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> |
db7e527d | 77 | #include <asm/mutex.h> |
e6e6685a GC |
78 | #ifdef CONFIG_PARAVIRT |
79 | #include <asm/paravirt.h> | |
80 | #endif | |
1da177e4 | 81 | |
029632fb | 82 | #include "sched.h" |
391e43da | 83 | #include "../workqueue_sched.h" |
6e0534f2 | 84 | |
a8d154b0 | 85 | #define CREATE_TRACE_POINTS |
ad8d75ff | 86 | #include <trace/events/sched.h> |
a8d154b0 | 87 | |
029632fb | 88 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 89 | { |
58088ad0 PT |
90 | unsigned long delta; |
91 | ktime_t soft, hard, now; | |
d0b27fa7 | 92 | |
58088ad0 PT |
93 | for (;;) { |
94 | if (hrtimer_active(period_timer)) | |
95 | break; | |
96 | ||
97 | now = hrtimer_cb_get_time(period_timer); | |
98 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 99 | |
58088ad0 PT |
100 | soft = hrtimer_get_softexpires(period_timer); |
101 | hard = hrtimer_get_expires(period_timer); | |
102 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
103 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
104 | HRTIMER_MODE_ABS_PINNED, 0); | |
105 | } | |
106 | } | |
107 | ||
029632fb PZ |
108 | DEFINE_MUTEX(sched_domains_mutex); |
109 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 110 | |
fe44d621 | 111 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 112 | |
029632fb | 113 | void update_rq_clock(struct rq *rq) |
3e51f33f | 114 | { |
fe44d621 | 115 | s64 delta; |
305e6835 | 116 | |
61eadef6 | 117 | if (rq->skip_clock_update > 0) |
f26f9aff | 118 | return; |
aa483808 | 119 | |
fe44d621 PZ |
120 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
121 | rq->clock += delta; | |
122 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
123 | } |
124 | ||
bf5c91ba IM |
125 | /* |
126 | * Debugging: various feature bits | |
127 | */ | |
f00b45c1 | 128 | |
f00b45c1 PZ |
129 | #define SCHED_FEAT(name, enabled) \ |
130 | (1UL << __SCHED_FEAT_##name) * enabled | | |
131 | ||
bf5c91ba | 132 | const_debug unsigned int sysctl_sched_features = |
391e43da | 133 | #include "features.h" |
f00b45c1 PZ |
134 | 0; |
135 | ||
136 | #undef SCHED_FEAT | |
137 | ||
138 | #ifdef CONFIG_SCHED_DEBUG | |
139 | #define SCHED_FEAT(name, enabled) \ | |
140 | #name , | |
141 | ||
983ed7a6 | 142 | static __read_mostly char *sched_feat_names[] = { |
391e43da | 143 | #include "features.h" |
f00b45c1 PZ |
144 | NULL |
145 | }; | |
146 | ||
147 | #undef SCHED_FEAT | |
148 | ||
34f3a814 | 149 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 150 | { |
f00b45c1 PZ |
151 | int i; |
152 | ||
f8b6d1cc | 153 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
154 | if (!(sysctl_sched_features & (1UL << i))) |
155 | seq_puts(m, "NO_"); | |
156 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 157 | } |
34f3a814 | 158 | seq_puts(m, "\n"); |
f00b45c1 | 159 | |
34f3a814 | 160 | return 0; |
f00b45c1 PZ |
161 | } |
162 | ||
f8b6d1cc PZ |
163 | #ifdef HAVE_JUMP_LABEL |
164 | ||
165 | #define jump_label_key__true jump_label_key_enabled | |
166 | #define jump_label_key__false jump_label_key_disabled | |
167 | ||
168 | #define SCHED_FEAT(name, enabled) \ | |
169 | jump_label_key__##enabled , | |
170 | ||
171 | struct jump_label_key sched_feat_keys[__SCHED_FEAT_NR] = { | |
172 | #include "features.h" | |
173 | }; | |
174 | ||
175 | #undef SCHED_FEAT | |
176 | ||
177 | static void sched_feat_disable(int i) | |
178 | { | |
179 | if (jump_label_enabled(&sched_feat_keys[i])) | |
180 | jump_label_dec(&sched_feat_keys[i]); | |
181 | } | |
182 | ||
183 | static void sched_feat_enable(int i) | |
184 | { | |
185 | if (!jump_label_enabled(&sched_feat_keys[i])) | |
186 | jump_label_inc(&sched_feat_keys[i]); | |
187 | } | |
188 | #else | |
189 | static void sched_feat_disable(int i) { }; | |
190 | static void sched_feat_enable(int i) { }; | |
191 | #endif /* HAVE_JUMP_LABEL */ | |
192 | ||
f00b45c1 PZ |
193 | static ssize_t |
194 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
195 | size_t cnt, loff_t *ppos) | |
196 | { | |
197 | char buf[64]; | |
7740191c | 198 | char *cmp; |
f00b45c1 PZ |
199 | int neg = 0; |
200 | int i; | |
201 | ||
202 | if (cnt > 63) | |
203 | cnt = 63; | |
204 | ||
205 | if (copy_from_user(&buf, ubuf, cnt)) | |
206 | return -EFAULT; | |
207 | ||
208 | buf[cnt] = 0; | |
7740191c | 209 | cmp = strstrip(buf); |
f00b45c1 | 210 | |
524429c3 | 211 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
212 | neg = 1; |
213 | cmp += 3; | |
214 | } | |
215 | ||
f8b6d1cc | 216 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 217 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 218 | if (neg) { |
f00b45c1 | 219 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
220 | sched_feat_disable(i); |
221 | } else { | |
f00b45c1 | 222 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
223 | sched_feat_enable(i); |
224 | } | |
f00b45c1 PZ |
225 | break; |
226 | } | |
227 | } | |
228 | ||
f8b6d1cc | 229 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
230 | return -EINVAL; |
231 | ||
42994724 | 232 | *ppos += cnt; |
f00b45c1 PZ |
233 | |
234 | return cnt; | |
235 | } | |
236 | ||
34f3a814 LZ |
237 | static int sched_feat_open(struct inode *inode, struct file *filp) |
238 | { | |
239 | return single_open(filp, sched_feat_show, NULL); | |
240 | } | |
241 | ||
828c0950 | 242 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
243 | .open = sched_feat_open, |
244 | .write = sched_feat_write, | |
245 | .read = seq_read, | |
246 | .llseek = seq_lseek, | |
247 | .release = single_release, | |
f00b45c1 PZ |
248 | }; |
249 | ||
250 | static __init int sched_init_debug(void) | |
251 | { | |
f00b45c1 PZ |
252 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
253 | &sched_feat_fops); | |
254 | ||
255 | return 0; | |
256 | } | |
257 | late_initcall(sched_init_debug); | |
f8b6d1cc | 258 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 259 | |
b82d9fdd PZ |
260 | /* |
261 | * Number of tasks to iterate in a single balance run. | |
262 | * Limited because this is done with IRQs disabled. | |
263 | */ | |
264 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
265 | ||
e9e9250b PZ |
266 | /* |
267 | * period over which we average the RT time consumption, measured | |
268 | * in ms. | |
269 | * | |
270 | * default: 1s | |
271 | */ | |
272 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
273 | ||
fa85ae24 | 274 | /* |
9f0c1e56 | 275 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
276 | * default: 1s |
277 | */ | |
9f0c1e56 | 278 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 279 | |
029632fb | 280 | __read_mostly int scheduler_running; |
6892b75e | 281 | |
9f0c1e56 PZ |
282 | /* |
283 | * part of the period that we allow rt tasks to run in us. | |
284 | * default: 0.95s | |
285 | */ | |
286 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 287 | |
fa85ae24 | 288 | |
1da177e4 | 289 | |
0970d299 | 290 | /* |
0122ec5b | 291 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 292 | */ |
70b97a7f | 293 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
294 | __acquires(rq->lock) |
295 | { | |
0970d299 PZ |
296 | struct rq *rq; |
297 | ||
0122ec5b PZ |
298 | lockdep_assert_held(&p->pi_lock); |
299 | ||
3a5c359a | 300 | for (;;) { |
0970d299 | 301 | rq = task_rq(p); |
05fa785c | 302 | raw_spin_lock(&rq->lock); |
65cc8e48 | 303 | if (likely(rq == task_rq(p))) |
3a5c359a | 304 | return rq; |
05fa785c | 305 | raw_spin_unlock(&rq->lock); |
b29739f9 | 306 | } |
b29739f9 IM |
307 | } |
308 | ||
1da177e4 | 309 | /* |
0122ec5b | 310 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 311 | */ |
70b97a7f | 312 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 313 | __acquires(p->pi_lock) |
1da177e4 LT |
314 | __acquires(rq->lock) |
315 | { | |
70b97a7f | 316 | struct rq *rq; |
1da177e4 | 317 | |
3a5c359a | 318 | for (;;) { |
0122ec5b | 319 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 320 | rq = task_rq(p); |
05fa785c | 321 | raw_spin_lock(&rq->lock); |
65cc8e48 | 322 | if (likely(rq == task_rq(p))) |
3a5c359a | 323 | return rq; |
0122ec5b PZ |
324 | raw_spin_unlock(&rq->lock); |
325 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 326 | } |
1da177e4 LT |
327 | } |
328 | ||
a9957449 | 329 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
330 | __releases(rq->lock) |
331 | { | |
05fa785c | 332 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
333 | } |
334 | ||
0122ec5b PZ |
335 | static inline void |
336 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 337 | __releases(rq->lock) |
0122ec5b | 338 | __releases(p->pi_lock) |
1da177e4 | 339 | { |
0122ec5b PZ |
340 | raw_spin_unlock(&rq->lock); |
341 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
342 | } |
343 | ||
1da177e4 | 344 | /* |
cc2a73b5 | 345 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 346 | */ |
a9957449 | 347 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
348 | __acquires(rq->lock) |
349 | { | |
70b97a7f | 350 | struct rq *rq; |
1da177e4 LT |
351 | |
352 | local_irq_disable(); | |
353 | rq = this_rq(); | |
05fa785c | 354 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
355 | |
356 | return rq; | |
357 | } | |
358 | ||
8f4d37ec PZ |
359 | #ifdef CONFIG_SCHED_HRTICK |
360 | /* | |
361 | * Use HR-timers to deliver accurate preemption points. | |
362 | * | |
363 | * Its all a bit involved since we cannot program an hrt while holding the | |
364 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
365 | * reschedule event. | |
366 | * | |
367 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
368 | * rq->lock. | |
369 | */ | |
8f4d37ec | 370 | |
8f4d37ec PZ |
371 | static void hrtick_clear(struct rq *rq) |
372 | { | |
373 | if (hrtimer_active(&rq->hrtick_timer)) | |
374 | hrtimer_cancel(&rq->hrtick_timer); | |
375 | } | |
376 | ||
8f4d37ec PZ |
377 | /* |
378 | * High-resolution timer tick. | |
379 | * Runs from hardirq context with interrupts disabled. | |
380 | */ | |
381 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
382 | { | |
383 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
384 | ||
385 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
386 | ||
05fa785c | 387 | raw_spin_lock(&rq->lock); |
3e51f33f | 388 | update_rq_clock(rq); |
8f4d37ec | 389 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 390 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
391 | |
392 | return HRTIMER_NORESTART; | |
393 | } | |
394 | ||
95e904c7 | 395 | #ifdef CONFIG_SMP |
31656519 PZ |
396 | /* |
397 | * called from hardirq (IPI) context | |
398 | */ | |
399 | static void __hrtick_start(void *arg) | |
b328ca18 | 400 | { |
31656519 | 401 | struct rq *rq = arg; |
b328ca18 | 402 | |
05fa785c | 403 | raw_spin_lock(&rq->lock); |
31656519 PZ |
404 | hrtimer_restart(&rq->hrtick_timer); |
405 | rq->hrtick_csd_pending = 0; | |
05fa785c | 406 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
407 | } |
408 | ||
31656519 PZ |
409 | /* |
410 | * Called to set the hrtick timer state. | |
411 | * | |
412 | * called with rq->lock held and irqs disabled | |
413 | */ | |
029632fb | 414 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 415 | { |
31656519 PZ |
416 | struct hrtimer *timer = &rq->hrtick_timer; |
417 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 418 | |
cc584b21 | 419 | hrtimer_set_expires(timer, time); |
31656519 PZ |
420 | |
421 | if (rq == this_rq()) { | |
422 | hrtimer_restart(timer); | |
423 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 424 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
425 | rq->hrtick_csd_pending = 1; |
426 | } | |
b328ca18 PZ |
427 | } |
428 | ||
429 | static int | |
430 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
431 | { | |
432 | int cpu = (int)(long)hcpu; | |
433 | ||
434 | switch (action) { | |
435 | case CPU_UP_CANCELED: | |
436 | case CPU_UP_CANCELED_FROZEN: | |
437 | case CPU_DOWN_PREPARE: | |
438 | case CPU_DOWN_PREPARE_FROZEN: | |
439 | case CPU_DEAD: | |
440 | case CPU_DEAD_FROZEN: | |
31656519 | 441 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
442 | return NOTIFY_OK; |
443 | } | |
444 | ||
445 | return NOTIFY_DONE; | |
446 | } | |
447 | ||
fa748203 | 448 | static __init void init_hrtick(void) |
b328ca18 PZ |
449 | { |
450 | hotcpu_notifier(hotplug_hrtick, 0); | |
451 | } | |
31656519 PZ |
452 | #else |
453 | /* | |
454 | * Called to set the hrtick timer state. | |
455 | * | |
456 | * called with rq->lock held and irqs disabled | |
457 | */ | |
029632fb | 458 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 459 | { |
7f1e2ca9 | 460 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 461 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 462 | } |
b328ca18 | 463 | |
006c75f1 | 464 | static inline void init_hrtick(void) |
8f4d37ec | 465 | { |
8f4d37ec | 466 | } |
31656519 | 467 | #endif /* CONFIG_SMP */ |
8f4d37ec | 468 | |
31656519 | 469 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 470 | { |
31656519 PZ |
471 | #ifdef CONFIG_SMP |
472 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 473 | |
31656519 PZ |
474 | rq->hrtick_csd.flags = 0; |
475 | rq->hrtick_csd.func = __hrtick_start; | |
476 | rq->hrtick_csd.info = rq; | |
477 | #endif | |
8f4d37ec | 478 | |
31656519 PZ |
479 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
480 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 481 | } |
006c75f1 | 482 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
483 | static inline void hrtick_clear(struct rq *rq) |
484 | { | |
485 | } | |
486 | ||
8f4d37ec PZ |
487 | static inline void init_rq_hrtick(struct rq *rq) |
488 | { | |
489 | } | |
490 | ||
b328ca18 PZ |
491 | static inline void init_hrtick(void) |
492 | { | |
493 | } | |
006c75f1 | 494 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 495 | |
c24d20db IM |
496 | /* |
497 | * resched_task - mark a task 'to be rescheduled now'. | |
498 | * | |
499 | * On UP this means the setting of the need_resched flag, on SMP it | |
500 | * might also involve a cross-CPU call to trigger the scheduler on | |
501 | * the target CPU. | |
502 | */ | |
503 | #ifdef CONFIG_SMP | |
504 | ||
505 | #ifndef tsk_is_polling | |
506 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
507 | #endif | |
508 | ||
029632fb | 509 | void resched_task(struct task_struct *p) |
c24d20db IM |
510 | { |
511 | int cpu; | |
512 | ||
05fa785c | 513 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 514 | |
5ed0cec0 | 515 | if (test_tsk_need_resched(p)) |
c24d20db IM |
516 | return; |
517 | ||
5ed0cec0 | 518 | set_tsk_need_resched(p); |
c24d20db IM |
519 | |
520 | cpu = task_cpu(p); | |
521 | if (cpu == smp_processor_id()) | |
522 | return; | |
523 | ||
524 | /* NEED_RESCHED must be visible before we test polling */ | |
525 | smp_mb(); | |
526 | if (!tsk_is_polling(p)) | |
527 | smp_send_reschedule(cpu); | |
528 | } | |
529 | ||
029632fb | 530 | void resched_cpu(int cpu) |
c24d20db IM |
531 | { |
532 | struct rq *rq = cpu_rq(cpu); | |
533 | unsigned long flags; | |
534 | ||
05fa785c | 535 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
536 | return; |
537 | resched_task(cpu_curr(cpu)); | |
05fa785c | 538 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 539 | } |
06d8308c TG |
540 | |
541 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
542 | /* |
543 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
544 | * from an idle cpu. This is good for power-savings. | |
545 | * | |
546 | * We don't do similar optimization for completely idle system, as | |
547 | * selecting an idle cpu will add more delays to the timers than intended | |
548 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
549 | */ | |
550 | int get_nohz_timer_target(void) | |
551 | { | |
552 | int cpu = smp_processor_id(); | |
553 | int i; | |
554 | struct sched_domain *sd; | |
555 | ||
057f3fad | 556 | rcu_read_lock(); |
83cd4fe2 | 557 | for_each_domain(cpu, sd) { |
057f3fad PZ |
558 | for_each_cpu(i, sched_domain_span(sd)) { |
559 | if (!idle_cpu(i)) { | |
560 | cpu = i; | |
561 | goto unlock; | |
562 | } | |
563 | } | |
83cd4fe2 | 564 | } |
057f3fad PZ |
565 | unlock: |
566 | rcu_read_unlock(); | |
83cd4fe2 VP |
567 | return cpu; |
568 | } | |
06d8308c TG |
569 | /* |
570 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
571 | * idle CPU then this timer might expire before the next timer event | |
572 | * which is scheduled to wake up that CPU. In case of a completely | |
573 | * idle system the next event might even be infinite time into the | |
574 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
575 | * leaves the inner idle loop so the newly added timer is taken into | |
576 | * account when the CPU goes back to idle and evaluates the timer | |
577 | * wheel for the next timer event. | |
578 | */ | |
579 | void wake_up_idle_cpu(int cpu) | |
580 | { | |
581 | struct rq *rq = cpu_rq(cpu); | |
582 | ||
583 | if (cpu == smp_processor_id()) | |
584 | return; | |
585 | ||
586 | /* | |
587 | * This is safe, as this function is called with the timer | |
588 | * wheel base lock of (cpu) held. When the CPU is on the way | |
589 | * to idle and has not yet set rq->curr to idle then it will | |
590 | * be serialized on the timer wheel base lock and take the new | |
591 | * timer into account automatically. | |
592 | */ | |
593 | if (rq->curr != rq->idle) | |
594 | return; | |
45bf76df | 595 | |
45bf76df | 596 | /* |
06d8308c TG |
597 | * We can set TIF_RESCHED on the idle task of the other CPU |
598 | * lockless. The worst case is that the other CPU runs the | |
599 | * idle task through an additional NOOP schedule() | |
45bf76df | 600 | */ |
5ed0cec0 | 601 | set_tsk_need_resched(rq->idle); |
45bf76df | 602 | |
06d8308c TG |
603 | /* NEED_RESCHED must be visible before we test polling */ |
604 | smp_mb(); | |
605 | if (!tsk_is_polling(rq->idle)) | |
606 | smp_send_reschedule(cpu); | |
45bf76df IM |
607 | } |
608 | ||
ca38062e | 609 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 610 | { |
1c792db7 SS |
611 | int cpu = smp_processor_id(); |
612 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
613 | } |
614 | ||
ca38062e | 615 | #else /* CONFIG_NO_HZ */ |
45bf76df | 616 | |
ca38062e | 617 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 618 | { |
ca38062e | 619 | return false; |
2069dd75 PZ |
620 | } |
621 | ||
6d6bc0ad | 622 | #endif /* CONFIG_NO_HZ */ |
d842de87 | 623 | |
029632fb | 624 | void sched_avg_update(struct rq *rq) |
18d95a28 | 625 | { |
e9e9250b PZ |
626 | s64 period = sched_avg_period(); |
627 | ||
628 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
629 | /* |
630 | * Inline assembly required to prevent the compiler | |
631 | * optimising this loop into a divmod call. | |
632 | * See __iter_div_u64_rem() for another example of this. | |
633 | */ | |
634 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
635 | rq->age_stamp += period; |
636 | rq->rt_avg /= 2; | |
637 | } | |
18d95a28 PZ |
638 | } |
639 | ||
6d6bc0ad | 640 | #else /* !CONFIG_SMP */ |
029632fb | 641 | void resched_task(struct task_struct *p) |
18d95a28 | 642 | { |
05fa785c | 643 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 644 | set_tsk_need_resched(p); |
18d95a28 | 645 | } |
6d6bc0ad | 646 | #endif /* CONFIG_SMP */ |
18d95a28 | 647 | |
a790de99 PT |
648 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
649 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 650 | /* |
8277434e PT |
651 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
652 | * node and @up when leaving it for the final time. | |
653 | * | |
654 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 655 | */ |
029632fb | 656 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 657 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
658 | { |
659 | struct task_group *parent, *child; | |
eb755805 | 660 | int ret; |
c09595f6 | 661 | |
8277434e PT |
662 | parent = from; |
663 | ||
c09595f6 | 664 | down: |
eb755805 PZ |
665 | ret = (*down)(parent, data); |
666 | if (ret) | |
8277434e | 667 | goto out; |
c09595f6 PZ |
668 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
669 | parent = child; | |
670 | goto down; | |
671 | ||
672 | up: | |
673 | continue; | |
674 | } | |
eb755805 | 675 | ret = (*up)(parent, data); |
8277434e PT |
676 | if (ret || parent == from) |
677 | goto out; | |
c09595f6 PZ |
678 | |
679 | child = parent; | |
680 | parent = parent->parent; | |
681 | if (parent) | |
682 | goto up; | |
8277434e | 683 | out: |
eb755805 | 684 | return ret; |
c09595f6 PZ |
685 | } |
686 | ||
029632fb | 687 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 688 | { |
e2b245f8 | 689 | return 0; |
eb755805 | 690 | } |
18d95a28 PZ |
691 | #endif |
692 | ||
029632fb | 693 | void update_cpu_load(struct rq *this_rq); |
9c217245 | 694 | |
45bf76df IM |
695 | static void set_load_weight(struct task_struct *p) |
696 | { | |
f05998d4 NR |
697 | int prio = p->static_prio - MAX_RT_PRIO; |
698 | struct load_weight *load = &p->se.load; | |
699 | ||
dd41f596 IM |
700 | /* |
701 | * SCHED_IDLE tasks get minimal weight: | |
702 | */ | |
703 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 704 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 705 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
706 | return; |
707 | } | |
71f8bd46 | 708 | |
c8b28116 | 709 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 710 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
711 | } |
712 | ||
371fd7e7 | 713 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 714 | { |
a64692a3 | 715 | update_rq_clock(rq); |
dd41f596 | 716 | sched_info_queued(p); |
371fd7e7 | 717 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
718 | } |
719 | ||
371fd7e7 | 720 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 721 | { |
a64692a3 | 722 | update_rq_clock(rq); |
46ac22ba | 723 | sched_info_dequeued(p); |
371fd7e7 | 724 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
725 | } |
726 | ||
029632fb | 727 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
728 | { |
729 | if (task_contributes_to_load(p)) | |
730 | rq->nr_uninterruptible--; | |
731 | ||
371fd7e7 | 732 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
733 | } |
734 | ||
029632fb | 735 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
736 | { |
737 | if (task_contributes_to_load(p)) | |
738 | rq->nr_uninterruptible++; | |
739 | ||
371fd7e7 | 740 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
741 | } |
742 | ||
b52bfee4 VP |
743 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
744 | ||
305e6835 VP |
745 | /* |
746 | * There are no locks covering percpu hardirq/softirq time. | |
747 | * They are only modified in account_system_vtime, on corresponding CPU | |
748 | * with interrupts disabled. So, writes are safe. | |
749 | * They are read and saved off onto struct rq in update_rq_clock(). | |
750 | * This may result in other CPU reading this CPU's irq time and can | |
751 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
752 | * or new value with a side effect of accounting a slice of irq time to wrong |
753 | * task when irq is in progress while we read rq->clock. That is a worthy | |
754 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 755 | */ |
b52bfee4 VP |
756 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
757 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
758 | ||
759 | static DEFINE_PER_CPU(u64, irq_start_time); | |
760 | static int sched_clock_irqtime; | |
761 | ||
762 | void enable_sched_clock_irqtime(void) | |
763 | { | |
764 | sched_clock_irqtime = 1; | |
765 | } | |
766 | ||
767 | void disable_sched_clock_irqtime(void) | |
768 | { | |
769 | sched_clock_irqtime = 0; | |
770 | } | |
771 | ||
8e92c201 PZ |
772 | #ifndef CONFIG_64BIT |
773 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
774 | ||
775 | static inline void irq_time_write_begin(void) | |
776 | { | |
777 | __this_cpu_inc(irq_time_seq.sequence); | |
778 | smp_wmb(); | |
779 | } | |
780 | ||
781 | static inline void irq_time_write_end(void) | |
782 | { | |
783 | smp_wmb(); | |
784 | __this_cpu_inc(irq_time_seq.sequence); | |
785 | } | |
786 | ||
787 | static inline u64 irq_time_read(int cpu) | |
788 | { | |
789 | u64 irq_time; | |
790 | unsigned seq; | |
791 | ||
792 | do { | |
793 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
794 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
795 | per_cpu(cpu_hardirq_time, cpu); | |
796 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
797 | ||
798 | return irq_time; | |
799 | } | |
800 | #else /* CONFIG_64BIT */ | |
801 | static inline void irq_time_write_begin(void) | |
802 | { | |
803 | } | |
804 | ||
805 | static inline void irq_time_write_end(void) | |
806 | { | |
807 | } | |
808 | ||
809 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 810 | { |
305e6835 VP |
811 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
812 | } | |
8e92c201 | 813 | #endif /* CONFIG_64BIT */ |
305e6835 | 814 | |
fe44d621 PZ |
815 | /* |
816 | * Called before incrementing preempt_count on {soft,}irq_enter | |
817 | * and before decrementing preempt_count on {soft,}irq_exit. | |
818 | */ | |
b52bfee4 VP |
819 | void account_system_vtime(struct task_struct *curr) |
820 | { | |
821 | unsigned long flags; | |
fe44d621 | 822 | s64 delta; |
b52bfee4 | 823 | int cpu; |
b52bfee4 VP |
824 | |
825 | if (!sched_clock_irqtime) | |
826 | return; | |
827 | ||
828 | local_irq_save(flags); | |
829 | ||
b52bfee4 | 830 | cpu = smp_processor_id(); |
fe44d621 PZ |
831 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
832 | __this_cpu_add(irq_start_time, delta); | |
833 | ||
8e92c201 | 834 | irq_time_write_begin(); |
b52bfee4 VP |
835 | /* |
836 | * We do not account for softirq time from ksoftirqd here. | |
837 | * We want to continue accounting softirq time to ksoftirqd thread | |
838 | * in that case, so as not to confuse scheduler with a special task | |
839 | * that do not consume any time, but still wants to run. | |
840 | */ | |
841 | if (hardirq_count()) | |
fe44d621 | 842 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 843 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 844 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 845 | |
8e92c201 | 846 | irq_time_write_end(); |
b52bfee4 VP |
847 | local_irq_restore(flags); |
848 | } | |
b7dadc38 | 849 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 850 | |
e6e6685a GC |
851 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
852 | ||
853 | #ifdef CONFIG_PARAVIRT | |
854 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 855 | { |
e6e6685a GC |
856 | if (unlikely(steal > NSEC_PER_SEC)) |
857 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 858 | |
e6e6685a GC |
859 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
860 | } | |
861 | #endif | |
862 | ||
fe44d621 | 863 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 864 | { |
095c0aa8 GC |
865 | /* |
866 | * In theory, the compile should just see 0 here, and optimize out the call | |
867 | * to sched_rt_avg_update. But I don't trust it... | |
868 | */ | |
869 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
870 | s64 steal = 0, irq_delta = 0; | |
871 | #endif | |
872 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 873 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
874 | |
875 | /* | |
876 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
877 | * this case when a previous update_rq_clock() happened inside a | |
878 | * {soft,}irq region. | |
879 | * | |
880 | * When this happens, we stop ->clock_task and only update the | |
881 | * prev_irq_time stamp to account for the part that fit, so that a next | |
882 | * update will consume the rest. This ensures ->clock_task is | |
883 | * monotonic. | |
884 | * | |
885 | * It does however cause some slight miss-attribution of {soft,}irq | |
886 | * time, a more accurate solution would be to update the irq_time using | |
887 | * the current rq->clock timestamp, except that would require using | |
888 | * atomic ops. | |
889 | */ | |
890 | if (irq_delta > delta) | |
891 | irq_delta = delta; | |
892 | ||
893 | rq->prev_irq_time += irq_delta; | |
894 | delta -= irq_delta; | |
095c0aa8 GC |
895 | #endif |
896 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
897 | if (static_branch((¶virt_steal_rq_enabled))) { | |
898 | u64 st; | |
899 | ||
900 | steal = paravirt_steal_clock(cpu_of(rq)); | |
901 | steal -= rq->prev_steal_time_rq; | |
902 | ||
903 | if (unlikely(steal > delta)) | |
904 | steal = delta; | |
905 | ||
906 | st = steal_ticks(steal); | |
907 | steal = st * TICK_NSEC; | |
908 | ||
909 | rq->prev_steal_time_rq += steal; | |
910 | ||
911 | delta -= steal; | |
912 | } | |
913 | #endif | |
914 | ||
fe44d621 PZ |
915 | rq->clock_task += delta; |
916 | ||
095c0aa8 GC |
917 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
918 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
919 | sched_rt_avg_update(rq, irq_delta + steal); | |
920 | #endif | |
aa483808 VP |
921 | } |
922 | ||
095c0aa8 | 923 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
924 | static int irqtime_account_hi_update(void) |
925 | { | |
3292beb3 | 926 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
927 | unsigned long flags; |
928 | u64 latest_ns; | |
929 | int ret = 0; | |
930 | ||
931 | local_irq_save(flags); | |
932 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
612ef28a | 933 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) |
abb74cef VP |
934 | ret = 1; |
935 | local_irq_restore(flags); | |
936 | return ret; | |
937 | } | |
938 | ||
939 | static int irqtime_account_si_update(void) | |
940 | { | |
3292beb3 | 941 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
942 | unsigned long flags; |
943 | u64 latest_ns; | |
944 | int ret = 0; | |
945 | ||
946 | local_irq_save(flags); | |
947 | latest_ns = this_cpu_read(cpu_softirq_time); | |
612ef28a | 948 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) |
abb74cef VP |
949 | ret = 1; |
950 | local_irq_restore(flags); | |
951 | return ret; | |
952 | } | |
953 | ||
fe44d621 | 954 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 955 | |
abb74cef VP |
956 | #define sched_clock_irqtime (0) |
957 | ||
095c0aa8 | 958 | #endif |
b52bfee4 | 959 | |
34f971f6 PZ |
960 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
961 | { | |
962 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
963 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
964 | ||
965 | if (stop) { | |
966 | /* | |
967 | * Make it appear like a SCHED_FIFO task, its something | |
968 | * userspace knows about and won't get confused about. | |
969 | * | |
970 | * Also, it will make PI more or less work without too | |
971 | * much confusion -- but then, stop work should not | |
972 | * rely on PI working anyway. | |
973 | */ | |
974 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
975 | ||
976 | stop->sched_class = &stop_sched_class; | |
977 | } | |
978 | ||
979 | cpu_rq(cpu)->stop = stop; | |
980 | ||
981 | if (old_stop) { | |
982 | /* | |
983 | * Reset it back to a normal scheduling class so that | |
984 | * it can die in pieces. | |
985 | */ | |
986 | old_stop->sched_class = &rt_sched_class; | |
987 | } | |
988 | } | |
989 | ||
14531189 | 990 | /* |
dd41f596 | 991 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 992 | */ |
14531189 IM |
993 | static inline int __normal_prio(struct task_struct *p) |
994 | { | |
dd41f596 | 995 | return p->static_prio; |
14531189 IM |
996 | } |
997 | ||
b29739f9 IM |
998 | /* |
999 | * Calculate the expected normal priority: i.e. priority | |
1000 | * without taking RT-inheritance into account. Might be | |
1001 | * boosted by interactivity modifiers. Changes upon fork, | |
1002 | * setprio syscalls, and whenever the interactivity | |
1003 | * estimator recalculates. | |
1004 | */ | |
36c8b586 | 1005 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1006 | { |
1007 | int prio; | |
1008 | ||
e05606d3 | 1009 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1010 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1011 | else | |
1012 | prio = __normal_prio(p); | |
1013 | return prio; | |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * Calculate the current priority, i.e. the priority | |
1018 | * taken into account by the scheduler. This value might | |
1019 | * be boosted by RT tasks, or might be boosted by | |
1020 | * interactivity modifiers. Will be RT if the task got | |
1021 | * RT-boosted. If not then it returns p->normal_prio. | |
1022 | */ | |
36c8b586 | 1023 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1024 | { |
1025 | p->normal_prio = normal_prio(p); | |
1026 | /* | |
1027 | * If we are RT tasks or we were boosted to RT priority, | |
1028 | * keep the priority unchanged. Otherwise, update priority | |
1029 | * to the normal priority: | |
1030 | */ | |
1031 | if (!rt_prio(p->prio)) | |
1032 | return p->normal_prio; | |
1033 | return p->prio; | |
1034 | } | |
1035 | ||
1da177e4 LT |
1036 | /** |
1037 | * task_curr - is this task currently executing on a CPU? | |
1038 | * @p: the task in question. | |
1039 | */ | |
36c8b586 | 1040 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1041 | { |
1042 | return cpu_curr(task_cpu(p)) == p; | |
1043 | } | |
1044 | ||
cb469845 SR |
1045 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1046 | const struct sched_class *prev_class, | |
da7a735e | 1047 | int oldprio) |
cb469845 SR |
1048 | { |
1049 | if (prev_class != p->sched_class) { | |
1050 | if (prev_class->switched_from) | |
da7a735e PZ |
1051 | prev_class->switched_from(rq, p); |
1052 | p->sched_class->switched_to(rq, p); | |
1053 | } else if (oldprio != p->prio) | |
1054 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
1055 | } |
1056 | ||
029632fb | 1057 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1058 | { |
1059 | const struct sched_class *class; | |
1060 | ||
1061 | if (p->sched_class == rq->curr->sched_class) { | |
1062 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1063 | } else { | |
1064 | for_each_class(class) { | |
1065 | if (class == rq->curr->sched_class) | |
1066 | break; | |
1067 | if (class == p->sched_class) { | |
1068 | resched_task(rq->curr); | |
1069 | break; | |
1070 | } | |
1071 | } | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * A queue event has occurred, and we're going to schedule. In | |
1076 | * this case, we can save a useless back to back clock update. | |
1077 | */ | |
fd2f4419 | 1078 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1079 | rq->skip_clock_update = 1; |
1080 | } | |
1081 | ||
1da177e4 | 1082 | #ifdef CONFIG_SMP |
dd41f596 | 1083 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1084 | { |
e2912009 PZ |
1085 | #ifdef CONFIG_SCHED_DEBUG |
1086 | /* | |
1087 | * We should never call set_task_cpu() on a blocked task, | |
1088 | * ttwu() will sort out the placement. | |
1089 | */ | |
077614ee PZ |
1090 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
1091 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
1092 | |
1093 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1094 | /* |
1095 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1096 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1097 | * | |
1098 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
1099 | * see set_task_rq(). | |
1100 | * | |
1101 | * Furthermore, all task_rq users should acquire both locks, see | |
1102 | * task_rq_lock(). | |
1103 | */ | |
0122ec5b PZ |
1104 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1105 | lockdep_is_held(&task_rq(p)->lock))); | |
1106 | #endif | |
e2912009 PZ |
1107 | #endif |
1108 | ||
de1d7286 | 1109 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1110 | |
0c69774e PZ |
1111 | if (task_cpu(p) != new_cpu) { |
1112 | p->se.nr_migrations++; | |
a8b0ca17 | 1113 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1114 | } |
dd41f596 IM |
1115 | |
1116 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1117 | } |
1118 | ||
969c7921 | 1119 | struct migration_arg { |
36c8b586 | 1120 | struct task_struct *task; |
1da177e4 | 1121 | int dest_cpu; |
70b97a7f | 1122 | }; |
1da177e4 | 1123 | |
969c7921 TH |
1124 | static int migration_cpu_stop(void *data); |
1125 | ||
1da177e4 LT |
1126 | /* |
1127 | * wait_task_inactive - wait for a thread to unschedule. | |
1128 | * | |
85ba2d86 RM |
1129 | * If @match_state is nonzero, it's the @p->state value just checked and |
1130 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1131 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1132 | * we return a positive number (its total switch count). If a second call | |
1133 | * a short while later returns the same number, the caller can be sure that | |
1134 | * @p has remained unscheduled the whole time. | |
1135 | * | |
1da177e4 LT |
1136 | * The caller must ensure that the task *will* unschedule sometime soon, |
1137 | * else this function might spin for a *long* time. This function can't | |
1138 | * be called with interrupts off, or it may introduce deadlock with | |
1139 | * smp_call_function() if an IPI is sent by the same process we are | |
1140 | * waiting to become inactive. | |
1141 | */ | |
85ba2d86 | 1142 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1143 | { |
1144 | unsigned long flags; | |
dd41f596 | 1145 | int running, on_rq; |
85ba2d86 | 1146 | unsigned long ncsw; |
70b97a7f | 1147 | struct rq *rq; |
1da177e4 | 1148 | |
3a5c359a AK |
1149 | for (;;) { |
1150 | /* | |
1151 | * We do the initial early heuristics without holding | |
1152 | * any task-queue locks at all. We'll only try to get | |
1153 | * the runqueue lock when things look like they will | |
1154 | * work out! | |
1155 | */ | |
1156 | rq = task_rq(p); | |
fa490cfd | 1157 | |
3a5c359a AK |
1158 | /* |
1159 | * If the task is actively running on another CPU | |
1160 | * still, just relax and busy-wait without holding | |
1161 | * any locks. | |
1162 | * | |
1163 | * NOTE! Since we don't hold any locks, it's not | |
1164 | * even sure that "rq" stays as the right runqueue! | |
1165 | * But we don't care, since "task_running()" will | |
1166 | * return false if the runqueue has changed and p | |
1167 | * is actually now running somewhere else! | |
1168 | */ | |
85ba2d86 RM |
1169 | while (task_running(rq, p)) { |
1170 | if (match_state && unlikely(p->state != match_state)) | |
1171 | return 0; | |
3a5c359a | 1172 | cpu_relax(); |
85ba2d86 | 1173 | } |
fa490cfd | 1174 | |
3a5c359a AK |
1175 | /* |
1176 | * Ok, time to look more closely! We need the rq | |
1177 | * lock now, to be *sure*. If we're wrong, we'll | |
1178 | * just go back and repeat. | |
1179 | */ | |
1180 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1181 | trace_sched_wait_task(p); |
3a5c359a | 1182 | running = task_running(rq, p); |
fd2f4419 | 1183 | on_rq = p->on_rq; |
85ba2d86 | 1184 | ncsw = 0; |
f31e11d8 | 1185 | if (!match_state || p->state == match_state) |
93dcf55f | 1186 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1187 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1188 | |
85ba2d86 RM |
1189 | /* |
1190 | * If it changed from the expected state, bail out now. | |
1191 | */ | |
1192 | if (unlikely(!ncsw)) | |
1193 | break; | |
1194 | ||
3a5c359a AK |
1195 | /* |
1196 | * Was it really running after all now that we | |
1197 | * checked with the proper locks actually held? | |
1198 | * | |
1199 | * Oops. Go back and try again.. | |
1200 | */ | |
1201 | if (unlikely(running)) { | |
1202 | cpu_relax(); | |
1203 | continue; | |
1204 | } | |
fa490cfd | 1205 | |
3a5c359a AK |
1206 | /* |
1207 | * It's not enough that it's not actively running, | |
1208 | * it must be off the runqueue _entirely_, and not | |
1209 | * preempted! | |
1210 | * | |
80dd99b3 | 1211 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1212 | * running right now), it's preempted, and we should |
1213 | * yield - it could be a while. | |
1214 | */ | |
1215 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1216 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1217 | ||
1218 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1219 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1220 | continue; |
1221 | } | |
fa490cfd | 1222 | |
3a5c359a AK |
1223 | /* |
1224 | * Ahh, all good. It wasn't running, and it wasn't | |
1225 | * runnable, which means that it will never become | |
1226 | * running in the future either. We're all done! | |
1227 | */ | |
1228 | break; | |
1229 | } | |
85ba2d86 RM |
1230 | |
1231 | return ncsw; | |
1da177e4 LT |
1232 | } |
1233 | ||
1234 | /*** | |
1235 | * kick_process - kick a running thread to enter/exit the kernel | |
1236 | * @p: the to-be-kicked thread | |
1237 | * | |
1238 | * Cause a process which is running on another CPU to enter | |
1239 | * kernel-mode, without any delay. (to get signals handled.) | |
1240 | * | |
25985edc | 1241 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1242 | * because all it wants to ensure is that the remote task enters |
1243 | * the kernel. If the IPI races and the task has been migrated | |
1244 | * to another CPU then no harm is done and the purpose has been | |
1245 | * achieved as well. | |
1246 | */ | |
36c8b586 | 1247 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1248 | { |
1249 | int cpu; | |
1250 | ||
1251 | preempt_disable(); | |
1252 | cpu = task_cpu(p); | |
1253 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1254 | smp_send_reschedule(cpu); | |
1255 | preempt_enable(); | |
1256 | } | |
b43e3521 | 1257 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1258 | #endif /* CONFIG_SMP */ |
1da177e4 | 1259 | |
970b13ba | 1260 | #ifdef CONFIG_SMP |
30da688e | 1261 | /* |
013fdb80 | 1262 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1263 | */ |
5da9a0fb PZ |
1264 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1265 | { | |
1266 | int dest_cpu; | |
1267 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
1268 | ||
1269 | /* Look for allowed, online CPU in same node. */ | |
1270 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
fa17b507 | 1271 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb PZ |
1272 | return dest_cpu; |
1273 | ||
1274 | /* Any allowed, online CPU? */ | |
fa17b507 | 1275 | dest_cpu = cpumask_any_and(tsk_cpus_allowed(p), cpu_active_mask); |
5da9a0fb PZ |
1276 | if (dest_cpu < nr_cpu_ids) |
1277 | return dest_cpu; | |
1278 | ||
1279 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
1280 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
1281 | /* | |
1282 | * Don't tell them about moving exiting tasks or | |
1283 | * kernel threads (both mm NULL), since they never | |
1284 | * leave kernel. | |
1285 | */ | |
1286 | if (p->mm && printk_ratelimit()) { | |
1287 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
1288 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
1289 | } |
1290 | ||
1291 | return dest_cpu; | |
1292 | } | |
1293 | ||
e2912009 | 1294 | /* |
013fdb80 | 1295 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1296 | */ |
970b13ba | 1297 | static inline |
7608dec2 | 1298 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1299 | { |
7608dec2 | 1300 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1301 | |
1302 | /* | |
1303 | * In order not to call set_task_cpu() on a blocking task we need | |
1304 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1305 | * cpu. | |
1306 | * | |
1307 | * Since this is common to all placement strategies, this lives here. | |
1308 | * | |
1309 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1310 | * not worry about this generic constraint ] | |
1311 | */ | |
fa17b507 | 1312 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1313 | !cpu_online(cpu))) |
5da9a0fb | 1314 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1315 | |
1316 | return cpu; | |
970b13ba | 1317 | } |
09a40af5 MG |
1318 | |
1319 | static void update_avg(u64 *avg, u64 sample) | |
1320 | { | |
1321 | s64 diff = sample - *avg; | |
1322 | *avg += diff >> 3; | |
1323 | } | |
970b13ba PZ |
1324 | #endif |
1325 | ||
d7c01d27 | 1326 | static void |
b84cb5df | 1327 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1328 | { |
d7c01d27 | 1329 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1330 | struct rq *rq = this_rq(); |
1331 | ||
d7c01d27 PZ |
1332 | #ifdef CONFIG_SMP |
1333 | int this_cpu = smp_processor_id(); | |
1334 | ||
1335 | if (cpu == this_cpu) { | |
1336 | schedstat_inc(rq, ttwu_local); | |
1337 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1338 | } else { | |
1339 | struct sched_domain *sd; | |
1340 | ||
1341 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1342 | rcu_read_lock(); |
d7c01d27 PZ |
1343 | for_each_domain(this_cpu, sd) { |
1344 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1345 | schedstat_inc(sd, ttwu_wake_remote); | |
1346 | break; | |
1347 | } | |
1348 | } | |
057f3fad | 1349 | rcu_read_unlock(); |
d7c01d27 | 1350 | } |
f339b9dc PZ |
1351 | |
1352 | if (wake_flags & WF_MIGRATED) | |
1353 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1354 | ||
d7c01d27 PZ |
1355 | #endif /* CONFIG_SMP */ |
1356 | ||
1357 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1358 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1359 | |
1360 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1361 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1362 | |
d7c01d27 PZ |
1363 | #endif /* CONFIG_SCHEDSTATS */ |
1364 | } | |
1365 | ||
1366 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1367 | { | |
9ed3811a | 1368 | activate_task(rq, p, en_flags); |
fd2f4419 | 1369 | p->on_rq = 1; |
c2f7115e PZ |
1370 | |
1371 | /* if a worker is waking up, notify workqueue */ | |
1372 | if (p->flags & PF_WQ_WORKER) | |
1373 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1374 | } |
1375 | ||
23f41eeb PZ |
1376 | /* |
1377 | * Mark the task runnable and perform wakeup-preemption. | |
1378 | */ | |
89363381 | 1379 | static void |
23f41eeb | 1380 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1381 | { |
89363381 | 1382 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1383 | check_preempt_curr(rq, p, wake_flags); |
1384 | ||
1385 | p->state = TASK_RUNNING; | |
1386 | #ifdef CONFIG_SMP | |
1387 | if (p->sched_class->task_woken) | |
1388 | p->sched_class->task_woken(rq, p); | |
1389 | ||
e69c6341 | 1390 | if (rq->idle_stamp) { |
9ed3811a TH |
1391 | u64 delta = rq->clock - rq->idle_stamp; |
1392 | u64 max = 2*sysctl_sched_migration_cost; | |
1393 | ||
1394 | if (delta > max) | |
1395 | rq->avg_idle = max; | |
1396 | else | |
1397 | update_avg(&rq->avg_idle, delta); | |
1398 | rq->idle_stamp = 0; | |
1399 | } | |
1400 | #endif | |
1401 | } | |
1402 | ||
c05fbafb PZ |
1403 | static void |
1404 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1405 | { | |
1406 | #ifdef CONFIG_SMP | |
1407 | if (p->sched_contributes_to_load) | |
1408 | rq->nr_uninterruptible--; | |
1409 | #endif | |
1410 | ||
1411 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1412 | ttwu_do_wakeup(rq, p, wake_flags); | |
1413 | } | |
1414 | ||
1415 | /* | |
1416 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1417 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1418 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1419 | * the task is still ->on_rq. | |
1420 | */ | |
1421 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1422 | { | |
1423 | struct rq *rq; | |
1424 | int ret = 0; | |
1425 | ||
1426 | rq = __task_rq_lock(p); | |
1427 | if (p->on_rq) { | |
1428 | ttwu_do_wakeup(rq, p, wake_flags); | |
1429 | ret = 1; | |
1430 | } | |
1431 | __task_rq_unlock(rq); | |
1432 | ||
1433 | return ret; | |
1434 | } | |
1435 | ||
317f3941 | 1436 | #ifdef CONFIG_SMP |
fa14ff4a | 1437 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1438 | { |
1439 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1440 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1441 | struct task_struct *p; | |
317f3941 PZ |
1442 | |
1443 | raw_spin_lock(&rq->lock); | |
1444 | ||
fa14ff4a PZ |
1445 | while (llist) { |
1446 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1447 | llist = llist_next(llist); | |
317f3941 PZ |
1448 | ttwu_do_activate(rq, p, 0); |
1449 | } | |
1450 | ||
1451 | raw_spin_unlock(&rq->lock); | |
1452 | } | |
1453 | ||
1454 | void scheduler_ipi(void) | |
1455 | { | |
ca38062e | 1456 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1457 | return; |
1458 | ||
1459 | /* | |
1460 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1461 | * traditionally all their work was done from the interrupt return | |
1462 | * path. Now that we actually do some work, we need to make sure | |
1463 | * we do call them. | |
1464 | * | |
1465 | * Some archs already do call them, luckily irq_enter/exit nest | |
1466 | * properly. | |
1467 | * | |
1468 | * Arguably we should visit all archs and update all handlers, | |
1469 | * however a fair share of IPIs are still resched only so this would | |
1470 | * somewhat pessimize the simple resched case. | |
1471 | */ | |
1472 | irq_enter(); | |
fa14ff4a | 1473 | sched_ttwu_pending(); |
ca38062e SS |
1474 | |
1475 | /* | |
1476 | * Check if someone kicked us for doing the nohz idle load balance. | |
1477 | */ | |
6eb57e0d SS |
1478 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1479 | this_rq()->idle_balance = 1; | |
ca38062e | 1480 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1481 | } |
c5d753a5 | 1482 | irq_exit(); |
317f3941 PZ |
1483 | } |
1484 | ||
1485 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1486 | { | |
fa14ff4a | 1487 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1488 | smp_send_reschedule(cpu); |
1489 | } | |
d6aa8f85 PZ |
1490 | |
1491 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1492 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
1493 | { | |
1494 | struct rq *rq; | |
1495 | int ret = 0; | |
1496 | ||
1497 | rq = __task_rq_lock(p); | |
1498 | if (p->on_cpu) { | |
1499 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
1500 | ttwu_do_wakeup(rq, p, wake_flags); | |
1501 | ret = 1; | |
1502 | } | |
1503 | __task_rq_unlock(rq); | |
1504 | ||
1505 | return ret; | |
1506 | ||
1507 | } | |
1508 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
518cd623 | 1509 | |
39be3501 | 1510 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1511 | { |
1512 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1513 | } | |
d6aa8f85 | 1514 | #endif /* CONFIG_SMP */ |
317f3941 | 1515 | |
c05fbafb PZ |
1516 | static void ttwu_queue(struct task_struct *p, int cpu) |
1517 | { | |
1518 | struct rq *rq = cpu_rq(cpu); | |
1519 | ||
17d9f311 | 1520 | #if defined(CONFIG_SMP) |
39be3501 | 1521 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1522 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1523 | ttwu_queue_remote(p, cpu); |
1524 | return; | |
1525 | } | |
1526 | #endif | |
1527 | ||
c05fbafb PZ |
1528 | raw_spin_lock(&rq->lock); |
1529 | ttwu_do_activate(rq, p, 0); | |
1530 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1531 | } |
1532 | ||
1533 | /** | |
1da177e4 | 1534 | * try_to_wake_up - wake up a thread |
9ed3811a | 1535 | * @p: the thread to be awakened |
1da177e4 | 1536 | * @state: the mask of task states that can be woken |
9ed3811a | 1537 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1538 | * |
1539 | * Put it on the run-queue if it's not already there. The "current" | |
1540 | * thread is always on the run-queue (except when the actual | |
1541 | * re-schedule is in progress), and as such you're allowed to do | |
1542 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1543 | * runnable without the overhead of this. | |
1544 | * | |
9ed3811a TH |
1545 | * Returns %true if @p was woken up, %false if it was already running |
1546 | * or @state didn't match @p's state. | |
1da177e4 | 1547 | */ |
e4a52bcb PZ |
1548 | static int |
1549 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1550 | { |
1da177e4 | 1551 | unsigned long flags; |
c05fbafb | 1552 | int cpu, success = 0; |
2398f2c6 | 1553 | |
04e2f174 | 1554 | smp_wmb(); |
013fdb80 | 1555 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1556 | if (!(p->state & state)) |
1da177e4 LT |
1557 | goto out; |
1558 | ||
c05fbafb | 1559 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1560 | cpu = task_cpu(p); |
1da177e4 | 1561 | |
c05fbafb PZ |
1562 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1563 | goto stat; | |
1da177e4 | 1564 | |
1da177e4 | 1565 | #ifdef CONFIG_SMP |
e9c84311 | 1566 | /* |
c05fbafb PZ |
1567 | * If the owning (remote) cpu is still in the middle of schedule() with |
1568 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1569 | */ |
e4a52bcb PZ |
1570 | while (p->on_cpu) { |
1571 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1572 | /* | |
d6aa8f85 PZ |
1573 | * In case the architecture enables interrupts in |
1574 | * context_switch(), we cannot busy wait, since that | |
1575 | * would lead to deadlocks when an interrupt hits and | |
1576 | * tries to wake up @prev. So bail and do a complete | |
1577 | * remote wakeup. | |
e4a52bcb | 1578 | */ |
d6aa8f85 | 1579 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 1580 | goto stat; |
d6aa8f85 | 1581 | #else |
e4a52bcb | 1582 | cpu_relax(); |
d6aa8f85 | 1583 | #endif |
371fd7e7 | 1584 | } |
0970d299 | 1585 | /* |
e4a52bcb | 1586 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1587 | */ |
e4a52bcb | 1588 | smp_rmb(); |
1da177e4 | 1589 | |
a8e4f2ea | 1590 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1591 | p->state = TASK_WAKING; |
e7693a36 | 1592 | |
e4a52bcb | 1593 | if (p->sched_class->task_waking) |
74f8e4b2 | 1594 | p->sched_class->task_waking(p); |
efbbd05a | 1595 | |
7608dec2 | 1596 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1597 | if (task_cpu(p) != cpu) { |
1598 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1599 | set_task_cpu(p, cpu); |
f339b9dc | 1600 | } |
1da177e4 | 1601 | #endif /* CONFIG_SMP */ |
1da177e4 | 1602 | |
c05fbafb PZ |
1603 | ttwu_queue(p, cpu); |
1604 | stat: | |
b84cb5df | 1605 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1606 | out: |
013fdb80 | 1607 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1608 | |
1609 | return success; | |
1610 | } | |
1611 | ||
21aa9af0 TH |
1612 | /** |
1613 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1614 | * @p: the thread to be awakened | |
1615 | * | |
2acca55e | 1616 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1617 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1618 | * the current task. |
21aa9af0 TH |
1619 | */ |
1620 | static void try_to_wake_up_local(struct task_struct *p) | |
1621 | { | |
1622 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1623 | |
1624 | BUG_ON(rq != this_rq()); | |
1625 | BUG_ON(p == current); | |
1626 | lockdep_assert_held(&rq->lock); | |
1627 | ||
2acca55e PZ |
1628 | if (!raw_spin_trylock(&p->pi_lock)) { |
1629 | raw_spin_unlock(&rq->lock); | |
1630 | raw_spin_lock(&p->pi_lock); | |
1631 | raw_spin_lock(&rq->lock); | |
1632 | } | |
1633 | ||
21aa9af0 | 1634 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1635 | goto out; |
21aa9af0 | 1636 | |
fd2f4419 | 1637 | if (!p->on_rq) |
d7c01d27 PZ |
1638 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1639 | ||
23f41eeb | 1640 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1641 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1642 | out: |
1643 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1644 | } |
1645 | ||
50fa610a DH |
1646 | /** |
1647 | * wake_up_process - Wake up a specific process | |
1648 | * @p: The process to be woken up. | |
1649 | * | |
1650 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1651 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1652 | * running. | |
1653 | * | |
1654 | * It may be assumed that this function implies a write memory barrier before | |
1655 | * changing the task state if and only if any tasks are woken up. | |
1656 | */ | |
7ad5b3a5 | 1657 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1658 | { |
d9514f6c | 1659 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1660 | } |
1da177e4 LT |
1661 | EXPORT_SYMBOL(wake_up_process); |
1662 | ||
7ad5b3a5 | 1663 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1664 | { |
1665 | return try_to_wake_up(p, state, 0); | |
1666 | } | |
1667 | ||
1da177e4 LT |
1668 | /* |
1669 | * Perform scheduler related setup for a newly forked process p. | |
1670 | * p is forked by current. | |
dd41f596 IM |
1671 | * |
1672 | * __sched_fork() is basic setup used by init_idle() too: | |
1673 | */ | |
1674 | static void __sched_fork(struct task_struct *p) | |
1675 | { | |
fd2f4419 PZ |
1676 | p->on_rq = 0; |
1677 | ||
1678 | p->se.on_rq = 0; | |
dd41f596 IM |
1679 | p->se.exec_start = 0; |
1680 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1681 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1682 | p->se.nr_migrations = 0; |
da7a735e | 1683 | p->se.vruntime = 0; |
fd2f4419 | 1684 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1685 | |
1686 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1687 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1688 | #endif |
476d139c | 1689 | |
fa717060 | 1690 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1691 | |
e107be36 AK |
1692 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1693 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1694 | #endif | |
dd41f596 IM |
1695 | } |
1696 | ||
1697 | /* | |
1698 | * fork()/clone()-time setup: | |
1699 | */ | |
3e51e3ed | 1700 | void sched_fork(struct task_struct *p) |
dd41f596 | 1701 | { |
0122ec5b | 1702 | unsigned long flags; |
dd41f596 IM |
1703 | int cpu = get_cpu(); |
1704 | ||
1705 | __sched_fork(p); | |
06b83b5f | 1706 | /* |
0017d735 | 1707 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1708 | * nobody will actually run it, and a signal or other external |
1709 | * event cannot wake it up and insert it on the runqueue either. | |
1710 | */ | |
0017d735 | 1711 | p->state = TASK_RUNNING; |
dd41f596 | 1712 | |
c350a04e MG |
1713 | /* |
1714 | * Make sure we do not leak PI boosting priority to the child. | |
1715 | */ | |
1716 | p->prio = current->normal_prio; | |
1717 | ||
b9dc29e7 MG |
1718 | /* |
1719 | * Revert to default priority/policy on fork if requested. | |
1720 | */ | |
1721 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1722 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1723 | p->policy = SCHED_NORMAL; |
6c697bdf | 1724 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1725 | p->rt_priority = 0; |
1726 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1727 | p->static_prio = NICE_TO_PRIO(0); | |
1728 | ||
1729 | p->prio = p->normal_prio = __normal_prio(p); | |
1730 | set_load_weight(p); | |
6c697bdf | 1731 | |
b9dc29e7 MG |
1732 | /* |
1733 | * We don't need the reset flag anymore after the fork. It has | |
1734 | * fulfilled its duty: | |
1735 | */ | |
1736 | p->sched_reset_on_fork = 0; | |
1737 | } | |
ca94c442 | 1738 | |
2ddbf952 HS |
1739 | if (!rt_prio(p->prio)) |
1740 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1741 | |
cd29fe6f PZ |
1742 | if (p->sched_class->task_fork) |
1743 | p->sched_class->task_fork(p); | |
1744 | ||
86951599 PZ |
1745 | /* |
1746 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1747 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1748 | * is ran before sched_fork(). | |
1749 | * | |
1750 | * Silence PROVE_RCU. | |
1751 | */ | |
0122ec5b | 1752 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1753 | set_task_cpu(p, cpu); |
0122ec5b | 1754 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1755 | |
52f17b6c | 1756 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1757 | if (likely(sched_info_on())) |
52f17b6c | 1758 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1759 | #endif |
3ca7a440 PZ |
1760 | #if defined(CONFIG_SMP) |
1761 | p->on_cpu = 0; | |
4866cde0 | 1762 | #endif |
bdd4e85d | 1763 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1764 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1765 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1766 | #endif |
806c09a7 | 1767 | #ifdef CONFIG_SMP |
917b627d | 1768 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1769 | #endif |
917b627d | 1770 | |
476d139c | 1771 | put_cpu(); |
1da177e4 LT |
1772 | } |
1773 | ||
1774 | /* | |
1775 | * wake_up_new_task - wake up a newly created task for the first time. | |
1776 | * | |
1777 | * This function will do some initial scheduler statistics housekeeping | |
1778 | * that must be done for every newly created context, then puts the task | |
1779 | * on the runqueue and wakes it. | |
1780 | */ | |
3e51e3ed | 1781 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1782 | { |
1783 | unsigned long flags; | |
dd41f596 | 1784 | struct rq *rq; |
fabf318e | 1785 | |
ab2515c4 | 1786 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1787 | #ifdef CONFIG_SMP |
1788 | /* | |
1789 | * Fork balancing, do it here and not earlier because: | |
1790 | * - cpus_allowed can change in the fork path | |
1791 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1792 | */ |
ab2515c4 | 1793 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1794 | #endif |
1795 | ||
ab2515c4 | 1796 | rq = __task_rq_lock(p); |
cd29fe6f | 1797 | activate_task(rq, p, 0); |
fd2f4419 | 1798 | p->on_rq = 1; |
89363381 | 1799 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1800 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1801 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1802 | if (p->sched_class->task_woken) |
1803 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1804 | #endif |
0122ec5b | 1805 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1806 | } |
1807 | ||
e107be36 AK |
1808 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1809 | ||
1810 | /** | |
80dd99b3 | 1811 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1812 | * @notifier: notifier struct to register |
e107be36 AK |
1813 | */ |
1814 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1815 | { | |
1816 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1817 | } | |
1818 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1819 | ||
1820 | /** | |
1821 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1822 | * @notifier: notifier struct to unregister |
e107be36 AK |
1823 | * |
1824 | * This is safe to call from within a preemption notifier. | |
1825 | */ | |
1826 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1827 | { | |
1828 | hlist_del(¬ifier->link); | |
1829 | } | |
1830 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1831 | ||
1832 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1833 | { | |
1834 | struct preempt_notifier *notifier; | |
1835 | struct hlist_node *node; | |
1836 | ||
1837 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1838 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1839 | } | |
1840 | ||
1841 | static void | |
1842 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1843 | struct task_struct *next) | |
1844 | { | |
1845 | struct preempt_notifier *notifier; | |
1846 | struct hlist_node *node; | |
1847 | ||
1848 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1849 | notifier->ops->sched_out(notifier, next); | |
1850 | } | |
1851 | ||
6d6bc0ad | 1852 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1853 | |
1854 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1855 | { | |
1856 | } | |
1857 | ||
1858 | static void | |
1859 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1860 | struct task_struct *next) | |
1861 | { | |
1862 | } | |
1863 | ||
6d6bc0ad | 1864 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1865 | |
4866cde0 NP |
1866 | /** |
1867 | * prepare_task_switch - prepare to switch tasks | |
1868 | * @rq: the runqueue preparing to switch | |
421cee29 | 1869 | * @prev: the current task that is being switched out |
4866cde0 NP |
1870 | * @next: the task we are going to switch to. |
1871 | * | |
1872 | * This is called with the rq lock held and interrupts off. It must | |
1873 | * be paired with a subsequent finish_task_switch after the context | |
1874 | * switch. | |
1875 | * | |
1876 | * prepare_task_switch sets up locking and calls architecture specific | |
1877 | * hooks. | |
1878 | */ | |
e107be36 AK |
1879 | static inline void |
1880 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1881 | struct task_struct *next) | |
4866cde0 | 1882 | { |
fe4b04fa PZ |
1883 | sched_info_switch(prev, next); |
1884 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1885 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1886 | prepare_lock_switch(rq, next); |
1887 | prepare_arch_switch(next); | |
fe4b04fa | 1888 | trace_sched_switch(prev, next); |
4866cde0 NP |
1889 | } |
1890 | ||
1da177e4 LT |
1891 | /** |
1892 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1893 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1894 | * @prev: the thread we just switched away from. |
1895 | * | |
4866cde0 NP |
1896 | * finish_task_switch must be called after the context switch, paired |
1897 | * with a prepare_task_switch call before the context switch. | |
1898 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1899 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1900 | * |
1901 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1902 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1903 | * with the lock held can cause deadlocks; see schedule() for |
1904 | * details.) | |
1905 | */ | |
a9957449 | 1906 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1907 | __releases(rq->lock) |
1908 | { | |
1da177e4 | 1909 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1910 | long prev_state; |
1da177e4 LT |
1911 | |
1912 | rq->prev_mm = NULL; | |
1913 | ||
1914 | /* | |
1915 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1916 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1917 | * schedule one last time. The schedule call will never return, and |
1918 | * the scheduled task must drop that reference. | |
c394cc9f | 1919 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1920 | * still held, otherwise prev could be scheduled on another cpu, die |
1921 | * there before we look at prev->state, and then the reference would | |
1922 | * be dropped twice. | |
1923 | * Manfred Spraul <manfred@colorfullife.com> | |
1924 | */ | |
55a101f8 | 1925 | prev_state = prev->state; |
4866cde0 | 1926 | finish_arch_switch(prev); |
8381f65d JI |
1927 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1928 | local_irq_disable(); | |
1929 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 1930 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
1931 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1932 | local_irq_enable(); | |
1933 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 1934 | finish_lock_switch(rq, prev); |
e8fa1362 | 1935 | |
e107be36 | 1936 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1937 | if (mm) |
1938 | mmdrop(mm); | |
c394cc9f | 1939 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1940 | /* |
1941 | * Remove function-return probe instances associated with this | |
1942 | * task and put them back on the free list. | |
9761eea8 | 1943 | */ |
c6fd91f0 | 1944 | kprobe_flush_task(prev); |
1da177e4 | 1945 | put_task_struct(prev); |
c6fd91f0 | 1946 | } |
1da177e4 LT |
1947 | } |
1948 | ||
3f029d3c GH |
1949 | #ifdef CONFIG_SMP |
1950 | ||
1951 | /* assumes rq->lock is held */ | |
1952 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1953 | { | |
1954 | if (prev->sched_class->pre_schedule) | |
1955 | prev->sched_class->pre_schedule(rq, prev); | |
1956 | } | |
1957 | ||
1958 | /* rq->lock is NOT held, but preemption is disabled */ | |
1959 | static inline void post_schedule(struct rq *rq) | |
1960 | { | |
1961 | if (rq->post_schedule) { | |
1962 | unsigned long flags; | |
1963 | ||
05fa785c | 1964 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1965 | if (rq->curr->sched_class->post_schedule) |
1966 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1967 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1968 | |
1969 | rq->post_schedule = 0; | |
1970 | } | |
1971 | } | |
1972 | ||
1973 | #else | |
da19ab51 | 1974 | |
3f029d3c GH |
1975 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1976 | { | |
1977 | } | |
1978 | ||
1979 | static inline void post_schedule(struct rq *rq) | |
1980 | { | |
1da177e4 LT |
1981 | } |
1982 | ||
3f029d3c GH |
1983 | #endif |
1984 | ||
1da177e4 LT |
1985 | /** |
1986 | * schedule_tail - first thing a freshly forked thread must call. | |
1987 | * @prev: the thread we just switched away from. | |
1988 | */ | |
36c8b586 | 1989 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1990 | __releases(rq->lock) |
1991 | { | |
70b97a7f IM |
1992 | struct rq *rq = this_rq(); |
1993 | ||
4866cde0 | 1994 | finish_task_switch(rq, prev); |
da19ab51 | 1995 | |
3f029d3c GH |
1996 | /* |
1997 | * FIXME: do we need to worry about rq being invalidated by the | |
1998 | * task_switch? | |
1999 | */ | |
2000 | post_schedule(rq); | |
70b97a7f | 2001 | |
4866cde0 NP |
2002 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2003 | /* In this case, finish_task_switch does not reenable preemption */ | |
2004 | preempt_enable(); | |
2005 | #endif | |
1da177e4 | 2006 | if (current->set_child_tid) |
b488893a | 2007 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2008 | } |
2009 | ||
2010 | /* | |
2011 | * context_switch - switch to the new MM and the new | |
2012 | * thread's register state. | |
2013 | */ | |
dd41f596 | 2014 | static inline void |
70b97a7f | 2015 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2016 | struct task_struct *next) |
1da177e4 | 2017 | { |
dd41f596 | 2018 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2019 | |
e107be36 | 2020 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2021 | |
dd41f596 IM |
2022 | mm = next->mm; |
2023 | oldmm = prev->active_mm; | |
9226d125 ZA |
2024 | /* |
2025 | * For paravirt, this is coupled with an exit in switch_to to | |
2026 | * combine the page table reload and the switch backend into | |
2027 | * one hypercall. | |
2028 | */ | |
224101ed | 2029 | arch_start_context_switch(prev); |
9226d125 | 2030 | |
31915ab4 | 2031 | if (!mm) { |
1da177e4 LT |
2032 | next->active_mm = oldmm; |
2033 | atomic_inc(&oldmm->mm_count); | |
2034 | enter_lazy_tlb(oldmm, next); | |
2035 | } else | |
2036 | switch_mm(oldmm, mm, next); | |
2037 | ||
31915ab4 | 2038 | if (!prev->mm) { |
1da177e4 | 2039 | prev->active_mm = NULL; |
1da177e4 LT |
2040 | rq->prev_mm = oldmm; |
2041 | } | |
3a5f5e48 IM |
2042 | /* |
2043 | * Since the runqueue lock will be released by the next | |
2044 | * task (which is an invalid locking op but in the case | |
2045 | * of the scheduler it's an obvious special-case), so we | |
2046 | * do an early lockdep release here: | |
2047 | */ | |
2048 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2049 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2050 | #endif |
1da177e4 LT |
2051 | |
2052 | /* Here we just switch the register state and the stack. */ | |
2053 | switch_to(prev, next, prev); | |
2054 | ||
dd41f596 IM |
2055 | barrier(); |
2056 | /* | |
2057 | * this_rq must be evaluated again because prev may have moved | |
2058 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2059 | * frame will be invalid. | |
2060 | */ | |
2061 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2062 | } |
2063 | ||
2064 | /* | |
2065 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2066 | * | |
2067 | * externally visible scheduler statistics: current number of runnable | |
2068 | * threads, current number of uninterruptible-sleeping threads, total | |
2069 | * number of context switches performed since bootup. | |
2070 | */ | |
2071 | unsigned long nr_running(void) | |
2072 | { | |
2073 | unsigned long i, sum = 0; | |
2074 | ||
2075 | for_each_online_cpu(i) | |
2076 | sum += cpu_rq(i)->nr_running; | |
2077 | ||
2078 | return sum; | |
f711f609 | 2079 | } |
1da177e4 LT |
2080 | |
2081 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2082 | { |
1da177e4 | 2083 | unsigned long i, sum = 0; |
f711f609 | 2084 | |
0a945022 | 2085 | for_each_possible_cpu(i) |
1da177e4 | 2086 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2087 | |
2088 | /* | |
1da177e4 LT |
2089 | * Since we read the counters lockless, it might be slightly |
2090 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2091 | */ |
1da177e4 LT |
2092 | if (unlikely((long)sum < 0)) |
2093 | sum = 0; | |
f711f609 | 2094 | |
1da177e4 | 2095 | return sum; |
f711f609 | 2096 | } |
f711f609 | 2097 | |
1da177e4 | 2098 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2099 | { |
cc94abfc SR |
2100 | int i; |
2101 | unsigned long long sum = 0; | |
46cb4b7c | 2102 | |
0a945022 | 2103 | for_each_possible_cpu(i) |
1da177e4 | 2104 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2105 | |
1da177e4 LT |
2106 | return sum; |
2107 | } | |
483b4ee6 | 2108 | |
1da177e4 LT |
2109 | unsigned long nr_iowait(void) |
2110 | { | |
2111 | unsigned long i, sum = 0; | |
483b4ee6 | 2112 | |
0a945022 | 2113 | for_each_possible_cpu(i) |
1da177e4 | 2114 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2115 | |
1da177e4 LT |
2116 | return sum; |
2117 | } | |
483b4ee6 | 2118 | |
8c215bd3 | 2119 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2120 | { |
8c215bd3 | 2121 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2122 | return atomic_read(&this->nr_iowait); |
2123 | } | |
46cb4b7c | 2124 | |
69d25870 AV |
2125 | unsigned long this_cpu_load(void) |
2126 | { | |
2127 | struct rq *this = this_rq(); | |
2128 | return this->cpu_load[0]; | |
2129 | } | |
e790fb0b | 2130 | |
46cb4b7c | 2131 | |
dce48a84 TG |
2132 | /* Variables and functions for calc_load */ |
2133 | static atomic_long_t calc_load_tasks; | |
2134 | static unsigned long calc_load_update; | |
2135 | unsigned long avenrun[3]; | |
2136 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2137 | |
74f5187a PZ |
2138 | static long calc_load_fold_active(struct rq *this_rq) |
2139 | { | |
2140 | long nr_active, delta = 0; | |
2141 | ||
2142 | nr_active = this_rq->nr_running; | |
2143 | nr_active += (long) this_rq->nr_uninterruptible; | |
2144 | ||
2145 | if (nr_active != this_rq->calc_load_active) { | |
2146 | delta = nr_active - this_rq->calc_load_active; | |
2147 | this_rq->calc_load_active = nr_active; | |
2148 | } | |
2149 | ||
2150 | return delta; | |
2151 | } | |
2152 | ||
0f004f5a PZ |
2153 | static unsigned long |
2154 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2155 | { | |
2156 | load *= exp; | |
2157 | load += active * (FIXED_1 - exp); | |
2158 | load += 1UL << (FSHIFT - 1); | |
2159 | return load >> FSHIFT; | |
2160 | } | |
2161 | ||
74f5187a PZ |
2162 | #ifdef CONFIG_NO_HZ |
2163 | /* | |
2164 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2165 | * | |
2166 | * When making the ILB scale, we should try to pull this in as well. | |
2167 | */ | |
2168 | static atomic_long_t calc_load_tasks_idle; | |
2169 | ||
029632fb | 2170 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2171 | { |
2172 | long delta; | |
2173 | ||
2174 | delta = calc_load_fold_active(this_rq); | |
2175 | if (delta) | |
2176 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2177 | } | |
2178 | ||
2179 | static long calc_load_fold_idle(void) | |
2180 | { | |
2181 | long delta = 0; | |
2182 | ||
2183 | /* | |
2184 | * Its got a race, we don't care... | |
2185 | */ | |
2186 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2187 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2188 | ||
2189 | return delta; | |
2190 | } | |
0f004f5a PZ |
2191 | |
2192 | /** | |
2193 | * fixed_power_int - compute: x^n, in O(log n) time | |
2194 | * | |
2195 | * @x: base of the power | |
2196 | * @frac_bits: fractional bits of @x | |
2197 | * @n: power to raise @x to. | |
2198 | * | |
2199 | * By exploiting the relation between the definition of the natural power | |
2200 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2201 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2202 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2203 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2204 | * of course trivially computable in O(log_2 n), the length of our binary | |
2205 | * vector. | |
2206 | */ | |
2207 | static unsigned long | |
2208 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2209 | { | |
2210 | unsigned long result = 1UL << frac_bits; | |
2211 | ||
2212 | if (n) for (;;) { | |
2213 | if (n & 1) { | |
2214 | result *= x; | |
2215 | result += 1UL << (frac_bits - 1); | |
2216 | result >>= frac_bits; | |
2217 | } | |
2218 | n >>= 1; | |
2219 | if (!n) | |
2220 | break; | |
2221 | x *= x; | |
2222 | x += 1UL << (frac_bits - 1); | |
2223 | x >>= frac_bits; | |
2224 | } | |
2225 | ||
2226 | return result; | |
2227 | } | |
2228 | ||
2229 | /* | |
2230 | * a1 = a0 * e + a * (1 - e) | |
2231 | * | |
2232 | * a2 = a1 * e + a * (1 - e) | |
2233 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2234 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2235 | * | |
2236 | * a3 = a2 * e + a * (1 - e) | |
2237 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2238 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2239 | * | |
2240 | * ... | |
2241 | * | |
2242 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2243 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2244 | * = a0 * e^n + a * (1 - e^n) | |
2245 | * | |
2246 | * [1] application of the geometric series: | |
2247 | * | |
2248 | * n 1 - x^(n+1) | |
2249 | * S_n := \Sum x^i = ------------- | |
2250 | * i=0 1 - x | |
2251 | */ | |
2252 | static unsigned long | |
2253 | calc_load_n(unsigned long load, unsigned long exp, | |
2254 | unsigned long active, unsigned int n) | |
2255 | { | |
2256 | ||
2257 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2262 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2263 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2264 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2265 | * | |
2266 | * Once we've updated the global active value, we need to apply the exponential | |
2267 | * weights adjusted to the number of cycles missed. | |
2268 | */ | |
2269 | static void calc_global_nohz(unsigned long ticks) | |
2270 | { | |
2271 | long delta, active, n; | |
2272 | ||
2273 | if (time_before(jiffies, calc_load_update)) | |
2274 | return; | |
2275 | ||
2276 | /* | |
2277 | * If we crossed a calc_load_update boundary, make sure to fold | |
2278 | * any pending idle changes, the respective CPUs might have | |
2279 | * missed the tick driven calc_load_account_active() update | |
2280 | * due to NO_HZ. | |
2281 | */ | |
2282 | delta = calc_load_fold_idle(); | |
2283 | if (delta) | |
2284 | atomic_long_add(delta, &calc_load_tasks); | |
2285 | ||
2286 | /* | |
2287 | * If we were idle for multiple load cycles, apply them. | |
2288 | */ | |
2289 | if (ticks >= LOAD_FREQ) { | |
2290 | n = ticks / LOAD_FREQ; | |
2291 | ||
2292 | active = atomic_long_read(&calc_load_tasks); | |
2293 | active = active > 0 ? active * FIXED_1 : 0; | |
2294 | ||
2295 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
2296 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2297 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
2298 | ||
2299 | calc_load_update += n * LOAD_FREQ; | |
2300 | } | |
2301 | ||
2302 | /* | |
2303 | * Its possible the remainder of the above division also crosses | |
2304 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
2305 | * which comes after this will take care of that. | |
2306 | * | |
2307 | * Consider us being 11 ticks before a cycle completion, and us | |
2308 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
2309 | * age us 4 cycles, and the test in calc_global_load() will | |
2310 | * pick up the final one. | |
2311 | */ | |
2312 | } | |
74f5187a | 2313 | #else |
029632fb | 2314 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2315 | { |
2316 | } | |
2317 | ||
2318 | static inline long calc_load_fold_idle(void) | |
2319 | { | |
2320 | return 0; | |
2321 | } | |
0f004f5a PZ |
2322 | |
2323 | static void calc_global_nohz(unsigned long ticks) | |
2324 | { | |
2325 | } | |
74f5187a PZ |
2326 | #endif |
2327 | ||
2d02494f TG |
2328 | /** |
2329 | * get_avenrun - get the load average array | |
2330 | * @loads: pointer to dest load array | |
2331 | * @offset: offset to add | |
2332 | * @shift: shift count to shift the result left | |
2333 | * | |
2334 | * These values are estimates at best, so no need for locking. | |
2335 | */ | |
2336 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2337 | { | |
2338 | loads[0] = (avenrun[0] + offset) << shift; | |
2339 | loads[1] = (avenrun[1] + offset) << shift; | |
2340 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2341 | } |
46cb4b7c | 2342 | |
46cb4b7c | 2343 | /* |
dce48a84 TG |
2344 | * calc_load - update the avenrun load estimates 10 ticks after the |
2345 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2346 | */ |
0f004f5a | 2347 | void calc_global_load(unsigned long ticks) |
7835b98b | 2348 | { |
dce48a84 | 2349 | long active; |
1da177e4 | 2350 | |
0f004f5a PZ |
2351 | calc_global_nohz(ticks); |
2352 | ||
2353 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 2354 | return; |
1da177e4 | 2355 | |
dce48a84 TG |
2356 | active = atomic_long_read(&calc_load_tasks); |
2357 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2358 | |
dce48a84 TG |
2359 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2360 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2361 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2362 | |
dce48a84 TG |
2363 | calc_load_update += LOAD_FREQ; |
2364 | } | |
1da177e4 | 2365 | |
dce48a84 | 2366 | /* |
74f5187a PZ |
2367 | * Called from update_cpu_load() to periodically update this CPU's |
2368 | * active count. | |
dce48a84 TG |
2369 | */ |
2370 | static void calc_load_account_active(struct rq *this_rq) | |
2371 | { | |
74f5187a | 2372 | long delta; |
08c183f3 | 2373 | |
74f5187a PZ |
2374 | if (time_before(jiffies, this_rq->calc_load_update)) |
2375 | return; | |
783609c6 | 2376 | |
74f5187a PZ |
2377 | delta = calc_load_fold_active(this_rq); |
2378 | delta += calc_load_fold_idle(); | |
2379 | if (delta) | |
dce48a84 | 2380 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2381 | |
2382 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2383 | } |
2384 | ||
fdf3e95d VP |
2385 | /* |
2386 | * The exact cpuload at various idx values, calculated at every tick would be | |
2387 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2388 | * | |
2389 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2390 | * on nth tick when cpu may be busy, then we have: | |
2391 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2392 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2393 | * | |
2394 | * decay_load_missed() below does efficient calculation of | |
2395 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2396 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2397 | * | |
2398 | * The calculation is approximated on a 128 point scale. | |
2399 | * degrade_zero_ticks is the number of ticks after which load at any | |
2400 | * particular idx is approximated to be zero. | |
2401 | * degrade_factor is a precomputed table, a row for each load idx. | |
2402 | * Each column corresponds to degradation factor for a power of two ticks, | |
2403 | * based on 128 point scale. | |
2404 | * Example: | |
2405 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2406 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2407 | * | |
2408 | * With this power of 2 load factors, we can degrade the load n times | |
2409 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2410 | * n mult/shifts needed by the exact degradation. | |
2411 | */ | |
2412 | #define DEGRADE_SHIFT 7 | |
2413 | static const unsigned char | |
2414 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2415 | static const unsigned char | |
2416 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2417 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2418 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2419 | {96, 72, 40, 12, 1, 0, 0}, | |
2420 | {112, 98, 75, 43, 15, 1, 0}, | |
2421 | {120, 112, 98, 76, 45, 16, 2} }; | |
2422 | ||
2423 | /* | |
2424 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2425 | * would be when CPU is idle and so we just decay the old load without | |
2426 | * adding any new load. | |
2427 | */ | |
2428 | static unsigned long | |
2429 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2430 | { | |
2431 | int j = 0; | |
2432 | ||
2433 | if (!missed_updates) | |
2434 | return load; | |
2435 | ||
2436 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2437 | return 0; | |
2438 | ||
2439 | if (idx == 1) | |
2440 | return load >> missed_updates; | |
2441 | ||
2442 | while (missed_updates) { | |
2443 | if (missed_updates % 2) | |
2444 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2445 | ||
2446 | missed_updates >>= 1; | |
2447 | j++; | |
2448 | } | |
2449 | return load; | |
2450 | } | |
2451 | ||
46cb4b7c | 2452 | /* |
dd41f596 | 2453 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2454 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2455 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2456 | */ |
029632fb | 2457 | void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 2458 | { |
495eca49 | 2459 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
2460 | unsigned long curr_jiffies = jiffies; |
2461 | unsigned long pending_updates; | |
dd41f596 | 2462 | int i, scale; |
46cb4b7c | 2463 | |
dd41f596 | 2464 | this_rq->nr_load_updates++; |
46cb4b7c | 2465 | |
fdf3e95d VP |
2466 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
2467 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2468 | return; | |
2469 | ||
2470 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2471 | this_rq->last_load_update_tick = curr_jiffies; | |
2472 | ||
dd41f596 | 2473 | /* Update our load: */ |
fdf3e95d VP |
2474 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2475 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2476 | unsigned long old_load, new_load; |
7d1e6a9b | 2477 | |
dd41f596 | 2478 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2479 | |
dd41f596 | 2480 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2481 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2482 | new_load = this_load; |
a25707f3 IM |
2483 | /* |
2484 | * Round up the averaging division if load is increasing. This | |
2485 | * prevents us from getting stuck on 9 if the load is 10, for | |
2486 | * example. | |
2487 | */ | |
2488 | if (new_load > old_load) | |
fdf3e95d VP |
2489 | new_load += scale - 1; |
2490 | ||
2491 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2492 | } |
da2b71ed SS |
2493 | |
2494 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2495 | } |
2496 | ||
2497 | static void update_cpu_load_active(struct rq *this_rq) | |
2498 | { | |
2499 | update_cpu_load(this_rq); | |
46cb4b7c | 2500 | |
74f5187a | 2501 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2502 | } |
2503 | ||
dd41f596 | 2504 | #ifdef CONFIG_SMP |
8a0be9ef | 2505 | |
46cb4b7c | 2506 | /* |
38022906 PZ |
2507 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2508 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2509 | */ |
38022906 | 2510 | void sched_exec(void) |
46cb4b7c | 2511 | { |
38022906 | 2512 | struct task_struct *p = current; |
1da177e4 | 2513 | unsigned long flags; |
0017d735 | 2514 | int dest_cpu; |
46cb4b7c | 2515 | |
8f42ced9 | 2516 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2517 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2518 | if (dest_cpu == smp_processor_id()) |
2519 | goto unlock; | |
38022906 | 2520 | |
8f42ced9 | 2521 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2522 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2523 | |
8f42ced9 PZ |
2524 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2525 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2526 | return; |
2527 | } | |
0017d735 | 2528 | unlock: |
8f42ced9 | 2529 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2530 | } |
dd41f596 | 2531 | |
1da177e4 LT |
2532 | #endif |
2533 | ||
1da177e4 | 2534 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2535 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2536 | |
2537 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2538 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2539 | |
2540 | /* | |
c5f8d995 | 2541 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2542 | * @p in case that task is currently running. |
c5f8d995 HS |
2543 | * |
2544 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2545 | */ |
c5f8d995 HS |
2546 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2547 | { | |
2548 | u64 ns = 0; | |
2549 | ||
2550 | if (task_current(rq, p)) { | |
2551 | update_rq_clock(rq); | |
305e6835 | 2552 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2553 | if ((s64)ns < 0) |
2554 | ns = 0; | |
2555 | } | |
2556 | ||
2557 | return ns; | |
2558 | } | |
2559 | ||
bb34d92f | 2560 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2561 | { |
1da177e4 | 2562 | unsigned long flags; |
41b86e9c | 2563 | struct rq *rq; |
bb34d92f | 2564 | u64 ns = 0; |
48f24c4d | 2565 | |
41b86e9c | 2566 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2567 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2568 | task_rq_unlock(rq, p, &flags); |
1508487e | 2569 | |
c5f8d995 HS |
2570 | return ns; |
2571 | } | |
f06febc9 | 2572 | |
c5f8d995 HS |
2573 | /* |
2574 | * Return accounted runtime for the task. | |
2575 | * In case the task is currently running, return the runtime plus current's | |
2576 | * pending runtime that have not been accounted yet. | |
2577 | */ | |
2578 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2579 | { | |
2580 | unsigned long flags; | |
2581 | struct rq *rq; | |
2582 | u64 ns = 0; | |
2583 | ||
2584 | rq = task_rq_lock(p, &flags); | |
2585 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2586 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2587 | |
2588 | return ns; | |
2589 | } | |
48f24c4d | 2590 | |
54c707e9 GC |
2591 | #ifdef CONFIG_CGROUP_CPUACCT |
2592 | struct cgroup_subsys cpuacct_subsys; | |
2593 | struct cpuacct root_cpuacct; | |
2594 | #endif | |
2595 | ||
be726ffd GC |
2596 | static inline void task_group_account_field(struct task_struct *p, int index, |
2597 | u64 tmp) | |
54c707e9 GC |
2598 | { |
2599 | #ifdef CONFIG_CGROUP_CPUACCT | |
2600 | struct kernel_cpustat *kcpustat; | |
2601 | struct cpuacct *ca; | |
2602 | #endif | |
2603 | /* | |
2604 | * Since all updates are sure to touch the root cgroup, we | |
2605 | * get ourselves ahead and touch it first. If the root cgroup | |
2606 | * is the only cgroup, then nothing else should be necessary. | |
2607 | * | |
2608 | */ | |
2609 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
2610 | ||
2611 | #ifdef CONFIG_CGROUP_CPUACCT | |
2612 | if (unlikely(!cpuacct_subsys.active)) | |
2613 | return; | |
2614 | ||
2615 | rcu_read_lock(); | |
2616 | ca = task_ca(p); | |
2617 | while (ca && (ca != &root_cpuacct)) { | |
2618 | kcpustat = this_cpu_ptr(ca->cpustat); | |
2619 | kcpustat->cpustat[index] += tmp; | |
2620 | ca = parent_ca(ca); | |
2621 | } | |
2622 | rcu_read_unlock(); | |
2623 | #endif | |
2624 | } | |
2625 | ||
2626 | ||
1da177e4 LT |
2627 | /* |
2628 | * Account user cpu time to a process. | |
2629 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 2630 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 2631 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 2632 | */ |
457533a7 MS |
2633 | void account_user_time(struct task_struct *p, cputime_t cputime, |
2634 | cputime_t cputime_scaled) | |
1da177e4 | 2635 | { |
3292beb3 | 2636 | int index; |
1da177e4 | 2637 | |
457533a7 | 2638 | /* Add user time to process. */ |
64861634 MS |
2639 | p->utime += cputime; |
2640 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2641 | account_group_user_time(p, cputime); |
1da177e4 | 2642 | |
3292beb3 | 2643 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
ef12fefa | 2644 | |
1da177e4 | 2645 | /* Add user time to cpustat. */ |
612ef28a | 2646 | task_group_account_field(p, index, (__force u64) cputime); |
ef12fefa | 2647 | |
49b5cf34 JL |
2648 | /* Account for user time used */ |
2649 | acct_update_integrals(p); | |
1da177e4 LT |
2650 | } |
2651 | ||
94886b84 LV |
2652 | /* |
2653 | * Account guest cpu time to a process. | |
2654 | * @p: the process that the cpu time gets accounted to | |
2655 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 2656 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 2657 | */ |
457533a7 MS |
2658 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
2659 | cputime_t cputime_scaled) | |
94886b84 | 2660 | { |
3292beb3 | 2661 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
94886b84 | 2662 | |
457533a7 | 2663 | /* Add guest time to process. */ |
64861634 MS |
2664 | p->utime += cputime; |
2665 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2666 | account_group_user_time(p, cputime); |
64861634 | 2667 | p->gtime += cputime; |
94886b84 | 2668 | |
457533a7 | 2669 | /* Add guest time to cpustat. */ |
ce0e7b28 | 2670 | if (TASK_NICE(p) > 0) { |
612ef28a MS |
2671 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
2672 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
ce0e7b28 | 2673 | } else { |
612ef28a MS |
2674 | cpustat[CPUTIME_USER] += (__force u64) cputime; |
2675 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
ce0e7b28 | 2676 | } |
94886b84 LV |
2677 | } |
2678 | ||
70a89a66 VP |
2679 | /* |
2680 | * Account system cpu time to a process and desired cpustat field | |
2681 | * @p: the process that the cpu time gets accounted to | |
2682 | * @cputime: the cpu time spent in kernel space since the last update | |
2683 | * @cputime_scaled: cputime scaled by cpu frequency | |
2684 | * @target_cputime64: pointer to cpustat field that has to be updated | |
2685 | */ | |
2686 | static inline | |
2687 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3292beb3 | 2688 | cputime_t cputime_scaled, int index) |
70a89a66 | 2689 | { |
70a89a66 | 2690 | /* Add system time to process. */ |
64861634 MS |
2691 | p->stime += cputime; |
2692 | p->stimescaled += cputime_scaled; | |
70a89a66 VP |
2693 | account_group_system_time(p, cputime); |
2694 | ||
2695 | /* Add system time to cpustat. */ | |
612ef28a | 2696 | task_group_account_field(p, index, (__force u64) cputime); |
70a89a66 VP |
2697 | |
2698 | /* Account for system time used */ | |
2699 | acct_update_integrals(p); | |
2700 | } | |
2701 | ||
1da177e4 LT |
2702 | /* |
2703 | * Account system cpu time to a process. | |
2704 | * @p: the process that the cpu time gets accounted to | |
2705 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
2706 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 2707 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
2708 | */ |
2709 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 2710 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 | 2711 | { |
3292beb3 | 2712 | int index; |
1da177e4 | 2713 | |
983ed7a6 | 2714 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 2715 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
2716 | return; |
2717 | } | |
94886b84 | 2718 | |
1da177e4 | 2719 | if (hardirq_count() - hardirq_offset) |
3292beb3 | 2720 | index = CPUTIME_IRQ; |
75e1056f | 2721 | else if (in_serving_softirq()) |
3292beb3 | 2722 | index = CPUTIME_SOFTIRQ; |
1da177e4 | 2723 | else |
3292beb3 | 2724 | index = CPUTIME_SYSTEM; |
ef12fefa | 2725 | |
3292beb3 | 2726 | __account_system_time(p, cputime, cputime_scaled, index); |
1da177e4 LT |
2727 | } |
2728 | ||
c66f08be | 2729 | /* |
1da177e4 | 2730 | * Account for involuntary wait time. |
544b4a1f | 2731 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 2732 | */ |
79741dd3 | 2733 | void account_steal_time(cputime_t cputime) |
c66f08be | 2734 | { |
3292beb3 | 2735 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
79741dd3 | 2736 | |
612ef28a | 2737 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; |
c66f08be MN |
2738 | } |
2739 | ||
1da177e4 | 2740 | /* |
79741dd3 MS |
2741 | * Account for idle time. |
2742 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 2743 | */ |
79741dd3 | 2744 | void account_idle_time(cputime_t cputime) |
1da177e4 | 2745 | { |
3292beb3 | 2746 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
70b97a7f | 2747 | struct rq *rq = this_rq(); |
1da177e4 | 2748 | |
79741dd3 | 2749 | if (atomic_read(&rq->nr_iowait) > 0) |
612ef28a | 2750 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; |
79741dd3 | 2751 | else |
612ef28a | 2752 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; |
1da177e4 LT |
2753 | } |
2754 | ||
e6e6685a GC |
2755 | static __always_inline bool steal_account_process_tick(void) |
2756 | { | |
2757 | #ifdef CONFIG_PARAVIRT | |
2758 | if (static_branch(¶virt_steal_enabled)) { | |
2759 | u64 steal, st = 0; | |
2760 | ||
2761 | steal = paravirt_steal_clock(smp_processor_id()); | |
2762 | steal -= this_rq()->prev_steal_time; | |
2763 | ||
2764 | st = steal_ticks(steal); | |
2765 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
2766 | ||
2767 | account_steal_time(st); | |
2768 | return st; | |
2769 | } | |
2770 | #endif | |
2771 | return false; | |
2772 | } | |
2773 | ||
79741dd3 MS |
2774 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
2775 | ||
abb74cef VP |
2776 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
2777 | /* | |
2778 | * Account a tick to a process and cpustat | |
2779 | * @p: the process that the cpu time gets accounted to | |
2780 | * @user_tick: is the tick from userspace | |
2781 | * @rq: the pointer to rq | |
2782 | * | |
2783 | * Tick demultiplexing follows the order | |
2784 | * - pending hardirq update | |
2785 | * - pending softirq update | |
2786 | * - user_time | |
2787 | * - idle_time | |
2788 | * - system time | |
2789 | * - check for guest_time | |
2790 | * - else account as system_time | |
2791 | * | |
2792 | * Check for hardirq is done both for system and user time as there is | |
2793 | * no timer going off while we are on hardirq and hence we may never get an | |
2794 | * opportunity to update it solely in system time. | |
2795 | * p->stime and friends are only updated on system time and not on irq | |
2796 | * softirq as those do not count in task exec_runtime any more. | |
2797 | */ | |
2798 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2799 | struct rq *rq) | |
2800 | { | |
2801 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3292beb3 | 2802 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef | 2803 | |
e6e6685a GC |
2804 | if (steal_account_process_tick()) |
2805 | return; | |
2806 | ||
abb74cef | 2807 | if (irqtime_account_hi_update()) { |
612ef28a | 2808 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; |
abb74cef | 2809 | } else if (irqtime_account_si_update()) { |
612ef28a | 2810 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; |
414bee9b VP |
2811 | } else if (this_cpu_ksoftirqd() == p) { |
2812 | /* | |
2813 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
2814 | * So, we have to handle it separately here. | |
2815 | * Also, p->stime needs to be updated for ksoftirqd. | |
2816 | */ | |
2817 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2818 | CPUTIME_SOFTIRQ); |
abb74cef VP |
2819 | } else if (user_tick) { |
2820 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2821 | } else if (p == rq->idle) { | |
2822 | account_idle_time(cputime_one_jiffy); | |
2823 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
2824 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2825 | } else { | |
2826 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2827 | CPUTIME_SYSTEM); |
abb74cef VP |
2828 | } |
2829 | } | |
2830 | ||
2831 | static void irqtime_account_idle_ticks(int ticks) | |
2832 | { | |
2833 | int i; | |
2834 | struct rq *rq = this_rq(); | |
2835 | ||
2836 | for (i = 0; i < ticks; i++) | |
2837 | irqtime_account_process_tick(current, 0, rq); | |
2838 | } | |
544b4a1f | 2839 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
2840 | static void irqtime_account_idle_ticks(int ticks) {} |
2841 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2842 | struct rq *rq) {} | |
544b4a1f | 2843 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
2844 | |
2845 | /* | |
2846 | * Account a single tick of cpu time. | |
2847 | * @p: the process that the cpu time gets accounted to | |
2848 | * @user_tick: indicates if the tick is a user or a system tick | |
2849 | */ | |
2850 | void account_process_tick(struct task_struct *p, int user_tick) | |
2851 | { | |
a42548a1 | 2852 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
2853 | struct rq *rq = this_rq(); |
2854 | ||
abb74cef VP |
2855 | if (sched_clock_irqtime) { |
2856 | irqtime_account_process_tick(p, user_tick, rq); | |
2857 | return; | |
2858 | } | |
2859 | ||
e6e6685a GC |
2860 | if (steal_account_process_tick()) |
2861 | return; | |
2862 | ||
79741dd3 | 2863 | if (user_tick) |
a42548a1 | 2864 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 2865 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 2866 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
2867 | one_jiffy_scaled); |
2868 | else | |
a42548a1 | 2869 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
2870 | } |
2871 | ||
2872 | /* | |
2873 | * Account multiple ticks of steal time. | |
2874 | * @p: the process from which the cpu time has been stolen | |
2875 | * @ticks: number of stolen ticks | |
2876 | */ | |
2877 | void account_steal_ticks(unsigned long ticks) | |
2878 | { | |
2879 | account_steal_time(jiffies_to_cputime(ticks)); | |
2880 | } | |
2881 | ||
2882 | /* | |
2883 | * Account multiple ticks of idle time. | |
2884 | * @ticks: number of stolen ticks | |
2885 | */ | |
2886 | void account_idle_ticks(unsigned long ticks) | |
2887 | { | |
abb74cef VP |
2888 | |
2889 | if (sched_clock_irqtime) { | |
2890 | irqtime_account_idle_ticks(ticks); | |
2891 | return; | |
2892 | } | |
2893 | ||
79741dd3 | 2894 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
2895 | } |
2896 | ||
79741dd3 MS |
2897 | #endif |
2898 | ||
49048622 BS |
2899 | /* |
2900 | * Use precise platform statistics if available: | |
2901 | */ | |
2902 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 2903 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2904 | { |
d99ca3b9 HS |
2905 | *ut = p->utime; |
2906 | *st = p->stime; | |
49048622 BS |
2907 | } |
2908 | ||
0cf55e1e | 2909 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2910 | { |
0cf55e1e HS |
2911 | struct task_cputime cputime; |
2912 | ||
2913 | thread_group_cputime(p, &cputime); | |
2914 | ||
2915 | *ut = cputime.utime; | |
2916 | *st = cputime.stime; | |
49048622 BS |
2917 | } |
2918 | #else | |
761b1d26 HS |
2919 | |
2920 | #ifndef nsecs_to_cputime | |
b7b20df9 | 2921 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
2922 | #endif |
2923 | ||
d180c5bc | 2924 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2925 | { |
64861634 | 2926 | cputime_t rtime, utime = p->utime, total = utime + p->stime; |
49048622 BS |
2927 | |
2928 | /* | |
2929 | * Use CFS's precise accounting: | |
2930 | */ | |
d180c5bc | 2931 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
2932 | |
2933 | if (total) { | |
64861634 | 2934 | u64 temp = (__force u64) rtime; |
d180c5bc | 2935 | |
64861634 MS |
2936 | temp *= (__force u64) utime; |
2937 | do_div(temp, (__force u32) total); | |
2938 | utime = (__force cputime_t) temp; | |
d180c5bc HS |
2939 | } else |
2940 | utime = rtime; | |
49048622 | 2941 | |
d180c5bc HS |
2942 | /* |
2943 | * Compare with previous values, to keep monotonicity: | |
2944 | */ | |
761b1d26 | 2945 | p->prev_utime = max(p->prev_utime, utime); |
64861634 | 2946 | p->prev_stime = max(p->prev_stime, rtime - p->prev_utime); |
49048622 | 2947 | |
d99ca3b9 HS |
2948 | *ut = p->prev_utime; |
2949 | *st = p->prev_stime; | |
49048622 BS |
2950 | } |
2951 | ||
0cf55e1e HS |
2952 | /* |
2953 | * Must be called with siglock held. | |
2954 | */ | |
2955 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 2956 | { |
0cf55e1e HS |
2957 | struct signal_struct *sig = p->signal; |
2958 | struct task_cputime cputime; | |
2959 | cputime_t rtime, utime, total; | |
49048622 | 2960 | |
0cf55e1e | 2961 | thread_group_cputime(p, &cputime); |
49048622 | 2962 | |
64861634 | 2963 | total = cputime.utime + cputime.stime; |
0cf55e1e | 2964 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); |
49048622 | 2965 | |
0cf55e1e | 2966 | if (total) { |
64861634 | 2967 | u64 temp = (__force u64) rtime; |
49048622 | 2968 | |
64861634 MS |
2969 | temp *= (__force u64) cputime.utime; |
2970 | do_div(temp, (__force u32) total); | |
2971 | utime = (__force cputime_t) temp; | |
0cf55e1e HS |
2972 | } else |
2973 | utime = rtime; | |
2974 | ||
2975 | sig->prev_utime = max(sig->prev_utime, utime); | |
64861634 | 2976 | sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime); |
0cf55e1e HS |
2977 | |
2978 | *ut = sig->prev_utime; | |
2979 | *st = sig->prev_stime; | |
49048622 | 2980 | } |
49048622 | 2981 | #endif |
49048622 | 2982 | |
7835b98b CL |
2983 | /* |
2984 | * This function gets called by the timer code, with HZ frequency. | |
2985 | * We call it with interrupts disabled. | |
7835b98b CL |
2986 | */ |
2987 | void scheduler_tick(void) | |
2988 | { | |
7835b98b CL |
2989 | int cpu = smp_processor_id(); |
2990 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2991 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2992 | |
2993 | sched_clock_tick(); | |
dd41f596 | 2994 | |
05fa785c | 2995 | raw_spin_lock(&rq->lock); |
3e51f33f | 2996 | update_rq_clock(rq); |
fdf3e95d | 2997 | update_cpu_load_active(rq); |
fa85ae24 | 2998 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2999 | raw_spin_unlock(&rq->lock); |
7835b98b | 3000 | |
e9d2b064 | 3001 | perf_event_task_tick(); |
e220d2dc | 3002 | |
e418e1c2 | 3003 | #ifdef CONFIG_SMP |
6eb57e0d | 3004 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 3005 | trigger_load_balance(rq, cpu); |
e418e1c2 | 3006 | #endif |
1da177e4 LT |
3007 | } |
3008 | ||
132380a0 | 3009 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3010 | { |
3011 | if (in_lock_functions(addr)) { | |
3012 | addr = CALLER_ADDR2; | |
3013 | if (in_lock_functions(addr)) | |
3014 | addr = CALLER_ADDR3; | |
3015 | } | |
3016 | return addr; | |
3017 | } | |
1da177e4 | 3018 | |
7e49fcce SR |
3019 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3020 | defined(CONFIG_PREEMPT_TRACER)) | |
3021 | ||
43627582 | 3022 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3023 | { |
6cd8a4bb | 3024 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3025 | /* |
3026 | * Underflow? | |
3027 | */ | |
9a11b49a IM |
3028 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3029 | return; | |
6cd8a4bb | 3030 | #endif |
1da177e4 | 3031 | preempt_count() += val; |
6cd8a4bb | 3032 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3033 | /* |
3034 | * Spinlock count overflowing soon? | |
3035 | */ | |
33859f7f MOS |
3036 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3037 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3038 | #endif |
3039 | if (preempt_count() == val) | |
3040 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3041 | } |
3042 | EXPORT_SYMBOL(add_preempt_count); | |
3043 | ||
43627582 | 3044 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3045 | { |
6cd8a4bb | 3046 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3047 | /* |
3048 | * Underflow? | |
3049 | */ | |
01e3eb82 | 3050 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3051 | return; |
1da177e4 LT |
3052 | /* |
3053 | * Is the spinlock portion underflowing? | |
3054 | */ | |
9a11b49a IM |
3055 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3056 | !(preempt_count() & PREEMPT_MASK))) | |
3057 | return; | |
6cd8a4bb | 3058 | #endif |
9a11b49a | 3059 | |
6cd8a4bb SR |
3060 | if (preempt_count() == val) |
3061 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3062 | preempt_count() -= val; |
3063 | } | |
3064 | EXPORT_SYMBOL(sub_preempt_count); | |
3065 | ||
3066 | #endif | |
3067 | ||
3068 | /* | |
dd41f596 | 3069 | * Print scheduling while atomic bug: |
1da177e4 | 3070 | */ |
dd41f596 | 3071 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3072 | { |
838225b4 SS |
3073 | struct pt_regs *regs = get_irq_regs(); |
3074 | ||
664dfa65 DJ |
3075 | if (oops_in_progress) |
3076 | return; | |
3077 | ||
3df0fc5b PZ |
3078 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3079 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3080 | |
dd41f596 | 3081 | debug_show_held_locks(prev); |
e21f5b15 | 3082 | print_modules(); |
dd41f596 IM |
3083 | if (irqs_disabled()) |
3084 | print_irqtrace_events(prev); | |
838225b4 SS |
3085 | |
3086 | if (regs) | |
3087 | show_regs(regs); | |
3088 | else | |
3089 | dump_stack(); | |
dd41f596 | 3090 | } |
1da177e4 | 3091 | |
dd41f596 IM |
3092 | /* |
3093 | * Various schedule()-time debugging checks and statistics: | |
3094 | */ | |
3095 | static inline void schedule_debug(struct task_struct *prev) | |
3096 | { | |
1da177e4 | 3097 | /* |
41a2d6cf | 3098 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3099 | * schedule() atomically, we ignore that path for now. |
3100 | * Otherwise, whine if we are scheduling when we should not be. | |
3101 | */ | |
3f33a7ce | 3102 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 3103 | __schedule_bug(prev); |
b3fbab05 | 3104 | rcu_sleep_check(); |
dd41f596 | 3105 | |
1da177e4 LT |
3106 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3107 | ||
2d72376b | 3108 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3109 | } |
3110 | ||
6cecd084 | 3111 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3112 | { |
61eadef6 | 3113 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 3114 | update_rq_clock(rq); |
6cecd084 | 3115 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3116 | } |
3117 | ||
dd41f596 IM |
3118 | /* |
3119 | * Pick up the highest-prio task: | |
3120 | */ | |
3121 | static inline struct task_struct * | |
b67802ea | 3122 | pick_next_task(struct rq *rq) |
dd41f596 | 3123 | { |
5522d5d5 | 3124 | const struct sched_class *class; |
dd41f596 | 3125 | struct task_struct *p; |
1da177e4 LT |
3126 | |
3127 | /* | |
dd41f596 IM |
3128 | * Optimization: we know that if all tasks are in |
3129 | * the fair class we can call that function directly: | |
1da177e4 | 3130 | */ |
953bfcd1 | 3131 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 3132 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3133 | if (likely(p)) |
3134 | return p; | |
1da177e4 LT |
3135 | } |
3136 | ||
34f971f6 | 3137 | for_each_class(class) { |
fb8d4724 | 3138 | p = class->pick_next_task(rq); |
dd41f596 IM |
3139 | if (p) |
3140 | return p; | |
dd41f596 | 3141 | } |
34f971f6 PZ |
3142 | |
3143 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3144 | } |
1da177e4 | 3145 | |
dd41f596 | 3146 | /* |
c259e01a | 3147 | * __schedule() is the main scheduler function. |
dd41f596 | 3148 | */ |
c259e01a | 3149 | static void __sched __schedule(void) |
dd41f596 IM |
3150 | { |
3151 | struct task_struct *prev, *next; | |
67ca7bde | 3152 | unsigned long *switch_count; |
dd41f596 | 3153 | struct rq *rq; |
31656519 | 3154 | int cpu; |
dd41f596 | 3155 | |
ff743345 PZ |
3156 | need_resched: |
3157 | preempt_disable(); | |
dd41f596 IM |
3158 | cpu = smp_processor_id(); |
3159 | rq = cpu_rq(cpu); | |
25502a6c | 3160 | rcu_note_context_switch(cpu); |
dd41f596 | 3161 | prev = rq->curr; |
dd41f596 | 3162 | |
dd41f596 | 3163 | schedule_debug(prev); |
1da177e4 | 3164 | |
31656519 | 3165 | if (sched_feat(HRTICK)) |
f333fdc9 | 3166 | hrtick_clear(rq); |
8f4d37ec | 3167 | |
05fa785c | 3168 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3169 | |
246d86b5 | 3170 | switch_count = &prev->nivcsw; |
1da177e4 | 3171 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3172 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3173 | prev->state = TASK_RUNNING; |
21aa9af0 | 3174 | } else { |
2acca55e PZ |
3175 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3176 | prev->on_rq = 0; | |
3177 | ||
21aa9af0 | 3178 | /* |
2acca55e PZ |
3179 | * If a worker went to sleep, notify and ask workqueue |
3180 | * whether it wants to wake up a task to maintain | |
3181 | * concurrency. | |
21aa9af0 TH |
3182 | */ |
3183 | if (prev->flags & PF_WQ_WORKER) { | |
3184 | struct task_struct *to_wakeup; | |
3185 | ||
3186 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3187 | if (to_wakeup) | |
3188 | try_to_wake_up_local(to_wakeup); | |
3189 | } | |
21aa9af0 | 3190 | } |
dd41f596 | 3191 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3192 | } |
3193 | ||
3f029d3c | 3194 | pre_schedule(rq, prev); |
f65eda4f | 3195 | |
dd41f596 | 3196 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3197 | idle_balance(cpu, rq); |
1da177e4 | 3198 | |
df1c99d4 | 3199 | put_prev_task(rq, prev); |
b67802ea | 3200 | next = pick_next_task(rq); |
f26f9aff MG |
3201 | clear_tsk_need_resched(prev); |
3202 | rq->skip_clock_update = 0; | |
1da177e4 | 3203 | |
1da177e4 | 3204 | if (likely(prev != next)) { |
1da177e4 LT |
3205 | rq->nr_switches++; |
3206 | rq->curr = next; | |
3207 | ++*switch_count; | |
3208 | ||
dd41f596 | 3209 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3210 | /* |
246d86b5 ON |
3211 | * The context switch have flipped the stack from under us |
3212 | * and restored the local variables which were saved when | |
3213 | * this task called schedule() in the past. prev == current | |
3214 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3215 | */ |
3216 | cpu = smp_processor_id(); | |
3217 | rq = cpu_rq(cpu); | |
1da177e4 | 3218 | } else |
05fa785c | 3219 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3220 | |
3f029d3c | 3221 | post_schedule(rq); |
1da177e4 | 3222 | |
ba74c144 | 3223 | sched_preempt_enable_no_resched(); |
ff743345 | 3224 | if (need_resched()) |
1da177e4 LT |
3225 | goto need_resched; |
3226 | } | |
c259e01a | 3227 | |
9c40cef2 TG |
3228 | static inline void sched_submit_work(struct task_struct *tsk) |
3229 | { | |
3c7d5184 | 3230 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3231 | return; |
3232 | /* | |
3233 | * If we are going to sleep and we have plugged IO queued, | |
3234 | * make sure to submit it to avoid deadlocks. | |
3235 | */ | |
3236 | if (blk_needs_flush_plug(tsk)) | |
3237 | blk_schedule_flush_plug(tsk); | |
3238 | } | |
3239 | ||
6ebbe7a0 | 3240 | asmlinkage void __sched schedule(void) |
c259e01a | 3241 | { |
9c40cef2 TG |
3242 | struct task_struct *tsk = current; |
3243 | ||
3244 | sched_submit_work(tsk); | |
c259e01a TG |
3245 | __schedule(); |
3246 | } | |
1da177e4 LT |
3247 | EXPORT_SYMBOL(schedule); |
3248 | ||
c5491ea7 TG |
3249 | /** |
3250 | * schedule_preempt_disabled - called with preemption disabled | |
3251 | * | |
3252 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3253 | */ | |
3254 | void __sched schedule_preempt_disabled(void) | |
3255 | { | |
ba74c144 | 3256 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3257 | schedule(); |
3258 | preempt_disable(); | |
3259 | } | |
3260 | ||
c08f7829 | 3261 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3262 | |
c6eb3dda PZ |
3263 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3264 | { | |
c6eb3dda | 3265 | if (lock->owner != owner) |
307bf980 | 3266 | return false; |
0d66bf6d PZ |
3267 | |
3268 | /* | |
c6eb3dda PZ |
3269 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3270 | * lock->owner still matches owner, if that fails, owner might | |
3271 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3272 | * ensures the memory stays valid. | |
0d66bf6d | 3273 | */ |
c6eb3dda | 3274 | barrier(); |
0d66bf6d | 3275 | |
307bf980 | 3276 | return owner->on_cpu; |
c6eb3dda | 3277 | } |
0d66bf6d | 3278 | |
c6eb3dda PZ |
3279 | /* |
3280 | * Look out! "owner" is an entirely speculative pointer | |
3281 | * access and not reliable. | |
3282 | */ | |
3283 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3284 | { | |
3285 | if (!sched_feat(OWNER_SPIN)) | |
3286 | return 0; | |
0d66bf6d | 3287 | |
307bf980 | 3288 | rcu_read_lock(); |
c6eb3dda PZ |
3289 | while (owner_running(lock, owner)) { |
3290 | if (need_resched()) | |
307bf980 | 3291 | break; |
0d66bf6d | 3292 | |
335d7afb | 3293 | arch_mutex_cpu_relax(); |
0d66bf6d | 3294 | } |
307bf980 | 3295 | rcu_read_unlock(); |
4b402210 | 3296 | |
c6eb3dda | 3297 | /* |
307bf980 TG |
3298 | * We break out the loop above on need_resched() and when the |
3299 | * owner changed, which is a sign for heavy contention. Return | |
3300 | * success only when lock->owner is NULL. | |
c6eb3dda | 3301 | */ |
307bf980 | 3302 | return lock->owner == NULL; |
0d66bf6d PZ |
3303 | } |
3304 | #endif | |
3305 | ||
1da177e4 LT |
3306 | #ifdef CONFIG_PREEMPT |
3307 | /* | |
2ed6e34f | 3308 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3309 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3310 | * occur there and call schedule directly. |
3311 | */ | |
d1f74e20 | 3312 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3313 | { |
3314 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3315 | |
1da177e4 LT |
3316 | /* |
3317 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3318 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3319 | */ |
beed33a8 | 3320 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3321 | return; |
3322 | ||
3a5c359a | 3323 | do { |
d1f74e20 | 3324 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3325 | __schedule(); |
d1f74e20 | 3326 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3327 | |
3a5c359a AK |
3328 | /* |
3329 | * Check again in case we missed a preemption opportunity | |
3330 | * between schedule and now. | |
3331 | */ | |
3332 | barrier(); | |
5ed0cec0 | 3333 | } while (need_resched()); |
1da177e4 | 3334 | } |
1da177e4 LT |
3335 | EXPORT_SYMBOL(preempt_schedule); |
3336 | ||
3337 | /* | |
2ed6e34f | 3338 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3339 | * off of irq context. |
3340 | * Note, that this is called and return with irqs disabled. This will | |
3341 | * protect us against recursive calling from irq. | |
3342 | */ | |
3343 | asmlinkage void __sched preempt_schedule_irq(void) | |
3344 | { | |
3345 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3346 | |
2ed6e34f | 3347 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3348 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3349 | ||
3a5c359a AK |
3350 | do { |
3351 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3352 | local_irq_enable(); |
c259e01a | 3353 | __schedule(); |
3a5c359a | 3354 | local_irq_disable(); |
3a5c359a | 3355 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3356 | |
3a5c359a AK |
3357 | /* |
3358 | * Check again in case we missed a preemption opportunity | |
3359 | * between schedule and now. | |
3360 | */ | |
3361 | barrier(); | |
5ed0cec0 | 3362 | } while (need_resched()); |
1da177e4 LT |
3363 | } |
3364 | ||
3365 | #endif /* CONFIG_PREEMPT */ | |
3366 | ||
63859d4f | 3367 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3368 | void *key) |
1da177e4 | 3369 | { |
63859d4f | 3370 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3371 | } |
1da177e4 LT |
3372 | EXPORT_SYMBOL(default_wake_function); |
3373 | ||
3374 | /* | |
41a2d6cf IM |
3375 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3376 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3377 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3378 | * | |
3379 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3380 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3381 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3382 | */ | |
78ddb08f | 3383 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3384 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3385 | { |
2e45874c | 3386 | wait_queue_t *curr, *next; |
1da177e4 | 3387 | |
2e45874c | 3388 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3389 | unsigned flags = curr->flags; |
3390 | ||
63859d4f | 3391 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3392 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3393 | break; |
3394 | } | |
3395 | } | |
3396 | ||
3397 | /** | |
3398 | * __wake_up - wake up threads blocked on a waitqueue. | |
3399 | * @q: the waitqueue | |
3400 | * @mode: which threads | |
3401 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3402 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3403 | * |
3404 | * It may be assumed that this function implies a write memory barrier before | |
3405 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3406 | */ |
7ad5b3a5 | 3407 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3408 | int nr_exclusive, void *key) |
1da177e4 LT |
3409 | { |
3410 | unsigned long flags; | |
3411 | ||
3412 | spin_lock_irqsave(&q->lock, flags); | |
3413 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3414 | spin_unlock_irqrestore(&q->lock, flags); | |
3415 | } | |
1da177e4 LT |
3416 | EXPORT_SYMBOL(__wake_up); |
3417 | ||
3418 | /* | |
3419 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3420 | */ | |
63b20011 | 3421 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3422 | { |
63b20011 | 3423 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3424 | } |
22c43c81 | 3425 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3426 | |
4ede816a DL |
3427 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3428 | { | |
3429 | __wake_up_common(q, mode, 1, 0, key); | |
3430 | } | |
bf294b41 | 3431 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3432 | |
1da177e4 | 3433 | /** |
4ede816a | 3434 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3435 | * @q: the waitqueue |
3436 | * @mode: which threads | |
3437 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3438 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3439 | * |
3440 | * The sync wakeup differs that the waker knows that it will schedule | |
3441 | * away soon, so while the target thread will be woken up, it will not | |
3442 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3443 | * with each other. This can prevent needless bouncing between CPUs. | |
3444 | * | |
3445 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3446 | * |
3447 | * It may be assumed that this function implies a write memory barrier before | |
3448 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3449 | */ |
4ede816a DL |
3450 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3451 | int nr_exclusive, void *key) | |
1da177e4 LT |
3452 | { |
3453 | unsigned long flags; | |
7d478721 | 3454 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3455 | |
3456 | if (unlikely(!q)) | |
3457 | return; | |
3458 | ||
3459 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3460 | wake_flags = 0; |
1da177e4 LT |
3461 | |
3462 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3463 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3464 | spin_unlock_irqrestore(&q->lock, flags); |
3465 | } | |
4ede816a DL |
3466 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3467 | ||
3468 | /* | |
3469 | * __wake_up_sync - see __wake_up_sync_key() | |
3470 | */ | |
3471 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3472 | { | |
3473 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3474 | } | |
1da177e4 LT |
3475 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3476 | ||
65eb3dc6 KD |
3477 | /** |
3478 | * complete: - signals a single thread waiting on this completion | |
3479 | * @x: holds the state of this particular completion | |
3480 | * | |
3481 | * This will wake up a single thread waiting on this completion. Threads will be | |
3482 | * awakened in the same order in which they were queued. | |
3483 | * | |
3484 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3485 | * |
3486 | * It may be assumed that this function implies a write memory barrier before | |
3487 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3488 | */ |
b15136e9 | 3489 | void complete(struct completion *x) |
1da177e4 LT |
3490 | { |
3491 | unsigned long flags; | |
3492 | ||
3493 | spin_lock_irqsave(&x->wait.lock, flags); | |
3494 | x->done++; | |
d9514f6c | 3495 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3496 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3497 | } | |
3498 | EXPORT_SYMBOL(complete); | |
3499 | ||
65eb3dc6 KD |
3500 | /** |
3501 | * complete_all: - signals all threads waiting on this completion | |
3502 | * @x: holds the state of this particular completion | |
3503 | * | |
3504 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3505 | * |
3506 | * It may be assumed that this function implies a write memory barrier before | |
3507 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3508 | */ |
b15136e9 | 3509 | void complete_all(struct completion *x) |
1da177e4 LT |
3510 | { |
3511 | unsigned long flags; | |
3512 | ||
3513 | spin_lock_irqsave(&x->wait.lock, flags); | |
3514 | x->done += UINT_MAX/2; | |
d9514f6c | 3515 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3516 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3517 | } | |
3518 | EXPORT_SYMBOL(complete_all); | |
3519 | ||
8cbbe86d AK |
3520 | static inline long __sched |
3521 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3522 | { |
1da177e4 LT |
3523 | if (!x->done) { |
3524 | DECLARE_WAITQUEUE(wait, current); | |
3525 | ||
a93d2f17 | 3526 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3527 | do { |
94d3d824 | 3528 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3529 | timeout = -ERESTARTSYS; |
3530 | break; | |
8cbbe86d AK |
3531 | } |
3532 | __set_current_state(state); | |
1da177e4 LT |
3533 | spin_unlock_irq(&x->wait.lock); |
3534 | timeout = schedule_timeout(timeout); | |
3535 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3536 | } while (!x->done && timeout); |
1da177e4 | 3537 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3538 | if (!x->done) |
3539 | return timeout; | |
1da177e4 LT |
3540 | } |
3541 | x->done--; | |
ea71a546 | 3542 | return timeout ?: 1; |
1da177e4 | 3543 | } |
1da177e4 | 3544 | |
8cbbe86d AK |
3545 | static long __sched |
3546 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3547 | { |
1da177e4 LT |
3548 | might_sleep(); |
3549 | ||
3550 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3551 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3552 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3553 | return timeout; |
3554 | } | |
1da177e4 | 3555 | |
65eb3dc6 KD |
3556 | /** |
3557 | * wait_for_completion: - waits for completion of a task | |
3558 | * @x: holds the state of this particular completion | |
3559 | * | |
3560 | * This waits to be signaled for completion of a specific task. It is NOT | |
3561 | * interruptible and there is no timeout. | |
3562 | * | |
3563 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3564 | * and interrupt capability. Also see complete(). | |
3565 | */ | |
b15136e9 | 3566 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3567 | { |
3568 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3569 | } |
8cbbe86d | 3570 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3571 | |
65eb3dc6 KD |
3572 | /** |
3573 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3574 | * @x: holds the state of this particular completion | |
3575 | * @timeout: timeout value in jiffies | |
3576 | * | |
3577 | * This waits for either a completion of a specific task to be signaled or for a | |
3578 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3579 | * interruptible. | |
c6dc7f05 BF |
3580 | * |
3581 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3582 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3583 | */ |
b15136e9 | 3584 | unsigned long __sched |
8cbbe86d | 3585 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3586 | { |
8cbbe86d | 3587 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3588 | } |
8cbbe86d | 3589 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3590 | |
65eb3dc6 KD |
3591 | /** |
3592 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3593 | * @x: holds the state of this particular completion | |
3594 | * | |
3595 | * This waits for completion of a specific task to be signaled. It is | |
3596 | * interruptible. | |
c6dc7f05 BF |
3597 | * |
3598 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3599 | */ |
8cbbe86d | 3600 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3601 | { |
51e97990 AK |
3602 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3603 | if (t == -ERESTARTSYS) | |
3604 | return t; | |
3605 | return 0; | |
0fec171c | 3606 | } |
8cbbe86d | 3607 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3608 | |
65eb3dc6 KD |
3609 | /** |
3610 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3611 | * @x: holds the state of this particular completion | |
3612 | * @timeout: timeout value in jiffies | |
3613 | * | |
3614 | * This waits for either a completion of a specific task to be signaled or for a | |
3615 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3616 | * |
3617 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3618 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3619 | */ |
6bf41237 | 3620 | long __sched |
8cbbe86d AK |
3621 | wait_for_completion_interruptible_timeout(struct completion *x, |
3622 | unsigned long timeout) | |
0fec171c | 3623 | { |
8cbbe86d | 3624 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3625 | } |
8cbbe86d | 3626 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3627 | |
65eb3dc6 KD |
3628 | /** |
3629 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3630 | * @x: holds the state of this particular completion | |
3631 | * | |
3632 | * This waits to be signaled for completion of a specific task. It can be | |
3633 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3634 | * |
3635 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3636 | */ |
009e577e MW |
3637 | int __sched wait_for_completion_killable(struct completion *x) |
3638 | { | |
3639 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3640 | if (t == -ERESTARTSYS) | |
3641 | return t; | |
3642 | return 0; | |
3643 | } | |
3644 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3645 | ||
0aa12fb4 SW |
3646 | /** |
3647 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3648 | * @x: holds the state of this particular completion | |
3649 | * @timeout: timeout value in jiffies | |
3650 | * | |
3651 | * This waits for either a completion of a specific task to be | |
3652 | * signaled or for a specified timeout to expire. It can be | |
3653 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3654 | * |
3655 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3656 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3657 | */ |
6bf41237 | 3658 | long __sched |
0aa12fb4 SW |
3659 | wait_for_completion_killable_timeout(struct completion *x, |
3660 | unsigned long timeout) | |
3661 | { | |
3662 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3663 | } | |
3664 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3665 | ||
be4de352 DC |
3666 | /** |
3667 | * try_wait_for_completion - try to decrement a completion without blocking | |
3668 | * @x: completion structure | |
3669 | * | |
3670 | * Returns: 0 if a decrement cannot be done without blocking | |
3671 | * 1 if a decrement succeeded. | |
3672 | * | |
3673 | * If a completion is being used as a counting completion, | |
3674 | * attempt to decrement the counter without blocking. This | |
3675 | * enables us to avoid waiting if the resource the completion | |
3676 | * is protecting is not available. | |
3677 | */ | |
3678 | bool try_wait_for_completion(struct completion *x) | |
3679 | { | |
7539a3b3 | 3680 | unsigned long flags; |
be4de352 DC |
3681 | int ret = 1; |
3682 | ||
7539a3b3 | 3683 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3684 | if (!x->done) |
3685 | ret = 0; | |
3686 | else | |
3687 | x->done--; | |
7539a3b3 | 3688 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3689 | return ret; |
3690 | } | |
3691 | EXPORT_SYMBOL(try_wait_for_completion); | |
3692 | ||
3693 | /** | |
3694 | * completion_done - Test to see if a completion has any waiters | |
3695 | * @x: completion structure | |
3696 | * | |
3697 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3698 | * 1 if there are no waiters. | |
3699 | * | |
3700 | */ | |
3701 | bool completion_done(struct completion *x) | |
3702 | { | |
7539a3b3 | 3703 | unsigned long flags; |
be4de352 DC |
3704 | int ret = 1; |
3705 | ||
7539a3b3 | 3706 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3707 | if (!x->done) |
3708 | ret = 0; | |
7539a3b3 | 3709 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3710 | return ret; |
3711 | } | |
3712 | EXPORT_SYMBOL(completion_done); | |
3713 | ||
8cbbe86d AK |
3714 | static long __sched |
3715 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3716 | { |
0fec171c IM |
3717 | unsigned long flags; |
3718 | wait_queue_t wait; | |
3719 | ||
3720 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3721 | |
8cbbe86d | 3722 | __set_current_state(state); |
1da177e4 | 3723 | |
8cbbe86d AK |
3724 | spin_lock_irqsave(&q->lock, flags); |
3725 | __add_wait_queue(q, &wait); | |
3726 | spin_unlock(&q->lock); | |
3727 | timeout = schedule_timeout(timeout); | |
3728 | spin_lock_irq(&q->lock); | |
3729 | __remove_wait_queue(q, &wait); | |
3730 | spin_unlock_irqrestore(&q->lock, flags); | |
3731 | ||
3732 | return timeout; | |
3733 | } | |
3734 | ||
3735 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3736 | { | |
3737 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3738 | } |
1da177e4 LT |
3739 | EXPORT_SYMBOL(interruptible_sleep_on); |
3740 | ||
0fec171c | 3741 | long __sched |
95cdf3b7 | 3742 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3743 | { |
8cbbe86d | 3744 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3745 | } |
1da177e4 LT |
3746 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3747 | ||
0fec171c | 3748 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3749 | { |
8cbbe86d | 3750 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3751 | } |
1da177e4 LT |
3752 | EXPORT_SYMBOL(sleep_on); |
3753 | ||
0fec171c | 3754 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3755 | { |
8cbbe86d | 3756 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3757 | } |
1da177e4 LT |
3758 | EXPORT_SYMBOL(sleep_on_timeout); |
3759 | ||
b29739f9 IM |
3760 | #ifdef CONFIG_RT_MUTEXES |
3761 | ||
3762 | /* | |
3763 | * rt_mutex_setprio - set the current priority of a task | |
3764 | * @p: task | |
3765 | * @prio: prio value (kernel-internal form) | |
3766 | * | |
3767 | * This function changes the 'effective' priority of a task. It does | |
3768 | * not touch ->normal_prio like __setscheduler(). | |
3769 | * | |
3770 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3771 | */ | |
36c8b586 | 3772 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3773 | { |
83b699ed | 3774 | int oldprio, on_rq, running; |
70b97a7f | 3775 | struct rq *rq; |
83ab0aa0 | 3776 | const struct sched_class *prev_class; |
b29739f9 IM |
3777 | |
3778 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3779 | ||
0122ec5b | 3780 | rq = __task_rq_lock(p); |
b29739f9 | 3781 | |
1c4dd99b TG |
3782 | /* |
3783 | * Idle task boosting is a nono in general. There is one | |
3784 | * exception, when PREEMPT_RT and NOHZ is active: | |
3785 | * | |
3786 | * The idle task calls get_next_timer_interrupt() and holds | |
3787 | * the timer wheel base->lock on the CPU and another CPU wants | |
3788 | * to access the timer (probably to cancel it). We can safely | |
3789 | * ignore the boosting request, as the idle CPU runs this code | |
3790 | * with interrupts disabled and will complete the lock | |
3791 | * protected section without being interrupted. So there is no | |
3792 | * real need to boost. | |
3793 | */ | |
3794 | if (unlikely(p == rq->idle)) { | |
3795 | WARN_ON(p != rq->curr); | |
3796 | WARN_ON(p->pi_blocked_on); | |
3797 | goto out_unlock; | |
3798 | } | |
3799 | ||
a8027073 | 3800 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3801 | oldprio = p->prio; |
83ab0aa0 | 3802 | prev_class = p->sched_class; |
fd2f4419 | 3803 | on_rq = p->on_rq; |
051a1d1a | 3804 | running = task_current(rq, p); |
0e1f3483 | 3805 | if (on_rq) |
69be72c1 | 3806 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3807 | if (running) |
3808 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3809 | |
3810 | if (rt_prio(prio)) | |
3811 | p->sched_class = &rt_sched_class; | |
3812 | else | |
3813 | p->sched_class = &fair_sched_class; | |
3814 | ||
b29739f9 IM |
3815 | p->prio = prio; |
3816 | ||
0e1f3483 HS |
3817 | if (running) |
3818 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3819 | if (on_rq) |
371fd7e7 | 3820 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3821 | |
da7a735e | 3822 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3823 | out_unlock: |
0122ec5b | 3824 | __task_rq_unlock(rq); |
b29739f9 | 3825 | } |
b29739f9 | 3826 | #endif |
36c8b586 | 3827 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3828 | { |
dd41f596 | 3829 | int old_prio, delta, on_rq; |
1da177e4 | 3830 | unsigned long flags; |
70b97a7f | 3831 | struct rq *rq; |
1da177e4 LT |
3832 | |
3833 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3834 | return; | |
3835 | /* | |
3836 | * We have to be careful, if called from sys_setpriority(), | |
3837 | * the task might be in the middle of scheduling on another CPU. | |
3838 | */ | |
3839 | rq = task_rq_lock(p, &flags); | |
3840 | /* | |
3841 | * The RT priorities are set via sched_setscheduler(), but we still | |
3842 | * allow the 'normal' nice value to be set - but as expected | |
3843 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3844 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3845 | */ |
e05606d3 | 3846 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3847 | p->static_prio = NICE_TO_PRIO(nice); |
3848 | goto out_unlock; | |
3849 | } | |
fd2f4419 | 3850 | on_rq = p->on_rq; |
c09595f6 | 3851 | if (on_rq) |
69be72c1 | 3852 | dequeue_task(rq, p, 0); |
1da177e4 | 3853 | |
1da177e4 | 3854 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3855 | set_load_weight(p); |
b29739f9 IM |
3856 | old_prio = p->prio; |
3857 | p->prio = effective_prio(p); | |
3858 | delta = p->prio - old_prio; | |
1da177e4 | 3859 | |
dd41f596 | 3860 | if (on_rq) { |
371fd7e7 | 3861 | enqueue_task(rq, p, 0); |
1da177e4 | 3862 | /* |
d5f9f942 AM |
3863 | * If the task increased its priority or is running and |
3864 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3865 | */ |
d5f9f942 | 3866 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3867 | resched_task(rq->curr); |
3868 | } | |
3869 | out_unlock: | |
0122ec5b | 3870 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3871 | } |
1da177e4 LT |
3872 | EXPORT_SYMBOL(set_user_nice); |
3873 | ||
e43379f1 MM |
3874 | /* |
3875 | * can_nice - check if a task can reduce its nice value | |
3876 | * @p: task | |
3877 | * @nice: nice value | |
3878 | */ | |
36c8b586 | 3879 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3880 | { |
024f4747 MM |
3881 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3882 | int nice_rlim = 20 - nice; | |
48f24c4d | 3883 | |
78d7d407 | 3884 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3885 | capable(CAP_SYS_NICE)); |
3886 | } | |
3887 | ||
1da177e4 LT |
3888 | #ifdef __ARCH_WANT_SYS_NICE |
3889 | ||
3890 | /* | |
3891 | * sys_nice - change the priority of the current process. | |
3892 | * @increment: priority increment | |
3893 | * | |
3894 | * sys_setpriority is a more generic, but much slower function that | |
3895 | * does similar things. | |
3896 | */ | |
5add95d4 | 3897 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3898 | { |
48f24c4d | 3899 | long nice, retval; |
1da177e4 LT |
3900 | |
3901 | /* | |
3902 | * Setpriority might change our priority at the same moment. | |
3903 | * We don't have to worry. Conceptually one call occurs first | |
3904 | * and we have a single winner. | |
3905 | */ | |
e43379f1 MM |
3906 | if (increment < -40) |
3907 | increment = -40; | |
1da177e4 LT |
3908 | if (increment > 40) |
3909 | increment = 40; | |
3910 | ||
2b8f836f | 3911 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3912 | if (nice < -20) |
3913 | nice = -20; | |
3914 | if (nice > 19) | |
3915 | nice = 19; | |
3916 | ||
e43379f1 MM |
3917 | if (increment < 0 && !can_nice(current, nice)) |
3918 | return -EPERM; | |
3919 | ||
1da177e4 LT |
3920 | retval = security_task_setnice(current, nice); |
3921 | if (retval) | |
3922 | return retval; | |
3923 | ||
3924 | set_user_nice(current, nice); | |
3925 | return 0; | |
3926 | } | |
3927 | ||
3928 | #endif | |
3929 | ||
3930 | /** | |
3931 | * task_prio - return the priority value of a given task. | |
3932 | * @p: the task in question. | |
3933 | * | |
3934 | * This is the priority value as seen by users in /proc. | |
3935 | * RT tasks are offset by -200. Normal tasks are centered | |
3936 | * around 0, value goes from -16 to +15. | |
3937 | */ | |
36c8b586 | 3938 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3939 | { |
3940 | return p->prio - MAX_RT_PRIO; | |
3941 | } | |
3942 | ||
3943 | /** | |
3944 | * task_nice - return the nice value of a given task. | |
3945 | * @p: the task in question. | |
3946 | */ | |
36c8b586 | 3947 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3948 | { |
3949 | return TASK_NICE(p); | |
3950 | } | |
150d8bed | 3951 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3952 | |
3953 | /** | |
3954 | * idle_cpu - is a given cpu idle currently? | |
3955 | * @cpu: the processor in question. | |
3956 | */ | |
3957 | int idle_cpu(int cpu) | |
3958 | { | |
908a3283 TG |
3959 | struct rq *rq = cpu_rq(cpu); |
3960 | ||
3961 | if (rq->curr != rq->idle) | |
3962 | return 0; | |
3963 | ||
3964 | if (rq->nr_running) | |
3965 | return 0; | |
3966 | ||
3967 | #ifdef CONFIG_SMP | |
3968 | if (!llist_empty(&rq->wake_list)) | |
3969 | return 0; | |
3970 | #endif | |
3971 | ||
3972 | return 1; | |
1da177e4 LT |
3973 | } |
3974 | ||
1da177e4 LT |
3975 | /** |
3976 | * idle_task - return the idle task for a given cpu. | |
3977 | * @cpu: the processor in question. | |
3978 | */ | |
36c8b586 | 3979 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3980 | { |
3981 | return cpu_rq(cpu)->idle; | |
3982 | } | |
3983 | ||
3984 | /** | |
3985 | * find_process_by_pid - find a process with a matching PID value. | |
3986 | * @pid: the pid in question. | |
3987 | */ | |
a9957449 | 3988 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3989 | { |
228ebcbe | 3990 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3991 | } |
3992 | ||
3993 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3994 | static void |
3995 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3996 | { |
1da177e4 LT |
3997 | p->policy = policy; |
3998 | p->rt_priority = prio; | |
b29739f9 IM |
3999 | p->normal_prio = normal_prio(p); |
4000 | /* we are holding p->pi_lock already */ | |
4001 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4002 | if (rt_prio(p->prio)) |
4003 | p->sched_class = &rt_sched_class; | |
4004 | else | |
4005 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4006 | set_load_weight(p); |
1da177e4 LT |
4007 | } |
4008 | ||
c69e8d9c DH |
4009 | /* |
4010 | * check the target process has a UID that matches the current process's | |
4011 | */ | |
4012 | static bool check_same_owner(struct task_struct *p) | |
4013 | { | |
4014 | const struct cred *cred = current_cred(), *pcred; | |
4015 | bool match; | |
4016 | ||
4017 | rcu_read_lock(); | |
4018 | pcred = __task_cred(p); | |
b0e77598 SH |
4019 | if (cred->user->user_ns == pcred->user->user_ns) |
4020 | match = (cred->euid == pcred->euid || | |
4021 | cred->euid == pcred->uid); | |
4022 | else | |
4023 | match = false; | |
c69e8d9c DH |
4024 | rcu_read_unlock(); |
4025 | return match; | |
4026 | } | |
4027 | ||
961ccddd | 4028 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4029 | const struct sched_param *param, bool user) |
1da177e4 | 4030 | { |
83b699ed | 4031 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4032 | unsigned long flags; |
83ab0aa0 | 4033 | const struct sched_class *prev_class; |
70b97a7f | 4034 | struct rq *rq; |
ca94c442 | 4035 | int reset_on_fork; |
1da177e4 | 4036 | |
66e5393a SR |
4037 | /* may grab non-irq protected spin_locks */ |
4038 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4039 | recheck: |
4040 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4041 | if (policy < 0) { |
4042 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4043 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4044 | } else { |
4045 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4046 | policy &= ~SCHED_RESET_ON_FORK; | |
4047 | ||
4048 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4049 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4050 | policy != SCHED_IDLE) | |
4051 | return -EINVAL; | |
4052 | } | |
4053 | ||
1da177e4 LT |
4054 | /* |
4055 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4056 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4057 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4058 | */ |
4059 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4060 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4061 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4062 | return -EINVAL; |
e05606d3 | 4063 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4064 | return -EINVAL; |
4065 | ||
37e4ab3f OC |
4066 | /* |
4067 | * Allow unprivileged RT tasks to decrease priority: | |
4068 | */ | |
961ccddd | 4069 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4070 | if (rt_policy(policy)) { |
a44702e8 ON |
4071 | unsigned long rlim_rtprio = |
4072 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4073 | |
4074 | /* can't set/change the rt policy */ | |
4075 | if (policy != p->policy && !rlim_rtprio) | |
4076 | return -EPERM; | |
4077 | ||
4078 | /* can't increase priority */ | |
4079 | if (param->sched_priority > p->rt_priority && | |
4080 | param->sched_priority > rlim_rtprio) | |
4081 | return -EPERM; | |
4082 | } | |
c02aa73b | 4083 | |
dd41f596 | 4084 | /* |
c02aa73b DH |
4085 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4086 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4087 | */ |
c02aa73b DH |
4088 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4089 | if (!can_nice(p, TASK_NICE(p))) | |
4090 | return -EPERM; | |
4091 | } | |
5fe1d75f | 4092 | |
37e4ab3f | 4093 | /* can't change other user's priorities */ |
c69e8d9c | 4094 | if (!check_same_owner(p)) |
37e4ab3f | 4095 | return -EPERM; |
ca94c442 LP |
4096 | |
4097 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4098 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4099 | return -EPERM; | |
37e4ab3f | 4100 | } |
1da177e4 | 4101 | |
725aad24 | 4102 | if (user) { |
b0ae1981 | 4103 | retval = security_task_setscheduler(p); |
725aad24 JF |
4104 | if (retval) |
4105 | return retval; | |
4106 | } | |
4107 | ||
b29739f9 IM |
4108 | /* |
4109 | * make sure no PI-waiters arrive (or leave) while we are | |
4110 | * changing the priority of the task: | |
0122ec5b | 4111 | * |
25985edc | 4112 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4113 | * runqueue lock must be held. |
4114 | */ | |
0122ec5b | 4115 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 4116 | |
34f971f6 PZ |
4117 | /* |
4118 | * Changing the policy of the stop threads its a very bad idea | |
4119 | */ | |
4120 | if (p == rq->stop) { | |
0122ec5b | 4121 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
4122 | return -EINVAL; |
4123 | } | |
4124 | ||
a51e9198 DF |
4125 | /* |
4126 | * If not changing anything there's no need to proceed further: | |
4127 | */ | |
4128 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
4129 | param->sched_priority == p->rt_priority))) { | |
4130 | ||
4131 | __task_rq_unlock(rq); | |
4132 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4133 | return 0; | |
4134 | } | |
4135 | ||
dc61b1d6 PZ |
4136 | #ifdef CONFIG_RT_GROUP_SCHED |
4137 | if (user) { | |
4138 | /* | |
4139 | * Do not allow realtime tasks into groups that have no runtime | |
4140 | * assigned. | |
4141 | */ | |
4142 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4143 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4144 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4145 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4146 | return -EPERM; |
4147 | } | |
4148 | } | |
4149 | #endif | |
4150 | ||
1da177e4 LT |
4151 | /* recheck policy now with rq lock held */ |
4152 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4153 | policy = oldpolicy = -1; | |
0122ec5b | 4154 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4155 | goto recheck; |
4156 | } | |
fd2f4419 | 4157 | on_rq = p->on_rq; |
051a1d1a | 4158 | running = task_current(rq, p); |
0e1f3483 | 4159 | if (on_rq) |
4ca9b72b | 4160 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4161 | if (running) |
4162 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4163 | |
ca94c442 LP |
4164 | p->sched_reset_on_fork = reset_on_fork; |
4165 | ||
1da177e4 | 4166 | oldprio = p->prio; |
83ab0aa0 | 4167 | prev_class = p->sched_class; |
dd41f596 | 4168 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4169 | |
0e1f3483 HS |
4170 | if (running) |
4171 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4172 | if (on_rq) |
4ca9b72b | 4173 | enqueue_task(rq, p, 0); |
cb469845 | 4174 | |
da7a735e | 4175 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4176 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4177 | |
95e02ca9 TG |
4178 | rt_mutex_adjust_pi(p); |
4179 | ||
1da177e4 LT |
4180 | return 0; |
4181 | } | |
961ccddd RR |
4182 | |
4183 | /** | |
4184 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4185 | * @p: the task in question. | |
4186 | * @policy: new policy. | |
4187 | * @param: structure containing the new RT priority. | |
4188 | * | |
4189 | * NOTE that the task may be already dead. | |
4190 | */ | |
4191 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4192 | const struct sched_param *param) |
961ccddd RR |
4193 | { |
4194 | return __sched_setscheduler(p, policy, param, true); | |
4195 | } | |
1da177e4 LT |
4196 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4197 | ||
961ccddd RR |
4198 | /** |
4199 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4200 | * @p: the task in question. | |
4201 | * @policy: new policy. | |
4202 | * @param: structure containing the new RT priority. | |
4203 | * | |
4204 | * Just like sched_setscheduler, only don't bother checking if the | |
4205 | * current context has permission. For example, this is needed in | |
4206 | * stop_machine(): we create temporary high priority worker threads, | |
4207 | * but our caller might not have that capability. | |
4208 | */ | |
4209 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4210 | const struct sched_param *param) |
961ccddd RR |
4211 | { |
4212 | return __sched_setscheduler(p, policy, param, false); | |
4213 | } | |
4214 | ||
95cdf3b7 IM |
4215 | static int |
4216 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4217 | { |
1da177e4 LT |
4218 | struct sched_param lparam; |
4219 | struct task_struct *p; | |
36c8b586 | 4220 | int retval; |
1da177e4 LT |
4221 | |
4222 | if (!param || pid < 0) | |
4223 | return -EINVAL; | |
4224 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4225 | return -EFAULT; | |
5fe1d75f ON |
4226 | |
4227 | rcu_read_lock(); | |
4228 | retval = -ESRCH; | |
1da177e4 | 4229 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4230 | if (p != NULL) |
4231 | retval = sched_setscheduler(p, policy, &lparam); | |
4232 | rcu_read_unlock(); | |
36c8b586 | 4233 | |
1da177e4 LT |
4234 | return retval; |
4235 | } | |
4236 | ||
4237 | /** | |
4238 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4239 | * @pid: the pid in question. | |
4240 | * @policy: new policy. | |
4241 | * @param: structure containing the new RT priority. | |
4242 | */ | |
5add95d4 HC |
4243 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4244 | struct sched_param __user *, param) | |
1da177e4 | 4245 | { |
c21761f1 JB |
4246 | /* negative values for policy are not valid */ |
4247 | if (policy < 0) | |
4248 | return -EINVAL; | |
4249 | ||
1da177e4 LT |
4250 | return do_sched_setscheduler(pid, policy, param); |
4251 | } | |
4252 | ||
4253 | /** | |
4254 | * sys_sched_setparam - set/change the RT priority of a thread | |
4255 | * @pid: the pid in question. | |
4256 | * @param: structure containing the new RT priority. | |
4257 | */ | |
5add95d4 | 4258 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4259 | { |
4260 | return do_sched_setscheduler(pid, -1, param); | |
4261 | } | |
4262 | ||
4263 | /** | |
4264 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4265 | * @pid: the pid in question. | |
4266 | */ | |
5add95d4 | 4267 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4268 | { |
36c8b586 | 4269 | struct task_struct *p; |
3a5c359a | 4270 | int retval; |
1da177e4 LT |
4271 | |
4272 | if (pid < 0) | |
3a5c359a | 4273 | return -EINVAL; |
1da177e4 LT |
4274 | |
4275 | retval = -ESRCH; | |
5fe85be0 | 4276 | rcu_read_lock(); |
1da177e4 LT |
4277 | p = find_process_by_pid(pid); |
4278 | if (p) { | |
4279 | retval = security_task_getscheduler(p); | |
4280 | if (!retval) | |
ca94c442 LP |
4281 | retval = p->policy |
4282 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4283 | } |
5fe85be0 | 4284 | rcu_read_unlock(); |
1da177e4 LT |
4285 | return retval; |
4286 | } | |
4287 | ||
4288 | /** | |
ca94c442 | 4289 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4290 | * @pid: the pid in question. |
4291 | * @param: structure containing the RT priority. | |
4292 | */ | |
5add95d4 | 4293 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4294 | { |
4295 | struct sched_param lp; | |
36c8b586 | 4296 | struct task_struct *p; |
3a5c359a | 4297 | int retval; |
1da177e4 LT |
4298 | |
4299 | if (!param || pid < 0) | |
3a5c359a | 4300 | return -EINVAL; |
1da177e4 | 4301 | |
5fe85be0 | 4302 | rcu_read_lock(); |
1da177e4 LT |
4303 | p = find_process_by_pid(pid); |
4304 | retval = -ESRCH; | |
4305 | if (!p) | |
4306 | goto out_unlock; | |
4307 | ||
4308 | retval = security_task_getscheduler(p); | |
4309 | if (retval) | |
4310 | goto out_unlock; | |
4311 | ||
4312 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4313 | rcu_read_unlock(); |
1da177e4 LT |
4314 | |
4315 | /* | |
4316 | * This one might sleep, we cannot do it with a spinlock held ... | |
4317 | */ | |
4318 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4319 | ||
1da177e4 LT |
4320 | return retval; |
4321 | ||
4322 | out_unlock: | |
5fe85be0 | 4323 | rcu_read_unlock(); |
1da177e4 LT |
4324 | return retval; |
4325 | } | |
4326 | ||
96f874e2 | 4327 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4328 | { |
5a16f3d3 | 4329 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4330 | struct task_struct *p; |
4331 | int retval; | |
1da177e4 | 4332 | |
95402b38 | 4333 | get_online_cpus(); |
23f5d142 | 4334 | rcu_read_lock(); |
1da177e4 LT |
4335 | |
4336 | p = find_process_by_pid(pid); | |
4337 | if (!p) { | |
23f5d142 | 4338 | rcu_read_unlock(); |
95402b38 | 4339 | put_online_cpus(); |
1da177e4 LT |
4340 | return -ESRCH; |
4341 | } | |
4342 | ||
23f5d142 | 4343 | /* Prevent p going away */ |
1da177e4 | 4344 | get_task_struct(p); |
23f5d142 | 4345 | rcu_read_unlock(); |
1da177e4 | 4346 | |
5a16f3d3 RR |
4347 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4348 | retval = -ENOMEM; | |
4349 | goto out_put_task; | |
4350 | } | |
4351 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4352 | retval = -ENOMEM; | |
4353 | goto out_free_cpus_allowed; | |
4354 | } | |
1da177e4 | 4355 | retval = -EPERM; |
f1c84dae | 4356 | if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE)) |
1da177e4 LT |
4357 | goto out_unlock; |
4358 | ||
b0ae1981 | 4359 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4360 | if (retval) |
4361 | goto out_unlock; | |
4362 | ||
5a16f3d3 RR |
4363 | cpuset_cpus_allowed(p, cpus_allowed); |
4364 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4365 | again: |
5a16f3d3 | 4366 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4367 | |
8707d8b8 | 4368 | if (!retval) { |
5a16f3d3 RR |
4369 | cpuset_cpus_allowed(p, cpus_allowed); |
4370 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4371 | /* |
4372 | * We must have raced with a concurrent cpuset | |
4373 | * update. Just reset the cpus_allowed to the | |
4374 | * cpuset's cpus_allowed | |
4375 | */ | |
5a16f3d3 | 4376 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4377 | goto again; |
4378 | } | |
4379 | } | |
1da177e4 | 4380 | out_unlock: |
5a16f3d3 RR |
4381 | free_cpumask_var(new_mask); |
4382 | out_free_cpus_allowed: | |
4383 | free_cpumask_var(cpus_allowed); | |
4384 | out_put_task: | |
1da177e4 | 4385 | put_task_struct(p); |
95402b38 | 4386 | put_online_cpus(); |
1da177e4 LT |
4387 | return retval; |
4388 | } | |
4389 | ||
4390 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4391 | struct cpumask *new_mask) |
1da177e4 | 4392 | { |
96f874e2 RR |
4393 | if (len < cpumask_size()) |
4394 | cpumask_clear(new_mask); | |
4395 | else if (len > cpumask_size()) | |
4396 | len = cpumask_size(); | |
4397 | ||
1da177e4 LT |
4398 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4399 | } | |
4400 | ||
4401 | /** | |
4402 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4403 | * @pid: pid of the process | |
4404 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4405 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4406 | */ | |
5add95d4 HC |
4407 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4408 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4409 | { |
5a16f3d3 | 4410 | cpumask_var_t new_mask; |
1da177e4 LT |
4411 | int retval; |
4412 | ||
5a16f3d3 RR |
4413 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4414 | return -ENOMEM; | |
1da177e4 | 4415 | |
5a16f3d3 RR |
4416 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4417 | if (retval == 0) | |
4418 | retval = sched_setaffinity(pid, new_mask); | |
4419 | free_cpumask_var(new_mask); | |
4420 | return retval; | |
1da177e4 LT |
4421 | } |
4422 | ||
96f874e2 | 4423 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4424 | { |
36c8b586 | 4425 | struct task_struct *p; |
31605683 | 4426 | unsigned long flags; |
1da177e4 | 4427 | int retval; |
1da177e4 | 4428 | |
95402b38 | 4429 | get_online_cpus(); |
23f5d142 | 4430 | rcu_read_lock(); |
1da177e4 LT |
4431 | |
4432 | retval = -ESRCH; | |
4433 | p = find_process_by_pid(pid); | |
4434 | if (!p) | |
4435 | goto out_unlock; | |
4436 | ||
e7834f8f DQ |
4437 | retval = security_task_getscheduler(p); |
4438 | if (retval) | |
4439 | goto out_unlock; | |
4440 | ||
013fdb80 | 4441 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4442 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4443 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4444 | |
4445 | out_unlock: | |
23f5d142 | 4446 | rcu_read_unlock(); |
95402b38 | 4447 | put_online_cpus(); |
1da177e4 | 4448 | |
9531b62f | 4449 | return retval; |
1da177e4 LT |
4450 | } |
4451 | ||
4452 | /** | |
4453 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4454 | * @pid: pid of the process | |
4455 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4456 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4457 | */ | |
5add95d4 HC |
4458 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4459 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4460 | { |
4461 | int ret; | |
f17c8607 | 4462 | cpumask_var_t mask; |
1da177e4 | 4463 | |
84fba5ec | 4464 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4465 | return -EINVAL; |
4466 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4467 | return -EINVAL; |
4468 | ||
f17c8607 RR |
4469 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4470 | return -ENOMEM; | |
1da177e4 | 4471 | |
f17c8607 RR |
4472 | ret = sched_getaffinity(pid, mask); |
4473 | if (ret == 0) { | |
8bc037fb | 4474 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4475 | |
4476 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4477 | ret = -EFAULT; |
4478 | else | |
cd3d8031 | 4479 | ret = retlen; |
f17c8607 RR |
4480 | } |
4481 | free_cpumask_var(mask); | |
1da177e4 | 4482 | |
f17c8607 | 4483 | return ret; |
1da177e4 LT |
4484 | } |
4485 | ||
4486 | /** | |
4487 | * sys_sched_yield - yield the current processor to other threads. | |
4488 | * | |
dd41f596 IM |
4489 | * This function yields the current CPU to other tasks. If there are no |
4490 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4491 | */ |
5add95d4 | 4492 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4493 | { |
70b97a7f | 4494 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4495 | |
2d72376b | 4496 | schedstat_inc(rq, yld_count); |
4530d7ab | 4497 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4498 | |
4499 | /* | |
4500 | * Since we are going to call schedule() anyway, there's | |
4501 | * no need to preempt or enable interrupts: | |
4502 | */ | |
4503 | __release(rq->lock); | |
8a25d5de | 4504 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4505 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4506 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4507 | |
4508 | schedule(); | |
4509 | ||
4510 | return 0; | |
4511 | } | |
4512 | ||
d86ee480 PZ |
4513 | static inline int should_resched(void) |
4514 | { | |
4515 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4516 | } | |
4517 | ||
e7b38404 | 4518 | static void __cond_resched(void) |
1da177e4 | 4519 | { |
e7aaaa69 | 4520 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4521 | __schedule(); |
e7aaaa69 | 4522 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4523 | } |
4524 | ||
02b67cc3 | 4525 | int __sched _cond_resched(void) |
1da177e4 | 4526 | { |
d86ee480 | 4527 | if (should_resched()) { |
1da177e4 LT |
4528 | __cond_resched(); |
4529 | return 1; | |
4530 | } | |
4531 | return 0; | |
4532 | } | |
02b67cc3 | 4533 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4534 | |
4535 | /* | |
613afbf8 | 4536 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4537 | * call schedule, and on return reacquire the lock. |
4538 | * | |
41a2d6cf | 4539 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4540 | * operations here to prevent schedule() from being called twice (once via |
4541 | * spin_unlock(), once by hand). | |
4542 | */ | |
613afbf8 | 4543 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4544 | { |
d86ee480 | 4545 | int resched = should_resched(); |
6df3cecb JK |
4546 | int ret = 0; |
4547 | ||
f607c668 PZ |
4548 | lockdep_assert_held(lock); |
4549 | ||
95c354fe | 4550 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4551 | spin_unlock(lock); |
d86ee480 | 4552 | if (resched) |
95c354fe NP |
4553 | __cond_resched(); |
4554 | else | |
4555 | cpu_relax(); | |
6df3cecb | 4556 | ret = 1; |
1da177e4 | 4557 | spin_lock(lock); |
1da177e4 | 4558 | } |
6df3cecb | 4559 | return ret; |
1da177e4 | 4560 | } |
613afbf8 | 4561 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4562 | |
613afbf8 | 4563 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4564 | { |
4565 | BUG_ON(!in_softirq()); | |
4566 | ||
d86ee480 | 4567 | if (should_resched()) { |
98d82567 | 4568 | local_bh_enable(); |
1da177e4 LT |
4569 | __cond_resched(); |
4570 | local_bh_disable(); | |
4571 | return 1; | |
4572 | } | |
4573 | return 0; | |
4574 | } | |
613afbf8 | 4575 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4576 | |
1da177e4 LT |
4577 | /** |
4578 | * yield - yield the current processor to other threads. | |
4579 | * | |
72fd4a35 | 4580 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4581 | * thread runnable and calls sys_sched_yield(). |
4582 | */ | |
4583 | void __sched yield(void) | |
4584 | { | |
4585 | set_current_state(TASK_RUNNING); | |
4586 | sys_sched_yield(); | |
4587 | } | |
1da177e4 LT |
4588 | EXPORT_SYMBOL(yield); |
4589 | ||
d95f4122 MG |
4590 | /** |
4591 | * yield_to - yield the current processor to another thread in | |
4592 | * your thread group, or accelerate that thread toward the | |
4593 | * processor it's on. | |
16addf95 RD |
4594 | * @p: target task |
4595 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4596 | * |
4597 | * It's the caller's job to ensure that the target task struct | |
4598 | * can't go away on us before we can do any checks. | |
4599 | * | |
4600 | * Returns true if we indeed boosted the target task. | |
4601 | */ | |
4602 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4603 | { | |
4604 | struct task_struct *curr = current; | |
4605 | struct rq *rq, *p_rq; | |
4606 | unsigned long flags; | |
4607 | bool yielded = 0; | |
4608 | ||
4609 | local_irq_save(flags); | |
4610 | rq = this_rq(); | |
4611 | ||
4612 | again: | |
4613 | p_rq = task_rq(p); | |
4614 | double_rq_lock(rq, p_rq); | |
4615 | while (task_rq(p) != p_rq) { | |
4616 | double_rq_unlock(rq, p_rq); | |
4617 | goto again; | |
4618 | } | |
4619 | ||
4620 | if (!curr->sched_class->yield_to_task) | |
4621 | goto out; | |
4622 | ||
4623 | if (curr->sched_class != p->sched_class) | |
4624 | goto out; | |
4625 | ||
4626 | if (task_running(p_rq, p) || p->state) | |
4627 | goto out; | |
4628 | ||
4629 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4630 | if (yielded) { |
d95f4122 | 4631 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4632 | /* |
4633 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4634 | * fairness. | |
4635 | */ | |
4636 | if (preempt && rq != p_rq) | |
4637 | resched_task(p_rq->curr); | |
916671c0 MG |
4638 | } else { |
4639 | /* | |
4640 | * We might have set it in task_yield_fair(), but are | |
4641 | * not going to schedule(), so don't want to skip | |
4642 | * the next update. | |
4643 | */ | |
4644 | rq->skip_clock_update = 0; | |
6d1cafd8 | 4645 | } |
d95f4122 MG |
4646 | |
4647 | out: | |
4648 | double_rq_unlock(rq, p_rq); | |
4649 | local_irq_restore(flags); | |
4650 | ||
4651 | if (yielded) | |
4652 | schedule(); | |
4653 | ||
4654 | return yielded; | |
4655 | } | |
4656 | EXPORT_SYMBOL_GPL(yield_to); | |
4657 | ||
1da177e4 | 4658 | /* |
41a2d6cf | 4659 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4660 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4661 | */ |
4662 | void __sched io_schedule(void) | |
4663 | { | |
54d35f29 | 4664 | struct rq *rq = raw_rq(); |
1da177e4 | 4665 | |
0ff92245 | 4666 | delayacct_blkio_start(); |
1da177e4 | 4667 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4668 | blk_flush_plug(current); |
8f0dfc34 | 4669 | current->in_iowait = 1; |
1da177e4 | 4670 | schedule(); |
8f0dfc34 | 4671 | current->in_iowait = 0; |
1da177e4 | 4672 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4673 | delayacct_blkio_end(); |
1da177e4 | 4674 | } |
1da177e4 LT |
4675 | EXPORT_SYMBOL(io_schedule); |
4676 | ||
4677 | long __sched io_schedule_timeout(long timeout) | |
4678 | { | |
54d35f29 | 4679 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4680 | long ret; |
4681 | ||
0ff92245 | 4682 | delayacct_blkio_start(); |
1da177e4 | 4683 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4684 | blk_flush_plug(current); |
8f0dfc34 | 4685 | current->in_iowait = 1; |
1da177e4 | 4686 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4687 | current->in_iowait = 0; |
1da177e4 | 4688 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4689 | delayacct_blkio_end(); |
1da177e4 LT |
4690 | return ret; |
4691 | } | |
4692 | ||
4693 | /** | |
4694 | * sys_sched_get_priority_max - return maximum RT priority. | |
4695 | * @policy: scheduling class. | |
4696 | * | |
4697 | * this syscall returns the maximum rt_priority that can be used | |
4698 | * by a given scheduling class. | |
4699 | */ | |
5add95d4 | 4700 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4701 | { |
4702 | int ret = -EINVAL; | |
4703 | ||
4704 | switch (policy) { | |
4705 | case SCHED_FIFO: | |
4706 | case SCHED_RR: | |
4707 | ret = MAX_USER_RT_PRIO-1; | |
4708 | break; | |
4709 | case SCHED_NORMAL: | |
b0a9499c | 4710 | case SCHED_BATCH: |
dd41f596 | 4711 | case SCHED_IDLE: |
1da177e4 LT |
4712 | ret = 0; |
4713 | break; | |
4714 | } | |
4715 | return ret; | |
4716 | } | |
4717 | ||
4718 | /** | |
4719 | * sys_sched_get_priority_min - return minimum RT priority. | |
4720 | * @policy: scheduling class. | |
4721 | * | |
4722 | * this syscall returns the minimum rt_priority that can be used | |
4723 | * by a given scheduling class. | |
4724 | */ | |
5add95d4 | 4725 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4726 | { |
4727 | int ret = -EINVAL; | |
4728 | ||
4729 | switch (policy) { | |
4730 | case SCHED_FIFO: | |
4731 | case SCHED_RR: | |
4732 | ret = 1; | |
4733 | break; | |
4734 | case SCHED_NORMAL: | |
b0a9499c | 4735 | case SCHED_BATCH: |
dd41f596 | 4736 | case SCHED_IDLE: |
1da177e4 LT |
4737 | ret = 0; |
4738 | } | |
4739 | return ret; | |
4740 | } | |
4741 | ||
4742 | /** | |
4743 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4744 | * @pid: pid of the process. | |
4745 | * @interval: userspace pointer to the timeslice value. | |
4746 | * | |
4747 | * this syscall writes the default timeslice value of a given process | |
4748 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4749 | */ | |
17da2bd9 | 4750 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4751 | struct timespec __user *, interval) |
1da177e4 | 4752 | { |
36c8b586 | 4753 | struct task_struct *p; |
a4ec24b4 | 4754 | unsigned int time_slice; |
dba091b9 TG |
4755 | unsigned long flags; |
4756 | struct rq *rq; | |
3a5c359a | 4757 | int retval; |
1da177e4 | 4758 | struct timespec t; |
1da177e4 LT |
4759 | |
4760 | if (pid < 0) | |
3a5c359a | 4761 | return -EINVAL; |
1da177e4 LT |
4762 | |
4763 | retval = -ESRCH; | |
1a551ae7 | 4764 | rcu_read_lock(); |
1da177e4 LT |
4765 | p = find_process_by_pid(pid); |
4766 | if (!p) | |
4767 | goto out_unlock; | |
4768 | ||
4769 | retval = security_task_getscheduler(p); | |
4770 | if (retval) | |
4771 | goto out_unlock; | |
4772 | ||
dba091b9 TG |
4773 | rq = task_rq_lock(p, &flags); |
4774 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4775 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4776 | |
1a551ae7 | 4777 | rcu_read_unlock(); |
a4ec24b4 | 4778 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4779 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4780 | return retval; |
3a5c359a | 4781 | |
1da177e4 | 4782 | out_unlock: |
1a551ae7 | 4783 | rcu_read_unlock(); |
1da177e4 LT |
4784 | return retval; |
4785 | } | |
4786 | ||
7c731e0a | 4787 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4788 | |
82a1fcb9 | 4789 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4790 | { |
1da177e4 | 4791 | unsigned long free = 0; |
36c8b586 | 4792 | unsigned state; |
1da177e4 | 4793 | |
1da177e4 | 4794 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4795 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4796 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4797 | #if BITS_PER_LONG == 32 |
1da177e4 | 4798 | if (state == TASK_RUNNING) |
3df0fc5b | 4799 | printk(KERN_CONT " running "); |
1da177e4 | 4800 | else |
3df0fc5b | 4801 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4802 | #else |
4803 | if (state == TASK_RUNNING) | |
3df0fc5b | 4804 | printk(KERN_CONT " running task "); |
1da177e4 | 4805 | else |
3df0fc5b | 4806 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4807 | #endif |
4808 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4809 | free = stack_not_used(p); |
1da177e4 | 4810 | #endif |
3df0fc5b | 4811 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
07cde260 | 4812 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), |
aa47b7e0 | 4813 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4814 | |
5fb5e6de | 4815 | show_stack(p, NULL); |
1da177e4 LT |
4816 | } |
4817 | ||
e59e2ae2 | 4818 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4819 | { |
36c8b586 | 4820 | struct task_struct *g, *p; |
1da177e4 | 4821 | |
4bd77321 | 4822 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4823 | printk(KERN_INFO |
4824 | " task PC stack pid father\n"); | |
1da177e4 | 4825 | #else |
3df0fc5b PZ |
4826 | printk(KERN_INFO |
4827 | " task PC stack pid father\n"); | |
1da177e4 | 4828 | #endif |
510f5acc | 4829 | rcu_read_lock(); |
1da177e4 LT |
4830 | do_each_thread(g, p) { |
4831 | /* | |
4832 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4833 | * console might take a lot of time: |
1da177e4 LT |
4834 | */ |
4835 | touch_nmi_watchdog(); | |
39bc89fd | 4836 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4837 | sched_show_task(p); |
1da177e4 LT |
4838 | } while_each_thread(g, p); |
4839 | ||
04c9167f JF |
4840 | touch_all_softlockup_watchdogs(); |
4841 | ||
dd41f596 IM |
4842 | #ifdef CONFIG_SCHED_DEBUG |
4843 | sysrq_sched_debug_show(); | |
4844 | #endif | |
510f5acc | 4845 | rcu_read_unlock(); |
e59e2ae2 IM |
4846 | /* |
4847 | * Only show locks if all tasks are dumped: | |
4848 | */ | |
93335a21 | 4849 | if (!state_filter) |
e59e2ae2 | 4850 | debug_show_all_locks(); |
1da177e4 LT |
4851 | } |
4852 | ||
1df21055 IM |
4853 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4854 | { | |
dd41f596 | 4855 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4856 | } |
4857 | ||
f340c0d1 IM |
4858 | /** |
4859 | * init_idle - set up an idle thread for a given CPU | |
4860 | * @idle: task in question | |
4861 | * @cpu: cpu the idle task belongs to | |
4862 | * | |
4863 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4864 | * flag, to make booting more robust. | |
4865 | */ | |
5c1e1767 | 4866 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4867 | { |
70b97a7f | 4868 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4869 | unsigned long flags; |
4870 | ||
05fa785c | 4871 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4872 | |
dd41f596 | 4873 | __sched_fork(idle); |
06b83b5f | 4874 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4875 | idle->se.exec_start = sched_clock(); |
4876 | ||
1e1b6c51 | 4877 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4878 | /* |
4879 | * We're having a chicken and egg problem, even though we are | |
4880 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4881 | * lockdep check in task_group() will fail. | |
4882 | * | |
4883 | * Similar case to sched_fork(). / Alternatively we could | |
4884 | * use task_rq_lock() here and obtain the other rq->lock. | |
4885 | * | |
4886 | * Silence PROVE_RCU | |
4887 | */ | |
4888 | rcu_read_lock(); | |
dd41f596 | 4889 | __set_task_cpu(idle, cpu); |
6506cf6c | 4890 | rcu_read_unlock(); |
1da177e4 | 4891 | |
1da177e4 | 4892 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4893 | #if defined(CONFIG_SMP) |
4894 | idle->on_cpu = 1; | |
4866cde0 | 4895 | #endif |
05fa785c | 4896 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4897 | |
4898 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4899 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4900 | |
dd41f596 IM |
4901 | /* |
4902 | * The idle tasks have their own, simple scheduling class: | |
4903 | */ | |
4904 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4905 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
4906 | #if defined(CONFIG_SMP) |
4907 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4908 | #endif | |
19978ca6 IM |
4909 | } |
4910 | ||
1da177e4 | 4911 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4912 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4913 | { | |
4914 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4915 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4916 | |
4917 | cpumask_copy(&p->cpus_allowed, new_mask); | |
4918 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
1e1b6c51 KM |
4919 | } |
4920 | ||
1da177e4 LT |
4921 | /* |
4922 | * This is how migration works: | |
4923 | * | |
969c7921 TH |
4924 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4925 | * stop_one_cpu(). | |
4926 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4927 | * off the CPU) | |
4928 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4929 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4930 | * it and puts it into the right queue. |
969c7921 TH |
4931 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4932 | * is done. | |
1da177e4 LT |
4933 | */ |
4934 | ||
4935 | /* | |
4936 | * Change a given task's CPU affinity. Migrate the thread to a | |
4937 | * proper CPU and schedule it away if the CPU it's executing on | |
4938 | * is removed from the allowed bitmask. | |
4939 | * | |
4940 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4941 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4942 | * call is not atomic; no spinlocks may be held. |
4943 | */ | |
96f874e2 | 4944 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4945 | { |
4946 | unsigned long flags; | |
70b97a7f | 4947 | struct rq *rq; |
969c7921 | 4948 | unsigned int dest_cpu; |
48f24c4d | 4949 | int ret = 0; |
1da177e4 LT |
4950 | |
4951 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4952 | |
db44fc01 YZ |
4953 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4954 | goto out; | |
4955 | ||
6ad4c188 | 4956 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4957 | ret = -EINVAL; |
4958 | goto out; | |
4959 | } | |
4960 | ||
db44fc01 | 4961 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
4962 | ret = -EINVAL; |
4963 | goto out; | |
4964 | } | |
4965 | ||
1e1b6c51 | 4966 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4967 | |
1da177e4 | 4968 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4969 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4970 | goto out; |
4971 | ||
969c7921 | 4972 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4973 | if (p->on_rq) { |
969c7921 | 4974 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4975 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4976 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4977 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4978 | tlb_migrate_finish(p->mm); |
4979 | return 0; | |
4980 | } | |
4981 | out: | |
0122ec5b | 4982 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4983 | |
1da177e4 LT |
4984 | return ret; |
4985 | } | |
cd8ba7cd | 4986 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4987 | |
4988 | /* | |
41a2d6cf | 4989 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4990 | * this because either it can't run here any more (set_cpus_allowed() |
4991 | * away from this CPU, or CPU going down), or because we're | |
4992 | * attempting to rebalance this task on exec (sched_exec). | |
4993 | * | |
4994 | * So we race with normal scheduler movements, but that's OK, as long | |
4995 | * as the task is no longer on this CPU. | |
efc30814 KK |
4996 | * |
4997 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4998 | */ |
efc30814 | 4999 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5000 | { |
70b97a7f | 5001 | struct rq *rq_dest, *rq_src; |
e2912009 | 5002 | int ret = 0; |
1da177e4 | 5003 | |
e761b772 | 5004 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5005 | return ret; |
1da177e4 LT |
5006 | |
5007 | rq_src = cpu_rq(src_cpu); | |
5008 | rq_dest = cpu_rq(dest_cpu); | |
5009 | ||
0122ec5b | 5010 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
5011 | double_rq_lock(rq_src, rq_dest); |
5012 | /* Already moved. */ | |
5013 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5014 | goto done; |
1da177e4 | 5015 | /* Affinity changed (again). */ |
fa17b507 | 5016 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 5017 | goto fail; |
1da177e4 | 5018 | |
e2912009 PZ |
5019 | /* |
5020 | * If we're not on a rq, the next wake-up will ensure we're | |
5021 | * placed properly. | |
5022 | */ | |
fd2f4419 | 5023 | if (p->on_rq) { |
4ca9b72b | 5024 | dequeue_task(rq_src, p, 0); |
e2912009 | 5025 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 5026 | enqueue_task(rq_dest, p, 0); |
15afe09b | 5027 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5028 | } |
b1e38734 | 5029 | done: |
efc30814 | 5030 | ret = 1; |
b1e38734 | 5031 | fail: |
1da177e4 | 5032 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 5033 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 5034 | return ret; |
1da177e4 LT |
5035 | } |
5036 | ||
5037 | /* | |
969c7921 TH |
5038 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5039 | * and performs thread migration by bumping thread off CPU then | |
5040 | * 'pushing' onto another runqueue. | |
1da177e4 | 5041 | */ |
969c7921 | 5042 | static int migration_cpu_stop(void *data) |
1da177e4 | 5043 | { |
969c7921 | 5044 | struct migration_arg *arg = data; |
f7b4cddc | 5045 | |
969c7921 TH |
5046 | /* |
5047 | * The original target cpu might have gone down and we might | |
5048 | * be on another cpu but it doesn't matter. | |
5049 | */ | |
f7b4cddc | 5050 | local_irq_disable(); |
969c7921 | 5051 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5052 | local_irq_enable(); |
1da177e4 | 5053 | return 0; |
f7b4cddc ON |
5054 | } |
5055 | ||
1da177e4 | 5056 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5057 | |
054b9108 | 5058 | /* |
48c5ccae PZ |
5059 | * Ensures that the idle task is using init_mm right before its cpu goes |
5060 | * offline. | |
054b9108 | 5061 | */ |
48c5ccae | 5062 | void idle_task_exit(void) |
1da177e4 | 5063 | { |
48c5ccae | 5064 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5065 | |
48c5ccae | 5066 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5067 | |
48c5ccae PZ |
5068 | if (mm != &init_mm) |
5069 | switch_mm(mm, &init_mm, current); | |
5070 | mmdrop(mm); | |
1da177e4 LT |
5071 | } |
5072 | ||
5073 | /* | |
5074 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5075 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5076 | * for performance reasons the counter is not stricly tracking tasks to | |
5077 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5078 | * to keep the global sum constant after CPU-down: | |
5079 | */ | |
70b97a7f | 5080 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5081 | { |
6ad4c188 | 5082 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5083 | |
1da177e4 LT |
5084 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5085 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5086 | } |
5087 | ||
dd41f596 | 5088 | /* |
48c5ccae | 5089 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5090 | */ |
48c5ccae | 5091 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5092 | { |
48c5ccae PZ |
5093 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5094 | rq->calc_load_active = 0; | |
1da177e4 LT |
5095 | } |
5096 | ||
48f24c4d | 5097 | /* |
48c5ccae PZ |
5098 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5099 | * try_to_wake_up()->select_task_rq(). | |
5100 | * | |
5101 | * Called with rq->lock held even though we'er in stop_machine() and | |
5102 | * there's no concurrency possible, we hold the required locks anyway | |
5103 | * because of lock validation efforts. | |
1da177e4 | 5104 | */ |
48c5ccae | 5105 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5106 | { |
70b97a7f | 5107 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5108 | struct task_struct *next, *stop = rq->stop; |
5109 | int dest_cpu; | |
1da177e4 LT |
5110 | |
5111 | /* | |
48c5ccae PZ |
5112 | * Fudge the rq selection such that the below task selection loop |
5113 | * doesn't get stuck on the currently eligible stop task. | |
5114 | * | |
5115 | * We're currently inside stop_machine() and the rq is either stuck | |
5116 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5117 | * either way we should never end up calling schedule() until we're | |
5118 | * done here. | |
1da177e4 | 5119 | */ |
48c5ccae | 5120 | rq->stop = NULL; |
48f24c4d | 5121 | |
8cb120d3 PT |
5122 | /* Ensure any throttled groups are reachable by pick_next_task */ |
5123 | unthrottle_offline_cfs_rqs(rq); | |
5124 | ||
dd41f596 | 5125 | for ( ; ; ) { |
48c5ccae PZ |
5126 | /* |
5127 | * There's this thread running, bail when that's the only | |
5128 | * remaining thread. | |
5129 | */ | |
5130 | if (rq->nr_running == 1) | |
dd41f596 | 5131 | break; |
48c5ccae | 5132 | |
b67802ea | 5133 | next = pick_next_task(rq); |
48c5ccae | 5134 | BUG_ON(!next); |
79c53799 | 5135 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5136 | |
48c5ccae PZ |
5137 | /* Find suitable destination for @next, with force if needed. */ |
5138 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5139 | raw_spin_unlock(&rq->lock); | |
5140 | ||
5141 | __migrate_task(next, dead_cpu, dest_cpu); | |
5142 | ||
5143 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5144 | } |
dce48a84 | 5145 | |
48c5ccae | 5146 | rq->stop = stop; |
dce48a84 | 5147 | } |
48c5ccae | 5148 | |
1da177e4 LT |
5149 | #endif /* CONFIG_HOTPLUG_CPU */ |
5150 | ||
e692ab53 NP |
5151 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5152 | ||
5153 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5154 | { |
5155 | .procname = "sched_domain", | |
c57baf1e | 5156 | .mode = 0555, |
e0361851 | 5157 | }, |
56992309 | 5158 | {} |
e692ab53 NP |
5159 | }; |
5160 | ||
5161 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5162 | { |
5163 | .procname = "kernel", | |
c57baf1e | 5164 | .mode = 0555, |
e0361851 AD |
5165 | .child = sd_ctl_dir, |
5166 | }, | |
56992309 | 5167 | {} |
e692ab53 NP |
5168 | }; |
5169 | ||
5170 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5171 | { | |
5172 | struct ctl_table *entry = | |
5cf9f062 | 5173 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5174 | |
e692ab53 NP |
5175 | return entry; |
5176 | } | |
5177 | ||
6382bc90 MM |
5178 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5179 | { | |
cd790076 | 5180 | struct ctl_table *entry; |
6382bc90 | 5181 | |
cd790076 MM |
5182 | /* |
5183 | * In the intermediate directories, both the child directory and | |
5184 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5185 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5186 | * static strings and all have proc handlers. |
5187 | */ | |
5188 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5189 | if (entry->child) |
5190 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5191 | if (entry->proc_handler == NULL) |
5192 | kfree(entry->procname); | |
5193 | } | |
6382bc90 MM |
5194 | |
5195 | kfree(*tablep); | |
5196 | *tablep = NULL; | |
5197 | } | |
5198 | ||
e692ab53 | 5199 | static void |
e0361851 | 5200 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5201 | const char *procname, void *data, int maxlen, |
36fcb589 | 5202 | umode_t mode, proc_handler *proc_handler) |
e692ab53 | 5203 | { |
e692ab53 NP |
5204 | entry->procname = procname; |
5205 | entry->data = data; | |
5206 | entry->maxlen = maxlen; | |
5207 | entry->mode = mode; | |
5208 | entry->proc_handler = proc_handler; | |
5209 | } | |
5210 | ||
5211 | static struct ctl_table * | |
5212 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5213 | { | |
a5d8c348 | 5214 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5215 | |
ad1cdc1d MM |
5216 | if (table == NULL) |
5217 | return NULL; | |
5218 | ||
e0361851 | 5219 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5220 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5221 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5222 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5223 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5224 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5225 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5226 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5227 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5228 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5229 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5230 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5231 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5232 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5233 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5234 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5235 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5236 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5237 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5238 | &sd->cache_nice_tries, |
5239 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5240 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5241 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5242 | set_table_entry(&table[11], "name", sd->name, |
5243 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5244 | /* &table[12] is terminator */ | |
e692ab53 NP |
5245 | |
5246 | return table; | |
5247 | } | |
5248 | ||
9a4e7159 | 5249 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5250 | { |
5251 | struct ctl_table *entry, *table; | |
5252 | struct sched_domain *sd; | |
5253 | int domain_num = 0, i; | |
5254 | char buf[32]; | |
5255 | ||
5256 | for_each_domain(cpu, sd) | |
5257 | domain_num++; | |
5258 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5259 | if (table == NULL) |
5260 | return NULL; | |
e692ab53 NP |
5261 | |
5262 | i = 0; | |
5263 | for_each_domain(cpu, sd) { | |
5264 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5265 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5266 | entry->mode = 0555; |
e692ab53 NP |
5267 | entry->child = sd_alloc_ctl_domain_table(sd); |
5268 | entry++; | |
5269 | i++; | |
5270 | } | |
5271 | return table; | |
5272 | } | |
5273 | ||
5274 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5275 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5276 | { |
6ad4c188 | 5277 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5278 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5279 | char buf[32]; | |
5280 | ||
7378547f MM |
5281 | WARN_ON(sd_ctl_dir[0].child); |
5282 | sd_ctl_dir[0].child = entry; | |
5283 | ||
ad1cdc1d MM |
5284 | if (entry == NULL) |
5285 | return; | |
5286 | ||
6ad4c188 | 5287 | for_each_possible_cpu(i) { |
e692ab53 | 5288 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5289 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5290 | entry->mode = 0555; |
e692ab53 | 5291 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5292 | entry++; |
e692ab53 | 5293 | } |
7378547f MM |
5294 | |
5295 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5296 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5297 | } | |
6382bc90 | 5298 | |
7378547f | 5299 | /* may be called multiple times per register */ |
6382bc90 MM |
5300 | static void unregister_sched_domain_sysctl(void) |
5301 | { | |
7378547f MM |
5302 | if (sd_sysctl_header) |
5303 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5304 | sd_sysctl_header = NULL; |
7378547f MM |
5305 | if (sd_ctl_dir[0].child) |
5306 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5307 | } |
e692ab53 | 5308 | #else |
6382bc90 MM |
5309 | static void register_sched_domain_sysctl(void) |
5310 | { | |
5311 | } | |
5312 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5313 | { |
5314 | } | |
5315 | #endif | |
5316 | ||
1f11eb6a GH |
5317 | static void set_rq_online(struct rq *rq) |
5318 | { | |
5319 | if (!rq->online) { | |
5320 | const struct sched_class *class; | |
5321 | ||
c6c4927b | 5322 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5323 | rq->online = 1; |
5324 | ||
5325 | for_each_class(class) { | |
5326 | if (class->rq_online) | |
5327 | class->rq_online(rq); | |
5328 | } | |
5329 | } | |
5330 | } | |
5331 | ||
5332 | static void set_rq_offline(struct rq *rq) | |
5333 | { | |
5334 | if (rq->online) { | |
5335 | const struct sched_class *class; | |
5336 | ||
5337 | for_each_class(class) { | |
5338 | if (class->rq_offline) | |
5339 | class->rq_offline(rq); | |
5340 | } | |
5341 | ||
c6c4927b | 5342 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5343 | rq->online = 0; |
5344 | } | |
5345 | } | |
5346 | ||
1da177e4 LT |
5347 | /* |
5348 | * migration_call - callback that gets triggered when a CPU is added. | |
5349 | * Here we can start up the necessary migration thread for the new CPU. | |
5350 | */ | |
48f24c4d IM |
5351 | static int __cpuinit |
5352 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5353 | { |
48f24c4d | 5354 | int cpu = (long)hcpu; |
1da177e4 | 5355 | unsigned long flags; |
969c7921 | 5356 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5357 | |
48c5ccae | 5358 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5359 | |
1da177e4 | 5360 | case CPU_UP_PREPARE: |
a468d389 | 5361 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5362 | break; |
48f24c4d | 5363 | |
1da177e4 | 5364 | case CPU_ONLINE: |
1f94ef59 | 5365 | /* Update our root-domain */ |
05fa785c | 5366 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5367 | if (rq->rd) { |
c6c4927b | 5368 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5369 | |
5370 | set_rq_online(rq); | |
1f94ef59 | 5371 | } |
05fa785c | 5372 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5373 | break; |
48f24c4d | 5374 | |
1da177e4 | 5375 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5376 | case CPU_DYING: |
317f3941 | 5377 | sched_ttwu_pending(); |
57d885fe | 5378 | /* Update our root-domain */ |
05fa785c | 5379 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5380 | if (rq->rd) { |
c6c4927b | 5381 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5382 | set_rq_offline(rq); |
57d885fe | 5383 | } |
48c5ccae PZ |
5384 | migrate_tasks(cpu); |
5385 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5386 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5387 | |
5388 | migrate_nr_uninterruptible(rq); | |
5389 | calc_global_load_remove(rq); | |
57d885fe | 5390 | break; |
1da177e4 LT |
5391 | #endif |
5392 | } | |
49c022e6 PZ |
5393 | |
5394 | update_max_interval(); | |
5395 | ||
1da177e4 LT |
5396 | return NOTIFY_OK; |
5397 | } | |
5398 | ||
f38b0820 PM |
5399 | /* |
5400 | * Register at high priority so that task migration (migrate_all_tasks) | |
5401 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5402 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5403 | */ |
26c2143b | 5404 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5405 | .notifier_call = migration_call, |
50a323b7 | 5406 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5407 | }; |
5408 | ||
3a101d05 TH |
5409 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5410 | unsigned long action, void *hcpu) | |
5411 | { | |
5412 | switch (action & ~CPU_TASKS_FROZEN) { | |
5413 | case CPU_ONLINE: | |
5414 | case CPU_DOWN_FAILED: | |
5415 | set_cpu_active((long)hcpu, true); | |
5416 | return NOTIFY_OK; | |
5417 | default: | |
5418 | return NOTIFY_DONE; | |
5419 | } | |
5420 | } | |
5421 | ||
5422 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5423 | unsigned long action, void *hcpu) | |
5424 | { | |
5425 | switch (action & ~CPU_TASKS_FROZEN) { | |
5426 | case CPU_DOWN_PREPARE: | |
5427 | set_cpu_active((long)hcpu, false); | |
5428 | return NOTIFY_OK; | |
5429 | default: | |
5430 | return NOTIFY_DONE; | |
5431 | } | |
5432 | } | |
5433 | ||
7babe8db | 5434 | static int __init migration_init(void) |
1da177e4 LT |
5435 | { |
5436 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5437 | int err; |
48f24c4d | 5438 | |
3a101d05 | 5439 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5440 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5441 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5442 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5443 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5444 | |
3a101d05 TH |
5445 | /* Register cpu active notifiers */ |
5446 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5447 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5448 | ||
a004cd42 | 5449 | return 0; |
1da177e4 | 5450 | } |
7babe8db | 5451 | early_initcall(migration_init); |
1da177e4 LT |
5452 | #endif |
5453 | ||
5454 | #ifdef CONFIG_SMP | |
476f3534 | 5455 | |
4cb98839 PZ |
5456 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5457 | ||
3e9830dc | 5458 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5459 | |
f6630114 MT |
5460 | static __read_mostly int sched_domain_debug_enabled; |
5461 | ||
5462 | static int __init sched_domain_debug_setup(char *str) | |
5463 | { | |
5464 | sched_domain_debug_enabled = 1; | |
5465 | ||
5466 | return 0; | |
5467 | } | |
5468 | early_param("sched_debug", sched_domain_debug_setup); | |
5469 | ||
7c16ec58 | 5470 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5471 | struct cpumask *groupmask) |
1da177e4 | 5472 | { |
4dcf6aff | 5473 | struct sched_group *group = sd->groups; |
434d53b0 | 5474 | char str[256]; |
1da177e4 | 5475 | |
968ea6d8 | 5476 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5477 | cpumask_clear(groupmask); |
4dcf6aff IM |
5478 | |
5479 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5480 | ||
5481 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5482 | printk("does not load-balance\n"); |
4dcf6aff | 5483 | if (sd->parent) |
3df0fc5b PZ |
5484 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5485 | " has parent"); | |
4dcf6aff | 5486 | return -1; |
41c7ce9a NP |
5487 | } |
5488 | ||
3df0fc5b | 5489 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5490 | |
758b2cdc | 5491 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5492 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5493 | "CPU%d\n", cpu); | |
4dcf6aff | 5494 | } |
758b2cdc | 5495 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5496 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5497 | " CPU%d\n", cpu); | |
4dcf6aff | 5498 | } |
1da177e4 | 5499 | |
4dcf6aff | 5500 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5501 | do { |
4dcf6aff | 5502 | if (!group) { |
3df0fc5b PZ |
5503 | printk("\n"); |
5504 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5505 | break; |
5506 | } | |
5507 | ||
9c3f75cb | 5508 | if (!group->sgp->power) { |
3df0fc5b PZ |
5509 | printk(KERN_CONT "\n"); |
5510 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5511 | "set\n"); | |
4dcf6aff IM |
5512 | break; |
5513 | } | |
1da177e4 | 5514 | |
758b2cdc | 5515 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5516 | printk(KERN_CONT "\n"); |
5517 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5518 | break; |
5519 | } | |
1da177e4 | 5520 | |
758b2cdc | 5521 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5522 | printk(KERN_CONT "\n"); |
5523 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5524 | break; |
5525 | } | |
1da177e4 | 5526 | |
758b2cdc | 5527 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5528 | |
968ea6d8 | 5529 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5530 | |
3df0fc5b | 5531 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5532 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5533 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5534 | group->sgp->power); |
381512cf | 5535 | } |
1da177e4 | 5536 | |
4dcf6aff IM |
5537 | group = group->next; |
5538 | } while (group != sd->groups); | |
3df0fc5b | 5539 | printk(KERN_CONT "\n"); |
1da177e4 | 5540 | |
758b2cdc | 5541 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5542 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5543 | |
758b2cdc RR |
5544 | if (sd->parent && |
5545 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5546 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5547 | "of domain->span\n"); | |
4dcf6aff IM |
5548 | return 0; |
5549 | } | |
1da177e4 | 5550 | |
4dcf6aff IM |
5551 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5552 | { | |
5553 | int level = 0; | |
1da177e4 | 5554 | |
f6630114 MT |
5555 | if (!sched_domain_debug_enabled) |
5556 | return; | |
5557 | ||
4dcf6aff IM |
5558 | if (!sd) { |
5559 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5560 | return; | |
5561 | } | |
1da177e4 | 5562 | |
4dcf6aff IM |
5563 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5564 | ||
5565 | for (;;) { | |
4cb98839 | 5566 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5567 | break; |
1da177e4 LT |
5568 | level++; |
5569 | sd = sd->parent; | |
33859f7f | 5570 | if (!sd) |
4dcf6aff IM |
5571 | break; |
5572 | } | |
1da177e4 | 5573 | } |
6d6bc0ad | 5574 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5575 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5576 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5577 | |
1a20ff27 | 5578 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5579 | { |
758b2cdc | 5580 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5581 | return 1; |
5582 | ||
5583 | /* Following flags need at least 2 groups */ | |
5584 | if (sd->flags & (SD_LOAD_BALANCE | | |
5585 | SD_BALANCE_NEWIDLE | | |
5586 | SD_BALANCE_FORK | | |
89c4710e SS |
5587 | SD_BALANCE_EXEC | |
5588 | SD_SHARE_CPUPOWER | | |
5589 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5590 | if (sd->groups != sd->groups->next) |
5591 | return 0; | |
5592 | } | |
5593 | ||
5594 | /* Following flags don't use groups */ | |
c88d5910 | 5595 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5596 | return 0; |
5597 | ||
5598 | return 1; | |
5599 | } | |
5600 | ||
48f24c4d IM |
5601 | static int |
5602 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5603 | { |
5604 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5605 | ||
5606 | if (sd_degenerate(parent)) | |
5607 | return 1; | |
5608 | ||
758b2cdc | 5609 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5610 | return 0; |
5611 | ||
245af2c7 SS |
5612 | /* Flags needing groups don't count if only 1 group in parent */ |
5613 | if (parent->groups == parent->groups->next) { | |
5614 | pflags &= ~(SD_LOAD_BALANCE | | |
5615 | SD_BALANCE_NEWIDLE | | |
5616 | SD_BALANCE_FORK | | |
89c4710e SS |
5617 | SD_BALANCE_EXEC | |
5618 | SD_SHARE_CPUPOWER | | |
5619 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5620 | if (nr_node_ids == 1) |
5621 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5622 | } |
5623 | if (~cflags & pflags) | |
5624 | return 0; | |
5625 | ||
5626 | return 1; | |
5627 | } | |
5628 | ||
dce840a0 | 5629 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5630 | { |
dce840a0 | 5631 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5632 | |
68e74568 | 5633 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5634 | free_cpumask_var(rd->rto_mask); |
5635 | free_cpumask_var(rd->online); | |
5636 | free_cpumask_var(rd->span); | |
5637 | kfree(rd); | |
5638 | } | |
5639 | ||
57d885fe GH |
5640 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5641 | { | |
a0490fa3 | 5642 | struct root_domain *old_rd = NULL; |
57d885fe | 5643 | unsigned long flags; |
57d885fe | 5644 | |
05fa785c | 5645 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5646 | |
5647 | if (rq->rd) { | |
a0490fa3 | 5648 | old_rd = rq->rd; |
57d885fe | 5649 | |
c6c4927b | 5650 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5651 | set_rq_offline(rq); |
57d885fe | 5652 | |
c6c4927b | 5653 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5654 | |
a0490fa3 IM |
5655 | /* |
5656 | * If we dont want to free the old_rt yet then | |
5657 | * set old_rd to NULL to skip the freeing later | |
5658 | * in this function: | |
5659 | */ | |
5660 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5661 | old_rd = NULL; | |
57d885fe GH |
5662 | } |
5663 | ||
5664 | atomic_inc(&rd->refcount); | |
5665 | rq->rd = rd; | |
5666 | ||
c6c4927b | 5667 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5668 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5669 | set_rq_online(rq); |
57d885fe | 5670 | |
05fa785c | 5671 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5672 | |
5673 | if (old_rd) | |
dce840a0 | 5674 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5675 | } |
5676 | ||
68c38fc3 | 5677 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5678 | { |
5679 | memset(rd, 0, sizeof(*rd)); | |
5680 | ||
68c38fc3 | 5681 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5682 | goto out; |
68c38fc3 | 5683 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5684 | goto free_span; |
68c38fc3 | 5685 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5686 | goto free_online; |
6e0534f2 | 5687 | |
68c38fc3 | 5688 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5689 | goto free_rto_mask; |
c6c4927b | 5690 | return 0; |
6e0534f2 | 5691 | |
68e74568 RR |
5692 | free_rto_mask: |
5693 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5694 | free_online: |
5695 | free_cpumask_var(rd->online); | |
5696 | free_span: | |
5697 | free_cpumask_var(rd->span); | |
0c910d28 | 5698 | out: |
c6c4927b | 5699 | return -ENOMEM; |
57d885fe GH |
5700 | } |
5701 | ||
029632fb PZ |
5702 | /* |
5703 | * By default the system creates a single root-domain with all cpus as | |
5704 | * members (mimicking the global state we have today). | |
5705 | */ | |
5706 | struct root_domain def_root_domain; | |
5707 | ||
57d885fe GH |
5708 | static void init_defrootdomain(void) |
5709 | { | |
68c38fc3 | 5710 | init_rootdomain(&def_root_domain); |
c6c4927b | 5711 | |
57d885fe GH |
5712 | atomic_set(&def_root_domain.refcount, 1); |
5713 | } | |
5714 | ||
dc938520 | 5715 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5716 | { |
5717 | struct root_domain *rd; | |
5718 | ||
5719 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5720 | if (!rd) | |
5721 | return NULL; | |
5722 | ||
68c38fc3 | 5723 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5724 | kfree(rd); |
5725 | return NULL; | |
5726 | } | |
57d885fe GH |
5727 | |
5728 | return rd; | |
5729 | } | |
5730 | ||
e3589f6c PZ |
5731 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5732 | { | |
5733 | struct sched_group *tmp, *first; | |
5734 | ||
5735 | if (!sg) | |
5736 | return; | |
5737 | ||
5738 | first = sg; | |
5739 | do { | |
5740 | tmp = sg->next; | |
5741 | ||
5742 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5743 | kfree(sg->sgp); | |
5744 | ||
5745 | kfree(sg); | |
5746 | sg = tmp; | |
5747 | } while (sg != first); | |
5748 | } | |
5749 | ||
dce840a0 PZ |
5750 | static void free_sched_domain(struct rcu_head *rcu) |
5751 | { | |
5752 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5753 | |
5754 | /* | |
5755 | * If its an overlapping domain it has private groups, iterate and | |
5756 | * nuke them all. | |
5757 | */ | |
5758 | if (sd->flags & SD_OVERLAP) { | |
5759 | free_sched_groups(sd->groups, 1); | |
5760 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5761 | kfree(sd->groups->sgp); |
dce840a0 | 5762 | kfree(sd->groups); |
9c3f75cb | 5763 | } |
dce840a0 PZ |
5764 | kfree(sd); |
5765 | } | |
5766 | ||
5767 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5768 | { | |
5769 | call_rcu(&sd->rcu, free_sched_domain); | |
5770 | } | |
5771 | ||
5772 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5773 | { | |
5774 | for (; sd; sd = sd->parent) | |
5775 | destroy_sched_domain(sd, cpu); | |
5776 | } | |
5777 | ||
518cd623 PZ |
5778 | /* |
5779 | * Keep a special pointer to the highest sched_domain that has | |
5780 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5781 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5782 | * | |
5783 | * Also keep a unique ID per domain (we use the first cpu number in | |
5784 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5785 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5786 | */ |
5787 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5788 | DEFINE_PER_CPU(int, sd_llc_id); | |
5789 | ||
5790 | static void update_top_cache_domain(int cpu) | |
5791 | { | |
5792 | struct sched_domain *sd; | |
5793 | int id = cpu; | |
5794 | ||
5795 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
5796 | if (sd) | |
5797 | id = cpumask_first(sched_domain_span(sd)); | |
5798 | ||
5799 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5800 | per_cpu(sd_llc_id, cpu) = id; | |
5801 | } | |
5802 | ||
1da177e4 | 5803 | /* |
0eab9146 | 5804 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5805 | * hold the hotplug lock. |
5806 | */ | |
0eab9146 IM |
5807 | static void |
5808 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5809 | { |
70b97a7f | 5810 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5811 | struct sched_domain *tmp; |
5812 | ||
5813 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5814 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5815 | struct sched_domain *parent = tmp->parent; |
5816 | if (!parent) | |
5817 | break; | |
f29c9b1c | 5818 | |
1a848870 | 5819 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5820 | tmp->parent = parent->parent; |
1a848870 SS |
5821 | if (parent->parent) |
5822 | parent->parent->child = tmp; | |
dce840a0 | 5823 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5824 | } else |
5825 | tmp = tmp->parent; | |
245af2c7 SS |
5826 | } |
5827 | ||
1a848870 | 5828 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5829 | tmp = sd; |
245af2c7 | 5830 | sd = sd->parent; |
dce840a0 | 5831 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5832 | if (sd) |
5833 | sd->child = NULL; | |
5834 | } | |
1da177e4 | 5835 | |
4cb98839 | 5836 | sched_domain_debug(sd, cpu); |
1da177e4 | 5837 | |
57d885fe | 5838 | rq_attach_root(rq, rd); |
dce840a0 | 5839 | tmp = rq->sd; |
674311d5 | 5840 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5841 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5842 | |
5843 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5844 | } |
5845 | ||
5846 | /* cpus with isolated domains */ | |
dcc30a35 | 5847 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5848 | |
5849 | /* Setup the mask of cpus configured for isolated domains */ | |
5850 | static int __init isolated_cpu_setup(char *str) | |
5851 | { | |
bdddd296 | 5852 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5853 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5854 | return 1; |
5855 | } | |
5856 | ||
8927f494 | 5857 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5858 | |
9c1cfda2 | 5859 | #ifdef CONFIG_NUMA |
198e2f18 | 5860 | |
9c1cfda2 JH |
5861 | /** |
5862 | * find_next_best_node - find the next node to include in a sched_domain | |
5863 | * @node: node whose sched_domain we're building | |
5864 | * @used_nodes: nodes already in the sched_domain | |
5865 | * | |
41a2d6cf | 5866 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5867 | * finds the closest node not already in the @used_nodes map. |
5868 | * | |
5869 | * Should use nodemask_t. | |
5870 | */ | |
c5f59f08 | 5871 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 5872 | { |
7142d17e | 5873 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
5874 | |
5875 | min_val = INT_MAX; | |
5876 | ||
076ac2af | 5877 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 5878 | /* Start at @node */ |
076ac2af | 5879 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
5880 | |
5881 | if (!nr_cpus_node(n)) | |
5882 | continue; | |
5883 | ||
5884 | /* Skip already used nodes */ | |
c5f59f08 | 5885 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
5886 | continue; |
5887 | ||
5888 | /* Simple min distance search */ | |
5889 | val = node_distance(node, n); | |
5890 | ||
5891 | if (val < min_val) { | |
5892 | min_val = val; | |
5893 | best_node = n; | |
5894 | } | |
5895 | } | |
5896 | ||
7142d17e HD |
5897 | if (best_node != -1) |
5898 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
5899 | return best_node; |
5900 | } | |
5901 | ||
5902 | /** | |
5903 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5904 | * @node: node whose cpumask we're constructing | |
73486722 | 5905 | * @span: resulting cpumask |
9c1cfda2 | 5906 | * |
41a2d6cf | 5907 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
5908 | * should be one that prevents unnecessary balancing, but also spreads tasks |
5909 | * out optimally. | |
5910 | */ | |
96f874e2 | 5911 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 5912 | { |
c5f59f08 | 5913 | nodemask_t used_nodes; |
48f24c4d | 5914 | int i; |
9c1cfda2 | 5915 | |
6ca09dfc | 5916 | cpumask_clear(span); |
c5f59f08 | 5917 | nodes_clear(used_nodes); |
9c1cfda2 | 5918 | |
6ca09dfc | 5919 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 5920 | node_set(node, used_nodes); |
9c1cfda2 JH |
5921 | |
5922 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 5923 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
5924 | if (next_node < 0) |
5925 | break; | |
6ca09dfc | 5926 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 5927 | } |
9c1cfda2 | 5928 | } |
d3081f52 PZ |
5929 | |
5930 | static const struct cpumask *cpu_node_mask(int cpu) | |
5931 | { | |
5932 | lockdep_assert_held(&sched_domains_mutex); | |
5933 | ||
5934 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
5935 | ||
5936 | return sched_domains_tmpmask; | |
5937 | } | |
2c402dc3 PZ |
5938 | |
5939 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
5940 | { | |
5941 | return cpu_possible_mask; | |
5942 | } | |
6d6bc0ad | 5943 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 5944 | |
d3081f52 PZ |
5945 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5946 | { | |
5947 | return cpumask_of_node(cpu_to_node(cpu)); | |
5948 | } | |
5949 | ||
5c45bf27 | 5950 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5951 | |
dce840a0 PZ |
5952 | struct sd_data { |
5953 | struct sched_domain **__percpu sd; | |
5954 | struct sched_group **__percpu sg; | |
9c3f75cb | 5955 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5956 | }; |
5957 | ||
49a02c51 | 5958 | struct s_data { |
21d42ccf | 5959 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5960 | struct root_domain *rd; |
5961 | }; | |
5962 | ||
2109b99e | 5963 | enum s_alloc { |
2109b99e | 5964 | sa_rootdomain, |
21d42ccf | 5965 | sa_sd, |
dce840a0 | 5966 | sa_sd_storage, |
2109b99e AH |
5967 | sa_none, |
5968 | }; | |
5969 | ||
54ab4ff4 PZ |
5970 | struct sched_domain_topology_level; |
5971 | ||
5972 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5973 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5974 | ||
e3589f6c PZ |
5975 | #define SDTL_OVERLAP 0x01 |
5976 | ||
eb7a74e6 | 5977 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5978 | sched_domain_init_f init; |
5979 | sched_domain_mask_f mask; | |
e3589f6c | 5980 | int flags; |
54ab4ff4 | 5981 | struct sd_data data; |
eb7a74e6 PZ |
5982 | }; |
5983 | ||
e3589f6c PZ |
5984 | static int |
5985 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5986 | { | |
5987 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5988 | const struct cpumask *span = sched_domain_span(sd); | |
5989 | struct cpumask *covered = sched_domains_tmpmask; | |
5990 | struct sd_data *sdd = sd->private; | |
5991 | struct sched_domain *child; | |
5992 | int i; | |
5993 | ||
5994 | cpumask_clear(covered); | |
5995 | ||
5996 | for_each_cpu(i, span) { | |
5997 | struct cpumask *sg_span; | |
5998 | ||
5999 | if (cpumask_test_cpu(i, covered)) | |
6000 | continue; | |
6001 | ||
6002 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
4d78a223 | 6003 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6004 | |
6005 | if (!sg) | |
6006 | goto fail; | |
6007 | ||
6008 | sg_span = sched_group_cpus(sg); | |
6009 | ||
6010 | child = *per_cpu_ptr(sdd->sd, i); | |
6011 | if (child->child) { | |
6012 | child = child->child; | |
6013 | cpumask_copy(sg_span, sched_domain_span(child)); | |
6014 | } else | |
6015 | cpumask_set_cpu(i, sg_span); | |
6016 | ||
6017 | cpumask_or(covered, covered, sg_span); | |
6018 | ||
6019 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
6020 | atomic_inc(&sg->sgp->ref); | |
6021 | ||
6022 | if (cpumask_test_cpu(cpu, sg_span)) | |
6023 | groups = sg; | |
6024 | ||
6025 | if (!first) | |
6026 | first = sg; | |
6027 | if (last) | |
6028 | last->next = sg; | |
6029 | last = sg; | |
6030 | last->next = first; | |
6031 | } | |
6032 | sd->groups = groups; | |
6033 | ||
6034 | return 0; | |
6035 | ||
6036 | fail: | |
6037 | free_sched_groups(first, 0); | |
6038 | ||
6039 | return -ENOMEM; | |
6040 | } | |
6041 | ||
dce840a0 | 6042 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6043 | { |
dce840a0 PZ |
6044 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6045 | struct sched_domain *child = sd->child; | |
1da177e4 | 6046 | |
dce840a0 PZ |
6047 | if (child) |
6048 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6049 | |
9c3f75cb | 6050 | if (sg) { |
dce840a0 | 6051 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 6052 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 6053 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 6054 | } |
dce840a0 PZ |
6055 | |
6056 | return cpu; | |
1e9f28fa | 6057 | } |
1e9f28fa | 6058 | |
01a08546 | 6059 | /* |
dce840a0 PZ |
6060 | * build_sched_groups will build a circular linked list of the groups |
6061 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6062 | * and ->cpu_power to 0. | |
e3589f6c PZ |
6063 | * |
6064 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6065 | */ |
e3589f6c PZ |
6066 | static int |
6067 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6068 | { |
dce840a0 PZ |
6069 | struct sched_group *first = NULL, *last = NULL; |
6070 | struct sd_data *sdd = sd->private; | |
6071 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6072 | struct cpumask *covered; |
dce840a0 | 6073 | int i; |
9c1cfda2 | 6074 | |
e3589f6c PZ |
6075 | get_group(cpu, sdd, &sd->groups); |
6076 | atomic_inc(&sd->groups->ref); | |
6077 | ||
6078 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
6079 | return 0; | |
6080 | ||
f96225fd PZ |
6081 | lockdep_assert_held(&sched_domains_mutex); |
6082 | covered = sched_domains_tmpmask; | |
6083 | ||
dce840a0 | 6084 | cpumask_clear(covered); |
6711cab4 | 6085 | |
dce840a0 PZ |
6086 | for_each_cpu(i, span) { |
6087 | struct sched_group *sg; | |
6088 | int group = get_group(i, sdd, &sg); | |
6089 | int j; | |
6711cab4 | 6090 | |
dce840a0 PZ |
6091 | if (cpumask_test_cpu(i, covered)) |
6092 | continue; | |
6711cab4 | 6093 | |
dce840a0 | 6094 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 6095 | sg->sgp->power = 0; |
0601a88d | 6096 | |
dce840a0 PZ |
6097 | for_each_cpu(j, span) { |
6098 | if (get_group(j, sdd, NULL) != group) | |
6099 | continue; | |
0601a88d | 6100 | |
dce840a0 PZ |
6101 | cpumask_set_cpu(j, covered); |
6102 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6103 | } | |
0601a88d | 6104 | |
dce840a0 PZ |
6105 | if (!first) |
6106 | first = sg; | |
6107 | if (last) | |
6108 | last->next = sg; | |
6109 | last = sg; | |
6110 | } | |
6111 | last->next = first; | |
e3589f6c PZ |
6112 | |
6113 | return 0; | |
0601a88d | 6114 | } |
51888ca2 | 6115 | |
89c4710e SS |
6116 | /* |
6117 | * Initialize sched groups cpu_power. | |
6118 | * | |
6119 | * cpu_power indicates the capacity of sched group, which is used while | |
6120 | * distributing the load between different sched groups in a sched domain. | |
6121 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6122 | * there are asymmetries in the topology. If there are asymmetries, group | |
6123 | * having more cpu_power will pickup more load compared to the group having | |
6124 | * less cpu_power. | |
89c4710e SS |
6125 | */ |
6126 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6127 | { | |
e3589f6c | 6128 | struct sched_group *sg = sd->groups; |
89c4710e | 6129 | |
e3589f6c PZ |
6130 | WARN_ON(!sd || !sg); |
6131 | ||
6132 | do { | |
6133 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6134 | sg = sg->next; | |
6135 | } while (sg != sd->groups); | |
89c4710e | 6136 | |
e3589f6c PZ |
6137 | if (cpu != group_first_cpu(sg)) |
6138 | return; | |
aae6d3dd | 6139 | |
d274cb30 | 6140 | update_group_power(sd, cpu); |
69e1e811 | 6141 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
6142 | } |
6143 | ||
029632fb PZ |
6144 | int __weak arch_sd_sibling_asym_packing(void) |
6145 | { | |
6146 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
6147 | } |
6148 | ||
7c16ec58 MT |
6149 | /* |
6150 | * Initializers for schedule domains | |
6151 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6152 | */ | |
6153 | ||
a5d8c348 IM |
6154 | #ifdef CONFIG_SCHED_DEBUG |
6155 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6156 | #else | |
6157 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6158 | #endif | |
6159 | ||
54ab4ff4 PZ |
6160 | #define SD_INIT_FUNC(type) \ |
6161 | static noinline struct sched_domain * \ | |
6162 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
6163 | { \ | |
6164 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
6165 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
6166 | SD_INIT_NAME(sd, type); \ |
6167 | sd->private = &tl->data; \ | |
6168 | return sd; \ | |
7c16ec58 MT |
6169 | } |
6170 | ||
6171 | SD_INIT_FUNC(CPU) | |
6172 | #ifdef CONFIG_NUMA | |
6173 | SD_INIT_FUNC(ALLNODES) | |
6174 | SD_INIT_FUNC(NODE) | |
6175 | #endif | |
6176 | #ifdef CONFIG_SCHED_SMT | |
6177 | SD_INIT_FUNC(SIBLING) | |
6178 | #endif | |
6179 | #ifdef CONFIG_SCHED_MC | |
6180 | SD_INIT_FUNC(MC) | |
6181 | #endif | |
01a08546 HC |
6182 | #ifdef CONFIG_SCHED_BOOK |
6183 | SD_INIT_FUNC(BOOK) | |
6184 | #endif | |
7c16ec58 | 6185 | |
1d3504fc | 6186 | static int default_relax_domain_level = -1; |
60495e77 | 6187 | int sched_domain_level_max; |
1d3504fc HS |
6188 | |
6189 | static int __init setup_relax_domain_level(char *str) | |
6190 | { | |
30e0e178 LZ |
6191 | unsigned long val; |
6192 | ||
6193 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 6194 | if (val < sched_domain_level_max) |
30e0e178 LZ |
6195 | default_relax_domain_level = val; |
6196 | ||
1d3504fc HS |
6197 | return 1; |
6198 | } | |
6199 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6200 | ||
6201 | static void set_domain_attribute(struct sched_domain *sd, | |
6202 | struct sched_domain_attr *attr) | |
6203 | { | |
6204 | int request; | |
6205 | ||
6206 | if (!attr || attr->relax_domain_level < 0) { | |
6207 | if (default_relax_domain_level < 0) | |
6208 | return; | |
6209 | else | |
6210 | request = default_relax_domain_level; | |
6211 | } else | |
6212 | request = attr->relax_domain_level; | |
6213 | if (request < sd->level) { | |
6214 | /* turn off idle balance on this domain */ | |
c88d5910 | 6215 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6216 | } else { |
6217 | /* turn on idle balance on this domain */ | |
c88d5910 | 6218 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6219 | } |
6220 | } | |
6221 | ||
54ab4ff4 PZ |
6222 | static void __sdt_free(const struct cpumask *cpu_map); |
6223 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6224 | ||
2109b99e AH |
6225 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6226 | const struct cpumask *cpu_map) | |
6227 | { | |
6228 | switch (what) { | |
2109b99e | 6229 | case sa_rootdomain: |
822ff793 PZ |
6230 | if (!atomic_read(&d->rd->refcount)) |
6231 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6232 | case sa_sd: |
6233 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6234 | case sa_sd_storage: |
54ab4ff4 | 6235 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6236 | case sa_none: |
6237 | break; | |
6238 | } | |
6239 | } | |
3404c8d9 | 6240 | |
2109b99e AH |
6241 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6242 | const struct cpumask *cpu_map) | |
6243 | { | |
dce840a0 PZ |
6244 | memset(d, 0, sizeof(*d)); |
6245 | ||
54ab4ff4 PZ |
6246 | if (__sdt_alloc(cpu_map)) |
6247 | return sa_sd_storage; | |
dce840a0 PZ |
6248 | d->sd = alloc_percpu(struct sched_domain *); |
6249 | if (!d->sd) | |
6250 | return sa_sd_storage; | |
2109b99e | 6251 | d->rd = alloc_rootdomain(); |
dce840a0 | 6252 | if (!d->rd) |
21d42ccf | 6253 | return sa_sd; |
2109b99e AH |
6254 | return sa_rootdomain; |
6255 | } | |
57d885fe | 6256 | |
dce840a0 PZ |
6257 | /* |
6258 | * NULL the sd_data elements we've used to build the sched_domain and | |
6259 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6260 | * will not free the data we're using. | |
6261 | */ | |
6262 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6263 | { | |
6264 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6265 | |
6266 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6267 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6268 | ||
e3589f6c | 6269 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6270 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6271 | |
6272 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6273 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6274 | } |
6275 | ||
2c402dc3 PZ |
6276 | #ifdef CONFIG_SCHED_SMT |
6277 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6278 | { |
2c402dc3 | 6279 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6280 | } |
2c402dc3 | 6281 | #endif |
7f4588f3 | 6282 | |
d069b916 PZ |
6283 | /* |
6284 | * Topology list, bottom-up. | |
6285 | */ | |
2c402dc3 | 6286 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6287 | #ifdef CONFIG_SCHED_SMT |
6288 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6289 | #endif |
1e9f28fa | 6290 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6291 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6292 | #endif |
d069b916 PZ |
6293 | #ifdef CONFIG_SCHED_BOOK |
6294 | { sd_init_BOOK, cpu_book_mask, }, | |
6295 | #endif | |
6296 | { sd_init_CPU, cpu_cpu_mask, }, | |
6297 | #ifdef CONFIG_NUMA | |
e3589f6c | 6298 | { sd_init_NODE, cpu_node_mask, SDTL_OVERLAP, }, |
d069b916 | 6299 | { sd_init_ALLNODES, cpu_allnodes_mask, }, |
1da177e4 | 6300 | #endif |
eb7a74e6 PZ |
6301 | { NULL, }, |
6302 | }; | |
6303 | ||
6304 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6305 | ||
54ab4ff4 PZ |
6306 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6307 | { | |
6308 | struct sched_domain_topology_level *tl; | |
6309 | int j; | |
6310 | ||
6311 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6312 | struct sd_data *sdd = &tl->data; | |
6313 | ||
6314 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6315 | if (!sdd->sd) | |
6316 | return -ENOMEM; | |
6317 | ||
6318 | sdd->sg = alloc_percpu(struct sched_group *); | |
6319 | if (!sdd->sg) | |
6320 | return -ENOMEM; | |
6321 | ||
9c3f75cb PZ |
6322 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6323 | if (!sdd->sgp) | |
6324 | return -ENOMEM; | |
6325 | ||
54ab4ff4 PZ |
6326 | for_each_cpu(j, cpu_map) { |
6327 | struct sched_domain *sd; | |
6328 | struct sched_group *sg; | |
9c3f75cb | 6329 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6330 | |
6331 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6332 | GFP_KERNEL, cpu_to_node(j)); | |
6333 | if (!sd) | |
6334 | return -ENOMEM; | |
6335 | ||
6336 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6337 | ||
6338 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6339 | GFP_KERNEL, cpu_to_node(j)); | |
6340 | if (!sg) | |
6341 | return -ENOMEM; | |
6342 | ||
6343 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
6344 | |
6345 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
6346 | GFP_KERNEL, cpu_to_node(j)); | |
6347 | if (!sgp) | |
6348 | return -ENOMEM; | |
6349 | ||
6350 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6351 | } |
6352 | } | |
6353 | ||
6354 | return 0; | |
6355 | } | |
6356 | ||
6357 | static void __sdt_free(const struct cpumask *cpu_map) | |
6358 | { | |
6359 | struct sched_domain_topology_level *tl; | |
6360 | int j; | |
6361 | ||
6362 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6363 | struct sd_data *sdd = &tl->data; | |
6364 | ||
6365 | for_each_cpu(j, cpu_map) { | |
e3589f6c PZ |
6366 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j); |
6367 | if (sd && (sd->flags & SD_OVERLAP)) | |
6368 | free_sched_groups(sd->groups, 0); | |
feff8fa0 | 6369 | kfree(*per_cpu_ptr(sdd->sd, j)); |
54ab4ff4 | 6370 | kfree(*per_cpu_ptr(sdd->sg, j)); |
9c3f75cb | 6371 | kfree(*per_cpu_ptr(sdd->sgp, j)); |
54ab4ff4 PZ |
6372 | } |
6373 | free_percpu(sdd->sd); | |
6374 | free_percpu(sdd->sg); | |
9c3f75cb | 6375 | free_percpu(sdd->sgp); |
54ab4ff4 PZ |
6376 | } |
6377 | } | |
6378 | ||
2c402dc3 PZ |
6379 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6380 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6381 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6382 | int cpu) |
6383 | { | |
54ab4ff4 | 6384 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6385 | if (!sd) |
d069b916 | 6386 | return child; |
2c402dc3 PZ |
6387 | |
6388 | set_domain_attribute(sd, attr); | |
6389 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
6390 | if (child) { |
6391 | sd->level = child->level + 1; | |
6392 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6393 | child->parent = sd; |
60495e77 | 6394 | } |
d069b916 | 6395 | sd->child = child; |
2c402dc3 PZ |
6396 | |
6397 | return sd; | |
6398 | } | |
6399 | ||
2109b99e AH |
6400 | /* |
6401 | * Build sched domains for a given set of cpus and attach the sched domains | |
6402 | * to the individual cpus | |
6403 | */ | |
dce840a0 PZ |
6404 | static int build_sched_domains(const struct cpumask *cpu_map, |
6405 | struct sched_domain_attr *attr) | |
2109b99e AH |
6406 | { |
6407 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6408 | struct sched_domain *sd; |
2109b99e | 6409 | struct s_data d; |
822ff793 | 6410 | int i, ret = -ENOMEM; |
9c1cfda2 | 6411 | |
2109b99e AH |
6412 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6413 | if (alloc_state != sa_rootdomain) | |
6414 | goto error; | |
9c1cfda2 | 6415 | |
dce840a0 | 6416 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6417 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6418 | struct sched_domain_topology_level *tl; |
6419 | ||
3bd65a80 | 6420 | sd = NULL; |
e3589f6c | 6421 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6422 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6423 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6424 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6425 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6426 | break; | |
e3589f6c | 6427 | } |
d274cb30 | 6428 | |
d069b916 PZ |
6429 | while (sd->child) |
6430 | sd = sd->child; | |
6431 | ||
21d42ccf | 6432 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6433 | } |
6434 | ||
6435 | /* Build the groups for the domains */ | |
6436 | for_each_cpu(i, cpu_map) { | |
6437 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6438 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6439 | if (sd->flags & SD_OVERLAP) { |
6440 | if (build_overlap_sched_groups(sd, i)) | |
6441 | goto error; | |
6442 | } else { | |
6443 | if (build_sched_groups(sd, i)) | |
6444 | goto error; | |
6445 | } | |
1cf51902 | 6446 | } |
a06dadbe | 6447 | } |
9c1cfda2 | 6448 | |
1da177e4 | 6449 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6450 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6451 | if (!cpumask_test_cpu(i, cpu_map)) | |
6452 | continue; | |
9c1cfda2 | 6453 | |
dce840a0 PZ |
6454 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6455 | claim_allocations(i, sd); | |
cd4ea6ae | 6456 | init_sched_groups_power(i, sd); |
dce840a0 | 6457 | } |
f712c0c7 | 6458 | } |
9c1cfda2 | 6459 | |
1da177e4 | 6460 | /* Attach the domains */ |
dce840a0 | 6461 | rcu_read_lock(); |
abcd083a | 6462 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6463 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6464 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6465 | } |
dce840a0 | 6466 | rcu_read_unlock(); |
51888ca2 | 6467 | |
822ff793 | 6468 | ret = 0; |
51888ca2 | 6469 | error: |
2109b99e | 6470 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6471 | return ret; |
1da177e4 | 6472 | } |
029190c5 | 6473 | |
acc3f5d7 | 6474 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6475 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6476 | static struct sched_domain_attr *dattr_cur; |
6477 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6478 | |
6479 | /* | |
6480 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6481 | * cpumask) fails, then fallback to a single sched domain, |
6482 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6483 | */ |
4212823f | 6484 | static cpumask_var_t fallback_doms; |
029190c5 | 6485 | |
ee79d1bd HC |
6486 | /* |
6487 | * arch_update_cpu_topology lets virtualized architectures update the | |
6488 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6489 | * or 0 if it stayed the same. | |
6490 | */ | |
6491 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6492 | { |
ee79d1bd | 6493 | return 0; |
22e52b07 HC |
6494 | } |
6495 | ||
acc3f5d7 RR |
6496 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6497 | { | |
6498 | int i; | |
6499 | cpumask_var_t *doms; | |
6500 | ||
6501 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6502 | if (!doms) | |
6503 | return NULL; | |
6504 | for (i = 0; i < ndoms; i++) { | |
6505 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6506 | free_sched_domains(doms, i); | |
6507 | return NULL; | |
6508 | } | |
6509 | } | |
6510 | return doms; | |
6511 | } | |
6512 | ||
6513 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6514 | { | |
6515 | unsigned int i; | |
6516 | for (i = 0; i < ndoms; i++) | |
6517 | free_cpumask_var(doms[i]); | |
6518 | kfree(doms); | |
6519 | } | |
6520 | ||
1a20ff27 | 6521 | /* |
41a2d6cf | 6522 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6523 | * For now this just excludes isolated cpus, but could be used to |
6524 | * exclude other special cases in the future. | |
1a20ff27 | 6525 | */ |
c4a8849a | 6526 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6527 | { |
7378547f MM |
6528 | int err; |
6529 | ||
22e52b07 | 6530 | arch_update_cpu_topology(); |
029190c5 | 6531 | ndoms_cur = 1; |
acc3f5d7 | 6532 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6533 | if (!doms_cur) |
acc3f5d7 RR |
6534 | doms_cur = &fallback_doms; |
6535 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 6536 | dattr_cur = NULL; |
dce840a0 | 6537 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6538 | register_sched_domain_sysctl(); |
7378547f MM |
6539 | |
6540 | return err; | |
1a20ff27 DG |
6541 | } |
6542 | ||
1a20ff27 DG |
6543 | /* |
6544 | * Detach sched domains from a group of cpus specified in cpu_map | |
6545 | * These cpus will now be attached to the NULL domain | |
6546 | */ | |
96f874e2 | 6547 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6548 | { |
6549 | int i; | |
6550 | ||
dce840a0 | 6551 | rcu_read_lock(); |
abcd083a | 6552 | for_each_cpu(i, cpu_map) |
57d885fe | 6553 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6554 | rcu_read_unlock(); |
1a20ff27 DG |
6555 | } |
6556 | ||
1d3504fc HS |
6557 | /* handle null as "default" */ |
6558 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6559 | struct sched_domain_attr *new, int idx_new) | |
6560 | { | |
6561 | struct sched_domain_attr tmp; | |
6562 | ||
6563 | /* fast path */ | |
6564 | if (!new && !cur) | |
6565 | return 1; | |
6566 | ||
6567 | tmp = SD_ATTR_INIT; | |
6568 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6569 | new ? (new + idx_new) : &tmp, | |
6570 | sizeof(struct sched_domain_attr)); | |
6571 | } | |
6572 | ||
029190c5 PJ |
6573 | /* |
6574 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6575 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6576 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6577 | * It destroys each deleted domain and builds each new domain. | |
6578 | * | |
acc3f5d7 | 6579 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6580 | * The masks don't intersect (don't overlap.) We should setup one |
6581 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6582 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6583 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6584 | * it as it is. | |
6585 | * | |
acc3f5d7 RR |
6586 | * The passed in 'doms_new' should be allocated using |
6587 | * alloc_sched_domains. This routine takes ownership of it and will | |
6588 | * free_sched_domains it when done with it. If the caller failed the | |
6589 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6590 | * and partition_sched_domains() will fallback to the single partition | |
6591 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6592 | * |
96f874e2 | 6593 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6594 | * ndoms_new == 0 is a special case for destroying existing domains, |
6595 | * and it will not create the default domain. | |
dfb512ec | 6596 | * |
029190c5 PJ |
6597 | * Call with hotplug lock held |
6598 | */ | |
acc3f5d7 | 6599 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6600 | struct sched_domain_attr *dattr_new) |
029190c5 | 6601 | { |
dfb512ec | 6602 | int i, j, n; |
d65bd5ec | 6603 | int new_topology; |
029190c5 | 6604 | |
712555ee | 6605 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6606 | |
7378547f MM |
6607 | /* always unregister in case we don't destroy any domains */ |
6608 | unregister_sched_domain_sysctl(); | |
6609 | ||
d65bd5ec HC |
6610 | /* Let architecture update cpu core mappings. */ |
6611 | new_topology = arch_update_cpu_topology(); | |
6612 | ||
dfb512ec | 6613 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6614 | |
6615 | /* Destroy deleted domains */ | |
6616 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6617 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6618 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6619 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6620 | goto match1; |
6621 | } | |
6622 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6623 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6624 | match1: |
6625 | ; | |
6626 | } | |
6627 | ||
e761b772 MK |
6628 | if (doms_new == NULL) { |
6629 | ndoms_cur = 0; | |
acc3f5d7 | 6630 | doms_new = &fallback_doms; |
6ad4c188 | 6631 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6632 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6633 | } |
6634 | ||
029190c5 PJ |
6635 | /* Build new domains */ |
6636 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6637 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6638 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6639 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6640 | goto match2; |
6641 | } | |
6642 | /* no match - add a new doms_new */ | |
dce840a0 | 6643 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6644 | match2: |
6645 | ; | |
6646 | } | |
6647 | ||
6648 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6649 | if (doms_cur != &fallback_doms) |
6650 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6651 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6652 | doms_cur = doms_new; |
1d3504fc | 6653 | dattr_cur = dattr_new; |
029190c5 | 6654 | ndoms_cur = ndoms_new; |
7378547f MM |
6655 | |
6656 | register_sched_domain_sysctl(); | |
a1835615 | 6657 | |
712555ee | 6658 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6659 | } |
6660 | ||
5c45bf27 | 6661 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 6662 | static void reinit_sched_domains(void) |
5c45bf27 | 6663 | { |
95402b38 | 6664 | get_online_cpus(); |
dfb512ec MK |
6665 | |
6666 | /* Destroy domains first to force the rebuild */ | |
6667 | partition_sched_domains(0, NULL, NULL); | |
6668 | ||
e761b772 | 6669 | rebuild_sched_domains(); |
95402b38 | 6670 | put_online_cpus(); |
5c45bf27 SS |
6671 | } |
6672 | ||
6673 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6674 | { | |
afb8a9b7 | 6675 | unsigned int level = 0; |
5c45bf27 | 6676 | |
afb8a9b7 GS |
6677 | if (sscanf(buf, "%u", &level) != 1) |
6678 | return -EINVAL; | |
6679 | ||
6680 | /* | |
6681 | * level is always be positive so don't check for | |
6682 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
6683 | * What happens on 0 or 1 byte write, | |
6684 | * need to check for count as well? | |
6685 | */ | |
6686 | ||
6687 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
6688 | return -EINVAL; |
6689 | ||
6690 | if (smt) | |
afb8a9b7 | 6691 | sched_smt_power_savings = level; |
5c45bf27 | 6692 | else |
afb8a9b7 | 6693 | sched_mc_power_savings = level; |
5c45bf27 | 6694 | |
c4a8849a | 6695 | reinit_sched_domains(); |
5c45bf27 | 6696 | |
c70f22d2 | 6697 | return count; |
5c45bf27 SS |
6698 | } |
6699 | ||
5c45bf27 | 6700 | #ifdef CONFIG_SCHED_MC |
8a25a2fd KS |
6701 | static ssize_t sched_mc_power_savings_show(struct device *dev, |
6702 | struct device_attribute *attr, | |
6703 | char *buf) | |
5c45bf27 | 6704 | { |
8a25a2fd | 6705 | return sprintf(buf, "%u\n", sched_mc_power_savings); |
5c45bf27 | 6706 | } |
8a25a2fd KS |
6707 | static ssize_t sched_mc_power_savings_store(struct device *dev, |
6708 | struct device_attribute *attr, | |
48f24c4d | 6709 | const char *buf, size_t count) |
5c45bf27 SS |
6710 | { |
6711 | return sched_power_savings_store(buf, count, 0); | |
6712 | } | |
8a25a2fd KS |
6713 | static DEVICE_ATTR(sched_mc_power_savings, 0644, |
6714 | sched_mc_power_savings_show, | |
6715 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6716 | #endif |
6717 | ||
6718 | #ifdef CONFIG_SCHED_SMT | |
8a25a2fd KS |
6719 | static ssize_t sched_smt_power_savings_show(struct device *dev, |
6720 | struct device_attribute *attr, | |
6721 | char *buf) | |
5c45bf27 | 6722 | { |
8a25a2fd | 6723 | return sprintf(buf, "%u\n", sched_smt_power_savings); |
5c45bf27 | 6724 | } |
8a25a2fd KS |
6725 | static ssize_t sched_smt_power_savings_store(struct device *dev, |
6726 | struct device_attribute *attr, | |
48f24c4d | 6727 | const char *buf, size_t count) |
5c45bf27 SS |
6728 | { |
6729 | return sched_power_savings_store(buf, count, 1); | |
6730 | } | |
8a25a2fd | 6731 | static DEVICE_ATTR(sched_smt_power_savings, 0644, |
f718cd4a | 6732 | sched_smt_power_savings_show, |
6707de00 AB |
6733 | sched_smt_power_savings_store); |
6734 | #endif | |
6735 | ||
8a25a2fd | 6736 | int __init sched_create_sysfs_power_savings_entries(struct device *dev) |
6707de00 AB |
6737 | { |
6738 | int err = 0; | |
6739 | ||
6740 | #ifdef CONFIG_SCHED_SMT | |
6741 | if (smt_capable()) | |
8a25a2fd | 6742 | err = device_create_file(dev, &dev_attr_sched_smt_power_savings); |
6707de00 AB |
6743 | #endif |
6744 | #ifdef CONFIG_SCHED_MC | |
6745 | if (!err && mc_capable()) | |
8a25a2fd | 6746 | err = device_create_file(dev, &dev_attr_sched_mc_power_savings); |
6707de00 AB |
6747 | #endif |
6748 | return err; | |
6749 | } | |
6d6bc0ad | 6750 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 6751 | |
1da177e4 | 6752 | /* |
3a101d05 TH |
6753 | * Update cpusets according to cpu_active mask. If cpusets are |
6754 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6755 | * around partition_sched_domains(). | |
1da177e4 | 6756 | */ |
0b2e918a TH |
6757 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6758 | void *hcpu) | |
e761b772 | 6759 | { |
3a101d05 | 6760 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 6761 | case CPU_ONLINE: |
6ad4c188 | 6762 | case CPU_DOWN_FAILED: |
3a101d05 | 6763 | cpuset_update_active_cpus(); |
e761b772 | 6764 | return NOTIFY_OK; |
3a101d05 TH |
6765 | default: |
6766 | return NOTIFY_DONE; | |
6767 | } | |
6768 | } | |
e761b772 | 6769 | |
0b2e918a TH |
6770 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6771 | void *hcpu) | |
3a101d05 TH |
6772 | { |
6773 | switch (action & ~CPU_TASKS_FROZEN) { | |
6774 | case CPU_DOWN_PREPARE: | |
6775 | cpuset_update_active_cpus(); | |
6776 | return NOTIFY_OK; | |
e761b772 MK |
6777 | default: |
6778 | return NOTIFY_DONE; | |
6779 | } | |
6780 | } | |
e761b772 | 6781 | |
1da177e4 LT |
6782 | void __init sched_init_smp(void) |
6783 | { | |
dcc30a35 RR |
6784 | cpumask_var_t non_isolated_cpus; |
6785 | ||
6786 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6787 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6788 | |
95402b38 | 6789 | get_online_cpus(); |
712555ee | 6790 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6791 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6792 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6793 | if (cpumask_empty(non_isolated_cpus)) | |
6794 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6795 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6796 | put_online_cpus(); |
e761b772 | 6797 | |
3a101d05 TH |
6798 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6799 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6800 | |
6801 | /* RT runtime code needs to handle some hotplug events */ | |
6802 | hotcpu_notifier(update_runtime, 0); | |
6803 | ||
b328ca18 | 6804 | init_hrtick(); |
5c1e1767 NP |
6805 | |
6806 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6807 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6808 | BUG(); |
19978ca6 | 6809 | sched_init_granularity(); |
dcc30a35 | 6810 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6811 | |
0e3900e6 | 6812 | init_sched_rt_class(); |
1da177e4 LT |
6813 | } |
6814 | #else | |
6815 | void __init sched_init_smp(void) | |
6816 | { | |
19978ca6 | 6817 | sched_init_granularity(); |
1da177e4 LT |
6818 | } |
6819 | #endif /* CONFIG_SMP */ | |
6820 | ||
cd1bb94b AB |
6821 | const_debug unsigned int sysctl_timer_migration = 1; |
6822 | ||
1da177e4 LT |
6823 | int in_sched_functions(unsigned long addr) |
6824 | { | |
1da177e4 LT |
6825 | return in_lock_functions(addr) || |
6826 | (addr >= (unsigned long)__sched_text_start | |
6827 | && addr < (unsigned long)__sched_text_end); | |
6828 | } | |
6829 | ||
029632fb PZ |
6830 | #ifdef CONFIG_CGROUP_SCHED |
6831 | struct task_group root_task_group; | |
052f1dc7 | 6832 | #endif |
6f505b16 | 6833 | |
029632fb | 6834 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6835 | |
1da177e4 LT |
6836 | void __init sched_init(void) |
6837 | { | |
dd41f596 | 6838 | int i, j; |
434d53b0 MT |
6839 | unsigned long alloc_size = 0, ptr; |
6840 | ||
6841 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6842 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6843 | #endif | |
6844 | #ifdef CONFIG_RT_GROUP_SCHED | |
6845 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6846 | #endif |
df7c8e84 | 6847 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6848 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6849 | #endif |
434d53b0 | 6850 | if (alloc_size) { |
36b7b6d4 | 6851 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6852 | |
6853 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6854 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6855 | ptr += nr_cpu_ids * sizeof(void **); |
6856 | ||
07e06b01 | 6857 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6858 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6859 | |
6d6bc0ad | 6860 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6861 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6862 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6863 | ptr += nr_cpu_ids * sizeof(void **); |
6864 | ||
07e06b01 | 6865 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6866 | ptr += nr_cpu_ids * sizeof(void **); |
6867 | ||
6d6bc0ad | 6868 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6869 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6870 | for_each_possible_cpu(i) { | |
6871 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
6872 | ptr += cpumask_size(); | |
6873 | } | |
6874 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6875 | } |
dd41f596 | 6876 | |
57d885fe GH |
6877 | #ifdef CONFIG_SMP |
6878 | init_defrootdomain(); | |
6879 | #endif | |
6880 | ||
d0b27fa7 PZ |
6881 | init_rt_bandwidth(&def_rt_bandwidth, |
6882 | global_rt_period(), global_rt_runtime()); | |
6883 | ||
6884 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6885 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6886 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6887 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6888 | |
7c941438 | 6889 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6890 | list_add(&root_task_group.list, &task_groups); |
6891 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6892 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6893 | autogroup_init(&init_task); |
54c707e9 | 6894 | |
7c941438 | 6895 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6896 | |
54c707e9 GC |
6897 | #ifdef CONFIG_CGROUP_CPUACCT |
6898 | root_cpuacct.cpustat = &kernel_cpustat; | |
6899 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
6900 | /* Too early, not expected to fail */ | |
6901 | BUG_ON(!root_cpuacct.cpuusage); | |
6902 | #endif | |
0a945022 | 6903 | for_each_possible_cpu(i) { |
70b97a7f | 6904 | struct rq *rq; |
1da177e4 LT |
6905 | |
6906 | rq = cpu_rq(i); | |
05fa785c | 6907 | raw_spin_lock_init(&rq->lock); |
7897986b | 6908 | rq->nr_running = 0; |
dce48a84 TG |
6909 | rq->calc_load_active = 0; |
6910 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6911 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6912 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6913 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6914 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6915 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6916 | /* |
07e06b01 | 6917 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6918 | * |
6919 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6920 | * gets 100% of the cpu resources in the system. This overall | |
6921 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6922 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6923 | * based on each entity's (task or task-group's) weight |
6924 | * (se->load.weight). | |
6925 | * | |
07e06b01 | 6926 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6927 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6928 | * then A0's share of the cpu resource is: | |
6929 | * | |
0d905bca | 6930 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6931 | * |
07e06b01 YZ |
6932 | * We achieve this by letting root_task_group's tasks sit |
6933 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6934 | */ |
ab84d31e | 6935 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6936 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6937 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6938 | ||
6939 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6940 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6941 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6942 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6943 | #endif |
1da177e4 | 6944 | |
dd41f596 IM |
6945 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6946 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6947 | |
6948 | rq->last_load_update_tick = jiffies; | |
6949 | ||
1da177e4 | 6950 | #ifdef CONFIG_SMP |
41c7ce9a | 6951 | rq->sd = NULL; |
57d885fe | 6952 | rq->rd = NULL; |
1399fa78 | 6953 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6954 | rq->post_schedule = 0; |
1da177e4 | 6955 | rq->active_balance = 0; |
dd41f596 | 6956 | rq->next_balance = jiffies; |
1da177e4 | 6957 | rq->push_cpu = 0; |
0a2966b4 | 6958 | rq->cpu = i; |
1f11eb6a | 6959 | rq->online = 0; |
eae0c9df MG |
6960 | rq->idle_stamp = 0; |
6961 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 6962 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 6963 | #ifdef CONFIG_NO_HZ |
1c792db7 | 6964 | rq->nohz_flags = 0; |
83cd4fe2 | 6965 | #endif |
1da177e4 | 6966 | #endif |
8f4d37ec | 6967 | init_rq_hrtick(rq); |
1da177e4 | 6968 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6969 | } |
6970 | ||
2dd73a4f | 6971 | set_load_weight(&init_task); |
b50f60ce | 6972 | |
e107be36 AK |
6973 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6974 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6975 | #endif | |
6976 | ||
b50f60ce | 6977 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6978 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6979 | #endif |
6980 | ||
1da177e4 LT |
6981 | /* |
6982 | * The boot idle thread does lazy MMU switching as well: | |
6983 | */ | |
6984 | atomic_inc(&init_mm.mm_count); | |
6985 | enter_lazy_tlb(&init_mm, current); | |
6986 | ||
6987 | /* | |
6988 | * Make us the idle thread. Technically, schedule() should not be | |
6989 | * called from this thread, however somewhere below it might be, | |
6990 | * but because we are the idle thread, we just pick up running again | |
6991 | * when this runqueue becomes "idle". | |
6992 | */ | |
6993 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6994 | |
6995 | calc_load_update = jiffies + LOAD_FREQ; | |
6996 | ||
dd41f596 IM |
6997 | /* |
6998 | * During early bootup we pretend to be a normal task: | |
6999 | */ | |
7000 | current->sched_class = &fair_sched_class; | |
6892b75e | 7001 | |
bf4d83f6 | 7002 | #ifdef CONFIG_SMP |
4cb98839 | 7003 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7004 | /* May be allocated at isolcpus cmdline parse time */ |
7005 | if (cpu_isolated_map == NULL) | |
7006 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
029632fb PZ |
7007 | #endif |
7008 | init_sched_fair_class(); | |
6a7b3dc3 | 7009 | |
6892b75e | 7010 | scheduler_running = 1; |
1da177e4 LT |
7011 | } |
7012 | ||
d902db1e | 7013 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7014 | static inline int preempt_count_equals(int preempt_offset) |
7015 | { | |
234da7bc | 7016 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7017 | |
4ba8216c | 7018 | return (nested == preempt_offset); |
e4aafea2 FW |
7019 | } |
7020 | ||
d894837f | 7021 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7022 | { |
1da177e4 LT |
7023 | static unsigned long prev_jiffy; /* ratelimiting */ |
7024 | ||
b3fbab05 | 7025 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7026 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7027 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7028 | return; |
7029 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7030 | return; | |
7031 | prev_jiffy = jiffies; | |
7032 | ||
3df0fc5b PZ |
7033 | printk(KERN_ERR |
7034 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7035 | file, line); | |
7036 | printk(KERN_ERR | |
7037 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7038 | in_atomic(), irqs_disabled(), | |
7039 | current->pid, current->comm); | |
aef745fc IM |
7040 | |
7041 | debug_show_held_locks(current); | |
7042 | if (irqs_disabled()) | |
7043 | print_irqtrace_events(current); | |
7044 | dump_stack(); | |
1da177e4 LT |
7045 | } |
7046 | EXPORT_SYMBOL(__might_sleep); | |
7047 | #endif | |
7048 | ||
7049 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7050 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7051 | { | |
da7a735e PZ |
7052 | const struct sched_class *prev_class = p->sched_class; |
7053 | int old_prio = p->prio; | |
3a5e4dc1 | 7054 | int on_rq; |
3e51f33f | 7055 | |
fd2f4419 | 7056 | on_rq = p->on_rq; |
3a5e4dc1 | 7057 | if (on_rq) |
4ca9b72b | 7058 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7059 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7060 | if (on_rq) { | |
4ca9b72b | 7061 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7062 | resched_task(rq->curr); |
7063 | } | |
da7a735e PZ |
7064 | |
7065 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7066 | } |
7067 | ||
1da177e4 LT |
7068 | void normalize_rt_tasks(void) |
7069 | { | |
a0f98a1c | 7070 | struct task_struct *g, *p; |
1da177e4 | 7071 | unsigned long flags; |
70b97a7f | 7072 | struct rq *rq; |
1da177e4 | 7073 | |
4cf5d77a | 7074 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7075 | do_each_thread(g, p) { |
178be793 IM |
7076 | /* |
7077 | * Only normalize user tasks: | |
7078 | */ | |
7079 | if (!p->mm) | |
7080 | continue; | |
7081 | ||
6cfb0d5d | 7082 | p->se.exec_start = 0; |
6cfb0d5d | 7083 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7084 | p->se.statistics.wait_start = 0; |
7085 | p->se.statistics.sleep_start = 0; | |
7086 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7087 | #endif |
dd41f596 IM |
7088 | |
7089 | if (!rt_task(p)) { | |
7090 | /* | |
7091 | * Renice negative nice level userspace | |
7092 | * tasks back to 0: | |
7093 | */ | |
7094 | if (TASK_NICE(p) < 0 && p->mm) | |
7095 | set_user_nice(p, 0); | |
1da177e4 | 7096 | continue; |
dd41f596 | 7097 | } |
1da177e4 | 7098 | |
1d615482 | 7099 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7100 | rq = __task_rq_lock(p); |
1da177e4 | 7101 | |
178be793 | 7102 | normalize_task(rq, p); |
3a5e4dc1 | 7103 | |
b29739f9 | 7104 | __task_rq_unlock(rq); |
1d615482 | 7105 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7106 | } while_each_thread(g, p); |
7107 | ||
4cf5d77a | 7108 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7109 | } |
7110 | ||
7111 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7112 | |
67fc4e0c | 7113 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7114 | /* |
67fc4e0c | 7115 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7116 | * |
7117 | * They can only be called when the whole system has been | |
7118 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7119 | * activity can take place. Using them for anything else would | |
7120 | * be a serious bug, and as a result, they aren't even visible | |
7121 | * under any other configuration. | |
7122 | */ | |
7123 | ||
7124 | /** | |
7125 | * curr_task - return the current task for a given cpu. | |
7126 | * @cpu: the processor in question. | |
7127 | * | |
7128 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7129 | */ | |
36c8b586 | 7130 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7131 | { |
7132 | return cpu_curr(cpu); | |
7133 | } | |
7134 | ||
67fc4e0c JW |
7135 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7136 | ||
7137 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7138 | /** |
7139 | * set_curr_task - set the current task for a given cpu. | |
7140 | * @cpu: the processor in question. | |
7141 | * @p: the task pointer to set. | |
7142 | * | |
7143 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7144 | * are serviced on a separate stack. It allows the architecture to switch the |
7145 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7146 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7147 | * and caller must save the original value of the current task (see | |
7148 | * curr_task() above) and restore that value before reenabling interrupts and | |
7149 | * re-starting the system. | |
7150 | * | |
7151 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7152 | */ | |
36c8b586 | 7153 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7154 | { |
7155 | cpu_curr(cpu) = p; | |
7156 | } | |
7157 | ||
7158 | #endif | |
29f59db3 | 7159 | |
7c941438 | 7160 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7161 | /* task_group_lock serializes the addition/removal of task groups */ |
7162 | static DEFINE_SPINLOCK(task_group_lock); | |
7163 | ||
bccbe08a PZ |
7164 | static void free_sched_group(struct task_group *tg) |
7165 | { | |
7166 | free_fair_sched_group(tg); | |
7167 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7168 | autogroup_free(tg); |
bccbe08a PZ |
7169 | kfree(tg); |
7170 | } | |
7171 | ||
7172 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7173 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7174 | { |
7175 | struct task_group *tg; | |
7176 | unsigned long flags; | |
bccbe08a PZ |
7177 | |
7178 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7179 | if (!tg) | |
7180 | return ERR_PTR(-ENOMEM); | |
7181 | ||
ec7dc8ac | 7182 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7183 | goto err; |
7184 | ||
ec7dc8ac | 7185 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7186 | goto err; |
7187 | ||
8ed36996 | 7188 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7189 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7190 | |
7191 | WARN_ON(!parent); /* root should already exist */ | |
7192 | ||
7193 | tg->parent = parent; | |
f473aa5e | 7194 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7195 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7196 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7197 | |
9b5b7751 | 7198 | return tg; |
29f59db3 SV |
7199 | |
7200 | err: | |
6f505b16 | 7201 | free_sched_group(tg); |
29f59db3 SV |
7202 | return ERR_PTR(-ENOMEM); |
7203 | } | |
7204 | ||
9b5b7751 | 7205 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7206 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7207 | { |
29f59db3 | 7208 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7209 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7210 | } |
7211 | ||
9b5b7751 | 7212 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7213 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7214 | { |
8ed36996 | 7215 | unsigned long flags; |
9b5b7751 | 7216 | int i; |
29f59db3 | 7217 | |
3d4b47b4 PZ |
7218 | /* end participation in shares distribution */ |
7219 | for_each_possible_cpu(i) | |
bccbe08a | 7220 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7221 | |
7222 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7223 | list_del_rcu(&tg->list); |
f473aa5e | 7224 | list_del_rcu(&tg->siblings); |
8ed36996 | 7225 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7226 | |
9b5b7751 | 7227 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7228 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7229 | } |
7230 | ||
9b5b7751 | 7231 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7232 | * The caller of this function should have put the task in its new group |
7233 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7234 | * reflect its new group. | |
9b5b7751 SV |
7235 | */ |
7236 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7237 | { |
7238 | int on_rq, running; | |
7239 | unsigned long flags; | |
7240 | struct rq *rq; | |
7241 | ||
7242 | rq = task_rq_lock(tsk, &flags); | |
7243 | ||
051a1d1a | 7244 | running = task_current(rq, tsk); |
fd2f4419 | 7245 | on_rq = tsk->on_rq; |
29f59db3 | 7246 | |
0e1f3483 | 7247 | if (on_rq) |
29f59db3 | 7248 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7249 | if (unlikely(running)) |
7250 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7251 | |
810b3817 | 7252 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7253 | if (tsk->sched_class->task_move_group) |
7254 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7255 | else | |
810b3817 | 7256 | #endif |
b2b5ce02 | 7257 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7258 | |
0e1f3483 HS |
7259 | if (unlikely(running)) |
7260 | tsk->sched_class->set_curr_task(rq); | |
7261 | if (on_rq) | |
371fd7e7 | 7262 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7263 | |
0122ec5b | 7264 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7265 | } |
7c941438 | 7266 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7267 | |
a790de99 | 7268 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7269 | static unsigned long to_ratio(u64 period, u64 runtime) |
7270 | { | |
7271 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7272 | return 1ULL << 20; |
9f0c1e56 | 7273 | |
9a7e0b18 | 7274 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7275 | } |
a790de99 PT |
7276 | #endif |
7277 | ||
7278 | #ifdef CONFIG_RT_GROUP_SCHED | |
7279 | /* | |
7280 | * Ensure that the real time constraints are schedulable. | |
7281 | */ | |
7282 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7283 | |
9a7e0b18 PZ |
7284 | /* Must be called with tasklist_lock held */ |
7285 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7286 | { |
9a7e0b18 | 7287 | struct task_struct *g, *p; |
b40b2e8e | 7288 | |
9a7e0b18 | 7289 | do_each_thread(g, p) { |
029632fb | 7290 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7291 | return 1; |
7292 | } while_each_thread(g, p); | |
b40b2e8e | 7293 | |
9a7e0b18 PZ |
7294 | return 0; |
7295 | } | |
b40b2e8e | 7296 | |
9a7e0b18 PZ |
7297 | struct rt_schedulable_data { |
7298 | struct task_group *tg; | |
7299 | u64 rt_period; | |
7300 | u64 rt_runtime; | |
7301 | }; | |
b40b2e8e | 7302 | |
a790de99 | 7303 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7304 | { |
7305 | struct rt_schedulable_data *d = data; | |
7306 | struct task_group *child; | |
7307 | unsigned long total, sum = 0; | |
7308 | u64 period, runtime; | |
b40b2e8e | 7309 | |
9a7e0b18 PZ |
7310 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7311 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7312 | |
9a7e0b18 PZ |
7313 | if (tg == d->tg) { |
7314 | period = d->rt_period; | |
7315 | runtime = d->rt_runtime; | |
b40b2e8e | 7316 | } |
b40b2e8e | 7317 | |
4653f803 PZ |
7318 | /* |
7319 | * Cannot have more runtime than the period. | |
7320 | */ | |
7321 | if (runtime > period && runtime != RUNTIME_INF) | |
7322 | return -EINVAL; | |
6f505b16 | 7323 | |
4653f803 PZ |
7324 | /* |
7325 | * Ensure we don't starve existing RT tasks. | |
7326 | */ | |
9a7e0b18 PZ |
7327 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7328 | return -EBUSY; | |
6f505b16 | 7329 | |
9a7e0b18 | 7330 | total = to_ratio(period, runtime); |
6f505b16 | 7331 | |
4653f803 PZ |
7332 | /* |
7333 | * Nobody can have more than the global setting allows. | |
7334 | */ | |
7335 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7336 | return -EINVAL; | |
6f505b16 | 7337 | |
4653f803 PZ |
7338 | /* |
7339 | * The sum of our children's runtime should not exceed our own. | |
7340 | */ | |
9a7e0b18 PZ |
7341 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7342 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7343 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7344 | |
9a7e0b18 PZ |
7345 | if (child == d->tg) { |
7346 | period = d->rt_period; | |
7347 | runtime = d->rt_runtime; | |
7348 | } | |
6f505b16 | 7349 | |
9a7e0b18 | 7350 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7351 | } |
6f505b16 | 7352 | |
9a7e0b18 PZ |
7353 | if (sum > total) |
7354 | return -EINVAL; | |
7355 | ||
7356 | return 0; | |
6f505b16 PZ |
7357 | } |
7358 | ||
9a7e0b18 | 7359 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7360 | { |
8277434e PT |
7361 | int ret; |
7362 | ||
9a7e0b18 PZ |
7363 | struct rt_schedulable_data data = { |
7364 | .tg = tg, | |
7365 | .rt_period = period, | |
7366 | .rt_runtime = runtime, | |
7367 | }; | |
7368 | ||
8277434e PT |
7369 | rcu_read_lock(); |
7370 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7371 | rcu_read_unlock(); | |
7372 | ||
7373 | return ret; | |
521f1a24 DG |
7374 | } |
7375 | ||
ab84d31e | 7376 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7377 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7378 | { |
ac086bc2 | 7379 | int i, err = 0; |
9f0c1e56 | 7380 | |
9f0c1e56 | 7381 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7382 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7383 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7384 | if (err) | |
9f0c1e56 | 7385 | goto unlock; |
ac086bc2 | 7386 | |
0986b11b | 7387 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7388 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7389 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7390 | |
7391 | for_each_possible_cpu(i) { | |
7392 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7393 | ||
0986b11b | 7394 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7395 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7396 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7397 | } |
0986b11b | 7398 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7399 | unlock: |
521f1a24 | 7400 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7401 | mutex_unlock(&rt_constraints_mutex); |
7402 | ||
7403 | return err; | |
6f505b16 PZ |
7404 | } |
7405 | ||
d0b27fa7 PZ |
7406 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7407 | { | |
7408 | u64 rt_runtime, rt_period; | |
7409 | ||
7410 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7411 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7412 | if (rt_runtime_us < 0) | |
7413 | rt_runtime = RUNTIME_INF; | |
7414 | ||
ab84d31e | 7415 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7416 | } |
7417 | ||
9f0c1e56 PZ |
7418 | long sched_group_rt_runtime(struct task_group *tg) |
7419 | { | |
7420 | u64 rt_runtime_us; | |
7421 | ||
d0b27fa7 | 7422 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7423 | return -1; |
7424 | ||
d0b27fa7 | 7425 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7426 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7427 | return rt_runtime_us; | |
7428 | } | |
d0b27fa7 PZ |
7429 | |
7430 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7431 | { | |
7432 | u64 rt_runtime, rt_period; | |
7433 | ||
7434 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7435 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7436 | ||
619b0488 R |
7437 | if (rt_period == 0) |
7438 | return -EINVAL; | |
7439 | ||
ab84d31e | 7440 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7441 | } |
7442 | ||
7443 | long sched_group_rt_period(struct task_group *tg) | |
7444 | { | |
7445 | u64 rt_period_us; | |
7446 | ||
7447 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7448 | do_div(rt_period_us, NSEC_PER_USEC); | |
7449 | return rt_period_us; | |
7450 | } | |
7451 | ||
7452 | static int sched_rt_global_constraints(void) | |
7453 | { | |
4653f803 | 7454 | u64 runtime, period; |
d0b27fa7 PZ |
7455 | int ret = 0; |
7456 | ||
ec5d4989 HS |
7457 | if (sysctl_sched_rt_period <= 0) |
7458 | return -EINVAL; | |
7459 | ||
4653f803 PZ |
7460 | runtime = global_rt_runtime(); |
7461 | period = global_rt_period(); | |
7462 | ||
7463 | /* | |
7464 | * Sanity check on the sysctl variables. | |
7465 | */ | |
7466 | if (runtime > period && runtime != RUNTIME_INF) | |
7467 | return -EINVAL; | |
10b612f4 | 7468 | |
d0b27fa7 | 7469 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7470 | read_lock(&tasklist_lock); |
4653f803 | 7471 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7472 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7473 | mutex_unlock(&rt_constraints_mutex); |
7474 | ||
7475 | return ret; | |
7476 | } | |
54e99124 DG |
7477 | |
7478 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7479 | { | |
7480 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7481 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7482 | return 0; | |
7483 | ||
7484 | return 1; | |
7485 | } | |
7486 | ||
6d6bc0ad | 7487 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7488 | static int sched_rt_global_constraints(void) |
7489 | { | |
ac086bc2 PZ |
7490 | unsigned long flags; |
7491 | int i; | |
7492 | ||
ec5d4989 HS |
7493 | if (sysctl_sched_rt_period <= 0) |
7494 | return -EINVAL; | |
7495 | ||
60aa605d PZ |
7496 | /* |
7497 | * There's always some RT tasks in the root group | |
7498 | * -- migration, kstopmachine etc.. | |
7499 | */ | |
7500 | if (sysctl_sched_rt_runtime == 0) | |
7501 | return -EBUSY; | |
7502 | ||
0986b11b | 7503 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7504 | for_each_possible_cpu(i) { |
7505 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7506 | ||
0986b11b | 7507 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7508 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7509 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7510 | } |
0986b11b | 7511 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7512 | |
d0b27fa7 PZ |
7513 | return 0; |
7514 | } | |
6d6bc0ad | 7515 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7516 | |
7517 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7518 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7519 | loff_t *ppos) |
7520 | { | |
7521 | int ret; | |
7522 | int old_period, old_runtime; | |
7523 | static DEFINE_MUTEX(mutex); | |
7524 | ||
7525 | mutex_lock(&mutex); | |
7526 | old_period = sysctl_sched_rt_period; | |
7527 | old_runtime = sysctl_sched_rt_runtime; | |
7528 | ||
8d65af78 | 7529 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7530 | |
7531 | if (!ret && write) { | |
7532 | ret = sched_rt_global_constraints(); | |
7533 | if (ret) { | |
7534 | sysctl_sched_rt_period = old_period; | |
7535 | sysctl_sched_rt_runtime = old_runtime; | |
7536 | } else { | |
7537 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7538 | def_rt_bandwidth.rt_period = | |
7539 | ns_to_ktime(global_rt_period()); | |
7540 | } | |
7541 | } | |
7542 | mutex_unlock(&mutex); | |
7543 | ||
7544 | return ret; | |
7545 | } | |
68318b8e | 7546 | |
052f1dc7 | 7547 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7548 | |
7549 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7550 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7551 | { |
2b01dfe3 PM |
7552 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7553 | struct task_group, css); | |
68318b8e SV |
7554 | } |
7555 | ||
7556 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7557 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 7558 | { |
ec7dc8ac | 7559 | struct task_group *tg, *parent; |
68318b8e | 7560 | |
2b01dfe3 | 7561 | if (!cgrp->parent) { |
68318b8e | 7562 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7563 | return &root_task_group.css; |
68318b8e SV |
7564 | } |
7565 | ||
ec7dc8ac DG |
7566 | parent = cgroup_tg(cgrp->parent); |
7567 | tg = sched_create_group(parent); | |
68318b8e SV |
7568 | if (IS_ERR(tg)) |
7569 | return ERR_PTR(-ENOMEM); | |
7570 | ||
68318b8e SV |
7571 | return &tg->css; |
7572 | } | |
7573 | ||
41a2d6cf IM |
7574 | static void |
7575 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7576 | { |
2b01dfe3 | 7577 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7578 | |
7579 | sched_destroy_group(tg); | |
7580 | } | |
7581 | ||
bb9d97b6 TH |
7582 | static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7583 | struct cgroup_taskset *tset) | |
68318b8e | 7584 | { |
bb9d97b6 TH |
7585 | struct task_struct *task; |
7586 | ||
7587 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7588 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7589 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7590 | return -EINVAL; | |
b68aa230 | 7591 | #else |
bb9d97b6 TH |
7592 | /* We don't support RT-tasks being in separate groups */ |
7593 | if (task->sched_class != &fair_sched_class) | |
7594 | return -EINVAL; | |
b68aa230 | 7595 | #endif |
bb9d97b6 | 7596 | } |
be367d09 BB |
7597 | return 0; |
7598 | } | |
68318b8e | 7599 | |
bb9d97b6 TH |
7600 | static void cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7601 | struct cgroup_taskset *tset) | |
68318b8e | 7602 | { |
bb9d97b6 TH |
7603 | struct task_struct *task; |
7604 | ||
7605 | cgroup_taskset_for_each(task, cgrp, tset) | |
7606 | sched_move_task(task); | |
68318b8e SV |
7607 | } |
7608 | ||
068c5cc5 | 7609 | static void |
d41d5a01 PZ |
7610 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
7611 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
7612 | { |
7613 | /* | |
7614 | * cgroup_exit() is called in the copy_process() failure path. | |
7615 | * Ignore this case since the task hasn't ran yet, this avoids | |
7616 | * trying to poke a half freed task state from generic code. | |
7617 | */ | |
7618 | if (!(task->flags & PF_EXITING)) | |
7619 | return; | |
7620 | ||
7621 | sched_move_task(task); | |
7622 | } | |
7623 | ||
052f1dc7 | 7624 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7625 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7626 | u64 shareval) |
68318b8e | 7627 | { |
c8b28116 | 7628 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7629 | } |
7630 | ||
f4c753b7 | 7631 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7632 | { |
2b01dfe3 | 7633 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7634 | |
c8b28116 | 7635 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7636 | } |
ab84d31e PT |
7637 | |
7638 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7639 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7640 | ||
ab84d31e PT |
7641 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7642 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7643 | ||
a790de99 PT |
7644 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7645 | ||
ab84d31e PT |
7646 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7647 | { | |
56f570e5 | 7648 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7649 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7650 | |
7651 | if (tg == &root_task_group) | |
7652 | return -EINVAL; | |
7653 | ||
7654 | /* | |
7655 | * Ensure we have at some amount of bandwidth every period. This is | |
7656 | * to prevent reaching a state of large arrears when throttled via | |
7657 | * entity_tick() resulting in prolonged exit starvation. | |
7658 | */ | |
7659 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7660 | return -EINVAL; | |
7661 | ||
7662 | /* | |
7663 | * Likewise, bound things on the otherside by preventing insane quota | |
7664 | * periods. This also allows us to normalize in computing quota | |
7665 | * feasibility. | |
7666 | */ | |
7667 | if (period > max_cfs_quota_period) | |
7668 | return -EINVAL; | |
7669 | ||
a790de99 PT |
7670 | mutex_lock(&cfs_constraints_mutex); |
7671 | ret = __cfs_schedulable(tg, period, quota); | |
7672 | if (ret) | |
7673 | goto out_unlock; | |
7674 | ||
58088ad0 | 7675 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7676 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7677 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7678 | raw_spin_lock_irq(&cfs_b->lock); |
7679 | cfs_b->period = ns_to_ktime(period); | |
7680 | cfs_b->quota = quota; | |
58088ad0 | 7681 | |
a9cf55b2 | 7682 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7683 | /* restart the period timer (if active) to handle new period expiry */ |
7684 | if (runtime_enabled && cfs_b->timer_active) { | |
7685 | /* force a reprogram */ | |
7686 | cfs_b->timer_active = 0; | |
7687 | __start_cfs_bandwidth(cfs_b); | |
7688 | } | |
ab84d31e PT |
7689 | raw_spin_unlock_irq(&cfs_b->lock); |
7690 | ||
7691 | for_each_possible_cpu(i) { | |
7692 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7693 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7694 | |
7695 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7696 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7697 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7698 | |
029632fb | 7699 | if (cfs_rq->throttled) |
671fd9da | 7700 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7701 | raw_spin_unlock_irq(&rq->lock); |
7702 | } | |
a790de99 PT |
7703 | out_unlock: |
7704 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7705 | |
a790de99 | 7706 | return ret; |
ab84d31e PT |
7707 | } |
7708 | ||
7709 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7710 | { | |
7711 | u64 quota, period; | |
7712 | ||
029632fb | 7713 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7714 | if (cfs_quota_us < 0) |
7715 | quota = RUNTIME_INF; | |
7716 | else | |
7717 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7718 | ||
7719 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7720 | } | |
7721 | ||
7722 | long tg_get_cfs_quota(struct task_group *tg) | |
7723 | { | |
7724 | u64 quota_us; | |
7725 | ||
029632fb | 7726 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7727 | return -1; |
7728 | ||
029632fb | 7729 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7730 | do_div(quota_us, NSEC_PER_USEC); |
7731 | ||
7732 | return quota_us; | |
7733 | } | |
7734 | ||
7735 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7736 | { | |
7737 | u64 quota, period; | |
7738 | ||
7739 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7740 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7741 | |
ab84d31e PT |
7742 | return tg_set_cfs_bandwidth(tg, period, quota); |
7743 | } | |
7744 | ||
7745 | long tg_get_cfs_period(struct task_group *tg) | |
7746 | { | |
7747 | u64 cfs_period_us; | |
7748 | ||
029632fb | 7749 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7750 | do_div(cfs_period_us, NSEC_PER_USEC); |
7751 | ||
7752 | return cfs_period_us; | |
7753 | } | |
7754 | ||
7755 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7756 | { | |
7757 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7758 | } | |
7759 | ||
7760 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7761 | s64 cfs_quota_us) | |
7762 | { | |
7763 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7764 | } | |
7765 | ||
7766 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7767 | { | |
7768 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7769 | } | |
7770 | ||
7771 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7772 | u64 cfs_period_us) | |
7773 | { | |
7774 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7775 | } | |
7776 | ||
a790de99 PT |
7777 | struct cfs_schedulable_data { |
7778 | struct task_group *tg; | |
7779 | u64 period, quota; | |
7780 | }; | |
7781 | ||
7782 | /* | |
7783 | * normalize group quota/period to be quota/max_period | |
7784 | * note: units are usecs | |
7785 | */ | |
7786 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7787 | struct cfs_schedulable_data *d) | |
7788 | { | |
7789 | u64 quota, period; | |
7790 | ||
7791 | if (tg == d->tg) { | |
7792 | period = d->period; | |
7793 | quota = d->quota; | |
7794 | } else { | |
7795 | period = tg_get_cfs_period(tg); | |
7796 | quota = tg_get_cfs_quota(tg); | |
7797 | } | |
7798 | ||
7799 | /* note: these should typically be equivalent */ | |
7800 | if (quota == RUNTIME_INF || quota == -1) | |
7801 | return RUNTIME_INF; | |
7802 | ||
7803 | return to_ratio(period, quota); | |
7804 | } | |
7805 | ||
7806 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7807 | { | |
7808 | struct cfs_schedulable_data *d = data; | |
029632fb | 7809 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7810 | s64 quota = 0, parent_quota = -1; |
7811 | ||
7812 | if (!tg->parent) { | |
7813 | quota = RUNTIME_INF; | |
7814 | } else { | |
029632fb | 7815 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7816 | |
7817 | quota = normalize_cfs_quota(tg, d); | |
7818 | parent_quota = parent_b->hierarchal_quota; | |
7819 | ||
7820 | /* | |
7821 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7822 | * limit is set | |
7823 | */ | |
7824 | if (quota == RUNTIME_INF) | |
7825 | quota = parent_quota; | |
7826 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7827 | return -EINVAL; | |
7828 | } | |
7829 | cfs_b->hierarchal_quota = quota; | |
7830 | ||
7831 | return 0; | |
7832 | } | |
7833 | ||
7834 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7835 | { | |
8277434e | 7836 | int ret; |
a790de99 PT |
7837 | struct cfs_schedulable_data data = { |
7838 | .tg = tg, | |
7839 | .period = period, | |
7840 | .quota = quota, | |
7841 | }; | |
7842 | ||
7843 | if (quota != RUNTIME_INF) { | |
7844 | do_div(data.period, NSEC_PER_USEC); | |
7845 | do_div(data.quota, NSEC_PER_USEC); | |
7846 | } | |
7847 | ||
8277434e PT |
7848 | rcu_read_lock(); |
7849 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7850 | rcu_read_unlock(); | |
7851 | ||
7852 | return ret; | |
a790de99 | 7853 | } |
e8da1b18 NR |
7854 | |
7855 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7856 | struct cgroup_map_cb *cb) | |
7857 | { | |
7858 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7859 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7860 | |
7861 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7862 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7863 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7864 | ||
7865 | return 0; | |
7866 | } | |
ab84d31e | 7867 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7868 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7869 | |
052f1dc7 | 7870 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7871 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7872 | s64 val) |
6f505b16 | 7873 | { |
06ecb27c | 7874 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7875 | } |
7876 | ||
06ecb27c | 7877 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7878 | { |
06ecb27c | 7879 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7880 | } |
d0b27fa7 PZ |
7881 | |
7882 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7883 | u64 rt_period_us) | |
7884 | { | |
7885 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7886 | } | |
7887 | ||
7888 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7889 | { | |
7890 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7891 | } | |
6d6bc0ad | 7892 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7893 | |
fe5c7cc2 | 7894 | static struct cftype cpu_files[] = { |
052f1dc7 | 7895 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7896 | { |
7897 | .name = "shares", | |
f4c753b7 PM |
7898 | .read_u64 = cpu_shares_read_u64, |
7899 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7900 | }, |
052f1dc7 | 7901 | #endif |
ab84d31e PT |
7902 | #ifdef CONFIG_CFS_BANDWIDTH |
7903 | { | |
7904 | .name = "cfs_quota_us", | |
7905 | .read_s64 = cpu_cfs_quota_read_s64, | |
7906 | .write_s64 = cpu_cfs_quota_write_s64, | |
7907 | }, | |
7908 | { | |
7909 | .name = "cfs_period_us", | |
7910 | .read_u64 = cpu_cfs_period_read_u64, | |
7911 | .write_u64 = cpu_cfs_period_write_u64, | |
7912 | }, | |
e8da1b18 NR |
7913 | { |
7914 | .name = "stat", | |
7915 | .read_map = cpu_stats_show, | |
7916 | }, | |
ab84d31e | 7917 | #endif |
052f1dc7 | 7918 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7919 | { |
9f0c1e56 | 7920 | .name = "rt_runtime_us", |
06ecb27c PM |
7921 | .read_s64 = cpu_rt_runtime_read, |
7922 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7923 | }, |
d0b27fa7 PZ |
7924 | { |
7925 | .name = "rt_period_us", | |
f4c753b7 PM |
7926 | .read_u64 = cpu_rt_period_read_uint, |
7927 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7928 | }, |
052f1dc7 | 7929 | #endif |
68318b8e SV |
7930 | }; |
7931 | ||
7932 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7933 | { | |
fe5c7cc2 | 7934 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7935 | } |
7936 | ||
7937 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7938 | .name = "cpu", |
7939 | .create = cpu_cgroup_create, | |
7940 | .destroy = cpu_cgroup_destroy, | |
bb9d97b6 TH |
7941 | .can_attach = cpu_cgroup_can_attach, |
7942 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7943 | .exit = cpu_cgroup_exit, |
38605cae IM |
7944 | .populate = cpu_cgroup_populate, |
7945 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7946 | .early_init = 1, |
7947 | }; | |
7948 | ||
052f1dc7 | 7949 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
7950 | |
7951 | #ifdef CONFIG_CGROUP_CPUACCT | |
7952 | ||
7953 | /* | |
7954 | * CPU accounting code for task groups. | |
7955 | * | |
7956 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7957 | * (balbir@in.ibm.com). | |
7958 | */ | |
7959 | ||
d842de87 SV |
7960 | /* create a new cpu accounting group */ |
7961 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 7962 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7963 | { |
54c707e9 | 7964 | struct cpuacct *ca; |
d842de87 | 7965 | |
54c707e9 GC |
7966 | if (!cgrp->parent) |
7967 | return &root_cpuacct.css; | |
7968 | ||
7969 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 7970 | if (!ca) |
ef12fefa | 7971 | goto out; |
d842de87 SV |
7972 | |
7973 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
7974 | if (!ca->cpuusage) |
7975 | goto out_free_ca; | |
7976 | ||
54c707e9 GC |
7977 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
7978 | if (!ca->cpustat) | |
7979 | goto out_free_cpuusage; | |
934352f2 | 7980 | |
d842de87 | 7981 | return &ca->css; |
ef12fefa | 7982 | |
54c707e9 | 7983 | out_free_cpuusage: |
ef12fefa BR |
7984 | free_percpu(ca->cpuusage); |
7985 | out_free_ca: | |
7986 | kfree(ca); | |
7987 | out: | |
7988 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
7989 | } |
7990 | ||
7991 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 7992 | static void |
32cd756a | 7993 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 7994 | { |
32cd756a | 7995 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 7996 | |
54c707e9 | 7997 | free_percpu(ca->cpustat); |
d842de87 SV |
7998 | free_percpu(ca->cpuusage); |
7999 | kfree(ca); | |
8000 | } | |
8001 | ||
720f5498 KC |
8002 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8003 | { | |
b36128c8 | 8004 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8005 | u64 data; |
8006 | ||
8007 | #ifndef CONFIG_64BIT | |
8008 | /* | |
8009 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8010 | */ | |
05fa785c | 8011 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8012 | data = *cpuusage; |
05fa785c | 8013 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8014 | #else |
8015 | data = *cpuusage; | |
8016 | #endif | |
8017 | ||
8018 | return data; | |
8019 | } | |
8020 | ||
8021 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8022 | { | |
b36128c8 | 8023 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8024 | |
8025 | #ifndef CONFIG_64BIT | |
8026 | /* | |
8027 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8028 | */ | |
05fa785c | 8029 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8030 | *cpuusage = val; |
05fa785c | 8031 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8032 | #else |
8033 | *cpuusage = val; | |
8034 | #endif | |
8035 | } | |
8036 | ||
d842de87 | 8037 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8038 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8039 | { |
32cd756a | 8040 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8041 | u64 totalcpuusage = 0; |
8042 | int i; | |
8043 | ||
720f5498 KC |
8044 | for_each_present_cpu(i) |
8045 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8046 | |
8047 | return totalcpuusage; | |
8048 | } | |
8049 | ||
0297b803 DG |
8050 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8051 | u64 reset) | |
8052 | { | |
8053 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8054 | int err = 0; | |
8055 | int i; | |
8056 | ||
8057 | if (reset) { | |
8058 | err = -EINVAL; | |
8059 | goto out; | |
8060 | } | |
8061 | ||
720f5498 KC |
8062 | for_each_present_cpu(i) |
8063 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8064 | |
0297b803 DG |
8065 | out: |
8066 | return err; | |
8067 | } | |
8068 | ||
e9515c3c KC |
8069 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8070 | struct seq_file *m) | |
8071 | { | |
8072 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8073 | u64 percpu; | |
8074 | int i; | |
8075 | ||
8076 | for_each_present_cpu(i) { | |
8077 | percpu = cpuacct_cpuusage_read(ca, i); | |
8078 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8079 | } | |
8080 | seq_printf(m, "\n"); | |
8081 | return 0; | |
8082 | } | |
8083 | ||
ef12fefa BR |
8084 | static const char *cpuacct_stat_desc[] = { |
8085 | [CPUACCT_STAT_USER] = "user", | |
8086 | [CPUACCT_STAT_SYSTEM] = "system", | |
8087 | }; | |
8088 | ||
8089 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8090 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8091 | { |
8092 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8093 | int cpu; |
8094 | s64 val = 0; | |
ef12fefa | 8095 | |
54c707e9 GC |
8096 | for_each_online_cpu(cpu) { |
8097 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8098 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8099 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8100 | } |
54c707e9 GC |
8101 | val = cputime64_to_clock_t(val); |
8102 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8103 | |
54c707e9 GC |
8104 | val = 0; |
8105 | for_each_online_cpu(cpu) { | |
8106 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8107 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8108 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8109 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8110 | } |
54c707e9 GC |
8111 | |
8112 | val = cputime64_to_clock_t(val); | |
8113 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8114 | ||
ef12fefa BR |
8115 | return 0; |
8116 | } | |
8117 | ||
d842de87 SV |
8118 | static struct cftype files[] = { |
8119 | { | |
8120 | .name = "usage", | |
f4c753b7 PM |
8121 | .read_u64 = cpuusage_read, |
8122 | .write_u64 = cpuusage_write, | |
d842de87 | 8123 | }, |
e9515c3c KC |
8124 | { |
8125 | .name = "usage_percpu", | |
8126 | .read_seq_string = cpuacct_percpu_seq_read, | |
8127 | }, | |
ef12fefa BR |
8128 | { |
8129 | .name = "stat", | |
8130 | .read_map = cpuacct_stats_show, | |
8131 | }, | |
d842de87 SV |
8132 | }; |
8133 | ||
32cd756a | 8134 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8135 | { |
32cd756a | 8136 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8137 | } |
8138 | ||
8139 | /* | |
8140 | * charge this task's execution time to its accounting group. | |
8141 | * | |
8142 | * called with rq->lock held. | |
8143 | */ | |
029632fb | 8144 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8145 | { |
8146 | struct cpuacct *ca; | |
934352f2 | 8147 | int cpu; |
d842de87 | 8148 | |
c40c6f85 | 8149 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8150 | return; |
8151 | ||
934352f2 | 8152 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8153 | |
8154 | rcu_read_lock(); | |
8155 | ||
d842de87 | 8156 | ca = task_ca(tsk); |
d842de87 | 8157 | |
44252e42 | 8158 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8159 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8160 | *cpuusage += cputime; |
8161 | } | |
a18b83b7 BR |
8162 | |
8163 | rcu_read_unlock(); | |
d842de87 SV |
8164 | } |
8165 | ||
8166 | struct cgroup_subsys cpuacct_subsys = { | |
8167 | .name = "cpuacct", | |
8168 | .create = cpuacct_create, | |
8169 | .destroy = cpuacct_destroy, | |
8170 | .populate = cpuacct_populate, | |
8171 | .subsys_id = cpuacct_subsys_id, | |
8172 | }; | |
8173 | #endif /* CONFIG_CGROUP_CPUACCT */ |