | |
1866 | { | |
1867 | } | |
1868 | ||
6d6bc0ad | 1869 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1870 | |
4866cde0 NP |
1871 | /** |
1872 | * prepare_task_switch - prepare to switch tasks | |
1873 | * @rq: the runqueue preparing to switch | |
421cee29 | 1874 | * @prev: the current task that is being switched out |
4866cde0 NP |
1875 | * @next: the task we are going to switch to. |
1876 | * | |
1877 | * This is called with the rq lock held and interrupts off. It must | |
1878 | * be paired with a subsequent finish_task_switch after the context | |
1879 | * switch. | |
1880 | * | |
1881 | * prepare_task_switch sets up locking and calls architecture specific | |
1882 | * hooks. | |
1883 | */ | |
e107be36 AK |
1884 | static inline void |
1885 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1886 | struct task_struct *next) | |
4866cde0 | 1887 | { |
fe4b04fa PZ |
1888 | sched_info_switch(prev, next); |
1889 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1890 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1891 | prepare_lock_switch(rq, next); |
1892 | prepare_arch_switch(next); | |
fe4b04fa | 1893 | trace_sched_switch(prev, next); |
4866cde0 NP |
1894 | } |
1895 | ||
1da177e4 LT |
1896 | /** |
1897 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1898 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1899 | * @prev: the thread we just switched away from. |
1900 | * | |
4866cde0 NP |
1901 | * finish_task_switch must be called after the context switch, paired |
1902 | * with a prepare_task_switch call before the context switch. | |
1903 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1904 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1905 | * |
1906 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1907 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1908 | * with the lock held can cause deadlocks; see schedule() for |
1909 | * details.) | |
1910 | */ | |
a9957449 | 1911 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1912 | __releases(rq->lock) |
1913 | { | |
1da177e4 | 1914 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1915 | long prev_state; |
1da177e4 LT |
1916 | |
1917 | rq->prev_mm = NULL; | |
1918 | ||
1919 | /* | |
1920 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1921 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1922 | * schedule one last time. The schedule call will never return, and |
1923 | * the scheduled task must drop that reference. | |
c394cc9f | 1924 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1925 | * still held, otherwise prev could be scheduled on another cpu, die |
1926 | * there before we look at prev->state, and then the reference would | |
1927 | * be dropped twice. | |
1928 | * Manfred Spraul <manfred@colorfullife.com> | |
1929 | */ | |
55a101f8 | 1930 | prev_state = prev->state; |
4866cde0 | 1931 | finish_arch_switch(prev); |
8381f65d JI |
1932 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1933 | local_irq_disable(); | |
1934 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 1935 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
1936 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1937 | local_irq_enable(); | |
1938 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 1939 | finish_lock_switch(rq, prev); |
1ac9bc69 | 1940 | trace_sched_stat_sleeptime(current, rq->clock); |
e8fa1362 | 1941 | |
e107be36 | 1942 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1943 | if (mm) |
1944 | mmdrop(mm); | |
c394cc9f | 1945 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1946 | /* |
1947 | * Remove function-return probe instances associated with this | |
1948 | * task and put them back on the free list. | |
9761eea8 | 1949 | */ |
c6fd91f0 | 1950 | kprobe_flush_task(prev); |
1da177e4 | 1951 | put_task_struct(prev); |
c6fd91f0 | 1952 | } |
1da177e4 LT |
1953 | } |
1954 | ||
3f029d3c GH |
1955 | #ifdef CONFIG_SMP |
1956 | ||
1957 | /* assumes rq->lock is held */ | |
1958 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1959 | { | |
1960 | if (prev->sched_class->pre_schedule) | |
1961 | prev->sched_class->pre_schedule(rq, prev); | |
1962 | } | |
1963 | ||
1964 | /* rq->lock is NOT held, but preemption is disabled */ | |
1965 | static inline void post_schedule(struct rq *rq) | |
1966 | { | |
1967 | if (rq->post_schedule) { | |
1968 | unsigned long flags; | |
1969 | ||
05fa785c | 1970 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1971 | if (rq->curr->sched_class->post_schedule) |
1972 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1973 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1974 | |
1975 | rq->post_schedule = 0; | |
1976 | } | |
1977 | } | |
1978 | ||
1979 | #else | |
da19ab51 | 1980 | |
3f029d3c GH |
1981 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1982 | { | |
1983 | } | |
1984 | ||
1985 | static inline void post_schedule(struct rq *rq) | |
1986 | { | |
1da177e4 LT |
1987 | } |
1988 | ||
3f029d3c GH |
1989 | #endif |
1990 | ||
1da177e4 LT |
1991 | /** |
1992 | * schedule_tail - first thing a freshly forked thread must call. | |
1993 | * @prev: the thread we just switched away from. | |
1994 | */ | |
36c8b586 | 1995 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1996 | __releases(rq->lock) |
1997 | { | |
70b97a7f IM |
1998 | struct rq *rq = this_rq(); |
1999 | ||
4866cde0 | 2000 | finish_task_switch(rq, prev); |
da19ab51 | 2001 | |
3f029d3c GH |
2002 | /* |
2003 | * FIXME: do we need to worry about rq being invalidated by the | |
2004 | * task_switch? | |
2005 | */ | |
2006 | post_schedule(rq); | |
70b97a7f | 2007 | |
4866cde0 NP |
2008 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2009 | /* In this case, finish_task_switch does not reenable preemption */ | |
2010 | preempt_enable(); | |
2011 | #endif | |
1da177e4 | 2012 | if (current->set_child_tid) |
b488893a | 2013 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2014 | } |
2015 | ||
2016 | /* | |
2017 | * context_switch - switch to the new MM and the new | |
2018 | * thread's register state. | |
2019 | */ | |
dd41f596 | 2020 | static inline void |
70b97a7f | 2021 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2022 | struct task_struct *next) |
1da177e4 | 2023 | { |
dd41f596 | 2024 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2025 | |
e107be36 | 2026 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2027 | |
dd41f596 IM |
2028 | mm = next->mm; |
2029 | oldmm = prev->active_mm; | |
9226d125 ZA |
2030 | /* |
2031 | * For paravirt, this is coupled with an exit in switch_to to | |
2032 | * combine the page table reload and the switch backend into | |
2033 | * one hypercall. | |
2034 | */ | |
224101ed | 2035 | arch_start_context_switch(prev); |
9226d125 | 2036 | |
31915ab4 | 2037 | if (!mm) { |
1da177e4 LT |
2038 | next->active_mm = oldmm; |
2039 | atomic_inc(&oldmm->mm_count); | |
2040 | enter_lazy_tlb(oldmm, next); | |
2041 | } else | |
2042 | switch_mm(oldmm, mm, next); | |
2043 | ||
31915ab4 | 2044 | if (!prev->mm) { |
1da177e4 | 2045 | prev->active_mm = NULL; |
1da177e4 LT |
2046 | rq->prev_mm = oldmm; |
2047 | } | |
3a5f5e48 IM |
2048 | /* |
2049 | * Since the runqueue lock will be released by the next | |
2050 | * task (which is an invalid locking op but in the case | |
2051 | * of the scheduler it's an obvious special-case), so we | |
2052 | * do an early lockdep release here: | |
2053 | */ | |
2054 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2055 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2056 | #endif |
1da177e4 LT |
2057 | |
2058 | /* Here we just switch the register state and the stack. */ | |
2059 | switch_to(prev, next, prev); | |
2060 | ||
dd41f596 IM |
2061 | barrier(); |
2062 | /* | |
2063 | * this_rq must be evaluated again because prev may have moved | |
2064 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2065 | * frame will be invalid. | |
2066 | */ | |
2067 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2068 | } |
2069 | ||
2070 | /* | |
2071 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2072 | * | |
2073 | * externally visible scheduler statistics: current number of runnable | |
2074 | * threads, current number of uninterruptible-sleeping threads, total | |
2075 | * number of context switches performed since bootup. | |
2076 | */ | |
2077 | unsigned long nr_running(void) | |
2078 | { | |
2079 | unsigned long i, sum = 0; | |
2080 | ||
2081 | for_each_online_cpu(i) | |
2082 | sum += cpu_rq(i)->nr_running; | |
2083 | ||
2084 | return sum; | |
f711f609 | 2085 | } |
1da177e4 LT |
2086 | |
2087 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2088 | { |
1da177e4 | 2089 | unsigned long i, sum = 0; |
f711f609 | 2090 | |
0a945022 | 2091 | for_each_possible_cpu(i) |
1da177e4 | 2092 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2093 | |
2094 | /* | |
1da177e4 LT |
2095 | * Since we read the counters lockless, it might be slightly |
2096 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2097 | */ |
1da177e4 LT |
2098 | if (unlikely((long)sum < 0)) |
2099 | sum = 0; | |
f711f609 | 2100 | |
1da177e4 | 2101 | return sum; |
f711f609 | 2102 | } |
f711f609 | 2103 | |
1da177e4 | 2104 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2105 | { |
cc94abfc SR |
2106 | int i; |
2107 | unsigned long long sum = 0; | |
46cb4b7c | 2108 | |
0a945022 | 2109 | for_each_possible_cpu(i) |
1da177e4 | 2110 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2111 | |
1da177e4 LT |
2112 | return sum; |
2113 | } | |
483b4ee6 | 2114 | |
1da177e4 LT |
2115 | unsigned long nr_iowait(void) |
2116 | { | |
2117 | unsigned long i, sum = 0; | |
483b4ee6 | 2118 | |
0a945022 | 2119 | for_each_possible_cpu(i) |
1da177e4 | 2120 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2121 | |
1da177e4 LT |
2122 | return sum; |
2123 | } | |
483b4ee6 | 2124 | |
8c215bd3 | 2125 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2126 | { |
8c215bd3 | 2127 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2128 | return atomic_read(&this->nr_iowait); |
2129 | } | |
46cb4b7c | 2130 | |
69d25870 AV |
2131 | unsigned long this_cpu_load(void) |
2132 | { | |
2133 | struct rq *this = this_rq(); | |
2134 | return this->cpu_load[0]; | |
2135 | } | |
e790fb0b | 2136 | |
46cb4b7c | 2137 | |
dce48a84 TG |
2138 | /* Variables and functions for calc_load */ |
2139 | static atomic_long_t calc_load_tasks; | |
2140 | static unsigned long calc_load_update; | |
2141 | unsigned long avenrun[3]; | |
2142 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2143 | |
74f5187a PZ |
2144 | static long calc_load_fold_active(struct rq *this_rq) |
2145 | { | |
2146 | long nr_active, delta = 0; | |
2147 | ||
2148 | nr_active = this_rq->nr_running; | |
2149 | nr_active += (long) this_rq->nr_uninterruptible; | |
2150 | ||
2151 | if (nr_active != this_rq->calc_load_active) { | |
2152 | delta = nr_active - this_rq->calc_load_active; | |
2153 | this_rq->calc_load_active = nr_active; | |
2154 | } | |
2155 | ||
2156 | return delta; | |
2157 | } | |
2158 | ||
0f004f5a PZ |
2159 | static unsigned long |
2160 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2161 | { | |
2162 | load *= exp; | |
2163 | load += active * (FIXED_1 - exp); | |
2164 | load += 1UL << (FSHIFT - 1); | |
2165 | return load >> FSHIFT; | |
2166 | } | |
2167 | ||
74f5187a PZ |
2168 | #ifdef CONFIG_NO_HZ |
2169 | /* | |
2170 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2171 | * | |
2172 | * When making the ILB scale, we should try to pull this in as well. | |
2173 | */ | |
2174 | static atomic_long_t calc_load_tasks_idle; | |
2175 | ||
029632fb | 2176 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2177 | { |
2178 | long delta; | |
2179 | ||
2180 | delta = calc_load_fold_active(this_rq); | |
2181 | if (delta) | |
2182 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2183 | } | |
2184 | ||
2185 | static long calc_load_fold_idle(void) | |
2186 | { | |
2187 | long delta = 0; | |
2188 | ||
2189 | /* | |
2190 | * Its got a race, we don't care... | |
2191 | */ | |
2192 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2193 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2194 | ||
2195 | return delta; | |
2196 | } | |
0f004f5a PZ |
2197 | |
2198 | /** | |
2199 | * fixed_power_int - compute: x^n, in O(log n) time | |
2200 | * | |
2201 | * @x: base of the power | |
2202 | * @frac_bits: fractional bits of @x | |
2203 | * @n: power to raise @x to. | |
2204 | * | |
2205 | * By exploiting the relation between the definition of the natural power | |
2206 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2207 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2208 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2209 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2210 | * of course trivially computable in O(log_2 n), the length of our binary | |
2211 | * vector. | |
2212 | */ | |
2213 | static unsigned long | |
2214 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2215 | { | |
2216 | unsigned long result = 1UL << frac_bits; | |
2217 | ||
2218 | if (n) for (;;) { | |
2219 | if (n & 1) { | |
2220 | result *= x; | |
2221 | result += 1UL << (frac_bits - 1); | |
2222 | result >>= frac_bits; | |
2223 | } | |
2224 | n >>= 1; | |
2225 | if (!n) | |
2226 | break; | |
2227 | x *= x; | |
2228 | x += 1UL << (frac_bits - 1); | |
2229 | x >>= frac_bits; | |
2230 | } | |
2231 | ||
2232 | return result; | |
2233 | } | |
2234 | ||
2235 | /* | |
2236 | * a1 = a0 * e + a * (1 - e) | |
2237 | * | |
2238 | * a2 = a1 * e + a * (1 - e) | |
2239 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2240 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2241 | * | |
2242 | * a3 = a2 * e + a * (1 - e) | |
2243 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2244 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2245 | * | |
2246 | * ... | |
2247 | * | |
2248 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2249 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2250 | * = a0 * e^n + a * (1 - e^n) | |
2251 | * | |
2252 | * [1] application of the geometric series: | |
2253 | * | |
2254 | * n 1 - x^(n+1) | |
2255 | * S_n := \Sum x^i = ------------- | |
2256 | * i=0 1 - x | |
2257 | */ | |
2258 | static unsigned long | |
2259 | calc_load_n(unsigned long load, unsigned long exp, | |
2260 | unsigned long active, unsigned int n) | |
2261 | { | |
2262 | ||
2263 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2264 | } | |
2265 | ||
2266 | /* | |
2267 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2268 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2269 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2270 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2271 | * | |
2272 | * Once we've updated the global active value, we need to apply the exponential | |
2273 | * weights adjusted to the number of cycles missed. | |
2274 | */ | |
2275 | static void calc_global_nohz(unsigned long ticks) | |
2276 | { | |
2277 | long delta, active, n; | |
2278 | ||
2279 | if (time_before(jiffies, calc_load_update)) | |
2280 | return; | |
2281 | ||
2282 | /* | |
2283 | * If we crossed a calc_load_update boundary, make sure to fold | |
2284 | * any pending idle changes, the respective CPUs might have | |
2285 | * missed the tick driven calc_load_account_active() update | |
2286 | * due to NO_HZ. | |
2287 | */ | |
2288 | delta = calc_load_fold_idle(); | |
2289 | if (delta) | |
2290 | atomic_long_add(delta, &calc_load_tasks); | |
2291 | ||
2292 | /* | |
2293 | * If we were idle for multiple load cycles, apply them. | |
2294 | */ | |
2295 | if (ticks >= LOAD_FREQ) { | |
2296 | n = ticks / LOAD_FREQ; | |
2297 | ||
2298 | active = atomic_long_read(&calc_load_tasks); | |
2299 | active = active > 0 ? active * FIXED_1 : 0; | |
2300 | ||
2301 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
2302 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2303 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
2304 | ||
2305 | calc_load_update += n * LOAD_FREQ; | |
2306 | } | |
2307 | ||
2308 | /* | |
2309 | * Its possible the remainder of the above division also crosses | |
2310 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
2311 | * which comes after this will take care of that. | |
2312 | * | |
2313 | * Consider us being 11 ticks before a cycle completion, and us | |
2314 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
2315 | * age us 4 cycles, and the test in calc_global_load() will | |
2316 | * pick up the final one. | |
2317 | */ | |
2318 | } | |
74f5187a | 2319 | #else |
029632fb | 2320 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2321 | { |
2322 | } | |
2323 | ||
2324 | static inline long calc_load_fold_idle(void) | |
2325 | { | |
2326 | return 0; | |
2327 | } | |
0f004f5a PZ |
2328 | |
2329 | static void calc_global_nohz(unsigned long ticks) | |
2330 | { | |
2331 | } | |
74f5187a PZ |
2332 | #endif |
2333 | ||
2d02494f TG |
2334 | /** |
2335 | * get_avenrun - get the load average array | |
2336 | * @loads: pointer to dest load array | |
2337 | * @offset: offset to add | |
2338 | * @shift: shift count to shift the result left | |
2339 | * | |
2340 | * These values are estimates at best, so no need for locking. | |
2341 | */ | |
2342 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2343 | { | |
2344 | loads[0] = (avenrun[0] + offset) << shift; | |
2345 | loads[1] = (avenrun[1] + offset) << shift; | |
2346 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2347 | } |
46cb4b7c | 2348 | |
46cb4b7c | 2349 | /* |
dce48a84 TG |
2350 | * calc_load - update the avenrun load estimates 10 ticks after the |
2351 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2352 | */ |
0f004f5a | 2353 | void calc_global_load(unsigned long ticks) |
7835b98b | 2354 | { |
dce48a84 | 2355 | long active; |
1da177e4 | 2356 | |
0f004f5a PZ |
2357 | calc_global_nohz(ticks); |
2358 | ||
2359 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 2360 | return; |
1da177e4 | 2361 | |
dce48a84 TG |
2362 | active = atomic_long_read(&calc_load_tasks); |
2363 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2364 | |
dce48a84 TG |
2365 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2366 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2367 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2368 | |
dce48a84 TG |
2369 | calc_load_update += LOAD_FREQ; |
2370 | } | |
1da177e4 | 2371 | |
dce48a84 | 2372 | /* |
74f5187a PZ |
2373 | * Called from update_cpu_load() to periodically update this CPU's |
2374 | * active count. | |
dce48a84 TG |
2375 | */ |
2376 | static void calc_load_account_active(struct rq *this_rq) | |
2377 | { | |
74f5187a | 2378 | long delta; |
08c183f3 | 2379 | |
74f5187a PZ |
2380 | if (time_before(jiffies, this_rq->calc_load_update)) |
2381 | return; | |
783609c6 | 2382 | |
74f5187a PZ |
2383 | delta = calc_load_fold_active(this_rq); |
2384 | delta += calc_load_fold_idle(); | |
2385 | if (delta) | |
dce48a84 | 2386 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2387 | |
2388 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2389 | } |
2390 | ||
fdf3e95d VP |
2391 | /* |
2392 | * The exact cpuload at various idx values, calculated at every tick would be | |
2393 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2394 | * | |
2395 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2396 | * on nth tick when cpu may be busy, then we have: | |
2397 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2398 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2399 | * | |
2400 | * decay_load_missed() below does efficient calculation of | |
2401 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2402 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2403 | * | |
2404 | * The calculation is approximated on a 128 point scale. | |
2405 | * degrade_zero_ticks is the number of ticks after which load at any | |
2406 | * particular idx is approximated to be zero. | |
2407 | * degrade_factor is a precomputed table, a row for each load idx. | |
2408 | * Each column corresponds to degradation factor for a power of two ticks, | |
2409 | * based on 128 point scale. | |
2410 | * Example: | |
2411 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2412 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2413 | * | |
2414 | * With this power of 2 load factors, we can degrade the load n times | |
2415 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2416 | * n mult/shifts needed by the exact degradation. | |
2417 | */ | |
2418 | #define DEGRADE_SHIFT 7 | |
2419 | static const unsigned char | |
2420 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2421 | static const unsigned char | |
2422 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2423 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2424 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2425 | {96, 72, 40, 12, 1, 0, 0}, | |
2426 | {112, 98, 75, 43, 15, 1, 0}, | |
2427 | {120, 112, 98, 76, 45, 16, 2} }; | |
2428 | ||
2429 | /* | |
2430 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2431 | * would be when CPU is idle and so we just decay the old load without | |
2432 | * adding any new load. | |
2433 | */ | |
2434 | static unsigned long | |
2435 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2436 | { | |
2437 | int j = 0; | |
2438 | ||
2439 | if (!missed_updates) | |
2440 | return load; | |
2441 | ||
2442 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2443 | return 0; | |
2444 | ||
2445 | if (idx == 1) | |
2446 | return load >> missed_updates; | |
2447 | ||
2448 | while (missed_updates) { | |
2449 | if (missed_updates % 2) | |
2450 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2451 | ||
2452 | missed_updates >>= 1; | |
2453 | j++; | |
2454 | } | |
2455 | return load; | |
2456 | } | |
2457 | ||
46cb4b7c | 2458 | /* |
dd41f596 | 2459 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2460 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2461 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2462 | */ |
029632fb | 2463 | void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 2464 | { |
495eca49 | 2465 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
2466 | unsigned long curr_jiffies = jiffies; |
2467 | unsigned long pending_updates; | |
dd41f596 | 2468 | int i, scale; |
46cb4b7c | 2469 | |
dd41f596 | 2470 | this_rq->nr_load_updates++; |
46cb4b7c | 2471 | |
fdf3e95d VP |
2472 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
2473 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2474 | return; | |
2475 | ||
2476 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2477 | this_rq->last_load_update_tick = curr_jiffies; | |
2478 | ||
dd41f596 | 2479 | /* Update our load: */ |
fdf3e95d VP |
2480 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2481 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2482 | unsigned long old_load, new_load; |
7d1e6a9b | 2483 | |
dd41f596 | 2484 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2485 | |
dd41f596 | 2486 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2487 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2488 | new_load = this_load; |
a25707f3 IM |
2489 | /* |
2490 | * Round up the averaging division if load is increasing. This | |
2491 | * prevents us from getting stuck on 9 if the load is 10, for | |
2492 | * example. | |
2493 | */ | |
2494 | if (new_load > old_load) | |
fdf3e95d VP |
2495 | new_load += scale - 1; |
2496 | ||
2497 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2498 | } |
da2b71ed SS |
2499 | |
2500 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2501 | } |
2502 | ||
2503 | static void update_cpu_load_active(struct rq *this_rq) | |
2504 | { | |
2505 | update_cpu_load(this_rq); | |
46cb4b7c | 2506 | |
74f5187a | 2507 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2508 | } |
2509 | ||
dd41f596 | 2510 | #ifdef CONFIG_SMP |
8a0be9ef | 2511 | |
46cb4b7c | 2512 | /* |
38022906 PZ |
2513 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2514 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2515 | */ |
38022906 | 2516 | void sched_exec(void) |
46cb4b7c | 2517 | { |
38022906 | 2518 | struct task_struct *p = current; |
1da177e4 | 2519 | unsigned long flags; |
0017d735 | 2520 | int dest_cpu; |
46cb4b7c | 2521 | |
8f42ced9 | 2522 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2523 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2524 | if (dest_cpu == smp_processor_id()) |
2525 | goto unlock; | |
38022906 | 2526 | |
8f42ced9 | 2527 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2528 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2529 | |
8f42ced9 PZ |
2530 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2531 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2532 | return; |
2533 | } | |
0017d735 | 2534 | unlock: |
8f42ced9 | 2535 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2536 | } |
dd41f596 | 2537 | |
1da177e4 LT |
2538 | #endif |
2539 | ||
1da177e4 | 2540 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2541 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2542 | |
2543 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2544 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2545 | |
2546 | /* | |
c5f8d995 | 2547 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2548 | * @p in case that task is currently running. |
c5f8d995 HS |
2549 | * |
2550 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2551 | */ |
c5f8d995 HS |
2552 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2553 | { | |
2554 | u64 ns = 0; | |
2555 | ||
2556 | if (task_current(rq, p)) { | |
2557 | update_rq_clock(rq); | |
305e6835 | 2558 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2559 | if ((s64)ns < 0) |
2560 | ns = 0; | |
2561 | } | |
2562 | ||
2563 | return ns; | |
2564 | } | |
2565 | ||
bb34d92f | 2566 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2567 | { |
1da177e4 | 2568 | unsigned long flags; |
41b86e9c | 2569 | struct rq *rq; |
bb34d92f | 2570 | u64 ns = 0; |
48f24c4d | 2571 | |
41b86e9c | 2572 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2573 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2574 | task_rq_unlock(rq, p, &flags); |
1508487e | 2575 | |
c5f8d995 HS |
2576 | return ns; |
2577 | } | |
f06febc9 | 2578 | |
c5f8d995 HS |
2579 | /* |
2580 | * Return accounted runtime for the task. | |
2581 | * In case the task is currently running, return the runtime plus current's | |
2582 | * pending runtime that have not been accounted yet. | |
2583 | */ | |
2584 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2585 | { | |
2586 | unsigned long flags; | |
2587 | struct rq *rq; | |
2588 | u64 ns = 0; | |
2589 | ||
2590 | rq = task_rq_lock(p, &flags); | |
2591 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2592 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2593 | |
2594 | return ns; | |
2595 | } | |
48f24c4d | 2596 | |
54c707e9 GC |
2597 | #ifdef CONFIG_CGROUP_CPUACCT |
2598 | struct cgroup_subsys cpuacct_subsys; | |
2599 | struct cpuacct root_cpuacct; | |
2600 | #endif | |
2601 | ||
be726ffd GC |
2602 | static inline void task_group_account_field(struct task_struct *p, int index, |
2603 | u64 tmp) | |
54c707e9 GC |
2604 | { |
2605 | #ifdef CONFIG_CGROUP_CPUACCT | |
2606 | struct kernel_cpustat *kcpustat; | |
2607 | struct cpuacct *ca; | |
2608 | #endif | |
2609 | /* | |
2610 | * Since all updates are sure to touch the root cgroup, we | |
2611 | * get ourselves ahead and touch it first. If the root cgroup | |
2612 | * is the only cgroup, then nothing else should be necessary. | |
2613 | * | |
2614 | */ | |
2615 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
2616 | ||
2617 | #ifdef CONFIG_CGROUP_CPUACCT | |
2618 | if (unlikely(!cpuacct_subsys.active)) | |
2619 | return; | |
2620 | ||
2621 | rcu_read_lock(); | |
2622 | ca = task_ca(p); | |
2623 | while (ca && (ca != &root_cpuacct)) { | |
2624 | kcpustat = this_cpu_ptr(ca->cpustat); | |
2625 | kcpustat->cpustat[index] += tmp; | |
2626 | ca = parent_ca(ca); | |
2627 | } | |
2628 | rcu_read_unlock(); | |
2629 | #endif | |
2630 | } | |
2631 | ||
2632 | ||
1da177e4 LT |
2633 | /* |
2634 | * Account user cpu time to a process. | |
2635 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 2636 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 2637 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 2638 | */ |
457533a7 MS |
2639 | void account_user_time(struct task_struct *p, cputime_t cputime, |
2640 | cputime_t cputime_scaled) | |
1da177e4 | 2641 | { |
3292beb3 | 2642 | int index; |
1da177e4 | 2643 | |
457533a7 | 2644 | /* Add user time to process. */ |
64861634 MS |
2645 | p->utime += cputime; |
2646 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2647 | account_group_user_time(p, cputime); |
1da177e4 | 2648 | |
3292beb3 | 2649 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
ef12fefa | 2650 | |
1da177e4 | 2651 | /* Add user time to cpustat. */ |
612ef28a | 2652 | task_group_account_field(p, index, (__force u64) cputime); |
ef12fefa | 2653 | |
49b5cf34 JL |
2654 | /* Account for user time used */ |
2655 | acct_update_integrals(p); | |
1da177e4 LT |
2656 | } |
2657 | ||
94886b84 LV |
2658 | /* |
2659 | * Account guest cpu time to a process. | |
2660 | * @p: the process that the cpu time gets accounted to | |
2661 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 2662 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 2663 | */ |
457533a7 MS |
2664 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
2665 | cputime_t cputime_scaled) | |
94886b84 | 2666 | { |
3292beb3 | 2667 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
94886b84 | 2668 | |
457533a7 | 2669 | /* Add guest time to process. */ |
64861634 MS |
2670 | p->utime += cputime; |
2671 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2672 | account_group_user_time(p, cputime); |
64861634 | 2673 | p->gtime += cputime; |
94886b84 | 2674 | |
457533a7 | 2675 | /* Add guest time to cpustat. */ |
ce0e7b28 | 2676 | if (TASK_NICE(p) > 0) { |
612ef28a MS |
2677 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
2678 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
ce0e7b28 | 2679 | } else { |
612ef28a MS |
2680 | cpustat[CPUTIME_USER] += (__force u64) cputime; |
2681 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
ce0e7b28 | 2682 | } |
94886b84 LV |
2683 | } |
2684 | ||
70a89a66 VP |
2685 | /* |
2686 | * Account system cpu time to a process and desired cpustat field | |
2687 | * @p: the process that the cpu time gets accounted to | |
2688 | * @cputime: the cpu time spent in kernel space since the last update | |
2689 | * @cputime_scaled: cputime scaled by cpu frequency | |
2690 | * @target_cputime64: pointer to cpustat field that has to be updated | |
2691 | */ | |
2692 | static inline | |
2693 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3292beb3 | 2694 | cputime_t cputime_scaled, int index) |
70a89a66 | 2695 | { |
70a89a66 | 2696 | /* Add system time to process. */ |
64861634 MS |
2697 | p->stime += cputime; |
2698 | p->stimescaled += cputime_scaled; | |
70a89a66 VP |
2699 | account_group_system_time(p, cputime); |
2700 | ||
2701 | /* Add system time to cpustat. */ | |
612ef28a | 2702 | task_group_account_field(p, index, (__force u64) cputime); |
70a89a66 VP |
2703 | |
2704 | /* Account for system time used */ | |
2705 | acct_update_integrals(p); | |
2706 | } | |
2707 | ||
1da177e4 LT |
2708 | /* |
2709 | * Account system cpu time to a process. | |
2710 | * @p: the process that the cpu time gets accounted to | |
2711 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
2712 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 2713 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
2714 | */ |
2715 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 2716 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 | 2717 | { |
3292beb3 | 2718 | int index; |
1da177e4 | 2719 | |
983ed7a6 | 2720 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 2721 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
2722 | return; |
2723 | } | |
94886b84 | 2724 | |
1da177e4 | 2725 | if (hardirq_count() - hardirq_offset) |
3292beb3 | 2726 | index = CPUTIME_IRQ; |
75e1056f | 2727 | else if (in_serving_softirq()) |
3292beb3 | 2728 | index = CPUTIME_SOFTIRQ; |
1da177e4 | 2729 | else |
3292beb3 | 2730 | index = CPUTIME_SYSTEM; |
ef12fefa | 2731 | |
3292beb3 | 2732 | __account_system_time(p, cputime, cputime_scaled, index); |
1da177e4 LT |
2733 | } |
2734 | ||
c66f08be | 2735 | /* |
1da177e4 | 2736 | * Account for involuntary wait time. |
544b4a1f | 2737 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 2738 | */ |
79741dd3 | 2739 | void account_steal_time(cputime_t cputime) |
c66f08be | 2740 | { |
3292beb3 | 2741 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
79741dd3 | 2742 | |
612ef28a | 2743 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; |
c66f08be MN |
2744 | } |
2745 | ||
1da177e4 | 2746 | /* |
79741dd3 MS |
2747 | * Account for idle time. |
2748 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 2749 | */ |
79741dd3 | 2750 | void account_idle_time(cputime_t cputime) |
1da177e4 | 2751 | { |
3292beb3 | 2752 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
70b97a7f | 2753 | struct rq *rq = this_rq(); |
1da177e4 | 2754 | |
79741dd3 | 2755 | if (atomic_read(&rq->nr_iowait) > 0) |
612ef28a | 2756 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; |
79741dd3 | 2757 | else |
612ef28a | 2758 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; |
1da177e4 LT |
2759 | } |
2760 | ||
e6e6685a GC |
2761 | static __always_inline bool steal_account_process_tick(void) |
2762 | { | |
2763 | #ifdef CONFIG_PARAVIRT | |
2764 | if (static_branch(¶virt_steal_enabled)) { | |
2765 | u64 steal, st = 0; | |
2766 | ||
2767 | steal = paravirt_steal_clock(smp_processor_id()); | |
2768 | steal -= this_rq()->prev_steal_time; | |
2769 | ||
2770 | st = steal_ticks(steal); | |
2771 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
2772 | ||
2773 | account_steal_time(st); | |
2774 | return st; | |
2775 | } | |
2776 | #endif | |
2777 | return false; | |
2778 | } | |
2779 | ||
79741dd3 MS |
2780 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
2781 | ||
abb74cef VP |
2782 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
2783 | /* | |
2784 | * Account a tick to a process and cpustat | |
2785 | * @p: the process that the cpu time gets accounted to | |
2786 | * @user_tick: is the tick from userspace | |
2787 | * @rq: the pointer to rq | |
2788 | * | |
2789 | * Tick demultiplexing follows the order | |
2790 | * - pending hardirq update | |
2791 | * - pending softirq update | |
2792 | * - user_time | |
2793 | * - idle_time | |
2794 | * - system time | |
2795 | * - check for guest_time | |
2796 | * - else account as system_time | |
2797 | * | |
2798 | * Check for hardirq is done both for system and user time as there is | |
2799 | * no timer going off while we are on hardirq and hence we may never get an | |
2800 | * opportunity to update it solely in system time. | |
2801 | * p->stime and friends are only updated on system time and not on irq | |
2802 | * softirq as those do not count in task exec_runtime any more. | |
2803 | */ | |
2804 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2805 | struct rq *rq) | |
2806 | { | |
2807 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3292beb3 | 2808 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef | 2809 | |
e6e6685a GC |
2810 | if (steal_account_process_tick()) |
2811 | return; | |
2812 | ||
abb74cef | 2813 | if (irqtime_account_hi_update()) { |
612ef28a | 2814 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; |
abb74cef | 2815 | } else if (irqtime_account_si_update()) { |
612ef28a | 2816 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; |
414bee9b VP |
2817 | } else if (this_cpu_ksoftirqd() == p) { |
2818 | /* | |
2819 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
2820 | * So, we have to handle it separately here. | |
2821 | * Also, p->stime needs to be updated for ksoftirqd. | |
2822 | */ | |
2823 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2824 | CPUTIME_SOFTIRQ); |
abb74cef VP |
2825 | } else if (user_tick) { |
2826 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2827 | } else if (p == rq->idle) { | |
2828 | account_idle_time(cputime_one_jiffy); | |
2829 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
2830 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2831 | } else { | |
2832 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2833 | CPUTIME_SYSTEM); |
abb74cef VP |
2834 | } |
2835 | } | |
2836 | ||
2837 | static void irqtime_account_idle_ticks(int ticks) | |
2838 | { | |
2839 | int i; | |
2840 | struct rq *rq = this_rq(); | |
2841 | ||
2842 | for (i = 0; i < ticks; i++) | |
2843 | irqtime_account_process_tick(current, 0, rq); | |
2844 | } | |
544b4a1f | 2845 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
2846 | static void irqtime_account_idle_ticks(int ticks) {} |
2847 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2848 | struct rq *rq) {} | |
544b4a1f | 2849 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
2850 | |
2851 | /* | |
2852 | * Account a single tick of cpu time. | |
2853 | * @p: the process that the cpu time gets accounted to | |
2854 | * @user_tick: indicates if the tick is a user or a system tick | |
2855 | */ | |
2856 | void account_process_tick(struct task_struct *p, int user_tick) | |
2857 | { | |
a42548a1 | 2858 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
2859 | struct rq *rq = this_rq(); |
2860 | ||
abb74cef VP |
2861 | if (sched_clock_irqtime) { |
2862 | irqtime_account_process_tick(p, user_tick, rq); | |
2863 | return; | |
2864 | } | |
2865 | ||
e6e6685a GC |
2866 | if (steal_account_process_tick()) |
2867 | return; | |
2868 | ||
79741dd3 | 2869 | if (user_tick) |
a42548a1 | 2870 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 2871 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 2872 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
2873 | one_jiffy_scaled); |
2874 | else | |
a42548a1 | 2875 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
2876 | } |
2877 | ||
2878 | /* | |
2879 | * Account multiple ticks of steal time. | |
2880 | * @p: the process from which the cpu time has been stolen | |
2881 | * @ticks: number of stolen ticks | |
2882 | */ | |
2883 | void account_steal_ticks(unsigned long ticks) | |
2884 | { | |
2885 | account_steal_time(jiffies_to_cputime(ticks)); | |
2886 | } | |
2887 | ||
2888 | /* | |
2889 | * Account multiple ticks of idle time. | |
2890 | * @ticks: number of stolen ticks | |
2891 | */ | |
2892 | void account_idle_ticks(unsigned long ticks) | |
2893 | { | |
abb74cef VP |
2894 | |
2895 | if (sched_clock_irqtime) { | |
2896 | irqtime_account_idle_ticks(ticks); | |
2897 | return; | |
2898 | } | |
2899 | ||
79741dd3 | 2900 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
2901 | } |
2902 | ||
79741dd3 MS |
2903 | #endif |
2904 | ||
49048622 BS |
2905 | /* |
2906 | * Use precise platform statistics if available: | |
2907 | */ | |
2908 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 2909 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2910 | { |
d99ca3b9 HS |
2911 | *ut = p->utime; |
2912 | *st = p->stime; | |
49048622 BS |
2913 | } |
2914 | ||
0cf55e1e | 2915 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2916 | { |
0cf55e1e HS |
2917 | struct task_cputime cputime; |
2918 | ||
2919 | thread_group_cputime(p, &cputime); | |
2920 | ||
2921 | *ut = cputime.utime; | |
2922 | *st = cputime.stime; | |
49048622 BS |
2923 | } |
2924 | #else | |
761b1d26 HS |
2925 | |
2926 | #ifndef nsecs_to_cputime | |
b7b20df9 | 2927 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
2928 | #endif |
2929 | ||
d180c5bc | 2930 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2931 | { |
64861634 | 2932 | cputime_t rtime, utime = p->utime, total = utime + p->stime; |
49048622 BS |
2933 | |
2934 | /* | |
2935 | * Use CFS's precise accounting: | |
2936 | */ | |
d180c5bc | 2937 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
2938 | |
2939 | if (total) { | |
64861634 | 2940 | u64 temp = (__force u64) rtime; |
d180c5bc | 2941 | |
64861634 MS |
2942 | temp *= (__force u64) utime; |
2943 | do_div(temp, (__force u32) total); | |
2944 | utime = (__force cputime_t) temp; | |
d180c5bc HS |
2945 | } else |
2946 | utime = rtime; | |
49048622 | 2947 | |
d180c5bc HS |
2948 | /* |
2949 | * Compare with previous values, to keep monotonicity: | |
2950 | */ | |
761b1d26 | 2951 | p->prev_utime = max(p->prev_utime, utime); |
64861634 | 2952 | p->prev_stime = max(p->prev_stime, rtime - p->prev_utime); |
49048622 | 2953 | |
d99ca3b9 HS |
2954 | *ut = p->prev_utime; |
2955 | *st = p->prev_stime; | |
49048622 BS |
2956 | } |
2957 | ||
0cf55e1e HS |
2958 | /* |
2959 | * Must be called with siglock held. | |
2960 | */ | |
2961 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 2962 | { |
0cf55e1e HS |
2963 | struct signal_struct *sig = p->signal; |
2964 | struct task_cputime cputime; | |
2965 | cputime_t rtime, utime, total; | |
49048622 | 2966 | |
0cf55e1e | 2967 | thread_group_cputime(p, &cputime); |
49048622 | 2968 | |
64861634 | 2969 | total = cputime.utime + cputime.stime; |
0cf55e1e | 2970 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); |
49048622 | 2971 | |
0cf55e1e | 2972 | if (total) { |
64861634 | 2973 | u64 temp = (__force u64) rtime; |
49048622 | 2974 | |
64861634 MS |
2975 | temp *= (__force u64) cputime.utime; |
2976 | do_div(temp, (__force u32) total); | |
2977 | utime = (__force cputime_t) temp; | |
0cf55e1e HS |
2978 | } else |
2979 | utime = rtime; | |
2980 | ||
2981 | sig->prev_utime = max(sig->prev_utime, utime); | |
64861634 | 2982 | sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime); |
0cf55e1e HS |
2983 | |
2984 | *ut = sig->prev_utime; | |
2985 | *st = sig->prev_stime; | |
49048622 | 2986 | } |
49048622 | 2987 | #endif |
49048622 | 2988 | |
7835b98b CL |
2989 | /* |
2990 | * This function gets called by the timer code, with HZ frequency. | |
2991 | * We call it with interrupts disabled. | |
7835b98b CL |
2992 | */ |
2993 | void scheduler_tick(void) | |
2994 | { | |
7835b98b CL |
2995 | int cpu = smp_processor_id(); |
2996 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2997 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2998 | |
2999 | sched_clock_tick(); | |
dd41f596 | 3000 | |
05fa785c | 3001 | raw_spin_lock(&rq->lock); |
3e51f33f | 3002 | update_rq_clock(rq); |
fdf3e95d | 3003 | update_cpu_load_active(rq); |
fa85ae24 | 3004 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3005 | raw_spin_unlock(&rq->lock); |
7835b98b | 3006 | |
e9d2b064 | 3007 | perf_event_task_tick(); |
e220d2dc | 3008 | |
e418e1c2 | 3009 | #ifdef CONFIG_SMP |
6eb57e0d | 3010 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 3011 | trigger_load_balance(rq, cpu); |
e418e1c2 | 3012 | #endif |
1da177e4 LT |
3013 | } |
3014 | ||
132380a0 | 3015 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3016 | { |
3017 | if (in_lock_functions(addr)) { | |
3018 | addr = CALLER_ADDR2; | |
3019 | if (in_lock_functions(addr)) | |
3020 | addr = CALLER_ADDR3; | |
3021 | } | |
3022 | return addr; | |
3023 | } | |
1da177e4 | 3024 | |
7e49fcce SR |
3025 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3026 | defined(CONFIG_PREEMPT_TRACER)) | |
3027 | ||
43627582 | 3028 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3029 | { |
6cd8a4bb | 3030 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3031 | /* |
3032 | * Underflow? | |
3033 | */ | |
9a11b49a IM |
3034 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3035 | return; | |
6cd8a4bb | 3036 | #endif |
1da177e4 | 3037 | preempt_count() += val; |
6cd8a4bb | 3038 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3039 | /* |
3040 | * Spinlock count overflowing soon? | |
3041 | */ | |
33859f7f MOS |
3042 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3043 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3044 | #endif |
3045 | if (preempt_count() == val) | |
3046 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3047 | } |
3048 | EXPORT_SYMBOL(add_preempt_count); | |
3049 | ||
43627582 | 3050 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3051 | { |
6cd8a4bb | 3052 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3053 | /* |
3054 | * Underflow? | |
3055 | */ | |
01e3eb82 | 3056 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3057 | return; |
1da177e4 LT |
3058 | /* |
3059 | * Is the spinlock portion underflowing? | |
3060 | */ | |
9a11b49a IM |
3061 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3062 | !(preempt_count() & PREEMPT_MASK))) | |
3063 | return; | |
6cd8a4bb | 3064 | #endif |
9a11b49a | 3065 | |
6cd8a4bb SR |
3066 | if (preempt_count() == val) |
3067 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3068 | preempt_count() -= val; |
3069 | } | |
3070 | EXPORT_SYMBOL(sub_preempt_count); | |
3071 | ||
3072 | #endif | |
3073 | ||
3074 | /* | |
dd41f596 | 3075 | * Print scheduling while atomic bug: |
1da177e4 | 3076 | */ |
dd41f596 | 3077 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3078 | { |
838225b4 SS |
3079 | struct pt_regs *regs = get_irq_regs(); |
3080 | ||
664dfa65 DJ |
3081 | if (oops_in_progress) |
3082 | return; | |
3083 | ||
3df0fc5b PZ |
3084 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3085 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3086 | |
dd41f596 | 3087 | debug_show_held_locks(prev); |
e21f5b15 | 3088 | print_modules(); |
dd41f596 IM |
3089 | if (irqs_disabled()) |
3090 | print_irqtrace_events(prev); | |
838225b4 SS |
3091 | |
3092 | if (regs) | |
3093 | show_regs(regs); | |
3094 | else | |
3095 | dump_stack(); | |
dd41f596 | 3096 | } |
1da177e4 | 3097 | |
dd41f596 IM |
3098 | /* |
3099 | * Various schedule()-time debugging checks and statistics: | |
3100 | */ | |
3101 | static inline void schedule_debug(struct task_struct *prev) | |
3102 | { | |
1da177e4 | 3103 | /* |
41a2d6cf | 3104 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3105 | * schedule() atomically, we ignore that path for now. |
3106 | * Otherwise, whine if we are scheduling when we should not be. | |
3107 | */ | |
3f33a7ce | 3108 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 3109 | __schedule_bug(prev); |
b3fbab05 | 3110 | rcu_sleep_check(); |
dd41f596 | 3111 | |
1da177e4 LT |
3112 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3113 | ||
2d72376b | 3114 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3115 | } |
3116 | ||
6cecd084 | 3117 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3118 | { |
61eadef6 | 3119 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 3120 | update_rq_clock(rq); |
6cecd084 | 3121 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3122 | } |
3123 | ||
dd41f596 IM |
3124 | /* |
3125 | * Pick up the highest-prio task: | |
3126 | */ | |
3127 | static inline struct task_struct * | |
b67802ea | 3128 | pick_next_task(struct rq *rq) |
dd41f596 | 3129 | { |
5522d5d5 | 3130 | const struct sched_class *class; |
dd41f596 | 3131 | struct task_struct *p; |
1da177e4 LT |
3132 | |
3133 | /* | |
dd41f596 IM |
3134 | * Optimization: we know that if all tasks are in |
3135 | * the fair class we can call that function directly: | |
1da177e4 | 3136 | */ |
953bfcd1 | 3137 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 3138 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3139 | if (likely(p)) |
3140 | return p; | |
1da177e4 LT |
3141 | } |
3142 | ||
34f971f6 | 3143 | for_each_class(class) { |
fb8d4724 | 3144 | p = class->pick_next_task(rq); |
dd41f596 IM |
3145 | if (p) |
3146 | return p; | |
dd41f596 | 3147 | } |
34f971f6 PZ |
3148 | |
3149 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3150 | } |
1da177e4 | 3151 | |
dd41f596 | 3152 | /* |
c259e01a | 3153 | * __schedule() is the main scheduler function. |
dd41f596 | 3154 | */ |
c259e01a | 3155 | static void __sched __schedule(void) |
dd41f596 IM |
3156 | { |
3157 | struct task_struct *prev, *next; | |
67ca7bde | 3158 | unsigned long *switch_count; |
dd41f596 | 3159 | struct rq *rq; |
31656519 | 3160 | int cpu; |
dd41f596 | 3161 | |
ff743345 PZ |
3162 | need_resched: |
3163 | preempt_disable(); | |
dd41f596 IM |
3164 | cpu = smp_processor_id(); |
3165 | rq = cpu_rq(cpu); | |
25502a6c | 3166 | rcu_note_context_switch(cpu); |
dd41f596 | 3167 | prev = rq->curr; |
dd41f596 | 3168 | |
dd41f596 | 3169 | schedule_debug(prev); |
1da177e4 | 3170 | |
31656519 | 3171 | if (sched_feat(HRTICK)) |
f333fdc9 | 3172 | hrtick_clear(rq); |
8f4d37ec | 3173 | |
05fa785c | 3174 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3175 | |
246d86b5 | 3176 | switch_count = &prev->nivcsw; |
1da177e4 | 3177 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3178 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3179 | prev->state = TASK_RUNNING; |
21aa9af0 | 3180 | } else { |
2acca55e PZ |
3181 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3182 | prev->on_rq = 0; | |
3183 | ||
21aa9af0 | 3184 | /* |
2acca55e PZ |
3185 | * If a worker went to sleep, notify and ask workqueue |
3186 | * whether it wants to wake up a task to maintain | |
3187 | * concurrency. | |
21aa9af0 TH |
3188 | */ |
3189 | if (prev->flags & PF_WQ_WORKER) { | |
3190 | struct task_struct *to_wakeup; | |
3191 | ||
3192 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3193 | if (to_wakeup) | |
3194 | try_to_wake_up_local(to_wakeup); | |
3195 | } | |
21aa9af0 | 3196 | } |
dd41f596 | 3197 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3198 | } |
3199 | ||
3f029d3c | 3200 | pre_schedule(rq, prev); |
f65eda4f | 3201 | |
dd41f596 | 3202 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3203 | idle_balance(cpu, rq); |
1da177e4 | 3204 | |
df1c99d4 | 3205 | put_prev_task(rq, prev); |
b67802ea | 3206 | next = pick_next_task(rq); |
f26f9aff MG |
3207 | clear_tsk_need_resched(prev); |
3208 | rq->skip_clock_update = 0; | |
1da177e4 | 3209 | |
1da177e4 | 3210 | if (likely(prev != next)) { |
1da177e4 LT |
3211 | rq->nr_switches++; |
3212 | rq->curr = next; | |
3213 | ++*switch_count; | |
3214 | ||
dd41f596 | 3215 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3216 | /* |
246d86b5 ON |
3217 | * The context switch have flipped the stack from under us |
3218 | * and restored the local variables which were saved when | |
3219 | * this task called schedule() in the past. prev == current | |
3220 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3221 | */ |
3222 | cpu = smp_processor_id(); | |
3223 | rq = cpu_rq(cpu); | |
1da177e4 | 3224 | } else |
05fa785c | 3225 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3226 | |
3f029d3c | 3227 | post_schedule(rq); |
1da177e4 | 3228 | |
1da177e4 | 3229 | preempt_enable_no_resched(); |
ff743345 | 3230 | if (need_resched()) |
1da177e4 LT |
3231 | goto need_resched; |
3232 | } | |
c259e01a | 3233 | |
9c40cef2 TG |
3234 | static inline void sched_submit_work(struct task_struct *tsk) |
3235 | { | |
3236 | if (!tsk->state) | |
3237 | return; | |
3238 | /* | |
3239 | * If we are going to sleep and we have plugged IO queued, | |
3240 | * make sure to submit it to avoid deadlocks. | |
3241 | */ | |
3242 | if (blk_needs_flush_plug(tsk)) | |
3243 | blk_schedule_flush_plug(tsk); | |
3244 | } | |
3245 | ||
6ebbe7a0 | 3246 | asmlinkage void __sched schedule(void) |
c259e01a | 3247 | { |
9c40cef2 TG |
3248 | struct task_struct *tsk = current; |
3249 | ||
3250 | sched_submit_work(tsk); | |
c259e01a TG |
3251 | __schedule(); |
3252 | } | |
1da177e4 LT |
3253 | EXPORT_SYMBOL(schedule); |
3254 | ||
c08f7829 | 3255 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3256 | |
c6eb3dda PZ |
3257 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3258 | { | |
c6eb3dda | 3259 | if (lock->owner != owner) |
307bf980 | 3260 | return false; |
0d66bf6d PZ |
3261 | |
3262 | /* | |
c6eb3dda PZ |
3263 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3264 | * lock->owner still matches owner, if that fails, owner might | |
3265 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3266 | * ensures the memory stays valid. | |
0d66bf6d | 3267 | */ |
c6eb3dda | 3268 | barrier(); |
0d66bf6d | 3269 | |
307bf980 | 3270 | return owner->on_cpu; |
c6eb3dda | 3271 | } |
0d66bf6d | 3272 | |
c6eb3dda PZ |
3273 | /* |
3274 | * Look out! "owner" is an entirely speculative pointer | |
3275 | * access and not reliable. | |
3276 | */ | |
3277 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3278 | { | |
3279 | if (!sched_feat(OWNER_SPIN)) | |
3280 | return 0; | |
0d66bf6d | 3281 | |
307bf980 | 3282 | rcu_read_lock(); |
c6eb3dda PZ |
3283 | while (owner_running(lock, owner)) { |
3284 | if (need_resched()) | |
307bf980 | 3285 | break; |
0d66bf6d | 3286 | |
335d7afb | 3287 | arch_mutex_cpu_relax(); |
0d66bf6d | 3288 | } |
307bf980 | 3289 | rcu_read_unlock(); |
4b402210 | 3290 | |
c6eb3dda | 3291 | /* |
307bf980 TG |
3292 | * We break out the loop above on need_resched() and when the |
3293 | * owner changed, which is a sign for heavy contention. Return | |
3294 | * success only when lock->owner is NULL. | |
c6eb3dda | 3295 | */ |
307bf980 | 3296 | return lock->owner == NULL; |
0d66bf6d PZ |
3297 | } |
3298 | #endif | |
3299 | ||
1da177e4 LT |
3300 | #ifdef CONFIG_PREEMPT |
3301 | /* | |
2ed6e34f | 3302 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3303 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3304 | * occur there and call schedule directly. |
3305 | */ | |
d1f74e20 | 3306 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3307 | { |
3308 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3309 | |
1da177e4 LT |
3310 | /* |
3311 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3312 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3313 | */ |
beed33a8 | 3314 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3315 | return; |
3316 | ||
3a5c359a | 3317 | do { |
d1f74e20 | 3318 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3319 | __schedule(); |
d1f74e20 | 3320 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3321 | |
3a5c359a AK |
3322 | /* |
3323 | * Check again in case we missed a preemption opportunity | |
3324 | * between schedule and now. | |
3325 | */ | |
3326 | barrier(); | |
5ed0cec0 | 3327 | } while (need_resched()); |
1da177e4 | 3328 | } |
1da177e4 LT |
3329 | EXPORT_SYMBOL(preempt_schedule); |
3330 | ||
3331 | /* | |
2ed6e34f | 3332 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3333 | * off of irq context. |
3334 | * Note, that this is called and return with irqs disabled. This will | |
3335 | * protect us against recursive calling from irq. | |
3336 | */ | |
3337 | asmlinkage void __sched preempt_schedule_irq(void) | |
3338 | { | |
3339 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3340 | |
2ed6e34f | 3341 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3342 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3343 | ||
3a5c359a AK |
3344 | do { |
3345 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3346 | local_irq_enable(); |
c259e01a | 3347 | __schedule(); |
3a5c359a | 3348 | local_irq_disable(); |
3a5c359a | 3349 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3350 | |
3a5c359a AK |
3351 | /* |
3352 | * Check again in case we missed a preemption opportunity | |
3353 | * between schedule and now. | |
3354 | */ | |
3355 | barrier(); | |
5ed0cec0 | 3356 | } while (need_resched()); |
1da177e4 LT |
3357 | } |
3358 | ||
3359 | #endif /* CONFIG_PREEMPT */ | |
3360 | ||
63859d4f | 3361 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3362 | void *key) |
1da177e4 | 3363 | { |
63859d4f | 3364 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3365 | } |
1da177e4 LT |
3366 | EXPORT_SYMBOL(default_wake_function); |
3367 | ||
3368 | /* | |
41a2d6cf IM |
3369 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3370 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3371 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3372 | * | |
3373 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3374 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3375 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3376 | */ | |
78ddb08f | 3377 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3378 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3379 | { |
2e45874c | 3380 | wait_queue_t *curr, *next; |
1da177e4 | 3381 | |
2e45874c | 3382 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3383 | unsigned flags = curr->flags; |
3384 | ||
63859d4f | 3385 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3386 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3387 | break; |
3388 | } | |
3389 | } | |
3390 | ||
3391 | /** | |
3392 | * __wake_up - wake up threads blocked on a waitqueue. | |
3393 | * @q: the waitqueue | |
3394 | * @mode: which threads | |
3395 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3396 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3397 | * |
3398 | * It may be assumed that this function implies a write memory barrier before | |
3399 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3400 | */ |
7ad5b3a5 | 3401 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3402 | int nr_exclusive, void *key) |
1da177e4 LT |
3403 | { |
3404 | unsigned long flags; | |
3405 | ||
3406 | spin_lock_irqsave(&q->lock, flags); | |
3407 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3408 | spin_unlock_irqrestore(&q->lock, flags); | |
3409 | } | |
1da177e4 LT |
3410 | EXPORT_SYMBOL(__wake_up); |
3411 | ||
3412 | /* | |
3413 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3414 | */ | |
7ad5b3a5 | 3415 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3416 | { |
3417 | __wake_up_common(q, mode, 1, 0, NULL); | |
3418 | } | |
22c43c81 | 3419 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3420 | |
4ede816a DL |
3421 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3422 | { | |
3423 | __wake_up_common(q, mode, 1, 0, key); | |
3424 | } | |
bf294b41 | 3425 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3426 | |
1da177e4 | 3427 | /** |
4ede816a | 3428 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3429 | * @q: the waitqueue |
3430 | * @mode: which threads | |
3431 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3432 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3433 | * |
3434 | * The sync wakeup differs that the waker knows that it will schedule | |
3435 | * away soon, so while the target thread will be woken up, it will not | |
3436 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3437 | * with each other. This can prevent needless bouncing between CPUs. | |
3438 | * | |
3439 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3440 | * |
3441 | * It may be assumed that this function implies a write memory barrier before | |
3442 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3443 | */ |
4ede816a DL |
3444 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3445 | int nr_exclusive, void *key) | |
1da177e4 LT |
3446 | { |
3447 | unsigned long flags; | |
7d478721 | 3448 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3449 | |
3450 | if (unlikely(!q)) | |
3451 | return; | |
3452 | ||
3453 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3454 | wake_flags = 0; |
1da177e4 LT |
3455 | |
3456 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3457 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3458 | spin_unlock_irqrestore(&q->lock, flags); |
3459 | } | |
4ede816a DL |
3460 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3461 | ||
3462 | /* | |
3463 | * __wake_up_sync - see __wake_up_sync_key() | |
3464 | */ | |
3465 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3466 | { | |
3467 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3468 | } | |
1da177e4 LT |
3469 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3470 | ||
65eb3dc6 KD |
3471 | /** |
3472 | * complete: - signals a single thread waiting on this completion | |
3473 | * @x: holds the state of this particular completion | |
3474 | * | |
3475 | * This will wake up a single thread waiting on this completion. Threads will be | |
3476 | * awakened in the same order in which they were queued. | |
3477 | * | |
3478 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3479 | * |
3480 | * It may be assumed that this function implies a write memory barrier before | |
3481 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3482 | */ |
b15136e9 | 3483 | void complete(struct completion *x) |
1da177e4 LT |
3484 | { |
3485 | unsigned long flags; | |
3486 | ||
3487 | spin_lock_irqsave(&x->wait.lock, flags); | |
3488 | x->done++; | |
d9514f6c | 3489 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3490 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3491 | } | |
3492 | EXPORT_SYMBOL(complete); | |
3493 | ||
65eb3dc6 KD |
3494 | /** |
3495 | * complete_all: - signals all threads waiting on this completion | |
3496 | * @x: holds the state of this particular completion | |
3497 | * | |
3498 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3499 | * |
3500 | * It may be assumed that this function implies a write memory barrier before | |
3501 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3502 | */ |
b15136e9 | 3503 | void complete_all(struct completion *x) |
1da177e4 LT |
3504 | { |
3505 | unsigned long flags; | |
3506 | ||
3507 | spin_lock_irqsave(&x->wait.lock, flags); | |
3508 | x->done += UINT_MAX/2; | |
d9514f6c | 3509 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3510 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3511 | } | |
3512 | EXPORT_SYMBOL(complete_all); | |
3513 | ||
8cbbe86d AK |
3514 | static inline long __sched |
3515 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3516 | { |
1da177e4 LT |
3517 | if (!x->done) { |
3518 | DECLARE_WAITQUEUE(wait, current); | |
3519 | ||
a93d2f17 | 3520 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3521 | do { |
94d3d824 | 3522 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3523 | timeout = -ERESTARTSYS; |
3524 | break; | |
8cbbe86d AK |
3525 | } |
3526 | __set_current_state(state); | |
1da177e4 LT |
3527 | spin_unlock_irq(&x->wait.lock); |
3528 | timeout = schedule_timeout(timeout); | |
3529 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3530 | } while (!x->done && timeout); |
1da177e4 | 3531 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3532 | if (!x->done) |
3533 | return timeout; | |
1da177e4 LT |
3534 | } |
3535 | x->done--; | |
ea71a546 | 3536 | return timeout ?: 1; |
1da177e4 | 3537 | } |
1da177e4 | 3538 | |
8cbbe86d AK |
3539 | static long __sched |
3540 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3541 | { |
1da177e4 LT |
3542 | might_sleep(); |
3543 | ||
3544 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3545 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3546 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3547 | return timeout; |
3548 | } | |
1da177e4 | 3549 | |
65eb3dc6 KD |
3550 | /** |
3551 | * wait_for_completion: - waits for completion of a task | |
3552 | * @x: holds the state of this particular completion | |
3553 | * | |
3554 | * This waits to be signaled for completion of a specific task. It is NOT | |
3555 | * interruptible and there is no timeout. | |
3556 | * | |
3557 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3558 | * and interrupt capability. Also see complete(). | |
3559 | */ | |
b15136e9 | 3560 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3561 | { |
3562 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3563 | } |
8cbbe86d | 3564 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3565 | |
65eb3dc6 KD |
3566 | /** |
3567 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3568 | * @x: holds the state of this particular completion | |
3569 | * @timeout: timeout value in jiffies | |
3570 | * | |
3571 | * This waits for either a completion of a specific task to be signaled or for a | |
3572 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3573 | * interruptible. | |
c6dc7f05 BF |
3574 | * |
3575 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3576 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3577 | */ |
b15136e9 | 3578 | unsigned long __sched |
8cbbe86d | 3579 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3580 | { |
8cbbe86d | 3581 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3582 | } |
8cbbe86d | 3583 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3584 | |
65eb3dc6 KD |
3585 | /** |
3586 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3587 | * @x: holds the state of this particular completion | |
3588 | * | |
3589 | * This waits for completion of a specific task to be signaled. It is | |
3590 | * interruptible. | |
c6dc7f05 BF |
3591 | * |
3592 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3593 | */ |
8cbbe86d | 3594 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3595 | { |
51e97990 AK |
3596 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3597 | if (t == -ERESTARTSYS) | |
3598 | return t; | |
3599 | return 0; | |
0fec171c | 3600 | } |
8cbbe86d | 3601 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3602 | |
65eb3dc6 KD |
3603 | /** |
3604 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3605 | * @x: holds the state of this particular completion | |
3606 | * @timeout: timeout value in jiffies | |
3607 | * | |
3608 | * This waits for either a completion of a specific task to be signaled or for a | |
3609 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3610 | * |
3611 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3612 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3613 | */ |
6bf41237 | 3614 | long __sched |
8cbbe86d AK |
3615 | wait_for_completion_interruptible_timeout(struct completion *x, |
3616 | unsigned long timeout) | |
0fec171c | 3617 | { |
8cbbe86d | 3618 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3619 | } |
8cbbe86d | 3620 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3621 | |
65eb3dc6 KD |
3622 | /** |
3623 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3624 | * @x: holds the state of this particular completion | |
3625 | * | |
3626 | * This waits to be signaled for completion of a specific task. It can be | |
3627 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3628 | * |
3629 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3630 | */ |
009e577e MW |
3631 | int __sched wait_for_completion_killable(struct completion *x) |
3632 | { | |
3633 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3634 | if (t == -ERESTARTSYS) | |
3635 | return t; | |
3636 | return 0; | |
3637 | } | |
3638 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3639 | ||
0aa12fb4 SW |
3640 | /** |
3641 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3642 | * @x: holds the state of this particular completion | |
3643 | * @timeout: timeout value in jiffies | |
3644 | * | |
3645 | * This waits for either a completion of a specific task to be | |
3646 | * signaled or for a specified timeout to expire. It can be | |
3647 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3648 | * |
3649 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3650 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3651 | */ |
6bf41237 | 3652 | long __sched |
0aa12fb4 SW |
3653 | wait_for_completion_killable_timeout(struct completion *x, |
3654 | unsigned long timeout) | |
3655 | { | |
3656 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3657 | } | |
3658 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3659 | ||
be4de352 DC |
3660 | /** |
3661 | * try_wait_for_completion - try to decrement a completion without blocking | |
3662 | * @x: completion structure | |
3663 | * | |
3664 | * Returns: 0 if a decrement cannot be done without blocking | |
3665 | * 1 if a decrement succeeded. | |
3666 | * | |
3667 | * If a completion is being used as a counting completion, | |
3668 | * attempt to decrement the counter without blocking. This | |
3669 | * enables us to avoid waiting if the resource the completion | |
3670 | * is protecting is not available. | |
3671 | */ | |
3672 | bool try_wait_for_completion(struct completion *x) | |
3673 | { | |
7539a3b3 | 3674 | unsigned long flags; |
be4de352 DC |
3675 | int ret = 1; |
3676 | ||
7539a3b3 | 3677 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3678 | if (!x->done) |
3679 | ret = 0; | |
3680 | else | |
3681 | x->done--; | |
7539a3b3 | 3682 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3683 | return ret; |
3684 | } | |
3685 | EXPORT_SYMBOL(try_wait_for_completion); | |
3686 | ||
3687 | /** | |
3688 | * completion_done - Test to see if a completion has any waiters | |
3689 | * @x: completion structure | |
3690 | * | |
3691 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3692 | * 1 if there are no waiters. | |
3693 | * | |
3694 | */ | |
3695 | bool completion_done(struct completion *x) | |
3696 | { | |
7539a3b3 | 3697 | unsigned long flags; |
be4de352 DC |
3698 | int ret = 1; |
3699 | ||
7539a3b3 | 3700 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3701 | if (!x->done) |
3702 | ret = 0; | |
7539a3b3 | 3703 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3704 | return ret; |
3705 | } | |
3706 | EXPORT_SYMBOL(completion_done); | |
3707 | ||
8cbbe86d AK |
3708 | static long __sched |
3709 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3710 | { |
0fec171c IM |
3711 | unsigned long flags; |
3712 | wait_queue_t wait; | |
3713 | ||
3714 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3715 | |
8cbbe86d | 3716 | __set_current_state(state); |
1da177e4 | 3717 | |
8cbbe86d AK |
3718 | spin_lock_irqsave(&q->lock, flags); |
3719 | __add_wait_queue(q, &wait); | |
3720 | spin_unlock(&q->lock); | |
3721 | timeout = schedule_timeout(timeout); | |
3722 | spin_lock_irq(&q->lock); | |
3723 | __remove_wait_queue(q, &wait); | |
3724 | spin_unlock_irqrestore(&q->lock, flags); | |
3725 | ||
3726 | return timeout; | |
3727 | } | |
3728 | ||
3729 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3730 | { | |
3731 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3732 | } |
1da177e4 LT |
3733 | EXPORT_SYMBOL(interruptible_sleep_on); |
3734 | ||
0fec171c | 3735 | long __sched |
95cdf3b7 | 3736 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3737 | { |
8cbbe86d | 3738 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3739 | } |
1da177e4 LT |
3740 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3741 | ||
0fec171c | 3742 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3743 | { |
8cbbe86d | 3744 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3745 | } |
1da177e4 LT |
3746 | EXPORT_SYMBOL(sleep_on); |
3747 | ||
0fec171c | 3748 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3749 | { |
8cbbe86d | 3750 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3751 | } |
1da177e4 LT |
3752 | EXPORT_SYMBOL(sleep_on_timeout); |
3753 | ||
b29739f9 IM |
3754 | #ifdef CONFIG_RT_MUTEXES |
3755 | ||
3756 | /* | |
3757 | * rt_mutex_setprio - set the current priority of a task | |
3758 | * @p: task | |
3759 | * @prio: prio value (kernel-internal form) | |
3760 | * | |
3761 | * This function changes the 'effective' priority of a task. It does | |
3762 | * not touch ->normal_prio like __setscheduler(). | |
3763 | * | |
3764 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3765 | */ | |
36c8b586 | 3766 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3767 | { |
83b699ed | 3768 | int oldprio, on_rq, running; |
70b97a7f | 3769 | struct rq *rq; |
83ab0aa0 | 3770 | const struct sched_class *prev_class; |
b29739f9 IM |
3771 | |
3772 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3773 | ||
0122ec5b | 3774 | rq = __task_rq_lock(p); |
b29739f9 | 3775 | |
a8027073 | 3776 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3777 | oldprio = p->prio; |
83ab0aa0 | 3778 | prev_class = p->sched_class; |
fd2f4419 | 3779 | on_rq = p->on_rq; |
051a1d1a | 3780 | running = task_current(rq, p); |
0e1f3483 | 3781 | if (on_rq) |
69be72c1 | 3782 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3783 | if (running) |
3784 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3785 | |
3786 | if (rt_prio(prio)) | |
3787 | p->sched_class = &rt_sched_class; | |
3788 | else | |
3789 | p->sched_class = &fair_sched_class; | |
3790 | ||
b29739f9 IM |
3791 | p->prio = prio; |
3792 | ||
0e1f3483 HS |
3793 | if (running) |
3794 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3795 | if (on_rq) |
371fd7e7 | 3796 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3797 | |
da7a735e | 3798 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3799 | __task_rq_unlock(rq); |
b29739f9 IM |
3800 | } |
3801 | ||
3802 | #endif | |
3803 | ||
36c8b586 | 3804 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3805 | { |
dd41f596 | 3806 | int old_prio, delta, on_rq; |
1da177e4 | 3807 | unsigned long flags; |
70b97a7f | 3808 | struct rq *rq; |
1da177e4 LT |
3809 | |
3810 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3811 | return; | |
3812 | /* | |
3813 | * We have to be careful, if called from sys_setpriority(), | |
3814 | * the task might be in the middle of scheduling on another CPU. | |
3815 | */ | |
3816 | rq = task_rq_lock(p, &flags); | |
3817 | /* | |
3818 | * The RT priorities are set via sched_setscheduler(), but we still | |
3819 | * allow the 'normal' nice value to be set - but as expected | |
3820 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3821 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3822 | */ |
e05606d3 | 3823 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3824 | p->static_prio = NICE_TO_PRIO(nice); |
3825 | goto out_unlock; | |
3826 | } | |
fd2f4419 | 3827 | on_rq = p->on_rq; |
c09595f6 | 3828 | if (on_rq) |
69be72c1 | 3829 | dequeue_task(rq, p, 0); |
1da177e4 | 3830 | |
1da177e4 | 3831 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3832 | set_load_weight(p); |
b29739f9 IM |
3833 | old_prio = p->prio; |
3834 | p->prio = effective_prio(p); | |
3835 | delta = p->prio - old_prio; | |
1da177e4 | 3836 | |
dd41f596 | 3837 | if (on_rq) { |
371fd7e7 | 3838 | enqueue_task(rq, p, 0); |
1da177e4 | 3839 | /* |
d5f9f942 AM |
3840 | * If the task increased its priority or is running and |
3841 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3842 | */ |
d5f9f942 | 3843 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3844 | resched_task(rq->curr); |
3845 | } | |
3846 | out_unlock: | |
0122ec5b | 3847 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3848 | } |
1da177e4 LT |
3849 | EXPORT_SYMBOL(set_user_nice); |
3850 | ||
e43379f1 MM |
3851 | /* |
3852 | * can_nice - check if a task can reduce its nice value | |
3853 | * @p: task | |
3854 | * @nice: nice value | |
3855 | */ | |
36c8b586 | 3856 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3857 | { |
024f4747 MM |
3858 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3859 | int nice_rlim = 20 - nice; | |
48f24c4d | 3860 | |
78d7d407 | 3861 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3862 | capable(CAP_SYS_NICE)); |
3863 | } | |
3864 | ||
1da177e4 LT |
3865 | #ifdef __ARCH_WANT_SYS_NICE |
3866 | ||
3867 | /* | |
3868 | * sys_nice - change the priority of the current process. | |
3869 | * @increment: priority increment | |
3870 | * | |
3871 | * sys_setpriority is a more generic, but much slower function that | |
3872 | * does similar things. | |
3873 | */ | |
5add95d4 | 3874 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3875 | { |
48f24c4d | 3876 | long nice, retval; |
1da177e4 LT |
3877 | |
3878 | /* | |
3879 | * Setpriority might change our priority at the same moment. | |
3880 | * We don't have to worry. Conceptually one call occurs first | |
3881 | * and we have a single winner. | |
3882 | */ | |
e43379f1 MM |
3883 | if (increment < -40) |
3884 | increment = -40; | |
1da177e4 LT |
3885 | if (increment > 40) |
3886 | increment = 40; | |
3887 | ||
2b8f836f | 3888 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3889 | if (nice < -20) |
3890 | nice = -20; | |
3891 | if (nice > 19) | |
3892 | nice = 19; | |
3893 | ||
e43379f1 MM |
3894 | if (increment < 0 && !can_nice(current, nice)) |
3895 | return -EPERM; | |
3896 | ||
1da177e4 LT |
3897 | retval = security_task_setnice(current, nice); |
3898 | if (retval) | |
3899 | return retval; | |
3900 | ||
3901 | set_user_nice(current, nice); | |
3902 | return 0; | |
3903 | } | |
3904 | ||
3905 | #endif | |
3906 | ||
3907 | /** | |
3908 | * task_prio - return the priority value of a given task. | |
3909 | * @p: the task in question. | |
3910 | * | |
3911 | * This is the priority value as seen by users in /proc. | |
3912 | * RT tasks are offset by -200. Normal tasks are centered | |
3913 | * around 0, value goes from -16 to +15. | |
3914 | */ | |
36c8b586 | 3915 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3916 | { |
3917 | return p->prio - MAX_RT_PRIO; | |
3918 | } | |
3919 | ||
3920 | /** | |
3921 | * task_nice - return the nice value of a given task. | |
3922 | * @p: the task in question. | |
3923 | */ | |
36c8b586 | 3924 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3925 | { |
3926 | return TASK_NICE(p); | |
3927 | } | |
150d8bed | 3928 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3929 | |
3930 | /** | |
3931 | * idle_cpu - is a given cpu idle currently? | |
3932 | * @cpu: the processor in question. | |
3933 | */ | |
3934 | int idle_cpu(int cpu) | |
3935 | { | |
908a3283 TG |
3936 | struct rq *rq = cpu_rq(cpu); |
3937 | ||
3938 | if (rq->curr != rq->idle) | |
3939 | return 0; | |
3940 | ||
3941 | if (rq->nr_running) | |
3942 | return 0; | |
3943 | ||
3944 | #ifdef CONFIG_SMP | |
3945 | if (!llist_empty(&rq->wake_list)) | |
3946 | return 0; | |
3947 | #endif | |
3948 | ||
3949 | return 1; | |
1da177e4 LT |
3950 | } |
3951 | ||
1da177e4 LT |
3952 | /** |
3953 | * idle_task - return the idle task for a given cpu. | |
3954 | * @cpu: the processor in question. | |
3955 | */ | |
36c8b586 | 3956 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3957 | { |
3958 | return cpu_rq(cpu)->idle; | |
3959 | } | |
3960 | ||
3961 | /** | |
3962 | * find_process_by_pid - find a process with a matching PID value. | |
3963 | * @pid: the pid in question. | |
3964 | */ | |
a9957449 | 3965 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3966 | { |
228ebcbe | 3967 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3968 | } |
3969 | ||
3970 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3971 | static void |
3972 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3973 | { |
1da177e4 LT |
3974 | p->policy = policy; |
3975 | p->rt_priority = prio; | |
b29739f9 IM |
3976 | p->normal_prio = normal_prio(p); |
3977 | /* we are holding p->pi_lock already */ | |
3978 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3979 | if (rt_prio(p->prio)) |
3980 | p->sched_class = &rt_sched_class; | |
3981 | else | |
3982 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3983 | set_load_weight(p); |
1da177e4 LT |
3984 | } |
3985 | ||
c69e8d9c DH |
3986 | /* |
3987 | * check the target process has a UID that matches the current process's | |
3988 | */ | |
3989 | static bool check_same_owner(struct task_struct *p) | |
3990 | { | |
3991 | const struct cred *cred = current_cred(), *pcred; | |
3992 | bool match; | |
3993 | ||
3994 | rcu_read_lock(); | |
3995 | pcred = __task_cred(p); | |
b0e77598 SH |
3996 | if (cred->user->user_ns == pcred->user->user_ns) |
3997 | match = (cred->euid == pcred->euid || | |
3998 | cred->euid == pcred->uid); | |
3999 | else | |
4000 | match = false; | |
c69e8d9c DH |
4001 | rcu_read_unlock(); |
4002 | return match; | |
4003 | } | |
4004 | ||
961ccddd | 4005 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4006 | const struct sched_param *param, bool user) |
1da177e4 | 4007 | { |
83b699ed | 4008 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4009 | unsigned long flags; |
83ab0aa0 | 4010 | const struct sched_class *prev_class; |
70b97a7f | 4011 | struct rq *rq; |
ca94c442 | 4012 | int reset_on_fork; |
1da177e4 | 4013 | |
66e5393a SR |
4014 | /* may grab non-irq protected spin_locks */ |
4015 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4016 | recheck: |
4017 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4018 | if (policy < 0) { |
4019 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4020 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4021 | } else { |
4022 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4023 | policy &= ~SCHED_RESET_ON_FORK; | |
4024 | ||
4025 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4026 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4027 | policy != SCHED_IDLE) | |
4028 | return -EINVAL; | |
4029 | } | |
4030 | ||
1da177e4 LT |
4031 | /* |
4032 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4033 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4034 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4035 | */ |
4036 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4037 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4038 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4039 | return -EINVAL; |
e05606d3 | 4040 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4041 | return -EINVAL; |
4042 | ||
37e4ab3f OC |
4043 | /* |
4044 | * Allow unprivileged RT tasks to decrease priority: | |
4045 | */ | |
961ccddd | 4046 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4047 | if (rt_policy(policy)) { |
a44702e8 ON |
4048 | unsigned long rlim_rtprio = |
4049 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4050 | |
4051 | /* can't set/change the rt policy */ | |
4052 | if (policy != p->policy && !rlim_rtprio) | |
4053 | return -EPERM; | |
4054 | ||
4055 | /* can't increase priority */ | |
4056 | if (param->sched_priority > p->rt_priority && | |
4057 | param->sched_priority > rlim_rtprio) | |
4058 | return -EPERM; | |
4059 | } | |
c02aa73b | 4060 | |
dd41f596 | 4061 | /* |
c02aa73b DH |
4062 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4063 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4064 | */ |
c02aa73b DH |
4065 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4066 | if (!can_nice(p, TASK_NICE(p))) | |
4067 | return -EPERM; | |
4068 | } | |
5fe1d75f | 4069 | |
37e4ab3f | 4070 | /* can't change other user's priorities */ |
c69e8d9c | 4071 | if (!check_same_owner(p)) |
37e4ab3f | 4072 | return -EPERM; |
ca94c442 LP |
4073 | |
4074 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4075 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4076 | return -EPERM; | |
37e4ab3f | 4077 | } |
1da177e4 | 4078 | |
725aad24 | 4079 | if (user) { |
b0ae1981 | 4080 | retval = security_task_setscheduler(p); |
725aad24 JF |
4081 | if (retval) |
4082 | return retval; | |
4083 | } | |
4084 | ||
b29739f9 IM |
4085 | /* |
4086 | * make sure no PI-waiters arrive (or leave) while we are | |
4087 | * changing the priority of the task: | |
0122ec5b | 4088 | * |
25985edc | 4089 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4090 | * runqueue lock must be held. |
4091 | */ | |
0122ec5b | 4092 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 4093 | |
34f971f6 PZ |
4094 | /* |
4095 | * Changing the policy of the stop threads its a very bad idea | |
4096 | */ | |
4097 | if (p == rq->stop) { | |
0122ec5b | 4098 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
4099 | return -EINVAL; |
4100 | } | |
4101 | ||
a51e9198 DF |
4102 | /* |
4103 | * If not changing anything there's no need to proceed further: | |
4104 | */ | |
4105 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
4106 | param->sched_priority == p->rt_priority))) { | |
4107 | ||
4108 | __task_rq_unlock(rq); | |
4109 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4110 | return 0; | |
4111 | } | |
4112 | ||
dc61b1d6 PZ |
4113 | #ifdef CONFIG_RT_GROUP_SCHED |
4114 | if (user) { | |
4115 | /* | |
4116 | * Do not allow realtime tasks into groups that have no runtime | |
4117 | * assigned. | |
4118 | */ | |
4119 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4120 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4121 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4122 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4123 | return -EPERM; |
4124 | } | |
4125 | } | |
4126 | #endif | |
4127 | ||
1da177e4 LT |
4128 | /* recheck policy now with rq lock held */ |
4129 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4130 | policy = oldpolicy = -1; | |
0122ec5b | 4131 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4132 | goto recheck; |
4133 | } | |
fd2f4419 | 4134 | on_rq = p->on_rq; |
051a1d1a | 4135 | running = task_current(rq, p); |
0e1f3483 | 4136 | if (on_rq) |
2e1cb74a | 4137 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4138 | if (running) |
4139 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4140 | |
ca94c442 LP |
4141 | p->sched_reset_on_fork = reset_on_fork; |
4142 | ||
1da177e4 | 4143 | oldprio = p->prio; |
83ab0aa0 | 4144 | prev_class = p->sched_class; |
dd41f596 | 4145 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4146 | |
0e1f3483 HS |
4147 | if (running) |
4148 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4149 | if (on_rq) |
dd41f596 | 4150 | activate_task(rq, p, 0); |
cb469845 | 4151 | |
da7a735e | 4152 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4153 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4154 | |
95e02ca9 TG |
4155 | rt_mutex_adjust_pi(p); |
4156 | ||
1da177e4 LT |
4157 | return 0; |
4158 | } | |
961ccddd RR |
4159 | |
4160 | /** | |
4161 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4162 | * @p: the task in question. | |
4163 | * @policy: new policy. | |
4164 | * @param: structure containing the new RT priority. | |
4165 | * | |
4166 | * NOTE that the task may be already dead. | |
4167 | */ | |
4168 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4169 | const struct sched_param *param) |
961ccddd RR |
4170 | { |
4171 | return __sched_setscheduler(p, policy, param, true); | |
4172 | } | |
1da177e4 LT |
4173 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4174 | ||
961ccddd RR |
4175 | /** |
4176 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4177 | * @p: the task in question. | |
4178 | * @policy: new policy. | |
4179 | * @param: structure containing the new RT priority. | |
4180 | * | |
4181 | * Just like sched_setscheduler, only don't bother checking if the | |
4182 | * current context has permission. For example, this is needed in | |
4183 | * stop_machine(): we create temporary high priority worker threads, | |
4184 | * but our caller might not have that capability. | |
4185 | */ | |
4186 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4187 | const struct sched_param *param) |
961ccddd RR |
4188 | { |
4189 | return __sched_setscheduler(p, policy, param, false); | |
4190 | } | |
4191 | ||
95cdf3b7 IM |
4192 | static int |
4193 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4194 | { |
1da177e4 LT |
4195 | struct sched_param lparam; |
4196 | struct task_struct *p; | |
36c8b586 | 4197 | int retval; |
1da177e4 LT |
4198 | |
4199 | if (!param || pid < 0) | |
4200 | return -EINVAL; | |
4201 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4202 | return -EFAULT; | |
5fe1d75f ON |
4203 | |
4204 | rcu_read_lock(); | |
4205 | retval = -ESRCH; | |
1da177e4 | 4206 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4207 | if (p != NULL) |
4208 | retval = sched_setscheduler(p, policy, &lparam); | |
4209 | rcu_read_unlock(); | |
36c8b586 | 4210 | |
1da177e4 LT |
4211 | return retval; |
4212 | } | |
4213 | ||
4214 | /** | |
4215 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4216 | * @pid: the pid in question. | |
4217 | * @policy: new policy. | |
4218 | * @param: structure containing the new RT priority. | |
4219 | */ | |
5add95d4 HC |
4220 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4221 | struct sched_param __user *, param) | |
1da177e4 | 4222 | { |
c21761f1 JB |
4223 | /* negative values for policy are not valid */ |
4224 | if (policy < 0) | |
4225 | return -EINVAL; | |
4226 | ||
1da177e4 LT |
4227 | return do_sched_setscheduler(pid, policy, param); |
4228 | } | |
4229 | ||
4230 | /** | |
4231 | * sys_sched_setparam - set/change the RT priority of a thread | |
4232 | * @pid: the pid in question. | |
4233 | * @param: structure containing the new RT priority. | |
4234 | */ | |
5add95d4 | 4235 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4236 | { |
4237 | return do_sched_setscheduler(pid, -1, param); | |
4238 | } | |
4239 | ||
4240 | /** | |
4241 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4242 | * @pid: the pid in question. | |
4243 | */ | |
5add95d4 | 4244 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4245 | { |
36c8b586 | 4246 | struct task_struct *p; |
3a5c359a | 4247 | int retval; |
1da177e4 LT |
4248 | |
4249 | if (pid < 0) | |
3a5c359a | 4250 | return -EINVAL; |
1da177e4 LT |
4251 | |
4252 | retval = -ESRCH; | |
5fe85be0 | 4253 | rcu_read_lock(); |
1da177e4 LT |
4254 | p = find_process_by_pid(pid); |
4255 | if (p) { | |
4256 | retval = security_task_getscheduler(p); | |
4257 | if (!retval) | |
ca94c442 LP |
4258 | retval = p->policy |
4259 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4260 | } |
5fe85be0 | 4261 | rcu_read_unlock(); |
1da177e4 LT |
4262 | return retval; |
4263 | } | |
4264 | ||
4265 | /** | |
ca94c442 | 4266 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4267 | * @pid: the pid in question. |
4268 | * @param: structure containing the RT priority. | |
4269 | */ | |
5add95d4 | 4270 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4271 | { |
4272 | struct sched_param lp; | |
36c8b586 | 4273 | struct task_struct *p; |
3a5c359a | 4274 | int retval; |
1da177e4 LT |
4275 | |
4276 | if (!param || pid < 0) | |
3a5c359a | 4277 | return -EINVAL; |
1da177e4 | 4278 | |
5fe85be0 | 4279 | rcu_read_lock(); |
1da177e4 LT |
4280 | p = find_process_by_pid(pid); |
4281 | retval = -ESRCH; | |
4282 | if (!p) | |
4283 | goto out_unlock; | |
4284 | ||
4285 | retval = security_task_getscheduler(p); | |
4286 | if (retval) | |
4287 | goto out_unlock; | |
4288 | ||
4289 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4290 | rcu_read_unlock(); |
1da177e4 LT |
4291 | |
4292 | /* | |
4293 | * This one might sleep, we cannot do it with a spinlock held ... | |
4294 | */ | |
4295 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4296 | ||
1da177e4 LT |
4297 | return retval; |
4298 | ||
4299 | out_unlock: | |
5fe85be0 | 4300 | rcu_read_unlock(); |
1da177e4 LT |
4301 | return retval; |
4302 | } | |
4303 | ||
96f874e2 | 4304 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4305 | { |
5a16f3d3 | 4306 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4307 | struct task_struct *p; |
4308 | int retval; | |
1da177e4 | 4309 | |
95402b38 | 4310 | get_online_cpus(); |
23f5d142 | 4311 | rcu_read_lock(); |
1da177e4 LT |
4312 | |
4313 | p = find_process_by_pid(pid); | |
4314 | if (!p) { | |
23f5d142 | 4315 | rcu_read_unlock(); |
95402b38 | 4316 | put_online_cpus(); |
1da177e4 LT |
4317 | return -ESRCH; |
4318 | } | |
4319 | ||
23f5d142 | 4320 | /* Prevent p going away */ |
1da177e4 | 4321 | get_task_struct(p); |
23f5d142 | 4322 | rcu_read_unlock(); |
1da177e4 | 4323 | |
5a16f3d3 RR |
4324 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4325 | retval = -ENOMEM; | |
4326 | goto out_put_task; | |
4327 | } | |
4328 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4329 | retval = -ENOMEM; | |
4330 | goto out_free_cpus_allowed; | |
4331 | } | |
1da177e4 | 4332 | retval = -EPERM; |
b0e77598 | 4333 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
4334 | goto out_unlock; |
4335 | ||
b0ae1981 | 4336 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4337 | if (retval) |
4338 | goto out_unlock; | |
4339 | ||
5a16f3d3 RR |
4340 | cpuset_cpus_allowed(p, cpus_allowed); |
4341 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4342 | again: |
5a16f3d3 | 4343 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4344 | |
8707d8b8 | 4345 | if (!retval) { |
5a16f3d3 RR |
4346 | cpuset_cpus_allowed(p, cpus_allowed); |
4347 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4348 | /* |
4349 | * We must have raced with a concurrent cpuset | |
4350 | * update. Just reset the cpus_allowed to the | |
4351 | * cpuset's cpus_allowed | |
4352 | */ | |
5a16f3d3 | 4353 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4354 | goto again; |
4355 | } | |
4356 | } | |
1da177e4 | 4357 | out_unlock: |
5a16f3d3 RR |
4358 | free_cpumask_var(new_mask); |
4359 | out_free_cpus_allowed: | |
4360 | free_cpumask_var(cpus_allowed); | |
4361 | out_put_task: | |
1da177e4 | 4362 | put_task_struct(p); |
95402b38 | 4363 | put_online_cpus(); |
1da177e4 LT |
4364 | return retval; |
4365 | } | |
4366 | ||
4367 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4368 | struct cpumask *new_mask) |
1da177e4 | 4369 | { |
96f874e2 RR |
4370 | if (len < cpumask_size()) |
4371 | cpumask_clear(new_mask); | |
4372 | else if (len > cpumask_size()) | |
4373 | len = cpumask_size(); | |
4374 | ||
1da177e4 LT |
4375 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4376 | } | |
4377 | ||
4378 | /** | |
4379 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4380 | * @pid: pid of the process | |
4381 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4382 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4383 | */ | |
5add95d4 HC |
4384 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4385 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4386 | { |
5a16f3d3 | 4387 | cpumask_var_t new_mask; |
1da177e4 LT |
4388 | int retval; |
4389 | ||
5a16f3d3 RR |
4390 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4391 | return -ENOMEM; | |
1da177e4 | 4392 | |
5a16f3d3 RR |
4393 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4394 | if (retval == 0) | |
4395 | retval = sched_setaffinity(pid, new_mask); | |
4396 | free_cpumask_var(new_mask); | |
4397 | return retval; | |
1da177e4 LT |
4398 | } |
4399 | ||
96f874e2 | 4400 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4401 | { |
36c8b586 | 4402 | struct task_struct *p; |
31605683 | 4403 | unsigned long flags; |
1da177e4 | 4404 | int retval; |
1da177e4 | 4405 | |
95402b38 | 4406 | get_online_cpus(); |
23f5d142 | 4407 | rcu_read_lock(); |
1da177e4 LT |
4408 | |
4409 | retval = -ESRCH; | |
4410 | p = find_process_by_pid(pid); | |
4411 | if (!p) | |
4412 | goto out_unlock; | |
4413 | ||
e7834f8f DQ |
4414 | retval = security_task_getscheduler(p); |
4415 | if (retval) | |
4416 | goto out_unlock; | |
4417 | ||
013fdb80 | 4418 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4419 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4420 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4421 | |
4422 | out_unlock: | |
23f5d142 | 4423 | rcu_read_unlock(); |
95402b38 | 4424 | put_online_cpus(); |
1da177e4 | 4425 | |
9531b62f | 4426 | return retval; |
1da177e4 LT |
4427 | } |
4428 | ||
4429 | /** | |
4430 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4431 | * @pid: pid of the process | |
4432 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4433 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4434 | */ | |
5add95d4 HC |
4435 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4436 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4437 | { |
4438 | int ret; | |
f17c8607 | 4439 | cpumask_var_t mask; |
1da177e4 | 4440 | |
84fba5ec | 4441 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4442 | return -EINVAL; |
4443 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4444 | return -EINVAL; |
4445 | ||
f17c8607 RR |
4446 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4447 | return -ENOMEM; | |
1da177e4 | 4448 | |
f17c8607 RR |
4449 | ret = sched_getaffinity(pid, mask); |
4450 | if (ret == 0) { | |
8bc037fb | 4451 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4452 | |
4453 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4454 | ret = -EFAULT; |
4455 | else | |
cd3d8031 | 4456 | ret = retlen; |
f17c8607 RR |
4457 | } |
4458 | free_cpumask_var(mask); | |
1da177e4 | 4459 | |
f17c8607 | 4460 | return ret; |
1da177e4 LT |
4461 | } |
4462 | ||
4463 | /** | |
4464 | * sys_sched_yield - yield the current processor to other threads. | |
4465 | * | |
dd41f596 IM |
4466 | * This function yields the current CPU to other tasks. If there are no |
4467 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4468 | */ |
5add95d4 | 4469 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4470 | { |
70b97a7f | 4471 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4472 | |
2d72376b | 4473 | schedstat_inc(rq, yld_count); |
4530d7ab | 4474 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4475 | |
4476 | /* | |
4477 | * Since we are going to call schedule() anyway, there's | |
4478 | * no need to preempt or enable interrupts: | |
4479 | */ | |
4480 | __release(rq->lock); | |
8a25d5de | 4481 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4482 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4483 | preempt_enable_no_resched(); |
4484 | ||
4485 | schedule(); | |
4486 | ||
4487 | return 0; | |
4488 | } | |
4489 | ||
d86ee480 PZ |
4490 | static inline int should_resched(void) |
4491 | { | |
4492 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4493 | } | |
4494 | ||
e7b38404 | 4495 | static void __cond_resched(void) |
1da177e4 | 4496 | { |
e7aaaa69 | 4497 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4498 | __schedule(); |
e7aaaa69 | 4499 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4500 | } |
4501 | ||
02b67cc3 | 4502 | int __sched _cond_resched(void) |
1da177e4 | 4503 | { |
d86ee480 | 4504 | if (should_resched()) { |
1da177e4 LT |
4505 | __cond_resched(); |
4506 | return 1; | |
4507 | } | |
4508 | return 0; | |
4509 | } | |
02b67cc3 | 4510 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4511 | |
4512 | /* | |
613afbf8 | 4513 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4514 | * call schedule, and on return reacquire the lock. |
4515 | * | |
41a2d6cf | 4516 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4517 | * operations here to prevent schedule() from being called twice (once via |
4518 | * spin_unlock(), once by hand). | |
4519 | */ | |
613afbf8 | 4520 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4521 | { |
d86ee480 | 4522 | int resched = should_resched(); |
6df3cecb JK |
4523 | int ret = 0; |
4524 | ||
f607c668 PZ |
4525 | lockdep_assert_held(lock); |
4526 | ||
95c354fe | 4527 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4528 | spin_unlock(lock); |
d86ee480 | 4529 | if (resched) |
95c354fe NP |
4530 | __cond_resched(); |
4531 | else | |
4532 | cpu_relax(); | |
6df3cecb | 4533 | ret = 1; |
1da177e4 | 4534 | spin_lock(lock); |
1da177e4 | 4535 | } |
6df3cecb | 4536 | return ret; |
1da177e4 | 4537 | } |
613afbf8 | 4538 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4539 | |
613afbf8 | 4540 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4541 | { |
4542 | BUG_ON(!in_softirq()); | |
4543 | ||
d86ee480 | 4544 | if (should_resched()) { |
98d82567 | 4545 | local_bh_enable(); |
1da177e4 LT |
4546 | __cond_resched(); |
4547 | local_bh_disable(); | |
4548 | return 1; | |
4549 | } | |
4550 | return 0; | |
4551 | } | |
613afbf8 | 4552 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4553 | |
1da177e4 LT |
4554 | /** |
4555 | * yield - yield the current processor to other threads. | |
4556 | * | |
72fd4a35 | 4557 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4558 | * thread runnable and calls sys_sched_yield(). |
4559 | */ | |
4560 | void __sched yield(void) | |
4561 | { | |
4562 | set_current_state(TASK_RUNNING); | |
4563 | sys_sched_yield(); | |
4564 | } | |
1da177e4 LT |
4565 | EXPORT_SYMBOL(yield); |
4566 | ||
d95f4122 MG |
4567 | /** |
4568 | * yield_to - yield the current processor to another thread in | |
4569 | * your thread group, or accelerate that thread toward the | |
4570 | * processor it's on. | |
16addf95 RD |
4571 | * @p: target task |
4572 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4573 | * |
4574 | * It's the caller's job to ensure that the target task struct | |
4575 | * can't go away on us before we can do any checks. | |
4576 | * | |
4577 | * Returns true if we indeed boosted the target task. | |
4578 | */ | |
4579 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4580 | { | |
4581 | struct task_struct *curr = current; | |
4582 | struct rq *rq, *p_rq; | |
4583 | unsigned long flags; | |
4584 | bool yielded = 0; | |
4585 | ||
4586 | local_irq_save(flags); | |
4587 | rq = this_rq(); | |
4588 | ||
4589 | again: | |
4590 | p_rq = task_rq(p); | |
4591 | double_rq_lock(rq, p_rq); | |
4592 | while (task_rq(p) != p_rq) { | |
4593 | double_rq_unlock(rq, p_rq); | |
4594 | goto again; | |
4595 | } | |
4596 | ||
4597 | if (!curr->sched_class->yield_to_task) | |
4598 | goto out; | |
4599 | ||
4600 | if (curr->sched_class != p->sched_class) | |
4601 | goto out; | |
4602 | ||
4603 | if (task_running(p_rq, p) || p->state) | |
4604 | goto out; | |
4605 | ||
4606 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4607 | if (yielded) { |
d95f4122 | 4608 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4609 | /* |
4610 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4611 | * fairness. | |
4612 | */ | |
4613 | if (preempt && rq != p_rq) | |
4614 | resched_task(p_rq->curr); | |
916671c0 MG |
4615 | } else { |
4616 | /* | |
4617 | * We might have set it in task_yield_fair(), but are | |
4618 | * not going to schedule(), so don't want to skip | |
4619 | * the next update. | |
4620 | */ | |
4621 | rq->skip_clock_update = 0; | |
6d1cafd8 | 4622 | } |
d95f4122 MG |
4623 | |
4624 | out: | |
4625 | double_rq_unlock(rq, p_rq); | |
4626 | local_irq_restore(flags); | |
4627 | ||
4628 | if (yielded) | |
4629 | schedule(); | |
4630 | ||
4631 | return yielded; | |
4632 | } | |
4633 | EXPORT_SYMBOL_GPL(yield_to); | |
4634 | ||
1da177e4 | 4635 | /* |
41a2d6cf | 4636 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4637 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4638 | */ |
4639 | void __sched io_schedule(void) | |
4640 | { | |
54d35f29 | 4641 | struct rq *rq = raw_rq(); |
1da177e4 | 4642 | |
0ff92245 | 4643 | delayacct_blkio_start(); |
1da177e4 | 4644 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4645 | blk_flush_plug(current); |
8f0dfc34 | 4646 | current->in_iowait = 1; |
1da177e4 | 4647 | schedule(); |
8f0dfc34 | 4648 | current->in_iowait = 0; |
1da177e4 | 4649 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4650 | delayacct_blkio_end(); |
1da177e4 | 4651 | } |
1da177e4 LT |
4652 | EXPORT_SYMBOL(io_schedule); |
4653 | ||
4654 | long __sched io_schedule_timeout(long timeout) | |
4655 | { | |
54d35f29 | 4656 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4657 | long ret; |
4658 | ||
0ff92245 | 4659 | delayacct_blkio_start(); |
1da177e4 | 4660 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4661 | blk_flush_plug(current); |
8f0dfc34 | 4662 | current->in_iowait = 1; |
1da177e4 | 4663 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4664 | current->in_iowait = 0; |
1da177e4 | 4665 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4666 | delayacct_blkio_end(); |
1da177e4 LT |
4667 | return ret; |
4668 | } | |
4669 | ||
4670 | /** | |
4671 | * sys_sched_get_priority_max - return maximum RT priority. | |
4672 | * @policy: scheduling class. | |
4673 | * | |
4674 | * this syscall returns the maximum rt_priority that can be used | |
4675 | * by a given scheduling class. | |
4676 | */ | |
5add95d4 | 4677 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4678 | { |
4679 | int ret = -EINVAL; | |
4680 | ||
4681 | switch (policy) { | |
4682 | case SCHED_FIFO: | |
4683 | case SCHED_RR: | |
4684 | ret = MAX_USER_RT_PRIO-1; | |
4685 | break; | |
4686 | case SCHED_NORMAL: | |
b0a9499c | 4687 | case SCHED_BATCH: |
dd41f596 | 4688 | case SCHED_IDLE: |
1da177e4 LT |
4689 | ret = 0; |
4690 | break; | |
4691 | } | |
4692 | return ret; | |
4693 | } | |
4694 | ||
4695 | /** | |
4696 | * sys_sched_get_priority_min - return minimum RT priority. | |
4697 | * @policy: scheduling class. | |
4698 | * | |
4699 | * this syscall returns the minimum rt_priority that can be used | |
4700 | * by a given scheduling class. | |
4701 | */ | |
5add95d4 | 4702 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4703 | { |
4704 | int ret = -EINVAL; | |
4705 | ||
4706 | switch (policy) { | |
4707 | case SCHED_FIFO: | |
4708 | case SCHED_RR: | |
4709 | ret = 1; | |
4710 | break; | |
4711 | case SCHED_NORMAL: | |
b0a9499c | 4712 | case SCHED_BATCH: |
dd41f596 | 4713 | case SCHED_IDLE: |
1da177e4 LT |
4714 | ret = 0; |
4715 | } | |
4716 | return ret; | |
4717 | } | |
4718 | ||
4719 | /** | |
4720 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4721 | * @pid: pid of the process. | |
4722 | * @interval: userspace pointer to the timeslice value. | |
4723 | * | |
4724 | * this syscall writes the default timeslice value of a given process | |
4725 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4726 | */ | |
17da2bd9 | 4727 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4728 | struct timespec __user *, interval) |
1da177e4 | 4729 | { |
36c8b586 | 4730 | struct task_struct *p; |
a4ec24b4 | 4731 | unsigned int time_slice; |
dba091b9 TG |
4732 | unsigned long flags; |
4733 | struct rq *rq; | |
3a5c359a | 4734 | int retval; |
1da177e4 | 4735 | struct timespec t; |
1da177e4 LT |
4736 | |
4737 | if (pid < 0) | |
3a5c359a | 4738 | return -EINVAL; |
1da177e4 LT |
4739 | |
4740 | retval = -ESRCH; | |
1a551ae7 | 4741 | rcu_read_lock(); |
1da177e4 LT |
4742 | p = find_process_by_pid(pid); |
4743 | if (!p) | |
4744 | goto out_unlock; | |
4745 | ||
4746 | retval = security_task_getscheduler(p); | |
4747 | if (retval) | |
4748 | goto out_unlock; | |
4749 | ||
dba091b9 TG |
4750 | rq = task_rq_lock(p, &flags); |
4751 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4752 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4753 | |
1a551ae7 | 4754 | rcu_read_unlock(); |
a4ec24b4 | 4755 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4756 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4757 | return retval; |
3a5c359a | 4758 | |
1da177e4 | 4759 | out_unlock: |
1a551ae7 | 4760 | rcu_read_unlock(); |
1da177e4 LT |
4761 | return retval; |
4762 | } | |
4763 | ||
7c731e0a | 4764 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4765 | |
82a1fcb9 | 4766 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4767 | { |
1da177e4 | 4768 | unsigned long free = 0; |
36c8b586 | 4769 | unsigned state; |
1da177e4 | 4770 | |
1da177e4 | 4771 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4772 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4773 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4774 | #if BITS_PER_LONG == 32 |
1da177e4 | 4775 | if (state == TASK_RUNNING) |
3df0fc5b | 4776 | printk(KERN_CONT " running "); |
1da177e4 | 4777 | else |
3df0fc5b | 4778 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4779 | #else |
4780 | if (state == TASK_RUNNING) | |
3df0fc5b | 4781 | printk(KERN_CONT " running task "); |
1da177e4 | 4782 | else |
3df0fc5b | 4783 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4784 | #endif |
4785 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4786 | free = stack_not_used(p); |
1da177e4 | 4787 | #endif |
3df0fc5b | 4788 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
07cde260 | 4789 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), |
aa47b7e0 | 4790 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4791 | |
5fb5e6de | 4792 | show_stack(p, NULL); |
1da177e4 LT |
4793 | } |
4794 | ||
e59e2ae2 | 4795 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4796 | { |
36c8b586 | 4797 | struct task_struct *g, *p; |
1da177e4 | 4798 | |
4bd77321 | 4799 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4800 | printk(KERN_INFO |
4801 | " task PC stack pid father\n"); | |
1da177e4 | 4802 | #else |
3df0fc5b PZ |
4803 | printk(KERN_INFO |
4804 | " task PC stack pid father\n"); | |
1da177e4 | 4805 | #endif |
510f5acc | 4806 | rcu_read_lock(); |
1da177e4 LT |
4807 | do_each_thread(g, p) { |
4808 | /* | |
4809 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4810 | * console might take a lot of time: |
1da177e4 LT |
4811 | */ |
4812 | touch_nmi_watchdog(); | |
39bc89fd | 4813 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4814 | sched_show_task(p); |
1da177e4 LT |
4815 | } while_each_thread(g, p); |
4816 | ||
04c9167f JF |
4817 | touch_all_softlockup_watchdogs(); |
4818 | ||
dd41f596 IM |
4819 | #ifdef CONFIG_SCHED_DEBUG |
4820 | sysrq_sched_debug_show(); | |
4821 | #endif | |
510f5acc | 4822 | rcu_read_unlock(); |
e59e2ae2 IM |
4823 | /* |
4824 | * Only show locks if all tasks are dumped: | |
4825 | */ | |
93335a21 | 4826 | if (!state_filter) |
e59e2ae2 | 4827 | debug_show_all_locks(); |
1da177e4 LT |
4828 | } |
4829 | ||
1df21055 IM |
4830 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4831 | { | |
dd41f596 | 4832 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4833 | } |
4834 | ||
f340c0d1 IM |
4835 | /** |
4836 | * init_idle - set up an idle thread for a given CPU | |
4837 | * @idle: task in question | |
4838 | * @cpu: cpu the idle task belongs to | |
4839 | * | |
4840 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4841 | * flag, to make booting more robust. | |
4842 | */ | |
5c1e1767 | 4843 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4844 | { |
70b97a7f | 4845 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4846 | unsigned long flags; |
4847 | ||
05fa785c | 4848 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4849 | |
dd41f596 | 4850 | __sched_fork(idle); |
06b83b5f | 4851 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4852 | idle->se.exec_start = sched_clock(); |
4853 | ||
1e1b6c51 | 4854 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4855 | /* |
4856 | * We're having a chicken and egg problem, even though we are | |
4857 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4858 | * lockdep check in task_group() will fail. | |
4859 | * | |
4860 | * Similar case to sched_fork(). / Alternatively we could | |
4861 | * use task_rq_lock() here and obtain the other rq->lock. | |
4862 | * | |
4863 | * Silence PROVE_RCU | |
4864 | */ | |
4865 | rcu_read_lock(); | |
dd41f596 | 4866 | __set_task_cpu(idle, cpu); |
6506cf6c | 4867 | rcu_read_unlock(); |
1da177e4 | 4868 | |
1da177e4 | 4869 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4870 | #if defined(CONFIG_SMP) |
4871 | idle->on_cpu = 1; | |
4866cde0 | 4872 | #endif |
05fa785c | 4873 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4874 | |
4875 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4876 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4877 | |
dd41f596 IM |
4878 | /* |
4879 | * The idle tasks have their own, simple scheduling class: | |
4880 | */ | |
4881 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4882 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
4883 | #if defined(CONFIG_SMP) |
4884 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4885 | #endif | |
19978ca6 IM |
4886 | } |
4887 | ||
1da177e4 | 4888 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4889 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4890 | { | |
4891 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4892 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4893 | |
4894 | cpumask_copy(&p->cpus_allowed, new_mask); | |
4895 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
1e1b6c51 KM |
4896 | } |
4897 | ||
1da177e4 LT |
4898 | /* |
4899 | * This is how migration works: | |
4900 | * | |
969c7921 TH |
4901 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4902 | * stop_one_cpu(). | |
4903 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4904 | * off the CPU) | |
4905 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4906 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4907 | * it and puts it into the right queue. |
969c7921 TH |
4908 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4909 | * is done. | |
1da177e4 LT |
4910 | */ |
4911 | ||
4912 | /* | |
4913 | * Change a given task's CPU affinity. Migrate the thread to a | |
4914 | * proper CPU and schedule it away if the CPU it's executing on | |
4915 | * is removed from the allowed bitmask. | |
4916 | * | |
4917 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4918 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4919 | * call is not atomic; no spinlocks may be held. |
4920 | */ | |
96f874e2 | 4921 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4922 | { |
4923 | unsigned long flags; | |
70b97a7f | 4924 | struct rq *rq; |
969c7921 | 4925 | unsigned int dest_cpu; |
48f24c4d | 4926 | int ret = 0; |
1da177e4 LT |
4927 | |
4928 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4929 | |
db44fc01 YZ |
4930 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4931 | goto out; | |
4932 | ||
6ad4c188 | 4933 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4934 | ret = -EINVAL; |
4935 | goto out; | |
4936 | } | |
4937 | ||
db44fc01 | 4938 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4939 | ret = -EINVAL; |
4940 | goto out; | |
4941 | } | |
4942 | ||
1e1b6c51 | 4943 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4944 | |
1da177e4 | 4945 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4946 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4947 | goto out; |
4948 | ||
969c7921 | 4949 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4950 | if (p->on_rq) { |
969c7921 | 4951 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4952 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4953 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4954 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4955 | tlb_migrate_finish(p->mm); |
4956 | return 0; | |
4957 | } | |
4958 | out: | |
0122ec5b | 4959 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4960 | |
1da177e4 LT |
4961 | return ret; |
4962 | } | |
cd8ba7cd | 4963 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4964 | |
4965 | /* | |
41a2d6cf | 4966 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4967 | * this because either it can't run here any more (set_cpus_allowed() |
4968 | * away from this CPU, or CPU going down), or because we're | |
4969 | * attempting to rebalance this task on exec (sched_exec). | |
4970 | * | |
4971 | * So we race with normal scheduler movements, but that's OK, as long | |
4972 | * as the task is no longer on this CPU. | |
efc30814 KK |
4973 | * |
4974 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4975 | */ |
efc30814 | 4976 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4977 | { |
70b97a7f | 4978 | struct rq *rq_dest, *rq_src; |
e2912009 | 4979 | int ret = 0; |
1da177e4 | 4980 | |
e761b772 | 4981 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4982 | return ret; |
1da177e4 LT |
4983 | |
4984 | rq_src = cpu_rq(src_cpu); | |
4985 | rq_dest = cpu_rq(dest_cpu); | |
4986 | ||
0122ec5b | 4987 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4988 | double_rq_lock(rq_src, rq_dest); |
4989 | /* Already moved. */ | |
4990 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4991 | goto done; |
1da177e4 | 4992 | /* Affinity changed (again). */ |
fa17b507 | 4993 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4994 | goto fail; |
1da177e4 | 4995 | |
e2912009 PZ |
4996 | /* |
4997 | * If we're not on a rq, the next wake-up will ensure we're | |
4998 | * placed properly. | |
4999 | */ | |
fd2f4419 | 5000 | if (p->on_rq) { |
2e1cb74a | 5001 | deactivate_task(rq_src, p, 0); |
e2912009 | 5002 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5003 | activate_task(rq_dest, p, 0); |
15afe09b | 5004 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5005 | } |
b1e38734 | 5006 | done: |
efc30814 | 5007 | ret = 1; |
b1e38734 | 5008 | fail: |
1da177e4 | 5009 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 5010 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 5011 | return ret; |
1da177e4 LT |
5012 | } |
5013 | ||
5014 | /* | |
969c7921 TH |
5015 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5016 | * and performs thread migration by bumping thread off CPU then | |
5017 | * 'pushing' onto another runqueue. | |
1da177e4 | 5018 | */ |
969c7921 | 5019 | static int migration_cpu_stop(void *data) |
1da177e4 | 5020 | { |
969c7921 | 5021 | struct migration_arg *arg = data; |
f7b4cddc | 5022 | |
969c7921 TH |
5023 | /* |
5024 | * The original target cpu might have gone down and we might | |
5025 | * be on another cpu but it doesn't matter. | |
5026 | */ | |
f7b4cddc | 5027 | local_irq_disable(); |
969c7921 | 5028 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5029 | local_irq_enable(); |
1da177e4 | 5030 | return 0; |
f7b4cddc ON |
5031 | } |
5032 | ||
1da177e4 | 5033 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5034 | |
054b9108 | 5035 | /* |
48c5ccae PZ |
5036 | * Ensures that the idle task is using init_mm right before its cpu goes |
5037 | * offline. | |
054b9108 | 5038 | */ |
48c5ccae | 5039 | void idle_task_exit(void) |
1da177e4 | 5040 | { |
48c5ccae | 5041 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5042 | |
48c5ccae | 5043 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5044 | |
48c5ccae PZ |
5045 | if (mm != &init_mm) |
5046 | switch_mm(mm, &init_mm, current); | |
5047 | mmdrop(mm); | |
1da177e4 LT |
5048 | } |
5049 | ||
5050 | /* | |
5051 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5052 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5053 | * for performance reasons the counter is not stricly tracking tasks to | |
5054 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5055 | * to keep the global sum constant after CPU-down: | |
5056 | */ | |
70b97a7f | 5057 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5058 | { |
6ad4c188 | 5059 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5060 | |
1da177e4 LT |
5061 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5062 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5063 | } |
5064 | ||
dd41f596 | 5065 | /* |
48c5ccae | 5066 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5067 | */ |
48c5ccae | 5068 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5069 | { |
48c5ccae PZ |
5070 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5071 | rq->calc_load_active = 0; | |
1da177e4 LT |
5072 | } |
5073 | ||
48f24c4d | 5074 | /* |
48c5ccae PZ |
5075 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5076 | * try_to_wake_up()->select_task_rq(). | |
5077 | * | |
5078 | * Called with rq->lock held even though we'er in stop_machine() and | |
5079 | * there's no concurrency possible, we hold the required locks anyway | |
5080 | * because of lock validation efforts. | |
1da177e4 | 5081 | */ |
48c5ccae | 5082 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5083 | { |
70b97a7f | 5084 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5085 | struct task_struct *next, *stop = rq->stop; |
5086 | int dest_cpu; | |
1da177e4 LT |
5087 | |
5088 | /* | |
48c5ccae PZ |
5089 | * Fudge the rq selection such that the below task selection loop |
5090 | * doesn't get stuck on the currently eligible stop task. | |
5091 | * | |
5092 | * We're currently inside stop_machine() and the rq is either stuck | |
5093 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5094 | * either way we should never end up calling schedule() until we're | |
5095 | * done here. | |
1da177e4 | 5096 | */ |
48c5ccae | 5097 | rq->stop = NULL; |
48f24c4d | 5098 | |
8cb120d3 PT |
5099 | /* Ensure any throttled groups are reachable by pick_next_task */ |
5100 | unthrottle_offline_cfs_rqs(rq); | |
5101 | ||
dd41f596 | 5102 | for ( ; ; ) { |
48c5ccae PZ |
5103 | /* |
5104 | * There's this thread running, bail when that's the only | |
5105 | * remaining thread. | |
5106 | */ | |
5107 | if (rq->nr_running == 1) | |
dd41f596 | 5108 | break; |
48c5ccae | 5109 | |
b67802ea | 5110 | next = pick_next_task(rq); |
48c5ccae | 5111 | BUG_ON(!next); |
79c53799 | 5112 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5113 | |
48c5ccae PZ |
5114 | /* Find suitable destination for @next, with force if needed. */ |
5115 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5116 | raw_spin_unlock(&rq->lock); | |
5117 | ||
5118 | __migrate_task(next, dead_cpu, dest_cpu); | |
5119 | ||
5120 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5121 | } |
dce48a84 | 5122 | |
48c5ccae | 5123 | rq->stop = stop; |
dce48a84 | 5124 | } |
48c5ccae | 5125 | |
1da177e4 LT |
5126 | #endif /* CONFIG_HOTPLUG_CPU */ |
5127 | ||
e692ab53 NP |
5128 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5129 | ||
5130 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5131 | { |
5132 | .procname = "sched_domain", | |
c57baf1e | 5133 | .mode = 0555, |
e0361851 | 5134 | }, |
56992309 | 5135 | {} |
e692ab53 NP |
5136 | }; |
5137 | ||
5138 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5139 | { |
5140 | .procname = "kernel", | |
c57baf1e | 5141 | .mode = 0555, |
e0361851 AD |
5142 | .child = sd_ctl_dir, |
5143 | }, | |
56992309 | 5144 | {} |
e692ab53 NP |
5145 | }; |
5146 | ||
5147 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5148 | { | |
5149 | struct ctl_table *entry = | |
5cf9f062 | 5150 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5151 | |
e692ab53 NP |
5152 | return entry; |
5153 | } | |
5154 | ||
6382bc90 MM |
5155 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5156 | { | |
cd790076 | 5157 | struct ctl_table *entry; |
6382bc90 | 5158 | |
cd790076 MM |
5159 | /* |
5160 | * In the intermediate directories, both the child directory and | |
5161 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5162 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5163 | * static strings and all have proc handlers. |
5164 | */ | |
5165 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5166 | if (entry->child) |
5167 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5168 | if (entry->proc_handler == NULL) |
5169 | kfree(entry->procname); | |
5170 | } | |
6382bc90 MM |
5171 | |
5172 | kfree(*tablep); | |
5173 | *tablep = NULL; | |
5174 | } | |
5175 | ||
e692ab53 | 5176 | static void |
e0361851 | 5177 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5178 | const char *procname, void *data, int maxlen, |
36fcb589 | 5179 | umode_t mode, proc_handler *proc_handler) |
e692ab53 | 5180 | { |
e692ab53 NP |
5181 | entry->procname = procname; |
5182 | entry->data = data; | |
5183 | entry->maxlen = maxlen; | |
5184 | entry->mode = mode; | |
5185 | entry->proc_handler = proc_handler; | |
5186 | } | |
5187 | ||
5188 | static struct ctl_table * | |
5189 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5190 | { | |
a5d8c348 | 5191 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5192 | |
ad1cdc1d MM |
5193 | if (table == NULL) |
5194 | return NULL; | |
5195 | ||
e0361851 | 5196 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5197 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5198 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5199 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5200 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5201 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5202 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5203 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5204 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5205 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5206 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5207 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5208 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5209 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5210 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5211 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5212 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5213 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5214 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5215 | &sd->cache_nice_tries, |
5216 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5217 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5218 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5219 | set_table_entry(&table[11], "name", sd->name, |
5220 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5221 | /* &table[12] is terminator */ | |
e692ab53 NP |
5222 | |
5223 | return table; | |
5224 | } | |
5225 | ||
9a4e7159 | 5226 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5227 | { |
5228 | struct ctl_table *entry, *table; | |
5229 | struct sched_domain *sd; | |
5230 | int domain_num = 0, i; | |
5231 | char buf[32]; | |
5232 | ||
5233 | for_each_domain(cpu, sd) | |
5234 | domain_num++; | |
5235 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5236 | if (table == NULL) |
5237 | return NULL; | |
e692ab53 NP |
5238 | |
5239 | i = 0; | |
5240 | for_each_domain(cpu, sd) { | |
5241 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5242 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5243 | entry->mode = 0555; |
e692ab53 NP |
5244 | entry->child = sd_alloc_ctl_domain_table(sd); |
5245 | entry++; | |
5246 | i++; | |
5247 | } | |
5248 | return table; | |
5249 | } | |
5250 | ||
5251 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5252 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5253 | { |
6ad4c188 | 5254 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5255 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5256 | char buf[32]; | |
5257 | ||
7378547f MM |
5258 | WARN_ON(sd_ctl_dir[0].child); |
5259 | sd_ctl_dir[0].child = entry; | |
5260 | ||
ad1cdc1d MM |
5261 | if (entry == NULL) |
5262 | return; | |
5263 | ||
6ad4c188 | 5264 | for_each_possible_cpu(i) { |
e692ab53 | 5265 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5266 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5267 | entry->mode = 0555; |
e692ab53 | 5268 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5269 | entry++; |
e692ab53 | 5270 | } |
7378547f MM |
5271 | |
5272 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5273 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5274 | } | |
6382bc90 | 5275 | |
7378547f | 5276 | /* may be called multiple times per register */ |
6382bc90 MM |
5277 | static void unregister_sched_domain_sysctl(void) |
5278 | { | |
7378547f MM |
5279 | if (sd_sysctl_header) |
5280 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5281 | sd_sysctl_header = NULL; |
7378547f MM |
5282 | if (sd_ctl_dir[0].child) |
5283 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5284 | } |
e692ab53 | 5285 | #else |
6382bc90 MM |
5286 | static void register_sched_domain_sysctl(void) |
5287 | { | |
5288 | } | |
5289 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5290 | { |
5291 | } | |
5292 | #endif | |
5293 | ||
1f11eb6a GH |
5294 | static void set_rq_online(struct rq *rq) |
5295 | { | |
5296 | if (!rq->online) { | |
5297 | const struct sched_class *class; | |
5298 | ||
c6c4927b | 5299 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5300 | rq->online = 1; |
5301 | ||
5302 | for_each_class(class) { | |
5303 | if (class->rq_online) | |
5304 | class->rq_online(rq); | |
5305 | } | |
5306 | } | |
5307 | } | |
5308 | ||
5309 | static void set_rq_offline(struct rq *rq) | |
5310 | { | |
5311 | if (rq->online) { | |
5312 | const struct sched_class *class; | |
5313 | ||
5314 | for_each_class(class) { | |
5315 | if (class->rq_offline) | |
5316 | class->rq_offline(rq); | |
5317 | } | |
5318 | ||
c6c4927b | 5319 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5320 | rq->online = 0; |
5321 | } | |
5322 | } | |
5323 | ||
1da177e4 LT |
5324 | /* |
5325 | * migration_call - callback that gets triggered when a CPU is added. | |
5326 | * Here we can start up the necessary migration thread for the new CPU. | |
5327 | */ | |
48f24c4d IM |
5328 | static int __cpuinit |
5329 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5330 | { |
48f24c4d | 5331 | int cpu = (long)hcpu; |
1da177e4 | 5332 | unsigned long flags; |
969c7921 | 5333 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5334 | |
48c5ccae | 5335 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5336 | |
1da177e4 | 5337 | case CPU_UP_PREPARE: |
a468d389 | 5338 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5339 | break; |
48f24c4d | 5340 | |
1da177e4 | 5341 | case CPU_ONLINE: |
1f94ef59 | 5342 | /* Update our root-domain */ |
05fa785c | 5343 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5344 | if (rq->rd) { |
c6c4927b | 5345 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5346 | |
5347 | set_rq_online(rq); | |
1f94ef59 | 5348 | } |
05fa785c | 5349 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5350 | break; |
48f24c4d | 5351 | |
1da177e4 | 5352 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5353 | case CPU_DYING: |
317f3941 | 5354 | sched_ttwu_pending(); |
57d885fe | 5355 | /* Update our root-domain */ |
05fa785c | 5356 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5357 | if (rq->rd) { |
c6c4927b | 5358 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5359 | set_rq_offline(rq); |
57d885fe | 5360 | } |
48c5ccae PZ |
5361 | migrate_tasks(cpu); |
5362 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5363 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5364 | |
5365 | migrate_nr_uninterruptible(rq); | |
5366 | calc_global_load_remove(rq); | |
57d885fe | 5367 | break; |
1da177e4 LT |
5368 | #endif |
5369 | } | |
49c022e6 PZ |
5370 | |
5371 | update_max_interval(); | |
5372 | ||
1da177e4 LT |
5373 | return NOTIFY_OK; |
5374 | } | |
5375 | ||
f38b0820 PM |
5376 | /* |
5377 | * Register at high priority so that task migration (migrate_all_tasks) | |
5378 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5379 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5380 | */ |
26c2143b | 5381 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5382 | .notifier_call = migration_call, |
50a323b7 | 5383 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5384 | }; |
5385 | ||
3a101d05 TH |
5386 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5387 | unsigned long action, void *hcpu) | |
5388 | { | |
5389 | switch (action & ~CPU_TASKS_FROZEN) { | |
5390 | case CPU_ONLINE: | |
5391 | case CPU_DOWN_FAILED: | |
5392 | set_cpu_active((long)hcpu, true); | |
5393 | return NOTIFY_OK; | |
5394 | default: | |
5395 | return NOTIFY_DONE; | |
5396 | } | |
5397 | } | |
5398 | ||
5399 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5400 | unsigned long action, void *hcpu) | |
5401 | { | |
5402 | switch (action & ~CPU_TASKS_FROZEN) { | |
5403 | case CPU_DOWN_PREPARE: | |
5404 | set_cpu_active((long)hcpu, false); | |
5405 | return NOTIFY_OK; | |
5406 | default: | |
5407 | return NOTIFY_DONE; | |
5408 | } | |
5409 | } | |
5410 | ||
7babe8db | 5411 | static int __init migration_init(void) |
1da177e4 LT |
5412 | { |
5413 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5414 | int err; |
48f24c4d | 5415 | |
3a101d05 | 5416 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5417 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5418 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5419 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5420 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5421 | |
3a101d05 TH |
5422 | /* Register cpu active notifiers */ |
5423 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5424 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5425 | ||
a004cd42 | 5426 | return 0; |
1da177e4 | 5427 | } |
7babe8db | 5428 | early_initcall(migration_init); |
1da177e4 LT |
5429 | #endif |
5430 | ||
5431 | #ifdef CONFIG_SMP | |
476f3534 | 5432 | |
4cb98839 PZ |
5433 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5434 | ||
3e9830dc | 5435 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5436 | |
f6630114 MT |
5437 | static __read_mostly int sched_domain_debug_enabled; |
5438 | ||
5439 | static int __init sched_domain_debug_setup(char *str) | |
5440 | { | |
5441 | sched_domain_debug_enabled = 1; | |
5442 | ||
5443 | return 0; | |
5444 | } | |
5445 | early_param("sched_debug", sched_domain_debug_setup); | |
5446 | ||
7c16ec58 | 5447 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5448 | struct cpumask *groupmask) |
1da177e4 | 5449 | { |
4dcf6aff | 5450 | struct sched_group *group = sd->groups; |
434d53b0 | 5451 | char str[256]; |
1da177e4 | 5452 | |
968ea6d8 | 5453 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5454 | cpumask_clear(groupmask); |
4dcf6aff IM |
5455 | |
5456 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5457 | ||
5458 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5459 | printk("does not load-balance\n"); |
4dcf6aff | 5460 | if (sd->parent) |
3df0fc5b PZ |
5461 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5462 | " has parent"); | |
4dcf6aff | 5463 | return -1; |
41c7ce9a NP |
5464 | } |
5465 | ||
3df0fc5b | 5466 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5467 | |
758b2cdc | 5468 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5469 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5470 | "CPU%d\n", cpu); | |
4dcf6aff | 5471 | } |
758b2cdc | 5472 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5473 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5474 | " CPU%d\n", cpu); | |
4dcf6aff | 5475 | } |
1da177e4 | 5476 | |
4dcf6aff | 5477 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5478 | do { |
4dcf6aff | 5479 | if (!group) { |
3df0fc5b PZ |
5480 | printk("\n"); |
5481 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5482 | break; |
5483 | } | |
5484 | ||
9c3f75cb | 5485 | if (!group->sgp->power) { |
3df0fc5b PZ |
5486 | printk(KERN_CONT "\n"); |
5487 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5488 | "set\n"); | |
4dcf6aff IM |
5489 | break; |
5490 | } | |
1da177e4 | 5491 | |
758b2cdc | 5492 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5493 | printk(KERN_CONT "\n"); |
5494 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5495 | break; |
5496 | } | |
1da177e4 | 5497 | |
758b2cdc | 5498 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5499 | printk(KERN_CONT "\n"); |
5500 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5501 | break; |
5502 | } | |
1da177e4 | 5503 | |
758b2cdc | 5504 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5505 | |
968ea6d8 | 5506 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5507 | |
3df0fc5b | 5508 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5509 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5510 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5511 | group->sgp->power); |
381512cf | 5512 | } |
1da177e4 | 5513 | |
4dcf6aff IM |
5514 | group = group->next; |
5515 | } while (group != sd->groups); | |
3df0fc5b | 5516 | printk(KERN_CONT "\n"); |
1da177e4 | 5517 | |
758b2cdc | 5518 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5519 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5520 | |
758b2cdc RR |
5521 | if (sd->parent && |
5522 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5523 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5524 | "of domain->span\n"); | |
4dcf6aff IM |
5525 | return 0; |
5526 | } | |
1da177e4 | 5527 | |
4dcf6aff IM |
5528 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5529 | { | |
5530 | int level = 0; | |
1da177e4 | 5531 | |
f6630114 MT |
5532 | if (!sched_domain_debug_enabled) |
5533 | return; | |
5534 | ||
4dcf6aff IM |
5535 | if (!sd) { |
5536 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5537 | return; | |
5538 | } | |
1da177e4 | 5539 | |
4dcf6aff IM |
5540 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5541 | ||
5542 | for (;;) { | |
4cb98839 | 5543 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5544 | break; |
1da177e4 LT |
5545 | level++; |
5546 | sd = sd->parent; | |
33859f7f | 5547 | if (!sd) |
4dcf6aff IM |
5548 | break; |
5549 | } | |
1da177e4 | 5550 | } |
6d6bc0ad | 5551 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5552 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5553 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5554 | |
1a20ff27 | 5555 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5556 | { |
758b2cdc | 5557 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5558 | return 1; |
5559 | ||
5560 | /* Following flags need at least 2 groups */ | |
5561 | if (sd->flags & (SD_LOAD_BALANCE | | |
5562 | SD_BALANCE_NEWIDLE | | |
5563 | SD_BALANCE_FORK | | |
89c4710e SS |
5564 | SD_BALANCE_EXEC | |
5565 | SD_SHARE_CPUPOWER | | |
5566 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5567 | if (sd->groups != sd->groups->next) |
5568 | return 0; | |
5569 | } | |
5570 | ||
5571 | /* Following flags don't use groups */ | |
c88d5910 | 5572 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5573 | return 0; |
5574 | ||
5575 | return 1; | |
5576 | } | |
5577 | ||
48f24c4d IM |
5578 | static int |
5579 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5580 | { |
5581 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5582 | ||
5583 | if (sd_degenerate(parent)) | |
5584 | return 1; | |
5585 | ||
758b2cdc | 5586 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5587 | return 0; |
5588 | ||
245af2c7 SS |
5589 | /* Flags needing groups don't count if only 1 group in parent */ |
5590 | if (parent->groups == parent->groups->next) { | |
5591 | pflags &= ~(SD_LOAD_BALANCE | | |
5592 | SD_BALANCE_NEWIDLE | | |
5593 | SD_BALANCE_FORK | | |
89c4710e SS |
5594 | SD_BALANCE_EXEC | |
5595 | SD_SHARE_CPUPOWER | | |
5596 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5597 | if (nr_node_ids == 1) |
5598 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5599 | } |
5600 | if (~cflags & pflags) | |
5601 | return 0; | |
5602 | ||
5603 | return 1; | |
5604 | } | |
5605 | ||
dce840a0 | 5606 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5607 | { |
dce840a0 | 5608 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5609 | |
68e74568 | 5610 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5611 | free_cpumask_var(rd->rto_mask); |
5612 | free_cpumask_var(rd->online); | |
5613 | free_cpumask_var(rd->span); | |
5614 | kfree(rd); | |
5615 | } | |
5616 | ||
57d885fe GH |
5617 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5618 | { | |
a0490fa3 | 5619 | struct root_domain *old_rd = NULL; |
57d885fe | 5620 | unsigned long flags; |
57d885fe | 5621 | |
05fa785c | 5622 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5623 | |
5624 | if (rq->rd) { | |
a0490fa3 | 5625 | old_rd = rq->rd; |
57d885fe | 5626 | |
c6c4927b | 5627 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5628 | set_rq_offline(rq); |
57d885fe | 5629 | |
c6c4927b | 5630 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5631 | |
a0490fa3 IM |
5632 | /* |
5633 | * If we dont want to free the old_rt yet then | |
5634 | * set old_rd to NULL to skip the freeing later | |
5635 | * in this function: | |
5636 | */ | |
5637 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5638 | old_rd = NULL; | |
57d885fe GH |
5639 | } |
5640 | ||
5641 | atomic_inc(&rd->refcount); | |
5642 | rq->rd = rd; | |
5643 | ||
c6c4927b | 5644 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5645 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5646 | set_rq_online(rq); |
57d885fe | 5647 | |
05fa785c | 5648 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5649 | |
5650 | if (old_rd) | |
dce840a0 | 5651 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5652 | } |
5653 | ||
68c38fc3 | 5654 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5655 | { |
5656 | memset(rd, 0, sizeof(*rd)); | |
5657 | ||
68c38fc3 | 5658 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5659 | goto out; |
68c38fc3 | 5660 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5661 | goto free_span; |
68c38fc3 | 5662 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5663 | goto free_online; |
6e0534f2 | 5664 | |
68c38fc3 | 5665 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5666 | goto free_rto_mask; |
c6c4927b | 5667 | return 0; |
6e0534f2 | 5668 | |
68e74568 RR |
5669 | free_rto_mask: |
5670 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5671 | free_online: |
5672 | free_cpumask_var(rd->online); | |
5673 | free_span: | |
5674 | free_cpumask_var(rd->span); | |
0c910d28 | 5675 | out: |
c6c4927b | 5676 | return -ENOMEM; |
57d885fe GH |
5677 | } |
5678 | ||
029632fb PZ |
5679 | /* |
5680 | * By default the system creates a single root-domain with all cpus as | |
5681 | * members (mimicking the global state we have today). | |
5682 | */ | |
5683 | struct root_domain def_root_domain; | |
5684 | ||
57d885fe GH |
5685 | static void init_defrootdomain(void) |
5686 | { | |
68c38fc3 | 5687 | init_rootdomain(&def_root_domain); |
c6c4927b | 5688 | |
57d885fe GH |
5689 | atomic_set(&def_root_domain.refcount, 1); |
5690 | } | |
5691 | ||
dc938520 | 5692 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5693 | { |
5694 | struct root_domain *rd; | |
5695 | ||
5696 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5697 | if (!rd) | |
5698 | return NULL; | |
5699 | ||
68c38fc3 | 5700 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5701 | kfree(rd); |
5702 | return NULL; | |
5703 | } | |
57d885fe GH |
5704 | |
5705 | return rd; | |
5706 | } | |
5707 | ||
e3589f6c PZ |
5708 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5709 | { | |
5710 | struct sched_group *tmp, *first; | |
5711 | ||
5712 | if (!sg) | |
5713 | return; | |
5714 | ||
5715 | first = sg; | |
5716 | do { | |
5717 | tmp = sg->next; | |
5718 | ||
5719 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5720 | kfree(sg->sgp); | |
5721 | ||
5722 | kfree(sg); | |
5723 | sg = tmp; | |
5724 | } while (sg != first); | |
5725 | } | |
5726 | ||
dce840a0 PZ |
5727 | static void free_sched_domain(struct rcu_head *rcu) |
5728 | { | |
5729 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5730 | |
5731 | /* | |
5732 | * If its an overlapping domain it has private groups, iterate and | |
5733 | * nuke them all. | |
5734 | */ | |
5735 | if (sd->flags & SD_OVERLAP) { | |
5736 | free_sched_groups(sd->groups, 1); | |
5737 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5738 | kfree(sd->groups->sgp); |
dce840a0 | 5739 | kfree(sd->groups); |
9c3f75cb | 5740 | } |
dce840a0 PZ |
5741 | kfree(sd); |
5742 | } | |
5743 | ||
5744 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5745 | { | |
5746 | call_rcu(&sd->rcu, free_sched_domain); | |
5747 | } | |
5748 | ||
5749 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5750 | { | |
5751 | for (; sd; sd = sd->parent) | |
5752 | destroy_sched_domain(sd, cpu); | |
5753 | } | |
5754 | ||
518cd623 PZ |
5755 | /* |
5756 | * Keep a special pointer to the highest sched_domain that has | |
5757 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5758 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5759 | * | |
5760 | * Also keep a unique ID per domain (we use the first cpu number in | |
5761 | * the cpumask of the domain), this allows us to quickly tell if | |
5762 | * two cpus are in the same cache domain, see ttwu_share_cache(). | |
5763 | */ | |
5764 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5765 | DEFINE_PER_CPU(int, sd_llc_id); | |
5766 | ||
5767 | static void update_top_cache_domain(int cpu) | |
5768 | { | |
5769 | struct sched_domain *sd; | |
5770 | int id = cpu; | |
5771 | ||
5772 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
5773 | if (sd) | |
5774 | id = cpumask_first(sched_domain_span(sd)); | |
5775 | ||
5776 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5777 | per_cpu(sd_llc_id, cpu) = id; | |
5778 | } | |
5779 | ||
1da177e4 | 5780 | /* |
0eab9146 | 5781 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5782 | * hold the hotplug lock. |
5783 | */ | |
0eab9146 IM |
5784 | static void |
5785 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5786 | { |
70b97a7f | 5787 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5788 | struct sched_domain *tmp; |
5789 | ||
5790 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5791 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5792 | struct sched_domain *parent = tmp->parent; |
5793 | if (!parent) | |
5794 | break; | |
f29c9b1c | 5795 | |
1a848870 | 5796 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5797 | tmp->parent = parent->parent; |
1a848870 SS |
5798 | if (parent->parent) |
5799 | parent->parent->child = tmp; | |
dce840a0 | 5800 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5801 | } else |
5802 | tmp = tmp->parent; | |
245af2c7 SS |
5803 | } |
5804 | ||
1a848870 | 5805 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5806 | tmp = sd; |
245af2c7 | 5807 | sd = sd->parent; |
dce840a0 | 5808 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5809 | if (sd) |
5810 | sd->child = NULL; | |
5811 | } | |
1da177e4 | 5812 | |
4cb98839 | 5813 | sched_domain_debug(sd, cpu); |
1da177e4 | 5814 | |
57d885fe | 5815 | rq_attach_root(rq, rd); |
dce840a0 | 5816 | tmp = rq->sd; |
674311d5 | 5817 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5818 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5819 | |
5820 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5821 | } |
5822 | ||
5823 | /* cpus with isolated domains */ | |
dcc30a35 | 5824 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5825 | |
5826 | /* Setup the mask of cpus configured for isolated domains */ | |
5827 | static int __init isolated_cpu_setup(char *str) | |
5828 | { | |
bdddd296 | 5829 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5830 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5831 | return 1; |
5832 | } | |
5833 | ||
8927f494 | 5834 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5835 | |
9c1cfda2 | 5836 | #ifdef CONFIG_NUMA |
198e2f18 | 5837 | |
9c1cfda2 JH |
5838 | /** |
5839 | * find_next_best_node - find the next node to include in a sched_domain | |
5840 | * @node: node whose sched_domain we're building | |
5841 | * @used_nodes: nodes already in the sched_domain | |
5842 | * | |
41a2d6cf | 5843 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5844 | * finds the closest node not already in the @used_nodes map. |
5845 | * | |
5846 | * Should use nodemask_t. | |
5847 | */ | |
c5f59f08 | 5848 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 5849 | { |
7142d17e | 5850 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
5851 | |
5852 | min_val = INT_MAX; | |
5853 | ||
076ac2af | 5854 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 5855 | /* Start at @node */ |
076ac2af | 5856 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
5857 | |
5858 | if (!nr_cpus_node(n)) | |
5859 | continue; | |
5860 | ||
5861 | /* Skip already used nodes */ | |
c5f59f08 | 5862 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
5863 | continue; |
5864 | ||
5865 | /* Simple min distance search */ | |
5866 | val = node_distance(node, n); | |
5867 | ||
5868 | if (val < min_val) { | |
5869 | min_val = val; | |
5870 | best_node = n; | |
5871 | } | |
5872 | } | |
5873 | ||
7142d17e HD |
5874 | if (best_node != -1) |
5875 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
5876 | return best_node; |
5877 | } | |
5878 | ||
5879 | /** | |
5880 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5881 | * @node: node whose cpumask we're constructing | |
73486722 | 5882 | * @span: resulting cpumask |
9c1cfda2 | 5883 | * |
41a2d6cf | 5884 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
5885 | * should be one that prevents unnecessary balancing, but also spreads tasks |
5886 | * out optimally. | |
5887 | */ | |
96f874e2 | 5888 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 5889 | { |
c5f59f08 | 5890 | nodemask_t used_nodes; |
48f24c4d | 5891 | int i; |
9c1cfda2 | 5892 | |
6ca09dfc | 5893 | cpumask_clear(span); |
c5f59f08 | 5894 | nodes_clear(used_nodes); |
9c1cfda2 | 5895 | |
6ca09dfc | 5896 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 5897 | node_set(node, used_nodes); |
9c1cfda2 JH |
5898 | |
5899 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 5900 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
5901 | if (next_node < 0) |
5902 | break; | |
6ca09dfc | 5903 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 5904 | } |
9c1cfda2 | 5905 | } |
d3081f52 PZ |
5906 | |
5907 | static const struct cpumask *cpu_node_mask(int cpu) | |
5908 | { | |
5909 | lockdep_assert_held(&sched_domains_mutex); | |
5910 | ||
5911 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
5912 | ||
5913 | return sched_domains_tmpmask; | |
5914 | } | |
2c402dc3 PZ |
5915 | |
5916 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
5917 | { | |
5918 | return cpu_possible_mask; | |
5919 | } | |
6d6bc0ad | 5920 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 5921 | |
d3081f52 PZ |
5922 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5923 | { | |
5924 | return cpumask_of_node(cpu_to_node(cpu)); | |
5925 | } | |
5926 | ||
5c45bf27 | 5927 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5928 | |
dce840a0 PZ |
5929 | struct sd_data { |
5930 | struct sched_domain **__percpu sd; | |
5931 | struct sched_group **__percpu sg; | |
9c3f75cb | 5932 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5933 | }; |
5934 | ||
49a02c51 | 5935 | struct s_data { |
21d42ccf | 5936 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5937 | struct root_domain *rd; |
5938 | }; | |
5939 | ||
2109b99e | 5940 | enum s_alloc { |
2109b99e | 5941 | sa_rootdomain, |
21d42ccf | 5942 | sa_sd, |
dce840a0 | 5943 | sa_sd_storage, |
2109b99e AH |
5944 | sa_none, |
5945 | }; | |
5946 | ||
54ab4ff4 PZ |
5947 | struct sched_domain_topology_level; |
5948 | ||
5949 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5950 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5951 | ||
e3589f6c PZ |
5952 | #define SDTL_OVERLAP 0x01 |
5953 | ||
eb7a74e6 | 5954 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5955 | sched_domain_init_f init; |
5956 | sched_domain_mask_f mask; | |
e3589f6c | 5957 | int flags; |
54ab4ff4 | 5958 | struct sd_data data; |
eb7a74e6 PZ |
5959 | }; |
5960 | ||
e3589f6c PZ |
5961 | static int |
5962 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5963 | { | |
5964 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5965 | const struct cpumask *span = sched_domain_span(sd); | |
5966 | struct cpumask *covered = sched_domains_tmpmask; | |
5967 | struct sd_data *sdd = sd->private; | |
5968 | struct sched_domain *child; | |
5969 | int i; | |
5970 | ||
5971 | cpumask_clear(covered); | |
5972 | ||
5973 | for_each_cpu(i, span) { | |
5974 | struct cpumask *sg_span; | |
5975 | ||
5976 | if (cpumask_test_cpu(i, covered)) | |
5977 | continue; | |
5978 | ||
5979 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
4d78a223 | 5980 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5981 | |
5982 | if (!sg) | |
5983 | goto fail; | |
5984 | ||
5985 | sg_span = sched_group_cpus(sg); | |
5986 | ||
5987 | child = *per_cpu_ptr(sdd->sd, i); | |
5988 | if (child->child) { | |
5989 | child = child->child; | |
5990 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5991 | } else | |
5992 | cpumask_set_cpu(i, sg_span); | |
5993 | ||
5994 | cpumask_or(covered, covered, sg_span); | |
5995 | ||
5996 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
5997 | atomic_inc(&sg->sgp->ref); | |
5998 | ||
5999 | if (cpumask_test_cpu(cpu, sg_span)) | |
6000 | groups = sg; | |
6001 | ||
6002 | if (!first) | |
6003 | first = sg; | |
6004 | if (last) | |
6005 | last->next = sg; | |
6006 | last = sg; | |
6007 | last->next = first; | |
6008 | } | |
6009 | sd->groups = groups; | |
6010 | ||
6011 | return 0; | |
6012 | ||
6013 | fail: | |
6014 | free_sched_groups(first, 0); | |
6015 | ||
6016 | return -ENOMEM; | |
6017 | } | |
6018 | ||
dce840a0 | 6019 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6020 | { |
dce840a0 PZ |
6021 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6022 | struct sched_domain *child = sd->child; | |
1da177e4 | 6023 | |
dce840a0 PZ |
6024 | if (child) |
6025 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6026 | |
9c3f75cb | 6027 | if (sg) { |
dce840a0 | 6028 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 6029 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 6030 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 6031 | } |
dce840a0 PZ |
6032 | |
6033 | return cpu; | |
1e9f28fa | 6034 | } |
1e9f28fa | 6035 | |
01a08546 | 6036 | /* |
dce840a0 PZ |
6037 | * build_sched_groups will build a circular linked list of the groups |
6038 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6039 | * and ->cpu_power to 0. | |
e3589f6c PZ |
6040 | * |
6041 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6042 | */ |
e3589f6c PZ |
6043 | static int |
6044 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6045 | { |
dce840a0 PZ |
6046 | struct sched_group *first = NULL, *last = NULL; |
6047 | struct sd_data *sdd = sd->private; | |
6048 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6049 | struct cpumask *covered; |
dce840a0 | 6050 | int i; |
9c1cfda2 | 6051 | |
e3589f6c PZ |
6052 | get_group(cpu, sdd, &sd->groups); |
6053 | atomic_inc(&sd->groups->ref); | |
6054 | ||
6055 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
6056 | return 0; | |
6057 | ||
f96225fd PZ |
6058 | lockdep_assert_held(&sched_domains_mutex); |
6059 | covered = sched_domains_tmpmask; | |
6060 | ||
dce840a0 | 6061 | cpumask_clear(covered); |
6711cab4 | 6062 | |
dce840a0 PZ |
6063 | for_each_cpu(i, span) { |
6064 | struct sched_group *sg; | |
6065 | int group = get_group(i, sdd, &sg); | |
6066 | int j; | |
6711cab4 | 6067 | |
dce840a0 PZ |
6068 | if (cpumask_test_cpu(i, covered)) |
6069 | continue; | |
6711cab4 | 6070 | |
dce840a0 | 6071 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 6072 | sg->sgp->power = 0; |
0601a88d | 6073 | |
dce840a0 PZ |
6074 | for_each_cpu(j, span) { |
6075 | if (get_group(j, sdd, NULL) != group) | |
6076 | continue; | |
0601a88d | 6077 | |
dce840a0 PZ |
6078 | cpumask_set_cpu(j, covered); |
6079 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6080 | } | |
0601a88d | 6081 | |
dce840a0 PZ |
6082 | if (!first) |
6083 | first = sg; | |
6084 | if (last) | |
6085 | last->next = sg; | |
6086 | last = sg; | |
6087 | } | |
6088 | last->next = first; | |
e3589f6c PZ |
6089 | |
6090 | return 0; | |
0601a88d | 6091 | } |
51888ca2 | 6092 | |
89c4710e SS |
6093 | /* |
6094 | * Initialize sched groups cpu_power. | |
6095 | * | |
6096 | * cpu_power indicates the capacity of sched group, which is used while | |
6097 | * distributing the load between different sched groups in a sched domain. | |
6098 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6099 | * there are asymmetries in the topology. If there are asymmetries, group | |
6100 | * having more cpu_power will pickup more load compared to the group having | |
6101 | * less cpu_power. | |
89c4710e SS |
6102 | */ |
6103 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6104 | { | |
e3589f6c | 6105 | struct sched_group *sg = sd->groups; |
89c4710e | 6106 | |
e3589f6c PZ |
6107 | WARN_ON(!sd || !sg); |
6108 | ||
6109 | do { | |
6110 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6111 | sg = sg->next; | |
6112 | } while (sg != sd->groups); | |
89c4710e | 6113 | |
e3589f6c PZ |
6114 | if (cpu != group_first_cpu(sg)) |
6115 | return; | |
aae6d3dd | 6116 | |
d274cb30 | 6117 | update_group_power(sd, cpu); |
69e1e811 | 6118 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
6119 | } |
6120 | ||
029632fb PZ |
6121 | int __weak arch_sd_sibling_asym_packing(void) |
6122 | { | |
6123 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
6124 | } |
6125 | ||
7c16ec58 MT |
6126 | /* |
6127 | * Initializers for schedule domains | |
6128 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6129 | */ | |
6130 | ||
a5d8c348 IM |
6131 | #ifdef CONFIG_SCHED_DEBUG |
6132 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6133 | #else | |
6134 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6135 | #endif | |
6136 | ||
54ab4ff4 PZ |
6137 | #define SD_INIT_FUNC(type) \ |
6138 | static noinline struct sched_domain * \ | |
6139 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
6140 | { \ | |
6141 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
6142 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
6143 | SD_INIT_NAME(sd, type); \ |
6144 | sd->private = &tl->data; \ | |
6145 | return sd; \ | |
7c16ec58 MT |
6146 | } |
6147 | ||
6148 | SD_INIT_FUNC(CPU) | |
6149 | #ifdef CONFIG_NUMA | |
6150 | SD_INIT_FUNC(ALLNODES) | |
6151 | SD_INIT_FUNC(NODE) | |
6152 | #endif | |
6153 | #ifdef CONFIG_SCHED_SMT | |
6154 | SD_INIT_FUNC(SIBLING) | |
6155 | #endif | |
6156 | #ifdef CONFIG_SCHED_MC | |
6157 | SD_INIT_FUNC(MC) | |
6158 | #endif | |
01a08546 HC |
6159 | #ifdef CONFIG_SCHED_BOOK |
6160 | SD_INIT_FUNC(BOOK) | |
6161 | #endif | |
7c16ec58 | 6162 | |
1d3504fc | 6163 | static int default_relax_domain_level = -1; |
60495e77 | 6164 | int sched_domain_level_max; |
1d3504fc HS |
6165 | |
6166 | static int __init setup_relax_domain_level(char *str) | |
6167 | { | |
30e0e178 LZ |
6168 | unsigned long val; |
6169 | ||
6170 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 6171 | if (val < sched_domain_level_max) |
30e0e178 LZ |
6172 | default_relax_domain_level = val; |
6173 | ||
1d3504fc HS |
6174 | return 1; |
6175 | } | |
6176 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6177 | ||
6178 | static void set_domain_attribute(struct sched_domain *sd, | |
6179 | struct sched_domain_attr *attr) | |
6180 | { | |
6181 | int request; | |
6182 | ||
6183 | if (!attr || attr->relax_domain_level < 0) { | |
6184 | if (default_relax_domain_level < 0) | |
6185 | return; | |
6186 | else | |
6187 | request = default_relax_domain_level; | |
6188 | } else | |
6189 | request = attr->relax_domain_level; | |
6190 | if (request < sd->level) { | |
6191 | /* turn off idle balance on this domain */ | |
c88d5910 | 6192 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6193 | } else { |
6194 | /* turn on idle balance on this domain */ | |
c88d5910 | 6195 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6196 | } |
6197 | } | |
6198 | ||
54ab4ff4 PZ |
6199 | static void __sdt_free(const struct cpumask *cpu_map); |
6200 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6201 | ||
2109b99e AH |
6202 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6203 | const struct cpumask *cpu_map) | |
6204 | { | |
6205 | switch (what) { | |
2109b99e | 6206 | case sa_rootdomain: |
822ff793 PZ |
6207 | if (!atomic_read(&d->rd->refcount)) |
6208 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6209 | case sa_sd: |
6210 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6211 | case sa_sd_storage: |
54ab4ff4 | 6212 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6213 | case sa_none: |
6214 | break; | |
6215 | } | |
6216 | } | |
3404c8d9 | 6217 | |
2109b99e AH |
6218 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6219 | const struct cpumask *cpu_map) | |
6220 | { | |
dce840a0 PZ |
6221 | memset(d, 0, sizeof(*d)); |
6222 | ||
54ab4ff4 PZ |
6223 | if (__sdt_alloc(cpu_map)) |
6224 | return sa_sd_storage; | |
dce840a0 PZ |
6225 | d->sd = alloc_percpu(struct sched_domain *); |
6226 | if (!d->sd) | |
6227 | return sa_sd_storage; | |
2109b99e | 6228 | d->rd = alloc_rootdomain(); |
dce840a0 | 6229 | if (!d->rd) |
21d42ccf | 6230 | return sa_sd; |
2109b99e AH |
6231 | return sa_rootdomain; |
6232 | } | |
57d885fe | 6233 | |
dce840a0 PZ |
6234 | /* |
6235 | * NULL the sd_data elements we've used to build the sched_domain and | |
6236 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6237 | * will not free the data we're using. | |
6238 | */ | |
6239 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6240 | { | |
6241 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6242 | |
6243 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6244 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6245 | ||
e3589f6c | 6246 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6247 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6248 | |
6249 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6250 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6251 | } |
6252 | ||
2c402dc3 PZ |
6253 | #ifdef CONFIG_SCHED_SMT |
6254 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6255 | { |
2c402dc3 | 6256 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6257 | } |
2c402dc3 | 6258 | #endif |
7f4588f3 | 6259 | |
d069b916 PZ |
6260 | /* |
6261 | * Topology list, bottom-up. | |
6262 | */ | |
2c402dc3 | 6263 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6264 | #ifdef CONFIG_SCHED_SMT |
6265 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6266 | #endif |
1e9f28fa | 6267 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6268 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6269 | #endif |
d069b916 PZ |
6270 | #ifdef CONFIG_SCHED_BOOK |
6271 | { sd_init_BOOK, cpu_book_mask, }, | |
6272 | #endif | |
6273 | { sd_init_CPU, cpu_cpu_mask, }, | |
6274 | #ifdef CONFIG_NUMA | |
e3589f6c | 6275 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, |
d069b916 | 6276 | { sd_init_ALLNODES, cpu_allnodes_mask, }, |
1da177e4 | 6277 | #endif |
eb7a74e6 PZ |
6278 | { NULL, }, |
6279 | }; | |
6280 | ||
6281 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6282 | ||
54ab4ff4 PZ |
6283 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6284 | { | |
6285 | struct sched_domain_topology_level *tl; | |
6286 | int j; | |
6287 | ||
6288 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6289 | struct sd_data *sdd = &tl->data; | |
6290 | ||
6291 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6292 | if (!sdd->sd) | |
6293 | return -ENOMEM; | |
6294 | ||
6295 | sdd->sg = alloc_percpu(struct sched_group *); | |
6296 | if (!sdd->sg) | |
6297 | return -ENOMEM; | |
6298 | ||
9c3f75cb PZ |
6299 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6300 | if (!sdd->sgp) | |
6301 | return -ENOMEM; | |
6302 | ||
54ab4ff4 PZ |
6303 | for_each_cpu(j, cpu_map) { |
6304 | struct sched_domain *sd; | |
6305 | struct sched_group *sg; | |
9c3f75cb | 6306 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6307 | |
6308 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6309 | GFP_KERNEL, cpu_to_node(j)); | |
6310 | if (!sd) | |
6311 | return -ENOMEM; | |
6312 | ||
6313 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6314 | ||
6315 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6316 | GFP_KERNEL, cpu_to_node(j)); | |
6317 | if (!sg) | |
6318 | return -ENOMEM; | |
6319 | ||
6320 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
6321 | |
6322 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
6323 | GFP_KERNEL, cpu_to_node(j)); | |
6324 | if (!sgp) | |
6325 | return -ENOMEM; | |
6326 | ||
6327 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6328 | } |
6329 | } | |
6330 | ||
6331 | return 0; | |
6332 | } | |
6333 | ||
6334 | static void __sdt_free(const struct cpumask *cpu_map) | |
6335 | { | |
6336 | struct sched_domain_topology_level *tl; | |
6337 | int j; | |
6338 | ||
6339 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6340 | struct sd_data *sdd = &tl->data; | |
6341 | ||
6342 | for_each_cpu(j, cpu_map) { | |
e3589f6c PZ |
6343 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); |
6344 | if (sd && (sd->flags & SD_OVERLAP)) | |
6345 | free_sched_groups(sd->groups, 0); | |
feff8fa0 | 6346 | kfree(*per_cpu_ptr(sdd->sd, j)); |
54ab4ff4 | 6347 | kfree(*per_cpu_ptr(sdd->sg, j)); |
9c3f75cb | 6348 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
54ab4ff4 PZ |
6349 | } |
6350 | free_percpu(sdd->sd); | |
6351 | free_percpu(sdd->sg); | |
9c3f75cb | 6352 | free_percpu(sdd->sgp); |
54ab4ff4 PZ |
6353 | } |
6354 | } | |
6355 | ||
2c402dc3 PZ |
6356 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6357 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6358 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6359 | int cpu) |
6360 | { | |
54ab4ff4 | 6361 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6362 | if (!sd) |
d069b916 | 6363 | return child; |
2c402dc3 PZ |
6364 | |
6365 | set_domain_attribute(sd, attr); | |
6366 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
6367 | if (child) { |
6368 | sd->level = child->level + 1; | |
6369 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6370 | child->parent = sd; |
60495e77 | 6371 | } |
d069b916 | 6372 | sd->child = child; |
2c402dc3 PZ |
6373 | |
6374 | return sd; | |
6375 | } | |
6376 | ||
2109b99e AH |
6377 | /* |
6378 | * Build sched domains for a given set of cpus and attach the sched domains | |
6379 | * to the individual cpus | |
6380 | */ | |
dce840a0 PZ |
6381 | static int build_sched_domains(const struct cpumask *cpu_map, |
6382 | struct sched_domain_attr *attr) | |
2109b99e AH |
6383 | { |
6384 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6385 | struct sched_domain *sd; |
2109b99e | 6386 | struct s_data d; |
822ff793 | 6387 | int i, ret = -ENOMEM; |
9c1cfda2 | 6388 | |
2109b99e AH |
6389 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6390 | if (alloc_state != sa_rootdomain) | |
6391 | goto error; | |
9c1cfda2 | 6392 | |
dce840a0 | 6393 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6394 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6395 | struct sched_domain_topology_level *tl; |
6396 | ||
3bd65a80 | 6397 | sd = NULL; |
e3589f6c | 6398 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6399 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6400 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6401 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6402 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6403 | break; | |
e3589f6c | 6404 | } |
d274cb30 | 6405 | |
d069b916 PZ |
6406 | while (sd->child) |
6407 | sd = sd->child; | |
6408 | ||
21d42ccf | 6409 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6410 | } |
6411 | ||
6412 | /* Build the groups for the domains */ | |
6413 | for_each_cpu(i, cpu_map) { | |
6414 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6415 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6416 | if (sd->flags & SD_OVERLAP) { |
6417 | if (build_overlap_sched_groups(sd, i)) | |
6418 | goto error; | |
6419 | } else { | |
6420 | if (build_sched_groups(sd, i)) | |
6421 | goto error; | |
6422 | } | |
1cf51902 | 6423 | } |
a06dadbe | 6424 | } |
9c1cfda2 | 6425 | |
1da177e4 | 6426 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6427 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6428 | if (!cpumask_test_cpu(i, cpu_map)) | |
6429 | continue; | |
9c1cfda2 | 6430 | |
dce840a0 PZ |
6431 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6432 | claim_allocations(i, sd); | |
cd4ea6ae | 6433 | init_sched_groups_power(i, sd); |
dce840a0 | 6434 | } |
f712c0c7 | 6435 | } |
9c1cfda2 | 6436 | |
1da177e4 | 6437 | /* Attach the domains */ |
dce840a0 | 6438 | rcu_read_lock(); |
abcd083a | 6439 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6440 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6441 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6442 | } |
dce840a0 | 6443 | rcu_read_unlock(); |
51888ca2 | 6444 | |
822ff793 | 6445 | ret = 0; |
51888ca2 | 6446 | error: |
2109b99e | 6447 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6448 | return ret; |
1da177e4 | 6449 | } |
029190c5 | 6450 | |
acc3f5d7 | 6451 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6452 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6453 | static struct sched_domain_attr *dattr_cur; |
6454 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6455 | |
6456 | /* | |
6457 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6458 | * cpumask) fails, then fallback to a single sched domain, |
6459 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6460 | */ |
4212823f | 6461 | static cpumask_var_t fallback_doms; |
029190c5 | 6462 | |
ee79d1bd HC |
6463 | /* |
6464 | * arch_update_cpu_topology lets virtualized architectures update the | |
6465 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6466 | * or 0 if it stayed the same. | |
6467 | */ | |
6468 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6469 | { |
ee79d1bd | 6470 | return 0; |
22e52b07 HC |
6471 | } |
6472 | ||
acc3f5d7 RR |
6473 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6474 | { | |
6475 | int i; | |
6476 | cpumask_var_t *doms; | |
6477 | ||
6478 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6479 | if (!doms) | |
6480 | return NULL; | |
6481 | for (i = 0; i < ndoms; i++) { | |
6482 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6483 | free_sched_domains(doms, i); | |
6484 | return NULL; | |
6485 | } | |
6486 | } | |
6487 | return doms; | |
6488 | } | |
6489 | ||
6490 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6491 | { | |
6492 | unsigned int i; | |
6493 | for (i = 0; i < ndoms; i++) | |
6494 | free_cpumask_var(doms[i]); | |
6495 | kfree(doms); | |
6496 | } | |
6497 | ||
1a20ff27 | 6498 | /* |
41a2d6cf | 6499 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6500 | * For now this just excludes isolated cpus, but could be used to |
6501 | * exclude other special cases in the future. | |
1a20ff27 | 6502 | */ |
c4a8849a | 6503 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6504 | { |
7378547f MM |
6505 | int err; |
6506 | ||
22e52b07 | 6507 | arch_update_cpu_topology(); |
029190c5 | 6508 | ndoms_cur = 1; |
acc3f5d7 | 6509 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6510 | if (!doms_cur) |
acc3f5d7 RR |
6511 | doms_cur = &fallback_doms; |
6512 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 6513 | dattr_cur = NULL; |
dce840a0 | 6514 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6515 | register_sched_domain_sysctl(); |
7378547f MM |
6516 | |
6517 | return err; | |
1a20ff27 DG |
6518 | } |
6519 | ||
1a20ff27 DG |
6520 | /* |
6521 | * Detach sched domains from a group of cpus specified in cpu_map | |
6522 | * These cpus will now be attached to the NULL domain | |
6523 | */ | |
96f874e2 | 6524 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6525 | { |
6526 | int i; | |
6527 | ||
dce840a0 | 6528 | rcu_read_lock(); |
abcd083a | 6529 | for_each_cpu(i, cpu_map) |
57d885fe | 6530 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6531 | rcu_read_unlock(); |
1a20ff27 DG |
6532 | } |
6533 | ||
1d3504fc HS |
6534 | /* handle null as "default" */ |
6535 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6536 | struct sched_domain_attr *new, int idx_new) | |
6537 | { | |
6538 | struct sched_domain_attr tmp; | |
6539 | ||
6540 | /* fast path */ | |
6541 | if (!new && !cur) | |
6542 | return 1; | |
6543 | ||
6544 | tmp = SD_ATTR_INIT; | |
6545 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6546 | new ? (new + idx_new) : &tmp, | |
6547 | sizeof(struct sched_domain_attr)); | |
6548 | } | |
6549 | ||
029190c5 PJ |
6550 | /* |
6551 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6552 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6553 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6554 | * It destroys each deleted domain and builds each new domain. | |
6555 | * | |
acc3f5d7 | 6556 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6557 | * The masks don't intersect (don't overlap.) We should setup one |
6558 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6559 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6560 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6561 | * it as it is. | |
6562 | * | |
acc3f5d7 RR |
6563 | * The passed in 'doms_new' should be allocated using |
6564 | * alloc_sched_domains. This routine takes ownership of it and will | |
6565 | * free_sched_domains it when done with it. If the caller failed the | |
6566 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6567 | * and partition_sched_domains() will fallback to the single partition | |
6568 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6569 | * |
96f874e2 | 6570 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6571 | * ndoms_new == 0 is a special case for destroying existing domains, |
6572 | * and it will not create the default domain. | |
dfb512ec | 6573 | * |
029190c5 PJ |
6574 | * Call with hotplug lock held |
6575 | */ | |
acc3f5d7 | 6576 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6577 | struct sched_domain_attr *dattr_new) |
029190c5 | 6578 | { |
dfb512ec | 6579 | int i, j, n; |
d65bd5ec | 6580 | int new_topology; |
029190c5 | 6581 | |
712555ee | 6582 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6583 | |
7378547f MM |
6584 | /* always unregister in case we don't destroy any domains */ |
6585 | unregister_sched_domain_sysctl(); | |
6586 | ||
d65bd5ec HC |
6587 | /* Let architecture update cpu core mappings. */ |
6588 | new_topology = arch_update_cpu_topology(); | |
6589 | ||
dfb512ec | 6590 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6591 | |
6592 | /* Destroy deleted domains */ | |
6593 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6594 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6595 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6596 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6597 | goto match1; |
6598 | } | |
6599 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6600 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6601 | match1: |
6602 | ; | |
6603 | } | |
6604 | ||
e761b772 MK |
6605 | if (doms_new == NULL) { |
6606 | ndoms_cur = 0; | |
acc3f5d7 | 6607 | doms_new = &fallback_doms; |
6ad4c188 | 6608 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6609 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6610 | } |
6611 | ||
029190c5 PJ |
6612 | /* Build new domains */ |
6613 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6614 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6615 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6616 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6617 | goto match2; |
6618 | } | |
6619 | /* no match - add a new doms_new */ | |
dce840a0 | 6620 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6621 | match2: |
6622 | ; | |
6623 | } | |
6624 | ||
6625 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6626 | if (doms_cur != &fallback_doms) |
6627 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6628 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6629 | doms_cur = doms_new; |
1d3504fc | 6630 | dattr_cur = dattr_new; |
029190c5 | 6631 | ndoms_cur = ndoms_new; |
7378547f MM |
6632 | |
6633 | register_sched_domain_sysctl(); | |
a1835615 | 6634 | |
712555ee | 6635 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6636 | } |
6637 | ||
5c45bf27 | 6638 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 6639 | static void reinit_sched_domains(void) |
5c45bf27 | 6640 | { |
95402b38 | 6641 | get_online_cpus(); |
dfb512ec MK |
6642 | |
6643 | /* Destroy domains first to force the rebuild */ | |
6644 | partition_sched_domains(0, NULL, NULL); | |
6645 | ||
e761b772 | 6646 | rebuild_sched_domains(); |
95402b38 | 6647 | put_online_cpus(); |
5c45bf27 SS |
6648 | } |
6649 | ||
6650 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6651 | { | |
afb8a9b7 | 6652 | unsigned int level = 0; |
5c45bf27 | 6653 | |
afb8a9b7 GS |
6654 | if (sscanf(buf, "%u", &level) != 1) |
6655 | return -EINVAL; | |
6656 | ||
6657 | /* | |
6658 | * level is always be positive so don't check for | |
6659 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
6660 | * What happens on 0 or 1 byte write, | |
6661 | * need to check for count as well? | |
6662 | */ | |
6663 | ||
6664 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
6665 | return -EINVAL; |
6666 | ||
6667 | if (smt) | |
afb8a9b7 | 6668 | sched_smt_power_savings = level; |
5c45bf27 | 6669 | else |
afb8a9b7 | 6670 | sched_mc_power_savings = level; |
5c45bf27 | 6671 | |
c4a8849a | 6672 | reinit_sched_domains(); |
5c45bf27 | 6673 | |
c70f22d2 | 6674 | return count; |
5c45bf27 SS |
6675 | } |
6676 | ||
5c45bf27 | 6677 | #ifdef CONFIG_SCHED_MC |
8a25a2fd KS |
6678 | static ssize_t sched_mc_power_savings_show(struct device *dev, |
6679 | struct device_attribute *attr, | |
6680 | char *buf) | |
5c45bf27 | 6681 | { |
8a25a2fd | 6682 | return sprintf(buf, "%u\n", sched_mc_power_savings); |
5c45bf27 | 6683 | } |
8a25a2fd KS |
6684 | static ssize_t sched_mc_power_savings_store(struct device *dev, |
6685 | struct device_attribute *attr, | |
48f24c4d | 6686 | const char *buf, size_t count) |
5c45bf27 SS |
6687 | { |
6688 | return sched_power_savings_store(buf, count, 0); | |
6689 | } | |
8a25a2fd KS |
6690 | static DEVICE_ATTR(sched_mc_power_savings, 0644, |
6691 | sched_mc_power_savings_show, | |
6692 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6693 | #endif |
6694 | ||
6695 | #ifdef CONFIG_SCHED_SMT | |
8a25a2fd KS |
6696 | static ssize_t sched_smt_power_savings_show(struct device *dev, |
6697 | struct device_attribute *attr, | |
6698 | char *buf) | |
5c45bf27 | 6699 | { |
8a25a2fd | 6700 | return sprintf(buf, "%u\n", sched_smt_power_savings); |
5c45bf27 | 6701 | } |
8a25a2fd KS |
6702 | static ssize_t sched_smt_power_savings_store(struct device *dev, |
6703 | struct device_attribute *attr, | |
48f24c4d | 6704 | const char *buf, size_t count) |
5c45bf27 SS |
6705 | { |
6706 | return sched_power_savings_store(buf, count, 1); | |
6707 | } | |
8a25a2fd | 6708 | static DEVICE_ATTR(sched_smt_power_savings, 0644, |
f718cd4a | 6709 | sched_smt_power_savings_show, |
6707de00 AB |
6710 | sched_smt_power_savings_store); |
6711 | #endif | |
6712 | ||
8a25a2fd | 6713 | int __init sched_create_sysfs_power_savings_entries(struct device *dev) |
6707de00 AB |
6714 | { |
6715 | int err = 0; | |
6716 | ||
6717 | #ifdef CONFIG_SCHED_SMT | |
6718 | if (smt_capable()) | |
8a25a2fd | 6719 | err = device_create_file(dev, &dev_attr_sched_smt_power_savings); |
6707de00 AB |
6720 | #endif |
6721 | #ifdef CONFIG_SCHED_MC | |
6722 | if (!err && mc_capable()) | |
8a25a2fd | 6723 | err = device_create_file(dev, &dev_attr_sched_mc_power_savings); |
6707de00 AB |
6724 | #endif |
6725 | return err; | |
6726 | } | |
6d6bc0ad | 6727 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 6728 | |
1da177e4 | 6729 | /* |
3a101d05 TH |
6730 | * Update cpusets according to cpu_active mask. If cpusets are |
6731 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6732 | * around partition_sched_domains(). | |
1da177e4 | 6733 | */ |
0b2e918a TH |
6734 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6735 | void *hcpu) | |
e761b772 | 6736 | { |
3a101d05 | 6737 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 6738 | case CPU_ONLINE: |
6ad4c188 | 6739 | case CPU_DOWN_FAILED: |
3a101d05 | 6740 | cpuset_update_active_cpus(); |
e761b772 | 6741 | return NOTIFY_OK; |
3a101d05 TH |
6742 | default: |
6743 | return NOTIFY_DONE; | |
6744 | } | |
6745 | } | |
e761b772 | 6746 | |
0b2e918a TH |
6747 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6748 | void *hcpu) | |
3a101d05 TH |
6749 | { |
6750 | switch (action & ~CPU_TASKS_FROZEN) { | |
6751 | case CPU_DOWN_PREPARE: | |
6752 | cpuset_update_active_cpus(); | |
6753 | return NOTIFY_OK; | |
e761b772 MK |
6754 | default: |
6755 | return NOTIFY_DONE; | |
6756 | } | |
6757 | } | |
e761b772 | 6758 | |
1da177e4 LT |
6759 | void __init sched_init_smp(void) |
6760 | { | |
dcc30a35 RR |
6761 | cpumask_var_t non_isolated_cpus; |
6762 | ||
6763 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6764 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6765 | |
95402b38 | 6766 | get_online_cpus(); |
712555ee | 6767 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6768 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6769 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6770 | if (cpumask_empty(non_isolated_cpus)) | |
6771 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6772 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6773 | put_online_cpus(); |
e761b772 | 6774 | |
3a101d05 TH |
6775 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6776 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6777 | |
6778 | /* RT runtime code needs to handle some hotplug events */ | |
6779 | hotcpu_notifier(update_runtime, 0); | |
6780 | ||
b328ca18 | 6781 | init_hrtick(); |
5c1e1767 NP |
6782 | |
6783 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6784 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6785 | BUG(); |
19978ca6 | 6786 | sched_init_granularity(); |
dcc30a35 | 6787 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6788 | |
0e3900e6 | 6789 | init_sched_rt_class(); |
1da177e4 LT |
6790 | } |
6791 | #else | |
6792 | void __init sched_init_smp(void) | |
6793 | { | |
19978ca6 | 6794 | sched_init_granularity(); |
1da177e4 LT |
6795 | } |
6796 | #endif /* CONFIG_SMP */ | |
6797 | ||
cd1bb94b AB |
6798 | const_debug unsigned int sysctl_timer_migration = 1; |
6799 | ||
1da177e4 LT |
6800 | int in_sched_functions(unsigned long addr) |
6801 | { | |
1da177e4 LT |
6802 | return in_lock_functions(addr) || |
6803 | (addr >= (unsigned long)__sched_text_start | |
6804 | && addr < (unsigned long)__sched_text_end); | |
6805 | } | |
6806 | ||
029632fb PZ |
6807 | #ifdef CONFIG_CGROUP_SCHED |
6808 | struct task_group root_task_group; | |
052f1dc7 | 6809 | #endif |
6f505b16 | 6810 | |
029632fb | 6811 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6812 | |
1da177e4 LT |
6813 | void __init sched_init(void) |
6814 | { | |
dd41f596 | 6815 | int i, j; |
434d53b0 MT |
6816 | unsigned long alloc_size = 0, ptr; |
6817 | ||
6818 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6819 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6820 | #endif | |
6821 | #ifdef CONFIG_RT_GROUP_SCHED | |
6822 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6823 | #endif |
df7c8e84 | 6824 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6825 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6826 | #endif |
434d53b0 | 6827 | if (alloc_size) { |
36b7b6d4 | 6828 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6829 | |
6830 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6831 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6832 | ptr += nr_cpu_ids * sizeof(void **); |
6833 | ||
07e06b01 | 6834 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6835 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6836 | |
6d6bc0ad | 6837 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6838 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6839 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6840 | ptr += nr_cpu_ids * sizeof(void **); |
6841 | ||
07e06b01 | 6842 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6843 | ptr += nr_cpu_ids * sizeof(void **); |
6844 | ||
6d6bc0ad | 6845 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6846 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6847 | for_each_possible_cpu(i) { | |
6848 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6849 | ptr += cpumask_size(); | |
6850 | } | |
6851 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6852 | } |
dd41f596 | 6853 | |
57d885fe GH |
6854 | #ifdef CONFIG_SMP |
6855 | init_defrootdomain(); | |
6856 | #endif | |
6857 | ||
d0b27fa7 PZ |
6858 | init_rt_bandwidth(&def_rt_bandwidth, |
6859 | global_rt_period(), global_rt_runtime()); | |
6860 | ||
6861 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6862 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6863 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6864 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6865 | |
7c941438 | 6866 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6867 | list_add(&root_task_group.list, &task_groups); |
6868 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6869 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6870 | autogroup_init(&init_task); |
54c707e9 | 6871 | |
7c941438 | 6872 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6873 | |
54c707e9 GC |
6874 | #ifdef CONFIG_CGROUP_CPUACCT |
6875 | root_cpuacct.cpustat = &kernel_cpustat; | |
6876 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
6877 | /* Too early, not expected to fail */ | |
6878 | BUG_ON(!root_cpuacct.cpuusage); | |
6879 | #endif | |
0a945022 | 6880 | for_each_possible_cpu(i) { |
70b97a7f | 6881 | struct rq *rq; |
1da177e4 LT |
6882 | |
6883 | rq = cpu_rq(i); | |
05fa785c | 6884 | raw_spin_lock_init(&rq->lock); |
7897986b | 6885 | rq->nr_running = 0; |
dce48a84 TG |
6886 | rq->calc_load_active = 0; |
6887 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6888 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6889 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6890 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6891 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6892 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6893 | /* |
07e06b01 | 6894 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6895 | * |
6896 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6897 | * gets 100% of the cpu resources in the system. This overall | |
6898 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6899 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6900 | * based on each entity's (task or task-group's) weight |
6901 | * (se->load.weight). | |
6902 | * | |
07e06b01 | 6903 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6904 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6905 | * then A0's share of the cpu resource is: | |
6906 | * | |
0d905bca | 6907 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6908 | * |
07e06b01 YZ |
6909 | * We achieve this by letting root_task_group's tasks sit |
6910 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6911 | */ |
ab84d31e | 6912 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6913 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6914 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6915 | ||
6916 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6917 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6918 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6919 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6920 | #endif |
1da177e4 | 6921 | |
dd41f596 IM |
6922 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6923 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6924 | |
6925 | rq->last_load_update_tick = jiffies; | |
6926 | ||
1da177e4 | 6927 | #ifdef CONFIG_SMP |
41c7ce9a | 6928 | rq->sd = NULL; |
57d885fe | 6929 | rq->rd = NULL; |
1399fa78 | 6930 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6931 | rq->post_schedule = 0; |
1da177e4 | 6932 | rq->active_balance = 0; |
dd41f596 | 6933 | rq->next_balance = jiffies; |
1da177e4 | 6934 | rq->push_cpu = 0; |
0a2966b4 | 6935 | rq->cpu = i; |
1f11eb6a | 6936 | rq->online = 0; |
eae0c9df MG |
6937 | rq->idle_stamp = 0; |
6938 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 6939 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 6940 | #ifdef CONFIG_NO_HZ |
1c792db7 | 6941 | rq->nohz_flags = 0; |
83cd4fe2 | 6942 | #endif |
1da177e4 | 6943 | #endif |
8f4d37ec | 6944 | init_rq_hrtick(rq); |
1da177e4 | 6945 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6946 | } |
6947 | ||
2dd73a4f | 6948 | set_load_weight(&init_task); |
b50f60ce | 6949 | |
e107be36 AK |
6950 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6951 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6952 | #endif | |
6953 | ||
b50f60ce | 6954 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6955 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6956 | #endif |
6957 | ||
1da177e4 LT |
6958 | /* |
6959 | * The boot idle thread does lazy MMU switching as well: | |
6960 | */ | |
6961 | atomic_inc(&init_mm.mm_count); | |
6962 | enter_lazy_tlb(&init_mm, current); | |
6963 | ||
6964 | /* | |
6965 | * Make us the idle thread. Technically, schedule() should not be | |
6966 | * called from this thread, however somewhere below it might be, | |
6967 | * but because we are the idle thread, we just pick up running again | |
6968 | * when this runqueue becomes "idle". | |
6969 | */ | |
6970 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6971 | |
6972 | calc_load_update = jiffies + LOAD_FREQ; | |
6973 | ||
dd41f596 IM |
6974 | /* |
6975 | * During early bootup we pretend to be a normal task: | |
6976 | */ | |
6977 | current->sched_class = &fair_sched_class; | |
6892b75e | 6978 | |
bf4d83f6 | 6979 | #ifdef CONFIG_SMP |
4cb98839 | 6980 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6981 | /* May be allocated at isolcpus cmdline parse time */ |
6982 | if (cpu_isolated_map == NULL) | |
6983 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
029632fb PZ |
6984 | #endif |
6985 | init_sched_fair_class(); | |
6a7b3dc3 | 6986 | |
6892b75e | 6987 | scheduler_running = 1; |
1da177e4 LT |
6988 | } |
6989 | ||
d902db1e | 6990 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6991 | static inline int preempt_count_equals(int preempt_offset) |
6992 | { | |
234da7bc | 6993 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6994 | |
4ba8216c | 6995 | return (nested == preempt_offset); |
e4aafea2 FW |
6996 | } |
6997 | ||
d894837f | 6998 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6999 | { |
1da177e4 LT |
7000 | static unsigned long prev_jiffy; /* ratelimiting */ |
7001 | ||
b3fbab05 | 7002 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7003 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7004 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7005 | return; |
7006 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7007 | return; | |
7008 | prev_jiffy = jiffies; | |
7009 | ||
3df0fc5b PZ |
7010 | printk(KERN_ERR |
7011 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7012 | file, line); | |
7013 | printk(KERN_ERR | |
7014 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7015 | in_atomic(), irqs_disabled(), | |
7016 | current->pid, current->comm); | |
aef745fc IM |
7017 | |
7018 | debug_show_held_locks(current); | |
7019 | if (irqs_disabled()) | |
7020 | print_irqtrace_events(current); | |
7021 | dump_stack(); | |
1da177e4 LT |
7022 | } |
7023 | EXPORT_SYMBOL(__might_sleep); | |
7024 | #endif | |
7025 | ||
7026 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7027 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7028 | { | |
da7a735e PZ |
7029 | const struct sched_class *prev_class = p->sched_class; |
7030 | int old_prio = p->prio; | |
3a5e4dc1 | 7031 | int on_rq; |
3e51f33f | 7032 | |
fd2f4419 | 7033 | on_rq = p->on_rq; |
3a5e4dc1 AK |
7034 | if (on_rq) |
7035 | deactivate_task(rq, p, 0); | |
7036 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7037 | if (on_rq) { | |
7038 | activate_task(rq, p, 0); | |
7039 | resched_task(rq->curr); | |
7040 | } | |
da7a735e PZ |
7041 | |
7042 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7043 | } |
7044 | ||
1da177e4 LT |
7045 | void normalize_rt_tasks(void) |
7046 | { | |
a0f98a1c | 7047 | struct task_struct *g, *p; |
1da177e4 | 7048 | unsigned long flags; |
70b97a7f | 7049 | struct rq *rq; |
1da177e4 | 7050 | |
4cf5d77a | 7051 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7052 | do_each_thread(g, p) { |
178be793 IM |
7053 | /* |
7054 | * Only normalize user tasks: | |
7055 | */ | |
7056 | if (!p->mm) | |
7057 | continue; | |
7058 | ||
6cfb0d5d | 7059 | p->se.exec_start = 0; |
6cfb0d5d | 7060 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7061 | p->se.statistics.wait_start = 0; |
7062 | p->se.statistics.sleep_start = 0; | |
7063 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7064 | #endif |
dd41f596 IM |
7065 | |
7066 | if (!rt_task(p)) { | |
7067 | /* | |
7068 | * Renice negative nice level userspace | |
7069 | * tasks back to 0: | |
7070 | */ | |
7071 | if (TASK_NICE(p) < 0 && p->mm) | |
7072 | set_user_nice(p, 0); | |
1da177e4 | 7073 | continue; |
dd41f596 | 7074 | } |
1da177e4 | 7075 | |
1d615482 | 7076 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7077 | rq = __task_rq_lock(p); |
1da177e4 | 7078 | |
178be793 | 7079 | normalize_task(rq, p); |
3a5e4dc1 | 7080 | |
b29739f9 | 7081 | __task_rq_unlock(rq); |
1d615482 | 7082 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7083 | } while_each_thread(g, p); |
7084 | ||
4cf5d77a | 7085 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7086 | } |
7087 | ||
7088 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7089 | |
67fc4e0c | 7090 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7091 | /* |
67fc4e0c | 7092 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7093 | * |
7094 | * They can only be called when the whole system has been | |
7095 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7096 | * activity can take place. Using them for anything else would | |
7097 | * be a serious bug, and as a result, they aren't even visible | |
7098 | * under any other configuration. | |
7099 | */ | |
7100 | ||
7101 | /** | |
7102 | * curr_task - return the current task for a given cpu. | |
7103 | * @cpu: the processor in question. | |
7104 | * | |
7105 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7106 | */ | |
36c8b586 | 7107 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7108 | { |
7109 | return cpu_curr(cpu); | |
7110 | } | |
7111 | ||
67fc4e0c JW |
7112 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7113 | ||
7114 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7115 | /** |
7116 | * set_curr_task - set the current task for a given cpu. | |
7117 | * @cpu: the processor in question. | |
7118 | * @p: the task pointer to set. | |
7119 | * | |
7120 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7121 | * are serviced on a separate stack. It allows the architecture to switch the |
7122 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7123 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7124 | * and caller must save the original value of the current task (see | |
7125 | * curr_task() above) and restore that value before reenabling interrupts and | |
7126 | * re-starting the system. | |
7127 | * | |
7128 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7129 | */ | |
36c8b586 | 7130 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7131 | { |
7132 | cpu_curr(cpu) = p; | |
7133 | } | |
7134 | ||
7135 | #endif | |
29f59db3 | 7136 | |
052f1dc7 | 7137 | #ifdef CONFIG_RT_GROUP_SCHED |
6d6bc0ad | 7138 | #else /* !CONFIG_RT_GROUP_SCHED */ |
6d6bc0ad | 7139 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 7140 | |
7c941438 | 7141 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7142 | /* task_group_lock serializes the addition/removal of task groups */ |
7143 | static DEFINE_SPINLOCK(task_group_lock); | |
7144 | ||
bccbe08a PZ |
7145 | static void free_sched_group(struct task_group *tg) |
7146 | { | |
7147 | free_fair_sched_group(tg); | |
7148 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7149 | autogroup_free(tg); |
bccbe08a PZ |
7150 | kfree(tg); |
7151 | } | |
7152 | ||
7153 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7154 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7155 | { |
7156 | struct task_group *tg; | |
7157 | unsigned long flags; | |
bccbe08a PZ |
7158 | |
7159 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7160 | if (!tg) | |
7161 | return ERR_PTR(-ENOMEM); | |
7162 | ||
ec7dc8ac | 7163 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7164 | goto err; |
7165 | ||
ec7dc8ac | 7166 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7167 | goto err; |
7168 | ||
8ed36996 | 7169 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7170 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7171 | |
7172 | WARN_ON(!parent); /* root should already exist */ | |
7173 | ||
7174 | tg->parent = parent; | |
f473aa5e | 7175 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7176 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7177 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7178 | |
9b5b7751 | 7179 | return tg; |
29f59db3 SV |
7180 | |
7181 | err: | |
6f505b16 | 7182 | free_sched_group(tg); |
29f59db3 SV |
7183 | return ERR_PTR(-ENOMEM); |
7184 | } | |
7185 | ||
9b5b7751 | 7186 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7187 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7188 | { |
29f59db3 | 7189 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7190 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7191 | } |
7192 | ||
9b5b7751 | 7193 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7194 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7195 | { |
8ed36996 | 7196 | unsigned long flags; |
9b5b7751 | 7197 | int i; |
29f59db3 | 7198 | |
3d4b47b4 PZ |
7199 | /* end participation in shares distribution */ |
7200 | for_each_possible_cpu(i) | |
bccbe08a | 7201 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7202 | |
7203 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7204 | list_del_rcu(&tg->list); |
f473aa5e | 7205 | list_del_rcu(&tg->siblings); |
8ed36996 | 7206 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7207 | |
9b5b7751 | 7208 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7209 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7210 | } |
7211 | ||
9b5b7751 | 7212 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7213 | * The caller of this function should have put the task in its new group |
7214 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7215 | * reflect its new group. | |
9b5b7751 SV |
7216 | */ |
7217 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7218 | { |
7219 | int on_rq, running; | |
7220 | unsigned long flags; | |
7221 | struct rq *rq; | |
7222 | ||
7223 | rq = task_rq_lock(tsk, &flags); | |
7224 | ||
051a1d1a | 7225 | running = task_current(rq, tsk); |
fd2f4419 | 7226 | on_rq = tsk->on_rq; |
29f59db3 | 7227 | |
0e1f3483 | 7228 | if (on_rq) |
29f59db3 | 7229 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7230 | if (unlikely(running)) |
7231 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7232 | |
810b3817 | 7233 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7234 | if (tsk->sched_class->task_move_group) |
7235 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7236 | else | |
810b3817 | 7237 | #endif |
b2b5ce02 | 7238 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7239 | |
0e1f3483 HS |
7240 | if (unlikely(running)) |
7241 | tsk->sched_class->set_curr_task(rq); | |
7242 | if (on_rq) | |
371fd7e7 | 7243 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7244 | |
0122ec5b | 7245 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7246 | } |
7c941438 | 7247 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7248 | |
052f1dc7 | 7249 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 | 7250 | #endif |
5cb350ba | 7251 | |
a790de99 | 7252 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7253 | static unsigned long to_ratio(u64 period, u64 runtime) |
7254 | { | |
7255 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7256 | return 1ULL << 20; |
9f0c1e56 | 7257 | |
9a7e0b18 | 7258 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7259 | } |
a790de99 PT |
7260 | #endif |
7261 | ||
7262 | #ifdef CONFIG_RT_GROUP_SCHED | |
7263 | /* | |
7264 | * Ensure that the real time constraints are schedulable. | |
7265 | */ | |
7266 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7267 | |
9a7e0b18 PZ |
7268 | /* Must be called with tasklist_lock held */ |
7269 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7270 | { |
9a7e0b18 | 7271 | struct task_struct *g, *p; |
b40b2e8e | 7272 | |
9a7e0b18 | 7273 | do_each_thread(g, p) { |
029632fb | 7274 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7275 | return 1; |
7276 | } while_each_thread(g, p); | |
b40b2e8e | 7277 | |
9a7e0b18 PZ |
7278 | return 0; |
7279 | } | |
b40b2e8e | 7280 | |
9a7e0b18 PZ |
7281 | struct rt_schedulable_data { |
7282 | struct task_group *tg; | |
7283 | u64 rt_period; | |
7284 | u64 rt_runtime; | |
7285 | }; | |
b40b2e8e | 7286 | |
a790de99 | 7287 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7288 | { |
7289 | struct rt_schedulable_data *d = data; | |
7290 | struct task_group *child; | |
7291 | unsigned long total, sum = 0; | |
7292 | u64 period, runtime; | |
b40b2e8e | 7293 | |
9a7e0b18 PZ |
7294 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7295 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7296 | |
9a7e0b18 PZ |
7297 | if (tg == d->tg) { |
7298 | period = d->rt_period; | |
7299 | runtime = d->rt_runtime; | |
b40b2e8e | 7300 | } |
b40b2e8e | 7301 | |
4653f803 PZ |
7302 | /* |
7303 | * Cannot have more runtime than the period. | |
7304 | */ | |
7305 | if (runtime > period && runtime != RUNTIME_INF) | |
7306 | return -EINVAL; | |
6f505b16 | 7307 | |
4653f803 PZ |
7308 | /* |
7309 | * Ensure we don't starve existing RT tasks. | |
7310 | */ | |
9a7e0b18 PZ |
7311 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7312 | return -EBUSY; | |
6f505b16 | 7313 | |
9a7e0b18 | 7314 | total = to_ratio(period, runtime); |
6f505b16 | 7315 | |
4653f803 PZ |
7316 | /* |
7317 | * Nobody can have more than the global setting allows. | |
7318 | */ | |
7319 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7320 | return -EINVAL; | |
6f505b16 | 7321 | |
4653f803 PZ |
7322 | /* |
7323 | * The sum of our children's runtime should not exceed our own. | |
7324 | */ | |
9a7e0b18 PZ |
7325 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7326 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7327 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7328 | |
9a7e0b18 PZ |
7329 | if (child == d->tg) { |
7330 | period = d->rt_period; | |
7331 | runtime = d->rt_runtime; | |
7332 | } | |
6f505b16 | 7333 | |
9a7e0b18 | 7334 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7335 | } |
6f505b16 | 7336 | |
9a7e0b18 PZ |
7337 | if (sum > total) |
7338 | return -EINVAL; | |
7339 | ||
7340 | return 0; | |
6f505b16 PZ |
7341 | } |
7342 | ||
9a7e0b18 | 7343 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7344 | { |
8277434e PT |
7345 | int ret; |
7346 | ||
9a7e0b18 PZ |
7347 | struct rt_schedulable_data data = { |
7348 | .tg = tg, | |
7349 | .rt_period = period, | |
7350 | .rt_runtime = runtime, | |
7351 | }; | |
7352 | ||
8277434e PT |
7353 | rcu_read_lock(); |
7354 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7355 | rcu_read_unlock(); | |
7356 | ||
7357 | return ret; | |
521f1a24 DG |
7358 | } |
7359 | ||
ab84d31e | 7360 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7361 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7362 | { |
ac086bc2 | 7363 | int i, err = 0; |
9f0c1e56 | 7364 | |
9f0c1e56 | 7365 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7366 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7367 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7368 | if (err) | |
9f0c1e56 | 7369 | goto unlock; |
ac086bc2 | 7370 | |
0986b11b | 7371 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7372 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7373 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7374 | |
7375 | for_each_possible_cpu(i) { | |
7376 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7377 | ||
0986b11b | 7378 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7379 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7380 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7381 | } |
0986b11b | 7382 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7383 | unlock: |
521f1a24 | 7384 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7385 | mutex_unlock(&rt_constraints_mutex); |
7386 | ||
7387 | return err; | |
6f505b16 PZ |
7388 | } |
7389 | ||
d0b27fa7 PZ |
7390 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7391 | { | |
7392 | u64 rt_runtime, rt_period; | |
7393 | ||
7394 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7395 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7396 | if (rt_runtime_us < 0) | |
7397 | rt_runtime = RUNTIME_INF; | |
7398 | ||
ab84d31e | 7399 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7400 | } |
7401 | ||
9f0c1e56 PZ |
7402 | long sched_group_rt_runtime(struct task_group *tg) |
7403 | { | |
7404 | u64 rt_runtime_us; | |
7405 | ||
d0b27fa7 | 7406 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7407 | return -1; |
7408 | ||
d0b27fa7 | 7409 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7410 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7411 | return rt_runtime_us; | |
7412 | } | |
d0b27fa7 PZ |
7413 | |
7414 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7415 | { | |
7416 | u64 rt_runtime, rt_period; | |
7417 | ||
7418 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7419 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7420 | ||
619b0488 R |
7421 | if (rt_period == 0) |
7422 | return -EINVAL; | |
7423 | ||
ab84d31e | 7424 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7425 | } |
7426 | ||
7427 | long sched_group_rt_period(struct task_group *tg) | |
7428 | { | |
7429 | u64 rt_period_us; | |
7430 | ||
7431 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7432 | do_div(rt_period_us, NSEC_PER_USEC); | |
7433 | return rt_period_us; | |
7434 | } | |
7435 | ||
7436 | static int sched_rt_global_constraints(void) | |
7437 | { | |
4653f803 | 7438 | u64 runtime, period; |
d0b27fa7 PZ |
7439 | int ret = 0; |
7440 | ||
ec5d4989 HS |
7441 | if (sysctl_sched_rt_period <= 0) |
7442 | return -EINVAL; | |
7443 | ||
4653f803 PZ |
7444 | runtime = global_rt_runtime(); |
7445 | period = global_rt_period(); | |
7446 | ||
7447 | /* | |
7448 | * Sanity check on the sysctl variables. | |
7449 | */ | |
7450 | if (runtime > period && runtime != RUNTIME_INF) | |
7451 | return -EINVAL; | |
10b612f4 | 7452 | |
d0b27fa7 | 7453 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7454 | read_lock(&tasklist_lock); |
4653f803 | 7455 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7456 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7457 | mutex_unlock(&rt_constraints_mutex); |
7458 | ||
7459 | return ret; | |
7460 | } | |
54e99124 DG |
7461 | |
7462 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7463 | { | |
7464 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7465 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7466 | return 0; | |
7467 | ||
7468 | return 1; | |
7469 | } | |
7470 | ||
6d6bc0ad | 7471 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7472 | static int sched_rt_global_constraints(void) |
7473 | { | |
ac086bc2 PZ |
7474 | unsigned long flags; |
7475 | int i; | |
7476 | ||
ec5d4989 HS |
7477 | if (sysctl_sched_rt_period <= 0) |
7478 | return -EINVAL; | |
7479 | ||
60aa605d PZ |
7480 | /* |
7481 | * There's always some RT tasks in the root group | |
7482 | * -- migration, kstopmachine etc.. | |
7483 | */ | |
7484 | if (sysctl_sched_rt_runtime == 0) | |
7485 | return -EBUSY; | |
7486 | ||
0986b11b | 7487 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7488 | for_each_possible_cpu(i) { |
7489 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7490 | ||
0986b11b | 7491 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7492 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7493 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7494 | } |
0986b11b | 7495 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7496 | |
d0b27fa7 PZ |
7497 | return 0; |
7498 | } | |
6d6bc0ad | 7499 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7500 | |
7501 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7502 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7503 | loff_t *ppos) |
7504 | { | |
7505 | int ret; | |
7506 | int old_period, old_runtime; | |
7507 | static DEFINE_MUTEX(mutex); | |
7508 | ||
7509 | mutex_lock(&mutex); | |
7510 | old_period = sysctl_sched_rt_period; | |
7511 | old_runtime = sysctl_sched_rt_runtime; | |
7512 | ||
8d65af78 | 7513 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7514 | |
7515 | if (!ret && write) { | |
7516 | ret = sched_rt_global_constraints(); | |
7517 | if (ret) { | |
7518 | sysctl_sched_rt_period = old_period; | |
7519 | sysctl_sched_rt_runtime = old_runtime; | |
7520 | } else { | |
7521 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7522 | def_rt_bandwidth.rt_period = | |
7523 | ns_to_ktime(global_rt_period()); | |
7524 | } | |
7525 | } | |
7526 | mutex_unlock(&mutex); | |
7527 | ||
7528 | return ret; | |
7529 | } | |
68318b8e | 7530 | |
052f1dc7 | 7531 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7532 | |
7533 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7534 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7535 | { |
2b01dfe3 PM |
7536 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7537 | struct task_group, css); | |
68318b8e SV |
7538 | } |
7539 | ||
7540 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7541 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 7542 | { |
ec7dc8ac | 7543 | struct task_group *tg, *parent; |
68318b8e | 7544 | |
2b01dfe3 | 7545 | if (!cgrp->parent) { |
68318b8e | 7546 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7547 | return &root_task_group.css; |
68318b8e SV |
7548 | } |
7549 | ||
ec7dc8ac DG |
7550 | parent = cgroup_tg(cgrp->parent); |
7551 | tg = sched_create_group(parent); | |
68318b8e SV |
7552 | if (IS_ERR(tg)) |
7553 | return ERR_PTR(-ENOMEM); | |
7554 | ||
68318b8e SV |
7555 | return &tg->css; |
7556 | } | |
7557 | ||
41a2d6cf IM |
7558 | static void |
7559 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7560 | { |
2b01dfe3 | 7561 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7562 | |
7563 | sched_destroy_group(tg); | |
7564 | } | |
7565 | ||
bb9d97b6 TH |
7566 | static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7567 | struct cgroup_taskset *tset) | |
68318b8e | 7568 | { |
bb9d97b6 TH |
7569 | struct task_struct *task; |
7570 | ||
7571 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7572 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7573 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7574 | return -EINVAL; | |
b68aa230 | 7575 | #else |
bb9d97b6 TH |
7576 | /* We don't support RT-tasks being in separate groups */ |
7577 | if (task->sched_class != &fair_sched_class) | |
7578 | return -EINVAL; | |
b68aa230 | 7579 | #endif |
bb9d97b6 | 7580 | } |
be367d09 BB |
7581 | return 0; |
7582 | } | |
68318b8e | 7583 | |
bb9d97b6 TH |
7584 | static void cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7585 | struct cgroup_taskset *tset) | |
68318b8e | 7586 | { |
bb9d97b6 TH |
7587 | struct task_struct *task; |
7588 | ||
7589 | cgroup_taskset_for_each(task, cgrp, tset) | |
7590 | sched_move_task(task); | |
68318b8e SV |
7591 | } |
7592 | ||
068c5cc5 | 7593 | static void |
d41d5a01 PZ |
7594 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7595 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
7596 | { |
7597 | /* | |
7598 | * cgroup_exit() is called in the copy_process() failure path. | |
7599 | * Ignore this case since the task hasn't ran yet, this avoids | |
7600 | * trying to poke a half freed task state from generic code. | |
7601 | */ | |
7602 | if (!(task->flags & PF_EXITING)) | |
7603 | return; | |
7604 | ||
7605 | sched_move_task(task); | |
7606 | } | |
7607 | ||
052f1dc7 | 7608 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7609 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7610 | u64 shareval) |
68318b8e | 7611 | { |
c8b28116 | 7612 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7613 | } |
7614 | ||
f4c753b7 | 7615 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7616 | { |
2b01dfe3 | 7617 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7618 | |
c8b28116 | 7619 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7620 | } |
ab84d31e PT |
7621 | |
7622 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7623 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7624 | ||
ab84d31e PT |
7625 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7626 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7627 | ||
a790de99 PT |
7628 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7629 | ||
ab84d31e PT |
7630 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7631 | { | |
56f570e5 | 7632 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7633 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7634 | |
7635 | if (tg == &root_task_group) | |
7636 | return -EINVAL; | |
7637 | ||
7638 | /* | |
7639 | * Ensure we have at some amount of bandwidth every period. This is | |
7640 | * to prevent reaching a state of large arrears when throttled via | |
7641 | * entity_tick() resulting in prolonged exit starvation. | |
7642 | */ | |
7643 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7644 | return -EINVAL; | |
7645 | ||
7646 | /* | |
7647 | * Likewise, bound things on the otherside by preventing insane quota | |
7648 | * periods. This also allows us to normalize in computing quota | |
7649 | * feasibility. | |
7650 | */ | |
7651 | if (period > max_cfs_quota_period) | |
7652 | return -EINVAL; | |
7653 | ||
a790de99 PT |
7654 | mutex_lock(&cfs_constraints_mutex); |
7655 | ret = __cfs_schedulable(tg, period, quota); | |
7656 | if (ret) | |
7657 | goto out_unlock; | |
7658 | ||
58088ad0 | 7659 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7660 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7661 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7662 | raw_spin_lock_irq(&cfs_b->lock); |
7663 | cfs_b->period = ns_to_ktime(period); | |
7664 | cfs_b->quota = quota; | |
58088ad0 | 7665 | |
a9cf55b2 | 7666 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7667 | /* restart the period timer (if active) to handle new period expiry */ |
7668 | if (runtime_enabled && cfs_b->timer_active) { | |
7669 | /* force a reprogram */ | |
7670 | cfs_b->timer_active = 0; | |
7671 | __start_cfs_bandwidth(cfs_b); | |
7672 | } | |
ab84d31e PT |
7673 | raw_spin_unlock_irq(&cfs_b->lock); |
7674 | ||
7675 | for_each_possible_cpu(i) { | |
7676 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7677 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7678 | |
7679 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7680 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7681 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7682 | |
029632fb | 7683 | if (cfs_rq->throttled) |
671fd9da | 7684 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7685 | raw_spin_unlock_irq(&rq->lock); |
7686 | } | |
a790de99 PT |
7687 | out_unlock: |
7688 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7689 | |
a790de99 | 7690 | return ret; |
ab84d31e PT |
7691 | } |
7692 | ||
7693 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7694 | { | |
7695 | u64 quota, period; | |
7696 | ||
029632fb | 7697 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7698 | if (cfs_quota_us < 0) |
7699 | quota = RUNTIME_INF; | |
7700 | else | |
7701 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7702 | ||
7703 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7704 | } | |
7705 | ||
7706 | long tg_get_cfs_quota(struct task_group *tg) | |
7707 | { | |
7708 | u64 quota_us; | |
7709 | ||
029632fb | 7710 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7711 | return -1; |
7712 | ||
029632fb | 7713 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7714 | do_div(quota_us, NSEC_PER_USEC); |
7715 | ||
7716 | return quota_us; | |
7717 | } | |
7718 | ||
7719 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7720 | { | |
7721 | u64 quota, period; | |
7722 | ||
7723 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7724 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7725 | |
ab84d31e PT |
7726 | return tg_set_cfs_bandwidth(tg, period, quota); |
7727 | } | |
7728 | ||
7729 | long tg_get_cfs_period(struct task_group *tg) | |
7730 | { | |
7731 | u64 cfs_period_us; | |
7732 | ||
029632fb | 7733 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7734 | do_div(cfs_period_us, NSEC_PER_USEC); |
7735 | ||
7736 | return cfs_period_us; | |
7737 | } | |
7738 | ||
7739 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7740 | { | |
7741 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7742 | } | |
7743 | ||
7744 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7745 | s64 cfs_quota_us) | |
7746 | { | |
7747 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7748 | } | |
7749 | ||
7750 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7751 | { | |
7752 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7753 | } | |
7754 | ||
7755 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7756 | u64 cfs_period_us) | |
7757 | { | |
7758 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7759 | } | |
7760 | ||
a790de99 PT |
7761 | struct cfs_schedulable_data { |
7762 | struct task_group *tg; | |
7763 | u64 period, quota; | |
7764 | }; | |
7765 | ||
7766 | /* | |
7767 | * normalize group quota/period to be quota/max_period | |
7768 | * note: units are usecs | |
7769 | */ | |
7770 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7771 | struct cfs_schedulable_data *d) | |
7772 | { | |
7773 | u64 quota, period; | |
7774 | ||
7775 | if (tg == d->tg) { | |
7776 | period = d->period; | |
7777 | quota = d->quota; | |
7778 | } else { | |
7779 | period = tg_get_cfs_period(tg); | |
7780 | quota = tg_get_cfs_quota(tg); | |
7781 | } | |
7782 | ||
7783 | /* note: these should typically be equivalent */ | |
7784 | if (quota == RUNTIME_INF || quota == -1) | |
7785 | return RUNTIME_INF; | |
7786 | ||
7787 | return to_ratio(period, quota); | |
7788 | } | |
7789 | ||
7790 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7791 | { | |
7792 | struct cfs_schedulable_data *d = data; | |
029632fb | 7793 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7794 | s64 quota = 0, parent_quota = -1; |
7795 | ||
7796 | if (!tg->parent) { | |
7797 | quota = RUNTIME_INF; | |
7798 | } else { | |
029632fb | 7799 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7800 | |
7801 | quota = normalize_cfs_quota(tg, d); | |
7802 | parent_quota = parent_b->hierarchal_quota; | |
7803 | ||
7804 | /* | |
7805 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7806 | * limit is set | |
7807 | */ | |
7808 | if (quota == RUNTIME_INF) | |
7809 | quota = parent_quota; | |
7810 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7811 | return -EINVAL; | |
7812 | } | |
7813 | cfs_b->hierarchal_quota = quota; | |
7814 | ||
7815 | return 0; | |
7816 | } | |
7817 | ||
7818 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7819 | { | |
8277434e | 7820 | int ret; |
a790de99 PT |
7821 | struct cfs_schedulable_data data = { |
7822 | .tg = tg, | |
7823 | .period = period, | |
7824 | .quota = quota, | |
7825 | }; | |
7826 | ||
7827 | if (quota != RUNTIME_INF) { | |
7828 | do_div(data.period, NSEC_PER_USEC); | |
7829 | do_div(data.quota, NSEC_PER_USEC); | |
7830 | } | |
7831 | ||
8277434e PT |
7832 | rcu_read_lock(); |
7833 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7834 | rcu_read_unlock(); | |
7835 | ||
7836 | return ret; | |
a790de99 | 7837 | } |
e8da1b18 NR |
7838 | |
7839 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7840 | struct cgroup_map_cb *cb) | |
7841 | { | |
7842 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7843 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7844 | |
7845 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7846 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7847 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7848 | ||
7849 | return 0; | |
7850 | } | |
ab84d31e | 7851 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7852 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7853 | |
052f1dc7 | 7854 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7855 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7856 | s64 val) |
6f505b16 | 7857 | { |
06ecb27c | 7858 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7859 | } |
7860 | ||
06ecb27c | 7861 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7862 | { |
06ecb27c | 7863 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7864 | } |
d0b27fa7 PZ |
7865 | |
7866 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7867 | u64 rt_period_us) | |
7868 | { | |
7869 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7870 | } | |
7871 | ||
7872 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7873 | { | |
7874 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7875 | } | |
6d6bc0ad | 7876 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7877 | |
fe5c7cc2 | 7878 | static struct cftype cpu_files[] = { |
052f1dc7 | 7879 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7880 | { |
7881 | .name = "shares", | |
f4c753b7 PM |
7882 | .read_u64 = cpu_shares_read_u64, |
7883 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7884 | }, |
052f1dc7 | 7885 | #endif |
ab84d31e PT |
7886 | #ifdef CONFIG_CFS_BANDWIDTH |
7887 | { | |
7888 | .name = "cfs_quota_us", | |
7889 | .read_s64 = cpu_cfs_quota_read_s64, | |
7890 | .write_s64 = cpu_cfs_quota_write_s64, | |
7891 | }, | |
7892 | { | |
7893 | .name = "cfs_period_us", | |
7894 | .read_u64 = cpu_cfs_period_read_u64, | |
7895 | .write_u64 = cpu_cfs_period_write_u64, | |
7896 | }, | |
e8da1b18 NR |
7897 | { |
7898 | .name = "stat", | |
7899 | .read_map = cpu_stats_show, | |
7900 | }, | |
ab84d31e | 7901 | #endif |
052f1dc7 | 7902 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7903 | { |
9f0c1e56 | 7904 | .name = "rt_runtime_us", |
06ecb27c PM |
7905 | .read_s64 = cpu_rt_runtime_read, |
7906 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7907 | }, |
d0b27fa7 PZ |
7908 | { |
7909 | .name = "rt_period_us", | |
f4c753b7 PM |
7910 | .read_u64 = cpu_rt_period_read_uint, |
7911 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7912 | }, |
052f1dc7 | 7913 | #endif |
68318b8e SV |
7914 | }; |
7915 | ||
7916 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7917 | { | |
fe5c7cc2 | 7918 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7919 | } |
7920 | ||
7921 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7922 | .name = "cpu", |
7923 | .create = cpu_cgroup_create, | |
7924 | .destroy = cpu_cgroup_destroy, | |
bb9d97b6 TH |
7925 | .can_attach = cpu_cgroup_can_attach, |
7926 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7927 | .exit = cpu_cgroup_exit, |
38605cae IM |
7928 | .populate = cpu_cgroup_populate, |
7929 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7930 | .early_init = 1, |
7931 | }; | |
7932 | ||
052f1dc7 | 7933 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7934 | |
7935 | #ifdef CONFIG_CGROUP_CPUACCT | |
7936 | ||
7937 | /* | |
7938 | * CPU accounting code for task groups. | |
7939 | * | |
7940 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7941 | * (balbir@in.ibm.com). | |
7942 | */ | |
7943 | ||
d842de87 SV |
7944 | /* create a new cpu accounting group */ |
7945 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 7946 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7947 | { |
54c707e9 | 7948 | struct cpuacct *ca; |
d842de87 | 7949 | |
54c707e9 GC |
7950 | if (!cgrp->parent) |
7951 | return &root_cpuacct.css; | |
7952 | ||
7953 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 7954 | if (!ca) |
ef12fefa | 7955 | goto out; |
d842de87 SV |
7956 | |
7957 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
7958 | if (!ca->cpuusage) |
7959 | goto out_free_ca; | |
7960 | ||
54c707e9 GC |
7961 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
7962 | if (!ca->cpustat) | |
7963 | goto out_free_cpuusage; | |
934352f2 | 7964 | |
d842de87 | 7965 | return &ca->css; |
ef12fefa | 7966 | |
54c707e9 | 7967 | out_free_cpuusage: |
ef12fefa BR |
7968 | free_percpu(ca->cpuusage); |
7969 | out_free_ca: | |
7970 | kfree(ca); | |
7971 | out: | |
7972 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
7973 | } |
7974 | ||
7975 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 7976 | static void |
32cd756a | 7977 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7978 | { |
32cd756a | 7979 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 7980 | |
54c707e9 | 7981 | free_percpu(ca->cpustat); |
d842de87 SV |
7982 | free_percpu(ca->cpuusage); |
7983 | kfree(ca); | |
7984 | } | |
7985 | ||
720f5498 KC |
7986 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
7987 | { | |
b36128c8 | 7988 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
7989 | u64 data; |
7990 | ||
7991 | #ifndef CONFIG_64BIT | |
7992 | /* | |
7993 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
7994 | */ | |
05fa785c | 7995 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 7996 | data = *cpuusage; |
05fa785c | 7997 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
7998 | #else |
7999 | data = *cpuusage; | |
8000 | #endif | |
8001 | ||
8002 | return data; | |
8003 | } | |
8004 | ||
8005 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8006 | { | |
b36128c8 | 8007 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8008 | |
8009 | #ifndef CONFIG_64BIT | |
8010 | /* | |
8011 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8012 | */ | |
05fa785c | 8013 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8014 | *cpuusage = val; |
05fa785c | 8015 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8016 | #else |
8017 | *cpuusage = val; | |
8018 | #endif | |
8019 | } | |
8020 | ||
d842de87 | 8021 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8022 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8023 | { |
32cd756a | 8024 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8025 | u64 totalcpuusage = 0; |
8026 | int i; | |
8027 | ||
720f5498 KC |
8028 | for_each_present_cpu(i) |
8029 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8030 | |
8031 | return totalcpuusage; | |
8032 | } | |
8033 | ||
0297b803 DG |
8034 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8035 | u64 reset) | |
8036 | { | |
8037 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8038 | int err = 0; | |
8039 | int i; | |
8040 | ||
8041 | if (reset) { | |
8042 | err = -EINVAL; | |
8043 | goto out; | |
8044 | } | |
8045 | ||
720f5498 KC |
8046 | for_each_present_cpu(i) |
8047 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8048 | |
0297b803 DG |
8049 | out: |
8050 | return err; | |
8051 | } | |
8052 | ||
e9515c3c KC |
8053 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8054 | struct seq_file *m) | |
8055 | { | |
8056 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8057 | u64 percpu; | |
8058 | int i; | |
8059 | ||
8060 | for_each_present_cpu(i) { | |
8061 | percpu = cpuacct_cpuusage_read(ca, i); | |
8062 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8063 | } | |
8064 | seq_printf(m, "\n"); | |
8065 | return 0; | |
8066 | } | |
8067 | ||
ef12fefa BR |
8068 | static const char *cpuacct_stat_desc[] = { |
8069 | [CPUACCT_STAT_USER] = "user", | |
8070 | [CPUACCT_STAT_SYSTEM] = "system", | |
8071 | }; | |
8072 | ||
8073 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8074 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8075 | { |
8076 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8077 | int cpu; |
8078 | s64 val = 0; | |
ef12fefa | 8079 | |
54c707e9 GC |
8080 | for_each_online_cpu(cpu) { |
8081 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8082 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8083 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8084 | } |
54c707e9 GC |
8085 | val = cputime64_to_clock_t(val); |
8086 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8087 | |
54c707e9 GC |
8088 | val = 0; |
8089 | for_each_online_cpu(cpu) { | |
8090 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8091 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8092 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8093 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8094 | } |
54c707e9 GC |
8095 | |
8096 | val = cputime64_to_clock_t(val); | |
8097 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8098 | ||
ef12fefa BR |
8099 | return 0; |
8100 | } | |
8101 | ||
d842de87 SV |
8102 | static struct cftype files[] = { |
8103 | { | |
8104 | .name = "usage", | |
f4c753b7 PM |
8105 | .read_u64 = cpuusage_read, |
8106 | .write_u64 = cpuusage_write, | |
d842de87 | 8107 | }, |
e9515c3c KC |
8108 | { |
8109 | .name = "usage_percpu", | |
8110 | .read_seq_string = cpuacct_percpu_seq_read, | |
8111 | }, | |
ef12fefa BR |
8112 | { |
8113 | .name = "stat", | |
8114 | .read_map = cpuacct_stats_show, | |
8115 | }, | |
d842de87 SV |
8116 | }; |
8117 | ||
32cd756a | 8118 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8119 | { |
32cd756a | 8120 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8121 | } |
8122 | ||
8123 | /* | |
8124 | * charge this task's execution time to its accounting group. | |
8125 | * | |
8126 | * called with rq->lock held. | |
8127 | */ | |
029632fb | 8128 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8129 | { |
8130 | struct cpuacct *ca; | |
934352f2 | 8131 | int cpu; |
d842de87 | 8132 | |
c40c6f85 | 8133 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8134 | return; |
8135 | ||
934352f2 | 8136 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8137 | |
8138 | rcu_read_lock(); | |
8139 | ||
d842de87 | 8140 | ca = task_ca(tsk); |
d842de87 | 8141 | |
44252e42 | 8142 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8143 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8144 | *cpuusage += cputime; |
8145 | } | |
a18b83b7 BR |
8146 | |
8147 | rcu_read_unlock(); | |
d842de87 SV |
8148 | } |
8149 | ||
8150 | struct cgroup_subsys cpuacct_subsys = { | |
8151 | .name = "cpuacct", | |
8152 | .create = cpuacct_create, | |
8153 | .destroy = cpuacct_destroy, | |
8154 | .populate = cpuacct_populate, | |
8155 | .subsys_id = cpuacct_subsys_id, | |
8156 | }; | |
8157 | #endif /* CONFIG_CGROUP_CPUACCT */ |