]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
1da177e4 | 76 | |
96f951ed | 77 | #include <asm/switch_to.h> |
5517d86b | 78 | #include <asm/tlb.h> |
838225b4 | 79 | #include <asm/irq_regs.h> |
db7e527d | 80 | #include <asm/mutex.h> |
e6e6685a GC |
81 | #ifdef CONFIG_PARAVIRT |
82 | #include <asm/paravirt.h> | |
83 | #endif | |
1da177e4 | 84 | |
029632fb | 85 | #include "sched.h" |
ea138446 | 86 | #include "../workqueue_internal.h" |
29d5e047 | 87 | #include "../smpboot.h" |
6e0534f2 | 88 | |
a8d154b0 | 89 | #define CREATE_TRACE_POINTS |
ad8d75ff | 90 | #include <trace/events/sched.h> |
a8d154b0 | 91 | |
029632fb | 92 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 93 | { |
58088ad0 PT |
94 | unsigned long delta; |
95 | ktime_t soft, hard, now; | |
d0b27fa7 | 96 | |
58088ad0 PT |
97 | for (;;) { |
98 | if (hrtimer_active(period_timer)) | |
99 | break; | |
100 | ||
101 | now = hrtimer_cb_get_time(period_timer); | |
102 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 103 | |
58088ad0 PT |
104 | soft = hrtimer_get_softexpires(period_timer); |
105 | hard = hrtimer_get_expires(period_timer); | |
106 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
107 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
108 | HRTIMER_MODE_ABS_PINNED, 0); | |
109 | } | |
110 | } | |
111 | ||
029632fb PZ |
112 | DEFINE_MUTEX(sched_domains_mutex); |
113 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 114 | |
fe44d621 | 115 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 116 | |
029632fb | 117 | void update_rq_clock(struct rq *rq) |
3e51f33f | 118 | { |
fe44d621 | 119 | s64 delta; |
305e6835 | 120 | |
61eadef6 | 121 | if (rq->skip_clock_update > 0) |
f26f9aff | 122 | return; |
aa483808 | 123 | |
fe44d621 PZ |
124 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
125 | rq->clock += delta; | |
126 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
127 | } |
128 | ||
bf5c91ba IM |
129 | /* |
130 | * Debugging: various feature bits | |
131 | */ | |
f00b45c1 | 132 | |
f00b45c1 PZ |
133 | #define SCHED_FEAT(name, enabled) \ |
134 | (1UL << __SCHED_FEAT_##name) * enabled | | |
135 | ||
bf5c91ba | 136 | const_debug unsigned int sysctl_sched_features = |
391e43da | 137 | #include "features.h" |
f00b45c1 PZ |
138 | 0; |
139 | ||
140 | #undef SCHED_FEAT | |
141 | ||
142 | #ifdef CONFIG_SCHED_DEBUG | |
143 | #define SCHED_FEAT(name, enabled) \ | |
144 | #name , | |
145 | ||
1292531f | 146 | static const char * const sched_feat_names[] = { |
391e43da | 147 | #include "features.h" |
f00b45c1 PZ |
148 | }; |
149 | ||
150 | #undef SCHED_FEAT | |
151 | ||
34f3a814 | 152 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 153 | { |
f00b45c1 PZ |
154 | int i; |
155 | ||
f8b6d1cc | 156 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
157 | if (!(sysctl_sched_features & (1UL << i))) |
158 | seq_puts(m, "NO_"); | |
159 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 160 | } |
34f3a814 | 161 | seq_puts(m, "\n"); |
f00b45c1 | 162 | |
34f3a814 | 163 | return 0; |
f00b45c1 PZ |
164 | } |
165 | ||
f8b6d1cc PZ |
166 | #ifdef HAVE_JUMP_LABEL |
167 | ||
c5905afb IM |
168 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
169 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
170 | |
171 | #define SCHED_FEAT(name, enabled) \ | |
172 | jump_label_key__##enabled , | |
173 | ||
c5905afb | 174 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
175 | #include "features.h" |
176 | }; | |
177 | ||
178 | #undef SCHED_FEAT | |
179 | ||
180 | static void sched_feat_disable(int i) | |
181 | { | |
c5905afb IM |
182 | if (static_key_enabled(&sched_feat_keys[i])) |
183 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
184 | } |
185 | ||
186 | static void sched_feat_enable(int i) | |
187 | { | |
c5905afb IM |
188 | if (!static_key_enabled(&sched_feat_keys[i])) |
189 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
190 | } |
191 | #else | |
192 | static void sched_feat_disable(int i) { }; | |
193 | static void sched_feat_enable(int i) { }; | |
194 | #endif /* HAVE_JUMP_LABEL */ | |
195 | ||
1a687c2e | 196 | static int sched_feat_set(char *cmp) |
f00b45c1 | 197 | { |
f00b45c1 | 198 | int i; |
1a687c2e | 199 | int neg = 0; |
f00b45c1 | 200 | |
524429c3 | 201 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
202 | neg = 1; |
203 | cmp += 3; | |
204 | } | |
205 | ||
f8b6d1cc | 206 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 207 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 208 | if (neg) { |
f00b45c1 | 209 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
210 | sched_feat_disable(i); |
211 | } else { | |
f00b45c1 | 212 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
213 | sched_feat_enable(i); |
214 | } | |
f00b45c1 PZ |
215 | break; |
216 | } | |
217 | } | |
218 | ||
1a687c2e MG |
219 | return i; |
220 | } | |
221 | ||
222 | static ssize_t | |
223 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
224 | size_t cnt, loff_t *ppos) | |
225 | { | |
226 | char buf[64]; | |
227 | char *cmp; | |
228 | int i; | |
229 | ||
230 | if (cnt > 63) | |
231 | cnt = 63; | |
232 | ||
233 | if (copy_from_user(&buf, ubuf, cnt)) | |
234 | return -EFAULT; | |
235 | ||
236 | buf[cnt] = 0; | |
237 | cmp = strstrip(buf); | |
238 | ||
239 | i = sched_feat_set(cmp); | |
f8b6d1cc | 240 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
241 | return -EINVAL; |
242 | ||
42994724 | 243 | *ppos += cnt; |
f00b45c1 PZ |
244 | |
245 | return cnt; | |
246 | } | |
247 | ||
34f3a814 LZ |
248 | static int sched_feat_open(struct inode *inode, struct file *filp) |
249 | { | |
250 | return single_open(filp, sched_feat_show, NULL); | |
251 | } | |
252 | ||
828c0950 | 253 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
254 | .open = sched_feat_open, |
255 | .write = sched_feat_write, | |
256 | .read = seq_read, | |
257 | .llseek = seq_lseek, | |
258 | .release = single_release, | |
f00b45c1 PZ |
259 | }; |
260 | ||
261 | static __init int sched_init_debug(void) | |
262 | { | |
f00b45c1 PZ |
263 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
264 | &sched_feat_fops); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | late_initcall(sched_init_debug); | |
f8b6d1cc | 269 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 270 | |
b82d9fdd PZ |
271 | /* |
272 | * Number of tasks to iterate in a single balance run. | |
273 | * Limited because this is done with IRQs disabled. | |
274 | */ | |
275 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
276 | ||
e9e9250b PZ |
277 | /* |
278 | * period over which we average the RT time consumption, measured | |
279 | * in ms. | |
280 | * | |
281 | * default: 1s | |
282 | */ | |
283 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
284 | ||
fa85ae24 | 285 | /* |
9f0c1e56 | 286 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
287 | * default: 1s |
288 | */ | |
9f0c1e56 | 289 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 290 | |
029632fb | 291 | __read_mostly int scheduler_running; |
6892b75e | 292 | |
9f0c1e56 PZ |
293 | /* |
294 | * part of the period that we allow rt tasks to run in us. | |
295 | * default: 0.95s | |
296 | */ | |
297 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 298 | |
fa85ae24 | 299 | |
1da177e4 | 300 | |
0970d299 | 301 | /* |
0122ec5b | 302 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 303 | */ |
70b97a7f | 304 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
305 | __acquires(rq->lock) |
306 | { | |
0970d299 PZ |
307 | struct rq *rq; |
308 | ||
0122ec5b PZ |
309 | lockdep_assert_held(&p->pi_lock); |
310 | ||
3a5c359a | 311 | for (;;) { |
0970d299 | 312 | rq = task_rq(p); |
05fa785c | 313 | raw_spin_lock(&rq->lock); |
65cc8e48 | 314 | if (likely(rq == task_rq(p))) |
3a5c359a | 315 | return rq; |
05fa785c | 316 | raw_spin_unlock(&rq->lock); |
b29739f9 | 317 | } |
b29739f9 IM |
318 | } |
319 | ||
1da177e4 | 320 | /* |
0122ec5b | 321 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 322 | */ |
70b97a7f | 323 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 324 | __acquires(p->pi_lock) |
1da177e4 LT |
325 | __acquires(rq->lock) |
326 | { | |
70b97a7f | 327 | struct rq *rq; |
1da177e4 | 328 | |
3a5c359a | 329 | for (;;) { |
0122ec5b | 330 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 331 | rq = task_rq(p); |
05fa785c | 332 | raw_spin_lock(&rq->lock); |
65cc8e48 | 333 | if (likely(rq == task_rq(p))) |
3a5c359a | 334 | return rq; |
0122ec5b PZ |
335 | raw_spin_unlock(&rq->lock); |
336 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 337 | } |
1da177e4 LT |
338 | } |
339 | ||
a9957449 | 340 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
341 | __releases(rq->lock) |
342 | { | |
05fa785c | 343 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
344 | } |
345 | ||
0122ec5b PZ |
346 | static inline void |
347 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 348 | __releases(rq->lock) |
0122ec5b | 349 | __releases(p->pi_lock) |
1da177e4 | 350 | { |
0122ec5b PZ |
351 | raw_spin_unlock(&rq->lock); |
352 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
353 | } |
354 | ||
1da177e4 | 355 | /* |
cc2a73b5 | 356 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 357 | */ |
a9957449 | 358 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
359 | __acquires(rq->lock) |
360 | { | |
70b97a7f | 361 | struct rq *rq; |
1da177e4 LT |
362 | |
363 | local_irq_disable(); | |
364 | rq = this_rq(); | |
05fa785c | 365 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
366 | |
367 | return rq; | |
368 | } | |
369 | ||
8f4d37ec PZ |
370 | #ifdef CONFIG_SCHED_HRTICK |
371 | /* | |
372 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 373 | */ |
8f4d37ec | 374 | |
8f4d37ec PZ |
375 | static void hrtick_clear(struct rq *rq) |
376 | { | |
377 | if (hrtimer_active(&rq->hrtick_timer)) | |
378 | hrtimer_cancel(&rq->hrtick_timer); | |
379 | } | |
380 | ||
8f4d37ec PZ |
381 | /* |
382 | * High-resolution timer tick. | |
383 | * Runs from hardirq context with interrupts disabled. | |
384 | */ | |
385 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
386 | { | |
387 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
388 | ||
389 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
390 | ||
05fa785c | 391 | raw_spin_lock(&rq->lock); |
3e51f33f | 392 | update_rq_clock(rq); |
8f4d37ec | 393 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 394 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
395 | |
396 | return HRTIMER_NORESTART; | |
397 | } | |
398 | ||
95e904c7 | 399 | #ifdef CONFIG_SMP |
971ee28c PZ |
400 | |
401 | static int __hrtick_restart(struct rq *rq) | |
402 | { | |
403 | struct hrtimer *timer = &rq->hrtick_timer; | |
404 | ktime_t time = hrtimer_get_softexpires(timer); | |
405 | ||
406 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
407 | } | |
408 | ||
31656519 PZ |
409 | /* |
410 | * called from hardirq (IPI) context | |
411 | */ | |
412 | static void __hrtick_start(void *arg) | |
b328ca18 | 413 | { |
31656519 | 414 | struct rq *rq = arg; |
b328ca18 | 415 | |
05fa785c | 416 | raw_spin_lock(&rq->lock); |
971ee28c | 417 | __hrtick_restart(rq); |
31656519 | 418 | rq->hrtick_csd_pending = 0; |
05fa785c | 419 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
420 | } |
421 | ||
31656519 PZ |
422 | /* |
423 | * Called to set the hrtick timer state. | |
424 | * | |
425 | * called with rq->lock held and irqs disabled | |
426 | */ | |
029632fb | 427 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 428 | { |
31656519 PZ |
429 | struct hrtimer *timer = &rq->hrtick_timer; |
430 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 431 | |
cc584b21 | 432 | hrtimer_set_expires(timer, time); |
31656519 PZ |
433 | |
434 | if (rq == this_rq()) { | |
971ee28c | 435 | __hrtick_restart(rq); |
31656519 | 436 | } else if (!rq->hrtick_csd_pending) { |
6e275637 | 437 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
438 | rq->hrtick_csd_pending = 1; |
439 | } | |
b328ca18 PZ |
440 | } |
441 | ||
442 | static int | |
443 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
444 | { | |
445 | int cpu = (int)(long)hcpu; | |
446 | ||
447 | switch (action) { | |
448 | case CPU_UP_CANCELED: | |
449 | case CPU_UP_CANCELED_FROZEN: | |
450 | case CPU_DOWN_PREPARE: | |
451 | case CPU_DOWN_PREPARE_FROZEN: | |
452 | case CPU_DEAD: | |
453 | case CPU_DEAD_FROZEN: | |
31656519 | 454 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
455 | return NOTIFY_OK; |
456 | } | |
457 | ||
458 | return NOTIFY_DONE; | |
459 | } | |
460 | ||
fa748203 | 461 | static __init void init_hrtick(void) |
b328ca18 PZ |
462 | { |
463 | hotcpu_notifier(hotplug_hrtick, 0); | |
464 | } | |
31656519 PZ |
465 | #else |
466 | /* | |
467 | * Called to set the hrtick timer state. | |
468 | * | |
469 | * called with rq->lock held and irqs disabled | |
470 | */ | |
029632fb | 471 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 472 | { |
7f1e2ca9 | 473 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 474 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 475 | } |
b328ca18 | 476 | |
006c75f1 | 477 | static inline void init_hrtick(void) |
8f4d37ec | 478 | { |
8f4d37ec | 479 | } |
31656519 | 480 | #endif /* CONFIG_SMP */ |
8f4d37ec | 481 | |
31656519 | 482 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 483 | { |
31656519 PZ |
484 | #ifdef CONFIG_SMP |
485 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 486 | |
31656519 PZ |
487 | rq->hrtick_csd.flags = 0; |
488 | rq->hrtick_csd.func = __hrtick_start; | |
489 | rq->hrtick_csd.info = rq; | |
490 | #endif | |
8f4d37ec | 491 | |
31656519 PZ |
492 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
493 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 494 | } |
006c75f1 | 495 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
496 | static inline void hrtick_clear(struct rq *rq) |
497 | { | |
498 | } | |
499 | ||
8f4d37ec PZ |
500 | static inline void init_rq_hrtick(struct rq *rq) |
501 | { | |
502 | } | |
503 | ||
b328ca18 PZ |
504 | static inline void init_hrtick(void) |
505 | { | |
506 | } | |
006c75f1 | 507 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 508 | |
c24d20db IM |
509 | /* |
510 | * resched_task - mark a task 'to be rescheduled now'. | |
511 | * | |
512 | * On UP this means the setting of the need_resched flag, on SMP it | |
513 | * might also involve a cross-CPU call to trigger the scheduler on | |
514 | * the target CPU. | |
515 | */ | |
029632fb | 516 | void resched_task(struct task_struct *p) |
c24d20db IM |
517 | { |
518 | int cpu; | |
519 | ||
b021fe3e | 520 | lockdep_assert_held(&task_rq(p)->lock); |
c24d20db | 521 | |
5ed0cec0 | 522 | if (test_tsk_need_resched(p)) |
c24d20db IM |
523 | return; |
524 | ||
5ed0cec0 | 525 | set_tsk_need_resched(p); |
c24d20db IM |
526 | |
527 | cpu = task_cpu(p); | |
f27dde8d PZ |
528 | if (cpu == smp_processor_id()) { |
529 | set_preempt_need_resched(); | |
c24d20db | 530 | return; |
f27dde8d | 531 | } |
c24d20db IM |
532 | |
533 | /* NEED_RESCHED must be visible before we test polling */ | |
534 | smp_mb(); | |
535 | if (!tsk_is_polling(p)) | |
536 | smp_send_reschedule(cpu); | |
537 | } | |
538 | ||
029632fb | 539 | void resched_cpu(int cpu) |
c24d20db IM |
540 | { |
541 | struct rq *rq = cpu_rq(cpu); | |
542 | unsigned long flags; | |
543 | ||
05fa785c | 544 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
545 | return; |
546 | resched_task(cpu_curr(cpu)); | |
05fa785c | 547 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 548 | } |
06d8308c | 549 | |
b021fe3e | 550 | #ifdef CONFIG_SMP |
3451d024 | 551 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
552 | /* |
553 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
554 | * from an idle cpu. This is good for power-savings. | |
555 | * | |
556 | * We don't do similar optimization for completely idle system, as | |
557 | * selecting an idle cpu will add more delays to the timers than intended | |
558 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
559 | */ | |
560 | int get_nohz_timer_target(void) | |
561 | { | |
562 | int cpu = smp_processor_id(); | |
563 | int i; | |
564 | struct sched_domain *sd; | |
565 | ||
057f3fad | 566 | rcu_read_lock(); |
83cd4fe2 | 567 | for_each_domain(cpu, sd) { |
057f3fad PZ |
568 | for_each_cpu(i, sched_domain_span(sd)) { |
569 | if (!idle_cpu(i)) { | |
570 | cpu = i; | |
571 | goto unlock; | |
572 | } | |
573 | } | |
83cd4fe2 | 574 | } |
057f3fad PZ |
575 | unlock: |
576 | rcu_read_unlock(); | |
83cd4fe2 VP |
577 | return cpu; |
578 | } | |
06d8308c TG |
579 | /* |
580 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
581 | * idle CPU then this timer might expire before the next timer event | |
582 | * which is scheduled to wake up that CPU. In case of a completely | |
583 | * idle system the next event might even be infinite time into the | |
584 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
585 | * leaves the inner idle loop so the newly added timer is taken into | |
586 | * account when the CPU goes back to idle and evaluates the timer | |
587 | * wheel for the next timer event. | |
588 | */ | |
1c20091e | 589 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
590 | { |
591 | struct rq *rq = cpu_rq(cpu); | |
592 | ||
593 | if (cpu == smp_processor_id()) | |
594 | return; | |
595 | ||
596 | /* | |
597 | * This is safe, as this function is called with the timer | |
598 | * wheel base lock of (cpu) held. When the CPU is on the way | |
599 | * to idle and has not yet set rq->curr to idle then it will | |
600 | * be serialized on the timer wheel base lock and take the new | |
601 | * timer into account automatically. | |
602 | */ | |
603 | if (rq->curr != rq->idle) | |
604 | return; | |
45bf76df | 605 | |
45bf76df | 606 | /* |
06d8308c TG |
607 | * We can set TIF_RESCHED on the idle task of the other CPU |
608 | * lockless. The worst case is that the other CPU runs the | |
609 | * idle task through an additional NOOP schedule() | |
45bf76df | 610 | */ |
5ed0cec0 | 611 | set_tsk_need_resched(rq->idle); |
45bf76df | 612 | |
06d8308c TG |
613 | /* NEED_RESCHED must be visible before we test polling */ |
614 | smp_mb(); | |
615 | if (!tsk_is_polling(rq->idle)) | |
616 | smp_send_reschedule(cpu); | |
45bf76df IM |
617 | } |
618 | ||
c5bfece2 | 619 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 620 | { |
c5bfece2 | 621 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
622 | if (cpu != smp_processor_id() || |
623 | tick_nohz_tick_stopped()) | |
624 | smp_send_reschedule(cpu); | |
625 | return true; | |
626 | } | |
627 | ||
628 | return false; | |
629 | } | |
630 | ||
631 | void wake_up_nohz_cpu(int cpu) | |
632 | { | |
c5bfece2 | 633 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
634 | wake_up_idle_cpu(cpu); |
635 | } | |
636 | ||
ca38062e | 637 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 638 | { |
1c792db7 | 639 | int cpu = smp_processor_id(); |
873b4c65 VG |
640 | |
641 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
642 | return false; | |
643 | ||
644 | if (idle_cpu(cpu) && !need_resched()) | |
645 | return true; | |
646 | ||
647 | /* | |
648 | * We can't run Idle Load Balance on this CPU for this time so we | |
649 | * cancel it and clear NOHZ_BALANCE_KICK | |
650 | */ | |
651 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
652 | return false; | |
45bf76df IM |
653 | } |
654 | ||
3451d024 | 655 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 656 | |
ca38062e | 657 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 658 | { |
ca38062e | 659 | return false; |
2069dd75 PZ |
660 | } |
661 | ||
3451d024 | 662 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 663 | |
ce831b38 FW |
664 | #ifdef CONFIG_NO_HZ_FULL |
665 | bool sched_can_stop_tick(void) | |
666 | { | |
667 | struct rq *rq; | |
668 | ||
669 | rq = this_rq(); | |
670 | ||
671 | /* Make sure rq->nr_running update is visible after the IPI */ | |
672 | smp_rmb(); | |
673 | ||
674 | /* More than one running task need preemption */ | |
675 | if (rq->nr_running > 1) | |
676 | return false; | |
677 | ||
678 | return true; | |
679 | } | |
680 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 681 | |
029632fb | 682 | void sched_avg_update(struct rq *rq) |
18d95a28 | 683 | { |
e9e9250b PZ |
684 | s64 period = sched_avg_period(); |
685 | ||
78becc27 | 686 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
687 | /* |
688 | * Inline assembly required to prevent the compiler | |
689 | * optimising this loop into a divmod call. | |
690 | * See __iter_div_u64_rem() for another example of this. | |
691 | */ | |
692 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
693 | rq->age_stamp += period; |
694 | rq->rt_avg /= 2; | |
695 | } | |
18d95a28 PZ |
696 | } |
697 | ||
6d6bc0ad | 698 | #endif /* CONFIG_SMP */ |
18d95a28 | 699 | |
a790de99 PT |
700 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
701 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 702 | /* |
8277434e PT |
703 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
704 | * node and @up when leaving it for the final time. | |
705 | * | |
706 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 707 | */ |
029632fb | 708 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 709 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
710 | { |
711 | struct task_group *parent, *child; | |
eb755805 | 712 | int ret; |
c09595f6 | 713 | |
8277434e PT |
714 | parent = from; |
715 | ||
c09595f6 | 716 | down: |
eb755805 PZ |
717 | ret = (*down)(parent, data); |
718 | if (ret) | |
8277434e | 719 | goto out; |
c09595f6 PZ |
720 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
721 | parent = child; | |
722 | goto down; | |
723 | ||
724 | up: | |
725 | continue; | |
726 | } | |
eb755805 | 727 | ret = (*up)(parent, data); |
8277434e PT |
728 | if (ret || parent == from) |
729 | goto out; | |
c09595f6 PZ |
730 | |
731 | child = parent; | |
732 | parent = parent->parent; | |
733 | if (parent) | |
734 | goto up; | |
8277434e | 735 | out: |
eb755805 | 736 | return ret; |
c09595f6 PZ |
737 | } |
738 | ||
029632fb | 739 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 740 | { |
e2b245f8 | 741 | return 0; |
eb755805 | 742 | } |
18d95a28 PZ |
743 | #endif |
744 | ||
45bf76df IM |
745 | static void set_load_weight(struct task_struct *p) |
746 | { | |
f05998d4 NR |
747 | int prio = p->static_prio - MAX_RT_PRIO; |
748 | struct load_weight *load = &p->se.load; | |
749 | ||
dd41f596 IM |
750 | /* |
751 | * SCHED_IDLE tasks get minimal weight: | |
752 | */ | |
753 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 754 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 755 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
756 | return; |
757 | } | |
71f8bd46 | 758 | |
c8b28116 | 759 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 760 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
761 | } |
762 | ||
371fd7e7 | 763 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 764 | { |
a64692a3 | 765 | update_rq_clock(rq); |
43148951 | 766 | sched_info_queued(rq, p); |
371fd7e7 | 767 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
768 | } |
769 | ||
371fd7e7 | 770 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 771 | { |
a64692a3 | 772 | update_rq_clock(rq); |
43148951 | 773 | sched_info_dequeued(rq, p); |
371fd7e7 | 774 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
775 | } |
776 | ||
029632fb | 777 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
778 | { |
779 | if (task_contributes_to_load(p)) | |
780 | rq->nr_uninterruptible--; | |
781 | ||
371fd7e7 | 782 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
783 | } |
784 | ||
029632fb | 785 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
786 | { |
787 | if (task_contributes_to_load(p)) | |
788 | rq->nr_uninterruptible++; | |
789 | ||
371fd7e7 | 790 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
791 | } |
792 | ||
fe44d621 | 793 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 794 | { |
095c0aa8 GC |
795 | /* |
796 | * In theory, the compile should just see 0 here, and optimize out the call | |
797 | * to sched_rt_avg_update. But I don't trust it... | |
798 | */ | |
799 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
800 | s64 steal = 0, irq_delta = 0; | |
801 | #endif | |
802 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 803 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
804 | |
805 | /* | |
806 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
807 | * this case when a previous update_rq_clock() happened inside a | |
808 | * {soft,}irq region. | |
809 | * | |
810 | * When this happens, we stop ->clock_task and only update the | |
811 | * prev_irq_time stamp to account for the part that fit, so that a next | |
812 | * update will consume the rest. This ensures ->clock_task is | |
813 | * monotonic. | |
814 | * | |
815 | * It does however cause some slight miss-attribution of {soft,}irq | |
816 | * time, a more accurate solution would be to update the irq_time using | |
817 | * the current rq->clock timestamp, except that would require using | |
818 | * atomic ops. | |
819 | */ | |
820 | if (irq_delta > delta) | |
821 | irq_delta = delta; | |
822 | ||
823 | rq->prev_irq_time += irq_delta; | |
824 | delta -= irq_delta; | |
095c0aa8 GC |
825 | #endif |
826 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 827 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
828 | u64 st; |
829 | ||
830 | steal = paravirt_steal_clock(cpu_of(rq)); | |
831 | steal -= rq->prev_steal_time_rq; | |
832 | ||
833 | if (unlikely(steal > delta)) | |
834 | steal = delta; | |
835 | ||
836 | st = steal_ticks(steal); | |
837 | steal = st * TICK_NSEC; | |
838 | ||
839 | rq->prev_steal_time_rq += steal; | |
840 | ||
841 | delta -= steal; | |
842 | } | |
843 | #endif | |
844 | ||
fe44d621 PZ |
845 | rq->clock_task += delta; |
846 | ||
095c0aa8 GC |
847 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
848 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
849 | sched_rt_avg_update(rq, irq_delta + steal); | |
850 | #endif | |
aa483808 VP |
851 | } |
852 | ||
34f971f6 PZ |
853 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
854 | { | |
855 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
856 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
857 | ||
858 | if (stop) { | |
859 | /* | |
860 | * Make it appear like a SCHED_FIFO task, its something | |
861 | * userspace knows about and won't get confused about. | |
862 | * | |
863 | * Also, it will make PI more or less work without too | |
864 | * much confusion -- but then, stop work should not | |
865 | * rely on PI working anyway. | |
866 | */ | |
867 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
868 | ||
869 | stop->sched_class = &stop_sched_class; | |
870 | } | |
871 | ||
872 | cpu_rq(cpu)->stop = stop; | |
873 | ||
874 | if (old_stop) { | |
875 | /* | |
876 | * Reset it back to a normal scheduling class so that | |
877 | * it can die in pieces. | |
878 | */ | |
879 | old_stop->sched_class = &rt_sched_class; | |
880 | } | |
881 | } | |
882 | ||
14531189 | 883 | /* |
dd41f596 | 884 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 885 | */ |
14531189 IM |
886 | static inline int __normal_prio(struct task_struct *p) |
887 | { | |
dd41f596 | 888 | return p->static_prio; |
14531189 IM |
889 | } |
890 | ||
b29739f9 IM |
891 | /* |
892 | * Calculate the expected normal priority: i.e. priority | |
893 | * without taking RT-inheritance into account. Might be | |
894 | * boosted by interactivity modifiers. Changes upon fork, | |
895 | * setprio syscalls, and whenever the interactivity | |
896 | * estimator recalculates. | |
897 | */ | |
36c8b586 | 898 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
899 | { |
900 | int prio; | |
901 | ||
aab03e05 DF |
902 | if (task_has_dl_policy(p)) |
903 | prio = MAX_DL_PRIO-1; | |
904 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
905 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
906 | else | |
907 | prio = __normal_prio(p); | |
908 | return prio; | |
909 | } | |
910 | ||
911 | /* | |
912 | * Calculate the current priority, i.e. the priority | |
913 | * taken into account by the scheduler. This value might | |
914 | * be boosted by RT tasks, or might be boosted by | |
915 | * interactivity modifiers. Will be RT if the task got | |
916 | * RT-boosted. If not then it returns p->normal_prio. | |
917 | */ | |
36c8b586 | 918 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
919 | { |
920 | p->normal_prio = normal_prio(p); | |
921 | /* | |
922 | * If we are RT tasks or we were boosted to RT priority, | |
923 | * keep the priority unchanged. Otherwise, update priority | |
924 | * to the normal priority: | |
925 | */ | |
926 | if (!rt_prio(p->prio)) | |
927 | return p->normal_prio; | |
928 | return p->prio; | |
929 | } | |
930 | ||
1da177e4 LT |
931 | /** |
932 | * task_curr - is this task currently executing on a CPU? | |
933 | * @p: the task in question. | |
e69f6186 YB |
934 | * |
935 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 936 | */ |
36c8b586 | 937 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
938 | { |
939 | return cpu_curr(task_cpu(p)) == p; | |
940 | } | |
941 | ||
cb469845 SR |
942 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
943 | const struct sched_class *prev_class, | |
da7a735e | 944 | int oldprio) |
cb469845 SR |
945 | { |
946 | if (prev_class != p->sched_class) { | |
947 | if (prev_class->switched_from) | |
da7a735e PZ |
948 | prev_class->switched_from(rq, p); |
949 | p->sched_class->switched_to(rq, p); | |
950 | } else if (oldprio != p->prio) | |
951 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
952 | } |
953 | ||
029632fb | 954 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
955 | { |
956 | const struct sched_class *class; | |
957 | ||
958 | if (p->sched_class == rq->curr->sched_class) { | |
959 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
960 | } else { | |
961 | for_each_class(class) { | |
962 | if (class == rq->curr->sched_class) | |
963 | break; | |
964 | if (class == p->sched_class) { | |
965 | resched_task(rq->curr); | |
966 | break; | |
967 | } | |
968 | } | |
969 | } | |
970 | ||
971 | /* | |
972 | * A queue event has occurred, and we're going to schedule. In | |
973 | * this case, we can save a useless back to back clock update. | |
974 | */ | |
fd2f4419 | 975 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
976 | rq->skip_clock_update = 1; |
977 | } | |
978 | ||
1da177e4 | 979 | #ifdef CONFIG_SMP |
dd41f596 | 980 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 981 | { |
e2912009 PZ |
982 | #ifdef CONFIG_SCHED_DEBUG |
983 | /* | |
984 | * We should never call set_task_cpu() on a blocked task, | |
985 | * ttwu() will sort out the placement. | |
986 | */ | |
077614ee | 987 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 988 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
989 | |
990 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
991 | /* |
992 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
993 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
994 | * | |
995 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 996 | * see task_group(). |
6c6c54e1 PZ |
997 | * |
998 | * Furthermore, all task_rq users should acquire both locks, see | |
999 | * task_rq_lock(). | |
1000 | */ | |
0122ec5b PZ |
1001 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1002 | lockdep_is_held(&task_rq(p)->lock))); | |
1003 | #endif | |
e2912009 PZ |
1004 | #endif |
1005 | ||
de1d7286 | 1006 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1007 | |
0c69774e | 1008 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1009 | if (p->sched_class->migrate_task_rq) |
1010 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1011 | p->se.nr_migrations++; |
a8b0ca17 | 1012 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1013 | } |
dd41f596 IM |
1014 | |
1015 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1016 | } |
1017 | ||
ac66f547 PZ |
1018 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1019 | { | |
1020 | if (p->on_rq) { | |
1021 | struct rq *src_rq, *dst_rq; | |
1022 | ||
1023 | src_rq = task_rq(p); | |
1024 | dst_rq = cpu_rq(cpu); | |
1025 | ||
1026 | deactivate_task(src_rq, p, 0); | |
1027 | set_task_cpu(p, cpu); | |
1028 | activate_task(dst_rq, p, 0); | |
1029 | check_preempt_curr(dst_rq, p, 0); | |
1030 | } else { | |
1031 | /* | |
1032 | * Task isn't running anymore; make it appear like we migrated | |
1033 | * it before it went to sleep. This means on wakeup we make the | |
1034 | * previous cpu our targer instead of where it really is. | |
1035 | */ | |
1036 | p->wake_cpu = cpu; | |
1037 | } | |
1038 | } | |
1039 | ||
1040 | struct migration_swap_arg { | |
1041 | struct task_struct *src_task, *dst_task; | |
1042 | int src_cpu, dst_cpu; | |
1043 | }; | |
1044 | ||
1045 | static int migrate_swap_stop(void *data) | |
1046 | { | |
1047 | struct migration_swap_arg *arg = data; | |
1048 | struct rq *src_rq, *dst_rq; | |
1049 | int ret = -EAGAIN; | |
1050 | ||
1051 | src_rq = cpu_rq(arg->src_cpu); | |
1052 | dst_rq = cpu_rq(arg->dst_cpu); | |
1053 | ||
74602315 PZ |
1054 | double_raw_lock(&arg->src_task->pi_lock, |
1055 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1056 | double_rq_lock(src_rq, dst_rq); |
1057 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1058 | goto unlock; | |
1059 | ||
1060 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1061 | goto unlock; | |
1062 | ||
1063 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1064 | goto unlock; | |
1065 | ||
1066 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1067 | goto unlock; | |
1068 | ||
1069 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1070 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1071 | ||
1072 | ret = 0; | |
1073 | ||
1074 | unlock: | |
1075 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1076 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1077 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1078 | |
1079 | return ret; | |
1080 | } | |
1081 | ||
1082 | /* | |
1083 | * Cross migrate two tasks | |
1084 | */ | |
1085 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1086 | { | |
1087 | struct migration_swap_arg arg; | |
1088 | int ret = -EINVAL; | |
1089 | ||
ac66f547 PZ |
1090 | arg = (struct migration_swap_arg){ |
1091 | .src_task = cur, | |
1092 | .src_cpu = task_cpu(cur), | |
1093 | .dst_task = p, | |
1094 | .dst_cpu = task_cpu(p), | |
1095 | }; | |
1096 | ||
1097 | if (arg.src_cpu == arg.dst_cpu) | |
1098 | goto out; | |
1099 | ||
6acce3ef PZ |
1100 | /* |
1101 | * These three tests are all lockless; this is OK since all of them | |
1102 | * will be re-checked with proper locks held further down the line. | |
1103 | */ | |
ac66f547 PZ |
1104 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1105 | goto out; | |
1106 | ||
1107 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1108 | goto out; | |
1109 | ||
1110 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1111 | goto out; | |
1112 | ||
1113 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); | |
1114 | ||
1115 | out: | |
ac66f547 PZ |
1116 | return ret; |
1117 | } | |
1118 | ||
969c7921 | 1119 | struct migration_arg { |
36c8b586 | 1120 | struct task_struct *task; |
1da177e4 | 1121 | int dest_cpu; |
70b97a7f | 1122 | }; |
1da177e4 | 1123 | |
969c7921 TH |
1124 | static int migration_cpu_stop(void *data); |
1125 | ||
1da177e4 LT |
1126 | /* |
1127 | * wait_task_inactive - wait for a thread to unschedule. | |
1128 | * | |
85ba2d86 RM |
1129 | * If @match_state is nonzero, it's the @p->state value just checked and |
1130 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1131 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1132 | * we return a positive number (its total switch count). If a second call | |
1133 | * a short while later returns the same number, the caller can be sure that | |
1134 | * @p has remained unscheduled the whole time. | |
1135 | * | |
1da177e4 LT |
1136 | * The caller must ensure that the task *will* unschedule sometime soon, |
1137 | * else this function might spin for a *long* time. This function can't | |
1138 | * be called with interrupts off, or it may introduce deadlock with | |
1139 | * smp_call_function() if an IPI is sent by the same process we are | |
1140 | * waiting to become inactive. | |
1141 | */ | |
85ba2d86 | 1142 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1143 | { |
1144 | unsigned long flags; | |
dd41f596 | 1145 | int running, on_rq; |
85ba2d86 | 1146 | unsigned long ncsw; |
70b97a7f | 1147 | struct rq *rq; |
1da177e4 | 1148 | |
3a5c359a AK |
1149 | for (;;) { |
1150 | /* | |
1151 | * We do the initial early heuristics without holding | |
1152 | * any task-queue locks at all. We'll only try to get | |
1153 | * the runqueue lock when things look like they will | |
1154 | * work out! | |
1155 | */ | |
1156 | rq = task_rq(p); | |
fa490cfd | 1157 | |
3a5c359a AK |
1158 | /* |
1159 | * If the task is actively running on another CPU | |
1160 | * still, just relax and busy-wait without holding | |
1161 | * any locks. | |
1162 | * | |
1163 | * NOTE! Since we don't hold any locks, it's not | |
1164 | * even sure that "rq" stays as the right runqueue! | |
1165 | * But we don't care, since "task_running()" will | |
1166 | * return false if the runqueue has changed and p | |
1167 | * is actually now running somewhere else! | |
1168 | */ | |
85ba2d86 RM |
1169 | while (task_running(rq, p)) { |
1170 | if (match_state && unlikely(p->state != match_state)) | |
1171 | return 0; | |
3a5c359a | 1172 | cpu_relax(); |
85ba2d86 | 1173 | } |
fa490cfd | 1174 | |
3a5c359a AK |
1175 | /* |
1176 | * Ok, time to look more closely! We need the rq | |
1177 | * lock now, to be *sure*. If we're wrong, we'll | |
1178 | * just go back and repeat. | |
1179 | */ | |
1180 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1181 | trace_sched_wait_task(p); |
3a5c359a | 1182 | running = task_running(rq, p); |
fd2f4419 | 1183 | on_rq = p->on_rq; |
85ba2d86 | 1184 | ncsw = 0; |
f31e11d8 | 1185 | if (!match_state || p->state == match_state) |
93dcf55f | 1186 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1187 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1188 | |
85ba2d86 RM |
1189 | /* |
1190 | * If it changed from the expected state, bail out now. | |
1191 | */ | |
1192 | if (unlikely(!ncsw)) | |
1193 | break; | |
1194 | ||
3a5c359a AK |
1195 | /* |
1196 | * Was it really running after all now that we | |
1197 | * checked with the proper locks actually held? | |
1198 | * | |
1199 | * Oops. Go back and try again.. | |
1200 | */ | |
1201 | if (unlikely(running)) { | |
1202 | cpu_relax(); | |
1203 | continue; | |
1204 | } | |
fa490cfd | 1205 | |
3a5c359a AK |
1206 | /* |
1207 | * It's not enough that it's not actively running, | |
1208 | * it must be off the runqueue _entirely_, and not | |
1209 | * preempted! | |
1210 | * | |
80dd99b3 | 1211 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1212 | * running right now), it's preempted, and we should |
1213 | * yield - it could be a while. | |
1214 | */ | |
1215 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1216 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1217 | ||
1218 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1219 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1220 | continue; |
1221 | } | |
fa490cfd | 1222 | |
3a5c359a AK |
1223 | /* |
1224 | * Ahh, all good. It wasn't running, and it wasn't | |
1225 | * runnable, which means that it will never become | |
1226 | * running in the future either. We're all done! | |
1227 | */ | |
1228 | break; | |
1229 | } | |
85ba2d86 RM |
1230 | |
1231 | return ncsw; | |
1da177e4 LT |
1232 | } |
1233 | ||
1234 | /*** | |
1235 | * kick_process - kick a running thread to enter/exit the kernel | |
1236 | * @p: the to-be-kicked thread | |
1237 | * | |
1238 | * Cause a process which is running on another CPU to enter | |
1239 | * kernel-mode, without any delay. (to get signals handled.) | |
1240 | * | |
25985edc | 1241 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1242 | * because all it wants to ensure is that the remote task enters |
1243 | * the kernel. If the IPI races and the task has been migrated | |
1244 | * to another CPU then no harm is done and the purpose has been | |
1245 | * achieved as well. | |
1246 | */ | |
36c8b586 | 1247 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1248 | { |
1249 | int cpu; | |
1250 | ||
1251 | preempt_disable(); | |
1252 | cpu = task_cpu(p); | |
1253 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1254 | smp_send_reschedule(cpu); | |
1255 | preempt_enable(); | |
1256 | } | |
b43e3521 | 1257 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1258 | #endif /* CONFIG_SMP */ |
1da177e4 | 1259 | |
970b13ba | 1260 | #ifdef CONFIG_SMP |
30da688e | 1261 | /* |
013fdb80 | 1262 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1263 | */ |
5da9a0fb PZ |
1264 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1265 | { | |
aa00d89c TC |
1266 | int nid = cpu_to_node(cpu); |
1267 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1268 | enum { cpuset, possible, fail } state = cpuset; |
1269 | int dest_cpu; | |
5da9a0fb | 1270 | |
aa00d89c TC |
1271 | /* |
1272 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1273 | * will return -1. There is no cpu on the node, and we should | |
1274 | * select the cpu on the other node. | |
1275 | */ | |
1276 | if (nid != -1) { | |
1277 | nodemask = cpumask_of_node(nid); | |
1278 | ||
1279 | /* Look for allowed, online CPU in same node. */ | |
1280 | for_each_cpu(dest_cpu, nodemask) { | |
1281 | if (!cpu_online(dest_cpu)) | |
1282 | continue; | |
1283 | if (!cpu_active(dest_cpu)) | |
1284 | continue; | |
1285 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1286 | return dest_cpu; | |
1287 | } | |
2baab4e9 | 1288 | } |
5da9a0fb | 1289 | |
2baab4e9 PZ |
1290 | for (;;) { |
1291 | /* Any allowed, online CPU? */ | |
e3831edd | 1292 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1293 | if (!cpu_online(dest_cpu)) |
1294 | continue; | |
1295 | if (!cpu_active(dest_cpu)) | |
1296 | continue; | |
1297 | goto out; | |
1298 | } | |
5da9a0fb | 1299 | |
2baab4e9 PZ |
1300 | switch (state) { |
1301 | case cpuset: | |
1302 | /* No more Mr. Nice Guy. */ | |
1303 | cpuset_cpus_allowed_fallback(p); | |
1304 | state = possible; | |
1305 | break; | |
1306 | ||
1307 | case possible: | |
1308 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1309 | state = fail; | |
1310 | break; | |
1311 | ||
1312 | case fail: | |
1313 | BUG(); | |
1314 | break; | |
1315 | } | |
1316 | } | |
1317 | ||
1318 | out: | |
1319 | if (state != cpuset) { | |
1320 | /* | |
1321 | * Don't tell them about moving exiting tasks or | |
1322 | * kernel threads (both mm NULL), since they never | |
1323 | * leave kernel. | |
1324 | */ | |
1325 | if (p->mm && printk_ratelimit()) { | |
1326 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1327 | task_pid_nr(p), p->comm, cpu); | |
1328 | } | |
5da9a0fb PZ |
1329 | } |
1330 | ||
1331 | return dest_cpu; | |
1332 | } | |
1333 | ||
e2912009 | 1334 | /* |
013fdb80 | 1335 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1336 | */ |
970b13ba | 1337 | static inline |
ac66f547 | 1338 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1339 | { |
ac66f547 | 1340 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1341 | |
1342 | /* | |
1343 | * In order not to call set_task_cpu() on a blocking task we need | |
1344 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1345 | * cpu. | |
1346 | * | |
1347 | * Since this is common to all placement strategies, this lives here. | |
1348 | * | |
1349 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1350 | * not worry about this generic constraint ] | |
1351 | */ | |
fa17b507 | 1352 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1353 | !cpu_online(cpu))) |
5da9a0fb | 1354 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1355 | |
1356 | return cpu; | |
970b13ba | 1357 | } |
09a40af5 MG |
1358 | |
1359 | static void update_avg(u64 *avg, u64 sample) | |
1360 | { | |
1361 | s64 diff = sample - *avg; | |
1362 | *avg += diff >> 3; | |
1363 | } | |
970b13ba PZ |
1364 | #endif |
1365 | ||
d7c01d27 | 1366 | static void |
b84cb5df | 1367 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1368 | { |
d7c01d27 | 1369 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1370 | struct rq *rq = this_rq(); |
1371 | ||
d7c01d27 PZ |
1372 | #ifdef CONFIG_SMP |
1373 | int this_cpu = smp_processor_id(); | |
1374 | ||
1375 | if (cpu == this_cpu) { | |
1376 | schedstat_inc(rq, ttwu_local); | |
1377 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1378 | } else { | |
1379 | struct sched_domain *sd; | |
1380 | ||
1381 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1382 | rcu_read_lock(); |
d7c01d27 PZ |
1383 | for_each_domain(this_cpu, sd) { |
1384 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1385 | schedstat_inc(sd, ttwu_wake_remote); | |
1386 | break; | |
1387 | } | |
1388 | } | |
057f3fad | 1389 | rcu_read_unlock(); |
d7c01d27 | 1390 | } |
f339b9dc PZ |
1391 | |
1392 | if (wake_flags & WF_MIGRATED) | |
1393 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1394 | ||
d7c01d27 PZ |
1395 | #endif /* CONFIG_SMP */ |
1396 | ||
1397 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1398 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1399 | |
1400 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1401 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1402 | |
d7c01d27 PZ |
1403 | #endif /* CONFIG_SCHEDSTATS */ |
1404 | } | |
1405 | ||
1406 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1407 | { | |
9ed3811a | 1408 | activate_task(rq, p, en_flags); |
fd2f4419 | 1409 | p->on_rq = 1; |
c2f7115e PZ |
1410 | |
1411 | /* if a worker is waking up, notify workqueue */ | |
1412 | if (p->flags & PF_WQ_WORKER) | |
1413 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1414 | } |
1415 | ||
23f41eeb PZ |
1416 | /* |
1417 | * Mark the task runnable and perform wakeup-preemption. | |
1418 | */ | |
89363381 | 1419 | static void |
23f41eeb | 1420 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1421 | { |
9ed3811a | 1422 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1423 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1424 | |
1425 | p->state = TASK_RUNNING; | |
1426 | #ifdef CONFIG_SMP | |
1427 | if (p->sched_class->task_woken) | |
1428 | p->sched_class->task_woken(rq, p); | |
1429 | ||
e69c6341 | 1430 | if (rq->idle_stamp) { |
78becc27 | 1431 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1432 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1433 | |
abfafa54 JL |
1434 | update_avg(&rq->avg_idle, delta); |
1435 | ||
1436 | if (rq->avg_idle > max) | |
9ed3811a | 1437 | rq->avg_idle = max; |
abfafa54 | 1438 | |
9ed3811a TH |
1439 | rq->idle_stamp = 0; |
1440 | } | |
1441 | #endif | |
1442 | } | |
1443 | ||
c05fbafb PZ |
1444 | static void |
1445 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1446 | { | |
1447 | #ifdef CONFIG_SMP | |
1448 | if (p->sched_contributes_to_load) | |
1449 | rq->nr_uninterruptible--; | |
1450 | #endif | |
1451 | ||
1452 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1453 | ttwu_do_wakeup(rq, p, wake_flags); | |
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1458 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1459 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1460 | * the task is still ->on_rq. | |
1461 | */ | |
1462 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1463 | { | |
1464 | struct rq *rq; | |
1465 | int ret = 0; | |
1466 | ||
1467 | rq = __task_rq_lock(p); | |
1468 | if (p->on_rq) { | |
1ad4ec0d FW |
1469 | /* check_preempt_curr() may use rq clock */ |
1470 | update_rq_clock(rq); | |
c05fbafb PZ |
1471 | ttwu_do_wakeup(rq, p, wake_flags); |
1472 | ret = 1; | |
1473 | } | |
1474 | __task_rq_unlock(rq); | |
1475 | ||
1476 | return ret; | |
1477 | } | |
1478 | ||
317f3941 | 1479 | #ifdef CONFIG_SMP |
fa14ff4a | 1480 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1481 | { |
1482 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1483 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1484 | struct task_struct *p; | |
317f3941 PZ |
1485 | |
1486 | raw_spin_lock(&rq->lock); | |
1487 | ||
fa14ff4a PZ |
1488 | while (llist) { |
1489 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1490 | llist = llist_next(llist); | |
317f3941 PZ |
1491 | ttwu_do_activate(rq, p, 0); |
1492 | } | |
1493 | ||
1494 | raw_spin_unlock(&rq->lock); | |
1495 | } | |
1496 | ||
1497 | void scheduler_ipi(void) | |
1498 | { | |
f27dde8d PZ |
1499 | /* |
1500 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1501 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1502 | * this IPI. | |
1503 | */ | |
1504 | if (tif_need_resched()) | |
1505 | set_preempt_need_resched(); | |
1506 | ||
873b4c65 VG |
1507 | if (llist_empty(&this_rq()->wake_list) |
1508 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1509 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1510 | return; |
1511 | ||
1512 | /* | |
1513 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1514 | * traditionally all their work was done from the interrupt return | |
1515 | * path. Now that we actually do some work, we need to make sure | |
1516 | * we do call them. | |
1517 | * | |
1518 | * Some archs already do call them, luckily irq_enter/exit nest | |
1519 | * properly. | |
1520 | * | |
1521 | * Arguably we should visit all archs and update all handlers, | |
1522 | * however a fair share of IPIs are still resched only so this would | |
1523 | * somewhat pessimize the simple resched case. | |
1524 | */ | |
1525 | irq_enter(); | |
ff442c51 | 1526 | tick_nohz_full_check(); |
fa14ff4a | 1527 | sched_ttwu_pending(); |
ca38062e SS |
1528 | |
1529 | /* | |
1530 | * Check if someone kicked us for doing the nohz idle load balance. | |
1531 | */ | |
873b4c65 | 1532 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1533 | this_rq()->idle_balance = 1; |
ca38062e | 1534 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1535 | } |
c5d753a5 | 1536 | irq_exit(); |
317f3941 PZ |
1537 | } |
1538 | ||
1539 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1540 | { | |
fa14ff4a | 1541 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1542 | smp_send_reschedule(cpu); |
1543 | } | |
d6aa8f85 | 1544 | |
39be3501 | 1545 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1546 | { |
1547 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1548 | } | |
d6aa8f85 | 1549 | #endif /* CONFIG_SMP */ |
317f3941 | 1550 | |
c05fbafb PZ |
1551 | static void ttwu_queue(struct task_struct *p, int cpu) |
1552 | { | |
1553 | struct rq *rq = cpu_rq(cpu); | |
1554 | ||
17d9f311 | 1555 | #if defined(CONFIG_SMP) |
39be3501 | 1556 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1557 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1558 | ttwu_queue_remote(p, cpu); |
1559 | return; | |
1560 | } | |
1561 | #endif | |
1562 | ||
c05fbafb PZ |
1563 | raw_spin_lock(&rq->lock); |
1564 | ttwu_do_activate(rq, p, 0); | |
1565 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1566 | } |
1567 | ||
1568 | /** | |
1da177e4 | 1569 | * try_to_wake_up - wake up a thread |
9ed3811a | 1570 | * @p: the thread to be awakened |
1da177e4 | 1571 | * @state: the mask of task states that can be woken |
9ed3811a | 1572 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1573 | * |
1574 | * Put it on the run-queue if it's not already there. The "current" | |
1575 | * thread is always on the run-queue (except when the actual | |
1576 | * re-schedule is in progress), and as such you're allowed to do | |
1577 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1578 | * runnable without the overhead of this. | |
1579 | * | |
e69f6186 | 1580 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1581 | * or @state didn't match @p's state. |
1da177e4 | 1582 | */ |
e4a52bcb PZ |
1583 | static int |
1584 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1585 | { |
1da177e4 | 1586 | unsigned long flags; |
c05fbafb | 1587 | int cpu, success = 0; |
2398f2c6 | 1588 | |
e0acd0a6 ON |
1589 | /* |
1590 | * If we are going to wake up a thread waiting for CONDITION we | |
1591 | * need to ensure that CONDITION=1 done by the caller can not be | |
1592 | * reordered with p->state check below. This pairs with mb() in | |
1593 | * set_current_state() the waiting thread does. | |
1594 | */ | |
1595 | smp_mb__before_spinlock(); | |
013fdb80 | 1596 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1597 | if (!(p->state & state)) |
1da177e4 LT |
1598 | goto out; |
1599 | ||
c05fbafb | 1600 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1601 | cpu = task_cpu(p); |
1da177e4 | 1602 | |
c05fbafb PZ |
1603 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1604 | goto stat; | |
1da177e4 | 1605 | |
1da177e4 | 1606 | #ifdef CONFIG_SMP |
e9c84311 | 1607 | /* |
c05fbafb PZ |
1608 | * If the owning (remote) cpu is still in the middle of schedule() with |
1609 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1610 | */ |
f3e94786 | 1611 | while (p->on_cpu) |
e4a52bcb | 1612 | cpu_relax(); |
0970d299 | 1613 | /* |
e4a52bcb | 1614 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1615 | */ |
e4a52bcb | 1616 | smp_rmb(); |
1da177e4 | 1617 | |
a8e4f2ea | 1618 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1619 | p->state = TASK_WAKING; |
e7693a36 | 1620 | |
e4a52bcb | 1621 | if (p->sched_class->task_waking) |
74f8e4b2 | 1622 | p->sched_class->task_waking(p); |
efbbd05a | 1623 | |
ac66f547 | 1624 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1625 | if (task_cpu(p) != cpu) { |
1626 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1627 | set_task_cpu(p, cpu); |
f339b9dc | 1628 | } |
1da177e4 | 1629 | #endif /* CONFIG_SMP */ |
1da177e4 | 1630 | |
c05fbafb PZ |
1631 | ttwu_queue(p, cpu); |
1632 | stat: | |
b84cb5df | 1633 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1634 | out: |
013fdb80 | 1635 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1636 | |
1637 | return success; | |
1638 | } | |
1639 | ||
21aa9af0 TH |
1640 | /** |
1641 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1642 | * @p: the thread to be awakened | |
1643 | * | |
2acca55e | 1644 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1645 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1646 | * the current task. |
21aa9af0 TH |
1647 | */ |
1648 | static void try_to_wake_up_local(struct task_struct *p) | |
1649 | { | |
1650 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1651 | |
383efcd0 TH |
1652 | if (WARN_ON_ONCE(rq != this_rq()) || |
1653 | WARN_ON_ONCE(p == current)) | |
1654 | return; | |
1655 | ||
21aa9af0 TH |
1656 | lockdep_assert_held(&rq->lock); |
1657 | ||
2acca55e PZ |
1658 | if (!raw_spin_trylock(&p->pi_lock)) { |
1659 | raw_spin_unlock(&rq->lock); | |
1660 | raw_spin_lock(&p->pi_lock); | |
1661 | raw_spin_lock(&rq->lock); | |
1662 | } | |
1663 | ||
21aa9af0 | 1664 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1665 | goto out; |
21aa9af0 | 1666 | |
fd2f4419 | 1667 | if (!p->on_rq) |
d7c01d27 PZ |
1668 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1669 | ||
23f41eeb | 1670 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1671 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1672 | out: |
1673 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1674 | } |
1675 | ||
50fa610a DH |
1676 | /** |
1677 | * wake_up_process - Wake up a specific process | |
1678 | * @p: The process to be woken up. | |
1679 | * | |
1680 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1681 | * processes. |
1682 | * | |
1683 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1684 | * |
1685 | * It may be assumed that this function implies a write memory barrier before | |
1686 | * changing the task state if and only if any tasks are woken up. | |
1687 | */ | |
7ad5b3a5 | 1688 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1689 | { |
9067ac85 ON |
1690 | WARN_ON(task_is_stopped_or_traced(p)); |
1691 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1692 | } |
1da177e4 LT |
1693 | EXPORT_SYMBOL(wake_up_process); |
1694 | ||
7ad5b3a5 | 1695 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1696 | { |
1697 | return try_to_wake_up(p, state, 0); | |
1698 | } | |
1699 | ||
1da177e4 LT |
1700 | /* |
1701 | * Perform scheduler related setup for a newly forked process p. | |
1702 | * p is forked by current. | |
dd41f596 IM |
1703 | * |
1704 | * __sched_fork() is basic setup used by init_idle() too: | |
1705 | */ | |
5e1576ed | 1706 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1707 | { |
fd2f4419 PZ |
1708 | p->on_rq = 0; |
1709 | ||
1710 | p->se.on_rq = 0; | |
dd41f596 IM |
1711 | p->se.exec_start = 0; |
1712 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1713 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1714 | p->se.nr_migrations = 0; |
da7a735e | 1715 | p->se.vruntime = 0; |
fd2f4419 | 1716 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1717 | |
1718 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1719 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1720 | #endif |
476d139c | 1721 | |
aab03e05 DF |
1722 | RB_CLEAR_NODE(&p->dl.rb_node); |
1723 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1724 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1725 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1726 | p->dl.dl_period = 0; |
aab03e05 DF |
1727 | p->dl.flags = 0; |
1728 | ||
fa717060 | 1729 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1730 | |
e107be36 AK |
1731 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1732 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1733 | #endif | |
cbee9f88 PZ |
1734 | |
1735 | #ifdef CONFIG_NUMA_BALANCING | |
1736 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1737 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1738 | p->mm->numa_scan_seq = 0; |
1739 | } | |
1740 | ||
5e1576ed RR |
1741 | if (clone_flags & CLONE_VM) |
1742 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1743 | else | |
1744 | p->numa_preferred_nid = -1; | |
1745 | ||
cbee9f88 PZ |
1746 | p->node_stamp = 0ULL; |
1747 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1748 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1749 | p->numa_work.next = &p->numa_work; |
f809ca9a | 1750 | p->numa_faults = NULL; |
745d6147 | 1751 | p->numa_faults_buffer = NULL; |
8c8a743c PZ |
1752 | |
1753 | INIT_LIST_HEAD(&p->numa_entry); | |
1754 | p->numa_group = NULL; | |
cbee9f88 | 1755 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1756 | } |
1757 | ||
1a687c2e | 1758 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1759 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1760 | void set_numabalancing_state(bool enabled) |
1761 | { | |
1762 | if (enabled) | |
1763 | sched_feat_set("NUMA"); | |
1764 | else | |
1765 | sched_feat_set("NO_NUMA"); | |
1766 | } | |
3105b86a MG |
1767 | #else |
1768 | __read_mostly bool numabalancing_enabled; | |
1769 | ||
1770 | void set_numabalancing_state(bool enabled) | |
1771 | { | |
1772 | numabalancing_enabled = enabled; | |
dd41f596 | 1773 | } |
3105b86a | 1774 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1775 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1776 | |
1777 | /* | |
1778 | * fork()/clone()-time setup: | |
1779 | */ | |
aab03e05 | 1780 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1781 | { |
0122ec5b | 1782 | unsigned long flags; |
dd41f596 IM |
1783 | int cpu = get_cpu(); |
1784 | ||
5e1576ed | 1785 | __sched_fork(clone_flags, p); |
06b83b5f | 1786 | /* |
0017d735 | 1787 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1788 | * nobody will actually run it, and a signal or other external |
1789 | * event cannot wake it up and insert it on the runqueue either. | |
1790 | */ | |
0017d735 | 1791 | p->state = TASK_RUNNING; |
dd41f596 | 1792 | |
c350a04e MG |
1793 | /* |
1794 | * Make sure we do not leak PI boosting priority to the child. | |
1795 | */ | |
1796 | p->prio = current->normal_prio; | |
1797 | ||
b9dc29e7 MG |
1798 | /* |
1799 | * Revert to default priority/policy on fork if requested. | |
1800 | */ | |
1801 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1802 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1803 | p->policy = SCHED_NORMAL; |
6c697bdf | 1804 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1805 | p->rt_priority = 0; |
1806 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1807 | p->static_prio = NICE_TO_PRIO(0); | |
1808 | ||
1809 | p->prio = p->normal_prio = __normal_prio(p); | |
1810 | set_load_weight(p); | |
6c697bdf | 1811 | |
b9dc29e7 MG |
1812 | /* |
1813 | * We don't need the reset flag anymore after the fork. It has | |
1814 | * fulfilled its duty: | |
1815 | */ | |
1816 | p->sched_reset_on_fork = 0; | |
1817 | } | |
ca94c442 | 1818 | |
aab03e05 DF |
1819 | if (dl_prio(p->prio)) { |
1820 | put_cpu(); | |
1821 | return -EAGAIN; | |
1822 | } else if (rt_prio(p->prio)) { | |
1823 | p->sched_class = &rt_sched_class; | |
1824 | } else { | |
2ddbf952 | 1825 | p->sched_class = &fair_sched_class; |
aab03e05 | 1826 | } |
b29739f9 | 1827 | |
cd29fe6f PZ |
1828 | if (p->sched_class->task_fork) |
1829 | p->sched_class->task_fork(p); | |
1830 | ||
86951599 PZ |
1831 | /* |
1832 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1833 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1834 | * is ran before sched_fork(). | |
1835 | * | |
1836 | * Silence PROVE_RCU. | |
1837 | */ | |
0122ec5b | 1838 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1839 | set_task_cpu(p, cpu); |
0122ec5b | 1840 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1841 | |
52f17b6c | 1842 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1843 | if (likely(sched_info_on())) |
52f17b6c | 1844 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1845 | #endif |
3ca7a440 PZ |
1846 | #if defined(CONFIG_SMP) |
1847 | p->on_cpu = 0; | |
4866cde0 | 1848 | #endif |
01028747 | 1849 | init_task_preempt_count(p); |
806c09a7 | 1850 | #ifdef CONFIG_SMP |
917b627d | 1851 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1852 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1853 | #endif |
917b627d | 1854 | |
476d139c | 1855 | put_cpu(); |
aab03e05 | 1856 | return 0; |
1da177e4 LT |
1857 | } |
1858 | ||
1859 | /* | |
1860 | * wake_up_new_task - wake up a newly created task for the first time. | |
1861 | * | |
1862 | * This function will do some initial scheduler statistics housekeeping | |
1863 | * that must be done for every newly created context, then puts the task | |
1864 | * on the runqueue and wakes it. | |
1865 | */ | |
3e51e3ed | 1866 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1867 | { |
1868 | unsigned long flags; | |
dd41f596 | 1869 | struct rq *rq; |
fabf318e | 1870 | |
ab2515c4 | 1871 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1872 | #ifdef CONFIG_SMP |
1873 | /* | |
1874 | * Fork balancing, do it here and not earlier because: | |
1875 | * - cpus_allowed can change in the fork path | |
1876 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1877 | */ |
ac66f547 | 1878 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1879 | #endif |
1880 | ||
a75cdaa9 AS |
1881 | /* Initialize new task's runnable average */ |
1882 | init_task_runnable_average(p); | |
ab2515c4 | 1883 | rq = __task_rq_lock(p); |
cd29fe6f | 1884 | activate_task(rq, p, 0); |
fd2f4419 | 1885 | p->on_rq = 1; |
89363381 | 1886 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1887 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1888 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1889 | if (p->sched_class->task_woken) |
1890 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1891 | #endif |
0122ec5b | 1892 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1893 | } |
1894 | ||
e107be36 AK |
1895 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1896 | ||
1897 | /** | |
80dd99b3 | 1898 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1899 | * @notifier: notifier struct to register |
e107be36 AK |
1900 | */ |
1901 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1902 | { | |
1903 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1904 | } | |
1905 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1906 | ||
1907 | /** | |
1908 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1909 | * @notifier: notifier struct to unregister |
e107be36 AK |
1910 | * |
1911 | * This is safe to call from within a preemption notifier. | |
1912 | */ | |
1913 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1914 | { | |
1915 | hlist_del(¬ifier->link); | |
1916 | } | |
1917 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1918 | ||
1919 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1920 | { | |
1921 | struct preempt_notifier *notifier; | |
e107be36 | 1922 | |
b67bfe0d | 1923 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1924 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
1925 | } | |
1926 | ||
1927 | static void | |
1928 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1929 | struct task_struct *next) | |
1930 | { | |
1931 | struct preempt_notifier *notifier; | |
e107be36 | 1932 | |
b67bfe0d | 1933 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
1934 | notifier->ops->sched_out(notifier, next); |
1935 | } | |
1936 | ||
6d6bc0ad | 1937 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1938 | |
1939 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1940 | { | |
1941 | } | |
1942 | ||
1943 | static void | |
1944 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1945 | struct task_struct *next) | |
1946 | { | |
1947 | } | |
1948 | ||
6d6bc0ad | 1949 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1950 | |
4866cde0 NP |
1951 | /** |
1952 | * prepare_task_switch - prepare to switch tasks | |
1953 | * @rq: the runqueue preparing to switch | |
421cee29 | 1954 | * @prev: the current task that is being switched out |
4866cde0 NP |
1955 | * @next: the task we are going to switch to. |
1956 | * | |
1957 | * This is called with the rq lock held and interrupts off. It must | |
1958 | * be paired with a subsequent finish_task_switch after the context | |
1959 | * switch. | |
1960 | * | |
1961 | * prepare_task_switch sets up locking and calls architecture specific | |
1962 | * hooks. | |
1963 | */ | |
e107be36 AK |
1964 | static inline void |
1965 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1966 | struct task_struct *next) | |
4866cde0 | 1967 | { |
895dd92c | 1968 | trace_sched_switch(prev, next); |
43148951 | 1969 | sched_info_switch(rq, prev, next); |
fe4b04fa | 1970 | perf_event_task_sched_out(prev, next); |
e107be36 | 1971 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1972 | prepare_lock_switch(rq, next); |
1973 | prepare_arch_switch(next); | |
1974 | } | |
1975 | ||
1da177e4 LT |
1976 | /** |
1977 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1978 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1979 | * @prev: the thread we just switched away from. |
1980 | * | |
4866cde0 NP |
1981 | * finish_task_switch must be called after the context switch, paired |
1982 | * with a prepare_task_switch call before the context switch. | |
1983 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1984 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1985 | * |
1986 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1987 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1988 | * with the lock held can cause deadlocks; see schedule() for |
1989 | * details.) | |
1990 | */ | |
a9957449 | 1991 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1992 | __releases(rq->lock) |
1993 | { | |
1da177e4 | 1994 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1995 | long prev_state; |
1da177e4 LT |
1996 | |
1997 | rq->prev_mm = NULL; | |
1998 | ||
1999 | /* | |
2000 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2001 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2002 | * schedule one last time. The schedule call will never return, and |
2003 | * the scheduled task must drop that reference. | |
c394cc9f | 2004 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2005 | * still held, otherwise prev could be scheduled on another cpu, die |
2006 | * there before we look at prev->state, and then the reference would | |
2007 | * be dropped twice. | |
2008 | * Manfred Spraul <manfred@colorfullife.com> | |
2009 | */ | |
55a101f8 | 2010 | prev_state = prev->state; |
bf9fae9f | 2011 | vtime_task_switch(prev); |
4866cde0 | 2012 | finish_arch_switch(prev); |
a8d757ef | 2013 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2014 | finish_lock_switch(rq, prev); |
01f23e16 | 2015 | finish_arch_post_lock_switch(); |
e8fa1362 | 2016 | |
e107be36 | 2017 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2018 | if (mm) |
2019 | mmdrop(mm); | |
c394cc9f | 2020 | if (unlikely(prev_state == TASK_DEAD)) { |
f809ca9a MG |
2021 | task_numa_free(prev); |
2022 | ||
e6c390f2 DF |
2023 | if (prev->sched_class->task_dead) |
2024 | prev->sched_class->task_dead(prev); | |
2025 | ||
c6fd91f0 | 2026 | /* |
2027 | * Remove function-return probe instances associated with this | |
2028 | * task and put them back on the free list. | |
9761eea8 | 2029 | */ |
c6fd91f0 | 2030 | kprobe_flush_task(prev); |
1da177e4 | 2031 | put_task_struct(prev); |
c6fd91f0 | 2032 | } |
99e5ada9 FW |
2033 | |
2034 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2035 | } |
2036 | ||
3f029d3c GH |
2037 | #ifdef CONFIG_SMP |
2038 | ||
2039 | /* assumes rq->lock is held */ | |
2040 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2041 | { | |
2042 | if (prev->sched_class->pre_schedule) | |
2043 | prev->sched_class->pre_schedule(rq, prev); | |
2044 | } | |
2045 | ||
2046 | /* rq->lock is NOT held, but preemption is disabled */ | |
2047 | static inline void post_schedule(struct rq *rq) | |
2048 | { | |
2049 | if (rq->post_schedule) { | |
2050 | unsigned long flags; | |
2051 | ||
05fa785c | 2052 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2053 | if (rq->curr->sched_class->post_schedule) |
2054 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2055 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2056 | |
2057 | rq->post_schedule = 0; | |
2058 | } | |
2059 | } | |
2060 | ||
2061 | #else | |
da19ab51 | 2062 | |
3f029d3c GH |
2063 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2064 | { | |
2065 | } | |
2066 | ||
2067 | static inline void post_schedule(struct rq *rq) | |
2068 | { | |
1da177e4 LT |
2069 | } |
2070 | ||
3f029d3c GH |
2071 | #endif |
2072 | ||
1da177e4 LT |
2073 | /** |
2074 | * schedule_tail - first thing a freshly forked thread must call. | |
2075 | * @prev: the thread we just switched away from. | |
2076 | */ | |
36c8b586 | 2077 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2078 | __releases(rq->lock) |
2079 | { | |
70b97a7f IM |
2080 | struct rq *rq = this_rq(); |
2081 | ||
4866cde0 | 2082 | finish_task_switch(rq, prev); |
da19ab51 | 2083 | |
3f029d3c GH |
2084 | /* |
2085 | * FIXME: do we need to worry about rq being invalidated by the | |
2086 | * task_switch? | |
2087 | */ | |
2088 | post_schedule(rq); | |
70b97a7f | 2089 | |
4866cde0 NP |
2090 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2091 | /* In this case, finish_task_switch does not reenable preemption */ | |
2092 | preempt_enable(); | |
2093 | #endif | |
1da177e4 | 2094 | if (current->set_child_tid) |
b488893a | 2095 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2096 | } |
2097 | ||
2098 | /* | |
2099 | * context_switch - switch to the new MM and the new | |
2100 | * thread's register state. | |
2101 | */ | |
dd41f596 | 2102 | static inline void |
70b97a7f | 2103 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2104 | struct task_struct *next) |
1da177e4 | 2105 | { |
dd41f596 | 2106 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2107 | |
e107be36 | 2108 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2109 | |
dd41f596 IM |
2110 | mm = next->mm; |
2111 | oldmm = prev->active_mm; | |
9226d125 ZA |
2112 | /* |
2113 | * For paravirt, this is coupled with an exit in switch_to to | |
2114 | * combine the page table reload and the switch backend into | |
2115 | * one hypercall. | |
2116 | */ | |
224101ed | 2117 | arch_start_context_switch(prev); |
9226d125 | 2118 | |
31915ab4 | 2119 | if (!mm) { |
1da177e4 LT |
2120 | next->active_mm = oldmm; |
2121 | atomic_inc(&oldmm->mm_count); | |
2122 | enter_lazy_tlb(oldmm, next); | |
2123 | } else | |
2124 | switch_mm(oldmm, mm, next); | |
2125 | ||
31915ab4 | 2126 | if (!prev->mm) { |
1da177e4 | 2127 | prev->active_mm = NULL; |
1da177e4 LT |
2128 | rq->prev_mm = oldmm; |
2129 | } | |
3a5f5e48 IM |
2130 | /* |
2131 | * Since the runqueue lock will be released by the next | |
2132 | * task (which is an invalid locking op but in the case | |
2133 | * of the scheduler it's an obvious special-case), so we | |
2134 | * do an early lockdep release here: | |
2135 | */ | |
2136 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2137 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2138 | #endif |
1da177e4 | 2139 | |
91d1aa43 | 2140 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2141 | /* Here we just switch the register state and the stack. */ |
2142 | switch_to(prev, next, prev); | |
2143 | ||
dd41f596 IM |
2144 | barrier(); |
2145 | /* | |
2146 | * this_rq must be evaluated again because prev may have moved | |
2147 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2148 | * frame will be invalid. | |
2149 | */ | |
2150 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2151 | } |
2152 | ||
2153 | /* | |
1c3e8264 | 2154 | * nr_running and nr_context_switches: |
1da177e4 LT |
2155 | * |
2156 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2157 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2158 | */ |
2159 | unsigned long nr_running(void) | |
2160 | { | |
2161 | unsigned long i, sum = 0; | |
2162 | ||
2163 | for_each_online_cpu(i) | |
2164 | sum += cpu_rq(i)->nr_running; | |
2165 | ||
2166 | return sum; | |
f711f609 | 2167 | } |
1da177e4 | 2168 | |
1da177e4 | 2169 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2170 | { |
cc94abfc SR |
2171 | int i; |
2172 | unsigned long long sum = 0; | |
46cb4b7c | 2173 | |
0a945022 | 2174 | for_each_possible_cpu(i) |
1da177e4 | 2175 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2176 | |
1da177e4 LT |
2177 | return sum; |
2178 | } | |
483b4ee6 | 2179 | |
1da177e4 LT |
2180 | unsigned long nr_iowait(void) |
2181 | { | |
2182 | unsigned long i, sum = 0; | |
483b4ee6 | 2183 | |
0a945022 | 2184 | for_each_possible_cpu(i) |
1da177e4 | 2185 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2186 | |
1da177e4 LT |
2187 | return sum; |
2188 | } | |
483b4ee6 | 2189 | |
8c215bd3 | 2190 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2191 | { |
8c215bd3 | 2192 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2193 | return atomic_read(&this->nr_iowait); |
2194 | } | |
46cb4b7c | 2195 | |
dd41f596 | 2196 | #ifdef CONFIG_SMP |
8a0be9ef | 2197 | |
46cb4b7c | 2198 | /* |
38022906 PZ |
2199 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2200 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2201 | */ |
38022906 | 2202 | void sched_exec(void) |
46cb4b7c | 2203 | { |
38022906 | 2204 | struct task_struct *p = current; |
1da177e4 | 2205 | unsigned long flags; |
0017d735 | 2206 | int dest_cpu; |
46cb4b7c | 2207 | |
8f42ced9 | 2208 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2209 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2210 | if (dest_cpu == smp_processor_id()) |
2211 | goto unlock; | |
38022906 | 2212 | |
8f42ced9 | 2213 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2214 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2215 | |
8f42ced9 PZ |
2216 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2217 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2218 | return; |
2219 | } | |
0017d735 | 2220 | unlock: |
8f42ced9 | 2221 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2222 | } |
dd41f596 | 2223 | |
1da177e4 LT |
2224 | #endif |
2225 | ||
1da177e4 | 2226 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2227 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2228 | |
2229 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2230 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2231 | |
2232 | /* | |
c5f8d995 | 2233 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2234 | * @p in case that task is currently running. |
c5f8d995 HS |
2235 | * |
2236 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2237 | */ |
c5f8d995 HS |
2238 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2239 | { | |
2240 | u64 ns = 0; | |
2241 | ||
2242 | if (task_current(rq, p)) { | |
2243 | update_rq_clock(rq); | |
78becc27 | 2244 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2245 | if ((s64)ns < 0) |
2246 | ns = 0; | |
2247 | } | |
2248 | ||
2249 | return ns; | |
2250 | } | |
2251 | ||
bb34d92f | 2252 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2253 | { |
1da177e4 | 2254 | unsigned long flags; |
41b86e9c | 2255 | struct rq *rq; |
bb34d92f | 2256 | u64 ns = 0; |
48f24c4d | 2257 | |
41b86e9c | 2258 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2259 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2260 | task_rq_unlock(rq, p, &flags); |
1508487e | 2261 | |
c5f8d995 HS |
2262 | return ns; |
2263 | } | |
f06febc9 | 2264 | |
c5f8d995 HS |
2265 | /* |
2266 | * Return accounted runtime for the task. | |
2267 | * In case the task is currently running, return the runtime plus current's | |
2268 | * pending runtime that have not been accounted yet. | |
2269 | */ | |
2270 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2271 | { | |
2272 | unsigned long flags; | |
2273 | struct rq *rq; | |
2274 | u64 ns = 0; | |
2275 | ||
911b2898 PZ |
2276 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2277 | /* | |
2278 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2279 | * So we have a optimization chance when the task's delta_exec is 0. | |
2280 | * Reading ->on_cpu is racy, but this is ok. | |
2281 | * | |
2282 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2283 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2284 | * indistinguishable from the read occurring a few cycles earlier. | |
2285 | */ | |
2286 | if (!p->on_cpu) | |
2287 | return p->se.sum_exec_runtime; | |
2288 | #endif | |
2289 | ||
c5f8d995 HS |
2290 | rq = task_rq_lock(p, &flags); |
2291 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2292 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2293 | |
2294 | return ns; | |
2295 | } | |
48f24c4d | 2296 | |
7835b98b CL |
2297 | /* |
2298 | * This function gets called by the timer code, with HZ frequency. | |
2299 | * We call it with interrupts disabled. | |
7835b98b CL |
2300 | */ |
2301 | void scheduler_tick(void) | |
2302 | { | |
7835b98b CL |
2303 | int cpu = smp_processor_id(); |
2304 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2305 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2306 | |
2307 | sched_clock_tick(); | |
dd41f596 | 2308 | |
05fa785c | 2309 | raw_spin_lock(&rq->lock); |
3e51f33f | 2310 | update_rq_clock(rq); |
fa85ae24 | 2311 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2312 | update_cpu_load_active(rq); |
05fa785c | 2313 | raw_spin_unlock(&rq->lock); |
7835b98b | 2314 | |
e9d2b064 | 2315 | perf_event_task_tick(); |
e220d2dc | 2316 | |
e418e1c2 | 2317 | #ifdef CONFIG_SMP |
6eb57e0d | 2318 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2319 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2320 | #endif |
265f22a9 | 2321 | rq_last_tick_reset(rq); |
1da177e4 LT |
2322 | } |
2323 | ||
265f22a9 FW |
2324 | #ifdef CONFIG_NO_HZ_FULL |
2325 | /** | |
2326 | * scheduler_tick_max_deferment | |
2327 | * | |
2328 | * Keep at least one tick per second when a single | |
2329 | * active task is running because the scheduler doesn't | |
2330 | * yet completely support full dynticks environment. | |
2331 | * | |
2332 | * This makes sure that uptime, CFS vruntime, load | |
2333 | * balancing, etc... continue to move forward, even | |
2334 | * with a very low granularity. | |
e69f6186 YB |
2335 | * |
2336 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2337 | */ |
2338 | u64 scheduler_tick_max_deferment(void) | |
2339 | { | |
2340 | struct rq *rq = this_rq(); | |
2341 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2342 | ||
2343 | next = rq->last_sched_tick + HZ; | |
2344 | ||
2345 | if (time_before_eq(next, now)) | |
2346 | return 0; | |
2347 | ||
2348 | return jiffies_to_usecs(next - now) * NSEC_PER_USEC; | |
1da177e4 | 2349 | } |
265f22a9 | 2350 | #endif |
1da177e4 | 2351 | |
132380a0 | 2352 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2353 | { |
2354 | if (in_lock_functions(addr)) { | |
2355 | addr = CALLER_ADDR2; | |
2356 | if (in_lock_functions(addr)) | |
2357 | addr = CALLER_ADDR3; | |
2358 | } | |
2359 | return addr; | |
2360 | } | |
1da177e4 | 2361 | |
7e49fcce SR |
2362 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2363 | defined(CONFIG_PREEMPT_TRACER)) | |
2364 | ||
bdb43806 | 2365 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2366 | { |
6cd8a4bb | 2367 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2368 | /* |
2369 | * Underflow? | |
2370 | */ | |
9a11b49a IM |
2371 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2372 | return; | |
6cd8a4bb | 2373 | #endif |
bdb43806 | 2374 | __preempt_count_add(val); |
6cd8a4bb | 2375 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2376 | /* |
2377 | * Spinlock count overflowing soon? | |
2378 | */ | |
33859f7f MOS |
2379 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2380 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2381 | #endif |
2382 | if (preempt_count() == val) | |
2383 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 | 2384 | } |
bdb43806 | 2385 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2386 | |
bdb43806 | 2387 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2388 | { |
6cd8a4bb | 2389 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2390 | /* |
2391 | * Underflow? | |
2392 | */ | |
01e3eb82 | 2393 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2394 | return; |
1da177e4 LT |
2395 | /* |
2396 | * Is the spinlock portion underflowing? | |
2397 | */ | |
9a11b49a IM |
2398 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2399 | !(preempt_count() & PREEMPT_MASK))) | |
2400 | return; | |
6cd8a4bb | 2401 | #endif |
9a11b49a | 2402 | |
6cd8a4bb SR |
2403 | if (preempt_count() == val) |
2404 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2405 | __preempt_count_sub(val); |
1da177e4 | 2406 | } |
bdb43806 | 2407 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2408 | |
2409 | #endif | |
2410 | ||
2411 | /* | |
dd41f596 | 2412 | * Print scheduling while atomic bug: |
1da177e4 | 2413 | */ |
dd41f596 | 2414 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2415 | { |
664dfa65 DJ |
2416 | if (oops_in_progress) |
2417 | return; | |
2418 | ||
3df0fc5b PZ |
2419 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2420 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2421 | |
dd41f596 | 2422 | debug_show_held_locks(prev); |
e21f5b15 | 2423 | print_modules(); |
dd41f596 IM |
2424 | if (irqs_disabled()) |
2425 | print_irqtrace_events(prev); | |
6135fc1e | 2426 | dump_stack(); |
373d4d09 | 2427 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2428 | } |
1da177e4 | 2429 | |
dd41f596 IM |
2430 | /* |
2431 | * Various schedule()-time debugging checks and statistics: | |
2432 | */ | |
2433 | static inline void schedule_debug(struct task_struct *prev) | |
2434 | { | |
1da177e4 | 2435 | /* |
41a2d6cf | 2436 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2437 | * schedule() atomically, we ignore that path. Otherwise whine |
2438 | * if we are scheduling when we should not. | |
1da177e4 | 2439 | */ |
192301e7 | 2440 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2441 | __schedule_bug(prev); |
b3fbab05 | 2442 | rcu_sleep_check(); |
dd41f596 | 2443 | |
1da177e4 LT |
2444 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2445 | ||
2d72376b | 2446 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2447 | } |
2448 | ||
6cecd084 | 2449 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2450 | { |
61eadef6 | 2451 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2452 | update_rq_clock(rq); |
6cecd084 | 2453 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2454 | } |
2455 | ||
dd41f596 IM |
2456 | /* |
2457 | * Pick up the highest-prio task: | |
2458 | */ | |
2459 | static inline struct task_struct * | |
b67802ea | 2460 | pick_next_task(struct rq *rq) |
dd41f596 | 2461 | { |
5522d5d5 | 2462 | const struct sched_class *class; |
dd41f596 | 2463 | struct task_struct *p; |
1da177e4 LT |
2464 | |
2465 | /* | |
dd41f596 IM |
2466 | * Optimization: we know that if all tasks are in |
2467 | * the fair class we can call that function directly: | |
1da177e4 | 2468 | */ |
953bfcd1 | 2469 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2470 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2471 | if (likely(p)) |
2472 | return p; | |
1da177e4 LT |
2473 | } |
2474 | ||
34f971f6 | 2475 | for_each_class(class) { |
fb8d4724 | 2476 | p = class->pick_next_task(rq); |
dd41f596 IM |
2477 | if (p) |
2478 | return p; | |
dd41f596 | 2479 | } |
34f971f6 PZ |
2480 | |
2481 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2482 | } |
1da177e4 | 2483 | |
dd41f596 | 2484 | /* |
c259e01a | 2485 | * __schedule() is the main scheduler function. |
edde96ea PE |
2486 | * |
2487 | * The main means of driving the scheduler and thus entering this function are: | |
2488 | * | |
2489 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2490 | * | |
2491 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2492 | * paths. For example, see arch/x86/entry_64.S. | |
2493 | * | |
2494 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2495 | * interrupt handler scheduler_tick(). | |
2496 | * | |
2497 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2498 | * task to the run-queue and that's it. | |
2499 | * | |
2500 | * Now, if the new task added to the run-queue preempts the current | |
2501 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2502 | * called on the nearest possible occasion: | |
2503 | * | |
2504 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2505 | * | |
2506 | * - in syscall or exception context, at the next outmost | |
2507 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2508 | * spin_unlock()!) | |
2509 | * | |
2510 | * - in IRQ context, return from interrupt-handler to | |
2511 | * preemptible context | |
2512 | * | |
2513 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2514 | * then at the next: | |
2515 | * | |
2516 | * - cond_resched() call | |
2517 | * - explicit schedule() call | |
2518 | * - return from syscall or exception to user-space | |
2519 | * - return from interrupt-handler to user-space | |
dd41f596 | 2520 | */ |
c259e01a | 2521 | static void __sched __schedule(void) |
dd41f596 IM |
2522 | { |
2523 | struct task_struct *prev, *next; | |
67ca7bde | 2524 | unsigned long *switch_count; |
dd41f596 | 2525 | struct rq *rq; |
31656519 | 2526 | int cpu; |
dd41f596 | 2527 | |
ff743345 PZ |
2528 | need_resched: |
2529 | preempt_disable(); | |
dd41f596 IM |
2530 | cpu = smp_processor_id(); |
2531 | rq = cpu_rq(cpu); | |
25502a6c | 2532 | rcu_note_context_switch(cpu); |
dd41f596 | 2533 | prev = rq->curr; |
dd41f596 | 2534 | |
dd41f596 | 2535 | schedule_debug(prev); |
1da177e4 | 2536 | |
31656519 | 2537 | if (sched_feat(HRTICK)) |
f333fdc9 | 2538 | hrtick_clear(rq); |
8f4d37ec | 2539 | |
e0acd0a6 ON |
2540 | /* |
2541 | * Make sure that signal_pending_state()->signal_pending() below | |
2542 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2543 | * done by the caller to avoid the race with signal_wake_up(). | |
2544 | */ | |
2545 | smp_mb__before_spinlock(); | |
05fa785c | 2546 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2547 | |
246d86b5 | 2548 | switch_count = &prev->nivcsw; |
1da177e4 | 2549 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2550 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2551 | prev->state = TASK_RUNNING; |
21aa9af0 | 2552 | } else { |
2acca55e PZ |
2553 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2554 | prev->on_rq = 0; | |
2555 | ||
21aa9af0 | 2556 | /* |
2acca55e PZ |
2557 | * If a worker went to sleep, notify and ask workqueue |
2558 | * whether it wants to wake up a task to maintain | |
2559 | * concurrency. | |
21aa9af0 TH |
2560 | */ |
2561 | if (prev->flags & PF_WQ_WORKER) { | |
2562 | struct task_struct *to_wakeup; | |
2563 | ||
2564 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2565 | if (to_wakeup) | |
2566 | try_to_wake_up_local(to_wakeup); | |
2567 | } | |
21aa9af0 | 2568 | } |
dd41f596 | 2569 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2570 | } |
2571 | ||
3f029d3c | 2572 | pre_schedule(rq, prev); |
f65eda4f | 2573 | |
dd41f596 | 2574 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2575 | idle_balance(cpu, rq); |
1da177e4 | 2576 | |
df1c99d4 | 2577 | put_prev_task(rq, prev); |
b67802ea | 2578 | next = pick_next_task(rq); |
f26f9aff | 2579 | clear_tsk_need_resched(prev); |
f27dde8d | 2580 | clear_preempt_need_resched(); |
f26f9aff | 2581 | rq->skip_clock_update = 0; |
1da177e4 | 2582 | |
1da177e4 | 2583 | if (likely(prev != next)) { |
1da177e4 LT |
2584 | rq->nr_switches++; |
2585 | rq->curr = next; | |
2586 | ++*switch_count; | |
2587 | ||
dd41f596 | 2588 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2589 | /* |
246d86b5 ON |
2590 | * The context switch have flipped the stack from under us |
2591 | * and restored the local variables which were saved when | |
2592 | * this task called schedule() in the past. prev == current | |
2593 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2594 | */ |
2595 | cpu = smp_processor_id(); | |
2596 | rq = cpu_rq(cpu); | |
1da177e4 | 2597 | } else |
05fa785c | 2598 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2599 | |
3f029d3c | 2600 | post_schedule(rq); |
1da177e4 | 2601 | |
ba74c144 | 2602 | sched_preempt_enable_no_resched(); |
ff743345 | 2603 | if (need_resched()) |
1da177e4 LT |
2604 | goto need_resched; |
2605 | } | |
c259e01a | 2606 | |
9c40cef2 TG |
2607 | static inline void sched_submit_work(struct task_struct *tsk) |
2608 | { | |
3c7d5184 | 2609 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2610 | return; |
2611 | /* | |
2612 | * If we are going to sleep and we have plugged IO queued, | |
2613 | * make sure to submit it to avoid deadlocks. | |
2614 | */ | |
2615 | if (blk_needs_flush_plug(tsk)) | |
2616 | blk_schedule_flush_plug(tsk); | |
2617 | } | |
2618 | ||
6ebbe7a0 | 2619 | asmlinkage void __sched schedule(void) |
c259e01a | 2620 | { |
9c40cef2 TG |
2621 | struct task_struct *tsk = current; |
2622 | ||
2623 | sched_submit_work(tsk); | |
c259e01a TG |
2624 | __schedule(); |
2625 | } | |
1da177e4 LT |
2626 | EXPORT_SYMBOL(schedule); |
2627 | ||
91d1aa43 | 2628 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2629 | asmlinkage void __sched schedule_user(void) |
2630 | { | |
2631 | /* | |
2632 | * If we come here after a random call to set_need_resched(), | |
2633 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2634 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2635 | * we find a better solution. | |
2636 | */ | |
91d1aa43 | 2637 | user_exit(); |
20ab65e3 | 2638 | schedule(); |
91d1aa43 | 2639 | user_enter(); |
20ab65e3 FW |
2640 | } |
2641 | #endif | |
2642 | ||
c5491ea7 TG |
2643 | /** |
2644 | * schedule_preempt_disabled - called with preemption disabled | |
2645 | * | |
2646 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2647 | */ | |
2648 | void __sched schedule_preempt_disabled(void) | |
2649 | { | |
ba74c144 | 2650 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2651 | schedule(); |
2652 | preempt_disable(); | |
2653 | } | |
2654 | ||
1da177e4 LT |
2655 | #ifdef CONFIG_PREEMPT |
2656 | /* | |
2ed6e34f | 2657 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2658 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2659 | * occur there and call schedule directly. |
2660 | */ | |
d1f74e20 | 2661 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 | 2662 | { |
1da177e4 LT |
2663 | /* |
2664 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2665 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2666 | */ |
fbb00b56 | 2667 | if (likely(!preemptible())) |
1da177e4 LT |
2668 | return; |
2669 | ||
3a5c359a | 2670 | do { |
bdb43806 | 2671 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2672 | __schedule(); |
bdb43806 | 2673 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2674 | |
3a5c359a AK |
2675 | /* |
2676 | * Check again in case we missed a preemption opportunity | |
2677 | * between schedule and now. | |
2678 | */ | |
2679 | barrier(); | |
5ed0cec0 | 2680 | } while (need_resched()); |
1da177e4 | 2681 | } |
1da177e4 | 2682 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2683 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2684 | |
2685 | /* | |
2ed6e34f | 2686 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2687 | * off of irq context. |
2688 | * Note, that this is called and return with irqs disabled. This will | |
2689 | * protect us against recursive calling from irq. | |
2690 | */ | |
2691 | asmlinkage void __sched preempt_schedule_irq(void) | |
2692 | { | |
b22366cd | 2693 | enum ctx_state prev_state; |
6478d880 | 2694 | |
2ed6e34f | 2695 | /* Catch callers which need to be fixed */ |
f27dde8d | 2696 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2697 | |
b22366cd FW |
2698 | prev_state = exception_enter(); |
2699 | ||
3a5c359a | 2700 | do { |
bdb43806 | 2701 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2702 | local_irq_enable(); |
c259e01a | 2703 | __schedule(); |
3a5c359a | 2704 | local_irq_disable(); |
bdb43806 | 2705 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2706 | |
3a5c359a AK |
2707 | /* |
2708 | * Check again in case we missed a preemption opportunity | |
2709 | * between schedule and now. | |
2710 | */ | |
2711 | barrier(); | |
5ed0cec0 | 2712 | } while (need_resched()); |
b22366cd FW |
2713 | |
2714 | exception_exit(prev_state); | |
1da177e4 LT |
2715 | } |
2716 | ||
63859d4f | 2717 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2718 | void *key) |
1da177e4 | 2719 | { |
63859d4f | 2720 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2721 | } |
1da177e4 LT |
2722 | EXPORT_SYMBOL(default_wake_function); |
2723 | ||
8cbbe86d AK |
2724 | static long __sched |
2725 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 2726 | { |
0fec171c IM |
2727 | unsigned long flags; |
2728 | wait_queue_t wait; | |
2729 | ||
2730 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 2731 | |
8cbbe86d | 2732 | __set_current_state(state); |
1da177e4 | 2733 | |
8cbbe86d AK |
2734 | spin_lock_irqsave(&q->lock, flags); |
2735 | __add_wait_queue(q, &wait); | |
2736 | spin_unlock(&q->lock); | |
2737 | timeout = schedule_timeout(timeout); | |
2738 | spin_lock_irq(&q->lock); | |
2739 | __remove_wait_queue(q, &wait); | |
2740 | spin_unlock_irqrestore(&q->lock, flags); | |
2741 | ||
2742 | return timeout; | |
2743 | } | |
2744 | ||
2745 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
2746 | { | |
2747 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 2748 | } |
1da177e4 LT |
2749 | EXPORT_SYMBOL(interruptible_sleep_on); |
2750 | ||
0fec171c | 2751 | long __sched |
95cdf3b7 | 2752 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2753 | { |
8cbbe86d | 2754 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 2755 | } |
1da177e4 LT |
2756 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
2757 | ||
0fec171c | 2758 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 2759 | { |
8cbbe86d | 2760 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 2761 | } |
1da177e4 LT |
2762 | EXPORT_SYMBOL(sleep_on); |
2763 | ||
0fec171c | 2764 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2765 | { |
8cbbe86d | 2766 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 2767 | } |
1da177e4 LT |
2768 | EXPORT_SYMBOL(sleep_on_timeout); |
2769 | ||
b29739f9 IM |
2770 | #ifdef CONFIG_RT_MUTEXES |
2771 | ||
2772 | /* | |
2773 | * rt_mutex_setprio - set the current priority of a task | |
2774 | * @p: task | |
2775 | * @prio: prio value (kernel-internal form) | |
2776 | * | |
2777 | * This function changes the 'effective' priority of a task. It does | |
2778 | * not touch ->normal_prio like __setscheduler(). | |
2779 | * | |
2780 | * Used by the rt_mutex code to implement priority inheritance logic. | |
2781 | */ | |
36c8b586 | 2782 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2783 | { |
83b699ed | 2784 | int oldprio, on_rq, running; |
70b97a7f | 2785 | struct rq *rq; |
83ab0aa0 | 2786 | const struct sched_class *prev_class; |
b29739f9 | 2787 | |
aab03e05 | 2788 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2789 | |
0122ec5b | 2790 | rq = __task_rq_lock(p); |
b29739f9 | 2791 | |
1c4dd99b TG |
2792 | /* |
2793 | * Idle task boosting is a nono in general. There is one | |
2794 | * exception, when PREEMPT_RT and NOHZ is active: | |
2795 | * | |
2796 | * The idle task calls get_next_timer_interrupt() and holds | |
2797 | * the timer wheel base->lock on the CPU and another CPU wants | |
2798 | * to access the timer (probably to cancel it). We can safely | |
2799 | * ignore the boosting request, as the idle CPU runs this code | |
2800 | * with interrupts disabled and will complete the lock | |
2801 | * protected section without being interrupted. So there is no | |
2802 | * real need to boost. | |
2803 | */ | |
2804 | if (unlikely(p == rq->idle)) { | |
2805 | WARN_ON(p != rq->curr); | |
2806 | WARN_ON(p->pi_blocked_on); | |
2807 | goto out_unlock; | |
2808 | } | |
2809 | ||
a8027073 | 2810 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 2811 | oldprio = p->prio; |
83ab0aa0 | 2812 | prev_class = p->sched_class; |
fd2f4419 | 2813 | on_rq = p->on_rq; |
051a1d1a | 2814 | running = task_current(rq, p); |
0e1f3483 | 2815 | if (on_rq) |
69be72c1 | 2816 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2817 | if (running) |
2818 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2819 | |
aab03e05 DF |
2820 | if (dl_prio(prio)) |
2821 | p->sched_class = &dl_sched_class; | |
2822 | else if (rt_prio(prio)) | |
dd41f596 IM |
2823 | p->sched_class = &rt_sched_class; |
2824 | else | |
2825 | p->sched_class = &fair_sched_class; | |
2826 | ||
b29739f9 IM |
2827 | p->prio = prio; |
2828 | ||
0e1f3483 HS |
2829 | if (running) |
2830 | p->sched_class->set_curr_task(rq); | |
da7a735e | 2831 | if (on_rq) |
371fd7e7 | 2832 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 2833 | |
da7a735e | 2834 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 2835 | out_unlock: |
0122ec5b | 2836 | __task_rq_unlock(rq); |
b29739f9 | 2837 | } |
b29739f9 | 2838 | #endif |
d50dde5a | 2839 | |
36c8b586 | 2840 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 2841 | { |
dd41f596 | 2842 | int old_prio, delta, on_rq; |
1da177e4 | 2843 | unsigned long flags; |
70b97a7f | 2844 | struct rq *rq; |
1da177e4 LT |
2845 | |
2846 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
2847 | return; | |
2848 | /* | |
2849 | * We have to be careful, if called from sys_setpriority(), | |
2850 | * the task might be in the middle of scheduling on another CPU. | |
2851 | */ | |
2852 | rq = task_rq_lock(p, &flags); | |
2853 | /* | |
2854 | * The RT priorities are set via sched_setscheduler(), but we still | |
2855 | * allow the 'normal' nice value to be set - but as expected | |
2856 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 2857 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 2858 | */ |
aab03e05 | 2859 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
2860 | p->static_prio = NICE_TO_PRIO(nice); |
2861 | goto out_unlock; | |
2862 | } | |
fd2f4419 | 2863 | on_rq = p->on_rq; |
c09595f6 | 2864 | if (on_rq) |
69be72c1 | 2865 | dequeue_task(rq, p, 0); |
1da177e4 | 2866 | |
1da177e4 | 2867 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 2868 | set_load_weight(p); |
b29739f9 IM |
2869 | old_prio = p->prio; |
2870 | p->prio = effective_prio(p); | |
2871 | delta = p->prio - old_prio; | |
1da177e4 | 2872 | |
dd41f596 | 2873 | if (on_rq) { |
371fd7e7 | 2874 | enqueue_task(rq, p, 0); |
1da177e4 | 2875 | /* |
d5f9f942 AM |
2876 | * If the task increased its priority or is running and |
2877 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 2878 | */ |
d5f9f942 | 2879 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
2880 | resched_task(rq->curr); |
2881 | } | |
2882 | out_unlock: | |
0122ec5b | 2883 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 2884 | } |
1da177e4 LT |
2885 | EXPORT_SYMBOL(set_user_nice); |
2886 | ||
e43379f1 MM |
2887 | /* |
2888 | * can_nice - check if a task can reduce its nice value | |
2889 | * @p: task | |
2890 | * @nice: nice value | |
2891 | */ | |
36c8b586 | 2892 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 2893 | { |
024f4747 MM |
2894 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
2895 | int nice_rlim = 20 - nice; | |
48f24c4d | 2896 | |
78d7d407 | 2897 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
2898 | capable(CAP_SYS_NICE)); |
2899 | } | |
2900 | ||
1da177e4 LT |
2901 | #ifdef __ARCH_WANT_SYS_NICE |
2902 | ||
2903 | /* | |
2904 | * sys_nice - change the priority of the current process. | |
2905 | * @increment: priority increment | |
2906 | * | |
2907 | * sys_setpriority is a more generic, but much slower function that | |
2908 | * does similar things. | |
2909 | */ | |
5add95d4 | 2910 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 2911 | { |
48f24c4d | 2912 | long nice, retval; |
1da177e4 LT |
2913 | |
2914 | /* | |
2915 | * Setpriority might change our priority at the same moment. | |
2916 | * We don't have to worry. Conceptually one call occurs first | |
2917 | * and we have a single winner. | |
2918 | */ | |
e43379f1 MM |
2919 | if (increment < -40) |
2920 | increment = -40; | |
1da177e4 LT |
2921 | if (increment > 40) |
2922 | increment = 40; | |
2923 | ||
2b8f836f | 2924 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
2925 | if (nice < -20) |
2926 | nice = -20; | |
2927 | if (nice > 19) | |
2928 | nice = 19; | |
2929 | ||
e43379f1 MM |
2930 | if (increment < 0 && !can_nice(current, nice)) |
2931 | return -EPERM; | |
2932 | ||
1da177e4 LT |
2933 | retval = security_task_setnice(current, nice); |
2934 | if (retval) | |
2935 | return retval; | |
2936 | ||
2937 | set_user_nice(current, nice); | |
2938 | return 0; | |
2939 | } | |
2940 | ||
2941 | #endif | |
2942 | ||
2943 | /** | |
2944 | * task_prio - return the priority value of a given task. | |
2945 | * @p: the task in question. | |
2946 | * | |
e69f6186 | 2947 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
2948 | * RT tasks are offset by -200. Normal tasks are centered |
2949 | * around 0, value goes from -16 to +15. | |
2950 | */ | |
36c8b586 | 2951 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
2952 | { |
2953 | return p->prio - MAX_RT_PRIO; | |
2954 | } | |
2955 | ||
2956 | /** | |
2957 | * task_nice - return the nice value of a given task. | |
2958 | * @p: the task in question. | |
e69f6186 YB |
2959 | * |
2960 | * Return: The nice value [ -20 ... 0 ... 19 ]. | |
1da177e4 | 2961 | */ |
36c8b586 | 2962 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
2963 | { |
2964 | return TASK_NICE(p); | |
2965 | } | |
150d8bed | 2966 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
2967 | |
2968 | /** | |
2969 | * idle_cpu - is a given cpu idle currently? | |
2970 | * @cpu: the processor in question. | |
e69f6186 YB |
2971 | * |
2972 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
2973 | */ |
2974 | int idle_cpu(int cpu) | |
2975 | { | |
908a3283 TG |
2976 | struct rq *rq = cpu_rq(cpu); |
2977 | ||
2978 | if (rq->curr != rq->idle) | |
2979 | return 0; | |
2980 | ||
2981 | if (rq->nr_running) | |
2982 | return 0; | |
2983 | ||
2984 | #ifdef CONFIG_SMP | |
2985 | if (!llist_empty(&rq->wake_list)) | |
2986 | return 0; | |
2987 | #endif | |
2988 | ||
2989 | return 1; | |
1da177e4 LT |
2990 | } |
2991 | ||
1da177e4 LT |
2992 | /** |
2993 | * idle_task - return the idle task for a given cpu. | |
2994 | * @cpu: the processor in question. | |
e69f6186 YB |
2995 | * |
2996 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 2997 | */ |
36c8b586 | 2998 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
2999 | { |
3000 | return cpu_rq(cpu)->idle; | |
3001 | } | |
3002 | ||
3003 | /** | |
3004 | * find_process_by_pid - find a process with a matching PID value. | |
3005 | * @pid: the pid in question. | |
e69f6186 YB |
3006 | * |
3007 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3008 | */ |
a9957449 | 3009 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3010 | { |
228ebcbe | 3011 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3012 | } |
3013 | ||
aab03e05 DF |
3014 | /* |
3015 | * This function initializes the sched_dl_entity of a newly becoming | |
3016 | * SCHED_DEADLINE task. | |
3017 | * | |
3018 | * Only the static values are considered here, the actual runtime and the | |
3019 | * absolute deadline will be properly calculated when the task is enqueued | |
3020 | * for the first time with its new policy. | |
3021 | */ | |
3022 | static void | |
3023 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3024 | { | |
3025 | struct sched_dl_entity *dl_se = &p->dl; | |
3026 | ||
3027 | init_dl_task_timer(dl_se); | |
3028 | dl_se->dl_runtime = attr->sched_runtime; | |
3029 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3030 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 DF |
3031 | dl_se->flags = attr->sched_flags; |
3032 | dl_se->dl_throttled = 0; | |
3033 | dl_se->dl_new = 1; | |
3034 | } | |
3035 | ||
d50dde5a DF |
3036 | /* Actually do priority change: must hold pi & rq lock. */ |
3037 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3038 | const struct sched_attr *attr) | |
1da177e4 | 3039 | { |
d50dde5a DF |
3040 | int policy = attr->sched_policy; |
3041 | ||
1da177e4 | 3042 | p->policy = policy; |
d50dde5a | 3043 | |
aab03e05 DF |
3044 | if (dl_policy(policy)) |
3045 | __setparam_dl(p, attr); | |
3046 | else if (rt_policy(policy)) | |
d50dde5a DF |
3047 | p->rt_priority = attr->sched_priority; |
3048 | else | |
3049 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); | |
3050 | ||
b29739f9 | 3051 | p->normal_prio = normal_prio(p); |
b29739f9 | 3052 | p->prio = rt_mutex_getprio(p); |
d50dde5a | 3053 | |
aab03e05 DF |
3054 | if (dl_prio(p->prio)) |
3055 | p->sched_class = &dl_sched_class; | |
3056 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3057 | p->sched_class = &rt_sched_class; |
3058 | else | |
3059 | p->sched_class = &fair_sched_class; | |
d50dde5a | 3060 | |
2dd73a4f | 3061 | set_load_weight(p); |
1da177e4 | 3062 | } |
aab03e05 DF |
3063 | |
3064 | static void | |
3065 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3066 | { | |
3067 | struct sched_dl_entity *dl_se = &p->dl; | |
3068 | ||
3069 | attr->sched_priority = p->rt_priority; | |
3070 | attr->sched_runtime = dl_se->dl_runtime; | |
3071 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3072 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3073 | attr->sched_flags = dl_se->flags; |
3074 | } | |
3075 | ||
3076 | /* | |
3077 | * This function validates the new parameters of a -deadline task. | |
3078 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 HG |
3079 | * than the runtime, as well as the period of being zero or |
3080 | * greater than deadline. | |
aab03e05 DF |
3081 | */ |
3082 | static bool | |
3083 | __checkparam_dl(const struct sched_attr *attr) | |
3084 | { | |
3085 | return attr && attr->sched_deadline != 0 && | |
755378a4 HG |
3086 | (attr->sched_period == 0 || |
3087 | (s64)(attr->sched_period - attr->sched_deadline) >= 0) && | |
3088 | (s64)(attr->sched_deadline - attr->sched_runtime ) >= 0; | |
aab03e05 DF |
3089 | } |
3090 | ||
c69e8d9c DH |
3091 | /* |
3092 | * check the target process has a UID that matches the current process's | |
3093 | */ | |
3094 | static bool check_same_owner(struct task_struct *p) | |
3095 | { | |
3096 | const struct cred *cred = current_cred(), *pcred; | |
3097 | bool match; | |
3098 | ||
3099 | rcu_read_lock(); | |
3100 | pcred = __task_cred(p); | |
9c806aa0 EB |
3101 | match = (uid_eq(cred->euid, pcred->euid) || |
3102 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3103 | rcu_read_unlock(); |
3104 | return match; | |
3105 | } | |
3106 | ||
d50dde5a DF |
3107 | static int __sched_setscheduler(struct task_struct *p, |
3108 | const struct sched_attr *attr, | |
3109 | bool user) | |
1da177e4 | 3110 | { |
83b699ed | 3111 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3112 | int policy = attr->sched_policy; |
1da177e4 | 3113 | unsigned long flags; |
83ab0aa0 | 3114 | const struct sched_class *prev_class; |
70b97a7f | 3115 | struct rq *rq; |
ca94c442 | 3116 | int reset_on_fork; |
1da177e4 | 3117 | |
66e5393a SR |
3118 | /* may grab non-irq protected spin_locks */ |
3119 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3120 | recheck: |
3121 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3122 | if (policy < 0) { |
3123 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3124 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3125 | } else { |
3126 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3127 | policy &= ~SCHED_RESET_ON_FORK; | |
3128 | ||
aab03e05 DF |
3129 | if (policy != SCHED_DEADLINE && |
3130 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3131 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3132 | policy != SCHED_IDLE) | |
3133 | return -EINVAL; | |
3134 | } | |
3135 | ||
1da177e4 LT |
3136 | /* |
3137 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3138 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3139 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3140 | */ |
d50dde5a DF |
3141 | if (attr->sched_priority < 0 || |
3142 | (p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || | |
3143 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) | |
1da177e4 | 3144 | return -EINVAL; |
aab03e05 DF |
3145 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3146 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3147 | return -EINVAL; |
3148 | ||
37e4ab3f OC |
3149 | /* |
3150 | * Allow unprivileged RT tasks to decrease priority: | |
3151 | */ | |
961ccddd | 3152 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a DF |
3153 | if (fair_policy(policy)) { |
3154 | if (!can_nice(p, attr->sched_nice)) | |
3155 | return -EPERM; | |
3156 | } | |
3157 | ||
e05606d3 | 3158 | if (rt_policy(policy)) { |
a44702e8 ON |
3159 | unsigned long rlim_rtprio = |
3160 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3161 | |
3162 | /* can't set/change the rt policy */ | |
3163 | if (policy != p->policy && !rlim_rtprio) | |
3164 | return -EPERM; | |
3165 | ||
3166 | /* can't increase priority */ | |
d50dde5a DF |
3167 | if (attr->sched_priority > p->rt_priority && |
3168 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3169 | return -EPERM; |
3170 | } | |
c02aa73b | 3171 | |
dd41f596 | 3172 | /* |
c02aa73b DH |
3173 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3174 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3175 | */ |
c02aa73b DH |
3176 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3177 | if (!can_nice(p, TASK_NICE(p))) | |
3178 | return -EPERM; | |
3179 | } | |
5fe1d75f | 3180 | |
37e4ab3f | 3181 | /* can't change other user's priorities */ |
c69e8d9c | 3182 | if (!check_same_owner(p)) |
37e4ab3f | 3183 | return -EPERM; |
ca94c442 LP |
3184 | |
3185 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3186 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3187 | return -EPERM; | |
37e4ab3f | 3188 | } |
1da177e4 | 3189 | |
725aad24 | 3190 | if (user) { |
b0ae1981 | 3191 | retval = security_task_setscheduler(p); |
725aad24 JF |
3192 | if (retval) |
3193 | return retval; | |
3194 | } | |
3195 | ||
b29739f9 IM |
3196 | /* |
3197 | * make sure no PI-waiters arrive (or leave) while we are | |
3198 | * changing the priority of the task: | |
0122ec5b | 3199 | * |
25985edc | 3200 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3201 | * runqueue lock must be held. |
3202 | */ | |
0122ec5b | 3203 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3204 | |
34f971f6 PZ |
3205 | /* |
3206 | * Changing the policy of the stop threads its a very bad idea | |
3207 | */ | |
3208 | if (p == rq->stop) { | |
0122ec5b | 3209 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3210 | return -EINVAL; |
3211 | } | |
3212 | ||
a51e9198 DF |
3213 | /* |
3214 | * If not changing anything there's no need to proceed further: | |
3215 | */ | |
d50dde5a DF |
3216 | if (unlikely(policy == p->policy)) { |
3217 | if (fair_policy(policy) && attr->sched_nice != TASK_NICE(p)) | |
3218 | goto change; | |
3219 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3220 | goto change; | |
aab03e05 DF |
3221 | if (dl_policy(policy)) |
3222 | goto change; | |
d50dde5a | 3223 | |
45afb173 | 3224 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3225 | return 0; |
3226 | } | |
d50dde5a | 3227 | change: |
a51e9198 | 3228 | |
dc61b1d6 PZ |
3229 | #ifdef CONFIG_RT_GROUP_SCHED |
3230 | if (user) { | |
3231 | /* | |
3232 | * Do not allow realtime tasks into groups that have no runtime | |
3233 | * assigned. | |
3234 | */ | |
3235 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3236 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3237 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3238 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3239 | return -EPERM; |
3240 | } | |
3241 | } | |
3242 | #endif | |
3243 | ||
1da177e4 LT |
3244 | /* recheck policy now with rq lock held */ |
3245 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3246 | policy = oldpolicy = -1; | |
0122ec5b | 3247 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3248 | goto recheck; |
3249 | } | |
fd2f4419 | 3250 | on_rq = p->on_rq; |
051a1d1a | 3251 | running = task_current(rq, p); |
0e1f3483 | 3252 | if (on_rq) |
4ca9b72b | 3253 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3254 | if (running) |
3255 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3256 | |
ca94c442 LP |
3257 | p->sched_reset_on_fork = reset_on_fork; |
3258 | ||
1da177e4 | 3259 | oldprio = p->prio; |
83ab0aa0 | 3260 | prev_class = p->sched_class; |
d50dde5a | 3261 | __setscheduler(rq, p, attr); |
f6b53205 | 3262 | |
0e1f3483 HS |
3263 | if (running) |
3264 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3265 | if (on_rq) |
4ca9b72b | 3266 | enqueue_task(rq, p, 0); |
cb469845 | 3267 | |
da7a735e | 3268 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3269 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3270 | |
95e02ca9 TG |
3271 | rt_mutex_adjust_pi(p); |
3272 | ||
1da177e4 LT |
3273 | return 0; |
3274 | } | |
961ccddd RR |
3275 | |
3276 | /** | |
3277 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3278 | * @p: the task in question. | |
3279 | * @policy: new policy. | |
3280 | * @param: structure containing the new RT priority. | |
3281 | * | |
e69f6186 YB |
3282 | * Return: 0 on success. An error code otherwise. |
3283 | * | |
961ccddd RR |
3284 | * NOTE that the task may be already dead. |
3285 | */ | |
3286 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3287 | const struct sched_param *param) |
961ccddd | 3288 | { |
d50dde5a DF |
3289 | struct sched_attr attr = { |
3290 | .sched_policy = policy, | |
3291 | .sched_priority = param->sched_priority | |
3292 | }; | |
3293 | return __sched_setscheduler(p, &attr, true); | |
961ccddd | 3294 | } |
1da177e4 LT |
3295 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3296 | ||
d50dde5a DF |
3297 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3298 | { | |
3299 | return __sched_setscheduler(p, attr, true); | |
3300 | } | |
3301 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3302 | ||
961ccddd RR |
3303 | /** |
3304 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3305 | * @p: the task in question. | |
3306 | * @policy: new policy. | |
3307 | * @param: structure containing the new RT priority. | |
3308 | * | |
3309 | * Just like sched_setscheduler, only don't bother checking if the | |
3310 | * current context has permission. For example, this is needed in | |
3311 | * stop_machine(): we create temporary high priority worker threads, | |
3312 | * but our caller might not have that capability. | |
e69f6186 YB |
3313 | * |
3314 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3315 | */ |
3316 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3317 | const struct sched_param *param) |
961ccddd | 3318 | { |
d50dde5a DF |
3319 | struct sched_attr attr = { |
3320 | .sched_policy = policy, | |
3321 | .sched_priority = param->sched_priority | |
3322 | }; | |
3323 | return __sched_setscheduler(p, &attr, false); | |
961ccddd RR |
3324 | } |
3325 | ||
95cdf3b7 IM |
3326 | static int |
3327 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3328 | { |
1da177e4 LT |
3329 | struct sched_param lparam; |
3330 | struct task_struct *p; | |
36c8b586 | 3331 | int retval; |
1da177e4 LT |
3332 | |
3333 | if (!param || pid < 0) | |
3334 | return -EINVAL; | |
3335 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3336 | return -EFAULT; | |
5fe1d75f ON |
3337 | |
3338 | rcu_read_lock(); | |
3339 | retval = -ESRCH; | |
1da177e4 | 3340 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3341 | if (p != NULL) |
3342 | retval = sched_setscheduler(p, policy, &lparam); | |
3343 | rcu_read_unlock(); | |
36c8b586 | 3344 | |
1da177e4 LT |
3345 | return retval; |
3346 | } | |
3347 | ||
d50dde5a DF |
3348 | /* |
3349 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3350 | */ | |
3351 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3352 | struct sched_attr *attr) | |
3353 | { | |
3354 | u32 size; | |
3355 | int ret; | |
3356 | ||
3357 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3358 | return -EFAULT; | |
3359 | ||
3360 | /* | |
3361 | * zero the full structure, so that a short copy will be nice. | |
3362 | */ | |
3363 | memset(attr, 0, sizeof(*attr)); | |
3364 | ||
3365 | ret = get_user(size, &uattr->size); | |
3366 | if (ret) | |
3367 | return ret; | |
3368 | ||
3369 | if (size > PAGE_SIZE) /* silly large */ | |
3370 | goto err_size; | |
3371 | ||
3372 | if (!size) /* abi compat */ | |
3373 | size = SCHED_ATTR_SIZE_VER0; | |
3374 | ||
3375 | if (size < SCHED_ATTR_SIZE_VER0) | |
3376 | goto err_size; | |
3377 | ||
3378 | /* | |
3379 | * If we're handed a bigger struct than we know of, | |
3380 | * ensure all the unknown bits are 0 - i.e. new | |
3381 | * user-space does not rely on any kernel feature | |
3382 | * extensions we dont know about yet. | |
3383 | */ | |
3384 | if (size > sizeof(*attr)) { | |
3385 | unsigned char __user *addr; | |
3386 | unsigned char __user *end; | |
3387 | unsigned char val; | |
3388 | ||
3389 | addr = (void __user *)uattr + sizeof(*attr); | |
3390 | end = (void __user *)uattr + size; | |
3391 | ||
3392 | for (; addr < end; addr++) { | |
3393 | ret = get_user(val, addr); | |
3394 | if (ret) | |
3395 | return ret; | |
3396 | if (val) | |
3397 | goto err_size; | |
3398 | } | |
3399 | size = sizeof(*attr); | |
3400 | } | |
3401 | ||
3402 | ret = copy_from_user(attr, uattr, size); | |
3403 | if (ret) | |
3404 | return -EFAULT; | |
3405 | ||
3406 | /* | |
3407 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3408 | * to be strict and return an error on out-of-bounds values? | |
3409 | */ | |
3410 | attr->sched_nice = clamp(attr->sched_nice, -20, 19); | |
3411 | ||
3412 | out: | |
3413 | return ret; | |
3414 | ||
3415 | err_size: | |
3416 | put_user(sizeof(*attr), &uattr->size); | |
3417 | ret = -E2BIG; | |
3418 | goto out; | |
3419 | } | |
3420 | ||
1da177e4 LT |
3421 | /** |
3422 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3423 | * @pid: the pid in question. | |
3424 | * @policy: new policy. | |
3425 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3426 | * |
3427 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3428 | */ |
5add95d4 HC |
3429 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3430 | struct sched_param __user *, param) | |
1da177e4 | 3431 | { |
c21761f1 JB |
3432 | /* negative values for policy are not valid */ |
3433 | if (policy < 0) | |
3434 | return -EINVAL; | |
3435 | ||
1da177e4 LT |
3436 | return do_sched_setscheduler(pid, policy, param); |
3437 | } | |
3438 | ||
3439 | /** | |
3440 | * sys_sched_setparam - set/change the RT priority of a thread | |
3441 | * @pid: the pid in question. | |
3442 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3443 | * |
3444 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3445 | */ |
5add95d4 | 3446 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3447 | { |
3448 | return do_sched_setscheduler(pid, -1, param); | |
3449 | } | |
3450 | ||
d50dde5a DF |
3451 | /** |
3452 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3453 | * @pid: the pid in question. | |
3454 | * @attr: structure containing the extended parameters. | |
3455 | */ | |
3456 | SYSCALL_DEFINE2(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr) | |
3457 | { | |
3458 | struct sched_attr attr; | |
3459 | struct task_struct *p; | |
3460 | int retval; | |
3461 | ||
3462 | if (!uattr || pid < 0) | |
3463 | return -EINVAL; | |
3464 | ||
3465 | if (sched_copy_attr(uattr, &attr)) | |
3466 | return -EFAULT; | |
3467 | ||
3468 | rcu_read_lock(); | |
3469 | retval = -ESRCH; | |
3470 | p = find_process_by_pid(pid); | |
3471 | if (p != NULL) | |
3472 | retval = sched_setattr(p, &attr); | |
3473 | rcu_read_unlock(); | |
3474 | ||
3475 | return retval; | |
3476 | } | |
3477 | ||
1da177e4 LT |
3478 | /** |
3479 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3480 | * @pid: the pid in question. | |
e69f6186 YB |
3481 | * |
3482 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3483 | * code. | |
1da177e4 | 3484 | */ |
5add95d4 | 3485 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3486 | { |
36c8b586 | 3487 | struct task_struct *p; |
3a5c359a | 3488 | int retval; |
1da177e4 LT |
3489 | |
3490 | if (pid < 0) | |
3a5c359a | 3491 | return -EINVAL; |
1da177e4 LT |
3492 | |
3493 | retval = -ESRCH; | |
5fe85be0 | 3494 | rcu_read_lock(); |
1da177e4 LT |
3495 | p = find_process_by_pid(pid); |
3496 | if (p) { | |
3497 | retval = security_task_getscheduler(p); | |
3498 | if (!retval) | |
ca94c442 LP |
3499 | retval = p->policy |
3500 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3501 | } |
5fe85be0 | 3502 | rcu_read_unlock(); |
1da177e4 LT |
3503 | return retval; |
3504 | } | |
3505 | ||
3506 | /** | |
ca94c442 | 3507 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3508 | * @pid: the pid in question. |
3509 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3510 | * |
3511 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3512 | * code. | |
1da177e4 | 3513 | */ |
5add95d4 | 3514 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3515 | { |
3516 | struct sched_param lp; | |
36c8b586 | 3517 | struct task_struct *p; |
3a5c359a | 3518 | int retval; |
1da177e4 LT |
3519 | |
3520 | if (!param || pid < 0) | |
3a5c359a | 3521 | return -EINVAL; |
1da177e4 | 3522 | |
5fe85be0 | 3523 | rcu_read_lock(); |
1da177e4 LT |
3524 | p = find_process_by_pid(pid); |
3525 | retval = -ESRCH; | |
3526 | if (!p) | |
3527 | goto out_unlock; | |
3528 | ||
3529 | retval = security_task_getscheduler(p); | |
3530 | if (retval) | |
3531 | goto out_unlock; | |
3532 | ||
aab03e05 DF |
3533 | if (task_has_dl_policy(p)) { |
3534 | retval = -EINVAL; | |
3535 | goto out_unlock; | |
3536 | } | |
1da177e4 | 3537 | lp.sched_priority = p->rt_priority; |
5fe85be0 | 3538 | rcu_read_unlock(); |
1da177e4 LT |
3539 | |
3540 | /* | |
3541 | * This one might sleep, we cannot do it with a spinlock held ... | |
3542 | */ | |
3543 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3544 | ||
1da177e4 LT |
3545 | return retval; |
3546 | ||
3547 | out_unlock: | |
5fe85be0 | 3548 | rcu_read_unlock(); |
1da177e4 LT |
3549 | return retval; |
3550 | } | |
3551 | ||
d50dde5a DF |
3552 | static int sched_read_attr(struct sched_attr __user *uattr, |
3553 | struct sched_attr *attr, | |
3554 | unsigned int usize) | |
3555 | { | |
3556 | int ret; | |
3557 | ||
3558 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3559 | return -EFAULT; | |
3560 | ||
3561 | /* | |
3562 | * If we're handed a smaller struct than we know of, | |
3563 | * ensure all the unknown bits are 0 - i.e. old | |
3564 | * user-space does not get uncomplete information. | |
3565 | */ | |
3566 | if (usize < sizeof(*attr)) { | |
3567 | unsigned char *addr; | |
3568 | unsigned char *end; | |
3569 | ||
3570 | addr = (void *)attr + usize; | |
3571 | end = (void *)attr + sizeof(*attr); | |
3572 | ||
3573 | for (; addr < end; addr++) { | |
3574 | if (*addr) | |
3575 | goto err_size; | |
3576 | } | |
3577 | ||
3578 | attr->size = usize; | |
3579 | } | |
3580 | ||
3581 | ret = copy_to_user(uattr, attr, usize); | |
3582 | if (ret) | |
3583 | return -EFAULT; | |
3584 | ||
3585 | out: | |
3586 | return ret; | |
3587 | ||
3588 | err_size: | |
3589 | ret = -E2BIG; | |
3590 | goto out; | |
3591 | } | |
3592 | ||
3593 | /** | |
aab03e05 | 3594 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a DF |
3595 | * @pid: the pid in question. |
3596 | * @attr: structure containing the extended parameters. | |
3597 | * @size: sizeof(attr) for fwd/bwd comp. | |
3598 | */ | |
3599 | SYSCALL_DEFINE3(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, | |
3600 | unsigned int, size) | |
3601 | { | |
3602 | struct sched_attr attr = { | |
3603 | .size = sizeof(struct sched_attr), | |
3604 | }; | |
3605 | struct task_struct *p; | |
3606 | int retval; | |
3607 | ||
3608 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
3609 | size < SCHED_ATTR_SIZE_VER0) | |
3610 | return -EINVAL; | |
3611 | ||
3612 | rcu_read_lock(); | |
3613 | p = find_process_by_pid(pid); | |
3614 | retval = -ESRCH; | |
3615 | if (!p) | |
3616 | goto out_unlock; | |
3617 | ||
3618 | retval = security_task_getscheduler(p); | |
3619 | if (retval) | |
3620 | goto out_unlock; | |
3621 | ||
3622 | attr.sched_policy = p->policy; | |
aab03e05 DF |
3623 | if (task_has_dl_policy(p)) |
3624 | __getparam_dl(p, &attr); | |
3625 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3626 | attr.sched_priority = p->rt_priority; |
3627 | else | |
3628 | attr.sched_nice = TASK_NICE(p); | |
3629 | ||
3630 | rcu_read_unlock(); | |
3631 | ||
3632 | retval = sched_read_attr(uattr, &attr, size); | |
3633 | return retval; | |
3634 | ||
3635 | out_unlock: | |
3636 | rcu_read_unlock(); | |
3637 | return retval; | |
3638 | } | |
3639 | ||
96f874e2 | 3640 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3641 | { |
5a16f3d3 | 3642 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3643 | struct task_struct *p; |
3644 | int retval; | |
1da177e4 | 3645 | |
23f5d142 | 3646 | rcu_read_lock(); |
1da177e4 LT |
3647 | |
3648 | p = find_process_by_pid(pid); | |
3649 | if (!p) { | |
23f5d142 | 3650 | rcu_read_unlock(); |
1da177e4 LT |
3651 | return -ESRCH; |
3652 | } | |
3653 | ||
23f5d142 | 3654 | /* Prevent p going away */ |
1da177e4 | 3655 | get_task_struct(p); |
23f5d142 | 3656 | rcu_read_unlock(); |
1da177e4 | 3657 | |
14a40ffc TH |
3658 | if (p->flags & PF_NO_SETAFFINITY) { |
3659 | retval = -EINVAL; | |
3660 | goto out_put_task; | |
3661 | } | |
5a16f3d3 RR |
3662 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3663 | retval = -ENOMEM; | |
3664 | goto out_put_task; | |
3665 | } | |
3666 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3667 | retval = -ENOMEM; | |
3668 | goto out_free_cpus_allowed; | |
3669 | } | |
1da177e4 | 3670 | retval = -EPERM; |
4c44aaaf EB |
3671 | if (!check_same_owner(p)) { |
3672 | rcu_read_lock(); | |
3673 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3674 | rcu_read_unlock(); | |
3675 | goto out_unlock; | |
3676 | } | |
3677 | rcu_read_unlock(); | |
3678 | } | |
1da177e4 | 3679 | |
b0ae1981 | 3680 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3681 | if (retval) |
3682 | goto out_unlock; | |
3683 | ||
5a16f3d3 RR |
3684 | cpuset_cpus_allowed(p, cpus_allowed); |
3685 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 3686 | again: |
5a16f3d3 | 3687 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3688 | |
8707d8b8 | 3689 | if (!retval) { |
5a16f3d3 RR |
3690 | cpuset_cpus_allowed(p, cpus_allowed); |
3691 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3692 | /* |
3693 | * We must have raced with a concurrent cpuset | |
3694 | * update. Just reset the cpus_allowed to the | |
3695 | * cpuset's cpus_allowed | |
3696 | */ | |
5a16f3d3 | 3697 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3698 | goto again; |
3699 | } | |
3700 | } | |
1da177e4 | 3701 | out_unlock: |
5a16f3d3 RR |
3702 | free_cpumask_var(new_mask); |
3703 | out_free_cpus_allowed: | |
3704 | free_cpumask_var(cpus_allowed); | |
3705 | out_put_task: | |
1da177e4 | 3706 | put_task_struct(p); |
1da177e4 LT |
3707 | return retval; |
3708 | } | |
3709 | ||
3710 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3711 | struct cpumask *new_mask) |
1da177e4 | 3712 | { |
96f874e2 RR |
3713 | if (len < cpumask_size()) |
3714 | cpumask_clear(new_mask); | |
3715 | else if (len > cpumask_size()) | |
3716 | len = cpumask_size(); | |
3717 | ||
1da177e4 LT |
3718 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3719 | } | |
3720 | ||
3721 | /** | |
3722 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3723 | * @pid: pid of the process | |
3724 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3725 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3726 | * |
3727 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3728 | */ |
5add95d4 HC |
3729 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3730 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3731 | { |
5a16f3d3 | 3732 | cpumask_var_t new_mask; |
1da177e4 LT |
3733 | int retval; |
3734 | ||
5a16f3d3 RR |
3735 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3736 | return -ENOMEM; | |
1da177e4 | 3737 | |
5a16f3d3 RR |
3738 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3739 | if (retval == 0) | |
3740 | retval = sched_setaffinity(pid, new_mask); | |
3741 | free_cpumask_var(new_mask); | |
3742 | return retval; | |
1da177e4 LT |
3743 | } |
3744 | ||
96f874e2 | 3745 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 3746 | { |
36c8b586 | 3747 | struct task_struct *p; |
31605683 | 3748 | unsigned long flags; |
1da177e4 | 3749 | int retval; |
1da177e4 | 3750 | |
23f5d142 | 3751 | rcu_read_lock(); |
1da177e4 LT |
3752 | |
3753 | retval = -ESRCH; | |
3754 | p = find_process_by_pid(pid); | |
3755 | if (!p) | |
3756 | goto out_unlock; | |
3757 | ||
e7834f8f DQ |
3758 | retval = security_task_getscheduler(p); |
3759 | if (retval) | |
3760 | goto out_unlock; | |
3761 | ||
013fdb80 | 3762 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 3763 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 3764 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
3765 | |
3766 | out_unlock: | |
23f5d142 | 3767 | rcu_read_unlock(); |
1da177e4 | 3768 | |
9531b62f | 3769 | return retval; |
1da177e4 LT |
3770 | } |
3771 | ||
3772 | /** | |
3773 | * sys_sched_getaffinity - get the cpu affinity of a process | |
3774 | * @pid: pid of the process | |
3775 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3776 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
3777 | * |
3778 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3779 | */ |
5add95d4 HC |
3780 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
3781 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
3782 | { |
3783 | int ret; | |
f17c8607 | 3784 | cpumask_var_t mask; |
1da177e4 | 3785 | |
84fba5ec | 3786 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
3787 | return -EINVAL; |
3788 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
3789 | return -EINVAL; |
3790 | ||
f17c8607 RR |
3791 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
3792 | return -ENOMEM; | |
1da177e4 | 3793 | |
f17c8607 RR |
3794 | ret = sched_getaffinity(pid, mask); |
3795 | if (ret == 0) { | |
8bc037fb | 3796 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
3797 | |
3798 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
3799 | ret = -EFAULT; |
3800 | else | |
cd3d8031 | 3801 | ret = retlen; |
f17c8607 RR |
3802 | } |
3803 | free_cpumask_var(mask); | |
1da177e4 | 3804 | |
f17c8607 | 3805 | return ret; |
1da177e4 LT |
3806 | } |
3807 | ||
3808 | /** | |
3809 | * sys_sched_yield - yield the current processor to other threads. | |
3810 | * | |
dd41f596 IM |
3811 | * This function yields the current CPU to other tasks. If there are no |
3812 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
3813 | * |
3814 | * Return: 0. | |
1da177e4 | 3815 | */ |
5add95d4 | 3816 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 3817 | { |
70b97a7f | 3818 | struct rq *rq = this_rq_lock(); |
1da177e4 | 3819 | |
2d72376b | 3820 | schedstat_inc(rq, yld_count); |
4530d7ab | 3821 | current->sched_class->yield_task(rq); |
1da177e4 LT |
3822 | |
3823 | /* | |
3824 | * Since we are going to call schedule() anyway, there's | |
3825 | * no need to preempt or enable interrupts: | |
3826 | */ | |
3827 | __release(rq->lock); | |
8a25d5de | 3828 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 3829 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 3830 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
3831 | |
3832 | schedule(); | |
3833 | ||
3834 | return 0; | |
3835 | } | |
3836 | ||
e7b38404 | 3837 | static void __cond_resched(void) |
1da177e4 | 3838 | { |
bdb43806 | 3839 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 3840 | __schedule(); |
bdb43806 | 3841 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
3842 | } |
3843 | ||
02b67cc3 | 3844 | int __sched _cond_resched(void) |
1da177e4 | 3845 | { |
d86ee480 | 3846 | if (should_resched()) { |
1da177e4 LT |
3847 | __cond_resched(); |
3848 | return 1; | |
3849 | } | |
3850 | return 0; | |
3851 | } | |
02b67cc3 | 3852 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
3853 | |
3854 | /* | |
613afbf8 | 3855 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
3856 | * call schedule, and on return reacquire the lock. |
3857 | * | |
41a2d6cf | 3858 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
3859 | * operations here to prevent schedule() from being called twice (once via |
3860 | * spin_unlock(), once by hand). | |
3861 | */ | |
613afbf8 | 3862 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 3863 | { |
d86ee480 | 3864 | int resched = should_resched(); |
6df3cecb JK |
3865 | int ret = 0; |
3866 | ||
f607c668 PZ |
3867 | lockdep_assert_held(lock); |
3868 | ||
95c354fe | 3869 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 3870 | spin_unlock(lock); |
d86ee480 | 3871 | if (resched) |
95c354fe NP |
3872 | __cond_resched(); |
3873 | else | |
3874 | cpu_relax(); | |
6df3cecb | 3875 | ret = 1; |
1da177e4 | 3876 | spin_lock(lock); |
1da177e4 | 3877 | } |
6df3cecb | 3878 | return ret; |
1da177e4 | 3879 | } |
613afbf8 | 3880 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 3881 | |
613afbf8 | 3882 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
3883 | { |
3884 | BUG_ON(!in_softirq()); | |
3885 | ||
d86ee480 | 3886 | if (should_resched()) { |
98d82567 | 3887 | local_bh_enable(); |
1da177e4 LT |
3888 | __cond_resched(); |
3889 | local_bh_disable(); | |
3890 | return 1; | |
3891 | } | |
3892 | return 0; | |
3893 | } | |
613afbf8 | 3894 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 3895 | |
1da177e4 LT |
3896 | /** |
3897 | * yield - yield the current processor to other threads. | |
3898 | * | |
8e3fabfd PZ |
3899 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
3900 | * | |
3901 | * The scheduler is at all times free to pick the calling task as the most | |
3902 | * eligible task to run, if removing the yield() call from your code breaks | |
3903 | * it, its already broken. | |
3904 | * | |
3905 | * Typical broken usage is: | |
3906 | * | |
3907 | * while (!event) | |
3908 | * yield(); | |
3909 | * | |
3910 | * where one assumes that yield() will let 'the other' process run that will | |
3911 | * make event true. If the current task is a SCHED_FIFO task that will never | |
3912 | * happen. Never use yield() as a progress guarantee!! | |
3913 | * | |
3914 | * If you want to use yield() to wait for something, use wait_event(). | |
3915 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
3916 | * If you still want to use yield(), do not! | |
1da177e4 LT |
3917 | */ |
3918 | void __sched yield(void) | |
3919 | { | |
3920 | set_current_state(TASK_RUNNING); | |
3921 | sys_sched_yield(); | |
3922 | } | |
1da177e4 LT |
3923 | EXPORT_SYMBOL(yield); |
3924 | ||
d95f4122 MG |
3925 | /** |
3926 | * yield_to - yield the current processor to another thread in | |
3927 | * your thread group, or accelerate that thread toward the | |
3928 | * processor it's on. | |
16addf95 RD |
3929 | * @p: target task |
3930 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
3931 | * |
3932 | * It's the caller's job to ensure that the target task struct | |
3933 | * can't go away on us before we can do any checks. | |
3934 | * | |
e69f6186 | 3935 | * Return: |
7b270f60 PZ |
3936 | * true (>0) if we indeed boosted the target task. |
3937 | * false (0) if we failed to boost the target. | |
3938 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
3939 | */ |
3940 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
3941 | { | |
3942 | struct task_struct *curr = current; | |
3943 | struct rq *rq, *p_rq; | |
3944 | unsigned long flags; | |
c3c18640 | 3945 | int yielded = 0; |
d95f4122 MG |
3946 | |
3947 | local_irq_save(flags); | |
3948 | rq = this_rq(); | |
3949 | ||
3950 | again: | |
3951 | p_rq = task_rq(p); | |
7b270f60 PZ |
3952 | /* |
3953 | * If we're the only runnable task on the rq and target rq also | |
3954 | * has only one task, there's absolutely no point in yielding. | |
3955 | */ | |
3956 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
3957 | yielded = -ESRCH; | |
3958 | goto out_irq; | |
3959 | } | |
3960 | ||
d95f4122 | 3961 | double_rq_lock(rq, p_rq); |
39e24d8f | 3962 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
3963 | double_rq_unlock(rq, p_rq); |
3964 | goto again; | |
3965 | } | |
3966 | ||
3967 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 3968 | goto out_unlock; |
d95f4122 MG |
3969 | |
3970 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 3971 | goto out_unlock; |
d95f4122 MG |
3972 | |
3973 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 3974 | goto out_unlock; |
d95f4122 MG |
3975 | |
3976 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 3977 | if (yielded) { |
d95f4122 | 3978 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
3979 | /* |
3980 | * Make p's CPU reschedule; pick_next_entity takes care of | |
3981 | * fairness. | |
3982 | */ | |
3983 | if (preempt && rq != p_rq) | |
3984 | resched_task(p_rq->curr); | |
3985 | } | |
d95f4122 | 3986 | |
7b270f60 | 3987 | out_unlock: |
d95f4122 | 3988 | double_rq_unlock(rq, p_rq); |
7b270f60 | 3989 | out_irq: |
d95f4122 MG |
3990 | local_irq_restore(flags); |
3991 | ||
7b270f60 | 3992 | if (yielded > 0) |
d95f4122 MG |
3993 | schedule(); |
3994 | ||
3995 | return yielded; | |
3996 | } | |
3997 | EXPORT_SYMBOL_GPL(yield_to); | |
3998 | ||
1da177e4 | 3999 | /* |
41a2d6cf | 4000 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4001 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4002 | */ |
4003 | void __sched io_schedule(void) | |
4004 | { | |
54d35f29 | 4005 | struct rq *rq = raw_rq(); |
1da177e4 | 4006 | |
0ff92245 | 4007 | delayacct_blkio_start(); |
1da177e4 | 4008 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4009 | blk_flush_plug(current); |
8f0dfc34 | 4010 | current->in_iowait = 1; |
1da177e4 | 4011 | schedule(); |
8f0dfc34 | 4012 | current->in_iowait = 0; |
1da177e4 | 4013 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4014 | delayacct_blkio_end(); |
1da177e4 | 4015 | } |
1da177e4 LT |
4016 | EXPORT_SYMBOL(io_schedule); |
4017 | ||
4018 | long __sched io_schedule_timeout(long timeout) | |
4019 | { | |
54d35f29 | 4020 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4021 | long ret; |
4022 | ||
0ff92245 | 4023 | delayacct_blkio_start(); |
1da177e4 | 4024 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4025 | blk_flush_plug(current); |
8f0dfc34 | 4026 | current->in_iowait = 1; |
1da177e4 | 4027 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4028 | current->in_iowait = 0; |
1da177e4 | 4029 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4030 | delayacct_blkio_end(); |
1da177e4 LT |
4031 | return ret; |
4032 | } | |
4033 | ||
4034 | /** | |
4035 | * sys_sched_get_priority_max - return maximum RT priority. | |
4036 | * @policy: scheduling class. | |
4037 | * | |
e69f6186 YB |
4038 | * Return: On success, this syscall returns the maximum |
4039 | * rt_priority that can be used by a given scheduling class. | |
4040 | * On failure, a negative error code is returned. | |
1da177e4 | 4041 | */ |
5add95d4 | 4042 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4043 | { |
4044 | int ret = -EINVAL; | |
4045 | ||
4046 | switch (policy) { | |
4047 | case SCHED_FIFO: | |
4048 | case SCHED_RR: | |
4049 | ret = MAX_USER_RT_PRIO-1; | |
4050 | break; | |
aab03e05 | 4051 | case SCHED_DEADLINE: |
1da177e4 | 4052 | case SCHED_NORMAL: |
b0a9499c | 4053 | case SCHED_BATCH: |
dd41f596 | 4054 | case SCHED_IDLE: |
1da177e4 LT |
4055 | ret = 0; |
4056 | break; | |
4057 | } | |
4058 | return ret; | |
4059 | } | |
4060 | ||
4061 | /** | |
4062 | * sys_sched_get_priority_min - return minimum RT priority. | |
4063 | * @policy: scheduling class. | |
4064 | * | |
e69f6186 YB |
4065 | * Return: On success, this syscall returns the minimum |
4066 | * rt_priority that can be used by a given scheduling class. | |
4067 | * On failure, a negative error code is returned. | |
1da177e4 | 4068 | */ |
5add95d4 | 4069 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4070 | { |
4071 | int ret = -EINVAL; | |
4072 | ||
4073 | switch (policy) { | |
4074 | case SCHED_FIFO: | |
4075 | case SCHED_RR: | |
4076 | ret = 1; | |
4077 | break; | |
aab03e05 | 4078 | case SCHED_DEADLINE: |
1da177e4 | 4079 | case SCHED_NORMAL: |
b0a9499c | 4080 | case SCHED_BATCH: |
dd41f596 | 4081 | case SCHED_IDLE: |
1da177e4 LT |
4082 | ret = 0; |
4083 | } | |
4084 | return ret; | |
4085 | } | |
4086 | ||
4087 | /** | |
4088 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4089 | * @pid: pid of the process. | |
4090 | * @interval: userspace pointer to the timeslice value. | |
4091 | * | |
4092 | * this syscall writes the default timeslice value of a given process | |
4093 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4094 | * |
4095 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4096 | * an error code. | |
1da177e4 | 4097 | */ |
17da2bd9 | 4098 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4099 | struct timespec __user *, interval) |
1da177e4 | 4100 | { |
36c8b586 | 4101 | struct task_struct *p; |
a4ec24b4 | 4102 | unsigned int time_slice; |
dba091b9 TG |
4103 | unsigned long flags; |
4104 | struct rq *rq; | |
3a5c359a | 4105 | int retval; |
1da177e4 | 4106 | struct timespec t; |
1da177e4 LT |
4107 | |
4108 | if (pid < 0) | |
3a5c359a | 4109 | return -EINVAL; |
1da177e4 LT |
4110 | |
4111 | retval = -ESRCH; | |
1a551ae7 | 4112 | rcu_read_lock(); |
1da177e4 LT |
4113 | p = find_process_by_pid(pid); |
4114 | if (!p) | |
4115 | goto out_unlock; | |
4116 | ||
4117 | retval = security_task_getscheduler(p); | |
4118 | if (retval) | |
4119 | goto out_unlock; | |
4120 | ||
dba091b9 TG |
4121 | rq = task_rq_lock(p, &flags); |
4122 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4123 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4124 | |
1a551ae7 | 4125 | rcu_read_unlock(); |
a4ec24b4 | 4126 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4127 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4128 | return retval; |
3a5c359a | 4129 | |
1da177e4 | 4130 | out_unlock: |
1a551ae7 | 4131 | rcu_read_unlock(); |
1da177e4 LT |
4132 | return retval; |
4133 | } | |
4134 | ||
7c731e0a | 4135 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4136 | |
82a1fcb9 | 4137 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4138 | { |
1da177e4 | 4139 | unsigned long free = 0; |
4e79752c | 4140 | int ppid; |
36c8b586 | 4141 | unsigned state; |
1da177e4 | 4142 | |
1da177e4 | 4143 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4144 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4145 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4146 | #if BITS_PER_LONG == 32 |
1da177e4 | 4147 | if (state == TASK_RUNNING) |
3df0fc5b | 4148 | printk(KERN_CONT " running "); |
1da177e4 | 4149 | else |
3df0fc5b | 4150 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4151 | #else |
4152 | if (state == TASK_RUNNING) | |
3df0fc5b | 4153 | printk(KERN_CONT " running task "); |
1da177e4 | 4154 | else |
3df0fc5b | 4155 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4156 | #endif |
4157 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4158 | free = stack_not_used(p); |
1da177e4 | 4159 | #endif |
4e79752c PM |
4160 | rcu_read_lock(); |
4161 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4162 | rcu_read_unlock(); | |
3df0fc5b | 4163 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4164 | task_pid_nr(p), ppid, |
aa47b7e0 | 4165 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4166 | |
3d1cb205 | 4167 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4168 | show_stack(p, NULL); |
1da177e4 LT |
4169 | } |
4170 | ||
e59e2ae2 | 4171 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4172 | { |
36c8b586 | 4173 | struct task_struct *g, *p; |
1da177e4 | 4174 | |
4bd77321 | 4175 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4176 | printk(KERN_INFO |
4177 | " task PC stack pid father\n"); | |
1da177e4 | 4178 | #else |
3df0fc5b PZ |
4179 | printk(KERN_INFO |
4180 | " task PC stack pid father\n"); | |
1da177e4 | 4181 | #endif |
510f5acc | 4182 | rcu_read_lock(); |
1da177e4 LT |
4183 | do_each_thread(g, p) { |
4184 | /* | |
4185 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4186 | * console might take a lot of time: |
1da177e4 LT |
4187 | */ |
4188 | touch_nmi_watchdog(); | |
39bc89fd | 4189 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4190 | sched_show_task(p); |
1da177e4 LT |
4191 | } while_each_thread(g, p); |
4192 | ||
04c9167f JF |
4193 | touch_all_softlockup_watchdogs(); |
4194 | ||
dd41f596 IM |
4195 | #ifdef CONFIG_SCHED_DEBUG |
4196 | sysrq_sched_debug_show(); | |
4197 | #endif | |
510f5acc | 4198 | rcu_read_unlock(); |
e59e2ae2 IM |
4199 | /* |
4200 | * Only show locks if all tasks are dumped: | |
4201 | */ | |
93335a21 | 4202 | if (!state_filter) |
e59e2ae2 | 4203 | debug_show_all_locks(); |
1da177e4 LT |
4204 | } |
4205 | ||
0db0628d | 4206 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4207 | { |
dd41f596 | 4208 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4209 | } |
4210 | ||
f340c0d1 IM |
4211 | /** |
4212 | * init_idle - set up an idle thread for a given CPU | |
4213 | * @idle: task in question | |
4214 | * @cpu: cpu the idle task belongs to | |
4215 | * | |
4216 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4217 | * flag, to make booting more robust. | |
4218 | */ | |
0db0628d | 4219 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4220 | { |
70b97a7f | 4221 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4222 | unsigned long flags; |
4223 | ||
05fa785c | 4224 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4225 | |
5e1576ed | 4226 | __sched_fork(0, idle); |
06b83b5f | 4227 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4228 | idle->se.exec_start = sched_clock(); |
4229 | ||
1e1b6c51 | 4230 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4231 | /* |
4232 | * We're having a chicken and egg problem, even though we are | |
4233 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4234 | * lockdep check in task_group() will fail. | |
4235 | * | |
4236 | * Similar case to sched_fork(). / Alternatively we could | |
4237 | * use task_rq_lock() here and obtain the other rq->lock. | |
4238 | * | |
4239 | * Silence PROVE_RCU | |
4240 | */ | |
4241 | rcu_read_lock(); | |
dd41f596 | 4242 | __set_task_cpu(idle, cpu); |
6506cf6c | 4243 | rcu_read_unlock(); |
1da177e4 | 4244 | |
1da177e4 | 4245 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4246 | #if defined(CONFIG_SMP) |
4247 | idle->on_cpu = 1; | |
4866cde0 | 4248 | #endif |
05fa785c | 4249 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4250 | |
4251 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4252 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4253 | |
dd41f596 IM |
4254 | /* |
4255 | * The idle tasks have their own, simple scheduling class: | |
4256 | */ | |
4257 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4258 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4259 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4260 | #if defined(CONFIG_SMP) |
4261 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4262 | #endif | |
19978ca6 IM |
4263 | } |
4264 | ||
1da177e4 | 4265 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4266 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4267 | { | |
4268 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4269 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4270 | |
4271 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4272 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4273 | } |
4274 | ||
1da177e4 LT |
4275 | /* |
4276 | * This is how migration works: | |
4277 | * | |
969c7921 TH |
4278 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4279 | * stop_one_cpu(). | |
4280 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4281 | * off the CPU) | |
4282 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4283 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4284 | * it and puts it into the right queue. |
969c7921 TH |
4285 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4286 | * is done. | |
1da177e4 LT |
4287 | */ |
4288 | ||
4289 | /* | |
4290 | * Change a given task's CPU affinity. Migrate the thread to a | |
4291 | * proper CPU and schedule it away if the CPU it's executing on | |
4292 | * is removed from the allowed bitmask. | |
4293 | * | |
4294 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4295 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4296 | * call is not atomic; no spinlocks may be held. |
4297 | */ | |
96f874e2 | 4298 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4299 | { |
4300 | unsigned long flags; | |
70b97a7f | 4301 | struct rq *rq; |
969c7921 | 4302 | unsigned int dest_cpu; |
48f24c4d | 4303 | int ret = 0; |
1da177e4 LT |
4304 | |
4305 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4306 | |
db44fc01 YZ |
4307 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4308 | goto out; | |
4309 | ||
6ad4c188 | 4310 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4311 | ret = -EINVAL; |
4312 | goto out; | |
4313 | } | |
4314 | ||
1e1b6c51 | 4315 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4316 | |
1da177e4 | 4317 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4318 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4319 | goto out; |
4320 | ||
969c7921 | 4321 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4322 | if (p->on_rq) { |
969c7921 | 4323 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4324 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4325 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4326 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4327 | tlb_migrate_finish(p->mm); |
4328 | return 0; | |
4329 | } | |
4330 | out: | |
0122ec5b | 4331 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4332 | |
1da177e4 LT |
4333 | return ret; |
4334 | } | |
cd8ba7cd | 4335 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4336 | |
4337 | /* | |
41a2d6cf | 4338 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4339 | * this because either it can't run here any more (set_cpus_allowed() |
4340 | * away from this CPU, or CPU going down), or because we're | |
4341 | * attempting to rebalance this task on exec (sched_exec). | |
4342 | * | |
4343 | * So we race with normal scheduler movements, but that's OK, as long | |
4344 | * as the task is no longer on this CPU. | |
efc30814 KK |
4345 | * |
4346 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4347 | */ |
efc30814 | 4348 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4349 | { |
70b97a7f | 4350 | struct rq *rq_dest, *rq_src; |
e2912009 | 4351 | int ret = 0; |
1da177e4 | 4352 | |
e761b772 | 4353 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4354 | return ret; |
1da177e4 LT |
4355 | |
4356 | rq_src = cpu_rq(src_cpu); | |
4357 | rq_dest = cpu_rq(dest_cpu); | |
4358 | ||
0122ec5b | 4359 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4360 | double_rq_lock(rq_src, rq_dest); |
4361 | /* Already moved. */ | |
4362 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4363 | goto done; |
1da177e4 | 4364 | /* Affinity changed (again). */ |
fa17b507 | 4365 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4366 | goto fail; |
1da177e4 | 4367 | |
e2912009 PZ |
4368 | /* |
4369 | * If we're not on a rq, the next wake-up will ensure we're | |
4370 | * placed properly. | |
4371 | */ | |
fd2f4419 | 4372 | if (p->on_rq) { |
4ca9b72b | 4373 | dequeue_task(rq_src, p, 0); |
e2912009 | 4374 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4375 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4376 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4377 | } |
b1e38734 | 4378 | done: |
efc30814 | 4379 | ret = 1; |
b1e38734 | 4380 | fail: |
1da177e4 | 4381 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4382 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4383 | return ret; |
1da177e4 LT |
4384 | } |
4385 | ||
e6628d5b MG |
4386 | #ifdef CONFIG_NUMA_BALANCING |
4387 | /* Migrate current task p to target_cpu */ | |
4388 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4389 | { | |
4390 | struct migration_arg arg = { p, target_cpu }; | |
4391 | int curr_cpu = task_cpu(p); | |
4392 | ||
4393 | if (curr_cpu == target_cpu) | |
4394 | return 0; | |
4395 | ||
4396 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4397 | return -EINVAL; | |
4398 | ||
4399 | /* TODO: This is not properly updating schedstats */ | |
4400 | ||
4401 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); | |
4402 | } | |
0ec8aa00 PZ |
4403 | |
4404 | /* | |
4405 | * Requeue a task on a given node and accurately track the number of NUMA | |
4406 | * tasks on the runqueues | |
4407 | */ | |
4408 | void sched_setnuma(struct task_struct *p, int nid) | |
4409 | { | |
4410 | struct rq *rq; | |
4411 | unsigned long flags; | |
4412 | bool on_rq, running; | |
4413 | ||
4414 | rq = task_rq_lock(p, &flags); | |
4415 | on_rq = p->on_rq; | |
4416 | running = task_current(rq, p); | |
4417 | ||
4418 | if (on_rq) | |
4419 | dequeue_task(rq, p, 0); | |
4420 | if (running) | |
4421 | p->sched_class->put_prev_task(rq, p); | |
4422 | ||
4423 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4424 | |
4425 | if (running) | |
4426 | p->sched_class->set_curr_task(rq); | |
4427 | if (on_rq) | |
4428 | enqueue_task(rq, p, 0); | |
4429 | task_rq_unlock(rq, p, &flags); | |
4430 | } | |
e6628d5b MG |
4431 | #endif |
4432 | ||
1da177e4 | 4433 | /* |
969c7921 TH |
4434 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4435 | * and performs thread migration by bumping thread off CPU then | |
4436 | * 'pushing' onto another runqueue. | |
1da177e4 | 4437 | */ |
969c7921 | 4438 | static int migration_cpu_stop(void *data) |
1da177e4 | 4439 | { |
969c7921 | 4440 | struct migration_arg *arg = data; |
f7b4cddc | 4441 | |
969c7921 TH |
4442 | /* |
4443 | * The original target cpu might have gone down and we might | |
4444 | * be on another cpu but it doesn't matter. | |
4445 | */ | |
f7b4cddc | 4446 | local_irq_disable(); |
969c7921 | 4447 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4448 | local_irq_enable(); |
1da177e4 | 4449 | return 0; |
f7b4cddc ON |
4450 | } |
4451 | ||
1da177e4 | 4452 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4453 | |
054b9108 | 4454 | /* |
48c5ccae PZ |
4455 | * Ensures that the idle task is using init_mm right before its cpu goes |
4456 | * offline. | |
054b9108 | 4457 | */ |
48c5ccae | 4458 | void idle_task_exit(void) |
1da177e4 | 4459 | { |
48c5ccae | 4460 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4461 | |
48c5ccae | 4462 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4463 | |
48c5ccae PZ |
4464 | if (mm != &init_mm) |
4465 | switch_mm(mm, &init_mm, current); | |
4466 | mmdrop(mm); | |
1da177e4 LT |
4467 | } |
4468 | ||
4469 | /* | |
5d180232 PZ |
4470 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4471 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4472 | * nr_active count is stable. | |
4473 | * | |
4474 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4475 | */ |
5d180232 | 4476 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4477 | { |
5d180232 PZ |
4478 | long delta = calc_load_fold_active(rq); |
4479 | if (delta) | |
4480 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4481 | } |
4482 | ||
48f24c4d | 4483 | /* |
48c5ccae PZ |
4484 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4485 | * try_to_wake_up()->select_task_rq(). | |
4486 | * | |
4487 | * Called with rq->lock held even though we'er in stop_machine() and | |
4488 | * there's no concurrency possible, we hold the required locks anyway | |
4489 | * because of lock validation efforts. | |
1da177e4 | 4490 | */ |
48c5ccae | 4491 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4492 | { |
70b97a7f | 4493 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4494 | struct task_struct *next, *stop = rq->stop; |
4495 | int dest_cpu; | |
1da177e4 LT |
4496 | |
4497 | /* | |
48c5ccae PZ |
4498 | * Fudge the rq selection such that the below task selection loop |
4499 | * doesn't get stuck on the currently eligible stop task. | |
4500 | * | |
4501 | * We're currently inside stop_machine() and the rq is either stuck | |
4502 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4503 | * either way we should never end up calling schedule() until we're | |
4504 | * done here. | |
1da177e4 | 4505 | */ |
48c5ccae | 4506 | rq->stop = NULL; |
48f24c4d | 4507 | |
77bd3970 FW |
4508 | /* |
4509 | * put_prev_task() and pick_next_task() sched | |
4510 | * class method both need to have an up-to-date | |
4511 | * value of rq->clock[_task] | |
4512 | */ | |
4513 | update_rq_clock(rq); | |
4514 | ||
dd41f596 | 4515 | for ( ; ; ) { |
48c5ccae PZ |
4516 | /* |
4517 | * There's this thread running, bail when that's the only | |
4518 | * remaining thread. | |
4519 | */ | |
4520 | if (rq->nr_running == 1) | |
dd41f596 | 4521 | break; |
48c5ccae | 4522 | |
b67802ea | 4523 | next = pick_next_task(rq); |
48c5ccae | 4524 | BUG_ON(!next); |
79c53799 | 4525 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4526 | |
48c5ccae PZ |
4527 | /* Find suitable destination for @next, with force if needed. */ |
4528 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4529 | raw_spin_unlock(&rq->lock); | |
4530 | ||
4531 | __migrate_task(next, dead_cpu, dest_cpu); | |
4532 | ||
4533 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4534 | } |
dce48a84 | 4535 | |
48c5ccae | 4536 | rq->stop = stop; |
dce48a84 | 4537 | } |
48c5ccae | 4538 | |
1da177e4 LT |
4539 | #endif /* CONFIG_HOTPLUG_CPU */ |
4540 | ||
e692ab53 NP |
4541 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4542 | ||
4543 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4544 | { |
4545 | .procname = "sched_domain", | |
c57baf1e | 4546 | .mode = 0555, |
e0361851 | 4547 | }, |
56992309 | 4548 | {} |
e692ab53 NP |
4549 | }; |
4550 | ||
4551 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4552 | { |
4553 | .procname = "kernel", | |
c57baf1e | 4554 | .mode = 0555, |
e0361851 AD |
4555 | .child = sd_ctl_dir, |
4556 | }, | |
56992309 | 4557 | {} |
e692ab53 NP |
4558 | }; |
4559 | ||
4560 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4561 | { | |
4562 | struct ctl_table *entry = | |
5cf9f062 | 4563 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4564 | |
e692ab53 NP |
4565 | return entry; |
4566 | } | |
4567 | ||
6382bc90 MM |
4568 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4569 | { | |
cd790076 | 4570 | struct ctl_table *entry; |
6382bc90 | 4571 | |
cd790076 MM |
4572 | /* |
4573 | * In the intermediate directories, both the child directory and | |
4574 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4575 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4576 | * static strings and all have proc handlers. |
4577 | */ | |
4578 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4579 | if (entry->child) |
4580 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4581 | if (entry->proc_handler == NULL) |
4582 | kfree(entry->procname); | |
4583 | } | |
6382bc90 MM |
4584 | |
4585 | kfree(*tablep); | |
4586 | *tablep = NULL; | |
4587 | } | |
4588 | ||
201c373e | 4589 | static int min_load_idx = 0; |
fd9b86d3 | 4590 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4591 | |
e692ab53 | 4592 | static void |
e0361851 | 4593 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4594 | const char *procname, void *data, int maxlen, |
201c373e NK |
4595 | umode_t mode, proc_handler *proc_handler, |
4596 | bool load_idx) | |
e692ab53 | 4597 | { |
e692ab53 NP |
4598 | entry->procname = procname; |
4599 | entry->data = data; | |
4600 | entry->maxlen = maxlen; | |
4601 | entry->mode = mode; | |
4602 | entry->proc_handler = proc_handler; | |
201c373e NK |
4603 | |
4604 | if (load_idx) { | |
4605 | entry->extra1 = &min_load_idx; | |
4606 | entry->extra2 = &max_load_idx; | |
4607 | } | |
e692ab53 NP |
4608 | } |
4609 | ||
4610 | static struct ctl_table * | |
4611 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4612 | { | |
a5d8c348 | 4613 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 4614 | |
ad1cdc1d MM |
4615 | if (table == NULL) |
4616 | return NULL; | |
4617 | ||
e0361851 | 4618 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4619 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4620 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4621 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4622 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4623 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4624 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4625 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4626 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4627 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4628 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4629 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4630 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4631 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4632 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4633 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4634 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4635 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4636 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4637 | &sd->cache_nice_tries, |
201c373e | 4638 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4639 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4640 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 4641 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 4642 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 4643 | /* &table[12] is terminator */ |
e692ab53 NP |
4644 | |
4645 | return table; | |
4646 | } | |
4647 | ||
be7002e6 | 4648 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4649 | { |
4650 | struct ctl_table *entry, *table; | |
4651 | struct sched_domain *sd; | |
4652 | int domain_num = 0, i; | |
4653 | char buf[32]; | |
4654 | ||
4655 | for_each_domain(cpu, sd) | |
4656 | domain_num++; | |
4657 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4658 | if (table == NULL) |
4659 | return NULL; | |
e692ab53 NP |
4660 | |
4661 | i = 0; | |
4662 | for_each_domain(cpu, sd) { | |
4663 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4664 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4665 | entry->mode = 0555; |
e692ab53 NP |
4666 | entry->child = sd_alloc_ctl_domain_table(sd); |
4667 | entry++; | |
4668 | i++; | |
4669 | } | |
4670 | return table; | |
4671 | } | |
4672 | ||
4673 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4674 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4675 | { |
6ad4c188 | 4676 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4677 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4678 | char buf[32]; | |
4679 | ||
7378547f MM |
4680 | WARN_ON(sd_ctl_dir[0].child); |
4681 | sd_ctl_dir[0].child = entry; | |
4682 | ||
ad1cdc1d MM |
4683 | if (entry == NULL) |
4684 | return; | |
4685 | ||
6ad4c188 | 4686 | for_each_possible_cpu(i) { |
e692ab53 | 4687 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4688 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4689 | entry->mode = 0555; |
e692ab53 | 4690 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4691 | entry++; |
e692ab53 | 4692 | } |
7378547f MM |
4693 | |
4694 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4695 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4696 | } | |
6382bc90 | 4697 | |
7378547f | 4698 | /* may be called multiple times per register */ |
6382bc90 MM |
4699 | static void unregister_sched_domain_sysctl(void) |
4700 | { | |
7378547f MM |
4701 | if (sd_sysctl_header) |
4702 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4703 | sd_sysctl_header = NULL; |
7378547f MM |
4704 | if (sd_ctl_dir[0].child) |
4705 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4706 | } |
e692ab53 | 4707 | #else |
6382bc90 MM |
4708 | static void register_sched_domain_sysctl(void) |
4709 | { | |
4710 | } | |
4711 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4712 | { |
4713 | } | |
4714 | #endif | |
4715 | ||
1f11eb6a GH |
4716 | static void set_rq_online(struct rq *rq) |
4717 | { | |
4718 | if (!rq->online) { | |
4719 | const struct sched_class *class; | |
4720 | ||
c6c4927b | 4721 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4722 | rq->online = 1; |
4723 | ||
4724 | for_each_class(class) { | |
4725 | if (class->rq_online) | |
4726 | class->rq_online(rq); | |
4727 | } | |
4728 | } | |
4729 | } | |
4730 | ||
4731 | static void set_rq_offline(struct rq *rq) | |
4732 | { | |
4733 | if (rq->online) { | |
4734 | const struct sched_class *class; | |
4735 | ||
4736 | for_each_class(class) { | |
4737 | if (class->rq_offline) | |
4738 | class->rq_offline(rq); | |
4739 | } | |
4740 | ||
c6c4927b | 4741 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4742 | rq->online = 0; |
4743 | } | |
4744 | } | |
4745 | ||
1da177e4 LT |
4746 | /* |
4747 | * migration_call - callback that gets triggered when a CPU is added. | |
4748 | * Here we can start up the necessary migration thread for the new CPU. | |
4749 | */ | |
0db0628d | 4750 | static int |
48f24c4d | 4751 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 4752 | { |
48f24c4d | 4753 | int cpu = (long)hcpu; |
1da177e4 | 4754 | unsigned long flags; |
969c7921 | 4755 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 4756 | |
48c5ccae | 4757 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 4758 | |
1da177e4 | 4759 | case CPU_UP_PREPARE: |
a468d389 | 4760 | rq->calc_load_update = calc_load_update; |
1da177e4 | 4761 | break; |
48f24c4d | 4762 | |
1da177e4 | 4763 | case CPU_ONLINE: |
1f94ef59 | 4764 | /* Update our root-domain */ |
05fa785c | 4765 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 4766 | if (rq->rd) { |
c6c4927b | 4767 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
4768 | |
4769 | set_rq_online(rq); | |
1f94ef59 | 4770 | } |
05fa785c | 4771 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 4772 | break; |
48f24c4d | 4773 | |
1da177e4 | 4774 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 4775 | case CPU_DYING: |
317f3941 | 4776 | sched_ttwu_pending(); |
57d885fe | 4777 | /* Update our root-domain */ |
05fa785c | 4778 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 4779 | if (rq->rd) { |
c6c4927b | 4780 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 4781 | set_rq_offline(rq); |
57d885fe | 4782 | } |
48c5ccae PZ |
4783 | migrate_tasks(cpu); |
4784 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 4785 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 4786 | break; |
48c5ccae | 4787 | |
5d180232 | 4788 | case CPU_DEAD: |
f319da0c | 4789 | calc_load_migrate(rq); |
57d885fe | 4790 | break; |
1da177e4 LT |
4791 | #endif |
4792 | } | |
49c022e6 PZ |
4793 | |
4794 | update_max_interval(); | |
4795 | ||
1da177e4 LT |
4796 | return NOTIFY_OK; |
4797 | } | |
4798 | ||
f38b0820 PM |
4799 | /* |
4800 | * Register at high priority so that task migration (migrate_all_tasks) | |
4801 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 4802 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 4803 | */ |
0db0628d | 4804 | static struct notifier_block migration_notifier = { |
1da177e4 | 4805 | .notifier_call = migration_call, |
50a323b7 | 4806 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
4807 | }; |
4808 | ||
0db0628d | 4809 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
4810 | unsigned long action, void *hcpu) |
4811 | { | |
4812 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 4813 | case CPU_STARTING: |
3a101d05 TH |
4814 | case CPU_DOWN_FAILED: |
4815 | set_cpu_active((long)hcpu, true); | |
4816 | return NOTIFY_OK; | |
4817 | default: | |
4818 | return NOTIFY_DONE; | |
4819 | } | |
4820 | } | |
4821 | ||
0db0628d | 4822 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
4823 | unsigned long action, void *hcpu) |
4824 | { | |
4825 | switch (action & ~CPU_TASKS_FROZEN) { | |
4826 | case CPU_DOWN_PREPARE: | |
4827 | set_cpu_active((long)hcpu, false); | |
4828 | return NOTIFY_OK; | |
4829 | default: | |
4830 | return NOTIFY_DONE; | |
4831 | } | |
4832 | } | |
4833 | ||
7babe8db | 4834 | static int __init migration_init(void) |
1da177e4 LT |
4835 | { |
4836 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 4837 | int err; |
48f24c4d | 4838 | |
3a101d05 | 4839 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
4840 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
4841 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
4842 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
4843 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 4844 | |
3a101d05 TH |
4845 | /* Register cpu active notifiers */ |
4846 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
4847 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
4848 | ||
a004cd42 | 4849 | return 0; |
1da177e4 | 4850 | } |
7babe8db | 4851 | early_initcall(migration_init); |
1da177e4 LT |
4852 | #endif |
4853 | ||
4854 | #ifdef CONFIG_SMP | |
476f3534 | 4855 | |
4cb98839 PZ |
4856 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
4857 | ||
3e9830dc | 4858 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 4859 | |
d039ac60 | 4860 | static __read_mostly int sched_debug_enabled; |
f6630114 | 4861 | |
d039ac60 | 4862 | static int __init sched_debug_setup(char *str) |
f6630114 | 4863 | { |
d039ac60 | 4864 | sched_debug_enabled = 1; |
f6630114 MT |
4865 | |
4866 | return 0; | |
4867 | } | |
d039ac60 PZ |
4868 | early_param("sched_debug", sched_debug_setup); |
4869 | ||
4870 | static inline bool sched_debug(void) | |
4871 | { | |
4872 | return sched_debug_enabled; | |
4873 | } | |
f6630114 | 4874 | |
7c16ec58 | 4875 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 4876 | struct cpumask *groupmask) |
1da177e4 | 4877 | { |
4dcf6aff | 4878 | struct sched_group *group = sd->groups; |
434d53b0 | 4879 | char str[256]; |
1da177e4 | 4880 | |
968ea6d8 | 4881 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 4882 | cpumask_clear(groupmask); |
4dcf6aff IM |
4883 | |
4884 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
4885 | ||
4886 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 4887 | printk("does not load-balance\n"); |
4dcf6aff | 4888 | if (sd->parent) |
3df0fc5b PZ |
4889 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
4890 | " has parent"); | |
4dcf6aff | 4891 | return -1; |
41c7ce9a NP |
4892 | } |
4893 | ||
3df0fc5b | 4894 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 4895 | |
758b2cdc | 4896 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
4897 | printk(KERN_ERR "ERROR: domain->span does not contain " |
4898 | "CPU%d\n", cpu); | |
4dcf6aff | 4899 | } |
758b2cdc | 4900 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
4901 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
4902 | " CPU%d\n", cpu); | |
4dcf6aff | 4903 | } |
1da177e4 | 4904 | |
4dcf6aff | 4905 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 4906 | do { |
4dcf6aff | 4907 | if (!group) { |
3df0fc5b PZ |
4908 | printk("\n"); |
4909 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
4910 | break; |
4911 | } | |
4912 | ||
c3decf0d PZ |
4913 | /* |
4914 | * Even though we initialize ->power to something semi-sane, | |
4915 | * we leave power_orig unset. This allows us to detect if | |
4916 | * domain iteration is still funny without causing /0 traps. | |
4917 | */ | |
4918 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
4919 | printk(KERN_CONT "\n"); |
4920 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
4921 | "set\n"); | |
4dcf6aff IM |
4922 | break; |
4923 | } | |
1da177e4 | 4924 | |
758b2cdc | 4925 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
4926 | printk(KERN_CONT "\n"); |
4927 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
4928 | break; |
4929 | } | |
1da177e4 | 4930 | |
cb83b629 PZ |
4931 | if (!(sd->flags & SD_OVERLAP) && |
4932 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
4933 | printk(KERN_CONT "\n"); |
4934 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
4935 | break; |
4936 | } | |
1da177e4 | 4937 | |
758b2cdc | 4938 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 4939 | |
968ea6d8 | 4940 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 4941 | |
3df0fc5b | 4942 | printk(KERN_CONT " %s", str); |
9c3f75cb | 4943 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 4944 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 4945 | group->sgp->power); |
381512cf | 4946 | } |
1da177e4 | 4947 | |
4dcf6aff IM |
4948 | group = group->next; |
4949 | } while (group != sd->groups); | |
3df0fc5b | 4950 | printk(KERN_CONT "\n"); |
1da177e4 | 4951 | |
758b2cdc | 4952 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 4953 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 4954 | |
758b2cdc RR |
4955 | if (sd->parent && |
4956 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
4957 | printk(KERN_ERR "ERROR: parent span is not a superset " |
4958 | "of domain->span\n"); | |
4dcf6aff IM |
4959 | return 0; |
4960 | } | |
1da177e4 | 4961 | |
4dcf6aff IM |
4962 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
4963 | { | |
4964 | int level = 0; | |
1da177e4 | 4965 | |
d039ac60 | 4966 | if (!sched_debug_enabled) |
f6630114 MT |
4967 | return; |
4968 | ||
4dcf6aff IM |
4969 | if (!sd) { |
4970 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
4971 | return; | |
4972 | } | |
1da177e4 | 4973 | |
4dcf6aff IM |
4974 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
4975 | ||
4976 | for (;;) { | |
4cb98839 | 4977 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 4978 | break; |
1da177e4 LT |
4979 | level++; |
4980 | sd = sd->parent; | |
33859f7f | 4981 | if (!sd) |
4dcf6aff IM |
4982 | break; |
4983 | } | |
1da177e4 | 4984 | } |
6d6bc0ad | 4985 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 4986 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
4987 | static inline bool sched_debug(void) |
4988 | { | |
4989 | return false; | |
4990 | } | |
6d6bc0ad | 4991 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 4992 | |
1a20ff27 | 4993 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 4994 | { |
758b2cdc | 4995 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
4996 | return 1; |
4997 | ||
4998 | /* Following flags need at least 2 groups */ | |
4999 | if (sd->flags & (SD_LOAD_BALANCE | | |
5000 | SD_BALANCE_NEWIDLE | | |
5001 | SD_BALANCE_FORK | | |
89c4710e SS |
5002 | SD_BALANCE_EXEC | |
5003 | SD_SHARE_CPUPOWER | | |
5004 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5005 | if (sd->groups != sd->groups->next) |
5006 | return 0; | |
5007 | } | |
5008 | ||
5009 | /* Following flags don't use groups */ | |
c88d5910 | 5010 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5011 | return 0; |
5012 | ||
5013 | return 1; | |
5014 | } | |
5015 | ||
48f24c4d IM |
5016 | static int |
5017 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5018 | { |
5019 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5020 | ||
5021 | if (sd_degenerate(parent)) | |
5022 | return 1; | |
5023 | ||
758b2cdc | 5024 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5025 | return 0; |
5026 | ||
245af2c7 SS |
5027 | /* Flags needing groups don't count if only 1 group in parent */ |
5028 | if (parent->groups == parent->groups->next) { | |
5029 | pflags &= ~(SD_LOAD_BALANCE | | |
5030 | SD_BALANCE_NEWIDLE | | |
5031 | SD_BALANCE_FORK | | |
89c4710e SS |
5032 | SD_BALANCE_EXEC | |
5033 | SD_SHARE_CPUPOWER | | |
10866e62 PZ |
5034 | SD_SHARE_PKG_RESOURCES | |
5035 | SD_PREFER_SIBLING); | |
5436499e KC |
5036 | if (nr_node_ids == 1) |
5037 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5038 | } |
5039 | if (~cflags & pflags) | |
5040 | return 0; | |
5041 | ||
5042 | return 1; | |
5043 | } | |
5044 | ||
dce840a0 | 5045 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5046 | { |
dce840a0 | 5047 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5048 | |
68e74568 | 5049 | cpupri_cleanup(&rd->cpupri); |
1baca4ce | 5050 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5051 | free_cpumask_var(rd->rto_mask); |
5052 | free_cpumask_var(rd->online); | |
5053 | free_cpumask_var(rd->span); | |
5054 | kfree(rd); | |
5055 | } | |
5056 | ||
57d885fe GH |
5057 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5058 | { | |
a0490fa3 | 5059 | struct root_domain *old_rd = NULL; |
57d885fe | 5060 | unsigned long flags; |
57d885fe | 5061 | |
05fa785c | 5062 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5063 | |
5064 | if (rq->rd) { | |
a0490fa3 | 5065 | old_rd = rq->rd; |
57d885fe | 5066 | |
c6c4927b | 5067 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5068 | set_rq_offline(rq); |
57d885fe | 5069 | |
c6c4927b | 5070 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5071 | |
a0490fa3 | 5072 | /* |
0515973f | 5073 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5074 | * set old_rd to NULL to skip the freeing later |
5075 | * in this function: | |
5076 | */ | |
5077 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5078 | old_rd = NULL; | |
57d885fe GH |
5079 | } |
5080 | ||
5081 | atomic_inc(&rd->refcount); | |
5082 | rq->rd = rd; | |
5083 | ||
c6c4927b | 5084 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5085 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5086 | set_rq_online(rq); |
57d885fe | 5087 | |
05fa785c | 5088 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5089 | |
5090 | if (old_rd) | |
dce840a0 | 5091 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5092 | } |
5093 | ||
68c38fc3 | 5094 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5095 | { |
5096 | memset(rd, 0, sizeof(*rd)); | |
5097 | ||
68c38fc3 | 5098 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5099 | goto out; |
68c38fc3 | 5100 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5101 | goto free_span; |
1baca4ce | 5102 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5103 | goto free_online; |
1baca4ce JL |
5104 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5105 | goto free_dlo_mask; | |
6e0534f2 | 5106 | |
68c38fc3 | 5107 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5108 | goto free_rto_mask; |
c6c4927b | 5109 | return 0; |
6e0534f2 | 5110 | |
68e74568 RR |
5111 | free_rto_mask: |
5112 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5113 | free_dlo_mask: |
5114 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5115 | free_online: |
5116 | free_cpumask_var(rd->online); | |
5117 | free_span: | |
5118 | free_cpumask_var(rd->span); | |
0c910d28 | 5119 | out: |
c6c4927b | 5120 | return -ENOMEM; |
57d885fe GH |
5121 | } |
5122 | ||
029632fb PZ |
5123 | /* |
5124 | * By default the system creates a single root-domain with all cpus as | |
5125 | * members (mimicking the global state we have today). | |
5126 | */ | |
5127 | struct root_domain def_root_domain; | |
5128 | ||
57d885fe GH |
5129 | static void init_defrootdomain(void) |
5130 | { | |
68c38fc3 | 5131 | init_rootdomain(&def_root_domain); |
c6c4927b | 5132 | |
57d885fe GH |
5133 | atomic_set(&def_root_domain.refcount, 1); |
5134 | } | |
5135 | ||
dc938520 | 5136 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5137 | { |
5138 | struct root_domain *rd; | |
5139 | ||
5140 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5141 | if (!rd) | |
5142 | return NULL; | |
5143 | ||
68c38fc3 | 5144 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5145 | kfree(rd); |
5146 | return NULL; | |
5147 | } | |
57d885fe GH |
5148 | |
5149 | return rd; | |
5150 | } | |
5151 | ||
e3589f6c PZ |
5152 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5153 | { | |
5154 | struct sched_group *tmp, *first; | |
5155 | ||
5156 | if (!sg) | |
5157 | return; | |
5158 | ||
5159 | first = sg; | |
5160 | do { | |
5161 | tmp = sg->next; | |
5162 | ||
5163 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5164 | kfree(sg->sgp); | |
5165 | ||
5166 | kfree(sg); | |
5167 | sg = tmp; | |
5168 | } while (sg != first); | |
5169 | } | |
5170 | ||
dce840a0 PZ |
5171 | static void free_sched_domain(struct rcu_head *rcu) |
5172 | { | |
5173 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5174 | |
5175 | /* | |
5176 | * If its an overlapping domain it has private groups, iterate and | |
5177 | * nuke them all. | |
5178 | */ | |
5179 | if (sd->flags & SD_OVERLAP) { | |
5180 | free_sched_groups(sd->groups, 1); | |
5181 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5182 | kfree(sd->groups->sgp); |
dce840a0 | 5183 | kfree(sd->groups); |
9c3f75cb | 5184 | } |
dce840a0 PZ |
5185 | kfree(sd); |
5186 | } | |
5187 | ||
5188 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5189 | { | |
5190 | call_rcu(&sd->rcu, free_sched_domain); | |
5191 | } | |
5192 | ||
5193 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5194 | { | |
5195 | for (; sd; sd = sd->parent) | |
5196 | destroy_sched_domain(sd, cpu); | |
5197 | } | |
5198 | ||
518cd623 PZ |
5199 | /* |
5200 | * Keep a special pointer to the highest sched_domain that has | |
5201 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5202 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5203 | * | |
5204 | * Also keep a unique ID per domain (we use the first cpu number in | |
5205 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5206 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5207 | */ |
5208 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5209 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5210 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5211 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5212 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5213 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5214 | |
5215 | static void update_top_cache_domain(int cpu) | |
5216 | { | |
5217 | struct sched_domain *sd; | |
5d4cf996 | 5218 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5219 | int id = cpu; |
7d9ffa89 | 5220 | int size = 1; |
518cd623 PZ |
5221 | |
5222 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5223 | if (sd) { |
518cd623 | 5224 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5225 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5226 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5227 | } |
5d4cf996 | 5228 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5229 | |
5230 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5231 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5232 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5233 | |
5234 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5235 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5236 | |
5237 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5238 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5239 | } |
5240 | ||
1da177e4 | 5241 | /* |
0eab9146 | 5242 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5243 | * hold the hotplug lock. |
5244 | */ | |
0eab9146 IM |
5245 | static void |
5246 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5247 | { |
70b97a7f | 5248 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5249 | struct sched_domain *tmp; |
5250 | ||
5251 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5252 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5253 | struct sched_domain *parent = tmp->parent; |
5254 | if (!parent) | |
5255 | break; | |
f29c9b1c | 5256 | |
1a848870 | 5257 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5258 | tmp->parent = parent->parent; |
1a848870 SS |
5259 | if (parent->parent) |
5260 | parent->parent->child = tmp; | |
10866e62 PZ |
5261 | /* |
5262 | * Transfer SD_PREFER_SIBLING down in case of a | |
5263 | * degenerate parent; the spans match for this | |
5264 | * so the property transfers. | |
5265 | */ | |
5266 | if (parent->flags & SD_PREFER_SIBLING) | |
5267 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5268 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5269 | } else |
5270 | tmp = tmp->parent; | |
245af2c7 SS |
5271 | } |
5272 | ||
1a848870 | 5273 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5274 | tmp = sd; |
245af2c7 | 5275 | sd = sd->parent; |
dce840a0 | 5276 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5277 | if (sd) |
5278 | sd->child = NULL; | |
5279 | } | |
1da177e4 | 5280 | |
4cb98839 | 5281 | sched_domain_debug(sd, cpu); |
1da177e4 | 5282 | |
57d885fe | 5283 | rq_attach_root(rq, rd); |
dce840a0 | 5284 | tmp = rq->sd; |
674311d5 | 5285 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5286 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5287 | |
5288 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5289 | } |
5290 | ||
5291 | /* cpus with isolated domains */ | |
dcc30a35 | 5292 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5293 | |
5294 | /* Setup the mask of cpus configured for isolated domains */ | |
5295 | static int __init isolated_cpu_setup(char *str) | |
5296 | { | |
bdddd296 | 5297 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5298 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5299 | return 1; |
5300 | } | |
5301 | ||
8927f494 | 5302 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5303 | |
d3081f52 PZ |
5304 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5305 | { | |
5306 | return cpumask_of_node(cpu_to_node(cpu)); | |
5307 | } | |
5308 | ||
dce840a0 PZ |
5309 | struct sd_data { |
5310 | struct sched_domain **__percpu sd; | |
5311 | struct sched_group **__percpu sg; | |
9c3f75cb | 5312 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5313 | }; |
5314 | ||
49a02c51 | 5315 | struct s_data { |
21d42ccf | 5316 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5317 | struct root_domain *rd; |
5318 | }; | |
5319 | ||
2109b99e | 5320 | enum s_alloc { |
2109b99e | 5321 | sa_rootdomain, |
21d42ccf | 5322 | sa_sd, |
dce840a0 | 5323 | sa_sd_storage, |
2109b99e AH |
5324 | sa_none, |
5325 | }; | |
5326 | ||
54ab4ff4 PZ |
5327 | struct sched_domain_topology_level; |
5328 | ||
5329 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5330 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5331 | ||
e3589f6c PZ |
5332 | #define SDTL_OVERLAP 0x01 |
5333 | ||
eb7a74e6 | 5334 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5335 | sched_domain_init_f init; |
5336 | sched_domain_mask_f mask; | |
e3589f6c | 5337 | int flags; |
cb83b629 | 5338 | int numa_level; |
54ab4ff4 | 5339 | struct sd_data data; |
eb7a74e6 PZ |
5340 | }; |
5341 | ||
c1174876 PZ |
5342 | /* |
5343 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5344 | * domain traversal. | |
5345 | * | |
5346 | * Asymmetric node setups can result in situations where the domain tree is of | |
5347 | * unequal depth, make sure to skip domains that already cover the entire | |
5348 | * range. | |
5349 | * | |
5350 | * In that case build_sched_domains() will have terminated the iteration early | |
5351 | * and our sibling sd spans will be empty. Domains should always include the | |
5352 | * cpu they're built on, so check that. | |
5353 | * | |
5354 | */ | |
5355 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5356 | { | |
5357 | const struct cpumask *span = sched_domain_span(sd); | |
5358 | struct sd_data *sdd = sd->private; | |
5359 | struct sched_domain *sibling; | |
5360 | int i; | |
5361 | ||
5362 | for_each_cpu(i, span) { | |
5363 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5364 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5365 | continue; | |
5366 | ||
5367 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5368 | } | |
5369 | } | |
5370 | ||
5371 | /* | |
5372 | * Return the canonical balance cpu for this group, this is the first cpu | |
5373 | * of this group that's also in the iteration mask. | |
5374 | */ | |
5375 | int group_balance_cpu(struct sched_group *sg) | |
5376 | { | |
5377 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5378 | } | |
5379 | ||
e3589f6c PZ |
5380 | static int |
5381 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5382 | { | |
5383 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5384 | const struct cpumask *span = sched_domain_span(sd); | |
5385 | struct cpumask *covered = sched_domains_tmpmask; | |
5386 | struct sd_data *sdd = sd->private; | |
5387 | struct sched_domain *child; | |
5388 | int i; | |
5389 | ||
5390 | cpumask_clear(covered); | |
5391 | ||
5392 | for_each_cpu(i, span) { | |
5393 | struct cpumask *sg_span; | |
5394 | ||
5395 | if (cpumask_test_cpu(i, covered)) | |
5396 | continue; | |
5397 | ||
c1174876 PZ |
5398 | child = *per_cpu_ptr(sdd->sd, i); |
5399 | ||
5400 | /* See the comment near build_group_mask(). */ | |
5401 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5402 | continue; | |
5403 | ||
e3589f6c | 5404 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5405 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5406 | |
5407 | if (!sg) | |
5408 | goto fail; | |
5409 | ||
5410 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5411 | if (child->child) { |
5412 | child = child->child; | |
5413 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5414 | } else | |
5415 | cpumask_set_cpu(i, sg_span); | |
5416 | ||
5417 | cpumask_or(covered, covered, sg_span); | |
5418 | ||
74a5ce20 | 5419 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5420 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5421 | build_group_mask(sd, sg); | |
5422 | ||
c3decf0d PZ |
5423 | /* |
5424 | * Initialize sgp->power such that even if we mess up the | |
5425 | * domains and no possible iteration will get us here, we won't | |
5426 | * die on a /0 trap. | |
5427 | */ | |
5428 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
8e8339a3 | 5429 | sg->sgp->power_orig = sg->sgp->power; |
e3589f6c | 5430 | |
c1174876 PZ |
5431 | /* |
5432 | * Make sure the first group of this domain contains the | |
5433 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5434 | * breaks. See update_sg_lb_stats(). | |
5435 | */ | |
74a5ce20 | 5436 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5437 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5438 | groups = sg; |
5439 | ||
5440 | if (!first) | |
5441 | first = sg; | |
5442 | if (last) | |
5443 | last->next = sg; | |
5444 | last = sg; | |
5445 | last->next = first; | |
5446 | } | |
5447 | sd->groups = groups; | |
5448 | ||
5449 | return 0; | |
5450 | ||
5451 | fail: | |
5452 | free_sched_groups(first, 0); | |
5453 | ||
5454 | return -ENOMEM; | |
5455 | } | |
5456 | ||
dce840a0 | 5457 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5458 | { |
dce840a0 PZ |
5459 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5460 | struct sched_domain *child = sd->child; | |
1da177e4 | 5461 | |
dce840a0 PZ |
5462 | if (child) |
5463 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5464 | |
9c3f75cb | 5465 | if (sg) { |
dce840a0 | 5466 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5467 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5468 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5469 | } |
dce840a0 PZ |
5470 | |
5471 | return cpu; | |
1e9f28fa | 5472 | } |
1e9f28fa | 5473 | |
01a08546 | 5474 | /* |
dce840a0 PZ |
5475 | * build_sched_groups will build a circular linked list of the groups |
5476 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5477 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5478 | * |
5479 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5480 | */ |
e3589f6c PZ |
5481 | static int |
5482 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5483 | { |
dce840a0 PZ |
5484 | struct sched_group *first = NULL, *last = NULL; |
5485 | struct sd_data *sdd = sd->private; | |
5486 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5487 | struct cpumask *covered; |
dce840a0 | 5488 | int i; |
9c1cfda2 | 5489 | |
e3589f6c PZ |
5490 | get_group(cpu, sdd, &sd->groups); |
5491 | atomic_inc(&sd->groups->ref); | |
5492 | ||
0936629f | 5493 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5494 | return 0; |
5495 | ||
f96225fd PZ |
5496 | lockdep_assert_held(&sched_domains_mutex); |
5497 | covered = sched_domains_tmpmask; | |
5498 | ||
dce840a0 | 5499 | cpumask_clear(covered); |
6711cab4 | 5500 | |
dce840a0 PZ |
5501 | for_each_cpu(i, span) { |
5502 | struct sched_group *sg; | |
cd08e923 | 5503 | int group, j; |
6711cab4 | 5504 | |
dce840a0 PZ |
5505 | if (cpumask_test_cpu(i, covered)) |
5506 | continue; | |
6711cab4 | 5507 | |
cd08e923 | 5508 | group = get_group(i, sdd, &sg); |
dce840a0 | 5509 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5510 | sg->sgp->power = 0; |
c1174876 | 5511 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5512 | |
dce840a0 PZ |
5513 | for_each_cpu(j, span) { |
5514 | if (get_group(j, sdd, NULL) != group) | |
5515 | continue; | |
0601a88d | 5516 | |
dce840a0 PZ |
5517 | cpumask_set_cpu(j, covered); |
5518 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5519 | } | |
0601a88d | 5520 | |
dce840a0 PZ |
5521 | if (!first) |
5522 | first = sg; | |
5523 | if (last) | |
5524 | last->next = sg; | |
5525 | last = sg; | |
5526 | } | |
5527 | last->next = first; | |
e3589f6c PZ |
5528 | |
5529 | return 0; | |
0601a88d | 5530 | } |
51888ca2 | 5531 | |
89c4710e SS |
5532 | /* |
5533 | * Initialize sched groups cpu_power. | |
5534 | * | |
5535 | * cpu_power indicates the capacity of sched group, which is used while | |
5536 | * distributing the load between different sched groups in a sched domain. | |
5537 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5538 | * there are asymmetries in the topology. If there are asymmetries, group | |
5539 | * having more cpu_power will pickup more load compared to the group having | |
5540 | * less cpu_power. | |
89c4710e SS |
5541 | */ |
5542 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5543 | { | |
e3589f6c | 5544 | struct sched_group *sg = sd->groups; |
89c4710e | 5545 | |
94c95ba6 | 5546 | WARN_ON(!sg); |
e3589f6c PZ |
5547 | |
5548 | do { | |
5549 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5550 | sg = sg->next; | |
5551 | } while (sg != sd->groups); | |
89c4710e | 5552 | |
c1174876 | 5553 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5554 | return; |
aae6d3dd | 5555 | |
d274cb30 | 5556 | update_group_power(sd, cpu); |
69e1e811 | 5557 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5558 | } |
5559 | ||
029632fb PZ |
5560 | int __weak arch_sd_sibling_asym_packing(void) |
5561 | { | |
5562 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5563 | } |
5564 | ||
7c16ec58 MT |
5565 | /* |
5566 | * Initializers for schedule domains | |
5567 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5568 | */ | |
5569 | ||
a5d8c348 IM |
5570 | #ifdef CONFIG_SCHED_DEBUG |
5571 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5572 | #else | |
5573 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5574 | #endif | |
5575 | ||
54ab4ff4 PZ |
5576 | #define SD_INIT_FUNC(type) \ |
5577 | static noinline struct sched_domain * \ | |
5578 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5579 | { \ | |
5580 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5581 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5582 | SD_INIT_NAME(sd, type); \ |
5583 | sd->private = &tl->data; \ | |
5584 | return sd; \ | |
7c16ec58 MT |
5585 | } |
5586 | ||
5587 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5588 | #ifdef CONFIG_SCHED_SMT |
5589 | SD_INIT_FUNC(SIBLING) | |
5590 | #endif | |
5591 | #ifdef CONFIG_SCHED_MC | |
5592 | SD_INIT_FUNC(MC) | |
5593 | #endif | |
01a08546 HC |
5594 | #ifdef CONFIG_SCHED_BOOK |
5595 | SD_INIT_FUNC(BOOK) | |
5596 | #endif | |
7c16ec58 | 5597 | |
1d3504fc | 5598 | static int default_relax_domain_level = -1; |
60495e77 | 5599 | int sched_domain_level_max; |
1d3504fc HS |
5600 | |
5601 | static int __init setup_relax_domain_level(char *str) | |
5602 | { | |
a841f8ce DS |
5603 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5604 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5605 | |
1d3504fc HS |
5606 | return 1; |
5607 | } | |
5608 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5609 | ||
5610 | static void set_domain_attribute(struct sched_domain *sd, | |
5611 | struct sched_domain_attr *attr) | |
5612 | { | |
5613 | int request; | |
5614 | ||
5615 | if (!attr || attr->relax_domain_level < 0) { | |
5616 | if (default_relax_domain_level < 0) | |
5617 | return; | |
5618 | else | |
5619 | request = default_relax_domain_level; | |
5620 | } else | |
5621 | request = attr->relax_domain_level; | |
5622 | if (request < sd->level) { | |
5623 | /* turn off idle balance on this domain */ | |
c88d5910 | 5624 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5625 | } else { |
5626 | /* turn on idle balance on this domain */ | |
c88d5910 | 5627 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5628 | } |
5629 | } | |
5630 | ||
54ab4ff4 PZ |
5631 | static void __sdt_free(const struct cpumask *cpu_map); |
5632 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5633 | ||
2109b99e AH |
5634 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5635 | const struct cpumask *cpu_map) | |
5636 | { | |
5637 | switch (what) { | |
2109b99e | 5638 | case sa_rootdomain: |
822ff793 PZ |
5639 | if (!atomic_read(&d->rd->refcount)) |
5640 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5641 | case sa_sd: |
5642 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5643 | case sa_sd_storage: |
54ab4ff4 | 5644 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5645 | case sa_none: |
5646 | break; | |
5647 | } | |
5648 | } | |
3404c8d9 | 5649 | |
2109b99e AH |
5650 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5651 | const struct cpumask *cpu_map) | |
5652 | { | |
dce840a0 PZ |
5653 | memset(d, 0, sizeof(*d)); |
5654 | ||
54ab4ff4 PZ |
5655 | if (__sdt_alloc(cpu_map)) |
5656 | return sa_sd_storage; | |
dce840a0 PZ |
5657 | d->sd = alloc_percpu(struct sched_domain *); |
5658 | if (!d->sd) | |
5659 | return sa_sd_storage; | |
2109b99e | 5660 | d->rd = alloc_rootdomain(); |
dce840a0 | 5661 | if (!d->rd) |
21d42ccf | 5662 | return sa_sd; |
2109b99e AH |
5663 | return sa_rootdomain; |
5664 | } | |
57d885fe | 5665 | |
dce840a0 PZ |
5666 | /* |
5667 | * NULL the sd_data elements we've used to build the sched_domain and | |
5668 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5669 | * will not free the data we're using. | |
5670 | */ | |
5671 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5672 | { | |
5673 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5674 | |
5675 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5676 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5677 | ||
e3589f6c | 5678 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5679 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5680 | |
5681 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5682 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5683 | } |
5684 | ||
2c402dc3 PZ |
5685 | #ifdef CONFIG_SCHED_SMT |
5686 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 5687 | { |
2c402dc3 | 5688 | return topology_thread_cpumask(cpu); |
3bd65a80 | 5689 | } |
2c402dc3 | 5690 | #endif |
7f4588f3 | 5691 | |
d069b916 PZ |
5692 | /* |
5693 | * Topology list, bottom-up. | |
5694 | */ | |
2c402dc3 | 5695 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
5696 | #ifdef CONFIG_SCHED_SMT |
5697 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 5698 | #endif |
1e9f28fa | 5699 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 5700 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 5701 | #endif |
d069b916 PZ |
5702 | #ifdef CONFIG_SCHED_BOOK |
5703 | { sd_init_BOOK, cpu_book_mask, }, | |
5704 | #endif | |
5705 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
5706 | { NULL, }, |
5707 | }; | |
5708 | ||
5709 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
5710 | ||
27723a68 VK |
5711 | #define for_each_sd_topology(tl) \ |
5712 | for (tl = sched_domain_topology; tl->init; tl++) | |
5713 | ||
cb83b629 PZ |
5714 | #ifdef CONFIG_NUMA |
5715 | ||
5716 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
5717 | static int *sched_domains_numa_distance; |
5718 | static struct cpumask ***sched_domains_numa_masks; | |
5719 | static int sched_domains_curr_level; | |
5720 | ||
cb83b629 PZ |
5721 | static inline int sd_local_flags(int level) |
5722 | { | |
10717dcd | 5723 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
5724 | return 0; |
5725 | ||
5726 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
5727 | } | |
5728 | ||
5729 | static struct sched_domain * | |
5730 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
5731 | { | |
5732 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
5733 | int level = tl->numa_level; | |
5734 | int sd_weight = cpumask_weight( | |
5735 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
5736 | ||
5737 | *sd = (struct sched_domain){ | |
5738 | .min_interval = sd_weight, | |
5739 | .max_interval = 2*sd_weight, | |
5740 | .busy_factor = 32, | |
870a0bb5 | 5741 | .imbalance_pct = 125, |
cb83b629 PZ |
5742 | .cache_nice_tries = 2, |
5743 | .busy_idx = 3, | |
5744 | .idle_idx = 2, | |
5745 | .newidle_idx = 0, | |
5746 | .wake_idx = 0, | |
5747 | .forkexec_idx = 0, | |
5748 | ||
5749 | .flags = 1*SD_LOAD_BALANCE | |
5750 | | 1*SD_BALANCE_NEWIDLE | |
5751 | | 0*SD_BALANCE_EXEC | |
5752 | | 0*SD_BALANCE_FORK | |
5753 | | 0*SD_BALANCE_WAKE | |
5754 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 5755 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
5756 | | 0*SD_SHARE_PKG_RESOURCES |
5757 | | 1*SD_SERIALIZE | |
5758 | | 0*SD_PREFER_SIBLING | |
3a7053b3 | 5759 | | 1*SD_NUMA |
cb83b629 PZ |
5760 | | sd_local_flags(level) |
5761 | , | |
5762 | .last_balance = jiffies, | |
5763 | .balance_interval = sd_weight, | |
5764 | }; | |
5765 | SD_INIT_NAME(sd, NUMA); | |
5766 | sd->private = &tl->data; | |
5767 | ||
5768 | /* | |
5769 | * Ugly hack to pass state to sd_numa_mask()... | |
5770 | */ | |
5771 | sched_domains_curr_level = tl->numa_level; | |
5772 | ||
5773 | return sd; | |
5774 | } | |
5775 | ||
5776 | static const struct cpumask *sd_numa_mask(int cpu) | |
5777 | { | |
5778 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
5779 | } | |
5780 | ||
d039ac60 PZ |
5781 | static void sched_numa_warn(const char *str) |
5782 | { | |
5783 | static int done = false; | |
5784 | int i,j; | |
5785 | ||
5786 | if (done) | |
5787 | return; | |
5788 | ||
5789 | done = true; | |
5790 | ||
5791 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
5792 | ||
5793 | for (i = 0; i < nr_node_ids; i++) { | |
5794 | printk(KERN_WARNING " "); | |
5795 | for (j = 0; j < nr_node_ids; j++) | |
5796 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
5797 | printk(KERN_CONT "\n"); | |
5798 | } | |
5799 | printk(KERN_WARNING "\n"); | |
5800 | } | |
5801 | ||
5802 | static bool find_numa_distance(int distance) | |
5803 | { | |
5804 | int i; | |
5805 | ||
5806 | if (distance == node_distance(0, 0)) | |
5807 | return true; | |
5808 | ||
5809 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5810 | if (sched_domains_numa_distance[i] == distance) | |
5811 | return true; | |
5812 | } | |
5813 | ||
5814 | return false; | |
5815 | } | |
5816 | ||
cb83b629 PZ |
5817 | static void sched_init_numa(void) |
5818 | { | |
5819 | int next_distance, curr_distance = node_distance(0, 0); | |
5820 | struct sched_domain_topology_level *tl; | |
5821 | int level = 0; | |
5822 | int i, j, k; | |
5823 | ||
cb83b629 PZ |
5824 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
5825 | if (!sched_domains_numa_distance) | |
5826 | return; | |
5827 | ||
5828 | /* | |
5829 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
5830 | * unique distances in the node_distance() table. | |
5831 | * | |
5832 | * Assumes node_distance(0,j) includes all distances in | |
5833 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
5834 | */ |
5835 | next_distance = curr_distance; | |
5836 | for (i = 0; i < nr_node_ids; i++) { | |
5837 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
5838 | for (k = 0; k < nr_node_ids; k++) { |
5839 | int distance = node_distance(i, k); | |
5840 | ||
5841 | if (distance > curr_distance && | |
5842 | (distance < next_distance || | |
5843 | next_distance == curr_distance)) | |
5844 | next_distance = distance; | |
5845 | ||
5846 | /* | |
5847 | * While not a strong assumption it would be nice to know | |
5848 | * about cases where if node A is connected to B, B is not | |
5849 | * equally connected to A. | |
5850 | */ | |
5851 | if (sched_debug() && node_distance(k, i) != distance) | |
5852 | sched_numa_warn("Node-distance not symmetric"); | |
5853 | ||
5854 | if (sched_debug() && i && !find_numa_distance(distance)) | |
5855 | sched_numa_warn("Node-0 not representative"); | |
5856 | } | |
5857 | if (next_distance != curr_distance) { | |
5858 | sched_domains_numa_distance[level++] = next_distance; | |
5859 | sched_domains_numa_levels = level; | |
5860 | curr_distance = next_distance; | |
5861 | } else break; | |
cb83b629 | 5862 | } |
d039ac60 PZ |
5863 | |
5864 | /* | |
5865 | * In case of sched_debug() we verify the above assumption. | |
5866 | */ | |
5867 | if (!sched_debug()) | |
5868 | break; | |
cb83b629 PZ |
5869 | } |
5870 | /* | |
5871 | * 'level' contains the number of unique distances, excluding the | |
5872 | * identity distance node_distance(i,i). | |
5873 | * | |
28b4a521 | 5874 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
5875 | * numbers. |
5876 | */ | |
5877 | ||
5f7865f3 TC |
5878 | /* |
5879 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
5880 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
5881 | * the array will contain less then 'level' members. This could be | |
5882 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
5883 | * in other functions. | |
5884 | * | |
5885 | * We reset it to 'level' at the end of this function. | |
5886 | */ | |
5887 | sched_domains_numa_levels = 0; | |
5888 | ||
cb83b629 PZ |
5889 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
5890 | if (!sched_domains_numa_masks) | |
5891 | return; | |
5892 | ||
5893 | /* | |
5894 | * Now for each level, construct a mask per node which contains all | |
5895 | * cpus of nodes that are that many hops away from us. | |
5896 | */ | |
5897 | for (i = 0; i < level; i++) { | |
5898 | sched_domains_numa_masks[i] = | |
5899 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
5900 | if (!sched_domains_numa_masks[i]) | |
5901 | return; | |
5902 | ||
5903 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 5904 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
5905 | if (!mask) |
5906 | return; | |
5907 | ||
5908 | sched_domains_numa_masks[i][j] = mask; | |
5909 | ||
5910 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 5911 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
5912 | continue; |
5913 | ||
5914 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
5915 | } | |
5916 | } | |
5917 | } | |
5918 | ||
5919 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
5920 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
5921 | if (!tl) | |
5922 | return; | |
5923 | ||
5924 | /* | |
5925 | * Copy the default topology bits.. | |
5926 | */ | |
5927 | for (i = 0; default_topology[i].init; i++) | |
5928 | tl[i] = default_topology[i]; | |
5929 | ||
5930 | /* | |
5931 | * .. and append 'j' levels of NUMA goodness. | |
5932 | */ | |
5933 | for (j = 0; j < level; i++, j++) { | |
5934 | tl[i] = (struct sched_domain_topology_level){ | |
5935 | .init = sd_numa_init, | |
5936 | .mask = sd_numa_mask, | |
5937 | .flags = SDTL_OVERLAP, | |
5938 | .numa_level = j, | |
5939 | }; | |
5940 | } | |
5941 | ||
5942 | sched_domain_topology = tl; | |
5f7865f3 TC |
5943 | |
5944 | sched_domains_numa_levels = level; | |
cb83b629 | 5945 | } |
301a5cba TC |
5946 | |
5947 | static void sched_domains_numa_masks_set(int cpu) | |
5948 | { | |
5949 | int i, j; | |
5950 | int node = cpu_to_node(cpu); | |
5951 | ||
5952 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5953 | for (j = 0; j < nr_node_ids; j++) { | |
5954 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
5955 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
5956 | } | |
5957 | } | |
5958 | } | |
5959 | ||
5960 | static void sched_domains_numa_masks_clear(int cpu) | |
5961 | { | |
5962 | int i, j; | |
5963 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
5964 | for (j = 0; j < nr_node_ids; j++) | |
5965 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
5966 | } | |
5967 | } | |
5968 | ||
5969 | /* | |
5970 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
5971 | * are onlined. | |
5972 | */ | |
5973 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
5974 | unsigned long action, | |
5975 | void *hcpu) | |
5976 | { | |
5977 | int cpu = (long)hcpu; | |
5978 | ||
5979 | switch (action & ~CPU_TASKS_FROZEN) { | |
5980 | case CPU_ONLINE: | |
5981 | sched_domains_numa_masks_set(cpu); | |
5982 | break; | |
5983 | ||
5984 | case CPU_DEAD: | |
5985 | sched_domains_numa_masks_clear(cpu); | |
5986 | break; | |
5987 | ||
5988 | default: | |
5989 | return NOTIFY_DONE; | |
5990 | } | |
5991 | ||
5992 | return NOTIFY_OK; | |
cb83b629 PZ |
5993 | } |
5994 | #else | |
5995 | static inline void sched_init_numa(void) | |
5996 | { | |
5997 | } | |
301a5cba TC |
5998 | |
5999 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6000 | unsigned long action, | |
6001 | void *hcpu) | |
6002 | { | |
6003 | return 0; | |
6004 | } | |
cb83b629 PZ |
6005 | #endif /* CONFIG_NUMA */ |
6006 | ||
54ab4ff4 PZ |
6007 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6008 | { | |
6009 | struct sched_domain_topology_level *tl; | |
6010 | int j; | |
6011 | ||
27723a68 | 6012 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6013 | struct sd_data *sdd = &tl->data; |
6014 | ||
6015 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6016 | if (!sdd->sd) | |
6017 | return -ENOMEM; | |
6018 | ||
6019 | sdd->sg = alloc_percpu(struct sched_group *); | |
6020 | if (!sdd->sg) | |
6021 | return -ENOMEM; | |
6022 | ||
9c3f75cb PZ |
6023 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6024 | if (!sdd->sgp) | |
6025 | return -ENOMEM; | |
6026 | ||
54ab4ff4 PZ |
6027 | for_each_cpu(j, cpu_map) { |
6028 | struct sched_domain *sd; | |
6029 | struct sched_group *sg; | |
9c3f75cb | 6030 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6031 | |
6032 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6033 | GFP_KERNEL, cpu_to_node(j)); | |
6034 | if (!sd) | |
6035 | return -ENOMEM; | |
6036 | ||
6037 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6038 | ||
6039 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6040 | GFP_KERNEL, cpu_to_node(j)); | |
6041 | if (!sg) | |
6042 | return -ENOMEM; | |
6043 | ||
30b4e9eb IM |
6044 | sg->next = sg; |
6045 | ||
54ab4ff4 | 6046 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6047 | |
c1174876 | 6048 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6049 | GFP_KERNEL, cpu_to_node(j)); |
6050 | if (!sgp) | |
6051 | return -ENOMEM; | |
6052 | ||
6053 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6054 | } |
6055 | } | |
6056 | ||
6057 | return 0; | |
6058 | } | |
6059 | ||
6060 | static void __sdt_free(const struct cpumask *cpu_map) | |
6061 | { | |
6062 | struct sched_domain_topology_level *tl; | |
6063 | int j; | |
6064 | ||
27723a68 | 6065 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6066 | struct sd_data *sdd = &tl->data; |
6067 | ||
6068 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6069 | struct sched_domain *sd; |
6070 | ||
6071 | if (sdd->sd) { | |
6072 | sd = *per_cpu_ptr(sdd->sd, j); | |
6073 | if (sd && (sd->flags & SD_OVERLAP)) | |
6074 | free_sched_groups(sd->groups, 0); | |
6075 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6076 | } | |
6077 | ||
6078 | if (sdd->sg) | |
6079 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6080 | if (sdd->sgp) | |
6081 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6082 | } |
6083 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6084 | sdd->sd = NULL; |
54ab4ff4 | 6085 | free_percpu(sdd->sg); |
fb2cf2c6 | 6086 | sdd->sg = NULL; |
9c3f75cb | 6087 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6088 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6089 | } |
6090 | } | |
6091 | ||
2c402dc3 | 6092 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6093 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6094 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6095 | { |
54ab4ff4 | 6096 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6097 | if (!sd) |
d069b916 | 6098 | return child; |
2c402dc3 | 6099 | |
2c402dc3 | 6100 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6101 | if (child) { |
6102 | sd->level = child->level + 1; | |
6103 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6104 | child->parent = sd; |
c75e0128 | 6105 | sd->child = child; |
60495e77 | 6106 | } |
a841f8ce | 6107 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6108 | |
6109 | return sd; | |
6110 | } | |
6111 | ||
2109b99e AH |
6112 | /* |
6113 | * Build sched domains for a given set of cpus and attach the sched domains | |
6114 | * to the individual cpus | |
6115 | */ | |
dce840a0 PZ |
6116 | static int build_sched_domains(const struct cpumask *cpu_map, |
6117 | struct sched_domain_attr *attr) | |
2109b99e | 6118 | { |
1c632169 | 6119 | enum s_alloc alloc_state; |
dce840a0 | 6120 | struct sched_domain *sd; |
2109b99e | 6121 | struct s_data d; |
822ff793 | 6122 | int i, ret = -ENOMEM; |
9c1cfda2 | 6123 | |
2109b99e AH |
6124 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6125 | if (alloc_state != sa_rootdomain) | |
6126 | goto error; | |
9c1cfda2 | 6127 | |
dce840a0 | 6128 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6129 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6130 | struct sched_domain_topology_level *tl; |
6131 | ||
3bd65a80 | 6132 | sd = NULL; |
27723a68 | 6133 | for_each_sd_topology(tl) { |
4a850cbe | 6134 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6135 | if (tl == sched_domain_topology) |
6136 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6137 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6138 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6139 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6140 | break; | |
e3589f6c | 6141 | } |
dce840a0 PZ |
6142 | } |
6143 | ||
6144 | /* Build the groups for the domains */ | |
6145 | for_each_cpu(i, cpu_map) { | |
6146 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6147 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6148 | if (sd->flags & SD_OVERLAP) { |
6149 | if (build_overlap_sched_groups(sd, i)) | |
6150 | goto error; | |
6151 | } else { | |
6152 | if (build_sched_groups(sd, i)) | |
6153 | goto error; | |
6154 | } | |
1cf51902 | 6155 | } |
a06dadbe | 6156 | } |
9c1cfda2 | 6157 | |
1da177e4 | 6158 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6159 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6160 | if (!cpumask_test_cpu(i, cpu_map)) | |
6161 | continue; | |
9c1cfda2 | 6162 | |
dce840a0 PZ |
6163 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6164 | claim_allocations(i, sd); | |
cd4ea6ae | 6165 | init_sched_groups_power(i, sd); |
dce840a0 | 6166 | } |
f712c0c7 | 6167 | } |
9c1cfda2 | 6168 | |
1da177e4 | 6169 | /* Attach the domains */ |
dce840a0 | 6170 | rcu_read_lock(); |
abcd083a | 6171 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6172 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6173 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6174 | } |
dce840a0 | 6175 | rcu_read_unlock(); |
51888ca2 | 6176 | |
822ff793 | 6177 | ret = 0; |
51888ca2 | 6178 | error: |
2109b99e | 6179 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6180 | return ret; |
1da177e4 | 6181 | } |
029190c5 | 6182 | |
acc3f5d7 | 6183 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6184 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6185 | static struct sched_domain_attr *dattr_cur; |
6186 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6187 | |
6188 | /* | |
6189 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6190 | * cpumask) fails, then fallback to a single sched domain, |
6191 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6192 | */ |
4212823f | 6193 | static cpumask_var_t fallback_doms; |
029190c5 | 6194 | |
ee79d1bd HC |
6195 | /* |
6196 | * arch_update_cpu_topology lets virtualized architectures update the | |
6197 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6198 | * or 0 if it stayed the same. | |
6199 | */ | |
6200 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6201 | { |
ee79d1bd | 6202 | return 0; |
22e52b07 HC |
6203 | } |
6204 | ||
acc3f5d7 RR |
6205 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6206 | { | |
6207 | int i; | |
6208 | cpumask_var_t *doms; | |
6209 | ||
6210 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6211 | if (!doms) | |
6212 | return NULL; | |
6213 | for (i = 0; i < ndoms; i++) { | |
6214 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6215 | free_sched_domains(doms, i); | |
6216 | return NULL; | |
6217 | } | |
6218 | } | |
6219 | return doms; | |
6220 | } | |
6221 | ||
6222 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6223 | { | |
6224 | unsigned int i; | |
6225 | for (i = 0; i < ndoms; i++) | |
6226 | free_cpumask_var(doms[i]); | |
6227 | kfree(doms); | |
6228 | } | |
6229 | ||
1a20ff27 | 6230 | /* |
41a2d6cf | 6231 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6232 | * For now this just excludes isolated cpus, but could be used to |
6233 | * exclude other special cases in the future. | |
1a20ff27 | 6234 | */ |
c4a8849a | 6235 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6236 | { |
7378547f MM |
6237 | int err; |
6238 | ||
22e52b07 | 6239 | arch_update_cpu_topology(); |
029190c5 | 6240 | ndoms_cur = 1; |
acc3f5d7 | 6241 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6242 | if (!doms_cur) |
acc3f5d7 RR |
6243 | doms_cur = &fallback_doms; |
6244 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6245 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6246 | register_sched_domain_sysctl(); |
7378547f MM |
6247 | |
6248 | return err; | |
1a20ff27 DG |
6249 | } |
6250 | ||
1a20ff27 DG |
6251 | /* |
6252 | * Detach sched domains from a group of cpus specified in cpu_map | |
6253 | * These cpus will now be attached to the NULL domain | |
6254 | */ | |
96f874e2 | 6255 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6256 | { |
6257 | int i; | |
6258 | ||
dce840a0 | 6259 | rcu_read_lock(); |
abcd083a | 6260 | for_each_cpu(i, cpu_map) |
57d885fe | 6261 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6262 | rcu_read_unlock(); |
1a20ff27 DG |
6263 | } |
6264 | ||
1d3504fc HS |
6265 | /* handle null as "default" */ |
6266 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6267 | struct sched_domain_attr *new, int idx_new) | |
6268 | { | |
6269 | struct sched_domain_attr tmp; | |
6270 | ||
6271 | /* fast path */ | |
6272 | if (!new && !cur) | |
6273 | return 1; | |
6274 | ||
6275 | tmp = SD_ATTR_INIT; | |
6276 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6277 | new ? (new + idx_new) : &tmp, | |
6278 | sizeof(struct sched_domain_attr)); | |
6279 | } | |
6280 | ||
029190c5 PJ |
6281 | /* |
6282 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6283 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6284 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6285 | * It destroys each deleted domain and builds each new domain. | |
6286 | * | |
acc3f5d7 | 6287 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6288 | * The masks don't intersect (don't overlap.) We should setup one |
6289 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6290 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6291 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6292 | * it as it is. | |
6293 | * | |
acc3f5d7 RR |
6294 | * The passed in 'doms_new' should be allocated using |
6295 | * alloc_sched_domains. This routine takes ownership of it and will | |
6296 | * free_sched_domains it when done with it. If the caller failed the | |
6297 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6298 | * and partition_sched_domains() will fallback to the single partition | |
6299 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6300 | * |
96f874e2 | 6301 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6302 | * ndoms_new == 0 is a special case for destroying existing domains, |
6303 | * and it will not create the default domain. | |
dfb512ec | 6304 | * |
029190c5 PJ |
6305 | * Call with hotplug lock held |
6306 | */ | |
acc3f5d7 | 6307 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6308 | struct sched_domain_attr *dattr_new) |
029190c5 | 6309 | { |
dfb512ec | 6310 | int i, j, n; |
d65bd5ec | 6311 | int new_topology; |
029190c5 | 6312 | |
712555ee | 6313 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6314 | |
7378547f MM |
6315 | /* always unregister in case we don't destroy any domains */ |
6316 | unregister_sched_domain_sysctl(); | |
6317 | ||
d65bd5ec HC |
6318 | /* Let architecture update cpu core mappings. */ |
6319 | new_topology = arch_update_cpu_topology(); | |
6320 | ||
dfb512ec | 6321 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6322 | |
6323 | /* Destroy deleted domains */ | |
6324 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6325 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6326 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6327 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6328 | goto match1; |
6329 | } | |
6330 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6331 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6332 | match1: |
6333 | ; | |
6334 | } | |
6335 | ||
c8d2d47a | 6336 | n = ndoms_cur; |
e761b772 | 6337 | if (doms_new == NULL) { |
c8d2d47a | 6338 | n = 0; |
acc3f5d7 | 6339 | doms_new = &fallback_doms; |
6ad4c188 | 6340 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6341 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6342 | } |
6343 | ||
029190c5 PJ |
6344 | /* Build new domains */ |
6345 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6346 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6347 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6348 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6349 | goto match2; |
6350 | } | |
6351 | /* no match - add a new doms_new */ | |
dce840a0 | 6352 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6353 | match2: |
6354 | ; | |
6355 | } | |
6356 | ||
6357 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6358 | if (doms_cur != &fallback_doms) |
6359 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6360 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6361 | doms_cur = doms_new; |
1d3504fc | 6362 | dattr_cur = dattr_new; |
029190c5 | 6363 | ndoms_cur = ndoms_new; |
7378547f MM |
6364 | |
6365 | register_sched_domain_sysctl(); | |
a1835615 | 6366 | |
712555ee | 6367 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6368 | } |
6369 | ||
d35be8ba SB |
6370 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6371 | ||
1da177e4 | 6372 | /* |
3a101d05 TH |
6373 | * Update cpusets according to cpu_active mask. If cpusets are |
6374 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6375 | * around partition_sched_domains(). | |
d35be8ba SB |
6376 | * |
6377 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6378 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6379 | */ |
0b2e918a TH |
6380 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6381 | void *hcpu) | |
e761b772 | 6382 | { |
d35be8ba SB |
6383 | switch (action) { |
6384 | case CPU_ONLINE_FROZEN: | |
6385 | case CPU_DOWN_FAILED_FROZEN: | |
6386 | ||
6387 | /* | |
6388 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6389 | * resume sequence. As long as this is not the last online | |
6390 | * operation in the resume sequence, just build a single sched | |
6391 | * domain, ignoring cpusets. | |
6392 | */ | |
6393 | num_cpus_frozen--; | |
6394 | if (likely(num_cpus_frozen)) { | |
6395 | partition_sched_domains(1, NULL, NULL); | |
6396 | break; | |
6397 | } | |
6398 | ||
6399 | /* | |
6400 | * This is the last CPU online operation. So fall through and | |
6401 | * restore the original sched domains by considering the | |
6402 | * cpuset configurations. | |
6403 | */ | |
6404 | ||
e761b772 | 6405 | case CPU_ONLINE: |
6ad4c188 | 6406 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6407 | cpuset_update_active_cpus(true); |
d35be8ba | 6408 | break; |
3a101d05 TH |
6409 | default: |
6410 | return NOTIFY_DONE; | |
6411 | } | |
d35be8ba | 6412 | return NOTIFY_OK; |
3a101d05 | 6413 | } |
e761b772 | 6414 | |
0b2e918a TH |
6415 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6416 | void *hcpu) | |
3a101d05 | 6417 | { |
d35be8ba | 6418 | switch (action) { |
3a101d05 | 6419 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6420 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6421 | break; |
6422 | case CPU_DOWN_PREPARE_FROZEN: | |
6423 | num_cpus_frozen++; | |
6424 | partition_sched_domains(1, NULL, NULL); | |
6425 | break; | |
e761b772 MK |
6426 | default: |
6427 | return NOTIFY_DONE; | |
6428 | } | |
d35be8ba | 6429 | return NOTIFY_OK; |
e761b772 | 6430 | } |
e761b772 | 6431 | |
1da177e4 LT |
6432 | void __init sched_init_smp(void) |
6433 | { | |
dcc30a35 RR |
6434 | cpumask_var_t non_isolated_cpus; |
6435 | ||
6436 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6437 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6438 | |
cb83b629 PZ |
6439 | sched_init_numa(); |
6440 | ||
6acce3ef PZ |
6441 | /* |
6442 | * There's no userspace yet to cause hotplug operations; hence all the | |
6443 | * cpu masks are stable and all blatant races in the below code cannot | |
6444 | * happen. | |
6445 | */ | |
712555ee | 6446 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6447 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6448 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6449 | if (cpumask_empty(non_isolated_cpus)) | |
6450 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6451 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6452 | |
301a5cba | 6453 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6454 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6455 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6456 | |
b328ca18 | 6457 | init_hrtick(); |
5c1e1767 NP |
6458 | |
6459 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6460 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6461 | BUG(); |
19978ca6 | 6462 | sched_init_granularity(); |
dcc30a35 | 6463 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6464 | |
0e3900e6 | 6465 | init_sched_rt_class(); |
1baca4ce | 6466 | init_sched_dl_class(); |
1da177e4 LT |
6467 | } |
6468 | #else | |
6469 | void __init sched_init_smp(void) | |
6470 | { | |
19978ca6 | 6471 | sched_init_granularity(); |
1da177e4 LT |
6472 | } |
6473 | #endif /* CONFIG_SMP */ | |
6474 | ||
cd1bb94b AB |
6475 | const_debug unsigned int sysctl_timer_migration = 1; |
6476 | ||
1da177e4 LT |
6477 | int in_sched_functions(unsigned long addr) |
6478 | { | |
1da177e4 LT |
6479 | return in_lock_functions(addr) || |
6480 | (addr >= (unsigned long)__sched_text_start | |
6481 | && addr < (unsigned long)__sched_text_end); | |
6482 | } | |
6483 | ||
029632fb | 6484 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6485 | /* |
6486 | * Default task group. | |
6487 | * Every task in system belongs to this group at bootup. | |
6488 | */ | |
029632fb | 6489 | struct task_group root_task_group; |
35cf4e50 | 6490 | LIST_HEAD(task_groups); |
052f1dc7 | 6491 | #endif |
6f505b16 | 6492 | |
e6252c3e | 6493 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6494 | |
1da177e4 LT |
6495 | void __init sched_init(void) |
6496 | { | |
dd41f596 | 6497 | int i, j; |
434d53b0 MT |
6498 | unsigned long alloc_size = 0, ptr; |
6499 | ||
6500 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6501 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6502 | #endif | |
6503 | #ifdef CONFIG_RT_GROUP_SCHED | |
6504 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6505 | #endif |
df7c8e84 | 6506 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6507 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6508 | #endif |
434d53b0 | 6509 | if (alloc_size) { |
36b7b6d4 | 6510 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6511 | |
6512 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6513 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6514 | ptr += nr_cpu_ids * sizeof(void **); |
6515 | ||
07e06b01 | 6516 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6517 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6518 | |
6d6bc0ad | 6519 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6520 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6521 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6522 | ptr += nr_cpu_ids * sizeof(void **); |
6523 | ||
07e06b01 | 6524 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6525 | ptr += nr_cpu_ids * sizeof(void **); |
6526 | ||
6d6bc0ad | 6527 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6528 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6529 | for_each_possible_cpu(i) { | |
e6252c3e | 6530 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6531 | ptr += cpumask_size(); |
6532 | } | |
6533 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6534 | } |
dd41f596 | 6535 | |
57d885fe GH |
6536 | #ifdef CONFIG_SMP |
6537 | init_defrootdomain(); | |
6538 | #endif | |
6539 | ||
d0b27fa7 PZ |
6540 | init_rt_bandwidth(&def_rt_bandwidth, |
6541 | global_rt_period(), global_rt_runtime()); | |
6542 | ||
6543 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 6544 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6545 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6546 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6547 | |
7c941438 | 6548 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6549 | list_add(&root_task_group.list, &task_groups); |
6550 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6551 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6552 | autogroup_init(&init_task); |
54c707e9 | 6553 | |
7c941438 | 6554 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6555 | |
0a945022 | 6556 | for_each_possible_cpu(i) { |
70b97a7f | 6557 | struct rq *rq; |
1da177e4 LT |
6558 | |
6559 | rq = cpu_rq(i); | |
05fa785c | 6560 | raw_spin_lock_init(&rq->lock); |
7897986b | 6561 | rq->nr_running = 0; |
dce48a84 TG |
6562 | rq->calc_load_active = 0; |
6563 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6564 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6565 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6566 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6567 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6568 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6569 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6570 | /* |
07e06b01 | 6571 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6572 | * |
6573 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6574 | * gets 100% of the cpu resources in the system. This overall | |
6575 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6576 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6577 | * based on each entity's (task or task-group's) weight |
6578 | * (se->load.weight). | |
6579 | * | |
07e06b01 | 6580 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6581 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6582 | * then A0's share of the cpu resource is: | |
6583 | * | |
0d905bca | 6584 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6585 | * |
07e06b01 YZ |
6586 | * We achieve this by letting root_task_group's tasks sit |
6587 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6588 | */ |
ab84d31e | 6589 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6590 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6591 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6592 | ||
6593 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6594 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6595 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6596 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6597 | #endif |
1da177e4 | 6598 | |
dd41f596 IM |
6599 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6600 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6601 | |
6602 | rq->last_load_update_tick = jiffies; | |
6603 | ||
1da177e4 | 6604 | #ifdef CONFIG_SMP |
41c7ce9a | 6605 | rq->sd = NULL; |
57d885fe | 6606 | rq->rd = NULL; |
1399fa78 | 6607 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6608 | rq->post_schedule = 0; |
1da177e4 | 6609 | rq->active_balance = 0; |
dd41f596 | 6610 | rq->next_balance = jiffies; |
1da177e4 | 6611 | rq->push_cpu = 0; |
0a2966b4 | 6612 | rq->cpu = i; |
1f11eb6a | 6613 | rq->online = 0; |
eae0c9df MG |
6614 | rq->idle_stamp = 0; |
6615 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6616 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6617 | |
6618 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6619 | ||
dc938520 | 6620 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6621 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6622 | rq->nohz_flags = 0; |
83cd4fe2 | 6623 | #endif |
265f22a9 FW |
6624 | #ifdef CONFIG_NO_HZ_FULL |
6625 | rq->last_sched_tick = 0; | |
6626 | #endif | |
1da177e4 | 6627 | #endif |
8f4d37ec | 6628 | init_rq_hrtick(rq); |
1da177e4 | 6629 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6630 | } |
6631 | ||
2dd73a4f | 6632 | set_load_weight(&init_task); |
b50f60ce | 6633 | |
e107be36 AK |
6634 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6635 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6636 | #endif | |
6637 | ||
1da177e4 LT |
6638 | /* |
6639 | * The boot idle thread does lazy MMU switching as well: | |
6640 | */ | |
6641 | atomic_inc(&init_mm.mm_count); | |
6642 | enter_lazy_tlb(&init_mm, current); | |
6643 | ||
6644 | /* | |
6645 | * Make us the idle thread. Technically, schedule() should not be | |
6646 | * called from this thread, however somewhere below it might be, | |
6647 | * but because we are the idle thread, we just pick up running again | |
6648 | * when this runqueue becomes "idle". | |
6649 | */ | |
6650 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6651 | |
6652 | calc_load_update = jiffies + LOAD_FREQ; | |
6653 | ||
dd41f596 IM |
6654 | /* |
6655 | * During early bootup we pretend to be a normal task: | |
6656 | */ | |
6657 | current->sched_class = &fair_sched_class; | |
6892b75e | 6658 | |
bf4d83f6 | 6659 | #ifdef CONFIG_SMP |
4cb98839 | 6660 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6661 | /* May be allocated at isolcpus cmdline parse time */ |
6662 | if (cpu_isolated_map == NULL) | |
6663 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6664 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6665 | #endif |
6666 | init_sched_fair_class(); | |
6a7b3dc3 | 6667 | |
6892b75e | 6668 | scheduler_running = 1; |
1da177e4 LT |
6669 | } |
6670 | ||
d902db1e | 6671 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6672 | static inline int preempt_count_equals(int preempt_offset) |
6673 | { | |
234da7bc | 6674 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6675 | |
4ba8216c | 6676 | return (nested == preempt_offset); |
e4aafea2 FW |
6677 | } |
6678 | ||
d894837f | 6679 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6680 | { |
1da177e4 LT |
6681 | static unsigned long prev_jiffy; /* ratelimiting */ |
6682 | ||
b3fbab05 | 6683 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6684 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
6685 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
6686 | return; |
6687 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6688 | return; | |
6689 | prev_jiffy = jiffies; | |
6690 | ||
3df0fc5b PZ |
6691 | printk(KERN_ERR |
6692 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6693 | file, line); | |
6694 | printk(KERN_ERR | |
6695 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6696 | in_atomic(), irqs_disabled(), | |
6697 | current->pid, current->comm); | |
aef745fc IM |
6698 | |
6699 | debug_show_held_locks(current); | |
6700 | if (irqs_disabled()) | |
6701 | print_irqtrace_events(current); | |
6702 | dump_stack(); | |
1da177e4 LT |
6703 | } |
6704 | EXPORT_SYMBOL(__might_sleep); | |
6705 | #endif | |
6706 | ||
6707 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6708 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6709 | { | |
da7a735e | 6710 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
6711 | struct sched_attr attr = { |
6712 | .sched_policy = SCHED_NORMAL, | |
6713 | }; | |
da7a735e | 6714 | int old_prio = p->prio; |
3a5e4dc1 | 6715 | int on_rq; |
3e51f33f | 6716 | |
fd2f4419 | 6717 | on_rq = p->on_rq; |
3a5e4dc1 | 6718 | if (on_rq) |
4ca9b72b | 6719 | dequeue_task(rq, p, 0); |
d50dde5a | 6720 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 6721 | if (on_rq) { |
4ca9b72b | 6722 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
6723 | resched_task(rq->curr); |
6724 | } | |
da7a735e PZ |
6725 | |
6726 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
6727 | } |
6728 | ||
1da177e4 LT |
6729 | void normalize_rt_tasks(void) |
6730 | { | |
a0f98a1c | 6731 | struct task_struct *g, *p; |
1da177e4 | 6732 | unsigned long flags; |
70b97a7f | 6733 | struct rq *rq; |
1da177e4 | 6734 | |
4cf5d77a | 6735 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 6736 | do_each_thread(g, p) { |
178be793 IM |
6737 | /* |
6738 | * Only normalize user tasks: | |
6739 | */ | |
6740 | if (!p->mm) | |
6741 | continue; | |
6742 | ||
6cfb0d5d | 6743 | p->se.exec_start = 0; |
6cfb0d5d | 6744 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
6745 | p->se.statistics.wait_start = 0; |
6746 | p->se.statistics.sleep_start = 0; | |
6747 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 6748 | #endif |
dd41f596 | 6749 | |
aab03e05 | 6750 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6751 | /* |
6752 | * Renice negative nice level userspace | |
6753 | * tasks back to 0: | |
6754 | */ | |
6755 | if (TASK_NICE(p) < 0 && p->mm) | |
6756 | set_user_nice(p, 0); | |
1da177e4 | 6757 | continue; |
dd41f596 | 6758 | } |
1da177e4 | 6759 | |
1d615482 | 6760 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 6761 | rq = __task_rq_lock(p); |
1da177e4 | 6762 | |
178be793 | 6763 | normalize_task(rq, p); |
3a5e4dc1 | 6764 | |
b29739f9 | 6765 | __task_rq_unlock(rq); |
1d615482 | 6766 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
6767 | } while_each_thread(g, p); |
6768 | ||
4cf5d77a | 6769 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
6770 | } |
6771 | ||
6772 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6773 | |
67fc4e0c | 6774 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6775 | /* |
67fc4e0c | 6776 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6777 | * |
6778 | * They can only be called when the whole system has been | |
6779 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6780 | * activity can take place. Using them for anything else would | |
6781 | * be a serious bug, and as a result, they aren't even visible | |
6782 | * under any other configuration. | |
6783 | */ | |
6784 | ||
6785 | /** | |
6786 | * curr_task - return the current task for a given cpu. | |
6787 | * @cpu: the processor in question. | |
6788 | * | |
6789 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6790 | * |
6791 | * Return: The current task for @cpu. | |
1df5c10a | 6792 | */ |
36c8b586 | 6793 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6794 | { |
6795 | return cpu_curr(cpu); | |
6796 | } | |
6797 | ||
67fc4e0c JW |
6798 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6799 | ||
6800 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
6801 | /** |
6802 | * set_curr_task - set the current task for a given cpu. | |
6803 | * @cpu: the processor in question. | |
6804 | * @p: the task pointer to set. | |
6805 | * | |
6806 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
6807 | * are serviced on a separate stack. It allows the architecture to switch the |
6808 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
6809 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6810 | * and caller must save the original value of the current task (see | |
6811 | * curr_task() above) and restore that value before reenabling interrupts and | |
6812 | * re-starting the system. | |
6813 | * | |
6814 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6815 | */ | |
36c8b586 | 6816 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6817 | { |
6818 | cpu_curr(cpu) = p; | |
6819 | } | |
6820 | ||
6821 | #endif | |
29f59db3 | 6822 | |
7c941438 | 6823 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6824 | /* task_group_lock serializes the addition/removal of task groups */ |
6825 | static DEFINE_SPINLOCK(task_group_lock); | |
6826 | ||
bccbe08a PZ |
6827 | static void free_sched_group(struct task_group *tg) |
6828 | { | |
6829 | free_fair_sched_group(tg); | |
6830 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6831 | autogroup_free(tg); |
bccbe08a PZ |
6832 | kfree(tg); |
6833 | } | |
6834 | ||
6835 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6836 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6837 | { |
6838 | struct task_group *tg; | |
bccbe08a PZ |
6839 | |
6840 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
6841 | if (!tg) | |
6842 | return ERR_PTR(-ENOMEM); | |
6843 | ||
ec7dc8ac | 6844 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6845 | goto err; |
6846 | ||
ec7dc8ac | 6847 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6848 | goto err; |
6849 | ||
ace783b9 LZ |
6850 | return tg; |
6851 | ||
6852 | err: | |
6853 | free_sched_group(tg); | |
6854 | return ERR_PTR(-ENOMEM); | |
6855 | } | |
6856 | ||
6857 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6858 | { | |
6859 | unsigned long flags; | |
6860 | ||
8ed36996 | 6861 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6862 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
6863 | |
6864 | WARN_ON(!parent); /* root should already exist */ | |
6865 | ||
6866 | tg->parent = parent; | |
f473aa5e | 6867 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6868 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6869 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6870 | } |
6871 | ||
9b5b7751 | 6872 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 6873 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6874 | { |
29f59db3 | 6875 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 6876 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6877 | } |
6878 | ||
9b5b7751 | 6879 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6880 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
6881 | { |
6882 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6883 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
6884 | } | |
6885 | ||
6886 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6887 | { |
8ed36996 | 6888 | unsigned long flags; |
9b5b7751 | 6889 | int i; |
29f59db3 | 6890 | |
3d4b47b4 PZ |
6891 | /* end participation in shares distribution */ |
6892 | for_each_possible_cpu(i) | |
bccbe08a | 6893 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
6894 | |
6895 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6896 | list_del_rcu(&tg->list); |
f473aa5e | 6897 | list_del_rcu(&tg->siblings); |
8ed36996 | 6898 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6899 | } |
6900 | ||
9b5b7751 | 6901 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6902 | * The caller of this function should have put the task in its new group |
6903 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6904 | * reflect its new group. | |
9b5b7751 SV |
6905 | */ |
6906 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 6907 | { |
8323f26c | 6908 | struct task_group *tg; |
29f59db3 SV |
6909 | int on_rq, running; |
6910 | unsigned long flags; | |
6911 | struct rq *rq; | |
6912 | ||
6913 | rq = task_rq_lock(tsk, &flags); | |
6914 | ||
051a1d1a | 6915 | running = task_current(rq, tsk); |
fd2f4419 | 6916 | on_rq = tsk->on_rq; |
29f59db3 | 6917 | |
0e1f3483 | 6918 | if (on_rq) |
29f59db3 | 6919 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
6920 | if (unlikely(running)) |
6921 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 6922 | |
8af01f56 | 6923 | tg = container_of(task_css_check(tsk, cpu_cgroup_subsys_id, |
8323f26c PZ |
6924 | lockdep_is_held(&tsk->sighand->siglock)), |
6925 | struct task_group, css); | |
6926 | tg = autogroup_task_group(tsk, tg); | |
6927 | tsk->sched_task_group = tg; | |
6928 | ||
810b3817 | 6929 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
6930 | if (tsk->sched_class->task_move_group) |
6931 | tsk->sched_class->task_move_group(tsk, on_rq); | |
6932 | else | |
810b3817 | 6933 | #endif |
b2b5ce02 | 6934 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 6935 | |
0e1f3483 HS |
6936 | if (unlikely(running)) |
6937 | tsk->sched_class->set_curr_task(rq); | |
6938 | if (on_rq) | |
371fd7e7 | 6939 | enqueue_task(rq, tsk, 0); |
29f59db3 | 6940 | |
0122ec5b | 6941 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 6942 | } |
7c941438 | 6943 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 6944 | |
a790de99 | 6945 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
6946 | static unsigned long to_ratio(u64 period, u64 runtime) |
6947 | { | |
6948 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 6949 | return 1ULL << 20; |
9f0c1e56 | 6950 | |
9a7e0b18 | 6951 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 6952 | } |
a790de99 PT |
6953 | #endif |
6954 | ||
6955 | #ifdef CONFIG_RT_GROUP_SCHED | |
6956 | /* | |
6957 | * Ensure that the real time constraints are schedulable. | |
6958 | */ | |
6959 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 6960 | |
9a7e0b18 PZ |
6961 | /* Must be called with tasklist_lock held */ |
6962 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 6963 | { |
9a7e0b18 | 6964 | struct task_struct *g, *p; |
b40b2e8e | 6965 | |
9a7e0b18 | 6966 | do_each_thread(g, p) { |
029632fb | 6967 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
6968 | return 1; |
6969 | } while_each_thread(g, p); | |
b40b2e8e | 6970 | |
9a7e0b18 PZ |
6971 | return 0; |
6972 | } | |
b40b2e8e | 6973 | |
9a7e0b18 PZ |
6974 | struct rt_schedulable_data { |
6975 | struct task_group *tg; | |
6976 | u64 rt_period; | |
6977 | u64 rt_runtime; | |
6978 | }; | |
b40b2e8e | 6979 | |
a790de99 | 6980 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
6981 | { |
6982 | struct rt_schedulable_data *d = data; | |
6983 | struct task_group *child; | |
6984 | unsigned long total, sum = 0; | |
6985 | u64 period, runtime; | |
b40b2e8e | 6986 | |
9a7e0b18 PZ |
6987 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
6988 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 6989 | |
9a7e0b18 PZ |
6990 | if (tg == d->tg) { |
6991 | period = d->rt_period; | |
6992 | runtime = d->rt_runtime; | |
b40b2e8e | 6993 | } |
b40b2e8e | 6994 | |
4653f803 PZ |
6995 | /* |
6996 | * Cannot have more runtime than the period. | |
6997 | */ | |
6998 | if (runtime > period && runtime != RUNTIME_INF) | |
6999 | return -EINVAL; | |
6f505b16 | 7000 | |
4653f803 PZ |
7001 | /* |
7002 | * Ensure we don't starve existing RT tasks. | |
7003 | */ | |
9a7e0b18 PZ |
7004 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7005 | return -EBUSY; | |
6f505b16 | 7006 | |
9a7e0b18 | 7007 | total = to_ratio(period, runtime); |
6f505b16 | 7008 | |
4653f803 PZ |
7009 | /* |
7010 | * Nobody can have more than the global setting allows. | |
7011 | */ | |
7012 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7013 | return -EINVAL; | |
6f505b16 | 7014 | |
4653f803 PZ |
7015 | /* |
7016 | * The sum of our children's runtime should not exceed our own. | |
7017 | */ | |
9a7e0b18 PZ |
7018 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7019 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7020 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7021 | |
9a7e0b18 PZ |
7022 | if (child == d->tg) { |
7023 | period = d->rt_period; | |
7024 | runtime = d->rt_runtime; | |
7025 | } | |
6f505b16 | 7026 | |
9a7e0b18 | 7027 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7028 | } |
6f505b16 | 7029 | |
9a7e0b18 PZ |
7030 | if (sum > total) |
7031 | return -EINVAL; | |
7032 | ||
7033 | return 0; | |
6f505b16 PZ |
7034 | } |
7035 | ||
9a7e0b18 | 7036 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7037 | { |
8277434e PT |
7038 | int ret; |
7039 | ||
9a7e0b18 PZ |
7040 | struct rt_schedulable_data data = { |
7041 | .tg = tg, | |
7042 | .rt_period = period, | |
7043 | .rt_runtime = runtime, | |
7044 | }; | |
7045 | ||
8277434e PT |
7046 | rcu_read_lock(); |
7047 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7048 | rcu_read_unlock(); | |
7049 | ||
7050 | return ret; | |
521f1a24 DG |
7051 | } |
7052 | ||
ab84d31e | 7053 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7054 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7055 | { |
ac086bc2 | 7056 | int i, err = 0; |
9f0c1e56 | 7057 | |
9f0c1e56 | 7058 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7059 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7060 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7061 | if (err) | |
9f0c1e56 | 7062 | goto unlock; |
ac086bc2 | 7063 | |
0986b11b | 7064 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7065 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7066 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7067 | |
7068 | for_each_possible_cpu(i) { | |
7069 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7070 | ||
0986b11b | 7071 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7072 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7073 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7074 | } |
0986b11b | 7075 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7076 | unlock: |
521f1a24 | 7077 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7078 | mutex_unlock(&rt_constraints_mutex); |
7079 | ||
7080 | return err; | |
6f505b16 PZ |
7081 | } |
7082 | ||
25cc7da7 | 7083 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7084 | { |
7085 | u64 rt_runtime, rt_period; | |
7086 | ||
7087 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7088 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7089 | if (rt_runtime_us < 0) | |
7090 | rt_runtime = RUNTIME_INF; | |
7091 | ||
ab84d31e | 7092 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7093 | } |
7094 | ||
25cc7da7 | 7095 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7096 | { |
7097 | u64 rt_runtime_us; | |
7098 | ||
d0b27fa7 | 7099 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7100 | return -1; |
7101 | ||
d0b27fa7 | 7102 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7103 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7104 | return rt_runtime_us; | |
7105 | } | |
d0b27fa7 | 7106 | |
25cc7da7 | 7107 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7108 | { |
7109 | u64 rt_runtime, rt_period; | |
7110 | ||
7111 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7112 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7113 | ||
619b0488 R |
7114 | if (rt_period == 0) |
7115 | return -EINVAL; | |
7116 | ||
ab84d31e | 7117 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7118 | } |
7119 | ||
25cc7da7 | 7120 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7121 | { |
7122 | u64 rt_period_us; | |
7123 | ||
7124 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7125 | do_div(rt_period_us, NSEC_PER_USEC); | |
7126 | return rt_period_us; | |
7127 | } | |
7128 | ||
7129 | static int sched_rt_global_constraints(void) | |
7130 | { | |
4653f803 | 7131 | u64 runtime, period; |
d0b27fa7 PZ |
7132 | int ret = 0; |
7133 | ||
ec5d4989 HS |
7134 | if (sysctl_sched_rt_period <= 0) |
7135 | return -EINVAL; | |
7136 | ||
4653f803 PZ |
7137 | runtime = global_rt_runtime(); |
7138 | period = global_rt_period(); | |
7139 | ||
7140 | /* | |
7141 | * Sanity check on the sysctl variables. | |
7142 | */ | |
7143 | if (runtime > period && runtime != RUNTIME_INF) | |
7144 | return -EINVAL; | |
10b612f4 | 7145 | |
d0b27fa7 | 7146 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7147 | read_lock(&tasklist_lock); |
4653f803 | 7148 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7149 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7150 | mutex_unlock(&rt_constraints_mutex); |
7151 | ||
7152 | return ret; | |
7153 | } | |
54e99124 | 7154 | |
25cc7da7 | 7155 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7156 | { |
7157 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7158 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7159 | return 0; | |
7160 | ||
7161 | return 1; | |
7162 | } | |
7163 | ||
6d6bc0ad | 7164 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7165 | static int sched_rt_global_constraints(void) |
7166 | { | |
ac086bc2 PZ |
7167 | unsigned long flags; |
7168 | int i; | |
7169 | ||
ec5d4989 HS |
7170 | if (sysctl_sched_rt_period <= 0) |
7171 | return -EINVAL; | |
7172 | ||
60aa605d PZ |
7173 | /* |
7174 | * There's always some RT tasks in the root group | |
7175 | * -- migration, kstopmachine etc.. | |
7176 | */ | |
7177 | if (sysctl_sched_rt_runtime == 0) | |
7178 | return -EBUSY; | |
7179 | ||
0986b11b | 7180 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7181 | for_each_possible_cpu(i) { |
7182 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7183 | ||
0986b11b | 7184 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7185 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7186 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7187 | } |
0986b11b | 7188 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7189 | |
d0b27fa7 PZ |
7190 | return 0; |
7191 | } | |
6d6bc0ad | 7192 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7193 | |
ce0dbbbb CW |
7194 | int sched_rr_handler(struct ctl_table *table, int write, |
7195 | void __user *buffer, size_t *lenp, | |
7196 | loff_t *ppos) | |
7197 | { | |
7198 | int ret; | |
7199 | static DEFINE_MUTEX(mutex); | |
7200 | ||
7201 | mutex_lock(&mutex); | |
7202 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7203 | /* make sure that internally we keep jiffies */ | |
7204 | /* also, writing zero resets timeslice to default */ | |
7205 | if (!ret && write) { | |
7206 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7207 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7208 | } | |
7209 | mutex_unlock(&mutex); | |
7210 | return ret; | |
7211 | } | |
7212 | ||
d0b27fa7 | 7213 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7214 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7215 | loff_t *ppos) |
7216 | { | |
7217 | int ret; | |
7218 | int old_period, old_runtime; | |
7219 | static DEFINE_MUTEX(mutex); | |
7220 | ||
7221 | mutex_lock(&mutex); | |
7222 | old_period = sysctl_sched_rt_period; | |
7223 | old_runtime = sysctl_sched_rt_runtime; | |
7224 | ||
8d65af78 | 7225 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7226 | |
7227 | if (!ret && write) { | |
7228 | ret = sched_rt_global_constraints(); | |
7229 | if (ret) { | |
7230 | sysctl_sched_rt_period = old_period; | |
7231 | sysctl_sched_rt_runtime = old_runtime; | |
7232 | } else { | |
7233 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7234 | def_rt_bandwidth.rt_period = | |
7235 | ns_to_ktime(global_rt_period()); | |
7236 | } | |
7237 | } | |
7238 | mutex_unlock(&mutex); | |
7239 | ||
7240 | return ret; | |
7241 | } | |
68318b8e | 7242 | |
052f1dc7 | 7243 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7244 | |
a7c6d554 | 7245 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7246 | { |
a7c6d554 | 7247 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7248 | } |
7249 | ||
eb95419b TH |
7250 | static struct cgroup_subsys_state * |
7251 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7252 | { |
eb95419b TH |
7253 | struct task_group *parent = css_tg(parent_css); |
7254 | struct task_group *tg; | |
68318b8e | 7255 | |
eb95419b | 7256 | if (!parent) { |
68318b8e | 7257 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7258 | return &root_task_group.css; |
68318b8e SV |
7259 | } |
7260 | ||
ec7dc8ac | 7261 | tg = sched_create_group(parent); |
68318b8e SV |
7262 | if (IS_ERR(tg)) |
7263 | return ERR_PTR(-ENOMEM); | |
7264 | ||
68318b8e SV |
7265 | return &tg->css; |
7266 | } | |
7267 | ||
eb95419b | 7268 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7269 | { |
eb95419b TH |
7270 | struct task_group *tg = css_tg(css); |
7271 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7272 | |
63876986 TH |
7273 | if (parent) |
7274 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7275 | return 0; |
7276 | } | |
7277 | ||
eb95419b | 7278 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7279 | { |
eb95419b | 7280 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7281 | |
7282 | sched_destroy_group(tg); | |
7283 | } | |
7284 | ||
eb95419b | 7285 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7286 | { |
eb95419b | 7287 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7288 | |
7289 | sched_offline_group(tg); | |
7290 | } | |
7291 | ||
eb95419b | 7292 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7293 | struct cgroup_taskset *tset) |
68318b8e | 7294 | { |
bb9d97b6 TH |
7295 | struct task_struct *task; |
7296 | ||
d99c8727 | 7297 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7298 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7299 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7300 | return -EINVAL; |
b68aa230 | 7301 | #else |
bb9d97b6 TH |
7302 | /* We don't support RT-tasks being in separate groups */ |
7303 | if (task->sched_class != &fair_sched_class) | |
7304 | return -EINVAL; | |
b68aa230 | 7305 | #endif |
bb9d97b6 | 7306 | } |
be367d09 BB |
7307 | return 0; |
7308 | } | |
68318b8e | 7309 | |
eb95419b | 7310 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7311 | struct cgroup_taskset *tset) |
68318b8e | 7312 | { |
bb9d97b6 TH |
7313 | struct task_struct *task; |
7314 | ||
d99c8727 | 7315 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7316 | sched_move_task(task); |
68318b8e SV |
7317 | } |
7318 | ||
eb95419b TH |
7319 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7320 | struct cgroup_subsys_state *old_css, | |
7321 | struct task_struct *task) | |
068c5cc5 PZ |
7322 | { |
7323 | /* | |
7324 | * cgroup_exit() is called in the copy_process() failure path. | |
7325 | * Ignore this case since the task hasn't ran yet, this avoids | |
7326 | * trying to poke a half freed task state from generic code. | |
7327 | */ | |
7328 | if (!(task->flags & PF_EXITING)) | |
7329 | return; | |
7330 | ||
7331 | sched_move_task(task); | |
7332 | } | |
7333 | ||
052f1dc7 | 7334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7335 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7336 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7337 | { |
182446d0 | 7338 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7339 | } |
7340 | ||
182446d0 TH |
7341 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7342 | struct cftype *cft) | |
68318b8e | 7343 | { |
182446d0 | 7344 | struct task_group *tg = css_tg(css); |
68318b8e | 7345 | |
c8b28116 | 7346 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7347 | } |
ab84d31e PT |
7348 | |
7349 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7350 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7351 | ||
ab84d31e PT |
7352 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7353 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7354 | ||
a790de99 PT |
7355 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7356 | ||
ab84d31e PT |
7357 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7358 | { | |
56f570e5 | 7359 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7360 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7361 | |
7362 | if (tg == &root_task_group) | |
7363 | return -EINVAL; | |
7364 | ||
7365 | /* | |
7366 | * Ensure we have at some amount of bandwidth every period. This is | |
7367 | * to prevent reaching a state of large arrears when throttled via | |
7368 | * entity_tick() resulting in prolonged exit starvation. | |
7369 | */ | |
7370 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7371 | return -EINVAL; | |
7372 | ||
7373 | /* | |
7374 | * Likewise, bound things on the otherside by preventing insane quota | |
7375 | * periods. This also allows us to normalize in computing quota | |
7376 | * feasibility. | |
7377 | */ | |
7378 | if (period > max_cfs_quota_period) | |
7379 | return -EINVAL; | |
7380 | ||
a790de99 PT |
7381 | mutex_lock(&cfs_constraints_mutex); |
7382 | ret = __cfs_schedulable(tg, period, quota); | |
7383 | if (ret) | |
7384 | goto out_unlock; | |
7385 | ||
58088ad0 | 7386 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7387 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7388 | /* |
7389 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7390 | * before making related changes, and on->off must occur afterwards | |
7391 | */ | |
7392 | if (runtime_enabled && !runtime_was_enabled) | |
7393 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7394 | raw_spin_lock_irq(&cfs_b->lock); |
7395 | cfs_b->period = ns_to_ktime(period); | |
7396 | cfs_b->quota = quota; | |
58088ad0 | 7397 | |
a9cf55b2 | 7398 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7399 | /* restart the period timer (if active) to handle new period expiry */ |
7400 | if (runtime_enabled && cfs_b->timer_active) { | |
7401 | /* force a reprogram */ | |
7402 | cfs_b->timer_active = 0; | |
7403 | __start_cfs_bandwidth(cfs_b); | |
7404 | } | |
ab84d31e PT |
7405 | raw_spin_unlock_irq(&cfs_b->lock); |
7406 | ||
7407 | for_each_possible_cpu(i) { | |
7408 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7409 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7410 | |
7411 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7412 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7413 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7414 | |
029632fb | 7415 | if (cfs_rq->throttled) |
671fd9da | 7416 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7417 | raw_spin_unlock_irq(&rq->lock); |
7418 | } | |
1ee14e6c BS |
7419 | if (runtime_was_enabled && !runtime_enabled) |
7420 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7421 | out_unlock: |
7422 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7423 | |
a790de99 | 7424 | return ret; |
ab84d31e PT |
7425 | } |
7426 | ||
7427 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7428 | { | |
7429 | u64 quota, period; | |
7430 | ||
029632fb | 7431 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7432 | if (cfs_quota_us < 0) |
7433 | quota = RUNTIME_INF; | |
7434 | else | |
7435 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7436 | ||
7437 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7438 | } | |
7439 | ||
7440 | long tg_get_cfs_quota(struct task_group *tg) | |
7441 | { | |
7442 | u64 quota_us; | |
7443 | ||
029632fb | 7444 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7445 | return -1; |
7446 | ||
029632fb | 7447 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7448 | do_div(quota_us, NSEC_PER_USEC); |
7449 | ||
7450 | return quota_us; | |
7451 | } | |
7452 | ||
7453 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7454 | { | |
7455 | u64 quota, period; | |
7456 | ||
7457 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7458 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7459 | |
ab84d31e PT |
7460 | return tg_set_cfs_bandwidth(tg, period, quota); |
7461 | } | |
7462 | ||
7463 | long tg_get_cfs_period(struct task_group *tg) | |
7464 | { | |
7465 | u64 cfs_period_us; | |
7466 | ||
029632fb | 7467 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7468 | do_div(cfs_period_us, NSEC_PER_USEC); |
7469 | ||
7470 | return cfs_period_us; | |
7471 | } | |
7472 | ||
182446d0 TH |
7473 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7474 | struct cftype *cft) | |
ab84d31e | 7475 | { |
182446d0 | 7476 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7477 | } |
7478 | ||
182446d0 TH |
7479 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7480 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7481 | { |
182446d0 | 7482 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7483 | } |
7484 | ||
182446d0 TH |
7485 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7486 | struct cftype *cft) | |
ab84d31e | 7487 | { |
182446d0 | 7488 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7489 | } |
7490 | ||
182446d0 TH |
7491 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7492 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7493 | { |
182446d0 | 7494 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7495 | } |
7496 | ||
a790de99 PT |
7497 | struct cfs_schedulable_data { |
7498 | struct task_group *tg; | |
7499 | u64 period, quota; | |
7500 | }; | |
7501 | ||
7502 | /* | |
7503 | * normalize group quota/period to be quota/max_period | |
7504 | * note: units are usecs | |
7505 | */ | |
7506 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7507 | struct cfs_schedulable_data *d) | |
7508 | { | |
7509 | u64 quota, period; | |
7510 | ||
7511 | if (tg == d->tg) { | |
7512 | period = d->period; | |
7513 | quota = d->quota; | |
7514 | } else { | |
7515 | period = tg_get_cfs_period(tg); | |
7516 | quota = tg_get_cfs_quota(tg); | |
7517 | } | |
7518 | ||
7519 | /* note: these should typically be equivalent */ | |
7520 | if (quota == RUNTIME_INF || quota == -1) | |
7521 | return RUNTIME_INF; | |
7522 | ||
7523 | return to_ratio(period, quota); | |
7524 | } | |
7525 | ||
7526 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7527 | { | |
7528 | struct cfs_schedulable_data *d = data; | |
029632fb | 7529 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7530 | s64 quota = 0, parent_quota = -1; |
7531 | ||
7532 | if (!tg->parent) { | |
7533 | quota = RUNTIME_INF; | |
7534 | } else { | |
029632fb | 7535 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7536 | |
7537 | quota = normalize_cfs_quota(tg, d); | |
7538 | parent_quota = parent_b->hierarchal_quota; | |
7539 | ||
7540 | /* | |
7541 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7542 | * limit is set | |
7543 | */ | |
7544 | if (quota == RUNTIME_INF) | |
7545 | quota = parent_quota; | |
7546 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7547 | return -EINVAL; | |
7548 | } | |
7549 | cfs_b->hierarchal_quota = quota; | |
7550 | ||
7551 | return 0; | |
7552 | } | |
7553 | ||
7554 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7555 | { | |
8277434e | 7556 | int ret; |
a790de99 PT |
7557 | struct cfs_schedulable_data data = { |
7558 | .tg = tg, | |
7559 | .period = period, | |
7560 | .quota = quota, | |
7561 | }; | |
7562 | ||
7563 | if (quota != RUNTIME_INF) { | |
7564 | do_div(data.period, NSEC_PER_USEC); | |
7565 | do_div(data.quota, NSEC_PER_USEC); | |
7566 | } | |
7567 | ||
8277434e PT |
7568 | rcu_read_lock(); |
7569 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7570 | rcu_read_unlock(); | |
7571 | ||
7572 | return ret; | |
a790de99 | 7573 | } |
e8da1b18 | 7574 | |
182446d0 | 7575 | static int cpu_stats_show(struct cgroup_subsys_state *css, struct cftype *cft, |
e8da1b18 NR |
7576 | struct cgroup_map_cb *cb) |
7577 | { | |
182446d0 | 7578 | struct task_group *tg = css_tg(css); |
029632fb | 7579 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7580 | |
7581 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7582 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7583 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7584 | ||
7585 | return 0; | |
7586 | } | |
ab84d31e | 7587 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7588 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7589 | |
052f1dc7 | 7590 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7591 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7592 | struct cftype *cft, s64 val) | |
6f505b16 | 7593 | { |
182446d0 | 7594 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7595 | } |
7596 | ||
182446d0 TH |
7597 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7598 | struct cftype *cft) | |
6f505b16 | 7599 | { |
182446d0 | 7600 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7601 | } |
d0b27fa7 | 7602 | |
182446d0 TH |
7603 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7604 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7605 | { |
182446d0 | 7606 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7607 | } |
7608 | ||
182446d0 TH |
7609 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7610 | struct cftype *cft) | |
d0b27fa7 | 7611 | { |
182446d0 | 7612 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7613 | } |
6d6bc0ad | 7614 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7615 | |
fe5c7cc2 | 7616 | static struct cftype cpu_files[] = { |
052f1dc7 | 7617 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7618 | { |
7619 | .name = "shares", | |
f4c753b7 PM |
7620 | .read_u64 = cpu_shares_read_u64, |
7621 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7622 | }, |
052f1dc7 | 7623 | #endif |
ab84d31e PT |
7624 | #ifdef CONFIG_CFS_BANDWIDTH |
7625 | { | |
7626 | .name = "cfs_quota_us", | |
7627 | .read_s64 = cpu_cfs_quota_read_s64, | |
7628 | .write_s64 = cpu_cfs_quota_write_s64, | |
7629 | }, | |
7630 | { | |
7631 | .name = "cfs_period_us", | |
7632 | .read_u64 = cpu_cfs_period_read_u64, | |
7633 | .write_u64 = cpu_cfs_period_write_u64, | |
7634 | }, | |
e8da1b18 NR |
7635 | { |
7636 | .name = "stat", | |
7637 | .read_map = cpu_stats_show, | |
7638 | }, | |
ab84d31e | 7639 | #endif |
052f1dc7 | 7640 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7641 | { |
9f0c1e56 | 7642 | .name = "rt_runtime_us", |
06ecb27c PM |
7643 | .read_s64 = cpu_rt_runtime_read, |
7644 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7645 | }, |
d0b27fa7 PZ |
7646 | { |
7647 | .name = "rt_period_us", | |
f4c753b7 PM |
7648 | .read_u64 = cpu_rt_period_read_uint, |
7649 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7650 | }, |
052f1dc7 | 7651 | #endif |
4baf6e33 | 7652 | { } /* terminate */ |
68318b8e SV |
7653 | }; |
7654 | ||
68318b8e | 7655 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 7656 | .name = "cpu", |
92fb9748 TH |
7657 | .css_alloc = cpu_cgroup_css_alloc, |
7658 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
7659 | .css_online = cpu_cgroup_css_online, |
7660 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
7661 | .can_attach = cpu_cgroup_can_attach, |
7662 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 7663 | .exit = cpu_cgroup_exit, |
38605cae | 7664 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 7665 | .base_cftypes = cpu_files, |
68318b8e SV |
7666 | .early_init = 1, |
7667 | }; | |
7668 | ||
052f1dc7 | 7669 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7670 | |
b637a328 PM |
7671 | void dump_cpu_task(int cpu) |
7672 | { | |
7673 | pr_info("Task dump for CPU %d:\n", cpu); | |
7674 | sched_show_task(cpu_curr(cpu)); | |
7675 | } |