Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
1da177e4 | 76 | |
96f951ed | 77 | #include <asm/switch_to.h> |
5517d86b | 78 | #include <asm/tlb.h> |
838225b4 | 79 | #include <asm/irq_regs.h> |
db7e527d | 80 | #include <asm/mutex.h> |
e6e6685a GC |
81 | #ifdef CONFIG_PARAVIRT |
82 | #include <asm/paravirt.h> | |
83 | #endif | |
1da177e4 | 84 | |
029632fb | 85 | #include "sched.h" |
ea138446 | 86 | #include "../workqueue_internal.h" |
29d5e047 | 87 | #include "../smpboot.h" |
6e0534f2 | 88 | |
a8d154b0 | 89 | #define CREATE_TRACE_POINTS |
ad8d75ff | 90 | #include <trace/events/sched.h> |
a8d154b0 | 91 | |
029632fb | 92 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 93 | { |
58088ad0 PT |
94 | unsigned long delta; |
95 | ktime_t soft, hard, now; | |
d0b27fa7 | 96 | |
58088ad0 PT |
97 | for (;;) { |
98 | if (hrtimer_active(period_timer)) | |
99 | break; | |
100 | ||
101 | now = hrtimer_cb_get_time(period_timer); | |
102 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 103 | |
58088ad0 PT |
104 | soft = hrtimer_get_softexpires(period_timer); |
105 | hard = hrtimer_get_expires(period_timer); | |
106 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
107 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
108 | HRTIMER_MODE_ABS_PINNED, 0); | |
109 | } | |
110 | } | |
111 | ||
029632fb PZ |
112 | DEFINE_MUTEX(sched_domains_mutex); |
113 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 114 | |
fe44d621 | 115 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 116 | |
029632fb | 117 | void update_rq_clock(struct rq *rq) |
3e51f33f | 118 | { |
fe44d621 | 119 | s64 delta; |
305e6835 | 120 | |
61eadef6 | 121 | if (rq->skip_clock_update > 0) |
f26f9aff | 122 | return; |
aa483808 | 123 | |
fe44d621 PZ |
124 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
125 | rq->clock += delta; | |
126 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
127 | } |
128 | ||
bf5c91ba IM |
129 | /* |
130 | * Debugging: various feature bits | |
131 | */ | |
f00b45c1 | 132 | |
f00b45c1 PZ |
133 | #define SCHED_FEAT(name, enabled) \ |
134 | (1UL << __SCHED_FEAT_##name) * enabled | | |
135 | ||
bf5c91ba | 136 | const_debug unsigned int sysctl_sched_features = |
391e43da | 137 | #include "features.h" |
f00b45c1 PZ |
138 | 0; |
139 | ||
140 | #undef SCHED_FEAT | |
141 | ||
142 | #ifdef CONFIG_SCHED_DEBUG | |
143 | #define SCHED_FEAT(name, enabled) \ | |
144 | #name , | |
145 | ||
1292531f | 146 | static const char * const sched_feat_names[] = { |
391e43da | 147 | #include "features.h" |
f00b45c1 PZ |
148 | }; |
149 | ||
150 | #undef SCHED_FEAT | |
151 | ||
34f3a814 | 152 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 153 | { |
f00b45c1 PZ |
154 | int i; |
155 | ||
f8b6d1cc | 156 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
157 | if (!(sysctl_sched_features & (1UL << i))) |
158 | seq_puts(m, "NO_"); | |
159 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 160 | } |
34f3a814 | 161 | seq_puts(m, "\n"); |
f00b45c1 | 162 | |
34f3a814 | 163 | return 0; |
f00b45c1 PZ |
164 | } |
165 | ||
f8b6d1cc PZ |
166 | #ifdef HAVE_JUMP_LABEL |
167 | ||
c5905afb IM |
168 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
169 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
170 | |
171 | #define SCHED_FEAT(name, enabled) \ | |
172 | jump_label_key__##enabled , | |
173 | ||
c5905afb | 174 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
175 | #include "features.h" |
176 | }; | |
177 | ||
178 | #undef SCHED_FEAT | |
179 | ||
180 | static void sched_feat_disable(int i) | |
181 | { | |
c5905afb IM |
182 | if (static_key_enabled(&sched_feat_keys[i])) |
183 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
184 | } |
185 | ||
186 | static void sched_feat_enable(int i) | |
187 | { | |
c5905afb IM |
188 | if (!static_key_enabled(&sched_feat_keys[i])) |
189 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
190 | } |
191 | #else | |
192 | static void sched_feat_disable(int i) { }; | |
193 | static void sched_feat_enable(int i) { }; | |
194 | #endif /* HAVE_JUMP_LABEL */ | |
195 | ||
1a687c2e | 196 | static int sched_feat_set(char *cmp) |
f00b45c1 | 197 | { |
f00b45c1 | 198 | int i; |
1a687c2e | 199 | int neg = 0; |
f00b45c1 | 200 | |
524429c3 | 201 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
202 | neg = 1; |
203 | cmp += 3; | |
204 | } | |
205 | ||
f8b6d1cc | 206 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 207 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 208 | if (neg) { |
f00b45c1 | 209 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
210 | sched_feat_disable(i); |
211 | } else { | |
f00b45c1 | 212 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
213 | sched_feat_enable(i); |
214 | } | |
f00b45c1 PZ |
215 | break; |
216 | } | |
217 | } | |
218 | ||
1a687c2e MG |
219 | return i; |
220 | } | |
221 | ||
222 | static ssize_t | |
223 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
224 | size_t cnt, loff_t *ppos) | |
225 | { | |
226 | char buf[64]; | |
227 | char *cmp; | |
228 | int i; | |
229 | ||
230 | if (cnt > 63) | |
231 | cnt = 63; | |
232 | ||
233 | if (copy_from_user(&buf, ubuf, cnt)) | |
234 | return -EFAULT; | |
235 | ||
236 | buf[cnt] = 0; | |
237 | cmp = strstrip(buf); | |
238 | ||
239 | i = sched_feat_set(cmp); | |
f8b6d1cc | 240 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
241 | return -EINVAL; |
242 | ||
42994724 | 243 | *ppos += cnt; |
f00b45c1 PZ |
244 | |
245 | return cnt; | |
246 | } | |
247 | ||
34f3a814 LZ |
248 | static int sched_feat_open(struct inode *inode, struct file *filp) |
249 | { | |
250 | return single_open(filp, sched_feat_show, NULL); | |
251 | } | |
252 | ||
828c0950 | 253 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
254 | .open = sched_feat_open, |
255 | .write = sched_feat_write, | |
256 | .read = seq_read, | |
257 | .llseek = seq_lseek, | |
258 | .release = single_release, | |
f00b45c1 PZ |
259 | }; |
260 | ||
261 | static __init int sched_init_debug(void) | |
262 | { | |
f00b45c1 PZ |
263 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
264 | &sched_feat_fops); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | late_initcall(sched_init_debug); | |
f8b6d1cc | 269 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 270 | |
b82d9fdd PZ |
271 | /* |
272 | * Number of tasks to iterate in a single balance run. | |
273 | * Limited because this is done with IRQs disabled. | |
274 | */ | |
275 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
276 | ||
e9e9250b PZ |
277 | /* |
278 | * period over which we average the RT time consumption, measured | |
279 | * in ms. | |
280 | * | |
281 | * default: 1s | |
282 | */ | |
283 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
284 | ||
fa85ae24 | 285 | /* |
9f0c1e56 | 286 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
287 | * default: 1s |
288 | */ | |
9f0c1e56 | 289 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 290 | |
029632fb | 291 | __read_mostly int scheduler_running; |
6892b75e | 292 | |
9f0c1e56 PZ |
293 | /* |
294 | * part of the period that we allow rt tasks to run in us. | |
295 | * default: 0.95s | |
296 | */ | |
297 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 298 | |
fa85ae24 | 299 | |
1da177e4 | 300 | |
0970d299 | 301 | /* |
0122ec5b | 302 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 303 | */ |
70b97a7f | 304 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
305 | __acquires(rq->lock) |
306 | { | |
0970d299 PZ |
307 | struct rq *rq; |
308 | ||
0122ec5b PZ |
309 | lockdep_assert_held(&p->pi_lock); |
310 | ||
3a5c359a | 311 | for (;;) { |
0970d299 | 312 | rq = task_rq(p); |
05fa785c | 313 | raw_spin_lock(&rq->lock); |
65cc8e48 | 314 | if (likely(rq == task_rq(p))) |
3a5c359a | 315 | return rq; |
05fa785c | 316 | raw_spin_unlock(&rq->lock); |
b29739f9 | 317 | } |
b29739f9 IM |
318 | } |
319 | ||
1da177e4 | 320 | /* |
0122ec5b | 321 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 322 | */ |
70b97a7f | 323 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 324 | __acquires(p->pi_lock) |
1da177e4 LT |
325 | __acquires(rq->lock) |
326 | { | |
70b97a7f | 327 | struct rq *rq; |
1da177e4 | 328 | |
3a5c359a | 329 | for (;;) { |
0122ec5b | 330 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 331 | rq = task_rq(p); |
05fa785c | 332 | raw_spin_lock(&rq->lock); |
65cc8e48 | 333 | if (likely(rq == task_rq(p))) |
3a5c359a | 334 | return rq; |
0122ec5b PZ |
335 | raw_spin_unlock(&rq->lock); |
336 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 337 | } |
1da177e4 LT |
338 | } |
339 | ||
a9957449 | 340 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
341 | __releases(rq->lock) |
342 | { | |
05fa785c | 343 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
344 | } |
345 | ||
0122ec5b PZ |
346 | static inline void |
347 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 348 | __releases(rq->lock) |
0122ec5b | 349 | __releases(p->pi_lock) |
1da177e4 | 350 | { |
0122ec5b PZ |
351 | raw_spin_unlock(&rq->lock); |
352 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
353 | } |
354 | ||
1da177e4 | 355 | /* |
cc2a73b5 | 356 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 357 | */ |
a9957449 | 358 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
359 | __acquires(rq->lock) |
360 | { | |
70b97a7f | 361 | struct rq *rq; |
1da177e4 LT |
362 | |
363 | local_irq_disable(); | |
364 | rq = this_rq(); | |
05fa785c | 365 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
366 | |
367 | return rq; | |
368 | } | |
369 | ||
8f4d37ec PZ |
370 | #ifdef CONFIG_SCHED_HRTICK |
371 | /* | |
372 | * Use HR-timers to deliver accurate preemption points. | |
373 | * | |
374 | * Its all a bit involved since we cannot program an hrt while holding the | |
375 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
376 | * reschedule event. | |
377 | * | |
378 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
379 | * rq->lock. | |
380 | */ | |
8f4d37ec | 381 | |
8f4d37ec PZ |
382 | static void hrtick_clear(struct rq *rq) |
383 | { | |
384 | if (hrtimer_active(&rq->hrtick_timer)) | |
385 | hrtimer_cancel(&rq->hrtick_timer); | |
386 | } | |
387 | ||
8f4d37ec PZ |
388 | /* |
389 | * High-resolution timer tick. | |
390 | * Runs from hardirq context with interrupts disabled. | |
391 | */ | |
392 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
393 | { | |
394 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
395 | ||
396 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
397 | ||
05fa785c | 398 | raw_spin_lock(&rq->lock); |
3e51f33f | 399 | update_rq_clock(rq); |
8f4d37ec | 400 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 401 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
402 | |
403 | return HRTIMER_NORESTART; | |
404 | } | |
405 | ||
95e904c7 | 406 | #ifdef CONFIG_SMP |
31656519 PZ |
407 | /* |
408 | * called from hardirq (IPI) context | |
409 | */ | |
410 | static void __hrtick_start(void *arg) | |
b328ca18 | 411 | { |
31656519 | 412 | struct rq *rq = arg; |
b328ca18 | 413 | |
05fa785c | 414 | raw_spin_lock(&rq->lock); |
31656519 PZ |
415 | hrtimer_restart(&rq->hrtick_timer); |
416 | rq->hrtick_csd_pending = 0; | |
05fa785c | 417 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
418 | } |
419 | ||
31656519 PZ |
420 | /* |
421 | * Called to set the hrtick timer state. | |
422 | * | |
423 | * called with rq->lock held and irqs disabled | |
424 | */ | |
029632fb | 425 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 426 | { |
31656519 PZ |
427 | struct hrtimer *timer = &rq->hrtick_timer; |
428 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 429 | |
cc584b21 | 430 | hrtimer_set_expires(timer, time); |
31656519 PZ |
431 | |
432 | if (rq == this_rq()) { | |
433 | hrtimer_restart(timer); | |
434 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 435 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
436 | rq->hrtick_csd_pending = 1; |
437 | } | |
b328ca18 PZ |
438 | } |
439 | ||
440 | static int | |
441 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
442 | { | |
443 | int cpu = (int)(long)hcpu; | |
444 | ||
445 | switch (action) { | |
446 | case CPU_UP_CANCELED: | |
447 | case CPU_UP_CANCELED_FROZEN: | |
448 | case CPU_DOWN_PREPARE: | |
449 | case CPU_DOWN_PREPARE_FROZEN: | |
450 | case CPU_DEAD: | |
451 | case CPU_DEAD_FROZEN: | |
31656519 | 452 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
453 | return NOTIFY_OK; |
454 | } | |
455 | ||
456 | return NOTIFY_DONE; | |
457 | } | |
458 | ||
fa748203 | 459 | static __init void init_hrtick(void) |
b328ca18 PZ |
460 | { |
461 | hotcpu_notifier(hotplug_hrtick, 0); | |
462 | } | |
31656519 PZ |
463 | #else |
464 | /* | |
465 | * Called to set the hrtick timer state. | |
466 | * | |
467 | * called with rq->lock held and irqs disabled | |
468 | */ | |
029632fb | 469 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 470 | { |
7f1e2ca9 | 471 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 472 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 473 | } |
b328ca18 | 474 | |
006c75f1 | 475 | static inline void init_hrtick(void) |
8f4d37ec | 476 | { |
8f4d37ec | 477 | } |
31656519 | 478 | #endif /* CONFIG_SMP */ |
8f4d37ec | 479 | |
31656519 | 480 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 481 | { |
31656519 PZ |
482 | #ifdef CONFIG_SMP |
483 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 484 | |
31656519 PZ |
485 | rq->hrtick_csd.flags = 0; |
486 | rq->hrtick_csd.func = __hrtick_start; | |
487 | rq->hrtick_csd.info = rq; | |
488 | #endif | |
8f4d37ec | 489 | |
31656519 PZ |
490 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
491 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 492 | } |
006c75f1 | 493 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
494 | static inline void hrtick_clear(struct rq *rq) |
495 | { | |
496 | } | |
497 | ||
8f4d37ec PZ |
498 | static inline void init_rq_hrtick(struct rq *rq) |
499 | { | |
500 | } | |
501 | ||
b328ca18 PZ |
502 | static inline void init_hrtick(void) |
503 | { | |
504 | } | |
006c75f1 | 505 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 506 | |
c24d20db IM |
507 | /* |
508 | * resched_task - mark a task 'to be rescheduled now'. | |
509 | * | |
510 | * On UP this means the setting of the need_resched flag, on SMP it | |
511 | * might also involve a cross-CPU call to trigger the scheduler on | |
512 | * the target CPU. | |
513 | */ | |
514 | #ifdef CONFIG_SMP | |
029632fb | 515 | void resched_task(struct task_struct *p) |
c24d20db IM |
516 | { |
517 | int cpu; | |
518 | ||
05fa785c | 519 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 520 | |
5ed0cec0 | 521 | if (test_tsk_need_resched(p)) |
c24d20db IM |
522 | return; |
523 | ||
5ed0cec0 | 524 | set_tsk_need_resched(p); |
c24d20db IM |
525 | |
526 | cpu = task_cpu(p); | |
527 | if (cpu == smp_processor_id()) | |
528 | return; | |
529 | ||
530 | /* NEED_RESCHED must be visible before we test polling */ | |
531 | smp_mb(); | |
532 | if (!tsk_is_polling(p)) | |
533 | smp_send_reschedule(cpu); | |
534 | } | |
535 | ||
029632fb | 536 | void resched_cpu(int cpu) |
c24d20db IM |
537 | { |
538 | struct rq *rq = cpu_rq(cpu); | |
539 | unsigned long flags; | |
540 | ||
05fa785c | 541 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
542 | return; |
543 | resched_task(cpu_curr(cpu)); | |
05fa785c | 544 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 545 | } |
06d8308c | 546 | |
3451d024 | 547 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
548 | /* |
549 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
550 | * from an idle cpu. This is good for power-savings. | |
551 | * | |
552 | * We don't do similar optimization for completely idle system, as | |
553 | * selecting an idle cpu will add more delays to the timers than intended | |
554 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
555 | */ | |
556 | int get_nohz_timer_target(void) | |
557 | { | |
558 | int cpu = smp_processor_id(); | |
559 | int i; | |
560 | struct sched_domain *sd; | |
561 | ||
057f3fad | 562 | rcu_read_lock(); |
83cd4fe2 | 563 | for_each_domain(cpu, sd) { |
057f3fad PZ |
564 | for_each_cpu(i, sched_domain_span(sd)) { |
565 | if (!idle_cpu(i)) { | |
566 | cpu = i; | |
567 | goto unlock; | |
568 | } | |
569 | } | |
83cd4fe2 | 570 | } |
057f3fad PZ |
571 | unlock: |
572 | rcu_read_unlock(); | |
83cd4fe2 VP |
573 | return cpu; |
574 | } | |
06d8308c TG |
575 | /* |
576 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
577 | * idle CPU then this timer might expire before the next timer event | |
578 | * which is scheduled to wake up that CPU. In case of a completely | |
579 | * idle system the next event might even be infinite time into the | |
580 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
581 | * leaves the inner idle loop so the newly added timer is taken into | |
582 | * account when the CPU goes back to idle and evaluates the timer | |
583 | * wheel for the next timer event. | |
584 | */ | |
1c20091e | 585 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
586 | { |
587 | struct rq *rq = cpu_rq(cpu); | |
588 | ||
589 | if (cpu == smp_processor_id()) | |
590 | return; | |
591 | ||
592 | /* | |
593 | * This is safe, as this function is called with the timer | |
594 | * wheel base lock of (cpu) held. When the CPU is on the way | |
595 | * to idle and has not yet set rq->curr to idle then it will | |
596 | * be serialized on the timer wheel base lock and take the new | |
597 | * timer into account automatically. | |
598 | */ | |
599 | if (rq->curr != rq->idle) | |
600 | return; | |
45bf76df | 601 | |
45bf76df | 602 | /* |
06d8308c TG |
603 | * We can set TIF_RESCHED on the idle task of the other CPU |
604 | * lockless. The worst case is that the other CPU runs the | |
605 | * idle task through an additional NOOP schedule() | |
45bf76df | 606 | */ |
5ed0cec0 | 607 | set_tsk_need_resched(rq->idle); |
45bf76df | 608 | |
06d8308c TG |
609 | /* NEED_RESCHED must be visible before we test polling */ |
610 | smp_mb(); | |
611 | if (!tsk_is_polling(rq->idle)) | |
612 | smp_send_reschedule(cpu); | |
45bf76df IM |
613 | } |
614 | ||
c5bfece2 | 615 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 616 | { |
c5bfece2 | 617 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
618 | if (cpu != smp_processor_id() || |
619 | tick_nohz_tick_stopped()) | |
620 | smp_send_reschedule(cpu); | |
621 | return true; | |
622 | } | |
623 | ||
624 | return false; | |
625 | } | |
626 | ||
627 | void wake_up_nohz_cpu(int cpu) | |
628 | { | |
c5bfece2 | 629 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
630 | wake_up_idle_cpu(cpu); |
631 | } | |
632 | ||
ca38062e | 633 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 634 | { |
1c792db7 SS |
635 | int cpu = smp_processor_id(); |
636 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
637 | } |
638 | ||
3451d024 | 639 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 640 | |
ca38062e | 641 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 642 | { |
ca38062e | 643 | return false; |
2069dd75 PZ |
644 | } |
645 | ||
3451d024 | 646 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 647 | |
ce831b38 FW |
648 | #ifdef CONFIG_NO_HZ_FULL |
649 | bool sched_can_stop_tick(void) | |
650 | { | |
651 | struct rq *rq; | |
652 | ||
653 | rq = this_rq(); | |
654 | ||
655 | /* Make sure rq->nr_running update is visible after the IPI */ | |
656 | smp_rmb(); | |
657 | ||
658 | /* More than one running task need preemption */ | |
659 | if (rq->nr_running > 1) | |
660 | return false; | |
661 | ||
662 | return true; | |
663 | } | |
664 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 665 | |
029632fb | 666 | void sched_avg_update(struct rq *rq) |
18d95a28 | 667 | { |
e9e9250b PZ |
668 | s64 period = sched_avg_period(); |
669 | ||
670 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
671 | /* |
672 | * Inline assembly required to prevent the compiler | |
673 | * optimising this loop into a divmod call. | |
674 | * See __iter_div_u64_rem() for another example of this. | |
675 | */ | |
676 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
677 | rq->age_stamp += period; |
678 | rq->rt_avg /= 2; | |
679 | } | |
18d95a28 PZ |
680 | } |
681 | ||
6d6bc0ad | 682 | #else /* !CONFIG_SMP */ |
029632fb | 683 | void resched_task(struct task_struct *p) |
18d95a28 | 684 | { |
05fa785c | 685 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 686 | set_tsk_need_resched(p); |
18d95a28 | 687 | } |
6d6bc0ad | 688 | #endif /* CONFIG_SMP */ |
18d95a28 | 689 | |
a790de99 PT |
690 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
691 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 692 | /* |
8277434e PT |
693 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
694 | * node and @up when leaving it for the final time. | |
695 | * | |
696 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 697 | */ |
029632fb | 698 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 699 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
700 | { |
701 | struct task_group *parent, *child; | |
eb755805 | 702 | int ret; |
c09595f6 | 703 | |
8277434e PT |
704 | parent = from; |
705 | ||
c09595f6 | 706 | down: |
eb755805 PZ |
707 | ret = (*down)(parent, data); |
708 | if (ret) | |
8277434e | 709 | goto out; |
c09595f6 PZ |
710 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
711 | parent = child; | |
712 | goto down; | |
713 | ||
714 | up: | |
715 | continue; | |
716 | } | |
eb755805 | 717 | ret = (*up)(parent, data); |
8277434e PT |
718 | if (ret || parent == from) |
719 | goto out; | |
c09595f6 PZ |
720 | |
721 | child = parent; | |
722 | parent = parent->parent; | |
723 | if (parent) | |
724 | goto up; | |
8277434e | 725 | out: |
eb755805 | 726 | return ret; |
c09595f6 PZ |
727 | } |
728 | ||
029632fb | 729 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 730 | { |
e2b245f8 | 731 | return 0; |
eb755805 | 732 | } |
18d95a28 PZ |
733 | #endif |
734 | ||
45bf76df IM |
735 | static void set_load_weight(struct task_struct *p) |
736 | { | |
f05998d4 NR |
737 | int prio = p->static_prio - MAX_RT_PRIO; |
738 | struct load_weight *load = &p->se.load; | |
739 | ||
dd41f596 IM |
740 | /* |
741 | * SCHED_IDLE tasks get minimal weight: | |
742 | */ | |
743 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 744 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 745 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
746 | return; |
747 | } | |
71f8bd46 | 748 | |
c8b28116 | 749 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 750 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
751 | } |
752 | ||
371fd7e7 | 753 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 754 | { |
a64692a3 | 755 | update_rq_clock(rq); |
dd41f596 | 756 | sched_info_queued(p); |
371fd7e7 | 757 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
758 | } |
759 | ||
371fd7e7 | 760 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 761 | { |
a64692a3 | 762 | update_rq_clock(rq); |
46ac22ba | 763 | sched_info_dequeued(p); |
371fd7e7 | 764 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
765 | } |
766 | ||
029632fb | 767 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
768 | { |
769 | if (task_contributes_to_load(p)) | |
770 | rq->nr_uninterruptible--; | |
771 | ||
371fd7e7 | 772 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
773 | } |
774 | ||
029632fb | 775 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
776 | { |
777 | if (task_contributes_to_load(p)) | |
778 | rq->nr_uninterruptible++; | |
779 | ||
371fd7e7 | 780 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
781 | } |
782 | ||
fe44d621 | 783 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 784 | { |
095c0aa8 GC |
785 | /* |
786 | * In theory, the compile should just see 0 here, and optimize out the call | |
787 | * to sched_rt_avg_update. But I don't trust it... | |
788 | */ | |
789 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
790 | s64 steal = 0, irq_delta = 0; | |
791 | #endif | |
792 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 793 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
794 | |
795 | /* | |
796 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
797 | * this case when a previous update_rq_clock() happened inside a | |
798 | * {soft,}irq region. | |
799 | * | |
800 | * When this happens, we stop ->clock_task and only update the | |
801 | * prev_irq_time stamp to account for the part that fit, so that a next | |
802 | * update will consume the rest. This ensures ->clock_task is | |
803 | * monotonic. | |
804 | * | |
805 | * It does however cause some slight miss-attribution of {soft,}irq | |
806 | * time, a more accurate solution would be to update the irq_time using | |
807 | * the current rq->clock timestamp, except that would require using | |
808 | * atomic ops. | |
809 | */ | |
810 | if (irq_delta > delta) | |
811 | irq_delta = delta; | |
812 | ||
813 | rq->prev_irq_time += irq_delta; | |
814 | delta -= irq_delta; | |
095c0aa8 GC |
815 | #endif |
816 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 817 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
818 | u64 st; |
819 | ||
820 | steal = paravirt_steal_clock(cpu_of(rq)); | |
821 | steal -= rq->prev_steal_time_rq; | |
822 | ||
823 | if (unlikely(steal > delta)) | |
824 | steal = delta; | |
825 | ||
826 | st = steal_ticks(steal); | |
827 | steal = st * TICK_NSEC; | |
828 | ||
829 | rq->prev_steal_time_rq += steal; | |
830 | ||
831 | delta -= steal; | |
832 | } | |
833 | #endif | |
834 | ||
fe44d621 PZ |
835 | rq->clock_task += delta; |
836 | ||
095c0aa8 GC |
837 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
838 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
839 | sched_rt_avg_update(rq, irq_delta + steal); | |
840 | #endif | |
aa483808 VP |
841 | } |
842 | ||
34f971f6 PZ |
843 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
844 | { | |
845 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
846 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
847 | ||
848 | if (stop) { | |
849 | /* | |
850 | * Make it appear like a SCHED_FIFO task, its something | |
851 | * userspace knows about and won't get confused about. | |
852 | * | |
853 | * Also, it will make PI more or less work without too | |
854 | * much confusion -- but then, stop work should not | |
855 | * rely on PI working anyway. | |
856 | */ | |
857 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
858 | ||
859 | stop->sched_class = &stop_sched_class; | |
860 | } | |
861 | ||
862 | cpu_rq(cpu)->stop = stop; | |
863 | ||
864 | if (old_stop) { | |
865 | /* | |
866 | * Reset it back to a normal scheduling class so that | |
867 | * it can die in pieces. | |
868 | */ | |
869 | old_stop->sched_class = &rt_sched_class; | |
870 | } | |
871 | } | |
872 | ||
14531189 | 873 | /* |
dd41f596 | 874 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 875 | */ |
14531189 IM |
876 | static inline int __normal_prio(struct task_struct *p) |
877 | { | |
dd41f596 | 878 | return p->static_prio; |
14531189 IM |
879 | } |
880 | ||
b29739f9 IM |
881 | /* |
882 | * Calculate the expected normal priority: i.e. priority | |
883 | * without taking RT-inheritance into account. Might be | |
884 | * boosted by interactivity modifiers. Changes upon fork, | |
885 | * setprio syscalls, and whenever the interactivity | |
886 | * estimator recalculates. | |
887 | */ | |
36c8b586 | 888 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
889 | { |
890 | int prio; | |
891 | ||
e05606d3 | 892 | if (task_has_rt_policy(p)) |
b29739f9 IM |
893 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
894 | else | |
895 | prio = __normal_prio(p); | |
896 | return prio; | |
897 | } | |
898 | ||
899 | /* | |
900 | * Calculate the current priority, i.e. the priority | |
901 | * taken into account by the scheduler. This value might | |
902 | * be boosted by RT tasks, or might be boosted by | |
903 | * interactivity modifiers. Will be RT if the task got | |
904 | * RT-boosted. If not then it returns p->normal_prio. | |
905 | */ | |
36c8b586 | 906 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
907 | { |
908 | p->normal_prio = normal_prio(p); | |
909 | /* | |
910 | * If we are RT tasks or we were boosted to RT priority, | |
911 | * keep the priority unchanged. Otherwise, update priority | |
912 | * to the normal priority: | |
913 | */ | |
914 | if (!rt_prio(p->prio)) | |
915 | return p->normal_prio; | |
916 | return p->prio; | |
917 | } | |
918 | ||
1da177e4 LT |
919 | /** |
920 | * task_curr - is this task currently executing on a CPU? | |
921 | * @p: the task in question. | |
922 | */ | |
36c8b586 | 923 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
924 | { |
925 | return cpu_curr(task_cpu(p)) == p; | |
926 | } | |
927 | ||
cb469845 SR |
928 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
929 | const struct sched_class *prev_class, | |
da7a735e | 930 | int oldprio) |
cb469845 SR |
931 | { |
932 | if (prev_class != p->sched_class) { | |
933 | if (prev_class->switched_from) | |
da7a735e PZ |
934 | prev_class->switched_from(rq, p); |
935 | p->sched_class->switched_to(rq, p); | |
936 | } else if (oldprio != p->prio) | |
937 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
938 | } |
939 | ||
029632fb | 940 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
941 | { |
942 | const struct sched_class *class; | |
943 | ||
944 | if (p->sched_class == rq->curr->sched_class) { | |
945 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
946 | } else { | |
947 | for_each_class(class) { | |
948 | if (class == rq->curr->sched_class) | |
949 | break; | |
950 | if (class == p->sched_class) { | |
951 | resched_task(rq->curr); | |
952 | break; | |
953 | } | |
954 | } | |
955 | } | |
956 | ||
957 | /* | |
958 | * A queue event has occurred, and we're going to schedule. In | |
959 | * this case, we can save a useless back to back clock update. | |
960 | */ | |
fd2f4419 | 961 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
962 | rq->skip_clock_update = 1; |
963 | } | |
964 | ||
582b336e MT |
965 | static ATOMIC_NOTIFIER_HEAD(task_migration_notifier); |
966 | ||
967 | void register_task_migration_notifier(struct notifier_block *n) | |
968 | { | |
969 | atomic_notifier_chain_register(&task_migration_notifier, n); | |
970 | } | |
971 | ||
1da177e4 | 972 | #ifdef CONFIG_SMP |
dd41f596 | 973 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 974 | { |
e2912009 PZ |
975 | #ifdef CONFIG_SCHED_DEBUG |
976 | /* | |
977 | * We should never call set_task_cpu() on a blocked task, | |
978 | * ttwu() will sort out the placement. | |
979 | */ | |
077614ee PZ |
980 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
981 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
982 | |
983 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
984 | /* |
985 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
986 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
987 | * | |
988 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 989 | * see task_group(). |
6c6c54e1 PZ |
990 | * |
991 | * Furthermore, all task_rq users should acquire both locks, see | |
992 | * task_rq_lock(). | |
993 | */ | |
0122ec5b PZ |
994 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
995 | lockdep_is_held(&task_rq(p)->lock))); | |
996 | #endif | |
e2912009 PZ |
997 | #endif |
998 | ||
de1d7286 | 999 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1000 | |
0c69774e | 1001 | if (task_cpu(p) != new_cpu) { |
582b336e MT |
1002 | struct task_migration_notifier tmn; |
1003 | ||
0a74bef8 PT |
1004 | if (p->sched_class->migrate_task_rq) |
1005 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1006 | p->se.nr_migrations++; |
a8b0ca17 | 1007 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
582b336e MT |
1008 | |
1009 | tmn.task = p; | |
1010 | tmn.from_cpu = task_cpu(p); | |
1011 | tmn.to_cpu = new_cpu; | |
1012 | ||
1013 | atomic_notifier_call_chain(&task_migration_notifier, 0, &tmn); | |
0c69774e | 1014 | } |
dd41f596 IM |
1015 | |
1016 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1017 | } |
1018 | ||
969c7921 | 1019 | struct migration_arg { |
36c8b586 | 1020 | struct task_struct *task; |
1da177e4 | 1021 | int dest_cpu; |
70b97a7f | 1022 | }; |
1da177e4 | 1023 | |
969c7921 TH |
1024 | static int migration_cpu_stop(void *data); |
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * wait_task_inactive - wait for a thread to unschedule. | |
1028 | * | |
85ba2d86 RM |
1029 | * If @match_state is nonzero, it's the @p->state value just checked and |
1030 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1031 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1032 | * we return a positive number (its total switch count). If a second call | |
1033 | * a short while later returns the same number, the caller can be sure that | |
1034 | * @p has remained unscheduled the whole time. | |
1035 | * | |
1da177e4 LT |
1036 | * The caller must ensure that the task *will* unschedule sometime soon, |
1037 | * else this function might spin for a *long* time. This function can't | |
1038 | * be called with interrupts off, or it may introduce deadlock with | |
1039 | * smp_call_function() if an IPI is sent by the same process we are | |
1040 | * waiting to become inactive. | |
1041 | */ | |
85ba2d86 | 1042 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1043 | { |
1044 | unsigned long flags; | |
dd41f596 | 1045 | int running, on_rq; |
85ba2d86 | 1046 | unsigned long ncsw; |
70b97a7f | 1047 | struct rq *rq; |
1da177e4 | 1048 | |
3a5c359a AK |
1049 | for (;;) { |
1050 | /* | |
1051 | * We do the initial early heuristics without holding | |
1052 | * any task-queue locks at all. We'll only try to get | |
1053 | * the runqueue lock when things look like they will | |
1054 | * work out! | |
1055 | */ | |
1056 | rq = task_rq(p); | |
fa490cfd | 1057 | |
3a5c359a AK |
1058 | /* |
1059 | * If the task is actively running on another CPU | |
1060 | * still, just relax and busy-wait without holding | |
1061 | * any locks. | |
1062 | * | |
1063 | * NOTE! Since we don't hold any locks, it's not | |
1064 | * even sure that "rq" stays as the right runqueue! | |
1065 | * But we don't care, since "task_running()" will | |
1066 | * return false if the runqueue has changed and p | |
1067 | * is actually now running somewhere else! | |
1068 | */ | |
85ba2d86 RM |
1069 | while (task_running(rq, p)) { |
1070 | if (match_state && unlikely(p->state != match_state)) | |
1071 | return 0; | |
3a5c359a | 1072 | cpu_relax(); |
85ba2d86 | 1073 | } |
fa490cfd | 1074 | |
3a5c359a AK |
1075 | /* |
1076 | * Ok, time to look more closely! We need the rq | |
1077 | * lock now, to be *sure*. If we're wrong, we'll | |
1078 | * just go back and repeat. | |
1079 | */ | |
1080 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1081 | trace_sched_wait_task(p); |
3a5c359a | 1082 | running = task_running(rq, p); |
fd2f4419 | 1083 | on_rq = p->on_rq; |
85ba2d86 | 1084 | ncsw = 0; |
f31e11d8 | 1085 | if (!match_state || p->state == match_state) |
93dcf55f | 1086 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1087 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1088 | |
85ba2d86 RM |
1089 | /* |
1090 | * If it changed from the expected state, bail out now. | |
1091 | */ | |
1092 | if (unlikely(!ncsw)) | |
1093 | break; | |
1094 | ||
3a5c359a AK |
1095 | /* |
1096 | * Was it really running after all now that we | |
1097 | * checked with the proper locks actually held? | |
1098 | * | |
1099 | * Oops. Go back and try again.. | |
1100 | */ | |
1101 | if (unlikely(running)) { | |
1102 | cpu_relax(); | |
1103 | continue; | |
1104 | } | |
fa490cfd | 1105 | |
3a5c359a AK |
1106 | /* |
1107 | * It's not enough that it's not actively running, | |
1108 | * it must be off the runqueue _entirely_, and not | |
1109 | * preempted! | |
1110 | * | |
80dd99b3 | 1111 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1112 | * running right now), it's preempted, and we should |
1113 | * yield - it could be a while. | |
1114 | */ | |
1115 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1116 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1117 | ||
1118 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1119 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1120 | continue; |
1121 | } | |
fa490cfd | 1122 | |
3a5c359a AK |
1123 | /* |
1124 | * Ahh, all good. It wasn't running, and it wasn't | |
1125 | * runnable, which means that it will never become | |
1126 | * running in the future either. We're all done! | |
1127 | */ | |
1128 | break; | |
1129 | } | |
85ba2d86 RM |
1130 | |
1131 | return ncsw; | |
1da177e4 LT |
1132 | } |
1133 | ||
1134 | /*** | |
1135 | * kick_process - kick a running thread to enter/exit the kernel | |
1136 | * @p: the to-be-kicked thread | |
1137 | * | |
1138 | * Cause a process which is running on another CPU to enter | |
1139 | * kernel-mode, without any delay. (to get signals handled.) | |
1140 | * | |
25985edc | 1141 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1142 | * because all it wants to ensure is that the remote task enters |
1143 | * the kernel. If the IPI races and the task has been migrated | |
1144 | * to another CPU then no harm is done and the purpose has been | |
1145 | * achieved as well. | |
1146 | */ | |
36c8b586 | 1147 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1148 | { |
1149 | int cpu; | |
1150 | ||
1151 | preempt_disable(); | |
1152 | cpu = task_cpu(p); | |
1153 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1154 | smp_send_reschedule(cpu); | |
1155 | preempt_enable(); | |
1156 | } | |
b43e3521 | 1157 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1158 | #endif /* CONFIG_SMP */ |
1da177e4 | 1159 | |
970b13ba | 1160 | #ifdef CONFIG_SMP |
30da688e | 1161 | /* |
013fdb80 | 1162 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1163 | */ |
5da9a0fb PZ |
1164 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1165 | { | |
aa00d89c TC |
1166 | int nid = cpu_to_node(cpu); |
1167 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1168 | enum { cpuset, possible, fail } state = cpuset; |
1169 | int dest_cpu; | |
5da9a0fb | 1170 | |
aa00d89c TC |
1171 | /* |
1172 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1173 | * will return -1. There is no cpu on the node, and we should | |
1174 | * select the cpu on the other node. | |
1175 | */ | |
1176 | if (nid != -1) { | |
1177 | nodemask = cpumask_of_node(nid); | |
1178 | ||
1179 | /* Look for allowed, online CPU in same node. */ | |
1180 | for_each_cpu(dest_cpu, nodemask) { | |
1181 | if (!cpu_online(dest_cpu)) | |
1182 | continue; | |
1183 | if (!cpu_active(dest_cpu)) | |
1184 | continue; | |
1185 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1186 | return dest_cpu; | |
1187 | } | |
2baab4e9 | 1188 | } |
5da9a0fb | 1189 | |
2baab4e9 PZ |
1190 | for (;;) { |
1191 | /* Any allowed, online CPU? */ | |
e3831edd | 1192 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1193 | if (!cpu_online(dest_cpu)) |
1194 | continue; | |
1195 | if (!cpu_active(dest_cpu)) | |
1196 | continue; | |
1197 | goto out; | |
1198 | } | |
5da9a0fb | 1199 | |
2baab4e9 PZ |
1200 | switch (state) { |
1201 | case cpuset: | |
1202 | /* No more Mr. Nice Guy. */ | |
1203 | cpuset_cpus_allowed_fallback(p); | |
1204 | state = possible; | |
1205 | break; | |
1206 | ||
1207 | case possible: | |
1208 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1209 | state = fail; | |
1210 | break; | |
1211 | ||
1212 | case fail: | |
1213 | BUG(); | |
1214 | break; | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | out: | |
1219 | if (state != cpuset) { | |
1220 | /* | |
1221 | * Don't tell them about moving exiting tasks or | |
1222 | * kernel threads (both mm NULL), since they never | |
1223 | * leave kernel. | |
1224 | */ | |
1225 | if (p->mm && printk_ratelimit()) { | |
1226 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1227 | task_pid_nr(p), p->comm, cpu); | |
1228 | } | |
5da9a0fb PZ |
1229 | } |
1230 | ||
1231 | return dest_cpu; | |
1232 | } | |
1233 | ||
e2912009 | 1234 | /* |
013fdb80 | 1235 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1236 | */ |
970b13ba | 1237 | static inline |
7608dec2 | 1238 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1239 | { |
7608dec2 | 1240 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1241 | |
1242 | /* | |
1243 | * In order not to call set_task_cpu() on a blocking task we need | |
1244 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1245 | * cpu. | |
1246 | * | |
1247 | * Since this is common to all placement strategies, this lives here. | |
1248 | * | |
1249 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1250 | * not worry about this generic constraint ] | |
1251 | */ | |
fa17b507 | 1252 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1253 | !cpu_online(cpu))) |
5da9a0fb | 1254 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1255 | |
1256 | return cpu; | |
970b13ba | 1257 | } |
09a40af5 MG |
1258 | |
1259 | static void update_avg(u64 *avg, u64 sample) | |
1260 | { | |
1261 | s64 diff = sample - *avg; | |
1262 | *avg += diff >> 3; | |
1263 | } | |
970b13ba PZ |
1264 | #endif |
1265 | ||
d7c01d27 | 1266 | static void |
b84cb5df | 1267 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1268 | { |
d7c01d27 | 1269 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1270 | struct rq *rq = this_rq(); |
1271 | ||
d7c01d27 PZ |
1272 | #ifdef CONFIG_SMP |
1273 | int this_cpu = smp_processor_id(); | |
1274 | ||
1275 | if (cpu == this_cpu) { | |
1276 | schedstat_inc(rq, ttwu_local); | |
1277 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1278 | } else { | |
1279 | struct sched_domain *sd; | |
1280 | ||
1281 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1282 | rcu_read_lock(); |
d7c01d27 PZ |
1283 | for_each_domain(this_cpu, sd) { |
1284 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1285 | schedstat_inc(sd, ttwu_wake_remote); | |
1286 | break; | |
1287 | } | |
1288 | } | |
057f3fad | 1289 | rcu_read_unlock(); |
d7c01d27 | 1290 | } |
f339b9dc PZ |
1291 | |
1292 | if (wake_flags & WF_MIGRATED) | |
1293 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1294 | ||
d7c01d27 PZ |
1295 | #endif /* CONFIG_SMP */ |
1296 | ||
1297 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1298 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1299 | |
1300 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1301 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1302 | |
d7c01d27 PZ |
1303 | #endif /* CONFIG_SCHEDSTATS */ |
1304 | } | |
1305 | ||
1306 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1307 | { | |
9ed3811a | 1308 | activate_task(rq, p, en_flags); |
fd2f4419 | 1309 | p->on_rq = 1; |
c2f7115e PZ |
1310 | |
1311 | /* if a worker is waking up, notify workqueue */ | |
1312 | if (p->flags & PF_WQ_WORKER) | |
1313 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1314 | } |
1315 | ||
23f41eeb PZ |
1316 | /* |
1317 | * Mark the task runnable and perform wakeup-preemption. | |
1318 | */ | |
89363381 | 1319 | static void |
23f41eeb | 1320 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1321 | { |
9ed3811a | 1322 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1323 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1324 | |
1325 | p->state = TASK_RUNNING; | |
1326 | #ifdef CONFIG_SMP | |
1327 | if (p->sched_class->task_woken) | |
1328 | p->sched_class->task_woken(rq, p); | |
1329 | ||
e69c6341 | 1330 | if (rq->idle_stamp) { |
9ed3811a TH |
1331 | u64 delta = rq->clock - rq->idle_stamp; |
1332 | u64 max = 2*sysctl_sched_migration_cost; | |
1333 | ||
1334 | if (delta > max) | |
1335 | rq->avg_idle = max; | |
1336 | else | |
1337 | update_avg(&rq->avg_idle, delta); | |
1338 | rq->idle_stamp = 0; | |
1339 | } | |
1340 | #endif | |
1341 | } | |
1342 | ||
c05fbafb PZ |
1343 | static void |
1344 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1345 | { | |
1346 | #ifdef CONFIG_SMP | |
1347 | if (p->sched_contributes_to_load) | |
1348 | rq->nr_uninterruptible--; | |
1349 | #endif | |
1350 | ||
1351 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1352 | ttwu_do_wakeup(rq, p, wake_flags); | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1357 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1358 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1359 | * the task is still ->on_rq. | |
1360 | */ | |
1361 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1362 | { | |
1363 | struct rq *rq; | |
1364 | int ret = 0; | |
1365 | ||
1366 | rq = __task_rq_lock(p); | |
1367 | if (p->on_rq) { | |
1368 | ttwu_do_wakeup(rq, p, wake_flags); | |
1369 | ret = 1; | |
1370 | } | |
1371 | __task_rq_unlock(rq); | |
1372 | ||
1373 | return ret; | |
1374 | } | |
1375 | ||
317f3941 | 1376 | #ifdef CONFIG_SMP |
fa14ff4a | 1377 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1378 | { |
1379 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1380 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1381 | struct task_struct *p; | |
317f3941 PZ |
1382 | |
1383 | raw_spin_lock(&rq->lock); | |
1384 | ||
fa14ff4a PZ |
1385 | while (llist) { |
1386 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1387 | llist = llist_next(llist); | |
317f3941 PZ |
1388 | ttwu_do_activate(rq, p, 0); |
1389 | } | |
1390 | ||
1391 | raw_spin_unlock(&rq->lock); | |
1392 | } | |
1393 | ||
1394 | void scheduler_ipi(void) | |
1395 | { | |
ff442c51 FW |
1396 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick() |
1397 | && !tick_nohz_full_cpu(smp_processor_id())) | |
c5d753a5 PZ |
1398 | return; |
1399 | ||
1400 | /* | |
1401 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1402 | * traditionally all their work was done from the interrupt return | |
1403 | * path. Now that we actually do some work, we need to make sure | |
1404 | * we do call them. | |
1405 | * | |
1406 | * Some archs already do call them, luckily irq_enter/exit nest | |
1407 | * properly. | |
1408 | * | |
1409 | * Arguably we should visit all archs and update all handlers, | |
1410 | * however a fair share of IPIs are still resched only so this would | |
1411 | * somewhat pessimize the simple resched case. | |
1412 | */ | |
1413 | irq_enter(); | |
ff442c51 | 1414 | tick_nohz_full_check(); |
fa14ff4a | 1415 | sched_ttwu_pending(); |
ca38062e SS |
1416 | |
1417 | /* | |
1418 | * Check if someone kicked us for doing the nohz idle load balance. | |
1419 | */ | |
6eb57e0d SS |
1420 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1421 | this_rq()->idle_balance = 1; | |
ca38062e | 1422 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1423 | } |
c5d753a5 | 1424 | irq_exit(); |
317f3941 PZ |
1425 | } |
1426 | ||
1427 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1428 | { | |
fa14ff4a | 1429 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1430 | smp_send_reschedule(cpu); |
1431 | } | |
d6aa8f85 | 1432 | |
39be3501 | 1433 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1434 | { |
1435 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1436 | } | |
d6aa8f85 | 1437 | #endif /* CONFIG_SMP */ |
317f3941 | 1438 | |
c05fbafb PZ |
1439 | static void ttwu_queue(struct task_struct *p, int cpu) |
1440 | { | |
1441 | struct rq *rq = cpu_rq(cpu); | |
1442 | ||
17d9f311 | 1443 | #if defined(CONFIG_SMP) |
39be3501 | 1444 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1445 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1446 | ttwu_queue_remote(p, cpu); |
1447 | return; | |
1448 | } | |
1449 | #endif | |
1450 | ||
c05fbafb PZ |
1451 | raw_spin_lock(&rq->lock); |
1452 | ttwu_do_activate(rq, p, 0); | |
1453 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1454 | } |
1455 | ||
1456 | /** | |
1da177e4 | 1457 | * try_to_wake_up - wake up a thread |
9ed3811a | 1458 | * @p: the thread to be awakened |
1da177e4 | 1459 | * @state: the mask of task states that can be woken |
9ed3811a | 1460 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1461 | * |
1462 | * Put it on the run-queue if it's not already there. The "current" | |
1463 | * thread is always on the run-queue (except when the actual | |
1464 | * re-schedule is in progress), and as such you're allowed to do | |
1465 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1466 | * runnable without the overhead of this. | |
1467 | * | |
9ed3811a TH |
1468 | * Returns %true if @p was woken up, %false if it was already running |
1469 | * or @state didn't match @p's state. | |
1da177e4 | 1470 | */ |
e4a52bcb PZ |
1471 | static int |
1472 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1473 | { |
1da177e4 | 1474 | unsigned long flags; |
c05fbafb | 1475 | int cpu, success = 0; |
2398f2c6 | 1476 | |
04e2f174 | 1477 | smp_wmb(); |
013fdb80 | 1478 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1479 | if (!(p->state & state)) |
1da177e4 LT |
1480 | goto out; |
1481 | ||
c05fbafb | 1482 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1483 | cpu = task_cpu(p); |
1da177e4 | 1484 | |
c05fbafb PZ |
1485 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1486 | goto stat; | |
1da177e4 | 1487 | |
1da177e4 | 1488 | #ifdef CONFIG_SMP |
e9c84311 | 1489 | /* |
c05fbafb PZ |
1490 | * If the owning (remote) cpu is still in the middle of schedule() with |
1491 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1492 | */ |
f3e94786 | 1493 | while (p->on_cpu) |
e4a52bcb | 1494 | cpu_relax(); |
0970d299 | 1495 | /* |
e4a52bcb | 1496 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1497 | */ |
e4a52bcb | 1498 | smp_rmb(); |
1da177e4 | 1499 | |
a8e4f2ea | 1500 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1501 | p->state = TASK_WAKING; |
e7693a36 | 1502 | |
e4a52bcb | 1503 | if (p->sched_class->task_waking) |
74f8e4b2 | 1504 | p->sched_class->task_waking(p); |
efbbd05a | 1505 | |
7608dec2 | 1506 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1507 | if (task_cpu(p) != cpu) { |
1508 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1509 | set_task_cpu(p, cpu); |
f339b9dc | 1510 | } |
1da177e4 | 1511 | #endif /* CONFIG_SMP */ |
1da177e4 | 1512 | |
c05fbafb PZ |
1513 | ttwu_queue(p, cpu); |
1514 | stat: | |
b84cb5df | 1515 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1516 | out: |
013fdb80 | 1517 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1518 | |
1519 | return success; | |
1520 | } | |
1521 | ||
21aa9af0 TH |
1522 | /** |
1523 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1524 | * @p: the thread to be awakened | |
1525 | * | |
2acca55e | 1526 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1527 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1528 | * the current task. |
21aa9af0 TH |
1529 | */ |
1530 | static void try_to_wake_up_local(struct task_struct *p) | |
1531 | { | |
1532 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1533 | |
383efcd0 TH |
1534 | if (WARN_ON_ONCE(rq != this_rq()) || |
1535 | WARN_ON_ONCE(p == current)) | |
1536 | return; | |
1537 | ||
21aa9af0 TH |
1538 | lockdep_assert_held(&rq->lock); |
1539 | ||
2acca55e PZ |
1540 | if (!raw_spin_trylock(&p->pi_lock)) { |
1541 | raw_spin_unlock(&rq->lock); | |
1542 | raw_spin_lock(&p->pi_lock); | |
1543 | raw_spin_lock(&rq->lock); | |
1544 | } | |
1545 | ||
21aa9af0 | 1546 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1547 | goto out; |
21aa9af0 | 1548 | |
fd2f4419 | 1549 | if (!p->on_rq) |
d7c01d27 PZ |
1550 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1551 | ||
23f41eeb | 1552 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1553 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1554 | out: |
1555 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1556 | } |
1557 | ||
50fa610a DH |
1558 | /** |
1559 | * wake_up_process - Wake up a specific process | |
1560 | * @p: The process to be woken up. | |
1561 | * | |
1562 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1563 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1564 | * running. | |
1565 | * | |
1566 | * It may be assumed that this function implies a write memory barrier before | |
1567 | * changing the task state if and only if any tasks are woken up. | |
1568 | */ | |
7ad5b3a5 | 1569 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1570 | { |
9067ac85 ON |
1571 | WARN_ON(task_is_stopped_or_traced(p)); |
1572 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1573 | } |
1da177e4 LT |
1574 | EXPORT_SYMBOL(wake_up_process); |
1575 | ||
7ad5b3a5 | 1576 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1577 | { |
1578 | return try_to_wake_up(p, state, 0); | |
1579 | } | |
1580 | ||
1da177e4 LT |
1581 | /* |
1582 | * Perform scheduler related setup for a newly forked process p. | |
1583 | * p is forked by current. | |
dd41f596 IM |
1584 | * |
1585 | * __sched_fork() is basic setup used by init_idle() too: | |
1586 | */ | |
1587 | static void __sched_fork(struct task_struct *p) | |
1588 | { | |
fd2f4419 PZ |
1589 | p->on_rq = 0; |
1590 | ||
1591 | p->se.on_rq = 0; | |
dd41f596 IM |
1592 | p->se.exec_start = 0; |
1593 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1594 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1595 | p->se.nr_migrations = 0; |
da7a735e | 1596 | p->se.vruntime = 0; |
fd2f4419 | 1597 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 1598 | |
f4e26b12 PT |
1599 | /* |
1600 | * Load-tracking only depends on SMP, FAIR_GROUP_SCHED dependency below may be | |
1601 | * removed when useful for applications beyond shares distribution (e.g. | |
1602 | * load-balance). | |
1603 | */ | |
1604 | #if defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED) | |
9d85f21c PT |
1605 | p->se.avg.runnable_avg_period = 0; |
1606 | p->se.avg.runnable_avg_sum = 0; | |
1607 | #endif | |
6cfb0d5d | 1608 | #ifdef CONFIG_SCHEDSTATS |
41acab88 | 1609 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1610 | #endif |
476d139c | 1611 | |
fa717060 | 1612 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1613 | |
e107be36 AK |
1614 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1615 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1616 | #endif | |
cbee9f88 PZ |
1617 | |
1618 | #ifdef CONFIG_NUMA_BALANCING | |
1619 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
1620 | p->mm->numa_next_scan = jiffies; | |
b8593bfd | 1621 | p->mm->numa_next_reset = jiffies; |
cbee9f88 PZ |
1622 | p->mm->numa_scan_seq = 0; |
1623 | } | |
1624 | ||
1625 | p->node_stamp = 0ULL; | |
1626 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
1627 | p->numa_migrate_seq = p->mm ? p->mm->numa_scan_seq - 1 : 0; | |
4b96a29b | 1628 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 PZ |
1629 | p->numa_work.next = &p->numa_work; |
1630 | #endif /* CONFIG_NUMA_BALANCING */ | |
dd41f596 IM |
1631 | } |
1632 | ||
1a687c2e | 1633 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1634 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1635 | void set_numabalancing_state(bool enabled) |
1636 | { | |
1637 | if (enabled) | |
1638 | sched_feat_set("NUMA"); | |
1639 | else | |
1640 | sched_feat_set("NO_NUMA"); | |
1641 | } | |
3105b86a MG |
1642 | #else |
1643 | __read_mostly bool numabalancing_enabled; | |
1644 | ||
1645 | void set_numabalancing_state(bool enabled) | |
1646 | { | |
1647 | numabalancing_enabled = enabled; | |
dd41f596 | 1648 | } |
3105b86a | 1649 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1650 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1651 | |
1652 | /* | |
1653 | * fork()/clone()-time setup: | |
1654 | */ | |
3e51e3ed | 1655 | void sched_fork(struct task_struct *p) |
dd41f596 | 1656 | { |
0122ec5b | 1657 | unsigned long flags; |
dd41f596 IM |
1658 | int cpu = get_cpu(); |
1659 | ||
1660 | __sched_fork(p); | |
06b83b5f | 1661 | /* |
0017d735 | 1662 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1663 | * nobody will actually run it, and a signal or other external |
1664 | * event cannot wake it up and insert it on the runqueue either. | |
1665 | */ | |
0017d735 | 1666 | p->state = TASK_RUNNING; |
dd41f596 | 1667 | |
c350a04e MG |
1668 | /* |
1669 | * Make sure we do not leak PI boosting priority to the child. | |
1670 | */ | |
1671 | p->prio = current->normal_prio; | |
1672 | ||
b9dc29e7 MG |
1673 | /* |
1674 | * Revert to default priority/policy on fork if requested. | |
1675 | */ | |
1676 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1677 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1678 | p->policy = SCHED_NORMAL; |
6c697bdf | 1679 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1680 | p->rt_priority = 0; |
1681 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1682 | p->static_prio = NICE_TO_PRIO(0); | |
1683 | ||
1684 | p->prio = p->normal_prio = __normal_prio(p); | |
1685 | set_load_weight(p); | |
6c697bdf | 1686 | |
b9dc29e7 MG |
1687 | /* |
1688 | * We don't need the reset flag anymore after the fork. It has | |
1689 | * fulfilled its duty: | |
1690 | */ | |
1691 | p->sched_reset_on_fork = 0; | |
1692 | } | |
ca94c442 | 1693 | |
2ddbf952 HS |
1694 | if (!rt_prio(p->prio)) |
1695 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1696 | |
cd29fe6f PZ |
1697 | if (p->sched_class->task_fork) |
1698 | p->sched_class->task_fork(p); | |
1699 | ||
86951599 PZ |
1700 | /* |
1701 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1702 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1703 | * is ran before sched_fork(). | |
1704 | * | |
1705 | * Silence PROVE_RCU. | |
1706 | */ | |
0122ec5b | 1707 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1708 | set_task_cpu(p, cpu); |
0122ec5b | 1709 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1710 | |
52f17b6c | 1711 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1712 | if (likely(sched_info_on())) |
52f17b6c | 1713 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1714 | #endif |
3ca7a440 PZ |
1715 | #if defined(CONFIG_SMP) |
1716 | p->on_cpu = 0; | |
4866cde0 | 1717 | #endif |
bdd4e85d | 1718 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1719 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1720 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1721 | #endif |
806c09a7 | 1722 | #ifdef CONFIG_SMP |
917b627d | 1723 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1724 | #endif |
917b627d | 1725 | |
476d139c | 1726 | put_cpu(); |
1da177e4 LT |
1727 | } |
1728 | ||
1729 | /* | |
1730 | * wake_up_new_task - wake up a newly created task for the first time. | |
1731 | * | |
1732 | * This function will do some initial scheduler statistics housekeeping | |
1733 | * that must be done for every newly created context, then puts the task | |
1734 | * on the runqueue and wakes it. | |
1735 | */ | |
3e51e3ed | 1736 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1737 | { |
1738 | unsigned long flags; | |
dd41f596 | 1739 | struct rq *rq; |
fabf318e | 1740 | |
ab2515c4 | 1741 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1742 | #ifdef CONFIG_SMP |
1743 | /* | |
1744 | * Fork balancing, do it here and not earlier because: | |
1745 | * - cpus_allowed can change in the fork path | |
1746 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1747 | */ |
ab2515c4 | 1748 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1749 | #endif |
1750 | ||
ab2515c4 | 1751 | rq = __task_rq_lock(p); |
cd29fe6f | 1752 | activate_task(rq, p, 0); |
fd2f4419 | 1753 | p->on_rq = 1; |
89363381 | 1754 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1755 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1756 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1757 | if (p->sched_class->task_woken) |
1758 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1759 | #endif |
0122ec5b | 1760 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1761 | } |
1762 | ||
e107be36 AK |
1763 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1764 | ||
1765 | /** | |
80dd99b3 | 1766 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1767 | * @notifier: notifier struct to register |
e107be36 AK |
1768 | */ |
1769 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1770 | { | |
1771 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1772 | } | |
1773 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1774 | ||
1775 | /** | |
1776 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1777 | * @notifier: notifier struct to unregister |
e107be36 AK |
1778 | * |
1779 | * This is safe to call from within a preemption notifier. | |
1780 | */ | |
1781 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1782 | { | |
1783 | hlist_del(¬ifier->link); | |
1784 | } | |
1785 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1786 | ||
1787 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1788 | { | |
1789 | struct preempt_notifier *notifier; | |
e107be36 | 1790 | |
b67bfe0d | 1791 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1792 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
1793 | } | |
1794 | ||
1795 | static void | |
1796 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1797 | struct task_struct *next) | |
1798 | { | |
1799 | struct preempt_notifier *notifier; | |
e107be36 | 1800 | |
b67bfe0d | 1801 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1802 | notifier->ops->sched_out(notifier, next); |
1803 | } | |
1804 | ||
6d6bc0ad | 1805 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1806 | |
1807 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1808 | { | |
1809 | } | |
1810 | ||
1811 | static void | |
1812 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1813 | struct task_struct *next) | |
1814 | { | |
1815 | } | |
1816 | ||
6d6bc0ad | 1817 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1818 | |
4866cde0 NP |
1819 | /** |
1820 | * prepare_task_switch - prepare to switch tasks | |
1821 | * @rq: the runqueue preparing to switch | |
421cee29 | 1822 | * @prev: the current task that is being switched out |
4866cde0 NP |
1823 | * @next: the task we are going to switch to. |
1824 | * | |
1825 | * This is called with the rq lock held and interrupts off. It must | |
1826 | * be paired with a subsequent finish_task_switch after the context | |
1827 | * switch. | |
1828 | * | |
1829 | * prepare_task_switch sets up locking and calls architecture specific | |
1830 | * hooks. | |
1831 | */ | |
e107be36 AK |
1832 | static inline void |
1833 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1834 | struct task_struct *next) | |
4866cde0 | 1835 | { |
895dd92c | 1836 | trace_sched_switch(prev, next); |
fe4b04fa PZ |
1837 | sched_info_switch(prev, next); |
1838 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1839 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1840 | prepare_lock_switch(rq, next); |
1841 | prepare_arch_switch(next); | |
1842 | } | |
1843 | ||
1da177e4 LT |
1844 | /** |
1845 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1846 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1847 | * @prev: the thread we just switched away from. |
1848 | * | |
4866cde0 NP |
1849 | * finish_task_switch must be called after the context switch, paired |
1850 | * with a prepare_task_switch call before the context switch. | |
1851 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1852 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1853 | * |
1854 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1855 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1856 | * with the lock held can cause deadlocks; see schedule() for |
1857 | * details.) | |
1858 | */ | |
a9957449 | 1859 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1860 | __releases(rq->lock) |
1861 | { | |
1da177e4 | 1862 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1863 | long prev_state; |
1da177e4 LT |
1864 | |
1865 | rq->prev_mm = NULL; | |
1866 | ||
1867 | /* | |
1868 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1869 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1870 | * schedule one last time. The schedule call will never return, and |
1871 | * the scheduled task must drop that reference. | |
c394cc9f | 1872 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1873 | * still held, otherwise prev could be scheduled on another cpu, die |
1874 | * there before we look at prev->state, and then the reference would | |
1875 | * be dropped twice. | |
1876 | * Manfred Spraul <manfred@colorfullife.com> | |
1877 | */ | |
55a101f8 | 1878 | prev_state = prev->state; |
bf9fae9f | 1879 | vtime_task_switch(prev); |
4866cde0 | 1880 | finish_arch_switch(prev); |
a8d757ef | 1881 | perf_event_task_sched_in(prev, current); |
4866cde0 | 1882 | finish_lock_switch(rq, prev); |
01f23e16 | 1883 | finish_arch_post_lock_switch(); |
e8fa1362 | 1884 | |
e107be36 | 1885 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1886 | if (mm) |
1887 | mmdrop(mm); | |
c394cc9f | 1888 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1889 | /* |
1890 | * Remove function-return probe instances associated with this | |
1891 | * task and put them back on the free list. | |
9761eea8 | 1892 | */ |
c6fd91f0 | 1893 | kprobe_flush_task(prev); |
1da177e4 | 1894 | put_task_struct(prev); |
c6fd91f0 | 1895 | } |
99e5ada9 FW |
1896 | |
1897 | tick_nohz_task_switch(current); | |
1da177e4 LT |
1898 | } |
1899 | ||
3f029d3c GH |
1900 | #ifdef CONFIG_SMP |
1901 | ||
1902 | /* assumes rq->lock is held */ | |
1903 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1904 | { | |
1905 | if (prev->sched_class->pre_schedule) | |
1906 | prev->sched_class->pre_schedule(rq, prev); | |
1907 | } | |
1908 | ||
1909 | /* rq->lock is NOT held, but preemption is disabled */ | |
1910 | static inline void post_schedule(struct rq *rq) | |
1911 | { | |
1912 | if (rq->post_schedule) { | |
1913 | unsigned long flags; | |
1914 | ||
05fa785c | 1915 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1916 | if (rq->curr->sched_class->post_schedule) |
1917 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1918 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1919 | |
1920 | rq->post_schedule = 0; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | #else | |
da19ab51 | 1925 | |
3f029d3c GH |
1926 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
1927 | { | |
1928 | } | |
1929 | ||
1930 | static inline void post_schedule(struct rq *rq) | |
1931 | { | |
1da177e4 LT |
1932 | } |
1933 | ||
3f029d3c GH |
1934 | #endif |
1935 | ||
1da177e4 LT |
1936 | /** |
1937 | * schedule_tail - first thing a freshly forked thread must call. | |
1938 | * @prev: the thread we just switched away from. | |
1939 | */ | |
36c8b586 | 1940 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1941 | __releases(rq->lock) |
1942 | { | |
70b97a7f IM |
1943 | struct rq *rq = this_rq(); |
1944 | ||
4866cde0 | 1945 | finish_task_switch(rq, prev); |
da19ab51 | 1946 | |
3f029d3c GH |
1947 | /* |
1948 | * FIXME: do we need to worry about rq being invalidated by the | |
1949 | * task_switch? | |
1950 | */ | |
1951 | post_schedule(rq); | |
70b97a7f | 1952 | |
4866cde0 NP |
1953 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
1954 | /* In this case, finish_task_switch does not reenable preemption */ | |
1955 | preempt_enable(); | |
1956 | #endif | |
1da177e4 | 1957 | if (current->set_child_tid) |
b488893a | 1958 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1959 | } |
1960 | ||
1961 | /* | |
1962 | * context_switch - switch to the new MM and the new | |
1963 | * thread's register state. | |
1964 | */ | |
dd41f596 | 1965 | static inline void |
70b97a7f | 1966 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1967 | struct task_struct *next) |
1da177e4 | 1968 | { |
dd41f596 | 1969 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1970 | |
e107be36 | 1971 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 1972 | |
dd41f596 IM |
1973 | mm = next->mm; |
1974 | oldmm = prev->active_mm; | |
9226d125 ZA |
1975 | /* |
1976 | * For paravirt, this is coupled with an exit in switch_to to | |
1977 | * combine the page table reload and the switch backend into | |
1978 | * one hypercall. | |
1979 | */ | |
224101ed | 1980 | arch_start_context_switch(prev); |
9226d125 | 1981 | |
31915ab4 | 1982 | if (!mm) { |
1da177e4 LT |
1983 | next->active_mm = oldmm; |
1984 | atomic_inc(&oldmm->mm_count); | |
1985 | enter_lazy_tlb(oldmm, next); | |
1986 | } else | |
1987 | switch_mm(oldmm, mm, next); | |
1988 | ||
31915ab4 | 1989 | if (!prev->mm) { |
1da177e4 | 1990 | prev->active_mm = NULL; |
1da177e4 LT |
1991 | rq->prev_mm = oldmm; |
1992 | } | |
3a5f5e48 IM |
1993 | /* |
1994 | * Since the runqueue lock will be released by the next | |
1995 | * task (which is an invalid locking op but in the case | |
1996 | * of the scheduler it's an obvious special-case), so we | |
1997 | * do an early lockdep release here: | |
1998 | */ | |
1999 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2000 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2001 | #endif |
1da177e4 | 2002 | |
91d1aa43 | 2003 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2004 | /* Here we just switch the register state and the stack. */ |
2005 | switch_to(prev, next, prev); | |
2006 | ||
dd41f596 IM |
2007 | barrier(); |
2008 | /* | |
2009 | * this_rq must be evaluated again because prev may have moved | |
2010 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2011 | * frame will be invalid. | |
2012 | */ | |
2013 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2014 | } |
2015 | ||
2016 | /* | |
1c3e8264 | 2017 | * nr_running and nr_context_switches: |
1da177e4 LT |
2018 | * |
2019 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2020 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2021 | */ |
2022 | unsigned long nr_running(void) | |
2023 | { | |
2024 | unsigned long i, sum = 0; | |
2025 | ||
2026 | for_each_online_cpu(i) | |
2027 | sum += cpu_rq(i)->nr_running; | |
2028 | ||
2029 | return sum; | |
f711f609 | 2030 | } |
1da177e4 | 2031 | |
1da177e4 | 2032 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2033 | { |
cc94abfc SR |
2034 | int i; |
2035 | unsigned long long sum = 0; | |
46cb4b7c | 2036 | |
0a945022 | 2037 | for_each_possible_cpu(i) |
1da177e4 | 2038 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2039 | |
1da177e4 LT |
2040 | return sum; |
2041 | } | |
483b4ee6 | 2042 | |
1da177e4 LT |
2043 | unsigned long nr_iowait(void) |
2044 | { | |
2045 | unsigned long i, sum = 0; | |
483b4ee6 | 2046 | |
0a945022 | 2047 | for_each_possible_cpu(i) |
1da177e4 | 2048 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2049 | |
1da177e4 LT |
2050 | return sum; |
2051 | } | |
483b4ee6 | 2052 | |
8c215bd3 | 2053 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2054 | { |
8c215bd3 | 2055 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2056 | return atomic_read(&this->nr_iowait); |
2057 | } | |
46cb4b7c | 2058 | |
dd41f596 | 2059 | #ifdef CONFIG_SMP |
8a0be9ef | 2060 | |
46cb4b7c | 2061 | /* |
38022906 PZ |
2062 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2063 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2064 | */ |
38022906 | 2065 | void sched_exec(void) |
46cb4b7c | 2066 | { |
38022906 | 2067 | struct task_struct *p = current; |
1da177e4 | 2068 | unsigned long flags; |
0017d735 | 2069 | int dest_cpu; |
46cb4b7c | 2070 | |
8f42ced9 | 2071 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2072 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2073 | if (dest_cpu == smp_processor_id()) |
2074 | goto unlock; | |
38022906 | 2075 | |
8f42ced9 | 2076 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2077 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2078 | |
8f42ced9 PZ |
2079 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2080 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2081 | return; |
2082 | } | |
0017d735 | 2083 | unlock: |
8f42ced9 | 2084 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2085 | } |
dd41f596 | 2086 | |
1da177e4 LT |
2087 | #endif |
2088 | ||
1da177e4 | 2089 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2090 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2091 | |
2092 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2093 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2094 | |
2095 | /* | |
c5f8d995 | 2096 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2097 | * @p in case that task is currently running. |
c5f8d995 HS |
2098 | * |
2099 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2100 | */ |
c5f8d995 HS |
2101 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2102 | { | |
2103 | u64 ns = 0; | |
2104 | ||
2105 | if (task_current(rq, p)) { | |
2106 | update_rq_clock(rq); | |
305e6835 | 2107 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2108 | if ((s64)ns < 0) |
2109 | ns = 0; | |
2110 | } | |
2111 | ||
2112 | return ns; | |
2113 | } | |
2114 | ||
bb34d92f | 2115 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2116 | { |
1da177e4 | 2117 | unsigned long flags; |
41b86e9c | 2118 | struct rq *rq; |
bb34d92f | 2119 | u64 ns = 0; |
48f24c4d | 2120 | |
41b86e9c | 2121 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2122 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2123 | task_rq_unlock(rq, p, &flags); |
1508487e | 2124 | |
c5f8d995 HS |
2125 | return ns; |
2126 | } | |
f06febc9 | 2127 | |
c5f8d995 HS |
2128 | /* |
2129 | * Return accounted runtime for the task. | |
2130 | * In case the task is currently running, return the runtime plus current's | |
2131 | * pending runtime that have not been accounted yet. | |
2132 | */ | |
2133 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2134 | { | |
2135 | unsigned long flags; | |
2136 | struct rq *rq; | |
2137 | u64 ns = 0; | |
2138 | ||
2139 | rq = task_rq_lock(p, &flags); | |
2140 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2141 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2142 | |
2143 | return ns; | |
2144 | } | |
48f24c4d | 2145 | |
7835b98b CL |
2146 | /* |
2147 | * This function gets called by the timer code, with HZ frequency. | |
2148 | * We call it with interrupts disabled. | |
7835b98b CL |
2149 | */ |
2150 | void scheduler_tick(void) | |
2151 | { | |
7835b98b CL |
2152 | int cpu = smp_processor_id(); |
2153 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2154 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2155 | |
2156 | sched_clock_tick(); | |
dd41f596 | 2157 | |
05fa785c | 2158 | raw_spin_lock(&rq->lock); |
3e51f33f | 2159 | update_rq_clock(rq); |
fdf3e95d | 2160 | update_cpu_load_active(rq); |
fa85ae24 | 2161 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 2162 | raw_spin_unlock(&rq->lock); |
7835b98b | 2163 | |
e9d2b064 | 2164 | perf_event_task_tick(); |
e220d2dc | 2165 | |
e418e1c2 | 2166 | #ifdef CONFIG_SMP |
6eb57e0d | 2167 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2168 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2169 | #endif |
265f22a9 | 2170 | rq_last_tick_reset(rq); |
1da177e4 LT |
2171 | } |
2172 | ||
265f22a9 FW |
2173 | #ifdef CONFIG_NO_HZ_FULL |
2174 | /** | |
2175 | * scheduler_tick_max_deferment | |
2176 | * | |
2177 | * Keep at least one tick per second when a single | |
2178 | * active task is running because the scheduler doesn't | |
2179 | * yet completely support full dynticks environment. | |
2180 | * | |
2181 | * This makes sure that uptime, CFS vruntime, load | |
2182 | * balancing, etc... continue to move forward, even | |
2183 | * with a very low granularity. | |
2184 | */ | |
2185 | u64 scheduler_tick_max_deferment(void) | |
2186 | { | |
2187 | struct rq *rq = this_rq(); | |
2188 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2189 | ||
2190 | next = rq->last_sched_tick + HZ; | |
2191 | ||
2192 | if (time_before_eq(next, now)) | |
2193 | return 0; | |
2194 | ||
2195 | return jiffies_to_usecs(next - now) * NSEC_PER_USEC; | |
1da177e4 | 2196 | } |
265f22a9 | 2197 | #endif |
1da177e4 | 2198 | |
132380a0 | 2199 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2200 | { |
2201 | if (in_lock_functions(addr)) { | |
2202 | addr = CALLER_ADDR2; | |
2203 | if (in_lock_functions(addr)) | |
2204 | addr = CALLER_ADDR3; | |
2205 | } | |
2206 | return addr; | |
2207 | } | |
1da177e4 | 2208 | |
7e49fcce SR |
2209 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2210 | defined(CONFIG_PREEMPT_TRACER)) | |
2211 | ||
43627582 | 2212 | void __kprobes add_preempt_count(int val) |
1da177e4 | 2213 | { |
6cd8a4bb | 2214 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2215 | /* |
2216 | * Underflow? | |
2217 | */ | |
9a11b49a IM |
2218 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2219 | return; | |
6cd8a4bb | 2220 | #endif |
1da177e4 | 2221 | preempt_count() += val; |
6cd8a4bb | 2222 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2223 | /* |
2224 | * Spinlock count overflowing soon? | |
2225 | */ | |
33859f7f MOS |
2226 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2227 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2228 | #endif |
2229 | if (preempt_count() == val) | |
2230 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2231 | } |
2232 | EXPORT_SYMBOL(add_preempt_count); | |
2233 | ||
43627582 | 2234 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 2235 | { |
6cd8a4bb | 2236 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2237 | /* |
2238 | * Underflow? | |
2239 | */ | |
01e3eb82 | 2240 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2241 | return; |
1da177e4 LT |
2242 | /* |
2243 | * Is the spinlock portion underflowing? | |
2244 | */ | |
9a11b49a IM |
2245 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2246 | !(preempt_count() & PREEMPT_MASK))) | |
2247 | return; | |
6cd8a4bb | 2248 | #endif |
9a11b49a | 2249 | |
6cd8a4bb SR |
2250 | if (preempt_count() == val) |
2251 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
2252 | preempt_count() -= val; |
2253 | } | |
2254 | EXPORT_SYMBOL(sub_preempt_count); | |
2255 | ||
2256 | #endif | |
2257 | ||
2258 | /* | |
dd41f596 | 2259 | * Print scheduling while atomic bug: |
1da177e4 | 2260 | */ |
dd41f596 | 2261 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2262 | { |
664dfa65 DJ |
2263 | if (oops_in_progress) |
2264 | return; | |
2265 | ||
3df0fc5b PZ |
2266 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2267 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2268 | |
dd41f596 | 2269 | debug_show_held_locks(prev); |
e21f5b15 | 2270 | print_modules(); |
dd41f596 IM |
2271 | if (irqs_disabled()) |
2272 | print_irqtrace_events(prev); | |
6135fc1e | 2273 | dump_stack(); |
373d4d09 | 2274 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2275 | } |
1da177e4 | 2276 | |
dd41f596 IM |
2277 | /* |
2278 | * Various schedule()-time debugging checks and statistics: | |
2279 | */ | |
2280 | static inline void schedule_debug(struct task_struct *prev) | |
2281 | { | |
1da177e4 | 2282 | /* |
41a2d6cf | 2283 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
2284 | * schedule() atomically, we ignore that path for now. |
2285 | * Otherwise, whine if we are scheduling when we should not be. | |
2286 | */ | |
3f33a7ce | 2287 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 2288 | __schedule_bug(prev); |
b3fbab05 | 2289 | rcu_sleep_check(); |
dd41f596 | 2290 | |
1da177e4 LT |
2291 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2292 | ||
2d72376b | 2293 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2294 | } |
2295 | ||
6cecd084 | 2296 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2297 | { |
61eadef6 | 2298 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2299 | update_rq_clock(rq); |
6cecd084 | 2300 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2301 | } |
2302 | ||
dd41f596 IM |
2303 | /* |
2304 | * Pick up the highest-prio task: | |
2305 | */ | |
2306 | static inline struct task_struct * | |
b67802ea | 2307 | pick_next_task(struct rq *rq) |
dd41f596 | 2308 | { |
5522d5d5 | 2309 | const struct sched_class *class; |
dd41f596 | 2310 | struct task_struct *p; |
1da177e4 LT |
2311 | |
2312 | /* | |
dd41f596 IM |
2313 | * Optimization: we know that if all tasks are in |
2314 | * the fair class we can call that function directly: | |
1da177e4 | 2315 | */ |
953bfcd1 | 2316 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2317 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2318 | if (likely(p)) |
2319 | return p; | |
1da177e4 LT |
2320 | } |
2321 | ||
34f971f6 | 2322 | for_each_class(class) { |
fb8d4724 | 2323 | p = class->pick_next_task(rq); |
dd41f596 IM |
2324 | if (p) |
2325 | return p; | |
dd41f596 | 2326 | } |
34f971f6 PZ |
2327 | |
2328 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2329 | } |
1da177e4 | 2330 | |
dd41f596 | 2331 | /* |
c259e01a | 2332 | * __schedule() is the main scheduler function. |
edde96ea PE |
2333 | * |
2334 | * The main means of driving the scheduler and thus entering this function are: | |
2335 | * | |
2336 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2337 | * | |
2338 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2339 | * paths. For example, see arch/x86/entry_64.S. | |
2340 | * | |
2341 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2342 | * interrupt handler scheduler_tick(). | |
2343 | * | |
2344 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2345 | * task to the run-queue and that's it. | |
2346 | * | |
2347 | * Now, if the new task added to the run-queue preempts the current | |
2348 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2349 | * called on the nearest possible occasion: | |
2350 | * | |
2351 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2352 | * | |
2353 | * - in syscall or exception context, at the next outmost | |
2354 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2355 | * spin_unlock()!) | |
2356 | * | |
2357 | * - in IRQ context, return from interrupt-handler to | |
2358 | * preemptible context | |
2359 | * | |
2360 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2361 | * then at the next: | |
2362 | * | |
2363 | * - cond_resched() call | |
2364 | * - explicit schedule() call | |
2365 | * - return from syscall or exception to user-space | |
2366 | * - return from interrupt-handler to user-space | |
dd41f596 | 2367 | */ |
c259e01a | 2368 | static void __sched __schedule(void) |
dd41f596 IM |
2369 | { |
2370 | struct task_struct *prev, *next; | |
67ca7bde | 2371 | unsigned long *switch_count; |
dd41f596 | 2372 | struct rq *rq; |
31656519 | 2373 | int cpu; |
dd41f596 | 2374 | |
ff743345 PZ |
2375 | need_resched: |
2376 | preempt_disable(); | |
dd41f596 IM |
2377 | cpu = smp_processor_id(); |
2378 | rq = cpu_rq(cpu); | |
25502a6c | 2379 | rcu_note_context_switch(cpu); |
dd41f596 | 2380 | prev = rq->curr; |
dd41f596 | 2381 | |
dd41f596 | 2382 | schedule_debug(prev); |
1da177e4 | 2383 | |
31656519 | 2384 | if (sched_feat(HRTICK)) |
f333fdc9 | 2385 | hrtick_clear(rq); |
8f4d37ec | 2386 | |
05fa785c | 2387 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2388 | |
246d86b5 | 2389 | switch_count = &prev->nivcsw; |
1da177e4 | 2390 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2391 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2392 | prev->state = TASK_RUNNING; |
21aa9af0 | 2393 | } else { |
2acca55e PZ |
2394 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2395 | prev->on_rq = 0; | |
2396 | ||
21aa9af0 | 2397 | /* |
2acca55e PZ |
2398 | * If a worker went to sleep, notify and ask workqueue |
2399 | * whether it wants to wake up a task to maintain | |
2400 | * concurrency. | |
21aa9af0 TH |
2401 | */ |
2402 | if (prev->flags & PF_WQ_WORKER) { | |
2403 | struct task_struct *to_wakeup; | |
2404 | ||
2405 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2406 | if (to_wakeup) | |
2407 | try_to_wake_up_local(to_wakeup); | |
2408 | } | |
21aa9af0 | 2409 | } |
dd41f596 | 2410 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2411 | } |
2412 | ||
3f029d3c | 2413 | pre_schedule(rq, prev); |
f65eda4f | 2414 | |
dd41f596 | 2415 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2416 | idle_balance(cpu, rq); |
1da177e4 | 2417 | |
df1c99d4 | 2418 | put_prev_task(rq, prev); |
b67802ea | 2419 | next = pick_next_task(rq); |
f26f9aff MG |
2420 | clear_tsk_need_resched(prev); |
2421 | rq->skip_clock_update = 0; | |
1da177e4 | 2422 | |
1da177e4 | 2423 | if (likely(prev != next)) { |
1da177e4 LT |
2424 | rq->nr_switches++; |
2425 | rq->curr = next; | |
2426 | ++*switch_count; | |
2427 | ||
dd41f596 | 2428 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2429 | /* |
246d86b5 ON |
2430 | * The context switch have flipped the stack from under us |
2431 | * and restored the local variables which were saved when | |
2432 | * this task called schedule() in the past. prev == current | |
2433 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2434 | */ |
2435 | cpu = smp_processor_id(); | |
2436 | rq = cpu_rq(cpu); | |
1da177e4 | 2437 | } else |
05fa785c | 2438 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2439 | |
3f029d3c | 2440 | post_schedule(rq); |
1da177e4 | 2441 | |
ba74c144 | 2442 | sched_preempt_enable_no_resched(); |
ff743345 | 2443 | if (need_resched()) |
1da177e4 LT |
2444 | goto need_resched; |
2445 | } | |
c259e01a | 2446 | |
9c40cef2 TG |
2447 | static inline void sched_submit_work(struct task_struct *tsk) |
2448 | { | |
3c7d5184 | 2449 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2450 | return; |
2451 | /* | |
2452 | * If we are going to sleep and we have plugged IO queued, | |
2453 | * make sure to submit it to avoid deadlocks. | |
2454 | */ | |
2455 | if (blk_needs_flush_plug(tsk)) | |
2456 | blk_schedule_flush_plug(tsk); | |
2457 | } | |
2458 | ||
6ebbe7a0 | 2459 | asmlinkage void __sched schedule(void) |
c259e01a | 2460 | { |
9c40cef2 TG |
2461 | struct task_struct *tsk = current; |
2462 | ||
2463 | sched_submit_work(tsk); | |
c259e01a TG |
2464 | __schedule(); |
2465 | } | |
1da177e4 LT |
2466 | EXPORT_SYMBOL(schedule); |
2467 | ||
91d1aa43 | 2468 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2469 | asmlinkage void __sched schedule_user(void) |
2470 | { | |
2471 | /* | |
2472 | * If we come here after a random call to set_need_resched(), | |
2473 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2474 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2475 | * we find a better solution. | |
2476 | */ | |
91d1aa43 | 2477 | user_exit(); |
20ab65e3 | 2478 | schedule(); |
91d1aa43 | 2479 | user_enter(); |
20ab65e3 FW |
2480 | } |
2481 | #endif | |
2482 | ||
c5491ea7 TG |
2483 | /** |
2484 | * schedule_preempt_disabled - called with preemption disabled | |
2485 | * | |
2486 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2487 | */ | |
2488 | void __sched schedule_preempt_disabled(void) | |
2489 | { | |
ba74c144 | 2490 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2491 | schedule(); |
2492 | preempt_disable(); | |
2493 | } | |
2494 | ||
1da177e4 LT |
2495 | #ifdef CONFIG_PREEMPT |
2496 | /* | |
2ed6e34f | 2497 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2498 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2499 | * occur there and call schedule directly. |
2500 | */ | |
d1f74e20 | 2501 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
2502 | { |
2503 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 2504 | |
1da177e4 LT |
2505 | /* |
2506 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2507 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2508 | */ |
beed33a8 | 2509 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
2510 | return; |
2511 | ||
3a5c359a | 2512 | do { |
d1f74e20 | 2513 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 2514 | __schedule(); |
d1f74e20 | 2515 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 2516 | |
3a5c359a AK |
2517 | /* |
2518 | * Check again in case we missed a preemption opportunity | |
2519 | * between schedule and now. | |
2520 | */ | |
2521 | barrier(); | |
5ed0cec0 | 2522 | } while (need_resched()); |
1da177e4 | 2523 | } |
1da177e4 LT |
2524 | EXPORT_SYMBOL(preempt_schedule); |
2525 | ||
2526 | /* | |
2ed6e34f | 2527 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2528 | * off of irq context. |
2529 | * Note, that this is called and return with irqs disabled. This will | |
2530 | * protect us against recursive calling from irq. | |
2531 | */ | |
2532 | asmlinkage void __sched preempt_schedule_irq(void) | |
2533 | { | |
2534 | struct thread_info *ti = current_thread_info(); | |
b22366cd | 2535 | enum ctx_state prev_state; |
6478d880 | 2536 | |
2ed6e34f | 2537 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
2538 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
2539 | ||
b22366cd FW |
2540 | prev_state = exception_enter(); |
2541 | ||
3a5c359a AK |
2542 | do { |
2543 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 2544 | local_irq_enable(); |
c259e01a | 2545 | __schedule(); |
3a5c359a | 2546 | local_irq_disable(); |
3a5c359a | 2547 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 2548 | |
3a5c359a AK |
2549 | /* |
2550 | * Check again in case we missed a preemption opportunity | |
2551 | * between schedule and now. | |
2552 | */ | |
2553 | barrier(); | |
5ed0cec0 | 2554 | } while (need_resched()); |
b22366cd FW |
2555 | |
2556 | exception_exit(prev_state); | |
1da177e4 LT |
2557 | } |
2558 | ||
2559 | #endif /* CONFIG_PREEMPT */ | |
2560 | ||
63859d4f | 2561 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2562 | void *key) |
1da177e4 | 2563 | { |
63859d4f | 2564 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2565 | } |
1da177e4 LT |
2566 | EXPORT_SYMBOL(default_wake_function); |
2567 | ||
2568 | /* | |
41a2d6cf IM |
2569 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
2570 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
2571 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
2572 | * | |
2573 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 2574 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
2575 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
2576 | */ | |
78ddb08f | 2577 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 2578 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 2579 | { |
2e45874c | 2580 | wait_queue_t *curr, *next; |
1da177e4 | 2581 | |
2e45874c | 2582 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
2583 | unsigned flags = curr->flags; |
2584 | ||
63859d4f | 2585 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 2586 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
2587 | break; |
2588 | } | |
2589 | } | |
2590 | ||
2591 | /** | |
2592 | * __wake_up - wake up threads blocked on a waitqueue. | |
2593 | * @q: the waitqueue | |
2594 | * @mode: which threads | |
2595 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 2596 | * @key: is directly passed to the wakeup function |
50fa610a DH |
2597 | * |
2598 | * It may be assumed that this function implies a write memory barrier before | |
2599 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 2600 | */ |
7ad5b3a5 | 2601 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 2602 | int nr_exclusive, void *key) |
1da177e4 LT |
2603 | { |
2604 | unsigned long flags; | |
2605 | ||
2606 | spin_lock_irqsave(&q->lock, flags); | |
2607 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
2608 | spin_unlock_irqrestore(&q->lock, flags); | |
2609 | } | |
1da177e4 LT |
2610 | EXPORT_SYMBOL(__wake_up); |
2611 | ||
2612 | /* | |
2613 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
2614 | */ | |
63b20011 | 2615 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 2616 | { |
63b20011 | 2617 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 2618 | } |
22c43c81 | 2619 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 2620 | |
4ede816a DL |
2621 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
2622 | { | |
2623 | __wake_up_common(q, mode, 1, 0, key); | |
2624 | } | |
bf294b41 | 2625 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 2626 | |
1da177e4 | 2627 | /** |
4ede816a | 2628 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
2629 | * @q: the waitqueue |
2630 | * @mode: which threads | |
2631 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 2632 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
2633 | * |
2634 | * The sync wakeup differs that the waker knows that it will schedule | |
2635 | * away soon, so while the target thread will be woken up, it will not | |
2636 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
2637 | * with each other. This can prevent needless bouncing between CPUs. | |
2638 | * | |
2639 | * On UP it can prevent extra preemption. | |
50fa610a DH |
2640 | * |
2641 | * It may be assumed that this function implies a write memory barrier before | |
2642 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 2643 | */ |
4ede816a DL |
2644 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
2645 | int nr_exclusive, void *key) | |
1da177e4 LT |
2646 | { |
2647 | unsigned long flags; | |
7d478721 | 2648 | int wake_flags = WF_SYNC; |
1da177e4 LT |
2649 | |
2650 | if (unlikely(!q)) | |
2651 | return; | |
2652 | ||
2653 | if (unlikely(!nr_exclusive)) | |
7d478721 | 2654 | wake_flags = 0; |
1da177e4 LT |
2655 | |
2656 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 2657 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
2658 | spin_unlock_irqrestore(&q->lock, flags); |
2659 | } | |
4ede816a DL |
2660 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
2661 | ||
2662 | /* | |
2663 | * __wake_up_sync - see __wake_up_sync_key() | |
2664 | */ | |
2665 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
2666 | { | |
2667 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
2668 | } | |
1da177e4 LT |
2669 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
2670 | ||
65eb3dc6 KD |
2671 | /** |
2672 | * complete: - signals a single thread waiting on this completion | |
2673 | * @x: holds the state of this particular completion | |
2674 | * | |
2675 | * This will wake up a single thread waiting on this completion. Threads will be | |
2676 | * awakened in the same order in which they were queued. | |
2677 | * | |
2678 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
2679 | * |
2680 | * It may be assumed that this function implies a write memory barrier before | |
2681 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 2682 | */ |
b15136e9 | 2683 | void complete(struct completion *x) |
1da177e4 LT |
2684 | { |
2685 | unsigned long flags; | |
2686 | ||
2687 | spin_lock_irqsave(&x->wait.lock, flags); | |
2688 | x->done++; | |
d9514f6c | 2689 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
2690 | spin_unlock_irqrestore(&x->wait.lock, flags); |
2691 | } | |
2692 | EXPORT_SYMBOL(complete); | |
2693 | ||
65eb3dc6 KD |
2694 | /** |
2695 | * complete_all: - signals all threads waiting on this completion | |
2696 | * @x: holds the state of this particular completion | |
2697 | * | |
2698 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
2699 | * |
2700 | * It may be assumed that this function implies a write memory barrier before | |
2701 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 2702 | */ |
b15136e9 | 2703 | void complete_all(struct completion *x) |
1da177e4 LT |
2704 | { |
2705 | unsigned long flags; | |
2706 | ||
2707 | spin_lock_irqsave(&x->wait.lock, flags); | |
2708 | x->done += UINT_MAX/2; | |
d9514f6c | 2709 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
2710 | spin_unlock_irqrestore(&x->wait.lock, flags); |
2711 | } | |
2712 | EXPORT_SYMBOL(complete_all); | |
2713 | ||
8cbbe86d | 2714 | static inline long __sched |
686855f5 VD |
2715 | do_wait_for_common(struct completion *x, |
2716 | long (*action)(long), long timeout, int state) | |
1da177e4 | 2717 | { |
1da177e4 LT |
2718 | if (!x->done) { |
2719 | DECLARE_WAITQUEUE(wait, current); | |
2720 | ||
a93d2f17 | 2721 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 2722 | do { |
94d3d824 | 2723 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
2724 | timeout = -ERESTARTSYS; |
2725 | break; | |
8cbbe86d AK |
2726 | } |
2727 | __set_current_state(state); | |
1da177e4 | 2728 | spin_unlock_irq(&x->wait.lock); |
686855f5 | 2729 | timeout = action(timeout); |
1da177e4 | 2730 | spin_lock_irq(&x->wait.lock); |
ea71a546 | 2731 | } while (!x->done && timeout); |
1da177e4 | 2732 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
2733 | if (!x->done) |
2734 | return timeout; | |
1da177e4 LT |
2735 | } |
2736 | x->done--; | |
ea71a546 | 2737 | return timeout ?: 1; |
1da177e4 | 2738 | } |
1da177e4 | 2739 | |
686855f5 VD |
2740 | static inline long __sched |
2741 | __wait_for_common(struct completion *x, | |
2742 | long (*action)(long), long timeout, int state) | |
1da177e4 | 2743 | { |
1da177e4 LT |
2744 | might_sleep(); |
2745 | ||
2746 | spin_lock_irq(&x->wait.lock); | |
686855f5 | 2747 | timeout = do_wait_for_common(x, action, timeout, state); |
1da177e4 | 2748 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
2749 | return timeout; |
2750 | } | |
1da177e4 | 2751 | |
686855f5 VD |
2752 | static long __sched |
2753 | wait_for_common(struct completion *x, long timeout, int state) | |
2754 | { | |
2755 | return __wait_for_common(x, schedule_timeout, timeout, state); | |
2756 | } | |
2757 | ||
2758 | static long __sched | |
2759 | wait_for_common_io(struct completion *x, long timeout, int state) | |
2760 | { | |
2761 | return __wait_for_common(x, io_schedule_timeout, timeout, state); | |
2762 | } | |
2763 | ||
65eb3dc6 KD |
2764 | /** |
2765 | * wait_for_completion: - waits for completion of a task | |
2766 | * @x: holds the state of this particular completion | |
2767 | * | |
2768 | * This waits to be signaled for completion of a specific task. It is NOT | |
2769 | * interruptible and there is no timeout. | |
2770 | * | |
2771 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
2772 | * and interrupt capability. Also see complete(). | |
2773 | */ | |
b15136e9 | 2774 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
2775 | { |
2776 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 2777 | } |
8cbbe86d | 2778 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 2779 | |
65eb3dc6 KD |
2780 | /** |
2781 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
2782 | * @x: holds the state of this particular completion | |
2783 | * @timeout: timeout value in jiffies | |
2784 | * | |
2785 | * This waits for either a completion of a specific task to be signaled or for a | |
2786 | * specified timeout to expire. The timeout is in jiffies. It is not | |
2787 | * interruptible. | |
c6dc7f05 BF |
2788 | * |
2789 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
2790 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 2791 | */ |
b15136e9 | 2792 | unsigned long __sched |
8cbbe86d | 2793 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 2794 | { |
8cbbe86d | 2795 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 2796 | } |
8cbbe86d | 2797 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 2798 | |
686855f5 VD |
2799 | /** |
2800 | * wait_for_completion_io: - waits for completion of a task | |
2801 | * @x: holds the state of this particular completion | |
2802 | * | |
2803 | * This waits to be signaled for completion of a specific task. It is NOT | |
2804 | * interruptible and there is no timeout. The caller is accounted as waiting | |
2805 | * for IO. | |
2806 | */ | |
2807 | void __sched wait_for_completion_io(struct completion *x) | |
2808 | { | |
2809 | wait_for_common_io(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
2810 | } | |
2811 | EXPORT_SYMBOL(wait_for_completion_io); | |
2812 | ||
2813 | /** | |
2814 | * wait_for_completion_io_timeout: - waits for completion of a task (w/timeout) | |
2815 | * @x: holds the state of this particular completion | |
2816 | * @timeout: timeout value in jiffies | |
2817 | * | |
2818 | * This waits for either a completion of a specific task to be signaled or for a | |
2819 | * specified timeout to expire. The timeout is in jiffies. It is not | |
2820 | * interruptible. The caller is accounted as waiting for IO. | |
2821 | * | |
2822 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
2823 | * jiffies left till timeout) if completed. | |
2824 | */ | |
2825 | unsigned long __sched | |
2826 | wait_for_completion_io_timeout(struct completion *x, unsigned long timeout) | |
2827 | { | |
2828 | return wait_for_common_io(x, timeout, TASK_UNINTERRUPTIBLE); | |
2829 | } | |
2830 | EXPORT_SYMBOL(wait_for_completion_io_timeout); | |
2831 | ||
65eb3dc6 KD |
2832 | /** |
2833 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
2834 | * @x: holds the state of this particular completion | |
2835 | * | |
2836 | * This waits for completion of a specific task to be signaled. It is | |
2837 | * interruptible. | |
c6dc7f05 BF |
2838 | * |
2839 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 2840 | */ |
8cbbe86d | 2841 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 2842 | { |
51e97990 AK |
2843 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
2844 | if (t == -ERESTARTSYS) | |
2845 | return t; | |
2846 | return 0; | |
0fec171c | 2847 | } |
8cbbe86d | 2848 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 2849 | |
65eb3dc6 KD |
2850 | /** |
2851 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
2852 | * @x: holds the state of this particular completion | |
2853 | * @timeout: timeout value in jiffies | |
2854 | * | |
2855 | * This waits for either a completion of a specific task to be signaled or for a | |
2856 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
2857 | * |
2858 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
2859 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 2860 | */ |
6bf41237 | 2861 | long __sched |
8cbbe86d AK |
2862 | wait_for_completion_interruptible_timeout(struct completion *x, |
2863 | unsigned long timeout) | |
0fec171c | 2864 | { |
8cbbe86d | 2865 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 2866 | } |
8cbbe86d | 2867 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 2868 | |
65eb3dc6 KD |
2869 | /** |
2870 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
2871 | * @x: holds the state of this particular completion | |
2872 | * | |
2873 | * This waits to be signaled for completion of a specific task. It can be | |
2874 | * interrupted by a kill signal. | |
c6dc7f05 BF |
2875 | * |
2876 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 2877 | */ |
009e577e MW |
2878 | int __sched wait_for_completion_killable(struct completion *x) |
2879 | { | |
2880 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
2881 | if (t == -ERESTARTSYS) | |
2882 | return t; | |
2883 | return 0; | |
2884 | } | |
2885 | EXPORT_SYMBOL(wait_for_completion_killable); | |
2886 | ||
0aa12fb4 SW |
2887 | /** |
2888 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
2889 | * @x: holds the state of this particular completion | |
2890 | * @timeout: timeout value in jiffies | |
2891 | * | |
2892 | * This waits for either a completion of a specific task to be | |
2893 | * signaled or for a specified timeout to expire. It can be | |
2894 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
2895 | * |
2896 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
2897 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 2898 | */ |
6bf41237 | 2899 | long __sched |
0aa12fb4 SW |
2900 | wait_for_completion_killable_timeout(struct completion *x, |
2901 | unsigned long timeout) | |
2902 | { | |
2903 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
2904 | } | |
2905 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
2906 | ||
be4de352 DC |
2907 | /** |
2908 | * try_wait_for_completion - try to decrement a completion without blocking | |
2909 | * @x: completion structure | |
2910 | * | |
2911 | * Returns: 0 if a decrement cannot be done without blocking | |
2912 | * 1 if a decrement succeeded. | |
2913 | * | |
2914 | * If a completion is being used as a counting completion, | |
2915 | * attempt to decrement the counter without blocking. This | |
2916 | * enables us to avoid waiting if the resource the completion | |
2917 | * is protecting is not available. | |
2918 | */ | |
2919 | bool try_wait_for_completion(struct completion *x) | |
2920 | { | |
7539a3b3 | 2921 | unsigned long flags; |
be4de352 DC |
2922 | int ret = 1; |
2923 | ||
7539a3b3 | 2924 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
2925 | if (!x->done) |
2926 | ret = 0; | |
2927 | else | |
2928 | x->done--; | |
7539a3b3 | 2929 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
2930 | return ret; |
2931 | } | |
2932 | EXPORT_SYMBOL(try_wait_for_completion); | |
2933 | ||
2934 | /** | |
2935 | * completion_done - Test to see if a completion has any waiters | |
2936 | * @x: completion structure | |
2937 | * | |
2938 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
2939 | * 1 if there are no waiters. | |
2940 | * | |
2941 | */ | |
2942 | bool completion_done(struct completion *x) | |
2943 | { | |
7539a3b3 | 2944 | unsigned long flags; |
be4de352 DC |
2945 | int ret = 1; |
2946 | ||
7539a3b3 | 2947 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
2948 | if (!x->done) |
2949 | ret = 0; | |
7539a3b3 | 2950 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
2951 | return ret; |
2952 | } | |
2953 | EXPORT_SYMBOL(completion_done); | |
2954 | ||
8cbbe86d AK |
2955 | static long __sched |
2956 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 2957 | { |
0fec171c IM |
2958 | unsigned long flags; |
2959 | wait_queue_t wait; | |
2960 | ||
2961 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 2962 | |
8cbbe86d | 2963 | __set_current_state(state); |
1da177e4 | 2964 | |
8cbbe86d AK |
2965 | spin_lock_irqsave(&q->lock, flags); |
2966 | __add_wait_queue(q, &wait); | |
2967 | spin_unlock(&q->lock); | |
2968 | timeout = schedule_timeout(timeout); | |
2969 | spin_lock_irq(&q->lock); | |
2970 | __remove_wait_queue(q, &wait); | |
2971 | spin_unlock_irqrestore(&q->lock, flags); | |
2972 | ||
2973 | return timeout; | |
2974 | } | |
2975 | ||
2976 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
2977 | { | |
2978 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 2979 | } |
1da177e4 LT |
2980 | EXPORT_SYMBOL(interruptible_sleep_on); |
2981 | ||
0fec171c | 2982 | long __sched |
95cdf3b7 | 2983 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2984 | { |
8cbbe86d | 2985 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 2986 | } |
1da177e4 LT |
2987 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
2988 | ||
0fec171c | 2989 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 2990 | { |
8cbbe86d | 2991 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 2992 | } |
1da177e4 LT |
2993 | EXPORT_SYMBOL(sleep_on); |
2994 | ||
0fec171c | 2995 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2996 | { |
8cbbe86d | 2997 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 2998 | } |
1da177e4 LT |
2999 | EXPORT_SYMBOL(sleep_on_timeout); |
3000 | ||
b29739f9 IM |
3001 | #ifdef CONFIG_RT_MUTEXES |
3002 | ||
3003 | /* | |
3004 | * rt_mutex_setprio - set the current priority of a task | |
3005 | * @p: task | |
3006 | * @prio: prio value (kernel-internal form) | |
3007 | * | |
3008 | * This function changes the 'effective' priority of a task. It does | |
3009 | * not touch ->normal_prio like __setscheduler(). | |
3010 | * | |
3011 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3012 | */ | |
36c8b586 | 3013 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3014 | { |
83b699ed | 3015 | int oldprio, on_rq, running; |
70b97a7f | 3016 | struct rq *rq; |
83ab0aa0 | 3017 | const struct sched_class *prev_class; |
b29739f9 IM |
3018 | |
3019 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3020 | ||
0122ec5b | 3021 | rq = __task_rq_lock(p); |
b29739f9 | 3022 | |
1c4dd99b TG |
3023 | /* |
3024 | * Idle task boosting is a nono in general. There is one | |
3025 | * exception, when PREEMPT_RT and NOHZ is active: | |
3026 | * | |
3027 | * The idle task calls get_next_timer_interrupt() and holds | |
3028 | * the timer wheel base->lock on the CPU and another CPU wants | |
3029 | * to access the timer (probably to cancel it). We can safely | |
3030 | * ignore the boosting request, as the idle CPU runs this code | |
3031 | * with interrupts disabled and will complete the lock | |
3032 | * protected section without being interrupted. So there is no | |
3033 | * real need to boost. | |
3034 | */ | |
3035 | if (unlikely(p == rq->idle)) { | |
3036 | WARN_ON(p != rq->curr); | |
3037 | WARN_ON(p->pi_blocked_on); | |
3038 | goto out_unlock; | |
3039 | } | |
3040 | ||
a8027073 | 3041 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3042 | oldprio = p->prio; |
83ab0aa0 | 3043 | prev_class = p->sched_class; |
fd2f4419 | 3044 | on_rq = p->on_rq; |
051a1d1a | 3045 | running = task_current(rq, p); |
0e1f3483 | 3046 | if (on_rq) |
69be72c1 | 3047 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3048 | if (running) |
3049 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3050 | |
3051 | if (rt_prio(prio)) | |
3052 | p->sched_class = &rt_sched_class; | |
3053 | else | |
3054 | p->sched_class = &fair_sched_class; | |
3055 | ||
b29739f9 IM |
3056 | p->prio = prio; |
3057 | ||
0e1f3483 HS |
3058 | if (running) |
3059 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3060 | if (on_rq) |
371fd7e7 | 3061 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3062 | |
da7a735e | 3063 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3064 | out_unlock: |
0122ec5b | 3065 | __task_rq_unlock(rq); |
b29739f9 | 3066 | } |
b29739f9 | 3067 | #endif |
36c8b586 | 3068 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3069 | { |
dd41f596 | 3070 | int old_prio, delta, on_rq; |
1da177e4 | 3071 | unsigned long flags; |
70b97a7f | 3072 | struct rq *rq; |
1da177e4 LT |
3073 | |
3074 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3075 | return; | |
3076 | /* | |
3077 | * We have to be careful, if called from sys_setpriority(), | |
3078 | * the task might be in the middle of scheduling on another CPU. | |
3079 | */ | |
3080 | rq = task_rq_lock(p, &flags); | |
3081 | /* | |
3082 | * The RT priorities are set via sched_setscheduler(), but we still | |
3083 | * allow the 'normal' nice value to be set - but as expected | |
3084 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3085 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3086 | */ |
e05606d3 | 3087 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3088 | p->static_prio = NICE_TO_PRIO(nice); |
3089 | goto out_unlock; | |
3090 | } | |
fd2f4419 | 3091 | on_rq = p->on_rq; |
c09595f6 | 3092 | if (on_rq) |
69be72c1 | 3093 | dequeue_task(rq, p, 0); |
1da177e4 | 3094 | |
1da177e4 | 3095 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3096 | set_load_weight(p); |
b29739f9 IM |
3097 | old_prio = p->prio; |
3098 | p->prio = effective_prio(p); | |
3099 | delta = p->prio - old_prio; | |
1da177e4 | 3100 | |
dd41f596 | 3101 | if (on_rq) { |
371fd7e7 | 3102 | enqueue_task(rq, p, 0); |
1da177e4 | 3103 | /* |
d5f9f942 AM |
3104 | * If the task increased its priority or is running and |
3105 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3106 | */ |
d5f9f942 | 3107 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3108 | resched_task(rq->curr); |
3109 | } | |
3110 | out_unlock: | |
0122ec5b | 3111 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3112 | } |
1da177e4 LT |
3113 | EXPORT_SYMBOL(set_user_nice); |
3114 | ||
e43379f1 MM |
3115 | /* |
3116 | * can_nice - check if a task can reduce its nice value | |
3117 | * @p: task | |
3118 | * @nice: nice value | |
3119 | */ | |
36c8b586 | 3120 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3121 | { |
024f4747 MM |
3122 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3123 | int nice_rlim = 20 - nice; | |
48f24c4d | 3124 | |
78d7d407 | 3125 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3126 | capable(CAP_SYS_NICE)); |
3127 | } | |
3128 | ||
1da177e4 LT |
3129 | #ifdef __ARCH_WANT_SYS_NICE |
3130 | ||
3131 | /* | |
3132 | * sys_nice - change the priority of the current process. | |
3133 | * @increment: priority increment | |
3134 | * | |
3135 | * sys_setpriority is a more generic, but much slower function that | |
3136 | * does similar things. | |
3137 | */ | |
5add95d4 | 3138 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3139 | { |
48f24c4d | 3140 | long nice, retval; |
1da177e4 LT |
3141 | |
3142 | /* | |
3143 | * Setpriority might change our priority at the same moment. | |
3144 | * We don't have to worry. Conceptually one call occurs first | |
3145 | * and we have a single winner. | |
3146 | */ | |
e43379f1 MM |
3147 | if (increment < -40) |
3148 | increment = -40; | |
1da177e4 LT |
3149 | if (increment > 40) |
3150 | increment = 40; | |
3151 | ||
2b8f836f | 3152 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3153 | if (nice < -20) |
3154 | nice = -20; | |
3155 | if (nice > 19) | |
3156 | nice = 19; | |
3157 | ||
e43379f1 MM |
3158 | if (increment < 0 && !can_nice(current, nice)) |
3159 | return -EPERM; | |
3160 | ||
1da177e4 LT |
3161 | retval = security_task_setnice(current, nice); |
3162 | if (retval) | |
3163 | return retval; | |
3164 | ||
3165 | set_user_nice(current, nice); | |
3166 | return 0; | |
3167 | } | |
3168 | ||
3169 | #endif | |
3170 | ||
3171 | /** | |
3172 | * task_prio - return the priority value of a given task. | |
3173 | * @p: the task in question. | |
3174 | * | |
3175 | * This is the priority value as seen by users in /proc. | |
3176 | * RT tasks are offset by -200. Normal tasks are centered | |
3177 | * around 0, value goes from -16 to +15. | |
3178 | */ | |
36c8b586 | 3179 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3180 | { |
3181 | return p->prio - MAX_RT_PRIO; | |
3182 | } | |
3183 | ||
3184 | /** | |
3185 | * task_nice - return the nice value of a given task. | |
3186 | * @p: the task in question. | |
3187 | */ | |
36c8b586 | 3188 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3189 | { |
3190 | return TASK_NICE(p); | |
3191 | } | |
150d8bed | 3192 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3193 | |
3194 | /** | |
3195 | * idle_cpu - is a given cpu idle currently? | |
3196 | * @cpu: the processor in question. | |
3197 | */ | |
3198 | int idle_cpu(int cpu) | |
3199 | { | |
908a3283 TG |
3200 | struct rq *rq = cpu_rq(cpu); |
3201 | ||
3202 | if (rq->curr != rq->idle) | |
3203 | return 0; | |
3204 | ||
3205 | if (rq->nr_running) | |
3206 | return 0; | |
3207 | ||
3208 | #ifdef CONFIG_SMP | |
3209 | if (!llist_empty(&rq->wake_list)) | |
3210 | return 0; | |
3211 | #endif | |
3212 | ||
3213 | return 1; | |
1da177e4 LT |
3214 | } |
3215 | ||
1da177e4 LT |
3216 | /** |
3217 | * idle_task - return the idle task for a given cpu. | |
3218 | * @cpu: the processor in question. | |
3219 | */ | |
36c8b586 | 3220 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3221 | { |
3222 | return cpu_rq(cpu)->idle; | |
3223 | } | |
3224 | ||
3225 | /** | |
3226 | * find_process_by_pid - find a process with a matching PID value. | |
3227 | * @pid: the pid in question. | |
3228 | */ | |
a9957449 | 3229 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3230 | { |
228ebcbe | 3231 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3232 | } |
3233 | ||
3234 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
3235 | static void |
3236 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 3237 | { |
1da177e4 LT |
3238 | p->policy = policy; |
3239 | p->rt_priority = prio; | |
b29739f9 IM |
3240 | p->normal_prio = normal_prio(p); |
3241 | /* we are holding p->pi_lock already */ | |
3242 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
3243 | if (rt_prio(p->prio)) |
3244 | p->sched_class = &rt_sched_class; | |
3245 | else | |
3246 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 3247 | set_load_weight(p); |
1da177e4 LT |
3248 | } |
3249 | ||
c69e8d9c DH |
3250 | /* |
3251 | * check the target process has a UID that matches the current process's | |
3252 | */ | |
3253 | static bool check_same_owner(struct task_struct *p) | |
3254 | { | |
3255 | const struct cred *cred = current_cred(), *pcred; | |
3256 | bool match; | |
3257 | ||
3258 | rcu_read_lock(); | |
3259 | pcred = __task_cred(p); | |
9c806aa0 EB |
3260 | match = (uid_eq(cred->euid, pcred->euid) || |
3261 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3262 | rcu_read_unlock(); |
3263 | return match; | |
3264 | } | |
3265 | ||
961ccddd | 3266 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 3267 | const struct sched_param *param, bool user) |
1da177e4 | 3268 | { |
83b699ed | 3269 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 3270 | unsigned long flags; |
83ab0aa0 | 3271 | const struct sched_class *prev_class; |
70b97a7f | 3272 | struct rq *rq; |
ca94c442 | 3273 | int reset_on_fork; |
1da177e4 | 3274 | |
66e5393a SR |
3275 | /* may grab non-irq protected spin_locks */ |
3276 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3277 | recheck: |
3278 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3279 | if (policy < 0) { |
3280 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3281 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3282 | } else { |
3283 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3284 | policy &= ~SCHED_RESET_ON_FORK; | |
3285 | ||
3286 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
3287 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
3288 | policy != SCHED_IDLE) | |
3289 | return -EINVAL; | |
3290 | } | |
3291 | ||
1da177e4 LT |
3292 | /* |
3293 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3294 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3295 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
3296 | */ |
3297 | if (param->sched_priority < 0 || | |
95cdf3b7 | 3298 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 3299 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3300 | return -EINVAL; |
e05606d3 | 3301 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
3302 | return -EINVAL; |
3303 | ||
37e4ab3f OC |
3304 | /* |
3305 | * Allow unprivileged RT tasks to decrease priority: | |
3306 | */ | |
961ccddd | 3307 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 3308 | if (rt_policy(policy)) { |
a44702e8 ON |
3309 | unsigned long rlim_rtprio = |
3310 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3311 | |
3312 | /* can't set/change the rt policy */ | |
3313 | if (policy != p->policy && !rlim_rtprio) | |
3314 | return -EPERM; | |
3315 | ||
3316 | /* can't increase priority */ | |
3317 | if (param->sched_priority > p->rt_priority && | |
3318 | param->sched_priority > rlim_rtprio) | |
3319 | return -EPERM; | |
3320 | } | |
c02aa73b | 3321 | |
dd41f596 | 3322 | /* |
c02aa73b DH |
3323 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3324 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3325 | */ |
c02aa73b DH |
3326 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3327 | if (!can_nice(p, TASK_NICE(p))) | |
3328 | return -EPERM; | |
3329 | } | |
5fe1d75f | 3330 | |
37e4ab3f | 3331 | /* can't change other user's priorities */ |
c69e8d9c | 3332 | if (!check_same_owner(p)) |
37e4ab3f | 3333 | return -EPERM; |
ca94c442 LP |
3334 | |
3335 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3336 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3337 | return -EPERM; | |
37e4ab3f | 3338 | } |
1da177e4 | 3339 | |
725aad24 | 3340 | if (user) { |
b0ae1981 | 3341 | retval = security_task_setscheduler(p); |
725aad24 JF |
3342 | if (retval) |
3343 | return retval; | |
3344 | } | |
3345 | ||
b29739f9 IM |
3346 | /* |
3347 | * make sure no PI-waiters arrive (or leave) while we are | |
3348 | * changing the priority of the task: | |
0122ec5b | 3349 | * |
25985edc | 3350 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3351 | * runqueue lock must be held. |
3352 | */ | |
0122ec5b | 3353 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3354 | |
34f971f6 PZ |
3355 | /* |
3356 | * Changing the policy of the stop threads its a very bad idea | |
3357 | */ | |
3358 | if (p == rq->stop) { | |
0122ec5b | 3359 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3360 | return -EINVAL; |
3361 | } | |
3362 | ||
a51e9198 DF |
3363 | /* |
3364 | * If not changing anything there's no need to proceed further: | |
3365 | */ | |
3366 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
3367 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 3368 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3369 | return 0; |
3370 | } | |
3371 | ||
dc61b1d6 PZ |
3372 | #ifdef CONFIG_RT_GROUP_SCHED |
3373 | if (user) { | |
3374 | /* | |
3375 | * Do not allow realtime tasks into groups that have no runtime | |
3376 | * assigned. | |
3377 | */ | |
3378 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3379 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3380 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3381 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3382 | return -EPERM; |
3383 | } | |
3384 | } | |
3385 | #endif | |
3386 | ||
1da177e4 LT |
3387 | /* recheck policy now with rq lock held */ |
3388 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3389 | policy = oldpolicy = -1; | |
0122ec5b | 3390 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3391 | goto recheck; |
3392 | } | |
fd2f4419 | 3393 | on_rq = p->on_rq; |
051a1d1a | 3394 | running = task_current(rq, p); |
0e1f3483 | 3395 | if (on_rq) |
4ca9b72b | 3396 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3397 | if (running) |
3398 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3399 | |
ca94c442 LP |
3400 | p->sched_reset_on_fork = reset_on_fork; |
3401 | ||
1da177e4 | 3402 | oldprio = p->prio; |
83ab0aa0 | 3403 | prev_class = p->sched_class; |
dd41f596 | 3404 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 3405 | |
0e1f3483 HS |
3406 | if (running) |
3407 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3408 | if (on_rq) |
4ca9b72b | 3409 | enqueue_task(rq, p, 0); |
cb469845 | 3410 | |
da7a735e | 3411 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3412 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3413 | |
95e02ca9 TG |
3414 | rt_mutex_adjust_pi(p); |
3415 | ||
1da177e4 LT |
3416 | return 0; |
3417 | } | |
961ccddd RR |
3418 | |
3419 | /** | |
3420 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3421 | * @p: the task in question. | |
3422 | * @policy: new policy. | |
3423 | * @param: structure containing the new RT priority. | |
3424 | * | |
3425 | * NOTE that the task may be already dead. | |
3426 | */ | |
3427 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3428 | const struct sched_param *param) |
961ccddd RR |
3429 | { |
3430 | return __sched_setscheduler(p, policy, param, true); | |
3431 | } | |
1da177e4 LT |
3432 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3433 | ||
961ccddd RR |
3434 | /** |
3435 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3436 | * @p: the task in question. | |
3437 | * @policy: new policy. | |
3438 | * @param: structure containing the new RT priority. | |
3439 | * | |
3440 | * Just like sched_setscheduler, only don't bother checking if the | |
3441 | * current context has permission. For example, this is needed in | |
3442 | * stop_machine(): we create temporary high priority worker threads, | |
3443 | * but our caller might not have that capability. | |
3444 | */ | |
3445 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3446 | const struct sched_param *param) |
961ccddd RR |
3447 | { |
3448 | return __sched_setscheduler(p, policy, param, false); | |
3449 | } | |
3450 | ||
95cdf3b7 IM |
3451 | static int |
3452 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3453 | { |
1da177e4 LT |
3454 | struct sched_param lparam; |
3455 | struct task_struct *p; | |
36c8b586 | 3456 | int retval; |
1da177e4 LT |
3457 | |
3458 | if (!param || pid < 0) | |
3459 | return -EINVAL; | |
3460 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3461 | return -EFAULT; | |
5fe1d75f ON |
3462 | |
3463 | rcu_read_lock(); | |
3464 | retval = -ESRCH; | |
1da177e4 | 3465 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3466 | if (p != NULL) |
3467 | retval = sched_setscheduler(p, policy, &lparam); | |
3468 | rcu_read_unlock(); | |
36c8b586 | 3469 | |
1da177e4 LT |
3470 | return retval; |
3471 | } | |
3472 | ||
3473 | /** | |
3474 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3475 | * @pid: the pid in question. | |
3476 | * @policy: new policy. | |
3477 | * @param: structure containing the new RT priority. | |
3478 | */ | |
5add95d4 HC |
3479 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3480 | struct sched_param __user *, param) | |
1da177e4 | 3481 | { |
c21761f1 JB |
3482 | /* negative values for policy are not valid */ |
3483 | if (policy < 0) | |
3484 | return -EINVAL; | |
3485 | ||
1da177e4 LT |
3486 | return do_sched_setscheduler(pid, policy, param); |
3487 | } | |
3488 | ||
3489 | /** | |
3490 | * sys_sched_setparam - set/change the RT priority of a thread | |
3491 | * @pid: the pid in question. | |
3492 | * @param: structure containing the new RT priority. | |
3493 | */ | |
5add95d4 | 3494 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3495 | { |
3496 | return do_sched_setscheduler(pid, -1, param); | |
3497 | } | |
3498 | ||
3499 | /** | |
3500 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3501 | * @pid: the pid in question. | |
3502 | */ | |
5add95d4 | 3503 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3504 | { |
36c8b586 | 3505 | struct task_struct *p; |
3a5c359a | 3506 | int retval; |
1da177e4 LT |
3507 | |
3508 | if (pid < 0) | |
3a5c359a | 3509 | return -EINVAL; |
1da177e4 LT |
3510 | |
3511 | retval = -ESRCH; | |
5fe85be0 | 3512 | rcu_read_lock(); |
1da177e4 LT |
3513 | p = find_process_by_pid(pid); |
3514 | if (p) { | |
3515 | retval = security_task_getscheduler(p); | |
3516 | if (!retval) | |
ca94c442 LP |
3517 | retval = p->policy |
3518 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3519 | } |
5fe85be0 | 3520 | rcu_read_unlock(); |
1da177e4 LT |
3521 | return retval; |
3522 | } | |
3523 | ||
3524 | /** | |
ca94c442 | 3525 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3526 | * @pid: the pid in question. |
3527 | * @param: structure containing the RT priority. | |
3528 | */ | |
5add95d4 | 3529 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3530 | { |
3531 | struct sched_param lp; | |
36c8b586 | 3532 | struct task_struct *p; |
3a5c359a | 3533 | int retval; |
1da177e4 LT |
3534 | |
3535 | if (!param || pid < 0) | |
3a5c359a | 3536 | return -EINVAL; |
1da177e4 | 3537 | |
5fe85be0 | 3538 | rcu_read_lock(); |
1da177e4 LT |
3539 | p = find_process_by_pid(pid); |
3540 | retval = -ESRCH; | |
3541 | if (!p) | |
3542 | goto out_unlock; | |
3543 | ||
3544 | retval = security_task_getscheduler(p); | |
3545 | if (retval) | |
3546 | goto out_unlock; | |
3547 | ||
3548 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3549 | rcu_read_unlock(); |
1da177e4 LT |
3550 | |
3551 | /* | |
3552 | * This one might sleep, we cannot do it with a spinlock held ... | |
3553 | */ | |
3554 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3555 | ||
1da177e4 LT |
3556 | return retval; |
3557 | ||
3558 | out_unlock: | |
5fe85be0 | 3559 | rcu_read_unlock(); |
1da177e4 LT |
3560 | return retval; |
3561 | } | |
3562 | ||
96f874e2 | 3563 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3564 | { |
5a16f3d3 | 3565 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3566 | struct task_struct *p; |
3567 | int retval; | |
1da177e4 | 3568 | |
95402b38 | 3569 | get_online_cpus(); |
23f5d142 | 3570 | rcu_read_lock(); |
1da177e4 LT |
3571 | |
3572 | p = find_process_by_pid(pid); | |
3573 | if (!p) { | |
23f5d142 | 3574 | rcu_read_unlock(); |
95402b38 | 3575 | put_online_cpus(); |
1da177e4 LT |
3576 | return -ESRCH; |
3577 | } | |
3578 | ||
23f5d142 | 3579 | /* Prevent p going away */ |
1da177e4 | 3580 | get_task_struct(p); |
23f5d142 | 3581 | rcu_read_unlock(); |
1da177e4 | 3582 | |
14a40ffc TH |
3583 | if (p->flags & PF_NO_SETAFFINITY) { |
3584 | retval = -EINVAL; | |
3585 | goto out_put_task; | |
3586 | } | |
5a16f3d3 RR |
3587 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3588 | retval = -ENOMEM; | |
3589 | goto out_put_task; | |
3590 | } | |
3591 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3592 | retval = -ENOMEM; | |
3593 | goto out_free_cpus_allowed; | |
3594 | } | |
1da177e4 | 3595 | retval = -EPERM; |
4c44aaaf EB |
3596 | if (!check_same_owner(p)) { |
3597 | rcu_read_lock(); | |
3598 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3599 | rcu_read_unlock(); | |
3600 | goto out_unlock; | |
3601 | } | |
3602 | rcu_read_unlock(); | |
3603 | } | |
1da177e4 | 3604 | |
b0ae1981 | 3605 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3606 | if (retval) |
3607 | goto out_unlock; | |
3608 | ||
5a16f3d3 RR |
3609 | cpuset_cpus_allowed(p, cpus_allowed); |
3610 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 3611 | again: |
5a16f3d3 | 3612 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3613 | |
8707d8b8 | 3614 | if (!retval) { |
5a16f3d3 RR |
3615 | cpuset_cpus_allowed(p, cpus_allowed); |
3616 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3617 | /* |
3618 | * We must have raced with a concurrent cpuset | |
3619 | * update. Just reset the cpus_allowed to the | |
3620 | * cpuset's cpus_allowed | |
3621 | */ | |
5a16f3d3 | 3622 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3623 | goto again; |
3624 | } | |
3625 | } | |
1da177e4 | 3626 | out_unlock: |
5a16f3d3 RR |
3627 | free_cpumask_var(new_mask); |
3628 | out_free_cpus_allowed: | |
3629 | free_cpumask_var(cpus_allowed); | |
3630 | out_put_task: | |
1da177e4 | 3631 | put_task_struct(p); |
95402b38 | 3632 | put_online_cpus(); |
1da177e4 LT |
3633 | return retval; |
3634 | } | |
3635 | ||
3636 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3637 | struct cpumask *new_mask) |
1da177e4 | 3638 | { |
96f874e2 RR |
3639 | if (len < cpumask_size()) |
3640 | cpumask_clear(new_mask); | |
3641 | else if (len > cpumask_size()) | |
3642 | len = cpumask_size(); | |
3643 | ||
1da177e4 LT |
3644 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3645 | } | |
3646 | ||
3647 | /** | |
3648 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3649 | * @pid: pid of the process | |
3650 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3651 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
3652 | */ | |
5add95d4 HC |
3653 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3654 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3655 | { |
5a16f3d3 | 3656 | cpumask_var_t new_mask; |
1da177e4 LT |
3657 | int retval; |
3658 | ||
5a16f3d3 RR |
3659 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3660 | return -ENOMEM; | |
1da177e4 | 3661 | |
5a16f3d3 RR |
3662 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3663 | if (retval == 0) | |
3664 | retval = sched_setaffinity(pid, new_mask); | |
3665 | free_cpumask_var(new_mask); | |
3666 | return retval; | |
1da177e4 LT |
3667 | } |
3668 | ||
96f874e2 | 3669 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 3670 | { |
36c8b586 | 3671 | struct task_struct *p; |
31605683 | 3672 | unsigned long flags; |
1da177e4 | 3673 | int retval; |
1da177e4 | 3674 | |
95402b38 | 3675 | get_online_cpus(); |
23f5d142 | 3676 | rcu_read_lock(); |
1da177e4 LT |
3677 | |
3678 | retval = -ESRCH; | |
3679 | p = find_process_by_pid(pid); | |
3680 | if (!p) | |
3681 | goto out_unlock; | |
3682 | ||
e7834f8f DQ |
3683 | retval = security_task_getscheduler(p); |
3684 | if (retval) | |
3685 | goto out_unlock; | |
3686 | ||
013fdb80 | 3687 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 3688 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 3689 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
3690 | |
3691 | out_unlock: | |
23f5d142 | 3692 | rcu_read_unlock(); |
95402b38 | 3693 | put_online_cpus(); |
1da177e4 | 3694 | |
9531b62f | 3695 | return retval; |
1da177e4 LT |
3696 | } |
3697 | ||
3698 | /** | |
3699 | * sys_sched_getaffinity - get the cpu affinity of a process | |
3700 | * @pid: pid of the process | |
3701 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3702 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
3703 | */ | |
5add95d4 HC |
3704 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
3705 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
3706 | { |
3707 | int ret; | |
f17c8607 | 3708 | cpumask_var_t mask; |
1da177e4 | 3709 | |
84fba5ec | 3710 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
3711 | return -EINVAL; |
3712 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
3713 | return -EINVAL; |
3714 | ||
f17c8607 RR |
3715 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
3716 | return -ENOMEM; | |
1da177e4 | 3717 | |
f17c8607 RR |
3718 | ret = sched_getaffinity(pid, mask); |
3719 | if (ret == 0) { | |
8bc037fb | 3720 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
3721 | |
3722 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
3723 | ret = -EFAULT; |
3724 | else | |
cd3d8031 | 3725 | ret = retlen; |
f17c8607 RR |
3726 | } |
3727 | free_cpumask_var(mask); | |
1da177e4 | 3728 | |
f17c8607 | 3729 | return ret; |
1da177e4 LT |
3730 | } |
3731 | ||
3732 | /** | |
3733 | * sys_sched_yield - yield the current processor to other threads. | |
3734 | * | |
dd41f596 IM |
3735 | * This function yields the current CPU to other tasks. If there are no |
3736 | * other threads running on this CPU then this function will return. | |
1da177e4 | 3737 | */ |
5add95d4 | 3738 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 3739 | { |
70b97a7f | 3740 | struct rq *rq = this_rq_lock(); |
1da177e4 | 3741 | |
2d72376b | 3742 | schedstat_inc(rq, yld_count); |
4530d7ab | 3743 | current->sched_class->yield_task(rq); |
1da177e4 LT |
3744 | |
3745 | /* | |
3746 | * Since we are going to call schedule() anyway, there's | |
3747 | * no need to preempt or enable interrupts: | |
3748 | */ | |
3749 | __release(rq->lock); | |
8a25d5de | 3750 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 3751 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 3752 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
3753 | |
3754 | schedule(); | |
3755 | ||
3756 | return 0; | |
3757 | } | |
3758 | ||
d86ee480 PZ |
3759 | static inline int should_resched(void) |
3760 | { | |
3761 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
3762 | } | |
3763 | ||
e7b38404 | 3764 | static void __cond_resched(void) |
1da177e4 | 3765 | { |
e7aaaa69 | 3766 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 3767 | __schedule(); |
e7aaaa69 | 3768 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
3769 | } |
3770 | ||
02b67cc3 | 3771 | int __sched _cond_resched(void) |
1da177e4 | 3772 | { |
d86ee480 | 3773 | if (should_resched()) { |
1da177e4 LT |
3774 | __cond_resched(); |
3775 | return 1; | |
3776 | } | |
3777 | return 0; | |
3778 | } | |
02b67cc3 | 3779 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
3780 | |
3781 | /* | |
613afbf8 | 3782 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
3783 | * call schedule, and on return reacquire the lock. |
3784 | * | |
41a2d6cf | 3785 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
3786 | * operations here to prevent schedule() from being called twice (once via |
3787 | * spin_unlock(), once by hand). | |
3788 | */ | |
613afbf8 | 3789 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 3790 | { |
d86ee480 | 3791 | int resched = should_resched(); |
6df3cecb JK |
3792 | int ret = 0; |
3793 | ||
f607c668 PZ |
3794 | lockdep_assert_held(lock); |
3795 | ||
95c354fe | 3796 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 3797 | spin_unlock(lock); |
d86ee480 | 3798 | if (resched) |
95c354fe NP |
3799 | __cond_resched(); |
3800 | else | |
3801 | cpu_relax(); | |
6df3cecb | 3802 | ret = 1; |
1da177e4 | 3803 | spin_lock(lock); |
1da177e4 | 3804 | } |
6df3cecb | 3805 | return ret; |
1da177e4 | 3806 | } |
613afbf8 | 3807 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 3808 | |
613afbf8 | 3809 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
3810 | { |
3811 | BUG_ON(!in_softirq()); | |
3812 | ||
d86ee480 | 3813 | if (should_resched()) { |
98d82567 | 3814 | local_bh_enable(); |
1da177e4 LT |
3815 | __cond_resched(); |
3816 | local_bh_disable(); | |
3817 | return 1; | |
3818 | } | |
3819 | return 0; | |
3820 | } | |
613afbf8 | 3821 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 3822 | |
1da177e4 LT |
3823 | /** |
3824 | * yield - yield the current processor to other threads. | |
3825 | * | |
8e3fabfd PZ |
3826 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
3827 | * | |
3828 | * The scheduler is at all times free to pick the calling task as the most | |
3829 | * eligible task to run, if removing the yield() call from your code breaks | |
3830 | * it, its already broken. | |
3831 | * | |
3832 | * Typical broken usage is: | |
3833 | * | |
3834 | * while (!event) | |
3835 | * yield(); | |
3836 | * | |
3837 | * where one assumes that yield() will let 'the other' process run that will | |
3838 | * make event true. If the current task is a SCHED_FIFO task that will never | |
3839 | * happen. Never use yield() as a progress guarantee!! | |
3840 | * | |
3841 | * If you want to use yield() to wait for something, use wait_event(). | |
3842 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
3843 | * If you still want to use yield(), do not! | |
1da177e4 LT |
3844 | */ |
3845 | void __sched yield(void) | |
3846 | { | |
3847 | set_current_state(TASK_RUNNING); | |
3848 | sys_sched_yield(); | |
3849 | } | |
1da177e4 LT |
3850 | EXPORT_SYMBOL(yield); |
3851 | ||
d95f4122 MG |
3852 | /** |
3853 | * yield_to - yield the current processor to another thread in | |
3854 | * your thread group, or accelerate that thread toward the | |
3855 | * processor it's on. | |
16addf95 RD |
3856 | * @p: target task |
3857 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
3858 | * |
3859 | * It's the caller's job to ensure that the target task struct | |
3860 | * can't go away on us before we can do any checks. | |
3861 | * | |
7b270f60 PZ |
3862 | * Returns: |
3863 | * true (>0) if we indeed boosted the target task. | |
3864 | * false (0) if we failed to boost the target. | |
3865 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
3866 | */ |
3867 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
3868 | { | |
3869 | struct task_struct *curr = current; | |
3870 | struct rq *rq, *p_rq; | |
3871 | unsigned long flags; | |
c3c18640 | 3872 | int yielded = 0; |
d95f4122 MG |
3873 | |
3874 | local_irq_save(flags); | |
3875 | rq = this_rq(); | |
3876 | ||
3877 | again: | |
3878 | p_rq = task_rq(p); | |
7b270f60 PZ |
3879 | /* |
3880 | * If we're the only runnable task on the rq and target rq also | |
3881 | * has only one task, there's absolutely no point in yielding. | |
3882 | */ | |
3883 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
3884 | yielded = -ESRCH; | |
3885 | goto out_irq; | |
3886 | } | |
3887 | ||
d95f4122 MG |
3888 | double_rq_lock(rq, p_rq); |
3889 | while (task_rq(p) != p_rq) { | |
3890 | double_rq_unlock(rq, p_rq); | |
3891 | goto again; | |
3892 | } | |
3893 | ||
3894 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 3895 | goto out_unlock; |
d95f4122 MG |
3896 | |
3897 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 3898 | goto out_unlock; |
d95f4122 MG |
3899 | |
3900 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 3901 | goto out_unlock; |
d95f4122 MG |
3902 | |
3903 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 3904 | if (yielded) { |
d95f4122 | 3905 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
3906 | /* |
3907 | * Make p's CPU reschedule; pick_next_entity takes care of | |
3908 | * fairness. | |
3909 | */ | |
3910 | if (preempt && rq != p_rq) | |
3911 | resched_task(p_rq->curr); | |
3912 | } | |
d95f4122 | 3913 | |
7b270f60 | 3914 | out_unlock: |
d95f4122 | 3915 | double_rq_unlock(rq, p_rq); |
7b270f60 | 3916 | out_irq: |
d95f4122 MG |
3917 | local_irq_restore(flags); |
3918 | ||
7b270f60 | 3919 | if (yielded > 0) |
d95f4122 MG |
3920 | schedule(); |
3921 | ||
3922 | return yielded; | |
3923 | } | |
3924 | EXPORT_SYMBOL_GPL(yield_to); | |
3925 | ||
1da177e4 | 3926 | /* |
41a2d6cf | 3927 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 3928 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
3929 | */ |
3930 | void __sched io_schedule(void) | |
3931 | { | |
54d35f29 | 3932 | struct rq *rq = raw_rq(); |
1da177e4 | 3933 | |
0ff92245 | 3934 | delayacct_blkio_start(); |
1da177e4 | 3935 | atomic_inc(&rq->nr_iowait); |
73c10101 | 3936 | blk_flush_plug(current); |
8f0dfc34 | 3937 | current->in_iowait = 1; |
1da177e4 | 3938 | schedule(); |
8f0dfc34 | 3939 | current->in_iowait = 0; |
1da177e4 | 3940 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 3941 | delayacct_blkio_end(); |
1da177e4 | 3942 | } |
1da177e4 LT |
3943 | EXPORT_SYMBOL(io_schedule); |
3944 | ||
3945 | long __sched io_schedule_timeout(long timeout) | |
3946 | { | |
54d35f29 | 3947 | struct rq *rq = raw_rq(); |
1da177e4 LT |
3948 | long ret; |
3949 | ||
0ff92245 | 3950 | delayacct_blkio_start(); |
1da177e4 | 3951 | atomic_inc(&rq->nr_iowait); |
73c10101 | 3952 | blk_flush_plug(current); |
8f0dfc34 | 3953 | current->in_iowait = 1; |
1da177e4 | 3954 | ret = schedule_timeout(timeout); |
8f0dfc34 | 3955 | current->in_iowait = 0; |
1da177e4 | 3956 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 3957 | delayacct_blkio_end(); |
1da177e4 LT |
3958 | return ret; |
3959 | } | |
3960 | ||
3961 | /** | |
3962 | * sys_sched_get_priority_max - return maximum RT priority. | |
3963 | * @policy: scheduling class. | |
3964 | * | |
3965 | * this syscall returns the maximum rt_priority that can be used | |
3966 | * by a given scheduling class. | |
3967 | */ | |
5add95d4 | 3968 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
3969 | { |
3970 | int ret = -EINVAL; | |
3971 | ||
3972 | switch (policy) { | |
3973 | case SCHED_FIFO: | |
3974 | case SCHED_RR: | |
3975 | ret = MAX_USER_RT_PRIO-1; | |
3976 | break; | |
3977 | case SCHED_NORMAL: | |
b0a9499c | 3978 | case SCHED_BATCH: |
dd41f596 | 3979 | case SCHED_IDLE: |
1da177e4 LT |
3980 | ret = 0; |
3981 | break; | |
3982 | } | |
3983 | return ret; | |
3984 | } | |
3985 | ||
3986 | /** | |
3987 | * sys_sched_get_priority_min - return minimum RT priority. | |
3988 | * @policy: scheduling class. | |
3989 | * | |
3990 | * this syscall returns the minimum rt_priority that can be used | |
3991 | * by a given scheduling class. | |
3992 | */ | |
5add95d4 | 3993 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
3994 | { |
3995 | int ret = -EINVAL; | |
3996 | ||
3997 | switch (policy) { | |
3998 | case SCHED_FIFO: | |
3999 | case SCHED_RR: | |
4000 | ret = 1; | |
4001 | break; | |
4002 | case SCHED_NORMAL: | |
b0a9499c | 4003 | case SCHED_BATCH: |
dd41f596 | 4004 | case SCHED_IDLE: |
1da177e4 LT |
4005 | ret = 0; |
4006 | } | |
4007 | return ret; | |
4008 | } | |
4009 | ||
4010 | /** | |
4011 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4012 | * @pid: pid of the process. | |
4013 | * @interval: userspace pointer to the timeslice value. | |
4014 | * | |
4015 | * this syscall writes the default timeslice value of a given process | |
4016 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4017 | */ | |
17da2bd9 | 4018 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4019 | struct timespec __user *, interval) |
1da177e4 | 4020 | { |
36c8b586 | 4021 | struct task_struct *p; |
a4ec24b4 | 4022 | unsigned int time_slice; |
dba091b9 TG |
4023 | unsigned long flags; |
4024 | struct rq *rq; | |
3a5c359a | 4025 | int retval; |
1da177e4 | 4026 | struct timespec t; |
1da177e4 LT |
4027 | |
4028 | if (pid < 0) | |
3a5c359a | 4029 | return -EINVAL; |
1da177e4 LT |
4030 | |
4031 | retval = -ESRCH; | |
1a551ae7 | 4032 | rcu_read_lock(); |
1da177e4 LT |
4033 | p = find_process_by_pid(pid); |
4034 | if (!p) | |
4035 | goto out_unlock; | |
4036 | ||
4037 | retval = security_task_getscheduler(p); | |
4038 | if (retval) | |
4039 | goto out_unlock; | |
4040 | ||
dba091b9 TG |
4041 | rq = task_rq_lock(p, &flags); |
4042 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4043 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4044 | |
1a551ae7 | 4045 | rcu_read_unlock(); |
a4ec24b4 | 4046 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4047 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4048 | return retval; |
3a5c359a | 4049 | |
1da177e4 | 4050 | out_unlock: |
1a551ae7 | 4051 | rcu_read_unlock(); |
1da177e4 LT |
4052 | return retval; |
4053 | } | |
4054 | ||
7c731e0a | 4055 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4056 | |
82a1fcb9 | 4057 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4058 | { |
1da177e4 | 4059 | unsigned long free = 0; |
4e79752c | 4060 | int ppid; |
36c8b586 | 4061 | unsigned state; |
1da177e4 | 4062 | |
1da177e4 | 4063 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4064 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4065 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4066 | #if BITS_PER_LONG == 32 |
1da177e4 | 4067 | if (state == TASK_RUNNING) |
3df0fc5b | 4068 | printk(KERN_CONT " running "); |
1da177e4 | 4069 | else |
3df0fc5b | 4070 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4071 | #else |
4072 | if (state == TASK_RUNNING) | |
3df0fc5b | 4073 | printk(KERN_CONT " running task "); |
1da177e4 | 4074 | else |
3df0fc5b | 4075 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4076 | #endif |
4077 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4078 | free = stack_not_used(p); |
1da177e4 | 4079 | #endif |
4e79752c PM |
4080 | rcu_read_lock(); |
4081 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4082 | rcu_read_unlock(); | |
3df0fc5b | 4083 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4084 | task_pid_nr(p), ppid, |
aa47b7e0 | 4085 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4086 | |
3d1cb205 | 4087 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4088 | show_stack(p, NULL); |
1da177e4 LT |
4089 | } |
4090 | ||
e59e2ae2 | 4091 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4092 | { |
36c8b586 | 4093 | struct task_struct *g, *p; |
1da177e4 | 4094 | |
4bd77321 | 4095 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4096 | printk(KERN_INFO |
4097 | " task PC stack pid father\n"); | |
1da177e4 | 4098 | #else |
3df0fc5b PZ |
4099 | printk(KERN_INFO |
4100 | " task PC stack pid father\n"); | |
1da177e4 | 4101 | #endif |
510f5acc | 4102 | rcu_read_lock(); |
1da177e4 LT |
4103 | do_each_thread(g, p) { |
4104 | /* | |
4105 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4106 | * console might take a lot of time: |
1da177e4 LT |
4107 | */ |
4108 | touch_nmi_watchdog(); | |
39bc89fd | 4109 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4110 | sched_show_task(p); |
1da177e4 LT |
4111 | } while_each_thread(g, p); |
4112 | ||
04c9167f JF |
4113 | touch_all_softlockup_watchdogs(); |
4114 | ||
dd41f596 IM |
4115 | #ifdef CONFIG_SCHED_DEBUG |
4116 | sysrq_sched_debug_show(); | |
4117 | #endif | |
510f5acc | 4118 | rcu_read_unlock(); |
e59e2ae2 IM |
4119 | /* |
4120 | * Only show locks if all tasks are dumped: | |
4121 | */ | |
93335a21 | 4122 | if (!state_filter) |
e59e2ae2 | 4123 | debug_show_all_locks(); |
1da177e4 LT |
4124 | } |
4125 | ||
1df21055 IM |
4126 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4127 | { | |
dd41f596 | 4128 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4129 | } |
4130 | ||
f340c0d1 IM |
4131 | /** |
4132 | * init_idle - set up an idle thread for a given CPU | |
4133 | * @idle: task in question | |
4134 | * @cpu: cpu the idle task belongs to | |
4135 | * | |
4136 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4137 | * flag, to make booting more robust. | |
4138 | */ | |
5c1e1767 | 4139 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4140 | { |
70b97a7f | 4141 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4142 | unsigned long flags; |
4143 | ||
05fa785c | 4144 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4145 | |
dd41f596 | 4146 | __sched_fork(idle); |
06b83b5f | 4147 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4148 | idle->se.exec_start = sched_clock(); |
4149 | ||
1e1b6c51 | 4150 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4151 | /* |
4152 | * We're having a chicken and egg problem, even though we are | |
4153 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4154 | * lockdep check in task_group() will fail. | |
4155 | * | |
4156 | * Similar case to sched_fork(). / Alternatively we could | |
4157 | * use task_rq_lock() here and obtain the other rq->lock. | |
4158 | * | |
4159 | * Silence PROVE_RCU | |
4160 | */ | |
4161 | rcu_read_lock(); | |
dd41f596 | 4162 | __set_task_cpu(idle, cpu); |
6506cf6c | 4163 | rcu_read_unlock(); |
1da177e4 | 4164 | |
1da177e4 | 4165 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4166 | #if defined(CONFIG_SMP) |
4167 | idle->on_cpu = 1; | |
4866cde0 | 4168 | #endif |
05fa785c | 4169 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4170 | |
4171 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4172 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4173 | |
dd41f596 IM |
4174 | /* |
4175 | * The idle tasks have their own, simple scheduling class: | |
4176 | */ | |
4177 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4178 | ftrace_graph_init_idle_task(idle, cpu); |
6a61671b | 4179 | vtime_init_idle(idle); |
f1c6f1a7 CE |
4180 | #if defined(CONFIG_SMP) |
4181 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4182 | #endif | |
19978ca6 IM |
4183 | } |
4184 | ||
1da177e4 | 4185 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4186 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4187 | { | |
4188 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4189 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4190 | |
4191 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4192 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4193 | } |
4194 | ||
1da177e4 LT |
4195 | /* |
4196 | * This is how migration works: | |
4197 | * | |
969c7921 TH |
4198 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4199 | * stop_one_cpu(). | |
4200 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4201 | * off the CPU) | |
4202 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4203 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4204 | * it and puts it into the right queue. |
969c7921 TH |
4205 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4206 | * is done. | |
1da177e4 LT |
4207 | */ |
4208 | ||
4209 | /* | |
4210 | * Change a given task's CPU affinity. Migrate the thread to a | |
4211 | * proper CPU and schedule it away if the CPU it's executing on | |
4212 | * is removed from the allowed bitmask. | |
4213 | * | |
4214 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4215 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4216 | * call is not atomic; no spinlocks may be held. |
4217 | */ | |
96f874e2 | 4218 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4219 | { |
4220 | unsigned long flags; | |
70b97a7f | 4221 | struct rq *rq; |
969c7921 | 4222 | unsigned int dest_cpu; |
48f24c4d | 4223 | int ret = 0; |
1da177e4 LT |
4224 | |
4225 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4226 | |
db44fc01 YZ |
4227 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4228 | goto out; | |
4229 | ||
6ad4c188 | 4230 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4231 | ret = -EINVAL; |
4232 | goto out; | |
4233 | } | |
4234 | ||
1e1b6c51 | 4235 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4236 | |
1da177e4 | 4237 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4238 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4239 | goto out; |
4240 | ||
969c7921 | 4241 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4242 | if (p->on_rq) { |
969c7921 | 4243 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4244 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4245 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4246 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4247 | tlb_migrate_finish(p->mm); |
4248 | return 0; | |
4249 | } | |
4250 | out: | |
0122ec5b | 4251 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4252 | |
1da177e4 LT |
4253 | return ret; |
4254 | } | |
cd8ba7cd | 4255 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4256 | |
4257 | /* | |
41a2d6cf | 4258 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4259 | * this because either it can't run here any more (set_cpus_allowed() |
4260 | * away from this CPU, or CPU going down), or because we're | |
4261 | * attempting to rebalance this task on exec (sched_exec). | |
4262 | * | |
4263 | * So we race with normal scheduler movements, but that's OK, as long | |
4264 | * as the task is no longer on this CPU. | |
efc30814 KK |
4265 | * |
4266 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4267 | */ |
efc30814 | 4268 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4269 | { |
70b97a7f | 4270 | struct rq *rq_dest, *rq_src; |
e2912009 | 4271 | int ret = 0; |
1da177e4 | 4272 | |
e761b772 | 4273 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4274 | return ret; |
1da177e4 LT |
4275 | |
4276 | rq_src = cpu_rq(src_cpu); | |
4277 | rq_dest = cpu_rq(dest_cpu); | |
4278 | ||
0122ec5b | 4279 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4280 | double_rq_lock(rq_src, rq_dest); |
4281 | /* Already moved. */ | |
4282 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4283 | goto done; |
1da177e4 | 4284 | /* Affinity changed (again). */ |
fa17b507 | 4285 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4286 | goto fail; |
1da177e4 | 4287 | |
e2912009 PZ |
4288 | /* |
4289 | * If we're not on a rq, the next wake-up will ensure we're | |
4290 | * placed properly. | |
4291 | */ | |
fd2f4419 | 4292 | if (p->on_rq) { |
4ca9b72b | 4293 | dequeue_task(rq_src, p, 0); |
e2912009 | 4294 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4295 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4296 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4297 | } |
b1e38734 | 4298 | done: |
efc30814 | 4299 | ret = 1; |
b1e38734 | 4300 | fail: |
1da177e4 | 4301 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4302 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4303 | return ret; |
1da177e4 LT |
4304 | } |
4305 | ||
4306 | /* | |
969c7921 TH |
4307 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4308 | * and performs thread migration by bumping thread off CPU then | |
4309 | * 'pushing' onto another runqueue. | |
1da177e4 | 4310 | */ |
969c7921 | 4311 | static int migration_cpu_stop(void *data) |
1da177e4 | 4312 | { |
969c7921 | 4313 | struct migration_arg *arg = data; |
f7b4cddc | 4314 | |
969c7921 TH |
4315 | /* |
4316 | * The original target cpu might have gone down and we might | |
4317 | * be on another cpu but it doesn't matter. | |
4318 | */ | |
f7b4cddc | 4319 | local_irq_disable(); |
969c7921 | 4320 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4321 | local_irq_enable(); |
1da177e4 | 4322 | return 0; |
f7b4cddc ON |
4323 | } |
4324 | ||
1da177e4 | 4325 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4326 | |
054b9108 | 4327 | /* |
48c5ccae PZ |
4328 | * Ensures that the idle task is using init_mm right before its cpu goes |
4329 | * offline. | |
054b9108 | 4330 | */ |
48c5ccae | 4331 | void idle_task_exit(void) |
1da177e4 | 4332 | { |
48c5ccae | 4333 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4334 | |
48c5ccae | 4335 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4336 | |
48c5ccae PZ |
4337 | if (mm != &init_mm) |
4338 | switch_mm(mm, &init_mm, current); | |
4339 | mmdrop(mm); | |
1da177e4 LT |
4340 | } |
4341 | ||
4342 | /* | |
5d180232 PZ |
4343 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4344 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4345 | * nr_active count is stable. | |
4346 | * | |
4347 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4348 | */ |
5d180232 | 4349 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4350 | { |
5d180232 PZ |
4351 | long delta = calc_load_fold_active(rq); |
4352 | if (delta) | |
4353 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4354 | } |
4355 | ||
48f24c4d | 4356 | /* |
48c5ccae PZ |
4357 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4358 | * try_to_wake_up()->select_task_rq(). | |
4359 | * | |
4360 | * Called with rq->lock held even though we'er in stop_machine() and | |
4361 | * there's no concurrency possible, we hold the required locks anyway | |
4362 | * because of lock validation efforts. | |
1da177e4 | 4363 | */ |
48c5ccae | 4364 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4365 | { |
70b97a7f | 4366 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4367 | struct task_struct *next, *stop = rq->stop; |
4368 | int dest_cpu; | |
1da177e4 LT |
4369 | |
4370 | /* | |
48c5ccae PZ |
4371 | * Fudge the rq selection such that the below task selection loop |
4372 | * doesn't get stuck on the currently eligible stop task. | |
4373 | * | |
4374 | * We're currently inside stop_machine() and the rq is either stuck | |
4375 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4376 | * either way we should never end up calling schedule() until we're | |
4377 | * done here. | |
1da177e4 | 4378 | */ |
48c5ccae | 4379 | rq->stop = NULL; |
48f24c4d | 4380 | |
77bd3970 FW |
4381 | /* |
4382 | * put_prev_task() and pick_next_task() sched | |
4383 | * class method both need to have an up-to-date | |
4384 | * value of rq->clock[_task] | |
4385 | */ | |
4386 | update_rq_clock(rq); | |
4387 | ||
dd41f596 | 4388 | for ( ; ; ) { |
48c5ccae PZ |
4389 | /* |
4390 | * There's this thread running, bail when that's the only | |
4391 | * remaining thread. | |
4392 | */ | |
4393 | if (rq->nr_running == 1) | |
dd41f596 | 4394 | break; |
48c5ccae | 4395 | |
b67802ea | 4396 | next = pick_next_task(rq); |
48c5ccae | 4397 | BUG_ON(!next); |
79c53799 | 4398 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4399 | |
48c5ccae PZ |
4400 | /* Find suitable destination for @next, with force if needed. */ |
4401 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4402 | raw_spin_unlock(&rq->lock); | |
4403 | ||
4404 | __migrate_task(next, dead_cpu, dest_cpu); | |
4405 | ||
4406 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4407 | } |
dce48a84 | 4408 | |
48c5ccae | 4409 | rq->stop = stop; |
dce48a84 | 4410 | } |
48c5ccae | 4411 | |
1da177e4 LT |
4412 | #endif /* CONFIG_HOTPLUG_CPU */ |
4413 | ||
e692ab53 NP |
4414 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4415 | ||
4416 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4417 | { |
4418 | .procname = "sched_domain", | |
c57baf1e | 4419 | .mode = 0555, |
e0361851 | 4420 | }, |
56992309 | 4421 | {} |
e692ab53 NP |
4422 | }; |
4423 | ||
4424 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4425 | { |
4426 | .procname = "kernel", | |
c57baf1e | 4427 | .mode = 0555, |
e0361851 AD |
4428 | .child = sd_ctl_dir, |
4429 | }, | |
56992309 | 4430 | {} |
e692ab53 NP |
4431 | }; |
4432 | ||
4433 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4434 | { | |
4435 | struct ctl_table *entry = | |
5cf9f062 | 4436 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4437 | |
e692ab53 NP |
4438 | return entry; |
4439 | } | |
4440 | ||
6382bc90 MM |
4441 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4442 | { | |
cd790076 | 4443 | struct ctl_table *entry; |
6382bc90 | 4444 | |
cd790076 MM |
4445 | /* |
4446 | * In the intermediate directories, both the child directory and | |
4447 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4448 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4449 | * static strings and all have proc handlers. |
4450 | */ | |
4451 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4452 | if (entry->child) |
4453 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4454 | if (entry->proc_handler == NULL) |
4455 | kfree(entry->procname); | |
4456 | } | |
6382bc90 MM |
4457 | |
4458 | kfree(*tablep); | |
4459 | *tablep = NULL; | |
4460 | } | |
4461 | ||
201c373e | 4462 | static int min_load_idx = 0; |
fd9b86d3 | 4463 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4464 | |
e692ab53 | 4465 | static void |
e0361851 | 4466 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4467 | const char *procname, void *data, int maxlen, |
201c373e NK |
4468 | umode_t mode, proc_handler *proc_handler, |
4469 | bool load_idx) | |
e692ab53 | 4470 | { |
e692ab53 NP |
4471 | entry->procname = procname; |
4472 | entry->data = data; | |
4473 | entry->maxlen = maxlen; | |
4474 | entry->mode = mode; | |
4475 | entry->proc_handler = proc_handler; | |
201c373e NK |
4476 | |
4477 | if (load_idx) { | |
4478 | entry->extra1 = &min_load_idx; | |
4479 | entry->extra2 = &max_load_idx; | |
4480 | } | |
e692ab53 NP |
4481 | } |
4482 | ||
4483 | static struct ctl_table * | |
4484 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4485 | { | |
a5d8c348 | 4486 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 4487 | |
ad1cdc1d MM |
4488 | if (table == NULL) |
4489 | return NULL; | |
4490 | ||
e0361851 | 4491 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4492 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4493 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4494 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4495 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4496 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4497 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4498 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4499 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4500 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4501 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4502 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4503 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4504 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4505 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4506 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4507 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4508 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4509 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4510 | &sd->cache_nice_tries, |
201c373e | 4511 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4512 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4513 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 4514 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 4515 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 4516 | /* &table[12] is terminator */ |
e692ab53 NP |
4517 | |
4518 | return table; | |
4519 | } | |
4520 | ||
9a4e7159 | 4521 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4522 | { |
4523 | struct ctl_table *entry, *table; | |
4524 | struct sched_domain *sd; | |
4525 | int domain_num = 0, i; | |
4526 | char buf[32]; | |
4527 | ||
4528 | for_each_domain(cpu, sd) | |
4529 | domain_num++; | |
4530 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4531 | if (table == NULL) |
4532 | return NULL; | |
e692ab53 NP |
4533 | |
4534 | i = 0; | |
4535 | for_each_domain(cpu, sd) { | |
4536 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4537 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4538 | entry->mode = 0555; |
e692ab53 NP |
4539 | entry->child = sd_alloc_ctl_domain_table(sd); |
4540 | entry++; | |
4541 | i++; | |
4542 | } | |
4543 | return table; | |
4544 | } | |
4545 | ||
4546 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4547 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4548 | { |
6ad4c188 | 4549 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4550 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4551 | char buf[32]; | |
4552 | ||
7378547f MM |
4553 | WARN_ON(sd_ctl_dir[0].child); |
4554 | sd_ctl_dir[0].child = entry; | |
4555 | ||
ad1cdc1d MM |
4556 | if (entry == NULL) |
4557 | return; | |
4558 | ||
6ad4c188 | 4559 | for_each_possible_cpu(i) { |
e692ab53 | 4560 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4561 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4562 | entry->mode = 0555; |
e692ab53 | 4563 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4564 | entry++; |
e692ab53 | 4565 | } |
7378547f MM |
4566 | |
4567 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4568 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4569 | } | |
6382bc90 | 4570 | |
7378547f | 4571 | /* may be called multiple times per register */ |
6382bc90 MM |
4572 | static void unregister_sched_domain_sysctl(void) |
4573 | { | |
7378547f MM |
4574 | if (sd_sysctl_header) |
4575 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4576 | sd_sysctl_header = NULL; |
7378547f MM |
4577 | if (sd_ctl_dir[0].child) |
4578 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4579 | } |
e692ab53 | 4580 | #else |
6382bc90 MM |
4581 | static void register_sched_domain_sysctl(void) |
4582 | { | |
4583 | } | |
4584 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4585 | { |
4586 | } | |
4587 | #endif | |
4588 | ||
1f11eb6a GH |
4589 | static void set_rq_online(struct rq *rq) |
4590 | { | |
4591 | if (!rq->online) { | |
4592 | const struct sched_class *class; | |
4593 | ||
c6c4927b | 4594 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4595 | rq->online = 1; |
4596 | ||
4597 | for_each_class(class) { | |
4598 | if (class->rq_online) | |
4599 | class->rq_online(rq); | |
4600 | } | |
4601 | } | |
4602 | } | |
4603 | ||
4604 | static void set_rq_offline(struct rq *rq) | |
4605 | { | |
4606 | if (rq->online) { | |
4607 | const struct sched_class *class; | |
4608 | ||
4609 | for_each_class(class) { | |
4610 | if (class->rq_offline) | |
4611 | class->rq_offline(rq); | |
4612 | } | |
4613 | ||
c6c4927b | 4614 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4615 | rq->online = 0; |
4616 | } | |
4617 | } | |
4618 | ||
1da177e4 LT |
4619 | /* |
4620 | * migration_call - callback that gets triggered when a CPU is added. | |
4621 | * Here we can start up the necessary migration thread for the new CPU. | |
4622 | */ | |
48f24c4d IM |
4623 | static int __cpuinit |
4624 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 4625 | { |
48f24c4d | 4626 | int cpu = (long)hcpu; |
1da177e4 | 4627 | unsigned long flags; |
969c7921 | 4628 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 4629 | |
48c5ccae | 4630 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 4631 | |
1da177e4 | 4632 | case CPU_UP_PREPARE: |
a468d389 | 4633 | rq->calc_load_update = calc_load_update; |
1da177e4 | 4634 | break; |
48f24c4d | 4635 | |
1da177e4 | 4636 | case CPU_ONLINE: |
1f94ef59 | 4637 | /* Update our root-domain */ |
05fa785c | 4638 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 4639 | if (rq->rd) { |
c6c4927b | 4640 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
4641 | |
4642 | set_rq_online(rq); | |
1f94ef59 | 4643 | } |
05fa785c | 4644 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 4645 | break; |
48f24c4d | 4646 | |
1da177e4 | 4647 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 4648 | case CPU_DYING: |
317f3941 | 4649 | sched_ttwu_pending(); |
57d885fe | 4650 | /* Update our root-domain */ |
05fa785c | 4651 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 4652 | if (rq->rd) { |
c6c4927b | 4653 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 4654 | set_rq_offline(rq); |
57d885fe | 4655 | } |
48c5ccae PZ |
4656 | migrate_tasks(cpu); |
4657 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 4658 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 4659 | break; |
48c5ccae | 4660 | |
5d180232 | 4661 | case CPU_DEAD: |
f319da0c | 4662 | calc_load_migrate(rq); |
57d885fe | 4663 | break; |
1da177e4 LT |
4664 | #endif |
4665 | } | |
49c022e6 PZ |
4666 | |
4667 | update_max_interval(); | |
4668 | ||
1da177e4 LT |
4669 | return NOTIFY_OK; |
4670 | } | |
4671 | ||
f38b0820 PM |
4672 | /* |
4673 | * Register at high priority so that task migration (migrate_all_tasks) | |
4674 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 4675 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 4676 | */ |
26c2143b | 4677 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 4678 | .notifier_call = migration_call, |
50a323b7 | 4679 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
4680 | }; |
4681 | ||
3a101d05 TH |
4682 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
4683 | unsigned long action, void *hcpu) | |
4684 | { | |
4685 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 4686 | case CPU_STARTING: |
3a101d05 TH |
4687 | case CPU_DOWN_FAILED: |
4688 | set_cpu_active((long)hcpu, true); | |
4689 | return NOTIFY_OK; | |
4690 | default: | |
4691 | return NOTIFY_DONE; | |
4692 | } | |
4693 | } | |
4694 | ||
4695 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
4696 | unsigned long action, void *hcpu) | |
4697 | { | |
4698 | switch (action & ~CPU_TASKS_FROZEN) { | |
4699 | case CPU_DOWN_PREPARE: | |
4700 | set_cpu_active((long)hcpu, false); | |
4701 | return NOTIFY_OK; | |
4702 | default: | |
4703 | return NOTIFY_DONE; | |
4704 | } | |
4705 | } | |
4706 | ||
7babe8db | 4707 | static int __init migration_init(void) |
1da177e4 LT |
4708 | { |
4709 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 4710 | int err; |
48f24c4d | 4711 | |
3a101d05 | 4712 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
4713 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
4714 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
4715 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
4716 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 4717 | |
3a101d05 TH |
4718 | /* Register cpu active notifiers */ |
4719 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
4720 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
4721 | ||
a004cd42 | 4722 | return 0; |
1da177e4 | 4723 | } |
7babe8db | 4724 | early_initcall(migration_init); |
1da177e4 LT |
4725 | #endif |
4726 | ||
4727 | #ifdef CONFIG_SMP | |
476f3534 | 4728 | |
4cb98839 PZ |
4729 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
4730 | ||
3e9830dc | 4731 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 4732 | |
d039ac60 | 4733 | static __read_mostly int sched_debug_enabled; |
f6630114 | 4734 | |
d039ac60 | 4735 | static int __init sched_debug_setup(char *str) |
f6630114 | 4736 | { |
d039ac60 | 4737 | sched_debug_enabled = 1; |
f6630114 MT |
4738 | |
4739 | return 0; | |
4740 | } | |
d039ac60 PZ |
4741 | early_param("sched_debug", sched_debug_setup); |
4742 | ||
4743 | static inline bool sched_debug(void) | |
4744 | { | |
4745 | return sched_debug_enabled; | |
4746 | } | |
f6630114 | 4747 | |
7c16ec58 | 4748 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 4749 | struct cpumask *groupmask) |
1da177e4 | 4750 | { |
4dcf6aff | 4751 | struct sched_group *group = sd->groups; |
434d53b0 | 4752 | char str[256]; |
1da177e4 | 4753 | |
968ea6d8 | 4754 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 4755 | cpumask_clear(groupmask); |
4dcf6aff IM |
4756 | |
4757 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
4758 | ||
4759 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 4760 | printk("does not load-balance\n"); |
4dcf6aff | 4761 | if (sd->parent) |
3df0fc5b PZ |
4762 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
4763 | " has parent"); | |
4dcf6aff | 4764 | return -1; |
41c7ce9a NP |
4765 | } |
4766 | ||
3df0fc5b | 4767 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 4768 | |
758b2cdc | 4769 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
4770 | printk(KERN_ERR "ERROR: domain->span does not contain " |
4771 | "CPU%d\n", cpu); | |
4dcf6aff | 4772 | } |
758b2cdc | 4773 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
4774 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
4775 | " CPU%d\n", cpu); | |
4dcf6aff | 4776 | } |
1da177e4 | 4777 | |
4dcf6aff | 4778 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 4779 | do { |
4dcf6aff | 4780 | if (!group) { |
3df0fc5b PZ |
4781 | printk("\n"); |
4782 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
4783 | break; |
4784 | } | |
4785 | ||
c3decf0d PZ |
4786 | /* |
4787 | * Even though we initialize ->power to something semi-sane, | |
4788 | * we leave power_orig unset. This allows us to detect if | |
4789 | * domain iteration is still funny without causing /0 traps. | |
4790 | */ | |
4791 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
4792 | printk(KERN_CONT "\n"); |
4793 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
4794 | "set\n"); | |
4dcf6aff IM |
4795 | break; |
4796 | } | |
1da177e4 | 4797 | |
758b2cdc | 4798 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
4799 | printk(KERN_CONT "\n"); |
4800 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
4801 | break; |
4802 | } | |
1da177e4 | 4803 | |
cb83b629 PZ |
4804 | if (!(sd->flags & SD_OVERLAP) && |
4805 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
4806 | printk(KERN_CONT "\n"); |
4807 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
4808 | break; |
4809 | } | |
1da177e4 | 4810 | |
758b2cdc | 4811 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 4812 | |
968ea6d8 | 4813 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 4814 | |
3df0fc5b | 4815 | printk(KERN_CONT " %s", str); |
9c3f75cb | 4816 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 4817 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 4818 | group->sgp->power); |
381512cf | 4819 | } |
1da177e4 | 4820 | |
4dcf6aff IM |
4821 | group = group->next; |
4822 | } while (group != sd->groups); | |
3df0fc5b | 4823 | printk(KERN_CONT "\n"); |
1da177e4 | 4824 | |
758b2cdc | 4825 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 4826 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 4827 | |
758b2cdc RR |
4828 | if (sd->parent && |
4829 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
4830 | printk(KERN_ERR "ERROR: parent span is not a superset " |
4831 | "of domain->span\n"); | |
4dcf6aff IM |
4832 | return 0; |
4833 | } | |
1da177e4 | 4834 | |
4dcf6aff IM |
4835 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
4836 | { | |
4837 | int level = 0; | |
1da177e4 | 4838 | |
d039ac60 | 4839 | if (!sched_debug_enabled) |
f6630114 MT |
4840 | return; |
4841 | ||
4dcf6aff IM |
4842 | if (!sd) { |
4843 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
4844 | return; | |
4845 | } | |
1da177e4 | 4846 | |
4dcf6aff IM |
4847 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
4848 | ||
4849 | for (;;) { | |
4cb98839 | 4850 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 4851 | break; |
1da177e4 LT |
4852 | level++; |
4853 | sd = sd->parent; | |
33859f7f | 4854 | if (!sd) |
4dcf6aff IM |
4855 | break; |
4856 | } | |
1da177e4 | 4857 | } |
6d6bc0ad | 4858 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 4859 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
4860 | static inline bool sched_debug(void) |
4861 | { | |
4862 | return false; | |
4863 | } | |
6d6bc0ad | 4864 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 4865 | |
1a20ff27 | 4866 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 4867 | { |
758b2cdc | 4868 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
4869 | return 1; |
4870 | ||
4871 | /* Following flags need at least 2 groups */ | |
4872 | if (sd->flags & (SD_LOAD_BALANCE | | |
4873 | SD_BALANCE_NEWIDLE | | |
4874 | SD_BALANCE_FORK | | |
89c4710e SS |
4875 | SD_BALANCE_EXEC | |
4876 | SD_SHARE_CPUPOWER | | |
4877 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
4878 | if (sd->groups != sd->groups->next) |
4879 | return 0; | |
4880 | } | |
4881 | ||
4882 | /* Following flags don't use groups */ | |
c88d5910 | 4883 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
4884 | return 0; |
4885 | ||
4886 | return 1; | |
4887 | } | |
4888 | ||
48f24c4d IM |
4889 | static int |
4890 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
4891 | { |
4892 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
4893 | ||
4894 | if (sd_degenerate(parent)) | |
4895 | return 1; | |
4896 | ||
758b2cdc | 4897 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
4898 | return 0; |
4899 | ||
245af2c7 SS |
4900 | /* Flags needing groups don't count if only 1 group in parent */ |
4901 | if (parent->groups == parent->groups->next) { | |
4902 | pflags &= ~(SD_LOAD_BALANCE | | |
4903 | SD_BALANCE_NEWIDLE | | |
4904 | SD_BALANCE_FORK | | |
89c4710e SS |
4905 | SD_BALANCE_EXEC | |
4906 | SD_SHARE_CPUPOWER | | |
4907 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
4908 | if (nr_node_ids == 1) |
4909 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
4910 | } |
4911 | if (~cflags & pflags) | |
4912 | return 0; | |
4913 | ||
4914 | return 1; | |
4915 | } | |
4916 | ||
dce840a0 | 4917 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 4918 | { |
dce840a0 | 4919 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 4920 | |
68e74568 | 4921 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
4922 | free_cpumask_var(rd->rto_mask); |
4923 | free_cpumask_var(rd->online); | |
4924 | free_cpumask_var(rd->span); | |
4925 | kfree(rd); | |
4926 | } | |
4927 | ||
57d885fe GH |
4928 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
4929 | { | |
a0490fa3 | 4930 | struct root_domain *old_rd = NULL; |
57d885fe | 4931 | unsigned long flags; |
57d885fe | 4932 | |
05fa785c | 4933 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
4934 | |
4935 | if (rq->rd) { | |
a0490fa3 | 4936 | old_rd = rq->rd; |
57d885fe | 4937 | |
c6c4927b | 4938 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 4939 | set_rq_offline(rq); |
57d885fe | 4940 | |
c6c4927b | 4941 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 4942 | |
a0490fa3 IM |
4943 | /* |
4944 | * If we dont want to free the old_rt yet then | |
4945 | * set old_rd to NULL to skip the freeing later | |
4946 | * in this function: | |
4947 | */ | |
4948 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
4949 | old_rd = NULL; | |
57d885fe GH |
4950 | } |
4951 | ||
4952 | atomic_inc(&rd->refcount); | |
4953 | rq->rd = rd; | |
4954 | ||
c6c4927b | 4955 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 4956 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 4957 | set_rq_online(rq); |
57d885fe | 4958 | |
05fa785c | 4959 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
4960 | |
4961 | if (old_rd) | |
dce840a0 | 4962 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
4963 | } |
4964 | ||
68c38fc3 | 4965 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
4966 | { |
4967 | memset(rd, 0, sizeof(*rd)); | |
4968 | ||
68c38fc3 | 4969 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 4970 | goto out; |
68c38fc3 | 4971 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 4972 | goto free_span; |
68c38fc3 | 4973 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 4974 | goto free_online; |
6e0534f2 | 4975 | |
68c38fc3 | 4976 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 4977 | goto free_rto_mask; |
c6c4927b | 4978 | return 0; |
6e0534f2 | 4979 | |
68e74568 RR |
4980 | free_rto_mask: |
4981 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
4982 | free_online: |
4983 | free_cpumask_var(rd->online); | |
4984 | free_span: | |
4985 | free_cpumask_var(rd->span); | |
0c910d28 | 4986 | out: |
c6c4927b | 4987 | return -ENOMEM; |
57d885fe GH |
4988 | } |
4989 | ||
029632fb PZ |
4990 | /* |
4991 | * By default the system creates a single root-domain with all cpus as | |
4992 | * members (mimicking the global state we have today). | |
4993 | */ | |
4994 | struct root_domain def_root_domain; | |
4995 | ||
57d885fe GH |
4996 | static void init_defrootdomain(void) |
4997 | { | |
68c38fc3 | 4998 | init_rootdomain(&def_root_domain); |
c6c4927b | 4999 | |
57d885fe GH |
5000 | atomic_set(&def_root_domain.refcount, 1); |
5001 | } | |
5002 | ||
dc938520 | 5003 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5004 | { |
5005 | struct root_domain *rd; | |
5006 | ||
5007 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5008 | if (!rd) | |
5009 | return NULL; | |
5010 | ||
68c38fc3 | 5011 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5012 | kfree(rd); |
5013 | return NULL; | |
5014 | } | |
57d885fe GH |
5015 | |
5016 | return rd; | |
5017 | } | |
5018 | ||
e3589f6c PZ |
5019 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5020 | { | |
5021 | struct sched_group *tmp, *first; | |
5022 | ||
5023 | if (!sg) | |
5024 | return; | |
5025 | ||
5026 | first = sg; | |
5027 | do { | |
5028 | tmp = sg->next; | |
5029 | ||
5030 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5031 | kfree(sg->sgp); | |
5032 | ||
5033 | kfree(sg); | |
5034 | sg = tmp; | |
5035 | } while (sg != first); | |
5036 | } | |
5037 | ||
dce840a0 PZ |
5038 | static void free_sched_domain(struct rcu_head *rcu) |
5039 | { | |
5040 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5041 | |
5042 | /* | |
5043 | * If its an overlapping domain it has private groups, iterate and | |
5044 | * nuke them all. | |
5045 | */ | |
5046 | if (sd->flags & SD_OVERLAP) { | |
5047 | free_sched_groups(sd->groups, 1); | |
5048 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5049 | kfree(sd->groups->sgp); |
dce840a0 | 5050 | kfree(sd->groups); |
9c3f75cb | 5051 | } |
dce840a0 PZ |
5052 | kfree(sd); |
5053 | } | |
5054 | ||
5055 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5056 | { | |
5057 | call_rcu(&sd->rcu, free_sched_domain); | |
5058 | } | |
5059 | ||
5060 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5061 | { | |
5062 | for (; sd; sd = sd->parent) | |
5063 | destroy_sched_domain(sd, cpu); | |
5064 | } | |
5065 | ||
518cd623 PZ |
5066 | /* |
5067 | * Keep a special pointer to the highest sched_domain that has | |
5068 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5069 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5070 | * | |
5071 | * Also keep a unique ID per domain (we use the first cpu number in | |
5072 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5073 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5074 | */ |
5075 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5076 | DEFINE_PER_CPU(int, sd_llc_id); | |
5077 | ||
5078 | static void update_top_cache_domain(int cpu) | |
5079 | { | |
5080 | struct sched_domain *sd; | |
5081 | int id = cpu; | |
5082 | ||
5083 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
37407ea7 | 5084 | if (sd) |
518cd623 PZ |
5085 | id = cpumask_first(sched_domain_span(sd)); |
5086 | ||
5087 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5088 | per_cpu(sd_llc_id, cpu) = id; | |
5089 | } | |
5090 | ||
1da177e4 | 5091 | /* |
0eab9146 | 5092 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5093 | * hold the hotplug lock. |
5094 | */ | |
0eab9146 IM |
5095 | static void |
5096 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5097 | { |
70b97a7f | 5098 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5099 | struct sched_domain *tmp; |
5100 | ||
5101 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5102 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5103 | struct sched_domain *parent = tmp->parent; |
5104 | if (!parent) | |
5105 | break; | |
f29c9b1c | 5106 | |
1a848870 | 5107 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5108 | tmp->parent = parent->parent; |
1a848870 SS |
5109 | if (parent->parent) |
5110 | parent->parent->child = tmp; | |
dce840a0 | 5111 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5112 | } else |
5113 | tmp = tmp->parent; | |
245af2c7 SS |
5114 | } |
5115 | ||
1a848870 | 5116 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5117 | tmp = sd; |
245af2c7 | 5118 | sd = sd->parent; |
dce840a0 | 5119 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5120 | if (sd) |
5121 | sd->child = NULL; | |
5122 | } | |
1da177e4 | 5123 | |
4cb98839 | 5124 | sched_domain_debug(sd, cpu); |
1da177e4 | 5125 | |
57d885fe | 5126 | rq_attach_root(rq, rd); |
dce840a0 | 5127 | tmp = rq->sd; |
674311d5 | 5128 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5129 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5130 | |
5131 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5132 | } |
5133 | ||
5134 | /* cpus with isolated domains */ | |
dcc30a35 | 5135 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5136 | |
5137 | /* Setup the mask of cpus configured for isolated domains */ | |
5138 | static int __init isolated_cpu_setup(char *str) | |
5139 | { | |
bdddd296 | 5140 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5141 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5142 | return 1; |
5143 | } | |
5144 | ||
8927f494 | 5145 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5146 | |
d3081f52 PZ |
5147 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5148 | { | |
5149 | return cpumask_of_node(cpu_to_node(cpu)); | |
5150 | } | |
5151 | ||
dce840a0 PZ |
5152 | struct sd_data { |
5153 | struct sched_domain **__percpu sd; | |
5154 | struct sched_group **__percpu sg; | |
9c3f75cb | 5155 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5156 | }; |
5157 | ||
49a02c51 | 5158 | struct s_data { |
21d42ccf | 5159 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5160 | struct root_domain *rd; |
5161 | }; | |
5162 | ||
2109b99e | 5163 | enum s_alloc { |
2109b99e | 5164 | sa_rootdomain, |
21d42ccf | 5165 | sa_sd, |
dce840a0 | 5166 | sa_sd_storage, |
2109b99e AH |
5167 | sa_none, |
5168 | }; | |
5169 | ||
54ab4ff4 PZ |
5170 | struct sched_domain_topology_level; |
5171 | ||
5172 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5173 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5174 | ||
e3589f6c PZ |
5175 | #define SDTL_OVERLAP 0x01 |
5176 | ||
eb7a74e6 | 5177 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5178 | sched_domain_init_f init; |
5179 | sched_domain_mask_f mask; | |
e3589f6c | 5180 | int flags; |
cb83b629 | 5181 | int numa_level; |
54ab4ff4 | 5182 | struct sd_data data; |
eb7a74e6 PZ |
5183 | }; |
5184 | ||
c1174876 PZ |
5185 | /* |
5186 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5187 | * domain traversal. | |
5188 | * | |
5189 | * Asymmetric node setups can result in situations where the domain tree is of | |
5190 | * unequal depth, make sure to skip domains that already cover the entire | |
5191 | * range. | |
5192 | * | |
5193 | * In that case build_sched_domains() will have terminated the iteration early | |
5194 | * and our sibling sd spans will be empty. Domains should always include the | |
5195 | * cpu they're built on, so check that. | |
5196 | * | |
5197 | */ | |
5198 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5199 | { | |
5200 | const struct cpumask *span = sched_domain_span(sd); | |
5201 | struct sd_data *sdd = sd->private; | |
5202 | struct sched_domain *sibling; | |
5203 | int i; | |
5204 | ||
5205 | for_each_cpu(i, span) { | |
5206 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5207 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5208 | continue; | |
5209 | ||
5210 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5211 | } | |
5212 | } | |
5213 | ||
5214 | /* | |
5215 | * Return the canonical balance cpu for this group, this is the first cpu | |
5216 | * of this group that's also in the iteration mask. | |
5217 | */ | |
5218 | int group_balance_cpu(struct sched_group *sg) | |
5219 | { | |
5220 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5221 | } | |
5222 | ||
e3589f6c PZ |
5223 | static int |
5224 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5225 | { | |
5226 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5227 | const struct cpumask *span = sched_domain_span(sd); | |
5228 | struct cpumask *covered = sched_domains_tmpmask; | |
5229 | struct sd_data *sdd = sd->private; | |
5230 | struct sched_domain *child; | |
5231 | int i; | |
5232 | ||
5233 | cpumask_clear(covered); | |
5234 | ||
5235 | for_each_cpu(i, span) { | |
5236 | struct cpumask *sg_span; | |
5237 | ||
5238 | if (cpumask_test_cpu(i, covered)) | |
5239 | continue; | |
5240 | ||
c1174876 PZ |
5241 | child = *per_cpu_ptr(sdd->sd, i); |
5242 | ||
5243 | /* See the comment near build_group_mask(). */ | |
5244 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5245 | continue; | |
5246 | ||
e3589f6c | 5247 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5248 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5249 | |
5250 | if (!sg) | |
5251 | goto fail; | |
5252 | ||
5253 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5254 | if (child->child) { |
5255 | child = child->child; | |
5256 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5257 | } else | |
5258 | cpumask_set_cpu(i, sg_span); | |
5259 | ||
5260 | cpumask_or(covered, covered, sg_span); | |
5261 | ||
74a5ce20 | 5262 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5263 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5264 | build_group_mask(sd, sg); | |
5265 | ||
c3decf0d PZ |
5266 | /* |
5267 | * Initialize sgp->power such that even if we mess up the | |
5268 | * domains and no possible iteration will get us here, we won't | |
5269 | * die on a /0 trap. | |
5270 | */ | |
5271 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 5272 | |
c1174876 PZ |
5273 | /* |
5274 | * Make sure the first group of this domain contains the | |
5275 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5276 | * breaks. See update_sg_lb_stats(). | |
5277 | */ | |
74a5ce20 | 5278 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5279 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5280 | groups = sg; |
5281 | ||
5282 | if (!first) | |
5283 | first = sg; | |
5284 | if (last) | |
5285 | last->next = sg; | |
5286 | last = sg; | |
5287 | last->next = first; | |
5288 | } | |
5289 | sd->groups = groups; | |
5290 | ||
5291 | return 0; | |
5292 | ||
5293 | fail: | |
5294 | free_sched_groups(first, 0); | |
5295 | ||
5296 | return -ENOMEM; | |
5297 | } | |
5298 | ||
dce840a0 | 5299 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5300 | { |
dce840a0 PZ |
5301 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5302 | struct sched_domain *child = sd->child; | |
1da177e4 | 5303 | |
dce840a0 PZ |
5304 | if (child) |
5305 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5306 | |
9c3f75cb | 5307 | if (sg) { |
dce840a0 | 5308 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5309 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5310 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5311 | } |
dce840a0 PZ |
5312 | |
5313 | return cpu; | |
1e9f28fa | 5314 | } |
1e9f28fa | 5315 | |
01a08546 | 5316 | /* |
dce840a0 PZ |
5317 | * build_sched_groups will build a circular linked list of the groups |
5318 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5319 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5320 | * |
5321 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5322 | */ |
e3589f6c PZ |
5323 | static int |
5324 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5325 | { |
dce840a0 PZ |
5326 | struct sched_group *first = NULL, *last = NULL; |
5327 | struct sd_data *sdd = sd->private; | |
5328 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5329 | struct cpumask *covered; |
dce840a0 | 5330 | int i; |
9c1cfda2 | 5331 | |
e3589f6c PZ |
5332 | get_group(cpu, sdd, &sd->groups); |
5333 | atomic_inc(&sd->groups->ref); | |
5334 | ||
5335 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
5336 | return 0; | |
5337 | ||
f96225fd PZ |
5338 | lockdep_assert_held(&sched_domains_mutex); |
5339 | covered = sched_domains_tmpmask; | |
5340 | ||
dce840a0 | 5341 | cpumask_clear(covered); |
6711cab4 | 5342 | |
dce840a0 PZ |
5343 | for_each_cpu(i, span) { |
5344 | struct sched_group *sg; | |
5345 | int group = get_group(i, sdd, &sg); | |
5346 | int j; | |
6711cab4 | 5347 | |
dce840a0 PZ |
5348 | if (cpumask_test_cpu(i, covered)) |
5349 | continue; | |
6711cab4 | 5350 | |
dce840a0 | 5351 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5352 | sg->sgp->power = 0; |
c1174876 | 5353 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5354 | |
dce840a0 PZ |
5355 | for_each_cpu(j, span) { |
5356 | if (get_group(j, sdd, NULL) != group) | |
5357 | continue; | |
0601a88d | 5358 | |
dce840a0 PZ |
5359 | cpumask_set_cpu(j, covered); |
5360 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5361 | } | |
0601a88d | 5362 | |
dce840a0 PZ |
5363 | if (!first) |
5364 | first = sg; | |
5365 | if (last) | |
5366 | last->next = sg; | |
5367 | last = sg; | |
5368 | } | |
5369 | last->next = first; | |
e3589f6c PZ |
5370 | |
5371 | return 0; | |
0601a88d | 5372 | } |
51888ca2 | 5373 | |
89c4710e SS |
5374 | /* |
5375 | * Initialize sched groups cpu_power. | |
5376 | * | |
5377 | * cpu_power indicates the capacity of sched group, which is used while | |
5378 | * distributing the load between different sched groups in a sched domain. | |
5379 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5380 | * there are asymmetries in the topology. If there are asymmetries, group | |
5381 | * having more cpu_power will pickup more load compared to the group having | |
5382 | * less cpu_power. | |
89c4710e SS |
5383 | */ |
5384 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5385 | { | |
e3589f6c | 5386 | struct sched_group *sg = sd->groups; |
89c4710e | 5387 | |
e3589f6c PZ |
5388 | WARN_ON(!sd || !sg); |
5389 | ||
5390 | do { | |
5391 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5392 | sg = sg->next; | |
5393 | } while (sg != sd->groups); | |
89c4710e | 5394 | |
c1174876 | 5395 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5396 | return; |
aae6d3dd | 5397 | |
d274cb30 | 5398 | update_group_power(sd, cpu); |
69e1e811 | 5399 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5400 | } |
5401 | ||
029632fb PZ |
5402 | int __weak arch_sd_sibling_asym_packing(void) |
5403 | { | |
5404 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5405 | } |
5406 | ||
7c16ec58 MT |
5407 | /* |
5408 | * Initializers for schedule domains | |
5409 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5410 | */ | |
5411 | ||
a5d8c348 IM |
5412 | #ifdef CONFIG_SCHED_DEBUG |
5413 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5414 | #else | |
5415 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5416 | #endif | |
5417 | ||
54ab4ff4 PZ |
5418 | #define SD_INIT_FUNC(type) \ |
5419 | static noinline struct sched_domain * \ | |
5420 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5421 | { \ | |
5422 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5423 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5424 | SD_INIT_NAME(sd, type); \ |
5425 | sd->private = &tl->data; \ | |
5426 | return sd; \ | |
7c16ec58 MT |
5427 | } |
5428 | ||
5429 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5430 | #ifdef CONFIG_SCHED_SMT |
5431 | SD_INIT_FUNC(SIBLING) | |
5432 | #endif | |
5433 | #ifdef CONFIG_SCHED_MC | |
5434 | SD_INIT_FUNC(MC) | |
5435 | #endif | |
01a08546 HC |
5436 | #ifdef CONFIG_SCHED_BOOK |
5437 | SD_INIT_FUNC(BOOK) | |
5438 | #endif | |
7c16ec58 | 5439 | |
1d3504fc | 5440 | static int default_relax_domain_level = -1; |
60495e77 | 5441 | int sched_domain_level_max; |
1d3504fc HS |
5442 | |
5443 | static int __init setup_relax_domain_level(char *str) | |
5444 | { | |
a841f8ce DS |
5445 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5446 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5447 | |
1d3504fc HS |
5448 | return 1; |
5449 | } | |
5450 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5451 | ||
5452 | static void set_domain_attribute(struct sched_domain *sd, | |
5453 | struct sched_domain_attr *attr) | |
5454 | { | |
5455 | int request; | |
5456 | ||
5457 | if (!attr || attr->relax_domain_level < 0) { | |
5458 | if (default_relax_domain_level < 0) | |
5459 | return; | |
5460 | else | |
5461 | request = default_relax_domain_level; | |
5462 | } else | |
5463 | request = attr->relax_domain_level; | |
5464 | if (request < sd->level) { | |
5465 | /* turn off idle balance on this domain */ | |
c88d5910 | 5466 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5467 | } else { |
5468 | /* turn on idle balance on this domain */ | |
c88d5910 | 5469 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5470 | } |
5471 | } | |
5472 | ||
54ab4ff4 PZ |
5473 | static void __sdt_free(const struct cpumask *cpu_map); |
5474 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5475 | ||
2109b99e AH |
5476 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5477 | const struct cpumask *cpu_map) | |
5478 | { | |
5479 | switch (what) { | |
2109b99e | 5480 | case sa_rootdomain: |
822ff793 PZ |
5481 | if (!atomic_read(&d->rd->refcount)) |
5482 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5483 | case sa_sd: |
5484 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5485 | case sa_sd_storage: |
54ab4ff4 | 5486 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5487 | case sa_none: |
5488 | break; | |
5489 | } | |
5490 | } | |
3404c8d9 | 5491 | |
2109b99e AH |
5492 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5493 | const struct cpumask *cpu_map) | |
5494 | { | |
dce840a0 PZ |
5495 | memset(d, 0, sizeof(*d)); |
5496 | ||
54ab4ff4 PZ |
5497 | if (__sdt_alloc(cpu_map)) |
5498 | return sa_sd_storage; | |
dce840a0 PZ |
5499 | d->sd = alloc_percpu(struct sched_domain *); |
5500 | if (!d->sd) | |
5501 | return sa_sd_storage; | |
2109b99e | 5502 | d->rd = alloc_rootdomain(); |
dce840a0 | 5503 | if (!d->rd) |
21d42ccf | 5504 | return sa_sd; |
2109b99e AH |
5505 | return sa_rootdomain; |
5506 | } | |
57d885fe | 5507 | |
dce840a0 PZ |
5508 | /* |
5509 | * NULL the sd_data elements we've used to build the sched_domain and | |
5510 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5511 | * will not free the data we're using. | |
5512 | */ | |
5513 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5514 | { | |
5515 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5516 | |
5517 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5518 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5519 | ||
e3589f6c | 5520 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5521 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5522 | |
5523 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5524 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5525 | } |
5526 | ||
2c402dc3 PZ |
5527 | #ifdef CONFIG_SCHED_SMT |
5528 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 5529 | { |
2c402dc3 | 5530 | return topology_thread_cpumask(cpu); |
3bd65a80 | 5531 | } |
2c402dc3 | 5532 | #endif |
7f4588f3 | 5533 | |
d069b916 PZ |
5534 | /* |
5535 | * Topology list, bottom-up. | |
5536 | */ | |
2c402dc3 | 5537 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
5538 | #ifdef CONFIG_SCHED_SMT |
5539 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 5540 | #endif |
1e9f28fa | 5541 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 5542 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 5543 | #endif |
d069b916 PZ |
5544 | #ifdef CONFIG_SCHED_BOOK |
5545 | { sd_init_BOOK, cpu_book_mask, }, | |
5546 | #endif | |
5547 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
5548 | { NULL, }, |
5549 | }; | |
5550 | ||
5551 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
5552 | ||
cb83b629 PZ |
5553 | #ifdef CONFIG_NUMA |
5554 | ||
5555 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
5556 | static int *sched_domains_numa_distance; |
5557 | static struct cpumask ***sched_domains_numa_masks; | |
5558 | static int sched_domains_curr_level; | |
5559 | ||
cb83b629 PZ |
5560 | static inline int sd_local_flags(int level) |
5561 | { | |
10717dcd | 5562 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
5563 | return 0; |
5564 | ||
5565 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
5566 | } | |
5567 | ||
5568 | static struct sched_domain * | |
5569 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
5570 | { | |
5571 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
5572 | int level = tl->numa_level; | |
5573 | int sd_weight = cpumask_weight( | |
5574 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
5575 | ||
5576 | *sd = (struct sched_domain){ | |
5577 | .min_interval = sd_weight, | |
5578 | .max_interval = 2*sd_weight, | |
5579 | .busy_factor = 32, | |
870a0bb5 | 5580 | .imbalance_pct = 125, |
cb83b629 PZ |
5581 | .cache_nice_tries = 2, |
5582 | .busy_idx = 3, | |
5583 | .idle_idx = 2, | |
5584 | .newidle_idx = 0, | |
5585 | .wake_idx = 0, | |
5586 | .forkexec_idx = 0, | |
5587 | ||
5588 | .flags = 1*SD_LOAD_BALANCE | |
5589 | | 1*SD_BALANCE_NEWIDLE | |
5590 | | 0*SD_BALANCE_EXEC | |
5591 | | 0*SD_BALANCE_FORK | |
5592 | | 0*SD_BALANCE_WAKE | |
5593 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 5594 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
5595 | | 0*SD_SHARE_PKG_RESOURCES |
5596 | | 1*SD_SERIALIZE | |
5597 | | 0*SD_PREFER_SIBLING | |
5598 | | sd_local_flags(level) | |
5599 | , | |
5600 | .last_balance = jiffies, | |
5601 | .balance_interval = sd_weight, | |
5602 | }; | |
5603 | SD_INIT_NAME(sd, NUMA); | |
5604 | sd->private = &tl->data; | |
5605 | ||
5606 | /* | |
5607 | * Ugly hack to pass state to sd_numa_mask()... | |
5608 | */ | |
5609 | sched_domains_curr_level = tl->numa_level; | |
5610 | ||
5611 | return sd; | |
5612 | } | |
5613 | ||
5614 | static const struct cpumask *sd_numa_mask(int cpu) | |
5615 | { | |
5616 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
5617 | } | |
5618 | ||
d039ac60 PZ |
5619 | static void sched_numa_warn(const char *str) |
5620 | { | |
5621 | static int done = false; | |
5622 | int i,j; | |
5623 | ||
5624 | if (done) | |
5625 | return; | |
5626 | ||
5627 | done = true; | |
5628 | ||
5629 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
5630 | ||
5631 | for (i = 0; i < nr_node_ids; i++) { | |
5632 | printk(KERN_WARNING " "); | |
5633 | for (j = 0; j < nr_node_ids; j++) | |
5634 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
5635 | printk(KERN_CONT "\n"); | |
5636 | } | |
5637 | printk(KERN_WARNING "\n"); | |
5638 | } | |
5639 | ||
5640 | static bool find_numa_distance(int distance) | |
5641 | { | |
5642 | int i; | |
5643 | ||
5644 | if (distance == node_distance(0, 0)) | |
5645 | return true; | |
5646 | ||
5647 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5648 | if (sched_domains_numa_distance[i] == distance) | |
5649 | return true; | |
5650 | } | |
5651 | ||
5652 | return false; | |
5653 | } | |
5654 | ||
cb83b629 PZ |
5655 | static void sched_init_numa(void) |
5656 | { | |
5657 | int next_distance, curr_distance = node_distance(0, 0); | |
5658 | struct sched_domain_topology_level *tl; | |
5659 | int level = 0; | |
5660 | int i, j, k; | |
5661 | ||
cb83b629 PZ |
5662 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
5663 | if (!sched_domains_numa_distance) | |
5664 | return; | |
5665 | ||
5666 | /* | |
5667 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
5668 | * unique distances in the node_distance() table. | |
5669 | * | |
5670 | * Assumes node_distance(0,j) includes all distances in | |
5671 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
5672 | */ |
5673 | next_distance = curr_distance; | |
5674 | for (i = 0; i < nr_node_ids; i++) { | |
5675 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
5676 | for (k = 0; k < nr_node_ids; k++) { |
5677 | int distance = node_distance(i, k); | |
5678 | ||
5679 | if (distance > curr_distance && | |
5680 | (distance < next_distance || | |
5681 | next_distance == curr_distance)) | |
5682 | next_distance = distance; | |
5683 | ||
5684 | /* | |
5685 | * While not a strong assumption it would be nice to know | |
5686 | * about cases where if node A is connected to B, B is not | |
5687 | * equally connected to A. | |
5688 | */ | |
5689 | if (sched_debug() && node_distance(k, i) != distance) | |
5690 | sched_numa_warn("Node-distance not symmetric"); | |
5691 | ||
5692 | if (sched_debug() && i && !find_numa_distance(distance)) | |
5693 | sched_numa_warn("Node-0 not representative"); | |
5694 | } | |
5695 | if (next_distance != curr_distance) { | |
5696 | sched_domains_numa_distance[level++] = next_distance; | |
5697 | sched_domains_numa_levels = level; | |
5698 | curr_distance = next_distance; | |
5699 | } else break; | |
cb83b629 | 5700 | } |
d039ac60 PZ |
5701 | |
5702 | /* | |
5703 | * In case of sched_debug() we verify the above assumption. | |
5704 | */ | |
5705 | if (!sched_debug()) | |
5706 | break; | |
cb83b629 PZ |
5707 | } |
5708 | /* | |
5709 | * 'level' contains the number of unique distances, excluding the | |
5710 | * identity distance node_distance(i,i). | |
5711 | * | |
28b4a521 | 5712 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
5713 | * numbers. |
5714 | */ | |
5715 | ||
5f7865f3 TC |
5716 | /* |
5717 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
5718 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
5719 | * the array will contain less then 'level' members. This could be | |
5720 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
5721 | * in other functions. | |
5722 | * | |
5723 | * We reset it to 'level' at the end of this function. | |
5724 | */ | |
5725 | sched_domains_numa_levels = 0; | |
5726 | ||
cb83b629 PZ |
5727 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
5728 | if (!sched_domains_numa_masks) | |
5729 | return; | |
5730 | ||
5731 | /* | |
5732 | * Now for each level, construct a mask per node which contains all | |
5733 | * cpus of nodes that are that many hops away from us. | |
5734 | */ | |
5735 | for (i = 0; i < level; i++) { | |
5736 | sched_domains_numa_masks[i] = | |
5737 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
5738 | if (!sched_domains_numa_masks[i]) | |
5739 | return; | |
5740 | ||
5741 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 5742 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
5743 | if (!mask) |
5744 | return; | |
5745 | ||
5746 | sched_domains_numa_masks[i][j] = mask; | |
5747 | ||
5748 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 5749 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
5750 | continue; |
5751 | ||
5752 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
5753 | } | |
5754 | } | |
5755 | } | |
5756 | ||
5757 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
5758 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
5759 | if (!tl) | |
5760 | return; | |
5761 | ||
5762 | /* | |
5763 | * Copy the default topology bits.. | |
5764 | */ | |
5765 | for (i = 0; default_topology[i].init; i++) | |
5766 | tl[i] = default_topology[i]; | |
5767 | ||
5768 | /* | |
5769 | * .. and append 'j' levels of NUMA goodness. | |
5770 | */ | |
5771 | for (j = 0; j < level; i++, j++) { | |
5772 | tl[i] = (struct sched_domain_topology_level){ | |
5773 | .init = sd_numa_init, | |
5774 | .mask = sd_numa_mask, | |
5775 | .flags = SDTL_OVERLAP, | |
5776 | .numa_level = j, | |
5777 | }; | |
5778 | } | |
5779 | ||
5780 | sched_domain_topology = tl; | |
5f7865f3 TC |
5781 | |
5782 | sched_domains_numa_levels = level; | |
cb83b629 | 5783 | } |
301a5cba TC |
5784 | |
5785 | static void sched_domains_numa_masks_set(int cpu) | |
5786 | { | |
5787 | int i, j; | |
5788 | int node = cpu_to_node(cpu); | |
5789 | ||
5790 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5791 | for (j = 0; j < nr_node_ids; j++) { | |
5792 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
5793 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
5794 | } | |
5795 | } | |
5796 | } | |
5797 | ||
5798 | static void sched_domains_numa_masks_clear(int cpu) | |
5799 | { | |
5800 | int i, j; | |
5801 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5802 | for (j = 0; j < nr_node_ids; j++) | |
5803 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
5804 | } | |
5805 | } | |
5806 | ||
5807 | /* | |
5808 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
5809 | * are onlined. | |
5810 | */ | |
5811 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
5812 | unsigned long action, | |
5813 | void *hcpu) | |
5814 | { | |
5815 | int cpu = (long)hcpu; | |
5816 | ||
5817 | switch (action & ~CPU_TASKS_FROZEN) { | |
5818 | case CPU_ONLINE: | |
5819 | sched_domains_numa_masks_set(cpu); | |
5820 | break; | |
5821 | ||
5822 | case CPU_DEAD: | |
5823 | sched_domains_numa_masks_clear(cpu); | |
5824 | break; | |
5825 | ||
5826 | default: | |
5827 | return NOTIFY_DONE; | |
5828 | } | |
5829 | ||
5830 | return NOTIFY_OK; | |
cb83b629 PZ |
5831 | } |
5832 | #else | |
5833 | static inline void sched_init_numa(void) | |
5834 | { | |
5835 | } | |
301a5cba TC |
5836 | |
5837 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
5838 | unsigned long action, | |
5839 | void *hcpu) | |
5840 | { | |
5841 | return 0; | |
5842 | } | |
cb83b629 PZ |
5843 | #endif /* CONFIG_NUMA */ |
5844 | ||
54ab4ff4 PZ |
5845 | static int __sdt_alloc(const struct cpumask *cpu_map) |
5846 | { | |
5847 | struct sched_domain_topology_level *tl; | |
5848 | int j; | |
5849 | ||
5850 | for (tl = sched_domain_topology; tl->init; tl++) { | |
5851 | struct sd_data *sdd = &tl->data; | |
5852 | ||
5853 | sdd->sd = alloc_percpu(struct sched_domain *); | |
5854 | if (!sdd->sd) | |
5855 | return -ENOMEM; | |
5856 | ||
5857 | sdd->sg = alloc_percpu(struct sched_group *); | |
5858 | if (!sdd->sg) | |
5859 | return -ENOMEM; | |
5860 | ||
9c3f75cb PZ |
5861 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
5862 | if (!sdd->sgp) | |
5863 | return -ENOMEM; | |
5864 | ||
54ab4ff4 PZ |
5865 | for_each_cpu(j, cpu_map) { |
5866 | struct sched_domain *sd; | |
5867 | struct sched_group *sg; | |
9c3f75cb | 5868 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
5869 | |
5870 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
5871 | GFP_KERNEL, cpu_to_node(j)); | |
5872 | if (!sd) | |
5873 | return -ENOMEM; | |
5874 | ||
5875 | *per_cpu_ptr(sdd->sd, j) = sd; | |
5876 | ||
5877 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
5878 | GFP_KERNEL, cpu_to_node(j)); | |
5879 | if (!sg) | |
5880 | return -ENOMEM; | |
5881 | ||
30b4e9eb IM |
5882 | sg->next = sg; |
5883 | ||
54ab4ff4 | 5884 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 5885 | |
c1174876 | 5886 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
5887 | GFP_KERNEL, cpu_to_node(j)); |
5888 | if (!sgp) | |
5889 | return -ENOMEM; | |
5890 | ||
5891 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
5892 | } |
5893 | } | |
5894 | ||
5895 | return 0; | |
5896 | } | |
5897 | ||
5898 | static void __sdt_free(const struct cpumask *cpu_map) | |
5899 | { | |
5900 | struct sched_domain_topology_level *tl; | |
5901 | int j; | |
5902 | ||
5903 | for (tl = sched_domain_topology; tl->init; tl++) { | |
5904 | struct sd_data *sdd = &tl->data; | |
5905 | ||
5906 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 5907 | struct sched_domain *sd; |
5908 | ||
5909 | if (sdd->sd) { | |
5910 | sd = *per_cpu_ptr(sdd->sd, j); | |
5911 | if (sd && (sd->flags & SD_OVERLAP)) | |
5912 | free_sched_groups(sd->groups, 0); | |
5913 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
5914 | } | |
5915 | ||
5916 | if (sdd->sg) | |
5917 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
5918 | if (sdd->sgp) | |
5919 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
5920 | } |
5921 | free_percpu(sdd->sd); | |
fb2cf2c6 | 5922 | sdd->sd = NULL; |
54ab4ff4 | 5923 | free_percpu(sdd->sg); |
fb2cf2c6 | 5924 | sdd->sg = NULL; |
9c3f75cb | 5925 | free_percpu(sdd->sgp); |
fb2cf2c6 | 5926 | sdd->sgp = NULL; |
54ab4ff4 PZ |
5927 | } |
5928 | } | |
5929 | ||
2c402dc3 PZ |
5930 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
5931 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 5932 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
5933 | int cpu) |
5934 | { | |
54ab4ff4 | 5935 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 5936 | if (!sd) |
d069b916 | 5937 | return child; |
2c402dc3 | 5938 | |
2c402dc3 | 5939 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
5940 | if (child) { |
5941 | sd->level = child->level + 1; | |
5942 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 5943 | child->parent = sd; |
60495e77 | 5944 | } |
d069b916 | 5945 | sd->child = child; |
a841f8ce | 5946 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
5947 | |
5948 | return sd; | |
5949 | } | |
5950 | ||
2109b99e AH |
5951 | /* |
5952 | * Build sched domains for a given set of cpus and attach the sched domains | |
5953 | * to the individual cpus | |
5954 | */ | |
dce840a0 PZ |
5955 | static int build_sched_domains(const struct cpumask *cpu_map, |
5956 | struct sched_domain_attr *attr) | |
2109b99e AH |
5957 | { |
5958 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 5959 | struct sched_domain *sd; |
2109b99e | 5960 | struct s_data d; |
822ff793 | 5961 | int i, ret = -ENOMEM; |
9c1cfda2 | 5962 | |
2109b99e AH |
5963 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
5964 | if (alloc_state != sa_rootdomain) | |
5965 | goto error; | |
9c1cfda2 | 5966 | |
dce840a0 | 5967 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 5968 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
5969 | struct sched_domain_topology_level *tl; |
5970 | ||
3bd65a80 | 5971 | sd = NULL; |
e3589f6c | 5972 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 5973 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
5974 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
5975 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
5976 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
5977 | break; | |
e3589f6c | 5978 | } |
d274cb30 | 5979 | |
d069b916 PZ |
5980 | while (sd->child) |
5981 | sd = sd->child; | |
5982 | ||
21d42ccf | 5983 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
5984 | } |
5985 | ||
5986 | /* Build the groups for the domains */ | |
5987 | for_each_cpu(i, cpu_map) { | |
5988 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
5989 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
5990 | if (sd->flags & SD_OVERLAP) { |
5991 | if (build_overlap_sched_groups(sd, i)) | |
5992 | goto error; | |
5993 | } else { | |
5994 | if (build_sched_groups(sd, i)) | |
5995 | goto error; | |
5996 | } | |
1cf51902 | 5997 | } |
a06dadbe | 5998 | } |
9c1cfda2 | 5999 | |
1da177e4 | 6000 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6001 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6002 | if (!cpumask_test_cpu(i, cpu_map)) | |
6003 | continue; | |
9c1cfda2 | 6004 | |
dce840a0 PZ |
6005 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6006 | claim_allocations(i, sd); | |
cd4ea6ae | 6007 | init_sched_groups_power(i, sd); |
dce840a0 | 6008 | } |
f712c0c7 | 6009 | } |
9c1cfda2 | 6010 | |
1da177e4 | 6011 | /* Attach the domains */ |
dce840a0 | 6012 | rcu_read_lock(); |
abcd083a | 6013 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6014 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6015 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6016 | } |
dce840a0 | 6017 | rcu_read_unlock(); |
51888ca2 | 6018 | |
822ff793 | 6019 | ret = 0; |
51888ca2 | 6020 | error: |
2109b99e | 6021 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6022 | return ret; |
1da177e4 | 6023 | } |
029190c5 | 6024 | |
acc3f5d7 | 6025 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6026 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6027 | static struct sched_domain_attr *dattr_cur; |
6028 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6029 | |
6030 | /* | |
6031 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6032 | * cpumask) fails, then fallback to a single sched domain, |
6033 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6034 | */ |
4212823f | 6035 | static cpumask_var_t fallback_doms; |
029190c5 | 6036 | |
ee79d1bd HC |
6037 | /* |
6038 | * arch_update_cpu_topology lets virtualized architectures update the | |
6039 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6040 | * or 0 if it stayed the same. | |
6041 | */ | |
6042 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6043 | { |
ee79d1bd | 6044 | return 0; |
22e52b07 HC |
6045 | } |
6046 | ||
acc3f5d7 RR |
6047 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6048 | { | |
6049 | int i; | |
6050 | cpumask_var_t *doms; | |
6051 | ||
6052 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6053 | if (!doms) | |
6054 | return NULL; | |
6055 | for (i = 0; i < ndoms; i++) { | |
6056 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6057 | free_sched_domains(doms, i); | |
6058 | return NULL; | |
6059 | } | |
6060 | } | |
6061 | return doms; | |
6062 | } | |
6063 | ||
6064 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6065 | { | |
6066 | unsigned int i; | |
6067 | for (i = 0; i < ndoms; i++) | |
6068 | free_cpumask_var(doms[i]); | |
6069 | kfree(doms); | |
6070 | } | |
6071 | ||
1a20ff27 | 6072 | /* |
41a2d6cf | 6073 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6074 | * For now this just excludes isolated cpus, but could be used to |
6075 | * exclude other special cases in the future. | |
1a20ff27 | 6076 | */ |
c4a8849a | 6077 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6078 | { |
7378547f MM |
6079 | int err; |
6080 | ||
22e52b07 | 6081 | arch_update_cpu_topology(); |
029190c5 | 6082 | ndoms_cur = 1; |
acc3f5d7 | 6083 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6084 | if (!doms_cur) |
acc3f5d7 RR |
6085 | doms_cur = &fallback_doms; |
6086 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6087 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6088 | register_sched_domain_sysctl(); |
7378547f MM |
6089 | |
6090 | return err; | |
1a20ff27 DG |
6091 | } |
6092 | ||
1a20ff27 DG |
6093 | /* |
6094 | * Detach sched domains from a group of cpus specified in cpu_map | |
6095 | * These cpus will now be attached to the NULL domain | |
6096 | */ | |
96f874e2 | 6097 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6098 | { |
6099 | int i; | |
6100 | ||
dce840a0 | 6101 | rcu_read_lock(); |
abcd083a | 6102 | for_each_cpu(i, cpu_map) |
57d885fe | 6103 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6104 | rcu_read_unlock(); |
1a20ff27 DG |
6105 | } |
6106 | ||
1d3504fc HS |
6107 | /* handle null as "default" */ |
6108 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6109 | struct sched_domain_attr *new, int idx_new) | |
6110 | { | |
6111 | struct sched_domain_attr tmp; | |
6112 | ||
6113 | /* fast path */ | |
6114 | if (!new && !cur) | |
6115 | return 1; | |
6116 | ||
6117 | tmp = SD_ATTR_INIT; | |
6118 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6119 | new ? (new + idx_new) : &tmp, | |
6120 | sizeof(struct sched_domain_attr)); | |
6121 | } | |
6122 | ||
029190c5 PJ |
6123 | /* |
6124 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6125 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6126 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6127 | * It destroys each deleted domain and builds each new domain. | |
6128 | * | |
acc3f5d7 | 6129 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6130 | * The masks don't intersect (don't overlap.) We should setup one |
6131 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6132 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6133 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6134 | * it as it is. | |
6135 | * | |
acc3f5d7 RR |
6136 | * The passed in 'doms_new' should be allocated using |
6137 | * alloc_sched_domains. This routine takes ownership of it and will | |
6138 | * free_sched_domains it when done with it. If the caller failed the | |
6139 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6140 | * and partition_sched_domains() will fallback to the single partition | |
6141 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6142 | * |
96f874e2 | 6143 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6144 | * ndoms_new == 0 is a special case for destroying existing domains, |
6145 | * and it will not create the default domain. | |
dfb512ec | 6146 | * |
029190c5 PJ |
6147 | * Call with hotplug lock held |
6148 | */ | |
acc3f5d7 | 6149 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6150 | struct sched_domain_attr *dattr_new) |
029190c5 | 6151 | { |
dfb512ec | 6152 | int i, j, n; |
d65bd5ec | 6153 | int new_topology; |
029190c5 | 6154 | |
712555ee | 6155 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6156 | |
7378547f MM |
6157 | /* always unregister in case we don't destroy any domains */ |
6158 | unregister_sched_domain_sysctl(); | |
6159 | ||
d65bd5ec HC |
6160 | /* Let architecture update cpu core mappings. */ |
6161 | new_topology = arch_update_cpu_topology(); | |
6162 | ||
dfb512ec | 6163 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6164 | |
6165 | /* Destroy deleted domains */ | |
6166 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6167 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6168 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6169 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6170 | goto match1; |
6171 | } | |
6172 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6173 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6174 | match1: |
6175 | ; | |
6176 | } | |
6177 | ||
e761b772 MK |
6178 | if (doms_new == NULL) { |
6179 | ndoms_cur = 0; | |
acc3f5d7 | 6180 | doms_new = &fallback_doms; |
6ad4c188 | 6181 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6182 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6183 | } |
6184 | ||
029190c5 PJ |
6185 | /* Build new domains */ |
6186 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6187 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6188 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6189 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6190 | goto match2; |
6191 | } | |
6192 | /* no match - add a new doms_new */ | |
dce840a0 | 6193 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6194 | match2: |
6195 | ; | |
6196 | } | |
6197 | ||
6198 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6199 | if (doms_cur != &fallback_doms) |
6200 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6201 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6202 | doms_cur = doms_new; |
1d3504fc | 6203 | dattr_cur = dattr_new; |
029190c5 | 6204 | ndoms_cur = ndoms_new; |
7378547f MM |
6205 | |
6206 | register_sched_domain_sysctl(); | |
a1835615 | 6207 | |
712555ee | 6208 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6209 | } |
6210 | ||
d35be8ba SB |
6211 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6212 | ||
1da177e4 | 6213 | /* |
3a101d05 TH |
6214 | * Update cpusets according to cpu_active mask. If cpusets are |
6215 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6216 | * around partition_sched_domains(). | |
d35be8ba SB |
6217 | * |
6218 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6219 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6220 | */ |
0b2e918a TH |
6221 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6222 | void *hcpu) | |
e761b772 | 6223 | { |
d35be8ba SB |
6224 | switch (action) { |
6225 | case CPU_ONLINE_FROZEN: | |
6226 | case CPU_DOWN_FAILED_FROZEN: | |
6227 | ||
6228 | /* | |
6229 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6230 | * resume sequence. As long as this is not the last online | |
6231 | * operation in the resume sequence, just build a single sched | |
6232 | * domain, ignoring cpusets. | |
6233 | */ | |
6234 | num_cpus_frozen--; | |
6235 | if (likely(num_cpus_frozen)) { | |
6236 | partition_sched_domains(1, NULL, NULL); | |
6237 | break; | |
6238 | } | |
6239 | ||
6240 | /* | |
6241 | * This is the last CPU online operation. So fall through and | |
6242 | * restore the original sched domains by considering the | |
6243 | * cpuset configurations. | |
6244 | */ | |
6245 | ||
e761b772 | 6246 | case CPU_ONLINE: |
6ad4c188 | 6247 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6248 | cpuset_update_active_cpus(true); |
d35be8ba | 6249 | break; |
3a101d05 TH |
6250 | default: |
6251 | return NOTIFY_DONE; | |
6252 | } | |
d35be8ba | 6253 | return NOTIFY_OK; |
3a101d05 | 6254 | } |
e761b772 | 6255 | |
0b2e918a TH |
6256 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6257 | void *hcpu) | |
3a101d05 | 6258 | { |
d35be8ba | 6259 | switch (action) { |
3a101d05 | 6260 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6261 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6262 | break; |
6263 | case CPU_DOWN_PREPARE_FROZEN: | |
6264 | num_cpus_frozen++; | |
6265 | partition_sched_domains(1, NULL, NULL); | |
6266 | break; | |
e761b772 MK |
6267 | default: |
6268 | return NOTIFY_DONE; | |
6269 | } | |
d35be8ba | 6270 | return NOTIFY_OK; |
e761b772 | 6271 | } |
e761b772 | 6272 | |
1da177e4 LT |
6273 | void __init sched_init_smp(void) |
6274 | { | |
dcc30a35 RR |
6275 | cpumask_var_t non_isolated_cpus; |
6276 | ||
6277 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6278 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6279 | |
cb83b629 PZ |
6280 | sched_init_numa(); |
6281 | ||
95402b38 | 6282 | get_online_cpus(); |
712555ee | 6283 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6284 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6285 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6286 | if (cpumask_empty(non_isolated_cpus)) | |
6287 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6288 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6289 | put_online_cpus(); |
e761b772 | 6290 | |
301a5cba | 6291 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6292 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6293 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6294 | |
b328ca18 | 6295 | init_hrtick(); |
5c1e1767 NP |
6296 | |
6297 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6298 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6299 | BUG(); |
19978ca6 | 6300 | sched_init_granularity(); |
dcc30a35 | 6301 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6302 | |
0e3900e6 | 6303 | init_sched_rt_class(); |
1da177e4 LT |
6304 | } |
6305 | #else | |
6306 | void __init sched_init_smp(void) | |
6307 | { | |
19978ca6 | 6308 | sched_init_granularity(); |
1da177e4 LT |
6309 | } |
6310 | #endif /* CONFIG_SMP */ | |
6311 | ||
cd1bb94b AB |
6312 | const_debug unsigned int sysctl_timer_migration = 1; |
6313 | ||
1da177e4 LT |
6314 | int in_sched_functions(unsigned long addr) |
6315 | { | |
1da177e4 LT |
6316 | return in_lock_functions(addr) || |
6317 | (addr >= (unsigned long)__sched_text_start | |
6318 | && addr < (unsigned long)__sched_text_end); | |
6319 | } | |
6320 | ||
029632fb | 6321 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6322 | /* |
6323 | * Default task group. | |
6324 | * Every task in system belongs to this group at bootup. | |
6325 | */ | |
029632fb | 6326 | struct task_group root_task_group; |
35cf4e50 | 6327 | LIST_HEAD(task_groups); |
052f1dc7 | 6328 | #endif |
6f505b16 | 6329 | |
e6252c3e | 6330 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6331 | |
1da177e4 LT |
6332 | void __init sched_init(void) |
6333 | { | |
dd41f596 | 6334 | int i, j; |
434d53b0 MT |
6335 | unsigned long alloc_size = 0, ptr; |
6336 | ||
6337 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6338 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6339 | #endif | |
6340 | #ifdef CONFIG_RT_GROUP_SCHED | |
6341 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6342 | #endif |
df7c8e84 | 6343 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6344 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6345 | #endif |
434d53b0 | 6346 | if (alloc_size) { |
36b7b6d4 | 6347 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6348 | |
6349 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6350 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6351 | ptr += nr_cpu_ids * sizeof(void **); |
6352 | ||
07e06b01 | 6353 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6354 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6355 | |
6d6bc0ad | 6356 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6357 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6358 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6359 | ptr += nr_cpu_ids * sizeof(void **); |
6360 | ||
07e06b01 | 6361 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6362 | ptr += nr_cpu_ids * sizeof(void **); |
6363 | ||
6d6bc0ad | 6364 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6365 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6366 | for_each_possible_cpu(i) { | |
e6252c3e | 6367 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6368 | ptr += cpumask_size(); |
6369 | } | |
6370 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6371 | } |
dd41f596 | 6372 | |
57d885fe GH |
6373 | #ifdef CONFIG_SMP |
6374 | init_defrootdomain(); | |
6375 | #endif | |
6376 | ||
d0b27fa7 PZ |
6377 | init_rt_bandwidth(&def_rt_bandwidth, |
6378 | global_rt_period(), global_rt_runtime()); | |
6379 | ||
6380 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6381 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6382 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6383 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6384 | |
7c941438 | 6385 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6386 | list_add(&root_task_group.list, &task_groups); |
6387 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6388 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6389 | autogroup_init(&init_task); |
54c707e9 | 6390 | |
7c941438 | 6391 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6392 | |
0a945022 | 6393 | for_each_possible_cpu(i) { |
70b97a7f | 6394 | struct rq *rq; |
1da177e4 LT |
6395 | |
6396 | rq = cpu_rq(i); | |
05fa785c | 6397 | raw_spin_lock_init(&rq->lock); |
7897986b | 6398 | rq->nr_running = 0; |
dce48a84 TG |
6399 | rq->calc_load_active = 0; |
6400 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6401 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6402 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 6403 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6404 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6405 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6406 | /* |
07e06b01 | 6407 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6408 | * |
6409 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6410 | * gets 100% of the cpu resources in the system. This overall | |
6411 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6412 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6413 | * based on each entity's (task or task-group's) weight |
6414 | * (se->load.weight). | |
6415 | * | |
07e06b01 | 6416 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6417 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6418 | * then A0's share of the cpu resource is: | |
6419 | * | |
0d905bca | 6420 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6421 | * |
07e06b01 YZ |
6422 | * We achieve this by letting root_task_group's tasks sit |
6423 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6424 | */ |
ab84d31e | 6425 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6426 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6427 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6428 | ||
6429 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6430 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6431 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6432 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6433 | #endif |
1da177e4 | 6434 | |
dd41f596 IM |
6435 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6436 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6437 | |
6438 | rq->last_load_update_tick = jiffies; | |
6439 | ||
1da177e4 | 6440 | #ifdef CONFIG_SMP |
41c7ce9a | 6441 | rq->sd = NULL; |
57d885fe | 6442 | rq->rd = NULL; |
1399fa78 | 6443 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6444 | rq->post_schedule = 0; |
1da177e4 | 6445 | rq->active_balance = 0; |
dd41f596 | 6446 | rq->next_balance = jiffies; |
1da177e4 | 6447 | rq->push_cpu = 0; |
0a2966b4 | 6448 | rq->cpu = i; |
1f11eb6a | 6449 | rq->online = 0; |
eae0c9df MG |
6450 | rq->idle_stamp = 0; |
6451 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
6452 | |
6453 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6454 | ||
dc938520 | 6455 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6456 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6457 | rq->nohz_flags = 0; |
83cd4fe2 | 6458 | #endif |
265f22a9 FW |
6459 | #ifdef CONFIG_NO_HZ_FULL |
6460 | rq->last_sched_tick = 0; | |
6461 | #endif | |
1da177e4 | 6462 | #endif |
8f4d37ec | 6463 | init_rq_hrtick(rq); |
1da177e4 | 6464 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6465 | } |
6466 | ||
2dd73a4f | 6467 | set_load_weight(&init_task); |
b50f60ce | 6468 | |
e107be36 AK |
6469 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6470 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6471 | #endif | |
6472 | ||
b50f60ce | 6473 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 6474 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
6475 | #endif |
6476 | ||
1da177e4 LT |
6477 | /* |
6478 | * The boot idle thread does lazy MMU switching as well: | |
6479 | */ | |
6480 | atomic_inc(&init_mm.mm_count); | |
6481 | enter_lazy_tlb(&init_mm, current); | |
6482 | ||
6483 | /* | |
6484 | * Make us the idle thread. Technically, schedule() should not be | |
6485 | * called from this thread, however somewhere below it might be, | |
6486 | * but because we are the idle thread, we just pick up running again | |
6487 | * when this runqueue becomes "idle". | |
6488 | */ | |
6489 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6490 | |
6491 | calc_load_update = jiffies + LOAD_FREQ; | |
6492 | ||
dd41f596 IM |
6493 | /* |
6494 | * During early bootup we pretend to be a normal task: | |
6495 | */ | |
6496 | current->sched_class = &fair_sched_class; | |
6892b75e | 6497 | |
bf4d83f6 | 6498 | #ifdef CONFIG_SMP |
4cb98839 | 6499 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6500 | /* May be allocated at isolcpus cmdline parse time */ |
6501 | if (cpu_isolated_map == NULL) | |
6502 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6503 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6504 | #endif |
6505 | init_sched_fair_class(); | |
6a7b3dc3 | 6506 | |
6892b75e | 6507 | scheduler_running = 1; |
1da177e4 LT |
6508 | } |
6509 | ||
d902db1e | 6510 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6511 | static inline int preempt_count_equals(int preempt_offset) |
6512 | { | |
234da7bc | 6513 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6514 | |
4ba8216c | 6515 | return (nested == preempt_offset); |
e4aafea2 FW |
6516 | } |
6517 | ||
d894837f | 6518 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6519 | { |
1da177e4 LT |
6520 | static unsigned long prev_jiffy; /* ratelimiting */ |
6521 | ||
b3fbab05 | 6522 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6523 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
6524 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
6525 | return; |
6526 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6527 | return; | |
6528 | prev_jiffy = jiffies; | |
6529 | ||
3df0fc5b PZ |
6530 | printk(KERN_ERR |
6531 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6532 | file, line); | |
6533 | printk(KERN_ERR | |
6534 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6535 | in_atomic(), irqs_disabled(), | |
6536 | current->pid, current->comm); | |
aef745fc IM |
6537 | |
6538 | debug_show_held_locks(current); | |
6539 | if (irqs_disabled()) | |
6540 | print_irqtrace_events(current); | |
6541 | dump_stack(); | |
1da177e4 LT |
6542 | } |
6543 | EXPORT_SYMBOL(__might_sleep); | |
6544 | #endif | |
6545 | ||
6546 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6547 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6548 | { | |
da7a735e PZ |
6549 | const struct sched_class *prev_class = p->sched_class; |
6550 | int old_prio = p->prio; | |
3a5e4dc1 | 6551 | int on_rq; |
3e51f33f | 6552 | |
fd2f4419 | 6553 | on_rq = p->on_rq; |
3a5e4dc1 | 6554 | if (on_rq) |
4ca9b72b | 6555 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
6556 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
6557 | if (on_rq) { | |
4ca9b72b | 6558 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
6559 | resched_task(rq->curr); |
6560 | } | |
da7a735e PZ |
6561 | |
6562 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
6563 | } |
6564 | ||
1da177e4 LT |
6565 | void normalize_rt_tasks(void) |
6566 | { | |
a0f98a1c | 6567 | struct task_struct *g, *p; |
1da177e4 | 6568 | unsigned long flags; |
70b97a7f | 6569 | struct rq *rq; |
1da177e4 | 6570 | |
4cf5d77a | 6571 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 6572 | do_each_thread(g, p) { |
178be793 IM |
6573 | /* |
6574 | * Only normalize user tasks: | |
6575 | */ | |
6576 | if (!p->mm) | |
6577 | continue; | |
6578 | ||
6cfb0d5d | 6579 | p->se.exec_start = 0; |
6cfb0d5d | 6580 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
6581 | p->se.statistics.wait_start = 0; |
6582 | p->se.statistics.sleep_start = 0; | |
6583 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 6584 | #endif |
dd41f596 IM |
6585 | |
6586 | if (!rt_task(p)) { | |
6587 | /* | |
6588 | * Renice negative nice level userspace | |
6589 | * tasks back to 0: | |
6590 | */ | |
6591 | if (TASK_NICE(p) < 0 && p->mm) | |
6592 | set_user_nice(p, 0); | |
1da177e4 | 6593 | continue; |
dd41f596 | 6594 | } |
1da177e4 | 6595 | |
1d615482 | 6596 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 6597 | rq = __task_rq_lock(p); |
1da177e4 | 6598 | |
178be793 | 6599 | normalize_task(rq, p); |
3a5e4dc1 | 6600 | |
b29739f9 | 6601 | __task_rq_unlock(rq); |
1d615482 | 6602 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
6603 | } while_each_thread(g, p); |
6604 | ||
4cf5d77a | 6605 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
6606 | } |
6607 | ||
6608 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6609 | |
67fc4e0c | 6610 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6611 | /* |
67fc4e0c | 6612 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6613 | * |
6614 | * They can only be called when the whole system has been | |
6615 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6616 | * activity can take place. Using them for anything else would | |
6617 | * be a serious bug, and as a result, they aren't even visible | |
6618 | * under any other configuration. | |
6619 | */ | |
6620 | ||
6621 | /** | |
6622 | * curr_task - return the current task for a given cpu. | |
6623 | * @cpu: the processor in question. | |
6624 | * | |
6625 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6626 | */ | |
36c8b586 | 6627 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6628 | { |
6629 | return cpu_curr(cpu); | |
6630 | } | |
6631 | ||
67fc4e0c JW |
6632 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6633 | ||
6634 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
6635 | /** |
6636 | * set_curr_task - set the current task for a given cpu. | |
6637 | * @cpu: the processor in question. | |
6638 | * @p: the task pointer to set. | |
6639 | * | |
6640 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
6641 | * are serviced on a separate stack. It allows the architecture to switch the |
6642 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
6643 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6644 | * and caller must save the original value of the current task (see | |
6645 | * curr_task() above) and restore that value before reenabling interrupts and | |
6646 | * re-starting the system. | |
6647 | * | |
6648 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6649 | */ | |
36c8b586 | 6650 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6651 | { |
6652 | cpu_curr(cpu) = p; | |
6653 | } | |
6654 | ||
6655 | #endif | |
29f59db3 | 6656 | |
7c941438 | 6657 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6658 | /* task_group_lock serializes the addition/removal of task groups */ |
6659 | static DEFINE_SPINLOCK(task_group_lock); | |
6660 | ||
bccbe08a PZ |
6661 | static void free_sched_group(struct task_group *tg) |
6662 | { | |
6663 | free_fair_sched_group(tg); | |
6664 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6665 | autogroup_free(tg); |
bccbe08a PZ |
6666 | kfree(tg); |
6667 | } | |
6668 | ||
6669 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6670 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6671 | { |
6672 | struct task_group *tg; | |
bccbe08a PZ |
6673 | |
6674 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
6675 | if (!tg) | |
6676 | return ERR_PTR(-ENOMEM); | |
6677 | ||
ec7dc8ac | 6678 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6679 | goto err; |
6680 | ||
ec7dc8ac | 6681 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6682 | goto err; |
6683 | ||
ace783b9 LZ |
6684 | return tg; |
6685 | ||
6686 | err: | |
6687 | free_sched_group(tg); | |
6688 | return ERR_PTR(-ENOMEM); | |
6689 | } | |
6690 | ||
6691 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6692 | { | |
6693 | unsigned long flags; | |
6694 | ||
8ed36996 | 6695 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6696 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
6697 | |
6698 | WARN_ON(!parent); /* root should already exist */ | |
6699 | ||
6700 | tg->parent = parent; | |
f473aa5e | 6701 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6702 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6703 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6704 | } |
6705 | ||
9b5b7751 | 6706 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 6707 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6708 | { |
29f59db3 | 6709 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 6710 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6711 | } |
6712 | ||
9b5b7751 | 6713 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6714 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
6715 | { |
6716 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6717 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
6718 | } | |
6719 | ||
6720 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6721 | { |
8ed36996 | 6722 | unsigned long flags; |
9b5b7751 | 6723 | int i; |
29f59db3 | 6724 | |
3d4b47b4 PZ |
6725 | /* end participation in shares distribution */ |
6726 | for_each_possible_cpu(i) | |
bccbe08a | 6727 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
6728 | |
6729 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6730 | list_del_rcu(&tg->list); |
f473aa5e | 6731 | list_del_rcu(&tg->siblings); |
8ed36996 | 6732 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6733 | } |
6734 | ||
9b5b7751 | 6735 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6736 | * The caller of this function should have put the task in its new group |
6737 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6738 | * reflect its new group. | |
9b5b7751 SV |
6739 | */ |
6740 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 6741 | { |
8323f26c | 6742 | struct task_group *tg; |
29f59db3 SV |
6743 | int on_rq, running; |
6744 | unsigned long flags; | |
6745 | struct rq *rq; | |
6746 | ||
6747 | rq = task_rq_lock(tsk, &flags); | |
6748 | ||
051a1d1a | 6749 | running = task_current(rq, tsk); |
fd2f4419 | 6750 | on_rq = tsk->on_rq; |
29f59db3 | 6751 | |
0e1f3483 | 6752 | if (on_rq) |
29f59db3 | 6753 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
6754 | if (unlikely(running)) |
6755 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 6756 | |
8323f26c PZ |
6757 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
6758 | lockdep_is_held(&tsk->sighand->siglock)), | |
6759 | struct task_group, css); | |
6760 | tg = autogroup_task_group(tsk, tg); | |
6761 | tsk->sched_task_group = tg; | |
6762 | ||
810b3817 | 6763 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
6764 | if (tsk->sched_class->task_move_group) |
6765 | tsk->sched_class->task_move_group(tsk, on_rq); | |
6766 | else | |
810b3817 | 6767 | #endif |
b2b5ce02 | 6768 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 6769 | |
0e1f3483 HS |
6770 | if (unlikely(running)) |
6771 | tsk->sched_class->set_curr_task(rq); | |
6772 | if (on_rq) | |
371fd7e7 | 6773 | enqueue_task(rq, tsk, 0); |
29f59db3 | 6774 | |
0122ec5b | 6775 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 6776 | } |
7c941438 | 6777 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 6778 | |
a790de99 | 6779 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
6780 | static unsigned long to_ratio(u64 period, u64 runtime) |
6781 | { | |
6782 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 6783 | return 1ULL << 20; |
9f0c1e56 | 6784 | |
9a7e0b18 | 6785 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 6786 | } |
a790de99 PT |
6787 | #endif |
6788 | ||
6789 | #ifdef CONFIG_RT_GROUP_SCHED | |
6790 | /* | |
6791 | * Ensure that the real time constraints are schedulable. | |
6792 | */ | |
6793 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 6794 | |
9a7e0b18 PZ |
6795 | /* Must be called with tasklist_lock held */ |
6796 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 6797 | { |
9a7e0b18 | 6798 | struct task_struct *g, *p; |
b40b2e8e | 6799 | |
9a7e0b18 | 6800 | do_each_thread(g, p) { |
029632fb | 6801 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
6802 | return 1; |
6803 | } while_each_thread(g, p); | |
b40b2e8e | 6804 | |
9a7e0b18 PZ |
6805 | return 0; |
6806 | } | |
b40b2e8e | 6807 | |
9a7e0b18 PZ |
6808 | struct rt_schedulable_data { |
6809 | struct task_group *tg; | |
6810 | u64 rt_period; | |
6811 | u64 rt_runtime; | |
6812 | }; | |
b40b2e8e | 6813 | |
a790de99 | 6814 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
6815 | { |
6816 | struct rt_schedulable_data *d = data; | |
6817 | struct task_group *child; | |
6818 | unsigned long total, sum = 0; | |
6819 | u64 period, runtime; | |
b40b2e8e | 6820 | |
9a7e0b18 PZ |
6821 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
6822 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 6823 | |
9a7e0b18 PZ |
6824 | if (tg == d->tg) { |
6825 | period = d->rt_period; | |
6826 | runtime = d->rt_runtime; | |
b40b2e8e | 6827 | } |
b40b2e8e | 6828 | |
4653f803 PZ |
6829 | /* |
6830 | * Cannot have more runtime than the period. | |
6831 | */ | |
6832 | if (runtime > period && runtime != RUNTIME_INF) | |
6833 | return -EINVAL; | |
6f505b16 | 6834 | |
4653f803 PZ |
6835 | /* |
6836 | * Ensure we don't starve existing RT tasks. | |
6837 | */ | |
9a7e0b18 PZ |
6838 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
6839 | return -EBUSY; | |
6f505b16 | 6840 | |
9a7e0b18 | 6841 | total = to_ratio(period, runtime); |
6f505b16 | 6842 | |
4653f803 PZ |
6843 | /* |
6844 | * Nobody can have more than the global setting allows. | |
6845 | */ | |
6846 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
6847 | return -EINVAL; | |
6f505b16 | 6848 | |
4653f803 PZ |
6849 | /* |
6850 | * The sum of our children's runtime should not exceed our own. | |
6851 | */ | |
9a7e0b18 PZ |
6852 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
6853 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
6854 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 6855 | |
9a7e0b18 PZ |
6856 | if (child == d->tg) { |
6857 | period = d->rt_period; | |
6858 | runtime = d->rt_runtime; | |
6859 | } | |
6f505b16 | 6860 | |
9a7e0b18 | 6861 | sum += to_ratio(period, runtime); |
9f0c1e56 | 6862 | } |
6f505b16 | 6863 | |
9a7e0b18 PZ |
6864 | if (sum > total) |
6865 | return -EINVAL; | |
6866 | ||
6867 | return 0; | |
6f505b16 PZ |
6868 | } |
6869 | ||
9a7e0b18 | 6870 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 6871 | { |
8277434e PT |
6872 | int ret; |
6873 | ||
9a7e0b18 PZ |
6874 | struct rt_schedulable_data data = { |
6875 | .tg = tg, | |
6876 | .rt_period = period, | |
6877 | .rt_runtime = runtime, | |
6878 | }; | |
6879 | ||
8277434e PT |
6880 | rcu_read_lock(); |
6881 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
6882 | rcu_read_unlock(); | |
6883 | ||
6884 | return ret; | |
521f1a24 DG |
6885 | } |
6886 | ||
ab84d31e | 6887 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 6888 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 6889 | { |
ac086bc2 | 6890 | int i, err = 0; |
9f0c1e56 | 6891 | |
9f0c1e56 | 6892 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 6893 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
6894 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
6895 | if (err) | |
9f0c1e56 | 6896 | goto unlock; |
ac086bc2 | 6897 | |
0986b11b | 6898 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
6899 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
6900 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
6901 | |
6902 | for_each_possible_cpu(i) { | |
6903 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
6904 | ||
0986b11b | 6905 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6906 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 6907 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6908 | } |
0986b11b | 6909 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 6910 | unlock: |
521f1a24 | 6911 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
6912 | mutex_unlock(&rt_constraints_mutex); |
6913 | ||
6914 | return err; | |
6f505b16 PZ |
6915 | } |
6916 | ||
25cc7da7 | 6917 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
6918 | { |
6919 | u64 rt_runtime, rt_period; | |
6920 | ||
6921 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
6922 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
6923 | if (rt_runtime_us < 0) | |
6924 | rt_runtime = RUNTIME_INF; | |
6925 | ||
ab84d31e | 6926 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
6927 | } |
6928 | ||
25cc7da7 | 6929 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
6930 | { |
6931 | u64 rt_runtime_us; | |
6932 | ||
d0b27fa7 | 6933 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
6934 | return -1; |
6935 | ||
d0b27fa7 | 6936 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
6937 | do_div(rt_runtime_us, NSEC_PER_USEC); |
6938 | return rt_runtime_us; | |
6939 | } | |
d0b27fa7 | 6940 | |
25cc7da7 | 6941 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
6942 | { |
6943 | u64 rt_runtime, rt_period; | |
6944 | ||
6945 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
6946 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
6947 | ||
619b0488 R |
6948 | if (rt_period == 0) |
6949 | return -EINVAL; | |
6950 | ||
ab84d31e | 6951 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
6952 | } |
6953 | ||
25cc7da7 | 6954 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
6955 | { |
6956 | u64 rt_period_us; | |
6957 | ||
6958 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
6959 | do_div(rt_period_us, NSEC_PER_USEC); | |
6960 | return rt_period_us; | |
6961 | } | |
6962 | ||
6963 | static int sched_rt_global_constraints(void) | |
6964 | { | |
4653f803 | 6965 | u64 runtime, period; |
d0b27fa7 PZ |
6966 | int ret = 0; |
6967 | ||
ec5d4989 HS |
6968 | if (sysctl_sched_rt_period <= 0) |
6969 | return -EINVAL; | |
6970 | ||
4653f803 PZ |
6971 | runtime = global_rt_runtime(); |
6972 | period = global_rt_period(); | |
6973 | ||
6974 | /* | |
6975 | * Sanity check on the sysctl variables. | |
6976 | */ | |
6977 | if (runtime > period && runtime != RUNTIME_INF) | |
6978 | return -EINVAL; | |
10b612f4 | 6979 | |
d0b27fa7 | 6980 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 6981 | read_lock(&tasklist_lock); |
4653f803 | 6982 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 6983 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
6984 | mutex_unlock(&rt_constraints_mutex); |
6985 | ||
6986 | return ret; | |
6987 | } | |
54e99124 | 6988 | |
25cc7da7 | 6989 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
6990 | { |
6991 | /* Don't accept realtime tasks when there is no way for them to run */ | |
6992 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
6993 | return 0; | |
6994 | ||
6995 | return 1; | |
6996 | } | |
6997 | ||
6d6bc0ad | 6998 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
6999 | static int sched_rt_global_constraints(void) |
7000 | { | |
ac086bc2 PZ |
7001 | unsigned long flags; |
7002 | int i; | |
7003 | ||
ec5d4989 HS |
7004 | if (sysctl_sched_rt_period <= 0) |
7005 | return -EINVAL; | |
7006 | ||
60aa605d PZ |
7007 | /* |
7008 | * There's always some RT tasks in the root group | |
7009 | * -- migration, kstopmachine etc.. | |
7010 | */ | |
7011 | if (sysctl_sched_rt_runtime == 0) | |
7012 | return -EBUSY; | |
7013 | ||
0986b11b | 7014 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7015 | for_each_possible_cpu(i) { |
7016 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7017 | ||
0986b11b | 7018 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7019 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7020 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7021 | } |
0986b11b | 7022 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7023 | |
d0b27fa7 PZ |
7024 | return 0; |
7025 | } | |
6d6bc0ad | 7026 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7027 | |
ce0dbbbb CW |
7028 | int sched_rr_handler(struct ctl_table *table, int write, |
7029 | void __user *buffer, size_t *lenp, | |
7030 | loff_t *ppos) | |
7031 | { | |
7032 | int ret; | |
7033 | static DEFINE_MUTEX(mutex); | |
7034 | ||
7035 | mutex_lock(&mutex); | |
7036 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7037 | /* make sure that internally we keep jiffies */ | |
7038 | /* also, writing zero resets timeslice to default */ | |
7039 | if (!ret && write) { | |
7040 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7041 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7042 | } | |
7043 | mutex_unlock(&mutex); | |
7044 | return ret; | |
7045 | } | |
7046 | ||
d0b27fa7 | 7047 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7048 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7049 | loff_t *ppos) |
7050 | { | |
7051 | int ret; | |
7052 | int old_period, old_runtime; | |
7053 | static DEFINE_MUTEX(mutex); | |
7054 | ||
7055 | mutex_lock(&mutex); | |
7056 | old_period = sysctl_sched_rt_period; | |
7057 | old_runtime = sysctl_sched_rt_runtime; | |
7058 | ||
8d65af78 | 7059 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7060 | |
7061 | if (!ret && write) { | |
7062 | ret = sched_rt_global_constraints(); | |
7063 | if (ret) { | |
7064 | sysctl_sched_rt_period = old_period; | |
7065 | sysctl_sched_rt_runtime = old_runtime; | |
7066 | } else { | |
7067 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7068 | def_rt_bandwidth.rt_period = | |
7069 | ns_to_ktime(global_rt_period()); | |
7070 | } | |
7071 | } | |
7072 | mutex_unlock(&mutex); | |
7073 | ||
7074 | return ret; | |
7075 | } | |
68318b8e | 7076 | |
052f1dc7 | 7077 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7078 | |
7079 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7080 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7081 | { |
2b01dfe3 PM |
7082 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7083 | struct task_group, css); | |
68318b8e SV |
7084 | } |
7085 | ||
92fb9748 | 7086 | static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp) |
68318b8e | 7087 | { |
ec7dc8ac | 7088 | struct task_group *tg, *parent; |
68318b8e | 7089 | |
2b01dfe3 | 7090 | if (!cgrp->parent) { |
68318b8e | 7091 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7092 | return &root_task_group.css; |
68318b8e SV |
7093 | } |
7094 | ||
ec7dc8ac DG |
7095 | parent = cgroup_tg(cgrp->parent); |
7096 | tg = sched_create_group(parent); | |
68318b8e SV |
7097 | if (IS_ERR(tg)) |
7098 | return ERR_PTR(-ENOMEM); | |
7099 | ||
68318b8e SV |
7100 | return &tg->css; |
7101 | } | |
7102 | ||
ace783b9 LZ |
7103 | static int cpu_cgroup_css_online(struct cgroup *cgrp) |
7104 | { | |
7105 | struct task_group *tg = cgroup_tg(cgrp); | |
7106 | struct task_group *parent; | |
7107 | ||
7108 | if (!cgrp->parent) | |
7109 | return 0; | |
7110 | ||
7111 | parent = cgroup_tg(cgrp->parent); | |
7112 | sched_online_group(tg, parent); | |
7113 | return 0; | |
7114 | } | |
7115 | ||
92fb9748 | 7116 | static void cpu_cgroup_css_free(struct cgroup *cgrp) |
68318b8e | 7117 | { |
2b01dfe3 | 7118 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7119 | |
7120 | sched_destroy_group(tg); | |
7121 | } | |
7122 | ||
ace783b9 LZ |
7123 | static void cpu_cgroup_css_offline(struct cgroup *cgrp) |
7124 | { | |
7125 | struct task_group *tg = cgroup_tg(cgrp); | |
7126 | ||
7127 | sched_offline_group(tg); | |
7128 | } | |
7129 | ||
761b3ef5 | 7130 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 7131 | struct cgroup_taskset *tset) |
68318b8e | 7132 | { |
bb9d97b6 TH |
7133 | struct task_struct *task; |
7134 | ||
7135 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7136 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7137 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7138 | return -EINVAL; | |
b68aa230 | 7139 | #else |
bb9d97b6 TH |
7140 | /* We don't support RT-tasks being in separate groups */ |
7141 | if (task->sched_class != &fair_sched_class) | |
7142 | return -EINVAL; | |
b68aa230 | 7143 | #endif |
bb9d97b6 | 7144 | } |
be367d09 BB |
7145 | return 0; |
7146 | } | |
68318b8e | 7147 | |
761b3ef5 | 7148 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 7149 | struct cgroup_taskset *tset) |
68318b8e | 7150 | { |
bb9d97b6 TH |
7151 | struct task_struct *task; |
7152 | ||
7153 | cgroup_taskset_for_each(task, cgrp, tset) | |
7154 | sched_move_task(task); | |
68318b8e SV |
7155 | } |
7156 | ||
068c5cc5 | 7157 | static void |
761b3ef5 LZ |
7158 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
7159 | struct task_struct *task) | |
068c5cc5 PZ |
7160 | { |
7161 | /* | |
7162 | * cgroup_exit() is called in the copy_process() failure path. | |
7163 | * Ignore this case since the task hasn't ran yet, this avoids | |
7164 | * trying to poke a half freed task state from generic code. | |
7165 | */ | |
7166 | if (!(task->flags & PF_EXITING)) | |
7167 | return; | |
7168 | ||
7169 | sched_move_task(task); | |
7170 | } | |
7171 | ||
052f1dc7 | 7172 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7173 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7174 | u64 shareval) |
68318b8e | 7175 | { |
c8b28116 | 7176 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7177 | } |
7178 | ||
f4c753b7 | 7179 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7180 | { |
2b01dfe3 | 7181 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7182 | |
c8b28116 | 7183 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7184 | } |
ab84d31e PT |
7185 | |
7186 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7187 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7188 | ||
ab84d31e PT |
7189 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7190 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7191 | ||
a790de99 PT |
7192 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7193 | ||
ab84d31e PT |
7194 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7195 | { | |
56f570e5 | 7196 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7197 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7198 | |
7199 | if (tg == &root_task_group) | |
7200 | return -EINVAL; | |
7201 | ||
7202 | /* | |
7203 | * Ensure we have at some amount of bandwidth every period. This is | |
7204 | * to prevent reaching a state of large arrears when throttled via | |
7205 | * entity_tick() resulting in prolonged exit starvation. | |
7206 | */ | |
7207 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7208 | return -EINVAL; | |
7209 | ||
7210 | /* | |
7211 | * Likewise, bound things on the otherside by preventing insane quota | |
7212 | * periods. This also allows us to normalize in computing quota | |
7213 | * feasibility. | |
7214 | */ | |
7215 | if (period > max_cfs_quota_period) | |
7216 | return -EINVAL; | |
7217 | ||
a790de99 PT |
7218 | mutex_lock(&cfs_constraints_mutex); |
7219 | ret = __cfs_schedulable(tg, period, quota); | |
7220 | if (ret) | |
7221 | goto out_unlock; | |
7222 | ||
58088ad0 | 7223 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7224 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7225 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7226 | raw_spin_lock_irq(&cfs_b->lock); |
7227 | cfs_b->period = ns_to_ktime(period); | |
7228 | cfs_b->quota = quota; | |
58088ad0 | 7229 | |
a9cf55b2 | 7230 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7231 | /* restart the period timer (if active) to handle new period expiry */ |
7232 | if (runtime_enabled && cfs_b->timer_active) { | |
7233 | /* force a reprogram */ | |
7234 | cfs_b->timer_active = 0; | |
7235 | __start_cfs_bandwidth(cfs_b); | |
7236 | } | |
ab84d31e PT |
7237 | raw_spin_unlock_irq(&cfs_b->lock); |
7238 | ||
7239 | for_each_possible_cpu(i) { | |
7240 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7241 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7242 | |
7243 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7244 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7245 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7246 | |
029632fb | 7247 | if (cfs_rq->throttled) |
671fd9da | 7248 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7249 | raw_spin_unlock_irq(&rq->lock); |
7250 | } | |
a790de99 PT |
7251 | out_unlock: |
7252 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7253 | |
a790de99 | 7254 | return ret; |
ab84d31e PT |
7255 | } |
7256 | ||
7257 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7258 | { | |
7259 | u64 quota, period; | |
7260 | ||
029632fb | 7261 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7262 | if (cfs_quota_us < 0) |
7263 | quota = RUNTIME_INF; | |
7264 | else | |
7265 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7266 | ||
7267 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7268 | } | |
7269 | ||
7270 | long tg_get_cfs_quota(struct task_group *tg) | |
7271 | { | |
7272 | u64 quota_us; | |
7273 | ||
029632fb | 7274 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7275 | return -1; |
7276 | ||
029632fb | 7277 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7278 | do_div(quota_us, NSEC_PER_USEC); |
7279 | ||
7280 | return quota_us; | |
7281 | } | |
7282 | ||
7283 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7284 | { | |
7285 | u64 quota, period; | |
7286 | ||
7287 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7288 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7289 | |
ab84d31e PT |
7290 | return tg_set_cfs_bandwidth(tg, period, quota); |
7291 | } | |
7292 | ||
7293 | long tg_get_cfs_period(struct task_group *tg) | |
7294 | { | |
7295 | u64 cfs_period_us; | |
7296 | ||
029632fb | 7297 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7298 | do_div(cfs_period_us, NSEC_PER_USEC); |
7299 | ||
7300 | return cfs_period_us; | |
7301 | } | |
7302 | ||
7303 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7304 | { | |
7305 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7306 | } | |
7307 | ||
7308 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7309 | s64 cfs_quota_us) | |
7310 | { | |
7311 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7312 | } | |
7313 | ||
7314 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7315 | { | |
7316 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7317 | } | |
7318 | ||
7319 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7320 | u64 cfs_period_us) | |
7321 | { | |
7322 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7323 | } | |
7324 | ||
a790de99 PT |
7325 | struct cfs_schedulable_data { |
7326 | struct task_group *tg; | |
7327 | u64 period, quota; | |
7328 | }; | |
7329 | ||
7330 | /* | |
7331 | * normalize group quota/period to be quota/max_period | |
7332 | * note: units are usecs | |
7333 | */ | |
7334 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7335 | struct cfs_schedulable_data *d) | |
7336 | { | |
7337 | u64 quota, period; | |
7338 | ||
7339 | if (tg == d->tg) { | |
7340 | period = d->period; | |
7341 | quota = d->quota; | |
7342 | } else { | |
7343 | period = tg_get_cfs_period(tg); | |
7344 | quota = tg_get_cfs_quota(tg); | |
7345 | } | |
7346 | ||
7347 | /* note: these should typically be equivalent */ | |
7348 | if (quota == RUNTIME_INF || quota == -1) | |
7349 | return RUNTIME_INF; | |
7350 | ||
7351 | return to_ratio(period, quota); | |
7352 | } | |
7353 | ||
7354 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7355 | { | |
7356 | struct cfs_schedulable_data *d = data; | |
029632fb | 7357 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7358 | s64 quota = 0, parent_quota = -1; |
7359 | ||
7360 | if (!tg->parent) { | |
7361 | quota = RUNTIME_INF; | |
7362 | } else { | |
029632fb | 7363 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7364 | |
7365 | quota = normalize_cfs_quota(tg, d); | |
7366 | parent_quota = parent_b->hierarchal_quota; | |
7367 | ||
7368 | /* | |
7369 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7370 | * limit is set | |
7371 | */ | |
7372 | if (quota == RUNTIME_INF) | |
7373 | quota = parent_quota; | |
7374 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7375 | return -EINVAL; | |
7376 | } | |
7377 | cfs_b->hierarchal_quota = quota; | |
7378 | ||
7379 | return 0; | |
7380 | } | |
7381 | ||
7382 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7383 | { | |
8277434e | 7384 | int ret; |
a790de99 PT |
7385 | struct cfs_schedulable_data data = { |
7386 | .tg = tg, | |
7387 | .period = period, | |
7388 | .quota = quota, | |
7389 | }; | |
7390 | ||
7391 | if (quota != RUNTIME_INF) { | |
7392 | do_div(data.period, NSEC_PER_USEC); | |
7393 | do_div(data.quota, NSEC_PER_USEC); | |
7394 | } | |
7395 | ||
8277434e PT |
7396 | rcu_read_lock(); |
7397 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7398 | rcu_read_unlock(); | |
7399 | ||
7400 | return ret; | |
a790de99 | 7401 | } |
e8da1b18 NR |
7402 | |
7403 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7404 | struct cgroup_map_cb *cb) | |
7405 | { | |
7406 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 7407 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7408 | |
7409 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7410 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7411 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7412 | ||
7413 | return 0; | |
7414 | } | |
ab84d31e | 7415 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7416 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7417 | |
052f1dc7 | 7418 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 7419 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 7420 | s64 val) |
6f505b16 | 7421 | { |
06ecb27c | 7422 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
7423 | } |
7424 | ||
06ecb27c | 7425 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 7426 | { |
06ecb27c | 7427 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 7428 | } |
d0b27fa7 PZ |
7429 | |
7430 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
7431 | u64 rt_period_us) | |
7432 | { | |
7433 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
7434 | } | |
7435 | ||
7436 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
7437 | { | |
7438 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
7439 | } | |
6d6bc0ad | 7440 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7441 | |
fe5c7cc2 | 7442 | static struct cftype cpu_files[] = { |
052f1dc7 | 7443 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7444 | { |
7445 | .name = "shares", | |
f4c753b7 PM |
7446 | .read_u64 = cpu_shares_read_u64, |
7447 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7448 | }, |
052f1dc7 | 7449 | #endif |
ab84d31e PT |
7450 | #ifdef CONFIG_CFS_BANDWIDTH |
7451 | { | |
7452 | .name = "cfs_quota_us", | |
7453 | .read_s64 = cpu_cfs_quota_read_s64, | |
7454 | .write_s64 = cpu_cfs_quota_write_s64, | |
7455 | }, | |
7456 | { | |
7457 | .name = "cfs_period_us", | |
7458 | .read_u64 = cpu_cfs_period_read_u64, | |
7459 | .write_u64 = cpu_cfs_period_write_u64, | |
7460 | }, | |
e8da1b18 NR |
7461 | { |
7462 | .name = "stat", | |
7463 | .read_map = cpu_stats_show, | |
7464 | }, | |
ab84d31e | 7465 | #endif |
052f1dc7 | 7466 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7467 | { |
9f0c1e56 | 7468 | .name = "rt_runtime_us", |
06ecb27c PM |
7469 | .read_s64 = cpu_rt_runtime_read, |
7470 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7471 | }, |
d0b27fa7 PZ |
7472 | { |
7473 | .name = "rt_period_us", | |
f4c753b7 PM |
7474 | .read_u64 = cpu_rt_period_read_uint, |
7475 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7476 | }, |
052f1dc7 | 7477 | #endif |
4baf6e33 | 7478 | { } /* terminate */ |
68318b8e SV |
7479 | }; |
7480 | ||
68318b8e | 7481 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 7482 | .name = "cpu", |
92fb9748 TH |
7483 | .css_alloc = cpu_cgroup_css_alloc, |
7484 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
7485 | .css_online = cpu_cgroup_css_online, |
7486 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
7487 | .can_attach = cpu_cgroup_can_attach, |
7488 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7489 | .exit = cpu_cgroup_exit, |
38605cae | 7490 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 7491 | .base_cftypes = cpu_files, |
68318b8e SV |
7492 | .early_init = 1, |
7493 | }; | |
7494 | ||
052f1dc7 | 7495 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7496 | |
b637a328 PM |
7497 | void dump_cpu_task(int cpu) |
7498 | { | |
7499 | pr_info("Task dump for CPU %d:\n", cpu); | |
7500 | sched_show_task(cpu_curr(cpu)); | |
7501 | } |