]>
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 | |
332ac17e DF |
299 | /* |
300 | * Maximum bandwidth available for all -deadline tasks and groups | |
301 | * (if group scheduling is configured) on each CPU. | |
302 | * | |
303 | * default: 5% | |
304 | */ | |
305 | unsigned int sysctl_sched_dl_period = 1000000; | |
306 | int sysctl_sched_dl_runtime = 50000; | |
307 | ||
fa85ae24 | 308 | |
1da177e4 | 309 | |
0970d299 | 310 | /* |
0122ec5b | 311 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 312 | */ |
70b97a7f | 313 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
314 | __acquires(rq->lock) |
315 | { | |
0970d299 PZ |
316 | struct rq *rq; |
317 | ||
0122ec5b PZ |
318 | lockdep_assert_held(&p->pi_lock); |
319 | ||
3a5c359a | 320 | for (;;) { |
0970d299 | 321 | rq = task_rq(p); |
05fa785c | 322 | raw_spin_lock(&rq->lock); |
65cc8e48 | 323 | if (likely(rq == task_rq(p))) |
3a5c359a | 324 | return rq; |
05fa785c | 325 | raw_spin_unlock(&rq->lock); |
b29739f9 | 326 | } |
b29739f9 IM |
327 | } |
328 | ||
1da177e4 | 329 | /* |
0122ec5b | 330 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 331 | */ |
70b97a7f | 332 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 333 | __acquires(p->pi_lock) |
1da177e4 LT |
334 | __acquires(rq->lock) |
335 | { | |
70b97a7f | 336 | struct rq *rq; |
1da177e4 | 337 | |
3a5c359a | 338 | for (;;) { |
0122ec5b | 339 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 340 | rq = task_rq(p); |
05fa785c | 341 | raw_spin_lock(&rq->lock); |
65cc8e48 | 342 | if (likely(rq == task_rq(p))) |
3a5c359a | 343 | return rq; |
0122ec5b PZ |
344 | raw_spin_unlock(&rq->lock); |
345 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 346 | } |
1da177e4 LT |
347 | } |
348 | ||
a9957449 | 349 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
350 | __releases(rq->lock) |
351 | { | |
05fa785c | 352 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
353 | } |
354 | ||
0122ec5b PZ |
355 | static inline void |
356 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 357 | __releases(rq->lock) |
0122ec5b | 358 | __releases(p->pi_lock) |
1da177e4 | 359 | { |
0122ec5b PZ |
360 | raw_spin_unlock(&rq->lock); |
361 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
362 | } |
363 | ||
1da177e4 | 364 | /* |
cc2a73b5 | 365 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 366 | */ |
a9957449 | 367 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
368 | __acquires(rq->lock) |
369 | { | |
70b97a7f | 370 | struct rq *rq; |
1da177e4 LT |
371 | |
372 | local_irq_disable(); | |
373 | rq = this_rq(); | |
05fa785c | 374 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
375 | |
376 | return rq; | |
377 | } | |
378 | ||
8f4d37ec PZ |
379 | #ifdef CONFIG_SCHED_HRTICK |
380 | /* | |
381 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 382 | */ |
8f4d37ec | 383 | |
8f4d37ec PZ |
384 | static void hrtick_clear(struct rq *rq) |
385 | { | |
386 | if (hrtimer_active(&rq->hrtick_timer)) | |
387 | hrtimer_cancel(&rq->hrtick_timer); | |
388 | } | |
389 | ||
8f4d37ec PZ |
390 | /* |
391 | * High-resolution timer tick. | |
392 | * Runs from hardirq context with interrupts disabled. | |
393 | */ | |
394 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
395 | { | |
396 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
397 | ||
398 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
399 | ||
05fa785c | 400 | raw_spin_lock(&rq->lock); |
3e51f33f | 401 | update_rq_clock(rq); |
8f4d37ec | 402 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 403 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
404 | |
405 | return HRTIMER_NORESTART; | |
406 | } | |
407 | ||
95e904c7 | 408 | #ifdef CONFIG_SMP |
971ee28c PZ |
409 | |
410 | static int __hrtick_restart(struct rq *rq) | |
411 | { | |
412 | struct hrtimer *timer = &rq->hrtick_timer; | |
413 | ktime_t time = hrtimer_get_softexpires(timer); | |
414 | ||
415 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
416 | } | |
417 | ||
31656519 PZ |
418 | /* |
419 | * called from hardirq (IPI) context | |
420 | */ | |
421 | static void __hrtick_start(void *arg) | |
b328ca18 | 422 | { |
31656519 | 423 | struct rq *rq = arg; |
b328ca18 | 424 | |
05fa785c | 425 | raw_spin_lock(&rq->lock); |
971ee28c | 426 | __hrtick_restart(rq); |
31656519 | 427 | rq->hrtick_csd_pending = 0; |
05fa785c | 428 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
429 | } |
430 | ||
31656519 PZ |
431 | /* |
432 | * Called to set the hrtick timer state. | |
433 | * | |
434 | * called with rq->lock held and irqs disabled | |
435 | */ | |
029632fb | 436 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 437 | { |
31656519 PZ |
438 | struct hrtimer *timer = &rq->hrtick_timer; |
439 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 440 | |
cc584b21 | 441 | hrtimer_set_expires(timer, time); |
31656519 PZ |
442 | |
443 | if (rq == this_rq()) { | |
971ee28c | 444 | __hrtick_restart(rq); |
31656519 | 445 | } else if (!rq->hrtick_csd_pending) { |
6e275637 | 446 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
447 | rq->hrtick_csd_pending = 1; |
448 | } | |
b328ca18 PZ |
449 | } |
450 | ||
451 | static int | |
452 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
453 | { | |
454 | int cpu = (int)(long)hcpu; | |
455 | ||
456 | switch (action) { | |
457 | case CPU_UP_CANCELED: | |
458 | case CPU_UP_CANCELED_FROZEN: | |
459 | case CPU_DOWN_PREPARE: | |
460 | case CPU_DOWN_PREPARE_FROZEN: | |
461 | case CPU_DEAD: | |
462 | case CPU_DEAD_FROZEN: | |
31656519 | 463 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
464 | return NOTIFY_OK; |
465 | } | |
466 | ||
467 | return NOTIFY_DONE; | |
468 | } | |
469 | ||
fa748203 | 470 | static __init void init_hrtick(void) |
b328ca18 PZ |
471 | { |
472 | hotcpu_notifier(hotplug_hrtick, 0); | |
473 | } | |
31656519 PZ |
474 | #else |
475 | /* | |
476 | * Called to set the hrtick timer state. | |
477 | * | |
478 | * called with rq->lock held and irqs disabled | |
479 | */ | |
029632fb | 480 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 481 | { |
7f1e2ca9 | 482 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 483 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 484 | } |
b328ca18 | 485 | |
006c75f1 | 486 | static inline void init_hrtick(void) |
8f4d37ec | 487 | { |
8f4d37ec | 488 | } |
31656519 | 489 | #endif /* CONFIG_SMP */ |
8f4d37ec | 490 | |
31656519 | 491 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 492 | { |
31656519 PZ |
493 | #ifdef CONFIG_SMP |
494 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 495 | |
31656519 PZ |
496 | rq->hrtick_csd.flags = 0; |
497 | rq->hrtick_csd.func = __hrtick_start; | |
498 | rq->hrtick_csd.info = rq; | |
499 | #endif | |
8f4d37ec | 500 | |
31656519 PZ |
501 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
502 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 503 | } |
006c75f1 | 504 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
505 | static inline void hrtick_clear(struct rq *rq) |
506 | { | |
507 | } | |
508 | ||
8f4d37ec PZ |
509 | static inline void init_rq_hrtick(struct rq *rq) |
510 | { | |
511 | } | |
512 | ||
b328ca18 PZ |
513 | static inline void init_hrtick(void) |
514 | { | |
515 | } | |
006c75f1 | 516 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 517 | |
c24d20db IM |
518 | /* |
519 | * resched_task - mark a task 'to be rescheduled now'. | |
520 | * | |
521 | * On UP this means the setting of the need_resched flag, on SMP it | |
522 | * might also involve a cross-CPU call to trigger the scheduler on | |
523 | * the target CPU. | |
524 | */ | |
029632fb | 525 | void resched_task(struct task_struct *p) |
c24d20db IM |
526 | { |
527 | int cpu; | |
528 | ||
b021fe3e | 529 | lockdep_assert_held(&task_rq(p)->lock); |
c24d20db | 530 | |
5ed0cec0 | 531 | if (test_tsk_need_resched(p)) |
c24d20db IM |
532 | return; |
533 | ||
5ed0cec0 | 534 | set_tsk_need_resched(p); |
c24d20db IM |
535 | |
536 | cpu = task_cpu(p); | |
f27dde8d PZ |
537 | if (cpu == smp_processor_id()) { |
538 | set_preempt_need_resched(); | |
c24d20db | 539 | return; |
f27dde8d | 540 | } |
c24d20db IM |
541 | |
542 | /* NEED_RESCHED must be visible before we test polling */ | |
543 | smp_mb(); | |
544 | if (!tsk_is_polling(p)) | |
545 | smp_send_reschedule(cpu); | |
546 | } | |
547 | ||
029632fb | 548 | void resched_cpu(int cpu) |
c24d20db IM |
549 | { |
550 | struct rq *rq = cpu_rq(cpu); | |
551 | unsigned long flags; | |
552 | ||
05fa785c | 553 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
554 | return; |
555 | resched_task(cpu_curr(cpu)); | |
05fa785c | 556 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 557 | } |
06d8308c | 558 | |
b021fe3e | 559 | #ifdef CONFIG_SMP |
3451d024 | 560 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
561 | /* |
562 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
563 | * from an idle cpu. This is good for power-savings. | |
564 | * | |
565 | * We don't do similar optimization for completely idle system, as | |
566 | * selecting an idle cpu will add more delays to the timers than intended | |
567 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
568 | */ | |
569 | int get_nohz_timer_target(void) | |
570 | { | |
571 | int cpu = smp_processor_id(); | |
572 | int i; | |
573 | struct sched_domain *sd; | |
574 | ||
057f3fad | 575 | rcu_read_lock(); |
83cd4fe2 | 576 | for_each_domain(cpu, sd) { |
057f3fad PZ |
577 | for_each_cpu(i, sched_domain_span(sd)) { |
578 | if (!idle_cpu(i)) { | |
579 | cpu = i; | |
580 | goto unlock; | |
581 | } | |
582 | } | |
83cd4fe2 | 583 | } |
057f3fad PZ |
584 | unlock: |
585 | rcu_read_unlock(); | |
83cd4fe2 VP |
586 | return cpu; |
587 | } | |
06d8308c TG |
588 | /* |
589 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
590 | * idle CPU then this timer might expire before the next timer event | |
591 | * which is scheduled to wake up that CPU. In case of a completely | |
592 | * idle system the next event might even be infinite time into the | |
593 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
594 | * leaves the inner idle loop so the newly added timer is taken into | |
595 | * account when the CPU goes back to idle and evaluates the timer | |
596 | * wheel for the next timer event. | |
597 | */ | |
1c20091e | 598 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
599 | { |
600 | struct rq *rq = cpu_rq(cpu); | |
601 | ||
602 | if (cpu == smp_processor_id()) | |
603 | return; | |
604 | ||
605 | /* | |
606 | * This is safe, as this function is called with the timer | |
607 | * wheel base lock of (cpu) held. When the CPU is on the way | |
608 | * to idle and has not yet set rq->curr to idle then it will | |
609 | * be serialized on the timer wheel base lock and take the new | |
610 | * timer into account automatically. | |
611 | */ | |
612 | if (rq->curr != rq->idle) | |
613 | return; | |
45bf76df | 614 | |
45bf76df | 615 | /* |
06d8308c TG |
616 | * We can set TIF_RESCHED on the idle task of the other CPU |
617 | * lockless. The worst case is that the other CPU runs the | |
618 | * idle task through an additional NOOP schedule() | |
45bf76df | 619 | */ |
5ed0cec0 | 620 | set_tsk_need_resched(rq->idle); |
45bf76df | 621 | |
06d8308c TG |
622 | /* NEED_RESCHED must be visible before we test polling */ |
623 | smp_mb(); | |
624 | if (!tsk_is_polling(rq->idle)) | |
625 | smp_send_reschedule(cpu); | |
45bf76df IM |
626 | } |
627 | ||
c5bfece2 | 628 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 629 | { |
c5bfece2 | 630 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
631 | if (cpu != smp_processor_id() || |
632 | tick_nohz_tick_stopped()) | |
633 | smp_send_reschedule(cpu); | |
634 | return true; | |
635 | } | |
636 | ||
637 | return false; | |
638 | } | |
639 | ||
640 | void wake_up_nohz_cpu(int cpu) | |
641 | { | |
c5bfece2 | 642 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
643 | wake_up_idle_cpu(cpu); |
644 | } | |
645 | ||
ca38062e | 646 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 647 | { |
1c792db7 | 648 | int cpu = smp_processor_id(); |
873b4c65 VG |
649 | |
650 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
651 | return false; | |
652 | ||
653 | if (idle_cpu(cpu) && !need_resched()) | |
654 | return true; | |
655 | ||
656 | /* | |
657 | * We can't run Idle Load Balance on this CPU for this time so we | |
658 | * cancel it and clear NOHZ_BALANCE_KICK | |
659 | */ | |
660 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
661 | return false; | |
45bf76df IM |
662 | } |
663 | ||
3451d024 | 664 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 665 | |
ca38062e | 666 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 667 | { |
ca38062e | 668 | return false; |
2069dd75 PZ |
669 | } |
670 | ||
3451d024 | 671 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 672 | |
ce831b38 FW |
673 | #ifdef CONFIG_NO_HZ_FULL |
674 | bool sched_can_stop_tick(void) | |
675 | { | |
676 | struct rq *rq; | |
677 | ||
678 | rq = this_rq(); | |
679 | ||
680 | /* Make sure rq->nr_running update is visible after the IPI */ | |
681 | smp_rmb(); | |
682 | ||
683 | /* More than one running task need preemption */ | |
684 | if (rq->nr_running > 1) | |
685 | return false; | |
686 | ||
687 | return true; | |
688 | } | |
689 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 690 | |
029632fb | 691 | void sched_avg_update(struct rq *rq) |
18d95a28 | 692 | { |
e9e9250b PZ |
693 | s64 period = sched_avg_period(); |
694 | ||
78becc27 | 695 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
696 | /* |
697 | * Inline assembly required to prevent the compiler | |
698 | * optimising this loop into a divmod call. | |
699 | * See __iter_div_u64_rem() for another example of this. | |
700 | */ | |
701 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
702 | rq->age_stamp += period; |
703 | rq->rt_avg /= 2; | |
704 | } | |
18d95a28 PZ |
705 | } |
706 | ||
6d6bc0ad | 707 | #endif /* CONFIG_SMP */ |
18d95a28 | 708 | |
a790de99 PT |
709 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
710 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 711 | /* |
8277434e PT |
712 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
713 | * node and @up when leaving it for the final time. | |
714 | * | |
715 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 716 | */ |
029632fb | 717 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 718 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
719 | { |
720 | struct task_group *parent, *child; | |
eb755805 | 721 | int ret; |
c09595f6 | 722 | |
8277434e PT |
723 | parent = from; |
724 | ||
c09595f6 | 725 | down: |
eb755805 PZ |
726 | ret = (*down)(parent, data); |
727 | if (ret) | |
8277434e | 728 | goto out; |
c09595f6 PZ |
729 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
730 | parent = child; | |
731 | goto down; | |
732 | ||
733 | up: | |
734 | continue; | |
735 | } | |
eb755805 | 736 | ret = (*up)(parent, data); |
8277434e PT |
737 | if (ret || parent == from) |
738 | goto out; | |
c09595f6 PZ |
739 | |
740 | child = parent; | |
741 | parent = parent->parent; | |
742 | if (parent) | |
743 | goto up; | |
8277434e | 744 | out: |
eb755805 | 745 | return ret; |
c09595f6 PZ |
746 | } |
747 | ||
029632fb | 748 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 749 | { |
e2b245f8 | 750 | return 0; |
eb755805 | 751 | } |
18d95a28 PZ |
752 | #endif |
753 | ||
45bf76df IM |
754 | static void set_load_weight(struct task_struct *p) |
755 | { | |
f05998d4 NR |
756 | int prio = p->static_prio - MAX_RT_PRIO; |
757 | struct load_weight *load = &p->se.load; | |
758 | ||
dd41f596 IM |
759 | /* |
760 | * SCHED_IDLE tasks get minimal weight: | |
761 | */ | |
762 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 763 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 764 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
765 | return; |
766 | } | |
71f8bd46 | 767 | |
c8b28116 | 768 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 769 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
770 | } |
771 | ||
371fd7e7 | 772 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 773 | { |
a64692a3 | 774 | update_rq_clock(rq); |
43148951 | 775 | sched_info_queued(rq, p); |
371fd7e7 | 776 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
777 | } |
778 | ||
371fd7e7 | 779 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 780 | { |
a64692a3 | 781 | update_rq_clock(rq); |
43148951 | 782 | sched_info_dequeued(rq, p); |
371fd7e7 | 783 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
784 | } |
785 | ||
029632fb | 786 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
787 | { |
788 | if (task_contributes_to_load(p)) | |
789 | rq->nr_uninterruptible--; | |
790 | ||
371fd7e7 | 791 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
792 | } |
793 | ||
029632fb | 794 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
795 | { |
796 | if (task_contributes_to_load(p)) | |
797 | rq->nr_uninterruptible++; | |
798 | ||
371fd7e7 | 799 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
800 | } |
801 | ||
fe44d621 | 802 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 803 | { |
095c0aa8 GC |
804 | /* |
805 | * In theory, the compile should just see 0 here, and optimize out the call | |
806 | * to sched_rt_avg_update. But I don't trust it... | |
807 | */ | |
808 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
809 | s64 steal = 0, irq_delta = 0; | |
810 | #endif | |
811 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 812 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
813 | |
814 | /* | |
815 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
816 | * this case when a previous update_rq_clock() happened inside a | |
817 | * {soft,}irq region. | |
818 | * | |
819 | * When this happens, we stop ->clock_task and only update the | |
820 | * prev_irq_time stamp to account for the part that fit, so that a next | |
821 | * update will consume the rest. This ensures ->clock_task is | |
822 | * monotonic. | |
823 | * | |
824 | * It does however cause some slight miss-attribution of {soft,}irq | |
825 | * time, a more accurate solution would be to update the irq_time using | |
826 | * the current rq->clock timestamp, except that would require using | |
827 | * atomic ops. | |
828 | */ | |
829 | if (irq_delta > delta) | |
830 | irq_delta = delta; | |
831 | ||
832 | rq->prev_irq_time += irq_delta; | |
833 | delta -= irq_delta; | |
095c0aa8 GC |
834 | #endif |
835 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 836 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
837 | u64 st; |
838 | ||
839 | steal = paravirt_steal_clock(cpu_of(rq)); | |
840 | steal -= rq->prev_steal_time_rq; | |
841 | ||
842 | if (unlikely(steal > delta)) | |
843 | steal = delta; | |
844 | ||
845 | st = steal_ticks(steal); | |
846 | steal = st * TICK_NSEC; | |
847 | ||
848 | rq->prev_steal_time_rq += steal; | |
849 | ||
850 | delta -= steal; | |
851 | } | |
852 | #endif | |
853 | ||
fe44d621 PZ |
854 | rq->clock_task += delta; |
855 | ||
095c0aa8 GC |
856 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
857 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
858 | sched_rt_avg_update(rq, irq_delta + steal); | |
859 | #endif | |
aa483808 VP |
860 | } |
861 | ||
34f971f6 PZ |
862 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
863 | { | |
864 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
865 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
866 | ||
867 | if (stop) { | |
868 | /* | |
869 | * Make it appear like a SCHED_FIFO task, its something | |
870 | * userspace knows about and won't get confused about. | |
871 | * | |
872 | * Also, it will make PI more or less work without too | |
873 | * much confusion -- but then, stop work should not | |
874 | * rely on PI working anyway. | |
875 | */ | |
876 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
877 | ||
878 | stop->sched_class = &stop_sched_class; | |
879 | } | |
880 | ||
881 | cpu_rq(cpu)->stop = stop; | |
882 | ||
883 | if (old_stop) { | |
884 | /* | |
885 | * Reset it back to a normal scheduling class so that | |
886 | * it can die in pieces. | |
887 | */ | |
888 | old_stop->sched_class = &rt_sched_class; | |
889 | } | |
890 | } | |
891 | ||
14531189 | 892 | /* |
dd41f596 | 893 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 894 | */ |
14531189 IM |
895 | static inline int __normal_prio(struct task_struct *p) |
896 | { | |
dd41f596 | 897 | return p->static_prio; |
14531189 IM |
898 | } |
899 | ||
b29739f9 IM |
900 | /* |
901 | * Calculate the expected normal priority: i.e. priority | |
902 | * without taking RT-inheritance into account. Might be | |
903 | * boosted by interactivity modifiers. Changes upon fork, | |
904 | * setprio syscalls, and whenever the interactivity | |
905 | * estimator recalculates. | |
906 | */ | |
36c8b586 | 907 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
908 | { |
909 | int prio; | |
910 | ||
aab03e05 DF |
911 | if (task_has_dl_policy(p)) |
912 | prio = MAX_DL_PRIO-1; | |
913 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
914 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
915 | else | |
916 | prio = __normal_prio(p); | |
917 | return prio; | |
918 | } | |
919 | ||
920 | /* | |
921 | * Calculate the current priority, i.e. the priority | |
922 | * taken into account by the scheduler. This value might | |
923 | * be boosted by RT tasks, or might be boosted by | |
924 | * interactivity modifiers. Will be RT if the task got | |
925 | * RT-boosted. If not then it returns p->normal_prio. | |
926 | */ | |
36c8b586 | 927 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
928 | { |
929 | p->normal_prio = normal_prio(p); | |
930 | /* | |
931 | * If we are RT tasks or we were boosted to RT priority, | |
932 | * keep the priority unchanged. Otherwise, update priority | |
933 | * to the normal priority: | |
934 | */ | |
935 | if (!rt_prio(p->prio)) | |
936 | return p->normal_prio; | |
937 | return p->prio; | |
938 | } | |
939 | ||
1da177e4 LT |
940 | /** |
941 | * task_curr - is this task currently executing on a CPU? | |
942 | * @p: the task in question. | |
e69f6186 YB |
943 | * |
944 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 945 | */ |
36c8b586 | 946 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
947 | { |
948 | return cpu_curr(task_cpu(p)) == p; | |
949 | } | |
950 | ||
cb469845 SR |
951 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
952 | const struct sched_class *prev_class, | |
da7a735e | 953 | int oldprio) |
cb469845 SR |
954 | { |
955 | if (prev_class != p->sched_class) { | |
956 | if (prev_class->switched_from) | |
da7a735e PZ |
957 | prev_class->switched_from(rq, p); |
958 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 959 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 960 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
961 | } |
962 | ||
029632fb | 963 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
964 | { |
965 | const struct sched_class *class; | |
966 | ||
967 | if (p->sched_class == rq->curr->sched_class) { | |
968 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
969 | } else { | |
970 | for_each_class(class) { | |
971 | if (class == rq->curr->sched_class) | |
972 | break; | |
973 | if (class == p->sched_class) { | |
974 | resched_task(rq->curr); | |
975 | break; | |
976 | } | |
977 | } | |
978 | } | |
979 | ||
980 | /* | |
981 | * A queue event has occurred, and we're going to schedule. In | |
982 | * this case, we can save a useless back to back clock update. | |
983 | */ | |
fd2f4419 | 984 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
985 | rq->skip_clock_update = 1; |
986 | } | |
987 | ||
1da177e4 | 988 | #ifdef CONFIG_SMP |
dd41f596 | 989 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 990 | { |
e2912009 PZ |
991 | #ifdef CONFIG_SCHED_DEBUG |
992 | /* | |
993 | * We should never call set_task_cpu() on a blocked task, | |
994 | * ttwu() will sort out the placement. | |
995 | */ | |
077614ee | 996 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 997 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
998 | |
999 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1000 | /* |
1001 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1002 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1003 | * | |
1004 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1005 | * see task_group(). |
6c6c54e1 PZ |
1006 | * |
1007 | * Furthermore, all task_rq users should acquire both locks, see | |
1008 | * task_rq_lock(). | |
1009 | */ | |
0122ec5b PZ |
1010 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1011 | lockdep_is_held(&task_rq(p)->lock))); | |
1012 | #endif | |
e2912009 PZ |
1013 | #endif |
1014 | ||
de1d7286 | 1015 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1016 | |
0c69774e | 1017 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1018 | if (p->sched_class->migrate_task_rq) |
1019 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1020 | p->se.nr_migrations++; |
a8b0ca17 | 1021 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1022 | } |
dd41f596 IM |
1023 | |
1024 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1025 | } |
1026 | ||
ac66f547 PZ |
1027 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1028 | { | |
1029 | if (p->on_rq) { | |
1030 | struct rq *src_rq, *dst_rq; | |
1031 | ||
1032 | src_rq = task_rq(p); | |
1033 | dst_rq = cpu_rq(cpu); | |
1034 | ||
1035 | deactivate_task(src_rq, p, 0); | |
1036 | set_task_cpu(p, cpu); | |
1037 | activate_task(dst_rq, p, 0); | |
1038 | check_preempt_curr(dst_rq, p, 0); | |
1039 | } else { | |
1040 | /* | |
1041 | * Task isn't running anymore; make it appear like we migrated | |
1042 | * it before it went to sleep. This means on wakeup we make the | |
1043 | * previous cpu our targer instead of where it really is. | |
1044 | */ | |
1045 | p->wake_cpu = cpu; | |
1046 | } | |
1047 | } | |
1048 | ||
1049 | struct migration_swap_arg { | |
1050 | struct task_struct *src_task, *dst_task; | |
1051 | int src_cpu, dst_cpu; | |
1052 | }; | |
1053 | ||
1054 | static int migrate_swap_stop(void *data) | |
1055 | { | |
1056 | struct migration_swap_arg *arg = data; | |
1057 | struct rq *src_rq, *dst_rq; | |
1058 | int ret = -EAGAIN; | |
1059 | ||
1060 | src_rq = cpu_rq(arg->src_cpu); | |
1061 | dst_rq = cpu_rq(arg->dst_cpu); | |
1062 | ||
74602315 PZ |
1063 | double_raw_lock(&arg->src_task->pi_lock, |
1064 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1065 | double_rq_lock(src_rq, dst_rq); |
1066 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1067 | goto unlock; | |
1068 | ||
1069 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1070 | goto unlock; | |
1071 | ||
1072 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1073 | goto unlock; | |
1074 | ||
1075 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1076 | goto unlock; | |
1077 | ||
1078 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1079 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1080 | ||
1081 | ret = 0; | |
1082 | ||
1083 | unlock: | |
1084 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1085 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1086 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1087 | |
1088 | return ret; | |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * Cross migrate two tasks | |
1093 | */ | |
1094 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1095 | { | |
1096 | struct migration_swap_arg arg; | |
1097 | int ret = -EINVAL; | |
1098 | ||
ac66f547 PZ |
1099 | arg = (struct migration_swap_arg){ |
1100 | .src_task = cur, | |
1101 | .src_cpu = task_cpu(cur), | |
1102 | .dst_task = p, | |
1103 | .dst_cpu = task_cpu(p), | |
1104 | }; | |
1105 | ||
1106 | if (arg.src_cpu == arg.dst_cpu) | |
1107 | goto out; | |
1108 | ||
6acce3ef PZ |
1109 | /* |
1110 | * These three tests are all lockless; this is OK since all of them | |
1111 | * will be re-checked with proper locks held further down the line. | |
1112 | */ | |
ac66f547 PZ |
1113 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1114 | goto out; | |
1115 | ||
1116 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1117 | goto out; | |
1118 | ||
1119 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1120 | goto out; | |
1121 | ||
1122 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); | |
1123 | ||
1124 | out: | |
ac66f547 PZ |
1125 | return ret; |
1126 | } | |
1127 | ||
969c7921 | 1128 | struct migration_arg { |
36c8b586 | 1129 | struct task_struct *task; |
1da177e4 | 1130 | int dest_cpu; |
70b97a7f | 1131 | }; |
1da177e4 | 1132 | |
969c7921 TH |
1133 | static int migration_cpu_stop(void *data); |
1134 | ||
1da177e4 LT |
1135 | /* |
1136 | * wait_task_inactive - wait for a thread to unschedule. | |
1137 | * | |
85ba2d86 RM |
1138 | * If @match_state is nonzero, it's the @p->state value just checked and |
1139 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1140 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1141 | * we return a positive number (its total switch count). If a second call | |
1142 | * a short while later returns the same number, the caller can be sure that | |
1143 | * @p has remained unscheduled the whole time. | |
1144 | * | |
1da177e4 LT |
1145 | * The caller must ensure that the task *will* unschedule sometime soon, |
1146 | * else this function might spin for a *long* time. This function can't | |
1147 | * be called with interrupts off, or it may introduce deadlock with | |
1148 | * smp_call_function() if an IPI is sent by the same process we are | |
1149 | * waiting to become inactive. | |
1150 | */ | |
85ba2d86 | 1151 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1152 | { |
1153 | unsigned long flags; | |
dd41f596 | 1154 | int running, on_rq; |
85ba2d86 | 1155 | unsigned long ncsw; |
70b97a7f | 1156 | struct rq *rq; |
1da177e4 | 1157 | |
3a5c359a AK |
1158 | for (;;) { |
1159 | /* | |
1160 | * We do the initial early heuristics without holding | |
1161 | * any task-queue locks at all. We'll only try to get | |
1162 | * the runqueue lock when things look like they will | |
1163 | * work out! | |
1164 | */ | |
1165 | rq = task_rq(p); | |
fa490cfd | 1166 | |
3a5c359a AK |
1167 | /* |
1168 | * If the task is actively running on another CPU | |
1169 | * still, just relax and busy-wait without holding | |
1170 | * any locks. | |
1171 | * | |
1172 | * NOTE! Since we don't hold any locks, it's not | |
1173 | * even sure that "rq" stays as the right runqueue! | |
1174 | * But we don't care, since "task_running()" will | |
1175 | * return false if the runqueue has changed and p | |
1176 | * is actually now running somewhere else! | |
1177 | */ | |
85ba2d86 RM |
1178 | while (task_running(rq, p)) { |
1179 | if (match_state && unlikely(p->state != match_state)) | |
1180 | return 0; | |
3a5c359a | 1181 | cpu_relax(); |
85ba2d86 | 1182 | } |
fa490cfd | 1183 | |
3a5c359a AK |
1184 | /* |
1185 | * Ok, time to look more closely! We need the rq | |
1186 | * lock now, to be *sure*. If we're wrong, we'll | |
1187 | * just go back and repeat. | |
1188 | */ | |
1189 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1190 | trace_sched_wait_task(p); |
3a5c359a | 1191 | running = task_running(rq, p); |
fd2f4419 | 1192 | on_rq = p->on_rq; |
85ba2d86 | 1193 | ncsw = 0; |
f31e11d8 | 1194 | if (!match_state || p->state == match_state) |
93dcf55f | 1195 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1196 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1197 | |
85ba2d86 RM |
1198 | /* |
1199 | * If it changed from the expected state, bail out now. | |
1200 | */ | |
1201 | if (unlikely(!ncsw)) | |
1202 | break; | |
1203 | ||
3a5c359a AK |
1204 | /* |
1205 | * Was it really running after all now that we | |
1206 | * checked with the proper locks actually held? | |
1207 | * | |
1208 | * Oops. Go back and try again.. | |
1209 | */ | |
1210 | if (unlikely(running)) { | |
1211 | cpu_relax(); | |
1212 | continue; | |
1213 | } | |
fa490cfd | 1214 | |
3a5c359a AK |
1215 | /* |
1216 | * It's not enough that it's not actively running, | |
1217 | * it must be off the runqueue _entirely_, and not | |
1218 | * preempted! | |
1219 | * | |
80dd99b3 | 1220 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1221 | * running right now), it's preempted, and we should |
1222 | * yield - it could be a while. | |
1223 | */ | |
1224 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1225 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1226 | ||
1227 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1228 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1229 | continue; |
1230 | } | |
fa490cfd | 1231 | |
3a5c359a AK |
1232 | /* |
1233 | * Ahh, all good. It wasn't running, and it wasn't | |
1234 | * runnable, which means that it will never become | |
1235 | * running in the future either. We're all done! | |
1236 | */ | |
1237 | break; | |
1238 | } | |
85ba2d86 RM |
1239 | |
1240 | return ncsw; | |
1da177e4 LT |
1241 | } |
1242 | ||
1243 | /*** | |
1244 | * kick_process - kick a running thread to enter/exit the kernel | |
1245 | * @p: the to-be-kicked thread | |
1246 | * | |
1247 | * Cause a process which is running on another CPU to enter | |
1248 | * kernel-mode, without any delay. (to get signals handled.) | |
1249 | * | |
25985edc | 1250 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1251 | * because all it wants to ensure is that the remote task enters |
1252 | * the kernel. If the IPI races and the task has been migrated | |
1253 | * to another CPU then no harm is done and the purpose has been | |
1254 | * achieved as well. | |
1255 | */ | |
36c8b586 | 1256 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1257 | { |
1258 | int cpu; | |
1259 | ||
1260 | preempt_disable(); | |
1261 | cpu = task_cpu(p); | |
1262 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1263 | smp_send_reschedule(cpu); | |
1264 | preempt_enable(); | |
1265 | } | |
b43e3521 | 1266 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1267 | #endif /* CONFIG_SMP */ |
1da177e4 | 1268 | |
970b13ba | 1269 | #ifdef CONFIG_SMP |
30da688e | 1270 | /* |
013fdb80 | 1271 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1272 | */ |
5da9a0fb PZ |
1273 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1274 | { | |
aa00d89c TC |
1275 | int nid = cpu_to_node(cpu); |
1276 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1277 | enum { cpuset, possible, fail } state = cpuset; |
1278 | int dest_cpu; | |
5da9a0fb | 1279 | |
aa00d89c TC |
1280 | /* |
1281 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1282 | * will return -1. There is no cpu on the node, and we should | |
1283 | * select the cpu on the other node. | |
1284 | */ | |
1285 | if (nid != -1) { | |
1286 | nodemask = cpumask_of_node(nid); | |
1287 | ||
1288 | /* Look for allowed, online CPU in same node. */ | |
1289 | for_each_cpu(dest_cpu, nodemask) { | |
1290 | if (!cpu_online(dest_cpu)) | |
1291 | continue; | |
1292 | if (!cpu_active(dest_cpu)) | |
1293 | continue; | |
1294 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1295 | return dest_cpu; | |
1296 | } | |
2baab4e9 | 1297 | } |
5da9a0fb | 1298 | |
2baab4e9 PZ |
1299 | for (;;) { |
1300 | /* Any allowed, online CPU? */ | |
e3831edd | 1301 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1302 | if (!cpu_online(dest_cpu)) |
1303 | continue; | |
1304 | if (!cpu_active(dest_cpu)) | |
1305 | continue; | |
1306 | goto out; | |
1307 | } | |
5da9a0fb | 1308 | |
2baab4e9 PZ |
1309 | switch (state) { |
1310 | case cpuset: | |
1311 | /* No more Mr. Nice Guy. */ | |
1312 | cpuset_cpus_allowed_fallback(p); | |
1313 | state = possible; | |
1314 | break; | |
1315 | ||
1316 | case possible: | |
1317 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1318 | state = fail; | |
1319 | break; | |
1320 | ||
1321 | case fail: | |
1322 | BUG(); | |
1323 | break; | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | out: | |
1328 | if (state != cpuset) { | |
1329 | /* | |
1330 | * Don't tell them about moving exiting tasks or | |
1331 | * kernel threads (both mm NULL), since they never | |
1332 | * leave kernel. | |
1333 | */ | |
1334 | if (p->mm && printk_ratelimit()) { | |
1335 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1336 | task_pid_nr(p), p->comm, cpu); | |
1337 | } | |
5da9a0fb PZ |
1338 | } |
1339 | ||
1340 | return dest_cpu; | |
1341 | } | |
1342 | ||
e2912009 | 1343 | /* |
013fdb80 | 1344 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1345 | */ |
970b13ba | 1346 | static inline |
ac66f547 | 1347 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1348 | { |
ac66f547 | 1349 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1350 | |
1351 | /* | |
1352 | * In order not to call set_task_cpu() on a blocking task we need | |
1353 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1354 | * cpu. | |
1355 | * | |
1356 | * Since this is common to all placement strategies, this lives here. | |
1357 | * | |
1358 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1359 | * not worry about this generic constraint ] | |
1360 | */ | |
fa17b507 | 1361 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1362 | !cpu_online(cpu))) |
5da9a0fb | 1363 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1364 | |
1365 | return cpu; | |
970b13ba | 1366 | } |
09a40af5 MG |
1367 | |
1368 | static void update_avg(u64 *avg, u64 sample) | |
1369 | { | |
1370 | s64 diff = sample - *avg; | |
1371 | *avg += diff >> 3; | |
1372 | } | |
970b13ba PZ |
1373 | #endif |
1374 | ||
d7c01d27 | 1375 | static void |
b84cb5df | 1376 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1377 | { |
d7c01d27 | 1378 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1379 | struct rq *rq = this_rq(); |
1380 | ||
d7c01d27 PZ |
1381 | #ifdef CONFIG_SMP |
1382 | int this_cpu = smp_processor_id(); | |
1383 | ||
1384 | if (cpu == this_cpu) { | |
1385 | schedstat_inc(rq, ttwu_local); | |
1386 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1387 | } else { | |
1388 | struct sched_domain *sd; | |
1389 | ||
1390 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1391 | rcu_read_lock(); |
d7c01d27 PZ |
1392 | for_each_domain(this_cpu, sd) { |
1393 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1394 | schedstat_inc(sd, ttwu_wake_remote); | |
1395 | break; | |
1396 | } | |
1397 | } | |
057f3fad | 1398 | rcu_read_unlock(); |
d7c01d27 | 1399 | } |
f339b9dc PZ |
1400 | |
1401 | if (wake_flags & WF_MIGRATED) | |
1402 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1403 | ||
d7c01d27 PZ |
1404 | #endif /* CONFIG_SMP */ |
1405 | ||
1406 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1407 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1408 | |
1409 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1410 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1411 | |
d7c01d27 PZ |
1412 | #endif /* CONFIG_SCHEDSTATS */ |
1413 | } | |
1414 | ||
1415 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1416 | { | |
9ed3811a | 1417 | activate_task(rq, p, en_flags); |
fd2f4419 | 1418 | p->on_rq = 1; |
c2f7115e PZ |
1419 | |
1420 | /* if a worker is waking up, notify workqueue */ | |
1421 | if (p->flags & PF_WQ_WORKER) | |
1422 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1423 | } |
1424 | ||
23f41eeb PZ |
1425 | /* |
1426 | * Mark the task runnable and perform wakeup-preemption. | |
1427 | */ | |
89363381 | 1428 | static void |
23f41eeb | 1429 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1430 | { |
9ed3811a | 1431 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1432 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1433 | |
1434 | p->state = TASK_RUNNING; | |
1435 | #ifdef CONFIG_SMP | |
1436 | if (p->sched_class->task_woken) | |
1437 | p->sched_class->task_woken(rq, p); | |
1438 | ||
e69c6341 | 1439 | if (rq->idle_stamp) { |
78becc27 | 1440 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1441 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1442 | |
abfafa54 JL |
1443 | update_avg(&rq->avg_idle, delta); |
1444 | ||
1445 | if (rq->avg_idle > max) | |
9ed3811a | 1446 | rq->avg_idle = max; |
abfafa54 | 1447 | |
9ed3811a TH |
1448 | rq->idle_stamp = 0; |
1449 | } | |
1450 | #endif | |
1451 | } | |
1452 | ||
c05fbafb PZ |
1453 | static void |
1454 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1455 | { | |
1456 | #ifdef CONFIG_SMP | |
1457 | if (p->sched_contributes_to_load) | |
1458 | rq->nr_uninterruptible--; | |
1459 | #endif | |
1460 | ||
1461 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1462 | ttwu_do_wakeup(rq, p, wake_flags); | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1467 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1468 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1469 | * the task is still ->on_rq. | |
1470 | */ | |
1471 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1472 | { | |
1473 | struct rq *rq; | |
1474 | int ret = 0; | |
1475 | ||
1476 | rq = __task_rq_lock(p); | |
1477 | if (p->on_rq) { | |
1ad4ec0d FW |
1478 | /* check_preempt_curr() may use rq clock */ |
1479 | update_rq_clock(rq); | |
c05fbafb PZ |
1480 | ttwu_do_wakeup(rq, p, wake_flags); |
1481 | ret = 1; | |
1482 | } | |
1483 | __task_rq_unlock(rq); | |
1484 | ||
1485 | return ret; | |
1486 | } | |
1487 | ||
317f3941 | 1488 | #ifdef CONFIG_SMP |
fa14ff4a | 1489 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1490 | { |
1491 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1492 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1493 | struct task_struct *p; | |
317f3941 PZ |
1494 | |
1495 | raw_spin_lock(&rq->lock); | |
1496 | ||
fa14ff4a PZ |
1497 | while (llist) { |
1498 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1499 | llist = llist_next(llist); | |
317f3941 PZ |
1500 | ttwu_do_activate(rq, p, 0); |
1501 | } | |
1502 | ||
1503 | raw_spin_unlock(&rq->lock); | |
1504 | } | |
1505 | ||
1506 | void scheduler_ipi(void) | |
1507 | { | |
f27dde8d PZ |
1508 | /* |
1509 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1510 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1511 | * this IPI. | |
1512 | */ | |
1513 | if (tif_need_resched()) | |
1514 | set_preempt_need_resched(); | |
1515 | ||
873b4c65 VG |
1516 | if (llist_empty(&this_rq()->wake_list) |
1517 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1518 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1519 | return; |
1520 | ||
1521 | /* | |
1522 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1523 | * traditionally all their work was done from the interrupt return | |
1524 | * path. Now that we actually do some work, we need to make sure | |
1525 | * we do call them. | |
1526 | * | |
1527 | * Some archs already do call them, luckily irq_enter/exit nest | |
1528 | * properly. | |
1529 | * | |
1530 | * Arguably we should visit all archs and update all handlers, | |
1531 | * however a fair share of IPIs are still resched only so this would | |
1532 | * somewhat pessimize the simple resched case. | |
1533 | */ | |
1534 | irq_enter(); | |
ff442c51 | 1535 | tick_nohz_full_check(); |
fa14ff4a | 1536 | sched_ttwu_pending(); |
ca38062e SS |
1537 | |
1538 | /* | |
1539 | * Check if someone kicked us for doing the nohz idle load balance. | |
1540 | */ | |
873b4c65 | 1541 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1542 | this_rq()->idle_balance = 1; |
ca38062e | 1543 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1544 | } |
c5d753a5 | 1545 | irq_exit(); |
317f3941 PZ |
1546 | } |
1547 | ||
1548 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1549 | { | |
fa14ff4a | 1550 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1551 | smp_send_reschedule(cpu); |
1552 | } | |
d6aa8f85 | 1553 | |
39be3501 | 1554 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1555 | { |
1556 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1557 | } | |
d6aa8f85 | 1558 | #endif /* CONFIG_SMP */ |
317f3941 | 1559 | |
c05fbafb PZ |
1560 | static void ttwu_queue(struct task_struct *p, int cpu) |
1561 | { | |
1562 | struct rq *rq = cpu_rq(cpu); | |
1563 | ||
17d9f311 | 1564 | #if defined(CONFIG_SMP) |
39be3501 | 1565 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1566 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1567 | ttwu_queue_remote(p, cpu); |
1568 | return; | |
1569 | } | |
1570 | #endif | |
1571 | ||
c05fbafb PZ |
1572 | raw_spin_lock(&rq->lock); |
1573 | ttwu_do_activate(rq, p, 0); | |
1574 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1575 | } |
1576 | ||
1577 | /** | |
1da177e4 | 1578 | * try_to_wake_up - wake up a thread |
9ed3811a | 1579 | * @p: the thread to be awakened |
1da177e4 | 1580 | * @state: the mask of task states that can be woken |
9ed3811a | 1581 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1582 | * |
1583 | * Put it on the run-queue if it's not already there. The "current" | |
1584 | * thread is always on the run-queue (except when the actual | |
1585 | * re-schedule is in progress), and as such you're allowed to do | |
1586 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1587 | * runnable without the overhead of this. | |
1588 | * | |
e69f6186 | 1589 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1590 | * or @state didn't match @p's state. |
1da177e4 | 1591 | */ |
e4a52bcb PZ |
1592 | static int |
1593 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1594 | { |
1da177e4 | 1595 | unsigned long flags; |
c05fbafb | 1596 | int cpu, success = 0; |
2398f2c6 | 1597 | |
e0acd0a6 ON |
1598 | /* |
1599 | * If we are going to wake up a thread waiting for CONDITION we | |
1600 | * need to ensure that CONDITION=1 done by the caller can not be | |
1601 | * reordered with p->state check below. This pairs with mb() in | |
1602 | * set_current_state() the waiting thread does. | |
1603 | */ | |
1604 | smp_mb__before_spinlock(); | |
013fdb80 | 1605 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1606 | if (!(p->state & state)) |
1da177e4 LT |
1607 | goto out; |
1608 | ||
c05fbafb | 1609 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1610 | cpu = task_cpu(p); |
1da177e4 | 1611 | |
c05fbafb PZ |
1612 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1613 | goto stat; | |
1da177e4 | 1614 | |
1da177e4 | 1615 | #ifdef CONFIG_SMP |
e9c84311 | 1616 | /* |
c05fbafb PZ |
1617 | * If the owning (remote) cpu is still in the middle of schedule() with |
1618 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1619 | */ |
f3e94786 | 1620 | while (p->on_cpu) |
e4a52bcb | 1621 | cpu_relax(); |
0970d299 | 1622 | /* |
e4a52bcb | 1623 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1624 | */ |
e4a52bcb | 1625 | smp_rmb(); |
1da177e4 | 1626 | |
a8e4f2ea | 1627 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1628 | p->state = TASK_WAKING; |
e7693a36 | 1629 | |
e4a52bcb | 1630 | if (p->sched_class->task_waking) |
74f8e4b2 | 1631 | p->sched_class->task_waking(p); |
efbbd05a | 1632 | |
ac66f547 | 1633 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1634 | if (task_cpu(p) != cpu) { |
1635 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1636 | set_task_cpu(p, cpu); |
f339b9dc | 1637 | } |
1da177e4 | 1638 | #endif /* CONFIG_SMP */ |
1da177e4 | 1639 | |
c05fbafb PZ |
1640 | ttwu_queue(p, cpu); |
1641 | stat: | |
b84cb5df | 1642 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1643 | out: |
013fdb80 | 1644 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1645 | |
1646 | return success; | |
1647 | } | |
1648 | ||
21aa9af0 TH |
1649 | /** |
1650 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1651 | * @p: the thread to be awakened | |
1652 | * | |
2acca55e | 1653 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1654 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1655 | * the current task. |
21aa9af0 TH |
1656 | */ |
1657 | static void try_to_wake_up_local(struct task_struct *p) | |
1658 | { | |
1659 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1660 | |
383efcd0 TH |
1661 | if (WARN_ON_ONCE(rq != this_rq()) || |
1662 | WARN_ON_ONCE(p == current)) | |
1663 | return; | |
1664 | ||
21aa9af0 TH |
1665 | lockdep_assert_held(&rq->lock); |
1666 | ||
2acca55e PZ |
1667 | if (!raw_spin_trylock(&p->pi_lock)) { |
1668 | raw_spin_unlock(&rq->lock); | |
1669 | raw_spin_lock(&p->pi_lock); | |
1670 | raw_spin_lock(&rq->lock); | |
1671 | } | |
1672 | ||
21aa9af0 | 1673 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1674 | goto out; |
21aa9af0 | 1675 | |
fd2f4419 | 1676 | if (!p->on_rq) |
d7c01d27 PZ |
1677 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1678 | ||
23f41eeb | 1679 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1680 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1681 | out: |
1682 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1683 | } |
1684 | ||
50fa610a DH |
1685 | /** |
1686 | * wake_up_process - Wake up a specific process | |
1687 | * @p: The process to be woken up. | |
1688 | * | |
1689 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1690 | * processes. |
1691 | * | |
1692 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1693 | * |
1694 | * It may be assumed that this function implies a write memory barrier before | |
1695 | * changing the task state if and only if any tasks are woken up. | |
1696 | */ | |
7ad5b3a5 | 1697 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1698 | { |
9067ac85 ON |
1699 | WARN_ON(task_is_stopped_or_traced(p)); |
1700 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1701 | } |
1da177e4 LT |
1702 | EXPORT_SYMBOL(wake_up_process); |
1703 | ||
7ad5b3a5 | 1704 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1705 | { |
1706 | return try_to_wake_up(p, state, 0); | |
1707 | } | |
1708 | ||
1da177e4 LT |
1709 | /* |
1710 | * Perform scheduler related setup for a newly forked process p. | |
1711 | * p is forked by current. | |
dd41f596 IM |
1712 | * |
1713 | * __sched_fork() is basic setup used by init_idle() too: | |
1714 | */ | |
5e1576ed | 1715 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1716 | { |
fd2f4419 PZ |
1717 | p->on_rq = 0; |
1718 | ||
1719 | p->se.on_rq = 0; | |
dd41f596 IM |
1720 | p->se.exec_start = 0; |
1721 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1722 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1723 | p->se.nr_migrations = 0; |
da7a735e | 1724 | p->se.vruntime = 0; |
fd2f4419 | 1725 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1726 | |
1727 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1728 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1729 | #endif |
476d139c | 1730 | |
aab03e05 DF |
1731 | RB_CLEAR_NODE(&p->dl.rb_node); |
1732 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1733 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1734 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1735 | p->dl.dl_period = 0; |
aab03e05 DF |
1736 | p->dl.flags = 0; |
1737 | ||
fa717060 | 1738 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1739 | |
e107be36 AK |
1740 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1741 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1742 | #endif | |
cbee9f88 PZ |
1743 | |
1744 | #ifdef CONFIG_NUMA_BALANCING | |
1745 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1746 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1747 | p->mm->numa_scan_seq = 0; |
1748 | } | |
1749 | ||
5e1576ed RR |
1750 | if (clone_flags & CLONE_VM) |
1751 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1752 | else | |
1753 | p->numa_preferred_nid = -1; | |
1754 | ||
cbee9f88 PZ |
1755 | p->node_stamp = 0ULL; |
1756 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1757 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1758 | p->numa_work.next = &p->numa_work; |
f809ca9a | 1759 | p->numa_faults = NULL; |
745d6147 | 1760 | p->numa_faults_buffer = NULL; |
8c8a743c PZ |
1761 | |
1762 | INIT_LIST_HEAD(&p->numa_entry); | |
1763 | p->numa_group = NULL; | |
cbee9f88 | 1764 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1765 | } |
1766 | ||
1a687c2e | 1767 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1768 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1769 | void set_numabalancing_state(bool enabled) |
1770 | { | |
1771 | if (enabled) | |
1772 | sched_feat_set("NUMA"); | |
1773 | else | |
1774 | sched_feat_set("NO_NUMA"); | |
1775 | } | |
3105b86a MG |
1776 | #else |
1777 | __read_mostly bool numabalancing_enabled; | |
1778 | ||
1779 | void set_numabalancing_state(bool enabled) | |
1780 | { | |
1781 | numabalancing_enabled = enabled; | |
dd41f596 | 1782 | } |
3105b86a | 1783 | #endif /* CONFIG_SCHED_DEBUG */ |
1a687c2e | 1784 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1785 | |
1786 | /* | |
1787 | * fork()/clone()-time setup: | |
1788 | */ | |
aab03e05 | 1789 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1790 | { |
0122ec5b | 1791 | unsigned long flags; |
dd41f596 IM |
1792 | int cpu = get_cpu(); |
1793 | ||
5e1576ed | 1794 | __sched_fork(clone_flags, p); |
06b83b5f | 1795 | /* |
0017d735 | 1796 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1797 | * nobody will actually run it, and a signal or other external |
1798 | * event cannot wake it up and insert it on the runqueue either. | |
1799 | */ | |
0017d735 | 1800 | p->state = TASK_RUNNING; |
dd41f596 | 1801 | |
c350a04e MG |
1802 | /* |
1803 | * Make sure we do not leak PI boosting priority to the child. | |
1804 | */ | |
1805 | p->prio = current->normal_prio; | |
1806 | ||
b9dc29e7 MG |
1807 | /* |
1808 | * Revert to default priority/policy on fork if requested. | |
1809 | */ | |
1810 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1811 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1812 | p->policy = SCHED_NORMAL; |
6c697bdf | 1813 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1814 | p->rt_priority = 0; |
1815 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1816 | p->static_prio = NICE_TO_PRIO(0); | |
1817 | ||
1818 | p->prio = p->normal_prio = __normal_prio(p); | |
1819 | set_load_weight(p); | |
6c697bdf | 1820 | |
b9dc29e7 MG |
1821 | /* |
1822 | * We don't need the reset flag anymore after the fork. It has | |
1823 | * fulfilled its duty: | |
1824 | */ | |
1825 | p->sched_reset_on_fork = 0; | |
1826 | } | |
ca94c442 | 1827 | |
aab03e05 DF |
1828 | if (dl_prio(p->prio)) { |
1829 | put_cpu(); | |
1830 | return -EAGAIN; | |
1831 | } else if (rt_prio(p->prio)) { | |
1832 | p->sched_class = &rt_sched_class; | |
1833 | } else { | |
2ddbf952 | 1834 | p->sched_class = &fair_sched_class; |
aab03e05 | 1835 | } |
b29739f9 | 1836 | |
cd29fe6f PZ |
1837 | if (p->sched_class->task_fork) |
1838 | p->sched_class->task_fork(p); | |
1839 | ||
86951599 PZ |
1840 | /* |
1841 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1842 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1843 | * is ran before sched_fork(). | |
1844 | * | |
1845 | * Silence PROVE_RCU. | |
1846 | */ | |
0122ec5b | 1847 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1848 | set_task_cpu(p, cpu); |
0122ec5b | 1849 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1850 | |
52f17b6c | 1851 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1852 | if (likely(sched_info_on())) |
52f17b6c | 1853 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1854 | #endif |
3ca7a440 PZ |
1855 | #if defined(CONFIG_SMP) |
1856 | p->on_cpu = 0; | |
4866cde0 | 1857 | #endif |
01028747 | 1858 | init_task_preempt_count(p); |
806c09a7 | 1859 | #ifdef CONFIG_SMP |
917b627d | 1860 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1861 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1862 | #endif |
917b627d | 1863 | |
476d139c | 1864 | put_cpu(); |
aab03e05 | 1865 | return 0; |
1da177e4 LT |
1866 | } |
1867 | ||
332ac17e DF |
1868 | unsigned long to_ratio(u64 period, u64 runtime) |
1869 | { | |
1870 | if (runtime == RUNTIME_INF) | |
1871 | return 1ULL << 20; | |
1872 | ||
1873 | /* | |
1874 | * Doing this here saves a lot of checks in all | |
1875 | * the calling paths, and returning zero seems | |
1876 | * safe for them anyway. | |
1877 | */ | |
1878 | if (period == 0) | |
1879 | return 0; | |
1880 | ||
1881 | return div64_u64(runtime << 20, period); | |
1882 | } | |
1883 | ||
1884 | #ifdef CONFIG_SMP | |
1885 | inline struct dl_bw *dl_bw_of(int i) | |
1886 | { | |
1887 | return &cpu_rq(i)->rd->dl_bw; | |
1888 | } | |
1889 | ||
1890 | static inline int __dl_span_weight(struct rq *rq) | |
1891 | { | |
1892 | return cpumask_weight(rq->rd->span); | |
1893 | } | |
1894 | #else | |
1895 | inline struct dl_bw *dl_bw_of(int i) | |
1896 | { | |
1897 | return &cpu_rq(i)->dl.dl_bw; | |
1898 | } | |
1899 | ||
1900 | static inline int __dl_span_weight(struct rq *rq) | |
1901 | { | |
1902 | return 1; | |
1903 | } | |
1904 | #endif | |
1905 | ||
1906 | static inline | |
1907 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
1908 | { | |
1909 | dl_b->total_bw -= tsk_bw; | |
1910 | } | |
1911 | ||
1912 | static inline | |
1913 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
1914 | { | |
1915 | dl_b->total_bw += tsk_bw; | |
1916 | } | |
1917 | ||
1918 | static inline | |
1919 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
1920 | { | |
1921 | return dl_b->bw != -1 && | |
1922 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
1923 | } | |
1924 | ||
1925 | /* | |
1926 | * We must be sure that accepting a new task (or allowing changing the | |
1927 | * parameters of an existing one) is consistent with the bandwidth | |
1928 | * constraints. If yes, this function also accordingly updates the currently | |
1929 | * allocated bandwidth to reflect the new situation. | |
1930 | * | |
1931 | * This function is called while holding p's rq->lock. | |
1932 | */ | |
1933 | static int dl_overflow(struct task_struct *p, int policy, | |
1934 | const struct sched_attr *attr) | |
1935 | { | |
1936 | ||
1937 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
1938 | u64 period = attr->sched_period; | |
1939 | u64 runtime = attr->sched_runtime; | |
1940 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
1941 | int cpus = __dl_span_weight(task_rq(p)); | |
1942 | int err = -1; | |
1943 | ||
1944 | if (new_bw == p->dl.dl_bw) | |
1945 | return 0; | |
1946 | ||
1947 | /* | |
1948 | * Either if a task, enters, leave, or stays -deadline but changes | |
1949 | * its parameters, we may need to update accordingly the total | |
1950 | * allocated bandwidth of the container. | |
1951 | */ | |
1952 | raw_spin_lock(&dl_b->lock); | |
1953 | if (dl_policy(policy) && !task_has_dl_policy(p) && | |
1954 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
1955 | __dl_add(dl_b, new_bw); | |
1956 | err = 0; | |
1957 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
1958 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
1959 | __dl_clear(dl_b, p->dl.dl_bw); | |
1960 | __dl_add(dl_b, new_bw); | |
1961 | err = 0; | |
1962 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
1963 | __dl_clear(dl_b, p->dl.dl_bw); | |
1964 | err = 0; | |
1965 | } | |
1966 | raw_spin_unlock(&dl_b->lock); | |
1967 | ||
1968 | return err; | |
1969 | } | |
1970 | ||
1971 | extern void init_dl_bw(struct dl_bw *dl_b); | |
1972 | ||
1da177e4 LT |
1973 | /* |
1974 | * wake_up_new_task - wake up a newly created task for the first time. | |
1975 | * | |
1976 | * This function will do some initial scheduler statistics housekeeping | |
1977 | * that must be done for every newly created context, then puts the task | |
1978 | * on the runqueue and wakes it. | |
1979 | */ | |
3e51e3ed | 1980 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1981 | { |
1982 | unsigned long flags; | |
dd41f596 | 1983 | struct rq *rq; |
fabf318e | 1984 | |
ab2515c4 | 1985 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1986 | #ifdef CONFIG_SMP |
1987 | /* | |
1988 | * Fork balancing, do it here and not earlier because: | |
1989 | * - cpus_allowed can change in the fork path | |
1990 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1991 | */ |
ac66f547 | 1992 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1993 | #endif |
1994 | ||
a75cdaa9 AS |
1995 | /* Initialize new task's runnable average */ |
1996 | init_task_runnable_average(p); | |
ab2515c4 | 1997 | rq = __task_rq_lock(p); |
cd29fe6f | 1998 | activate_task(rq, p, 0); |
fd2f4419 | 1999 | p->on_rq = 1; |
89363381 | 2000 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2001 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2002 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2003 | if (p->sched_class->task_woken) |
2004 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2005 | #endif |
0122ec5b | 2006 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2007 | } |
2008 | ||
e107be36 AK |
2009 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2010 | ||
2011 | /** | |
80dd99b3 | 2012 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2013 | * @notifier: notifier struct to register |
e107be36 AK |
2014 | */ |
2015 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2016 | { | |
2017 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2018 | } | |
2019 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2020 | ||
2021 | /** | |
2022 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2023 | * @notifier: notifier struct to unregister |
e107be36 AK |
2024 | * |
2025 | * This is safe to call from within a preemption notifier. | |
2026 | */ | |
2027 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2028 | { | |
2029 | hlist_del(¬ifier->link); | |
2030 | } | |
2031 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2032 | ||
2033 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2034 | { | |
2035 | struct preempt_notifier *notifier; | |
e107be36 | 2036 | |
b67bfe0d | 2037 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2038 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2039 | } | |
2040 | ||
2041 | static void | |
2042 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2043 | struct task_struct *next) | |
2044 | { | |
2045 | struct preempt_notifier *notifier; | |
e107be36 | 2046 | |
b67bfe0d | 2047 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2048 | notifier->ops->sched_out(notifier, next); |
2049 | } | |
2050 | ||
6d6bc0ad | 2051 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2052 | |
2053 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2054 | { | |
2055 | } | |
2056 | ||
2057 | static void | |
2058 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2059 | struct task_struct *next) | |
2060 | { | |
2061 | } | |
2062 | ||
6d6bc0ad | 2063 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2064 | |
4866cde0 NP |
2065 | /** |
2066 | * prepare_task_switch - prepare to switch tasks | |
2067 | * @rq: the runqueue preparing to switch | |
421cee29 | 2068 | * @prev: the current task that is being switched out |
4866cde0 NP |
2069 | * @next: the task we are going to switch to. |
2070 | * | |
2071 | * This is called with the rq lock held and interrupts off. It must | |
2072 | * be paired with a subsequent finish_task_switch after the context | |
2073 | * switch. | |
2074 | * | |
2075 | * prepare_task_switch sets up locking and calls architecture specific | |
2076 | * hooks. | |
2077 | */ | |
e107be36 AK |
2078 | static inline void |
2079 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2080 | struct task_struct *next) | |
4866cde0 | 2081 | { |
895dd92c | 2082 | trace_sched_switch(prev, next); |
43148951 | 2083 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2084 | perf_event_task_sched_out(prev, next); |
e107be36 | 2085 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2086 | prepare_lock_switch(rq, next); |
2087 | prepare_arch_switch(next); | |
2088 | } | |
2089 | ||
1da177e4 LT |
2090 | /** |
2091 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2092 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2093 | * @prev: the thread we just switched away from. |
2094 | * | |
4866cde0 NP |
2095 | * finish_task_switch must be called after the context switch, paired |
2096 | * with a prepare_task_switch call before the context switch. | |
2097 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2098 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2099 | * |
2100 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2101 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2102 | * with the lock held can cause deadlocks; see schedule() for |
2103 | * details.) | |
2104 | */ | |
a9957449 | 2105 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2106 | __releases(rq->lock) |
2107 | { | |
1da177e4 | 2108 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2109 | long prev_state; |
1da177e4 LT |
2110 | |
2111 | rq->prev_mm = NULL; | |
2112 | ||
2113 | /* | |
2114 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2115 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2116 | * schedule one last time. The schedule call will never return, and |
2117 | * the scheduled task must drop that reference. | |
c394cc9f | 2118 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2119 | * still held, otherwise prev could be scheduled on another cpu, die |
2120 | * there before we look at prev->state, and then the reference would | |
2121 | * be dropped twice. | |
2122 | * Manfred Spraul <manfred@colorfullife.com> | |
2123 | */ | |
55a101f8 | 2124 | prev_state = prev->state; |
bf9fae9f | 2125 | vtime_task_switch(prev); |
4866cde0 | 2126 | finish_arch_switch(prev); |
a8d757ef | 2127 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2128 | finish_lock_switch(rq, prev); |
01f23e16 | 2129 | finish_arch_post_lock_switch(); |
e8fa1362 | 2130 | |
e107be36 | 2131 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2132 | if (mm) |
2133 | mmdrop(mm); | |
c394cc9f | 2134 | if (unlikely(prev_state == TASK_DEAD)) { |
f809ca9a MG |
2135 | task_numa_free(prev); |
2136 | ||
e6c390f2 DF |
2137 | if (prev->sched_class->task_dead) |
2138 | prev->sched_class->task_dead(prev); | |
2139 | ||
c6fd91f0 | 2140 | /* |
2141 | * Remove function-return probe instances associated with this | |
2142 | * task and put them back on the free list. | |
9761eea8 | 2143 | */ |
c6fd91f0 | 2144 | kprobe_flush_task(prev); |
1da177e4 | 2145 | put_task_struct(prev); |
c6fd91f0 | 2146 | } |
99e5ada9 FW |
2147 | |
2148 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2149 | } |
2150 | ||
3f029d3c GH |
2151 | #ifdef CONFIG_SMP |
2152 | ||
2153 | /* assumes rq->lock is held */ | |
2154 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2155 | { | |
2156 | if (prev->sched_class->pre_schedule) | |
2157 | prev->sched_class->pre_schedule(rq, prev); | |
2158 | } | |
2159 | ||
2160 | /* rq->lock is NOT held, but preemption is disabled */ | |
2161 | static inline void post_schedule(struct rq *rq) | |
2162 | { | |
2163 | if (rq->post_schedule) { | |
2164 | unsigned long flags; | |
2165 | ||
05fa785c | 2166 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2167 | if (rq->curr->sched_class->post_schedule) |
2168 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2169 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2170 | |
2171 | rq->post_schedule = 0; | |
2172 | } | |
2173 | } | |
2174 | ||
2175 | #else | |
da19ab51 | 2176 | |
3f029d3c GH |
2177 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2178 | { | |
2179 | } | |
2180 | ||
2181 | static inline void post_schedule(struct rq *rq) | |
2182 | { | |
1da177e4 LT |
2183 | } |
2184 | ||
3f029d3c GH |
2185 | #endif |
2186 | ||
1da177e4 LT |
2187 | /** |
2188 | * schedule_tail - first thing a freshly forked thread must call. | |
2189 | * @prev: the thread we just switched away from. | |
2190 | */ | |
36c8b586 | 2191 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2192 | __releases(rq->lock) |
2193 | { | |
70b97a7f IM |
2194 | struct rq *rq = this_rq(); |
2195 | ||
4866cde0 | 2196 | finish_task_switch(rq, prev); |
da19ab51 | 2197 | |
3f029d3c GH |
2198 | /* |
2199 | * FIXME: do we need to worry about rq being invalidated by the | |
2200 | * task_switch? | |
2201 | */ | |
2202 | post_schedule(rq); | |
70b97a7f | 2203 | |
4866cde0 NP |
2204 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2205 | /* In this case, finish_task_switch does not reenable preemption */ | |
2206 | preempt_enable(); | |
2207 | #endif | |
1da177e4 | 2208 | if (current->set_child_tid) |
b488893a | 2209 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2210 | } |
2211 | ||
2212 | /* | |
2213 | * context_switch - switch to the new MM and the new | |
2214 | * thread's register state. | |
2215 | */ | |
dd41f596 | 2216 | static inline void |
70b97a7f | 2217 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2218 | struct task_struct *next) |
1da177e4 | 2219 | { |
dd41f596 | 2220 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2221 | |
e107be36 | 2222 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2223 | |
dd41f596 IM |
2224 | mm = next->mm; |
2225 | oldmm = prev->active_mm; | |
9226d125 ZA |
2226 | /* |
2227 | * For paravirt, this is coupled with an exit in switch_to to | |
2228 | * combine the page table reload and the switch backend into | |
2229 | * one hypercall. | |
2230 | */ | |
224101ed | 2231 | arch_start_context_switch(prev); |
9226d125 | 2232 | |
31915ab4 | 2233 | if (!mm) { |
1da177e4 LT |
2234 | next->active_mm = oldmm; |
2235 | atomic_inc(&oldmm->mm_count); | |
2236 | enter_lazy_tlb(oldmm, next); | |
2237 | } else | |
2238 | switch_mm(oldmm, mm, next); | |
2239 | ||
31915ab4 | 2240 | if (!prev->mm) { |
1da177e4 | 2241 | prev->active_mm = NULL; |
1da177e4 LT |
2242 | rq->prev_mm = oldmm; |
2243 | } | |
3a5f5e48 IM |
2244 | /* |
2245 | * Since the runqueue lock will be released by the next | |
2246 | * task (which is an invalid locking op but in the case | |
2247 | * of the scheduler it's an obvious special-case), so we | |
2248 | * do an early lockdep release here: | |
2249 | */ | |
2250 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2251 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2252 | #endif |
1da177e4 | 2253 | |
91d1aa43 | 2254 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2255 | /* Here we just switch the register state and the stack. */ |
2256 | switch_to(prev, next, prev); | |
2257 | ||
dd41f596 IM |
2258 | barrier(); |
2259 | /* | |
2260 | * this_rq must be evaluated again because prev may have moved | |
2261 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2262 | * frame will be invalid. | |
2263 | */ | |
2264 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2265 | } |
2266 | ||
2267 | /* | |
1c3e8264 | 2268 | * nr_running and nr_context_switches: |
1da177e4 LT |
2269 | * |
2270 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2271 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2272 | */ |
2273 | unsigned long nr_running(void) | |
2274 | { | |
2275 | unsigned long i, sum = 0; | |
2276 | ||
2277 | for_each_online_cpu(i) | |
2278 | sum += cpu_rq(i)->nr_running; | |
2279 | ||
2280 | return sum; | |
f711f609 | 2281 | } |
1da177e4 | 2282 | |
1da177e4 | 2283 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2284 | { |
cc94abfc SR |
2285 | int i; |
2286 | unsigned long long sum = 0; | |
46cb4b7c | 2287 | |
0a945022 | 2288 | for_each_possible_cpu(i) |
1da177e4 | 2289 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2290 | |
1da177e4 LT |
2291 | return sum; |
2292 | } | |
483b4ee6 | 2293 | |
1da177e4 LT |
2294 | unsigned long nr_iowait(void) |
2295 | { | |
2296 | unsigned long i, sum = 0; | |
483b4ee6 | 2297 | |
0a945022 | 2298 | for_each_possible_cpu(i) |
1da177e4 | 2299 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2300 | |
1da177e4 LT |
2301 | return sum; |
2302 | } | |
483b4ee6 | 2303 | |
8c215bd3 | 2304 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2305 | { |
8c215bd3 | 2306 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2307 | return atomic_read(&this->nr_iowait); |
2308 | } | |
46cb4b7c | 2309 | |
dd41f596 | 2310 | #ifdef CONFIG_SMP |
8a0be9ef | 2311 | |
46cb4b7c | 2312 | /* |
38022906 PZ |
2313 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2314 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2315 | */ |
38022906 | 2316 | void sched_exec(void) |
46cb4b7c | 2317 | { |
38022906 | 2318 | struct task_struct *p = current; |
1da177e4 | 2319 | unsigned long flags; |
0017d735 | 2320 | int dest_cpu; |
46cb4b7c | 2321 | |
8f42ced9 | 2322 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2323 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2324 | if (dest_cpu == smp_processor_id()) |
2325 | goto unlock; | |
38022906 | 2326 | |
8f42ced9 | 2327 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2328 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2329 | |
8f42ced9 PZ |
2330 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2331 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2332 | return; |
2333 | } | |
0017d735 | 2334 | unlock: |
8f42ced9 | 2335 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2336 | } |
dd41f596 | 2337 | |
1da177e4 LT |
2338 | #endif |
2339 | ||
1da177e4 | 2340 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2341 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2342 | |
2343 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2344 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2345 | |
2346 | /* | |
c5f8d995 | 2347 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2348 | * @p in case that task is currently running. |
c5f8d995 HS |
2349 | * |
2350 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2351 | */ |
c5f8d995 HS |
2352 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2353 | { | |
2354 | u64 ns = 0; | |
2355 | ||
2356 | if (task_current(rq, p)) { | |
2357 | update_rq_clock(rq); | |
78becc27 | 2358 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2359 | if ((s64)ns < 0) |
2360 | ns = 0; | |
2361 | } | |
2362 | ||
2363 | return ns; | |
2364 | } | |
2365 | ||
bb34d92f | 2366 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2367 | { |
1da177e4 | 2368 | unsigned long flags; |
41b86e9c | 2369 | struct rq *rq; |
bb34d92f | 2370 | u64 ns = 0; |
48f24c4d | 2371 | |
41b86e9c | 2372 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2373 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2374 | task_rq_unlock(rq, p, &flags); |
1508487e | 2375 | |
c5f8d995 HS |
2376 | return ns; |
2377 | } | |
f06febc9 | 2378 | |
c5f8d995 HS |
2379 | /* |
2380 | * Return accounted runtime for the task. | |
2381 | * In case the task is currently running, return the runtime plus current's | |
2382 | * pending runtime that have not been accounted yet. | |
2383 | */ | |
2384 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2385 | { | |
2386 | unsigned long flags; | |
2387 | struct rq *rq; | |
2388 | u64 ns = 0; | |
2389 | ||
911b2898 PZ |
2390 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2391 | /* | |
2392 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2393 | * So we have a optimization chance when the task's delta_exec is 0. | |
2394 | * Reading ->on_cpu is racy, but this is ok. | |
2395 | * | |
2396 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2397 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2398 | * indistinguishable from the read occurring a few cycles earlier. | |
2399 | */ | |
2400 | if (!p->on_cpu) | |
2401 | return p->se.sum_exec_runtime; | |
2402 | #endif | |
2403 | ||
c5f8d995 HS |
2404 | rq = task_rq_lock(p, &flags); |
2405 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2406 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2407 | |
2408 | return ns; | |
2409 | } | |
48f24c4d | 2410 | |
7835b98b CL |
2411 | /* |
2412 | * This function gets called by the timer code, with HZ frequency. | |
2413 | * We call it with interrupts disabled. | |
7835b98b CL |
2414 | */ |
2415 | void scheduler_tick(void) | |
2416 | { | |
7835b98b CL |
2417 | int cpu = smp_processor_id(); |
2418 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2419 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2420 | |
2421 | sched_clock_tick(); | |
dd41f596 | 2422 | |
05fa785c | 2423 | raw_spin_lock(&rq->lock); |
3e51f33f | 2424 | update_rq_clock(rq); |
fa85ae24 | 2425 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2426 | update_cpu_load_active(rq); |
05fa785c | 2427 | raw_spin_unlock(&rq->lock); |
7835b98b | 2428 | |
e9d2b064 | 2429 | perf_event_task_tick(); |
e220d2dc | 2430 | |
e418e1c2 | 2431 | #ifdef CONFIG_SMP |
6eb57e0d | 2432 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 2433 | trigger_load_balance(rq, cpu); |
e418e1c2 | 2434 | #endif |
265f22a9 | 2435 | rq_last_tick_reset(rq); |
1da177e4 LT |
2436 | } |
2437 | ||
265f22a9 FW |
2438 | #ifdef CONFIG_NO_HZ_FULL |
2439 | /** | |
2440 | * scheduler_tick_max_deferment | |
2441 | * | |
2442 | * Keep at least one tick per second when a single | |
2443 | * active task is running because the scheduler doesn't | |
2444 | * yet completely support full dynticks environment. | |
2445 | * | |
2446 | * This makes sure that uptime, CFS vruntime, load | |
2447 | * balancing, etc... continue to move forward, even | |
2448 | * with a very low granularity. | |
e69f6186 YB |
2449 | * |
2450 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2451 | */ |
2452 | u64 scheduler_tick_max_deferment(void) | |
2453 | { | |
2454 | struct rq *rq = this_rq(); | |
2455 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2456 | ||
2457 | next = rq->last_sched_tick + HZ; | |
2458 | ||
2459 | if (time_before_eq(next, now)) | |
2460 | return 0; | |
2461 | ||
2462 | return jiffies_to_usecs(next - now) * NSEC_PER_USEC; | |
1da177e4 | 2463 | } |
265f22a9 | 2464 | #endif |
1da177e4 | 2465 | |
132380a0 | 2466 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2467 | { |
2468 | if (in_lock_functions(addr)) { | |
2469 | addr = CALLER_ADDR2; | |
2470 | if (in_lock_functions(addr)) | |
2471 | addr = CALLER_ADDR3; | |
2472 | } | |
2473 | return addr; | |
2474 | } | |
1da177e4 | 2475 | |
7e49fcce SR |
2476 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2477 | defined(CONFIG_PREEMPT_TRACER)) | |
2478 | ||
bdb43806 | 2479 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2480 | { |
6cd8a4bb | 2481 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2482 | /* |
2483 | * Underflow? | |
2484 | */ | |
9a11b49a IM |
2485 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2486 | return; | |
6cd8a4bb | 2487 | #endif |
bdb43806 | 2488 | __preempt_count_add(val); |
6cd8a4bb | 2489 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2490 | /* |
2491 | * Spinlock count overflowing soon? | |
2492 | */ | |
33859f7f MOS |
2493 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2494 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
2495 | #endif |
2496 | if (preempt_count() == val) | |
2497 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 | 2498 | } |
bdb43806 | 2499 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2500 | |
bdb43806 | 2501 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2502 | { |
6cd8a4bb | 2503 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2504 | /* |
2505 | * Underflow? | |
2506 | */ | |
01e3eb82 | 2507 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2508 | return; |
1da177e4 LT |
2509 | /* |
2510 | * Is the spinlock portion underflowing? | |
2511 | */ | |
9a11b49a IM |
2512 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2513 | !(preempt_count() & PREEMPT_MASK))) | |
2514 | return; | |
6cd8a4bb | 2515 | #endif |
9a11b49a | 2516 | |
6cd8a4bb SR |
2517 | if (preempt_count() == val) |
2518 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2519 | __preempt_count_sub(val); |
1da177e4 | 2520 | } |
bdb43806 | 2521 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2522 | |
2523 | #endif | |
2524 | ||
2525 | /* | |
dd41f596 | 2526 | * Print scheduling while atomic bug: |
1da177e4 | 2527 | */ |
dd41f596 | 2528 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2529 | { |
664dfa65 DJ |
2530 | if (oops_in_progress) |
2531 | return; | |
2532 | ||
3df0fc5b PZ |
2533 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2534 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2535 | |
dd41f596 | 2536 | debug_show_held_locks(prev); |
e21f5b15 | 2537 | print_modules(); |
dd41f596 IM |
2538 | if (irqs_disabled()) |
2539 | print_irqtrace_events(prev); | |
6135fc1e | 2540 | dump_stack(); |
373d4d09 | 2541 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2542 | } |
1da177e4 | 2543 | |
dd41f596 IM |
2544 | /* |
2545 | * Various schedule()-time debugging checks and statistics: | |
2546 | */ | |
2547 | static inline void schedule_debug(struct task_struct *prev) | |
2548 | { | |
1da177e4 | 2549 | /* |
41a2d6cf | 2550 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2551 | * schedule() atomically, we ignore that path. Otherwise whine |
2552 | * if we are scheduling when we should not. | |
1da177e4 | 2553 | */ |
192301e7 | 2554 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2555 | __schedule_bug(prev); |
b3fbab05 | 2556 | rcu_sleep_check(); |
dd41f596 | 2557 | |
1da177e4 LT |
2558 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2559 | ||
2d72376b | 2560 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2561 | } |
2562 | ||
6cecd084 | 2563 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 2564 | { |
61eadef6 | 2565 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 2566 | update_rq_clock(rq); |
6cecd084 | 2567 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
2568 | } |
2569 | ||
dd41f596 IM |
2570 | /* |
2571 | * Pick up the highest-prio task: | |
2572 | */ | |
2573 | static inline struct task_struct * | |
b67802ea | 2574 | pick_next_task(struct rq *rq) |
dd41f596 | 2575 | { |
5522d5d5 | 2576 | const struct sched_class *class; |
dd41f596 | 2577 | struct task_struct *p; |
1da177e4 LT |
2578 | |
2579 | /* | |
dd41f596 IM |
2580 | * Optimization: we know that if all tasks are in |
2581 | * the fair class we can call that function directly: | |
1da177e4 | 2582 | */ |
953bfcd1 | 2583 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 2584 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
2585 | if (likely(p)) |
2586 | return p; | |
1da177e4 LT |
2587 | } |
2588 | ||
34f971f6 | 2589 | for_each_class(class) { |
fb8d4724 | 2590 | p = class->pick_next_task(rq); |
dd41f596 IM |
2591 | if (p) |
2592 | return p; | |
dd41f596 | 2593 | } |
34f971f6 PZ |
2594 | |
2595 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2596 | } |
1da177e4 | 2597 | |
dd41f596 | 2598 | /* |
c259e01a | 2599 | * __schedule() is the main scheduler function. |
edde96ea PE |
2600 | * |
2601 | * The main means of driving the scheduler and thus entering this function are: | |
2602 | * | |
2603 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2604 | * | |
2605 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2606 | * paths. For example, see arch/x86/entry_64.S. | |
2607 | * | |
2608 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2609 | * interrupt handler scheduler_tick(). | |
2610 | * | |
2611 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2612 | * task to the run-queue and that's it. | |
2613 | * | |
2614 | * Now, if the new task added to the run-queue preempts the current | |
2615 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2616 | * called on the nearest possible occasion: | |
2617 | * | |
2618 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2619 | * | |
2620 | * - in syscall or exception context, at the next outmost | |
2621 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2622 | * spin_unlock()!) | |
2623 | * | |
2624 | * - in IRQ context, return from interrupt-handler to | |
2625 | * preemptible context | |
2626 | * | |
2627 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2628 | * then at the next: | |
2629 | * | |
2630 | * - cond_resched() call | |
2631 | * - explicit schedule() call | |
2632 | * - return from syscall or exception to user-space | |
2633 | * - return from interrupt-handler to user-space | |
dd41f596 | 2634 | */ |
c259e01a | 2635 | static void __sched __schedule(void) |
dd41f596 IM |
2636 | { |
2637 | struct task_struct *prev, *next; | |
67ca7bde | 2638 | unsigned long *switch_count; |
dd41f596 | 2639 | struct rq *rq; |
31656519 | 2640 | int cpu; |
dd41f596 | 2641 | |
ff743345 PZ |
2642 | need_resched: |
2643 | preempt_disable(); | |
dd41f596 IM |
2644 | cpu = smp_processor_id(); |
2645 | rq = cpu_rq(cpu); | |
25502a6c | 2646 | rcu_note_context_switch(cpu); |
dd41f596 | 2647 | prev = rq->curr; |
dd41f596 | 2648 | |
dd41f596 | 2649 | schedule_debug(prev); |
1da177e4 | 2650 | |
31656519 | 2651 | if (sched_feat(HRTICK)) |
f333fdc9 | 2652 | hrtick_clear(rq); |
8f4d37ec | 2653 | |
e0acd0a6 ON |
2654 | /* |
2655 | * Make sure that signal_pending_state()->signal_pending() below | |
2656 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2657 | * done by the caller to avoid the race with signal_wake_up(). | |
2658 | */ | |
2659 | smp_mb__before_spinlock(); | |
05fa785c | 2660 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2661 | |
246d86b5 | 2662 | switch_count = &prev->nivcsw; |
1da177e4 | 2663 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2664 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2665 | prev->state = TASK_RUNNING; |
21aa9af0 | 2666 | } else { |
2acca55e PZ |
2667 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2668 | prev->on_rq = 0; | |
2669 | ||
21aa9af0 | 2670 | /* |
2acca55e PZ |
2671 | * If a worker went to sleep, notify and ask workqueue |
2672 | * whether it wants to wake up a task to maintain | |
2673 | * concurrency. | |
21aa9af0 TH |
2674 | */ |
2675 | if (prev->flags & PF_WQ_WORKER) { | |
2676 | struct task_struct *to_wakeup; | |
2677 | ||
2678 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2679 | if (to_wakeup) | |
2680 | try_to_wake_up_local(to_wakeup); | |
2681 | } | |
21aa9af0 | 2682 | } |
dd41f596 | 2683 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2684 | } |
2685 | ||
3f029d3c | 2686 | pre_schedule(rq, prev); |
f65eda4f | 2687 | |
dd41f596 | 2688 | if (unlikely(!rq->nr_running)) |
1da177e4 | 2689 | idle_balance(cpu, rq); |
1da177e4 | 2690 | |
df1c99d4 | 2691 | put_prev_task(rq, prev); |
b67802ea | 2692 | next = pick_next_task(rq); |
f26f9aff | 2693 | clear_tsk_need_resched(prev); |
f27dde8d | 2694 | clear_preempt_need_resched(); |
f26f9aff | 2695 | rq->skip_clock_update = 0; |
1da177e4 | 2696 | |
1da177e4 | 2697 | if (likely(prev != next)) { |
1da177e4 LT |
2698 | rq->nr_switches++; |
2699 | rq->curr = next; | |
2700 | ++*switch_count; | |
2701 | ||
dd41f596 | 2702 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2703 | /* |
246d86b5 ON |
2704 | * The context switch have flipped the stack from under us |
2705 | * and restored the local variables which were saved when | |
2706 | * this task called schedule() in the past. prev == current | |
2707 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2708 | */ |
2709 | cpu = smp_processor_id(); | |
2710 | rq = cpu_rq(cpu); | |
1da177e4 | 2711 | } else |
05fa785c | 2712 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2713 | |
3f029d3c | 2714 | post_schedule(rq); |
1da177e4 | 2715 | |
ba74c144 | 2716 | sched_preempt_enable_no_resched(); |
ff743345 | 2717 | if (need_resched()) |
1da177e4 LT |
2718 | goto need_resched; |
2719 | } | |
c259e01a | 2720 | |
9c40cef2 TG |
2721 | static inline void sched_submit_work(struct task_struct *tsk) |
2722 | { | |
3c7d5184 | 2723 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2724 | return; |
2725 | /* | |
2726 | * If we are going to sleep and we have plugged IO queued, | |
2727 | * make sure to submit it to avoid deadlocks. | |
2728 | */ | |
2729 | if (blk_needs_flush_plug(tsk)) | |
2730 | blk_schedule_flush_plug(tsk); | |
2731 | } | |
2732 | ||
6ebbe7a0 | 2733 | asmlinkage void __sched schedule(void) |
c259e01a | 2734 | { |
9c40cef2 TG |
2735 | struct task_struct *tsk = current; |
2736 | ||
2737 | sched_submit_work(tsk); | |
c259e01a TG |
2738 | __schedule(); |
2739 | } | |
1da177e4 LT |
2740 | EXPORT_SYMBOL(schedule); |
2741 | ||
91d1aa43 | 2742 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2743 | asmlinkage void __sched schedule_user(void) |
2744 | { | |
2745 | /* | |
2746 | * If we come here after a random call to set_need_resched(), | |
2747 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2748 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2749 | * we find a better solution. | |
2750 | */ | |
91d1aa43 | 2751 | user_exit(); |
20ab65e3 | 2752 | schedule(); |
91d1aa43 | 2753 | user_enter(); |
20ab65e3 FW |
2754 | } |
2755 | #endif | |
2756 | ||
c5491ea7 TG |
2757 | /** |
2758 | * schedule_preempt_disabled - called with preemption disabled | |
2759 | * | |
2760 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2761 | */ | |
2762 | void __sched schedule_preempt_disabled(void) | |
2763 | { | |
ba74c144 | 2764 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2765 | schedule(); |
2766 | preempt_disable(); | |
2767 | } | |
2768 | ||
1da177e4 LT |
2769 | #ifdef CONFIG_PREEMPT |
2770 | /* | |
2ed6e34f | 2771 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2772 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2773 | * occur there and call schedule directly. |
2774 | */ | |
d1f74e20 | 2775 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 | 2776 | { |
1da177e4 LT |
2777 | /* |
2778 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2779 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2780 | */ |
fbb00b56 | 2781 | if (likely(!preemptible())) |
1da177e4 LT |
2782 | return; |
2783 | ||
3a5c359a | 2784 | do { |
bdb43806 | 2785 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2786 | __schedule(); |
bdb43806 | 2787 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2788 | |
3a5c359a AK |
2789 | /* |
2790 | * Check again in case we missed a preemption opportunity | |
2791 | * between schedule and now. | |
2792 | */ | |
2793 | barrier(); | |
5ed0cec0 | 2794 | } while (need_resched()); |
1da177e4 | 2795 | } |
1da177e4 | 2796 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2797 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2798 | |
2799 | /* | |
2ed6e34f | 2800 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2801 | * off of irq context. |
2802 | * Note, that this is called and return with irqs disabled. This will | |
2803 | * protect us against recursive calling from irq. | |
2804 | */ | |
2805 | asmlinkage void __sched preempt_schedule_irq(void) | |
2806 | { | |
b22366cd | 2807 | enum ctx_state prev_state; |
6478d880 | 2808 | |
2ed6e34f | 2809 | /* Catch callers which need to be fixed */ |
f27dde8d | 2810 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2811 | |
b22366cd FW |
2812 | prev_state = exception_enter(); |
2813 | ||
3a5c359a | 2814 | do { |
bdb43806 | 2815 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2816 | local_irq_enable(); |
c259e01a | 2817 | __schedule(); |
3a5c359a | 2818 | local_irq_disable(); |
bdb43806 | 2819 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2820 | |
3a5c359a AK |
2821 | /* |
2822 | * Check again in case we missed a preemption opportunity | |
2823 | * between schedule and now. | |
2824 | */ | |
2825 | barrier(); | |
5ed0cec0 | 2826 | } while (need_resched()); |
b22366cd FW |
2827 | |
2828 | exception_exit(prev_state); | |
1da177e4 LT |
2829 | } |
2830 | ||
63859d4f | 2831 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2832 | void *key) |
1da177e4 | 2833 | { |
63859d4f | 2834 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2835 | } |
1da177e4 LT |
2836 | EXPORT_SYMBOL(default_wake_function); |
2837 | ||
8cbbe86d AK |
2838 | static long __sched |
2839 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 2840 | { |
0fec171c IM |
2841 | unsigned long flags; |
2842 | wait_queue_t wait; | |
2843 | ||
2844 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 2845 | |
8cbbe86d | 2846 | __set_current_state(state); |
1da177e4 | 2847 | |
8cbbe86d AK |
2848 | spin_lock_irqsave(&q->lock, flags); |
2849 | __add_wait_queue(q, &wait); | |
2850 | spin_unlock(&q->lock); | |
2851 | timeout = schedule_timeout(timeout); | |
2852 | spin_lock_irq(&q->lock); | |
2853 | __remove_wait_queue(q, &wait); | |
2854 | spin_unlock_irqrestore(&q->lock, flags); | |
2855 | ||
2856 | return timeout; | |
2857 | } | |
2858 | ||
2859 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
2860 | { | |
2861 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 2862 | } |
1da177e4 LT |
2863 | EXPORT_SYMBOL(interruptible_sleep_on); |
2864 | ||
0fec171c | 2865 | long __sched |
95cdf3b7 | 2866 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2867 | { |
8cbbe86d | 2868 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 2869 | } |
1da177e4 LT |
2870 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
2871 | ||
0fec171c | 2872 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 2873 | { |
8cbbe86d | 2874 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 2875 | } |
1da177e4 LT |
2876 | EXPORT_SYMBOL(sleep_on); |
2877 | ||
0fec171c | 2878 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2879 | { |
8cbbe86d | 2880 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 2881 | } |
1da177e4 LT |
2882 | EXPORT_SYMBOL(sleep_on_timeout); |
2883 | ||
b29739f9 IM |
2884 | #ifdef CONFIG_RT_MUTEXES |
2885 | ||
2886 | /* | |
2887 | * rt_mutex_setprio - set the current priority of a task | |
2888 | * @p: task | |
2889 | * @prio: prio value (kernel-internal form) | |
2890 | * | |
2891 | * This function changes the 'effective' priority of a task. It does | |
2892 | * not touch ->normal_prio like __setscheduler(). | |
2893 | * | |
2894 | * Used by the rt_mutex code to implement priority inheritance logic. | |
2895 | */ | |
36c8b586 | 2896 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2897 | { |
2d3d891d | 2898 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2899 | struct rq *rq; |
83ab0aa0 | 2900 | const struct sched_class *prev_class; |
b29739f9 | 2901 | |
aab03e05 | 2902 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2903 | |
0122ec5b | 2904 | rq = __task_rq_lock(p); |
b29739f9 | 2905 | |
1c4dd99b TG |
2906 | /* |
2907 | * Idle task boosting is a nono in general. There is one | |
2908 | * exception, when PREEMPT_RT and NOHZ is active: | |
2909 | * | |
2910 | * The idle task calls get_next_timer_interrupt() and holds | |
2911 | * the timer wheel base->lock on the CPU and another CPU wants | |
2912 | * to access the timer (probably to cancel it). We can safely | |
2913 | * ignore the boosting request, as the idle CPU runs this code | |
2914 | * with interrupts disabled and will complete the lock | |
2915 | * protected section without being interrupted. So there is no | |
2916 | * real need to boost. | |
2917 | */ | |
2918 | if (unlikely(p == rq->idle)) { | |
2919 | WARN_ON(p != rq->curr); | |
2920 | WARN_ON(p->pi_blocked_on); | |
2921 | goto out_unlock; | |
2922 | } | |
2923 | ||
a8027073 | 2924 | trace_sched_pi_setprio(p, prio); |
2d3d891d | 2925 | p->pi_top_task = rt_mutex_get_top_task(p); |
d5f9f942 | 2926 | oldprio = p->prio; |
83ab0aa0 | 2927 | prev_class = p->sched_class; |
fd2f4419 | 2928 | on_rq = p->on_rq; |
051a1d1a | 2929 | running = task_current(rq, p); |
0e1f3483 | 2930 | if (on_rq) |
69be72c1 | 2931 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2932 | if (running) |
2933 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2934 | |
2d3d891d DF |
2935 | /* |
2936 | * Boosting condition are: | |
2937 | * 1. -rt task is running and holds mutex A | |
2938 | * --> -dl task blocks on mutex A | |
2939 | * | |
2940 | * 2. -dl task is running and holds mutex A | |
2941 | * --> -dl task blocks on mutex A and could preempt the | |
2942 | * running task | |
2943 | */ | |
2944 | if (dl_prio(prio)) { | |
2945 | if (!dl_prio(p->normal_prio) || (p->pi_top_task && | |
2946 | dl_entity_preempt(&p->pi_top_task->dl, &p->dl))) { | |
2947 | p->dl.dl_boosted = 1; | |
2948 | p->dl.dl_throttled = 0; | |
2949 | enqueue_flag = ENQUEUE_REPLENISH; | |
2950 | } else | |
2951 | p->dl.dl_boosted = 0; | |
aab03e05 | 2952 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
2953 | } else if (rt_prio(prio)) { |
2954 | if (dl_prio(oldprio)) | |
2955 | p->dl.dl_boosted = 0; | |
2956 | if (oldprio < prio) | |
2957 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 2958 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
2959 | } else { |
2960 | if (dl_prio(oldprio)) | |
2961 | p->dl.dl_boosted = 0; | |
dd41f596 | 2962 | p->sched_class = &fair_sched_class; |
2d3d891d | 2963 | } |
dd41f596 | 2964 | |
b29739f9 IM |
2965 | p->prio = prio; |
2966 | ||
0e1f3483 HS |
2967 | if (running) |
2968 | p->sched_class->set_curr_task(rq); | |
da7a735e | 2969 | if (on_rq) |
2d3d891d | 2970 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 2971 | |
da7a735e | 2972 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 2973 | out_unlock: |
0122ec5b | 2974 | __task_rq_unlock(rq); |
b29739f9 | 2975 | } |
b29739f9 | 2976 | #endif |
d50dde5a | 2977 | |
36c8b586 | 2978 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 2979 | { |
dd41f596 | 2980 | int old_prio, delta, on_rq; |
1da177e4 | 2981 | unsigned long flags; |
70b97a7f | 2982 | struct rq *rq; |
1da177e4 LT |
2983 | |
2984 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
2985 | return; | |
2986 | /* | |
2987 | * We have to be careful, if called from sys_setpriority(), | |
2988 | * the task might be in the middle of scheduling on another CPU. | |
2989 | */ | |
2990 | rq = task_rq_lock(p, &flags); | |
2991 | /* | |
2992 | * The RT priorities are set via sched_setscheduler(), but we still | |
2993 | * allow the 'normal' nice value to be set - but as expected | |
2994 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 2995 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 2996 | */ |
aab03e05 | 2997 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
2998 | p->static_prio = NICE_TO_PRIO(nice); |
2999 | goto out_unlock; | |
3000 | } | |
fd2f4419 | 3001 | on_rq = p->on_rq; |
c09595f6 | 3002 | if (on_rq) |
69be72c1 | 3003 | dequeue_task(rq, p, 0); |
1da177e4 | 3004 | |
1da177e4 | 3005 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3006 | set_load_weight(p); |
b29739f9 IM |
3007 | old_prio = p->prio; |
3008 | p->prio = effective_prio(p); | |
3009 | delta = p->prio - old_prio; | |
1da177e4 | 3010 | |
dd41f596 | 3011 | if (on_rq) { |
371fd7e7 | 3012 | enqueue_task(rq, p, 0); |
1da177e4 | 3013 | /* |
d5f9f942 AM |
3014 | * If the task increased its priority or is running and |
3015 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3016 | */ |
d5f9f942 | 3017 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3018 | resched_task(rq->curr); |
3019 | } | |
3020 | out_unlock: | |
0122ec5b | 3021 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3022 | } |
1da177e4 LT |
3023 | EXPORT_SYMBOL(set_user_nice); |
3024 | ||
e43379f1 MM |
3025 | /* |
3026 | * can_nice - check if a task can reduce its nice value | |
3027 | * @p: task | |
3028 | * @nice: nice value | |
3029 | */ | |
36c8b586 | 3030 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3031 | { |
024f4747 MM |
3032 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3033 | int nice_rlim = 20 - nice; | |
48f24c4d | 3034 | |
78d7d407 | 3035 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3036 | capable(CAP_SYS_NICE)); |
3037 | } | |
3038 | ||
1da177e4 LT |
3039 | #ifdef __ARCH_WANT_SYS_NICE |
3040 | ||
3041 | /* | |
3042 | * sys_nice - change the priority of the current process. | |
3043 | * @increment: priority increment | |
3044 | * | |
3045 | * sys_setpriority is a more generic, but much slower function that | |
3046 | * does similar things. | |
3047 | */ | |
5add95d4 | 3048 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3049 | { |
48f24c4d | 3050 | long nice, retval; |
1da177e4 LT |
3051 | |
3052 | /* | |
3053 | * Setpriority might change our priority at the same moment. | |
3054 | * We don't have to worry. Conceptually one call occurs first | |
3055 | * and we have a single winner. | |
3056 | */ | |
e43379f1 MM |
3057 | if (increment < -40) |
3058 | increment = -40; | |
1da177e4 LT |
3059 | if (increment > 40) |
3060 | increment = 40; | |
3061 | ||
2b8f836f | 3062 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3063 | if (nice < -20) |
3064 | nice = -20; | |
3065 | if (nice > 19) | |
3066 | nice = 19; | |
3067 | ||
e43379f1 MM |
3068 | if (increment < 0 && !can_nice(current, nice)) |
3069 | return -EPERM; | |
3070 | ||
1da177e4 LT |
3071 | retval = security_task_setnice(current, nice); |
3072 | if (retval) | |
3073 | return retval; | |
3074 | ||
3075 | set_user_nice(current, nice); | |
3076 | return 0; | |
3077 | } | |
3078 | ||
3079 | #endif | |
3080 | ||
3081 | /** | |
3082 | * task_prio - return the priority value of a given task. | |
3083 | * @p: the task in question. | |
3084 | * | |
e69f6186 | 3085 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3086 | * RT tasks are offset by -200. Normal tasks are centered |
3087 | * around 0, value goes from -16 to +15. | |
3088 | */ | |
36c8b586 | 3089 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3090 | { |
3091 | return p->prio - MAX_RT_PRIO; | |
3092 | } | |
3093 | ||
3094 | /** | |
3095 | * task_nice - return the nice value of a given task. | |
3096 | * @p: the task in question. | |
e69f6186 YB |
3097 | * |
3098 | * Return: The nice value [ -20 ... 0 ... 19 ]. | |
1da177e4 | 3099 | */ |
36c8b586 | 3100 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3101 | { |
3102 | return TASK_NICE(p); | |
3103 | } | |
150d8bed | 3104 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3105 | |
3106 | /** | |
3107 | * idle_cpu - is a given cpu idle currently? | |
3108 | * @cpu: the processor in question. | |
e69f6186 YB |
3109 | * |
3110 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3111 | */ |
3112 | int idle_cpu(int cpu) | |
3113 | { | |
908a3283 TG |
3114 | struct rq *rq = cpu_rq(cpu); |
3115 | ||
3116 | if (rq->curr != rq->idle) | |
3117 | return 0; | |
3118 | ||
3119 | if (rq->nr_running) | |
3120 | return 0; | |
3121 | ||
3122 | #ifdef CONFIG_SMP | |
3123 | if (!llist_empty(&rq->wake_list)) | |
3124 | return 0; | |
3125 | #endif | |
3126 | ||
3127 | return 1; | |
1da177e4 LT |
3128 | } |
3129 | ||
1da177e4 LT |
3130 | /** |
3131 | * idle_task - return the idle task for a given cpu. | |
3132 | * @cpu: the processor in question. | |
e69f6186 YB |
3133 | * |
3134 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3135 | */ |
36c8b586 | 3136 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3137 | { |
3138 | return cpu_rq(cpu)->idle; | |
3139 | } | |
3140 | ||
3141 | /** | |
3142 | * find_process_by_pid - find a process with a matching PID value. | |
3143 | * @pid: the pid in question. | |
e69f6186 YB |
3144 | * |
3145 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3146 | */ |
a9957449 | 3147 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3148 | { |
228ebcbe | 3149 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3150 | } |
3151 | ||
aab03e05 DF |
3152 | /* |
3153 | * This function initializes the sched_dl_entity of a newly becoming | |
3154 | * SCHED_DEADLINE task. | |
3155 | * | |
3156 | * Only the static values are considered here, the actual runtime and the | |
3157 | * absolute deadline will be properly calculated when the task is enqueued | |
3158 | * for the first time with its new policy. | |
3159 | */ | |
3160 | static void | |
3161 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3162 | { | |
3163 | struct sched_dl_entity *dl_se = &p->dl; | |
3164 | ||
3165 | init_dl_task_timer(dl_se); | |
3166 | dl_se->dl_runtime = attr->sched_runtime; | |
3167 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3168 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3169 | dl_se->flags = attr->sched_flags; |
332ac17e | 3170 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3171 | dl_se->dl_throttled = 0; |
3172 | dl_se->dl_new = 1; | |
3173 | } | |
3174 | ||
d50dde5a DF |
3175 | /* Actually do priority change: must hold pi & rq lock. */ |
3176 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3177 | const struct sched_attr *attr) | |
1da177e4 | 3178 | { |
d50dde5a DF |
3179 | int policy = attr->sched_policy; |
3180 | ||
1da177e4 | 3181 | p->policy = policy; |
d50dde5a | 3182 | |
aab03e05 DF |
3183 | if (dl_policy(policy)) |
3184 | __setparam_dl(p, attr); | |
3185 | else if (rt_policy(policy)) | |
d50dde5a DF |
3186 | p->rt_priority = attr->sched_priority; |
3187 | else | |
3188 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); | |
3189 | ||
b29739f9 | 3190 | p->normal_prio = normal_prio(p); |
b29739f9 | 3191 | p->prio = rt_mutex_getprio(p); |
d50dde5a | 3192 | |
aab03e05 DF |
3193 | if (dl_prio(p->prio)) |
3194 | p->sched_class = &dl_sched_class; | |
3195 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3196 | p->sched_class = &rt_sched_class; |
3197 | else | |
3198 | p->sched_class = &fair_sched_class; | |
d50dde5a | 3199 | |
2dd73a4f | 3200 | set_load_weight(p); |
1da177e4 | 3201 | } |
aab03e05 DF |
3202 | |
3203 | static void | |
3204 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3205 | { | |
3206 | struct sched_dl_entity *dl_se = &p->dl; | |
3207 | ||
3208 | attr->sched_priority = p->rt_priority; | |
3209 | attr->sched_runtime = dl_se->dl_runtime; | |
3210 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3211 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3212 | attr->sched_flags = dl_se->flags; |
3213 | } | |
3214 | ||
3215 | /* | |
3216 | * This function validates the new parameters of a -deadline task. | |
3217 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3218 | * than the runtime, as well as the period of being zero or |
332ac17e DF |
3219 | * greater than deadline. Furthermore, we have to be sure that |
3220 | * user parameters are above the internal resolution (1us); we | |
3221 | * check sched_runtime only since it is always the smaller one. | |
aab03e05 DF |
3222 | */ |
3223 | static bool | |
3224 | __checkparam_dl(const struct sched_attr *attr) | |
3225 | { | |
3226 | return attr && attr->sched_deadline != 0 && | |
755378a4 HG |
3227 | (attr->sched_period == 0 || |
3228 | (s64)(attr->sched_period - attr->sched_deadline) >= 0) && | |
332ac17e DF |
3229 | (s64)(attr->sched_deadline - attr->sched_runtime ) >= 0 && |
3230 | attr->sched_runtime >= (2 << (DL_SCALE - 1)); | |
aab03e05 DF |
3231 | } |
3232 | ||
c69e8d9c DH |
3233 | /* |
3234 | * check the target process has a UID that matches the current process's | |
3235 | */ | |
3236 | static bool check_same_owner(struct task_struct *p) | |
3237 | { | |
3238 | const struct cred *cred = current_cred(), *pcred; | |
3239 | bool match; | |
3240 | ||
3241 | rcu_read_lock(); | |
3242 | pcred = __task_cred(p); | |
9c806aa0 EB |
3243 | match = (uid_eq(cred->euid, pcred->euid) || |
3244 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3245 | rcu_read_unlock(); |
3246 | return match; | |
3247 | } | |
3248 | ||
d50dde5a DF |
3249 | static int __sched_setscheduler(struct task_struct *p, |
3250 | const struct sched_attr *attr, | |
3251 | bool user) | |
1da177e4 | 3252 | { |
83b699ed | 3253 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3254 | int policy = attr->sched_policy; |
1da177e4 | 3255 | unsigned long flags; |
83ab0aa0 | 3256 | const struct sched_class *prev_class; |
70b97a7f | 3257 | struct rq *rq; |
ca94c442 | 3258 | int reset_on_fork; |
1da177e4 | 3259 | |
66e5393a SR |
3260 | /* may grab non-irq protected spin_locks */ |
3261 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3262 | recheck: |
3263 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3264 | if (policy < 0) { |
3265 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3266 | policy = oldpolicy = p->policy; |
ca94c442 LP |
3267 | } else { |
3268 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
3269 | policy &= ~SCHED_RESET_ON_FORK; | |
3270 | ||
aab03e05 DF |
3271 | if (policy != SCHED_DEADLINE && |
3272 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3273 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3274 | policy != SCHED_IDLE) | |
3275 | return -EINVAL; | |
3276 | } | |
3277 | ||
1da177e4 LT |
3278 | /* |
3279 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3280 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3281 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3282 | */ |
d50dde5a DF |
3283 | if (attr->sched_priority < 0 || |
3284 | (p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || | |
3285 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) | |
1da177e4 | 3286 | return -EINVAL; |
aab03e05 DF |
3287 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3288 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3289 | return -EINVAL; |
3290 | ||
37e4ab3f OC |
3291 | /* |
3292 | * Allow unprivileged RT tasks to decrease priority: | |
3293 | */ | |
961ccddd | 3294 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a DF |
3295 | if (fair_policy(policy)) { |
3296 | if (!can_nice(p, attr->sched_nice)) | |
3297 | return -EPERM; | |
3298 | } | |
3299 | ||
e05606d3 | 3300 | if (rt_policy(policy)) { |
a44702e8 ON |
3301 | unsigned long rlim_rtprio = |
3302 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3303 | |
3304 | /* can't set/change the rt policy */ | |
3305 | if (policy != p->policy && !rlim_rtprio) | |
3306 | return -EPERM; | |
3307 | ||
3308 | /* can't increase priority */ | |
d50dde5a DF |
3309 | if (attr->sched_priority > p->rt_priority && |
3310 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3311 | return -EPERM; |
3312 | } | |
c02aa73b | 3313 | |
dd41f596 | 3314 | /* |
c02aa73b DH |
3315 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3316 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3317 | */ |
c02aa73b DH |
3318 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
3319 | if (!can_nice(p, TASK_NICE(p))) | |
3320 | return -EPERM; | |
3321 | } | |
5fe1d75f | 3322 | |
37e4ab3f | 3323 | /* can't change other user's priorities */ |
c69e8d9c | 3324 | if (!check_same_owner(p)) |
37e4ab3f | 3325 | return -EPERM; |
ca94c442 LP |
3326 | |
3327 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3328 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3329 | return -EPERM; | |
37e4ab3f | 3330 | } |
1da177e4 | 3331 | |
725aad24 | 3332 | if (user) { |
b0ae1981 | 3333 | retval = security_task_setscheduler(p); |
725aad24 JF |
3334 | if (retval) |
3335 | return retval; | |
3336 | } | |
3337 | ||
b29739f9 IM |
3338 | /* |
3339 | * make sure no PI-waiters arrive (or leave) while we are | |
3340 | * changing the priority of the task: | |
0122ec5b | 3341 | * |
25985edc | 3342 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3343 | * runqueue lock must be held. |
3344 | */ | |
0122ec5b | 3345 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3346 | |
34f971f6 PZ |
3347 | /* |
3348 | * Changing the policy of the stop threads its a very bad idea | |
3349 | */ | |
3350 | if (p == rq->stop) { | |
0122ec5b | 3351 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3352 | return -EINVAL; |
3353 | } | |
3354 | ||
a51e9198 DF |
3355 | /* |
3356 | * If not changing anything there's no need to proceed further: | |
3357 | */ | |
d50dde5a DF |
3358 | if (unlikely(policy == p->policy)) { |
3359 | if (fair_policy(policy) && attr->sched_nice != TASK_NICE(p)) | |
3360 | goto change; | |
3361 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3362 | goto change; | |
aab03e05 DF |
3363 | if (dl_policy(policy)) |
3364 | goto change; | |
d50dde5a | 3365 | |
45afb173 | 3366 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3367 | return 0; |
3368 | } | |
d50dde5a | 3369 | change: |
a51e9198 | 3370 | |
dc61b1d6 | 3371 | if (user) { |
332ac17e | 3372 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3373 | /* |
3374 | * Do not allow realtime tasks into groups that have no runtime | |
3375 | * assigned. | |
3376 | */ | |
3377 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3378 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3379 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3380 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3381 | return -EPERM; |
3382 | } | |
dc61b1d6 | 3383 | #endif |
332ac17e DF |
3384 | #ifdef CONFIG_SMP |
3385 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3386 | cpumask_t *span = rq->rd->span; | |
3387 | cpumask_t act_affinity; | |
3388 | ||
3389 | /* | |
3390 | * cpus_allowed mask is statically initialized with | |
3391 | * CPU_MASK_ALL, span is instead dynamic. Here we | |
3392 | * compute the "dynamic" affinity of a task. | |
3393 | */ | |
3394 | cpumask_and(&act_affinity, &p->cpus_allowed, | |
3395 | cpu_active_mask); | |
3396 | ||
3397 | /* | |
3398 | * Don't allow tasks with an affinity mask smaller than | |
3399 | * the entire root_domain to become SCHED_DEADLINE. We | |
3400 | * will also fail if there's no bandwidth available. | |
3401 | */ | |
3402 | if (!cpumask_equal(&act_affinity, span) || | |
3403 | rq->rd->dl_bw.bw == 0) { | |
3404 | task_rq_unlock(rq, p, &flags); | |
3405 | return -EPERM; | |
3406 | } | |
3407 | } | |
3408 | #endif | |
3409 | } | |
dc61b1d6 | 3410 | |
1da177e4 LT |
3411 | /* recheck policy now with rq lock held */ |
3412 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3413 | policy = oldpolicy = -1; | |
0122ec5b | 3414 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3415 | goto recheck; |
3416 | } | |
332ac17e DF |
3417 | |
3418 | /* | |
3419 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3420 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3421 | * is available. | |
3422 | */ | |
3423 | if ((dl_policy(policy) || dl_task(p)) && | |
3424 | dl_overflow(p, policy, attr)) { | |
3425 | task_rq_unlock(rq, p, &flags); | |
3426 | return -EBUSY; | |
3427 | } | |
3428 | ||
fd2f4419 | 3429 | on_rq = p->on_rq; |
051a1d1a | 3430 | running = task_current(rq, p); |
0e1f3483 | 3431 | if (on_rq) |
4ca9b72b | 3432 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3433 | if (running) |
3434 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3435 | |
ca94c442 LP |
3436 | p->sched_reset_on_fork = reset_on_fork; |
3437 | ||
1da177e4 | 3438 | oldprio = p->prio; |
83ab0aa0 | 3439 | prev_class = p->sched_class; |
d50dde5a | 3440 | __setscheduler(rq, p, attr); |
f6b53205 | 3441 | |
0e1f3483 HS |
3442 | if (running) |
3443 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3444 | if (on_rq) |
4ca9b72b | 3445 | enqueue_task(rq, p, 0); |
cb469845 | 3446 | |
da7a735e | 3447 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3448 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3449 | |
95e02ca9 TG |
3450 | rt_mutex_adjust_pi(p); |
3451 | ||
1da177e4 LT |
3452 | return 0; |
3453 | } | |
961ccddd RR |
3454 | |
3455 | /** | |
3456 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3457 | * @p: the task in question. | |
3458 | * @policy: new policy. | |
3459 | * @param: structure containing the new RT priority. | |
3460 | * | |
e69f6186 YB |
3461 | * Return: 0 on success. An error code otherwise. |
3462 | * | |
961ccddd RR |
3463 | * NOTE that the task may be already dead. |
3464 | */ | |
3465 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3466 | const struct sched_param *param) |
961ccddd | 3467 | { |
d50dde5a DF |
3468 | struct sched_attr attr = { |
3469 | .sched_policy = policy, | |
3470 | .sched_priority = param->sched_priority | |
3471 | }; | |
3472 | return __sched_setscheduler(p, &attr, true); | |
961ccddd | 3473 | } |
1da177e4 LT |
3474 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3475 | ||
d50dde5a DF |
3476 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3477 | { | |
3478 | return __sched_setscheduler(p, attr, true); | |
3479 | } | |
3480 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3481 | ||
961ccddd RR |
3482 | /** |
3483 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3484 | * @p: the task in question. | |
3485 | * @policy: new policy. | |
3486 | * @param: structure containing the new RT priority. | |
3487 | * | |
3488 | * Just like sched_setscheduler, only don't bother checking if the | |
3489 | * current context has permission. For example, this is needed in | |
3490 | * stop_machine(): we create temporary high priority worker threads, | |
3491 | * but our caller might not have that capability. | |
e69f6186 YB |
3492 | * |
3493 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3494 | */ |
3495 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3496 | const struct sched_param *param) |
961ccddd | 3497 | { |
d50dde5a DF |
3498 | struct sched_attr attr = { |
3499 | .sched_policy = policy, | |
3500 | .sched_priority = param->sched_priority | |
3501 | }; | |
3502 | return __sched_setscheduler(p, &attr, false); | |
961ccddd RR |
3503 | } |
3504 | ||
95cdf3b7 IM |
3505 | static int |
3506 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3507 | { |
1da177e4 LT |
3508 | struct sched_param lparam; |
3509 | struct task_struct *p; | |
36c8b586 | 3510 | int retval; |
1da177e4 LT |
3511 | |
3512 | if (!param || pid < 0) | |
3513 | return -EINVAL; | |
3514 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3515 | return -EFAULT; | |
5fe1d75f ON |
3516 | |
3517 | rcu_read_lock(); | |
3518 | retval = -ESRCH; | |
1da177e4 | 3519 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3520 | if (p != NULL) |
3521 | retval = sched_setscheduler(p, policy, &lparam); | |
3522 | rcu_read_unlock(); | |
36c8b586 | 3523 | |
1da177e4 LT |
3524 | return retval; |
3525 | } | |
3526 | ||
d50dde5a DF |
3527 | /* |
3528 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3529 | */ | |
3530 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3531 | struct sched_attr *attr) | |
3532 | { | |
3533 | u32 size; | |
3534 | int ret; | |
3535 | ||
3536 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3537 | return -EFAULT; | |
3538 | ||
3539 | /* | |
3540 | * zero the full structure, so that a short copy will be nice. | |
3541 | */ | |
3542 | memset(attr, 0, sizeof(*attr)); | |
3543 | ||
3544 | ret = get_user(size, &uattr->size); | |
3545 | if (ret) | |
3546 | return ret; | |
3547 | ||
3548 | if (size > PAGE_SIZE) /* silly large */ | |
3549 | goto err_size; | |
3550 | ||
3551 | if (!size) /* abi compat */ | |
3552 | size = SCHED_ATTR_SIZE_VER0; | |
3553 | ||
3554 | if (size < SCHED_ATTR_SIZE_VER0) | |
3555 | goto err_size; | |
3556 | ||
3557 | /* | |
3558 | * If we're handed a bigger struct than we know of, | |
3559 | * ensure all the unknown bits are 0 - i.e. new | |
3560 | * user-space does not rely on any kernel feature | |
3561 | * extensions we dont know about yet. | |
3562 | */ | |
3563 | if (size > sizeof(*attr)) { | |
3564 | unsigned char __user *addr; | |
3565 | unsigned char __user *end; | |
3566 | unsigned char val; | |
3567 | ||
3568 | addr = (void __user *)uattr + sizeof(*attr); | |
3569 | end = (void __user *)uattr + size; | |
3570 | ||
3571 | for (; addr < end; addr++) { | |
3572 | ret = get_user(val, addr); | |
3573 | if (ret) | |
3574 | return ret; | |
3575 | if (val) | |
3576 | goto err_size; | |
3577 | } | |
3578 | size = sizeof(*attr); | |
3579 | } | |
3580 | ||
3581 | ret = copy_from_user(attr, uattr, size); | |
3582 | if (ret) | |
3583 | return -EFAULT; | |
3584 | ||
3585 | /* | |
3586 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3587 | * to be strict and return an error on out-of-bounds values? | |
3588 | */ | |
3589 | attr->sched_nice = clamp(attr->sched_nice, -20, 19); | |
3590 | ||
3591 | out: | |
3592 | return ret; | |
3593 | ||
3594 | err_size: | |
3595 | put_user(sizeof(*attr), &uattr->size); | |
3596 | ret = -E2BIG; | |
3597 | goto out; | |
3598 | } | |
3599 | ||
1da177e4 LT |
3600 | /** |
3601 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3602 | * @pid: the pid in question. | |
3603 | * @policy: new policy. | |
3604 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3605 | * |
3606 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3607 | */ |
5add95d4 HC |
3608 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3609 | struct sched_param __user *, param) | |
1da177e4 | 3610 | { |
c21761f1 JB |
3611 | /* negative values for policy are not valid */ |
3612 | if (policy < 0) | |
3613 | return -EINVAL; | |
3614 | ||
1da177e4 LT |
3615 | return do_sched_setscheduler(pid, policy, param); |
3616 | } | |
3617 | ||
3618 | /** | |
3619 | * sys_sched_setparam - set/change the RT priority of a thread | |
3620 | * @pid: the pid in question. | |
3621 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3622 | * |
3623 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3624 | */ |
5add95d4 | 3625 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3626 | { |
3627 | return do_sched_setscheduler(pid, -1, param); | |
3628 | } | |
3629 | ||
d50dde5a DF |
3630 | /** |
3631 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3632 | * @pid: the pid in question. | |
3633 | * @attr: structure containing the extended parameters. | |
3634 | */ | |
3635 | SYSCALL_DEFINE2(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr) | |
3636 | { | |
3637 | struct sched_attr attr; | |
3638 | struct task_struct *p; | |
3639 | int retval; | |
3640 | ||
3641 | if (!uattr || pid < 0) | |
3642 | return -EINVAL; | |
3643 | ||
3644 | if (sched_copy_attr(uattr, &attr)) | |
3645 | return -EFAULT; | |
3646 | ||
3647 | rcu_read_lock(); | |
3648 | retval = -ESRCH; | |
3649 | p = find_process_by_pid(pid); | |
3650 | if (p != NULL) | |
3651 | retval = sched_setattr(p, &attr); | |
3652 | rcu_read_unlock(); | |
3653 | ||
3654 | return retval; | |
3655 | } | |
3656 | ||
1da177e4 LT |
3657 | /** |
3658 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3659 | * @pid: the pid in question. | |
e69f6186 YB |
3660 | * |
3661 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3662 | * code. | |
1da177e4 | 3663 | */ |
5add95d4 | 3664 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3665 | { |
36c8b586 | 3666 | struct task_struct *p; |
3a5c359a | 3667 | int retval; |
1da177e4 LT |
3668 | |
3669 | if (pid < 0) | |
3a5c359a | 3670 | return -EINVAL; |
1da177e4 LT |
3671 | |
3672 | retval = -ESRCH; | |
5fe85be0 | 3673 | rcu_read_lock(); |
1da177e4 LT |
3674 | p = find_process_by_pid(pid); |
3675 | if (p) { | |
3676 | retval = security_task_getscheduler(p); | |
3677 | if (!retval) | |
ca94c442 LP |
3678 | retval = p->policy |
3679 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3680 | } |
5fe85be0 | 3681 | rcu_read_unlock(); |
1da177e4 LT |
3682 | return retval; |
3683 | } | |
3684 | ||
3685 | /** | |
ca94c442 | 3686 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3687 | * @pid: the pid in question. |
3688 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3689 | * |
3690 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3691 | * code. | |
1da177e4 | 3692 | */ |
5add95d4 | 3693 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3694 | { |
3695 | struct sched_param lp; | |
36c8b586 | 3696 | struct task_struct *p; |
3a5c359a | 3697 | int retval; |
1da177e4 LT |
3698 | |
3699 | if (!param || pid < 0) | |
3a5c359a | 3700 | return -EINVAL; |
1da177e4 | 3701 | |
5fe85be0 | 3702 | rcu_read_lock(); |
1da177e4 LT |
3703 | p = find_process_by_pid(pid); |
3704 | retval = -ESRCH; | |
3705 | if (!p) | |
3706 | goto out_unlock; | |
3707 | ||
3708 | retval = security_task_getscheduler(p); | |
3709 | if (retval) | |
3710 | goto out_unlock; | |
3711 | ||
aab03e05 DF |
3712 | if (task_has_dl_policy(p)) { |
3713 | retval = -EINVAL; | |
3714 | goto out_unlock; | |
3715 | } | |
1da177e4 | 3716 | lp.sched_priority = p->rt_priority; |
5fe85be0 | 3717 | rcu_read_unlock(); |
1da177e4 LT |
3718 | |
3719 | /* | |
3720 | * This one might sleep, we cannot do it with a spinlock held ... | |
3721 | */ | |
3722 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3723 | ||
1da177e4 LT |
3724 | return retval; |
3725 | ||
3726 | out_unlock: | |
5fe85be0 | 3727 | rcu_read_unlock(); |
1da177e4 LT |
3728 | return retval; |
3729 | } | |
3730 | ||
d50dde5a DF |
3731 | static int sched_read_attr(struct sched_attr __user *uattr, |
3732 | struct sched_attr *attr, | |
3733 | unsigned int usize) | |
3734 | { | |
3735 | int ret; | |
3736 | ||
3737 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3738 | return -EFAULT; | |
3739 | ||
3740 | /* | |
3741 | * If we're handed a smaller struct than we know of, | |
3742 | * ensure all the unknown bits are 0 - i.e. old | |
3743 | * user-space does not get uncomplete information. | |
3744 | */ | |
3745 | if (usize < sizeof(*attr)) { | |
3746 | unsigned char *addr; | |
3747 | unsigned char *end; | |
3748 | ||
3749 | addr = (void *)attr + usize; | |
3750 | end = (void *)attr + sizeof(*attr); | |
3751 | ||
3752 | for (; addr < end; addr++) { | |
3753 | if (*addr) | |
3754 | goto err_size; | |
3755 | } | |
3756 | ||
3757 | attr->size = usize; | |
3758 | } | |
3759 | ||
3760 | ret = copy_to_user(uattr, attr, usize); | |
3761 | if (ret) | |
3762 | return -EFAULT; | |
3763 | ||
3764 | out: | |
3765 | return ret; | |
3766 | ||
3767 | err_size: | |
3768 | ret = -E2BIG; | |
3769 | goto out; | |
3770 | } | |
3771 | ||
3772 | /** | |
aab03e05 | 3773 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a DF |
3774 | * @pid: the pid in question. |
3775 | * @attr: structure containing the extended parameters. | |
3776 | * @size: sizeof(attr) for fwd/bwd comp. | |
3777 | */ | |
3778 | SYSCALL_DEFINE3(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, | |
3779 | unsigned int, size) | |
3780 | { | |
3781 | struct sched_attr attr = { | |
3782 | .size = sizeof(struct sched_attr), | |
3783 | }; | |
3784 | struct task_struct *p; | |
3785 | int retval; | |
3786 | ||
3787 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
3788 | size < SCHED_ATTR_SIZE_VER0) | |
3789 | return -EINVAL; | |
3790 | ||
3791 | rcu_read_lock(); | |
3792 | p = find_process_by_pid(pid); | |
3793 | retval = -ESRCH; | |
3794 | if (!p) | |
3795 | goto out_unlock; | |
3796 | ||
3797 | retval = security_task_getscheduler(p); | |
3798 | if (retval) | |
3799 | goto out_unlock; | |
3800 | ||
3801 | attr.sched_policy = p->policy; | |
aab03e05 DF |
3802 | if (task_has_dl_policy(p)) |
3803 | __getparam_dl(p, &attr); | |
3804 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3805 | attr.sched_priority = p->rt_priority; |
3806 | else | |
3807 | attr.sched_nice = TASK_NICE(p); | |
3808 | ||
3809 | rcu_read_unlock(); | |
3810 | ||
3811 | retval = sched_read_attr(uattr, &attr, size); | |
3812 | return retval; | |
3813 | ||
3814 | out_unlock: | |
3815 | rcu_read_unlock(); | |
3816 | return retval; | |
3817 | } | |
3818 | ||
96f874e2 | 3819 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3820 | { |
5a16f3d3 | 3821 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3822 | struct task_struct *p; |
3823 | int retval; | |
1da177e4 | 3824 | |
23f5d142 | 3825 | rcu_read_lock(); |
1da177e4 LT |
3826 | |
3827 | p = find_process_by_pid(pid); | |
3828 | if (!p) { | |
23f5d142 | 3829 | rcu_read_unlock(); |
1da177e4 LT |
3830 | return -ESRCH; |
3831 | } | |
3832 | ||
23f5d142 | 3833 | /* Prevent p going away */ |
1da177e4 | 3834 | get_task_struct(p); |
23f5d142 | 3835 | rcu_read_unlock(); |
1da177e4 | 3836 | |
14a40ffc TH |
3837 | if (p->flags & PF_NO_SETAFFINITY) { |
3838 | retval = -EINVAL; | |
3839 | goto out_put_task; | |
3840 | } | |
5a16f3d3 RR |
3841 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3842 | retval = -ENOMEM; | |
3843 | goto out_put_task; | |
3844 | } | |
3845 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3846 | retval = -ENOMEM; | |
3847 | goto out_free_cpus_allowed; | |
3848 | } | |
1da177e4 | 3849 | retval = -EPERM; |
4c44aaaf EB |
3850 | if (!check_same_owner(p)) { |
3851 | rcu_read_lock(); | |
3852 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3853 | rcu_read_unlock(); | |
3854 | goto out_unlock; | |
3855 | } | |
3856 | rcu_read_unlock(); | |
3857 | } | |
1da177e4 | 3858 | |
b0ae1981 | 3859 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3860 | if (retval) |
3861 | goto out_unlock; | |
3862 | ||
332ac17e DF |
3863 | /* |
3864 | * Since bandwidth control happens on root_domain basis, | |
3865 | * if admission test is enabled, we only admit -deadline | |
3866 | * tasks allowed to run on all the CPUs in the task's | |
3867 | * root_domain. | |
3868 | */ | |
3869 | #ifdef CONFIG_SMP | |
3870 | if (task_has_dl_policy(p)) { | |
3871 | const struct cpumask *span = task_rq(p)->rd->span; | |
3872 | ||
3873 | if (dl_bandwidth_enabled() && | |
3874 | !cpumask_equal(in_mask, span)) { | |
3875 | retval = -EBUSY; | |
3876 | goto out_unlock; | |
3877 | } | |
3878 | } | |
3879 | #endif | |
3880 | ||
5a16f3d3 RR |
3881 | cpuset_cpus_allowed(p, cpus_allowed); |
3882 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 3883 | again: |
5a16f3d3 | 3884 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3885 | |
8707d8b8 | 3886 | if (!retval) { |
5a16f3d3 RR |
3887 | cpuset_cpus_allowed(p, cpus_allowed); |
3888 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3889 | /* |
3890 | * We must have raced with a concurrent cpuset | |
3891 | * update. Just reset the cpus_allowed to the | |
3892 | * cpuset's cpus_allowed | |
3893 | */ | |
5a16f3d3 | 3894 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3895 | goto again; |
3896 | } | |
3897 | } | |
1da177e4 | 3898 | out_unlock: |
5a16f3d3 RR |
3899 | free_cpumask_var(new_mask); |
3900 | out_free_cpus_allowed: | |
3901 | free_cpumask_var(cpus_allowed); | |
3902 | out_put_task: | |
1da177e4 | 3903 | put_task_struct(p); |
1da177e4 LT |
3904 | return retval; |
3905 | } | |
3906 | ||
3907 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3908 | struct cpumask *new_mask) |
1da177e4 | 3909 | { |
96f874e2 RR |
3910 | if (len < cpumask_size()) |
3911 | cpumask_clear(new_mask); | |
3912 | else if (len > cpumask_size()) | |
3913 | len = cpumask_size(); | |
3914 | ||
1da177e4 LT |
3915 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3916 | } | |
3917 | ||
3918 | /** | |
3919 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3920 | * @pid: pid of the process | |
3921 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3922 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3923 | * |
3924 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3925 | */ |
5add95d4 HC |
3926 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3927 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3928 | { |
5a16f3d3 | 3929 | cpumask_var_t new_mask; |
1da177e4 LT |
3930 | int retval; |
3931 | ||
5a16f3d3 RR |
3932 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3933 | return -ENOMEM; | |
1da177e4 | 3934 | |
5a16f3d3 RR |
3935 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3936 | if (retval == 0) | |
3937 | retval = sched_setaffinity(pid, new_mask); | |
3938 | free_cpumask_var(new_mask); | |
3939 | return retval; | |
1da177e4 LT |
3940 | } |
3941 | ||
96f874e2 | 3942 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 3943 | { |
36c8b586 | 3944 | struct task_struct *p; |
31605683 | 3945 | unsigned long flags; |
1da177e4 | 3946 | int retval; |
1da177e4 | 3947 | |
23f5d142 | 3948 | rcu_read_lock(); |
1da177e4 LT |
3949 | |
3950 | retval = -ESRCH; | |
3951 | p = find_process_by_pid(pid); | |
3952 | if (!p) | |
3953 | goto out_unlock; | |
3954 | ||
e7834f8f DQ |
3955 | retval = security_task_getscheduler(p); |
3956 | if (retval) | |
3957 | goto out_unlock; | |
3958 | ||
013fdb80 | 3959 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 3960 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 3961 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
3962 | |
3963 | out_unlock: | |
23f5d142 | 3964 | rcu_read_unlock(); |
1da177e4 | 3965 | |
9531b62f | 3966 | return retval; |
1da177e4 LT |
3967 | } |
3968 | ||
3969 | /** | |
3970 | * sys_sched_getaffinity - get the cpu affinity of a process | |
3971 | * @pid: pid of the process | |
3972 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3973 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
3974 | * |
3975 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3976 | */ |
5add95d4 HC |
3977 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
3978 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
3979 | { |
3980 | int ret; | |
f17c8607 | 3981 | cpumask_var_t mask; |
1da177e4 | 3982 | |
84fba5ec | 3983 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
3984 | return -EINVAL; |
3985 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
3986 | return -EINVAL; |
3987 | ||
f17c8607 RR |
3988 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
3989 | return -ENOMEM; | |
1da177e4 | 3990 | |
f17c8607 RR |
3991 | ret = sched_getaffinity(pid, mask); |
3992 | if (ret == 0) { | |
8bc037fb | 3993 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
3994 | |
3995 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
3996 | ret = -EFAULT; |
3997 | else | |
cd3d8031 | 3998 | ret = retlen; |
f17c8607 RR |
3999 | } |
4000 | free_cpumask_var(mask); | |
1da177e4 | 4001 | |
f17c8607 | 4002 | return ret; |
1da177e4 LT |
4003 | } |
4004 | ||
4005 | /** | |
4006 | * sys_sched_yield - yield the current processor to other threads. | |
4007 | * | |
dd41f596 IM |
4008 | * This function yields the current CPU to other tasks. If there are no |
4009 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4010 | * |
4011 | * Return: 0. | |
1da177e4 | 4012 | */ |
5add95d4 | 4013 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4014 | { |
70b97a7f | 4015 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4016 | |
2d72376b | 4017 | schedstat_inc(rq, yld_count); |
4530d7ab | 4018 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4019 | |
4020 | /* | |
4021 | * Since we are going to call schedule() anyway, there's | |
4022 | * no need to preempt or enable interrupts: | |
4023 | */ | |
4024 | __release(rq->lock); | |
8a25d5de | 4025 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4026 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4027 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4028 | |
4029 | schedule(); | |
4030 | ||
4031 | return 0; | |
4032 | } | |
4033 | ||
e7b38404 | 4034 | static void __cond_resched(void) |
1da177e4 | 4035 | { |
bdb43806 | 4036 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4037 | __schedule(); |
bdb43806 | 4038 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4039 | } |
4040 | ||
02b67cc3 | 4041 | int __sched _cond_resched(void) |
1da177e4 | 4042 | { |
d86ee480 | 4043 | if (should_resched()) { |
1da177e4 LT |
4044 | __cond_resched(); |
4045 | return 1; | |
4046 | } | |
4047 | return 0; | |
4048 | } | |
02b67cc3 | 4049 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4050 | |
4051 | /* | |
613afbf8 | 4052 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4053 | * call schedule, and on return reacquire the lock. |
4054 | * | |
41a2d6cf | 4055 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4056 | * operations here to prevent schedule() from being called twice (once via |
4057 | * spin_unlock(), once by hand). | |
4058 | */ | |
613afbf8 | 4059 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4060 | { |
d86ee480 | 4061 | int resched = should_resched(); |
6df3cecb JK |
4062 | int ret = 0; |
4063 | ||
f607c668 PZ |
4064 | lockdep_assert_held(lock); |
4065 | ||
95c354fe | 4066 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4067 | spin_unlock(lock); |
d86ee480 | 4068 | if (resched) |
95c354fe NP |
4069 | __cond_resched(); |
4070 | else | |
4071 | cpu_relax(); | |
6df3cecb | 4072 | ret = 1; |
1da177e4 | 4073 | spin_lock(lock); |
1da177e4 | 4074 | } |
6df3cecb | 4075 | return ret; |
1da177e4 | 4076 | } |
613afbf8 | 4077 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4078 | |
613afbf8 | 4079 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4080 | { |
4081 | BUG_ON(!in_softirq()); | |
4082 | ||
d86ee480 | 4083 | if (should_resched()) { |
98d82567 | 4084 | local_bh_enable(); |
1da177e4 LT |
4085 | __cond_resched(); |
4086 | local_bh_disable(); | |
4087 | return 1; | |
4088 | } | |
4089 | return 0; | |
4090 | } | |
613afbf8 | 4091 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4092 | |
1da177e4 LT |
4093 | /** |
4094 | * yield - yield the current processor to other threads. | |
4095 | * | |
8e3fabfd PZ |
4096 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4097 | * | |
4098 | * The scheduler is at all times free to pick the calling task as the most | |
4099 | * eligible task to run, if removing the yield() call from your code breaks | |
4100 | * it, its already broken. | |
4101 | * | |
4102 | * Typical broken usage is: | |
4103 | * | |
4104 | * while (!event) | |
4105 | * yield(); | |
4106 | * | |
4107 | * where one assumes that yield() will let 'the other' process run that will | |
4108 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4109 | * happen. Never use yield() as a progress guarantee!! | |
4110 | * | |
4111 | * If you want to use yield() to wait for something, use wait_event(). | |
4112 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4113 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4114 | */ |
4115 | void __sched yield(void) | |
4116 | { | |
4117 | set_current_state(TASK_RUNNING); | |
4118 | sys_sched_yield(); | |
4119 | } | |
1da177e4 LT |
4120 | EXPORT_SYMBOL(yield); |
4121 | ||
d95f4122 MG |
4122 | /** |
4123 | * yield_to - yield the current processor to another thread in | |
4124 | * your thread group, or accelerate that thread toward the | |
4125 | * processor it's on. | |
16addf95 RD |
4126 | * @p: target task |
4127 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4128 | * |
4129 | * It's the caller's job to ensure that the target task struct | |
4130 | * can't go away on us before we can do any checks. | |
4131 | * | |
e69f6186 | 4132 | * Return: |
7b270f60 PZ |
4133 | * true (>0) if we indeed boosted the target task. |
4134 | * false (0) if we failed to boost the target. | |
4135 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4136 | */ |
4137 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4138 | { | |
4139 | struct task_struct *curr = current; | |
4140 | struct rq *rq, *p_rq; | |
4141 | unsigned long flags; | |
c3c18640 | 4142 | int yielded = 0; |
d95f4122 MG |
4143 | |
4144 | local_irq_save(flags); | |
4145 | rq = this_rq(); | |
4146 | ||
4147 | again: | |
4148 | p_rq = task_rq(p); | |
7b270f60 PZ |
4149 | /* |
4150 | * If we're the only runnable task on the rq and target rq also | |
4151 | * has only one task, there's absolutely no point in yielding. | |
4152 | */ | |
4153 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4154 | yielded = -ESRCH; | |
4155 | goto out_irq; | |
4156 | } | |
4157 | ||
d95f4122 | 4158 | double_rq_lock(rq, p_rq); |
39e24d8f | 4159 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4160 | double_rq_unlock(rq, p_rq); |
4161 | goto again; | |
4162 | } | |
4163 | ||
4164 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4165 | goto out_unlock; |
d95f4122 MG |
4166 | |
4167 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4168 | goto out_unlock; |
d95f4122 MG |
4169 | |
4170 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4171 | goto out_unlock; |
d95f4122 MG |
4172 | |
4173 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4174 | if (yielded) { |
d95f4122 | 4175 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4176 | /* |
4177 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4178 | * fairness. | |
4179 | */ | |
4180 | if (preempt && rq != p_rq) | |
4181 | resched_task(p_rq->curr); | |
4182 | } | |
d95f4122 | 4183 | |
7b270f60 | 4184 | out_unlock: |
d95f4122 | 4185 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4186 | out_irq: |
d95f4122 MG |
4187 | local_irq_restore(flags); |
4188 | ||
7b270f60 | 4189 | if (yielded > 0) |
d95f4122 MG |
4190 | schedule(); |
4191 | ||
4192 | return yielded; | |
4193 | } | |
4194 | EXPORT_SYMBOL_GPL(yield_to); | |
4195 | ||
1da177e4 | 4196 | /* |
41a2d6cf | 4197 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4198 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4199 | */ |
4200 | void __sched io_schedule(void) | |
4201 | { | |
54d35f29 | 4202 | struct rq *rq = raw_rq(); |
1da177e4 | 4203 | |
0ff92245 | 4204 | delayacct_blkio_start(); |
1da177e4 | 4205 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4206 | blk_flush_plug(current); |
8f0dfc34 | 4207 | current->in_iowait = 1; |
1da177e4 | 4208 | schedule(); |
8f0dfc34 | 4209 | current->in_iowait = 0; |
1da177e4 | 4210 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4211 | delayacct_blkio_end(); |
1da177e4 | 4212 | } |
1da177e4 LT |
4213 | EXPORT_SYMBOL(io_schedule); |
4214 | ||
4215 | long __sched io_schedule_timeout(long timeout) | |
4216 | { | |
54d35f29 | 4217 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4218 | long ret; |
4219 | ||
0ff92245 | 4220 | delayacct_blkio_start(); |
1da177e4 | 4221 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4222 | blk_flush_plug(current); |
8f0dfc34 | 4223 | current->in_iowait = 1; |
1da177e4 | 4224 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4225 | current->in_iowait = 0; |
1da177e4 | 4226 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4227 | delayacct_blkio_end(); |
1da177e4 LT |
4228 | return ret; |
4229 | } | |
4230 | ||
4231 | /** | |
4232 | * sys_sched_get_priority_max - return maximum RT priority. | |
4233 | * @policy: scheduling class. | |
4234 | * | |
e69f6186 YB |
4235 | * Return: On success, this syscall returns the maximum |
4236 | * rt_priority that can be used by a given scheduling class. | |
4237 | * On failure, a negative error code is returned. | |
1da177e4 | 4238 | */ |
5add95d4 | 4239 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4240 | { |
4241 | int ret = -EINVAL; | |
4242 | ||
4243 | switch (policy) { | |
4244 | case SCHED_FIFO: | |
4245 | case SCHED_RR: | |
4246 | ret = MAX_USER_RT_PRIO-1; | |
4247 | break; | |
aab03e05 | 4248 | case SCHED_DEADLINE: |
1da177e4 | 4249 | case SCHED_NORMAL: |
b0a9499c | 4250 | case SCHED_BATCH: |
dd41f596 | 4251 | case SCHED_IDLE: |
1da177e4 LT |
4252 | ret = 0; |
4253 | break; | |
4254 | } | |
4255 | return ret; | |
4256 | } | |
4257 | ||
4258 | /** | |
4259 | * sys_sched_get_priority_min - return minimum RT priority. | |
4260 | * @policy: scheduling class. | |
4261 | * | |
e69f6186 YB |
4262 | * Return: On success, this syscall returns the minimum |
4263 | * rt_priority that can be used by a given scheduling class. | |
4264 | * On failure, a negative error code is returned. | |
1da177e4 | 4265 | */ |
5add95d4 | 4266 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4267 | { |
4268 | int ret = -EINVAL; | |
4269 | ||
4270 | switch (policy) { | |
4271 | case SCHED_FIFO: | |
4272 | case SCHED_RR: | |
4273 | ret = 1; | |
4274 | break; | |
aab03e05 | 4275 | case SCHED_DEADLINE: |
1da177e4 | 4276 | case SCHED_NORMAL: |
b0a9499c | 4277 | case SCHED_BATCH: |
dd41f596 | 4278 | case SCHED_IDLE: |
1da177e4 LT |
4279 | ret = 0; |
4280 | } | |
4281 | return ret; | |
4282 | } | |
4283 | ||
4284 | /** | |
4285 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4286 | * @pid: pid of the process. | |
4287 | * @interval: userspace pointer to the timeslice value. | |
4288 | * | |
4289 | * this syscall writes the default timeslice value of a given process | |
4290 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4291 | * |
4292 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4293 | * an error code. | |
1da177e4 | 4294 | */ |
17da2bd9 | 4295 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4296 | struct timespec __user *, interval) |
1da177e4 | 4297 | { |
36c8b586 | 4298 | struct task_struct *p; |
a4ec24b4 | 4299 | unsigned int time_slice; |
dba091b9 TG |
4300 | unsigned long flags; |
4301 | struct rq *rq; | |
3a5c359a | 4302 | int retval; |
1da177e4 | 4303 | struct timespec t; |
1da177e4 LT |
4304 | |
4305 | if (pid < 0) | |
3a5c359a | 4306 | return -EINVAL; |
1da177e4 LT |
4307 | |
4308 | retval = -ESRCH; | |
1a551ae7 | 4309 | rcu_read_lock(); |
1da177e4 LT |
4310 | p = find_process_by_pid(pid); |
4311 | if (!p) | |
4312 | goto out_unlock; | |
4313 | ||
4314 | retval = security_task_getscheduler(p); | |
4315 | if (retval) | |
4316 | goto out_unlock; | |
4317 | ||
dba091b9 TG |
4318 | rq = task_rq_lock(p, &flags); |
4319 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4320 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4321 | |
1a551ae7 | 4322 | rcu_read_unlock(); |
a4ec24b4 | 4323 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4324 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4325 | return retval; |
3a5c359a | 4326 | |
1da177e4 | 4327 | out_unlock: |
1a551ae7 | 4328 | rcu_read_unlock(); |
1da177e4 LT |
4329 | return retval; |
4330 | } | |
4331 | ||
7c731e0a | 4332 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4333 | |
82a1fcb9 | 4334 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4335 | { |
1da177e4 | 4336 | unsigned long free = 0; |
4e79752c | 4337 | int ppid; |
36c8b586 | 4338 | unsigned state; |
1da177e4 | 4339 | |
1da177e4 | 4340 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4341 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4342 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4343 | #if BITS_PER_LONG == 32 |
1da177e4 | 4344 | if (state == TASK_RUNNING) |
3df0fc5b | 4345 | printk(KERN_CONT " running "); |
1da177e4 | 4346 | else |
3df0fc5b | 4347 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4348 | #else |
4349 | if (state == TASK_RUNNING) | |
3df0fc5b | 4350 | printk(KERN_CONT " running task "); |
1da177e4 | 4351 | else |
3df0fc5b | 4352 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4353 | #endif |
4354 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4355 | free = stack_not_used(p); |
1da177e4 | 4356 | #endif |
4e79752c PM |
4357 | rcu_read_lock(); |
4358 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4359 | rcu_read_unlock(); | |
3df0fc5b | 4360 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4361 | task_pid_nr(p), ppid, |
aa47b7e0 | 4362 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4363 | |
3d1cb205 | 4364 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4365 | show_stack(p, NULL); |
1da177e4 LT |
4366 | } |
4367 | ||
e59e2ae2 | 4368 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4369 | { |
36c8b586 | 4370 | struct task_struct *g, *p; |
1da177e4 | 4371 | |
4bd77321 | 4372 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4373 | printk(KERN_INFO |
4374 | " task PC stack pid father\n"); | |
1da177e4 | 4375 | #else |
3df0fc5b PZ |
4376 | printk(KERN_INFO |
4377 | " task PC stack pid father\n"); | |
1da177e4 | 4378 | #endif |
510f5acc | 4379 | rcu_read_lock(); |
1da177e4 LT |
4380 | do_each_thread(g, p) { |
4381 | /* | |
4382 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4383 | * console might take a lot of time: |
1da177e4 LT |
4384 | */ |
4385 | touch_nmi_watchdog(); | |
39bc89fd | 4386 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4387 | sched_show_task(p); |
1da177e4 LT |
4388 | } while_each_thread(g, p); |
4389 | ||
04c9167f JF |
4390 | touch_all_softlockup_watchdogs(); |
4391 | ||
dd41f596 IM |
4392 | #ifdef CONFIG_SCHED_DEBUG |
4393 | sysrq_sched_debug_show(); | |
4394 | #endif | |
510f5acc | 4395 | rcu_read_unlock(); |
e59e2ae2 IM |
4396 | /* |
4397 | * Only show locks if all tasks are dumped: | |
4398 | */ | |
93335a21 | 4399 | if (!state_filter) |
e59e2ae2 | 4400 | debug_show_all_locks(); |
1da177e4 LT |
4401 | } |
4402 | ||
0db0628d | 4403 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4404 | { |
dd41f596 | 4405 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4406 | } |
4407 | ||
f340c0d1 IM |
4408 | /** |
4409 | * init_idle - set up an idle thread for a given CPU | |
4410 | * @idle: task in question | |
4411 | * @cpu: cpu the idle task belongs to | |
4412 | * | |
4413 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4414 | * flag, to make booting more robust. | |
4415 | */ | |
0db0628d | 4416 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4417 | { |
70b97a7f | 4418 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4419 | unsigned long flags; |
4420 | ||
05fa785c | 4421 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4422 | |
5e1576ed | 4423 | __sched_fork(0, idle); |
06b83b5f | 4424 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4425 | idle->se.exec_start = sched_clock(); |
4426 | ||
1e1b6c51 | 4427 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4428 | /* |
4429 | * We're having a chicken and egg problem, even though we are | |
4430 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4431 | * lockdep check in task_group() will fail. | |
4432 | * | |
4433 | * Similar case to sched_fork(). / Alternatively we could | |
4434 | * use task_rq_lock() here and obtain the other rq->lock. | |
4435 | * | |
4436 | * Silence PROVE_RCU | |
4437 | */ | |
4438 | rcu_read_lock(); | |
dd41f596 | 4439 | __set_task_cpu(idle, cpu); |
6506cf6c | 4440 | rcu_read_unlock(); |
1da177e4 | 4441 | |
1da177e4 | 4442 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4443 | #if defined(CONFIG_SMP) |
4444 | idle->on_cpu = 1; | |
4866cde0 | 4445 | #endif |
05fa785c | 4446 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4447 | |
4448 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4449 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4450 | |
dd41f596 IM |
4451 | /* |
4452 | * The idle tasks have their own, simple scheduling class: | |
4453 | */ | |
4454 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4455 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4456 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4457 | #if defined(CONFIG_SMP) |
4458 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4459 | #endif | |
19978ca6 IM |
4460 | } |
4461 | ||
1da177e4 | 4462 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4463 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4464 | { | |
4465 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4466 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4467 | |
4468 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4469 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4470 | } |
4471 | ||
1da177e4 LT |
4472 | /* |
4473 | * This is how migration works: | |
4474 | * | |
969c7921 TH |
4475 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4476 | * stop_one_cpu(). | |
4477 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4478 | * off the CPU) | |
4479 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4480 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4481 | * it and puts it into the right queue. |
969c7921 TH |
4482 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4483 | * is done. | |
1da177e4 LT |
4484 | */ |
4485 | ||
4486 | /* | |
4487 | * Change a given task's CPU affinity. Migrate the thread to a | |
4488 | * proper CPU and schedule it away if the CPU it's executing on | |
4489 | * is removed from the allowed bitmask. | |
4490 | * | |
4491 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4492 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4493 | * call is not atomic; no spinlocks may be held. |
4494 | */ | |
96f874e2 | 4495 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4496 | { |
4497 | unsigned long flags; | |
70b97a7f | 4498 | struct rq *rq; |
969c7921 | 4499 | unsigned int dest_cpu; |
48f24c4d | 4500 | int ret = 0; |
1da177e4 LT |
4501 | |
4502 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4503 | |
db44fc01 YZ |
4504 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4505 | goto out; | |
4506 | ||
6ad4c188 | 4507 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4508 | ret = -EINVAL; |
4509 | goto out; | |
4510 | } | |
4511 | ||
1e1b6c51 | 4512 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4513 | |
1da177e4 | 4514 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4515 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4516 | goto out; |
4517 | ||
969c7921 | 4518 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4519 | if (p->on_rq) { |
969c7921 | 4520 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4521 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4522 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4523 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4524 | tlb_migrate_finish(p->mm); |
4525 | return 0; | |
4526 | } | |
4527 | out: | |
0122ec5b | 4528 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4529 | |
1da177e4 LT |
4530 | return ret; |
4531 | } | |
cd8ba7cd | 4532 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 | 4533 | |
332ac17e DF |
4534 | /* |
4535 | * When dealing with a -deadline task, we have to check if moving it to | |
4536 | * a new CPU is possible or not. In fact, this is only true iff there | |
4537 | * is enough bandwidth available on such CPU, otherwise we want the | |
4538 | * whole migration progedure to fail over. | |
4539 | */ | |
4540 | static inline | |
4541 | bool set_task_cpu_dl(struct task_struct *p, unsigned int cpu) | |
4542 | { | |
4543 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4544 | struct dl_bw *cpu_b = dl_bw_of(cpu); | |
4545 | int ret = 1; | |
4546 | u64 bw; | |
4547 | ||
4548 | if (dl_b == cpu_b) | |
4549 | return 1; | |
4550 | ||
4551 | raw_spin_lock(&dl_b->lock); | |
4552 | raw_spin_lock(&cpu_b->lock); | |
4553 | ||
4554 | bw = cpu_b->bw * cpumask_weight(cpu_rq(cpu)->rd->span); | |
4555 | if (dl_bandwidth_enabled() && | |
4556 | bw < cpu_b->total_bw + p->dl.dl_bw) { | |
4557 | ret = 0; | |
4558 | goto unlock; | |
4559 | } | |
4560 | dl_b->total_bw -= p->dl.dl_bw; | |
4561 | cpu_b->total_bw += p->dl.dl_bw; | |
4562 | ||
4563 | unlock: | |
4564 | raw_spin_unlock(&cpu_b->lock); | |
4565 | raw_spin_unlock(&dl_b->lock); | |
4566 | ||
4567 | return ret; | |
4568 | } | |
4569 | ||
1da177e4 | 4570 | /* |
41a2d6cf | 4571 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4572 | * this because either it can't run here any more (set_cpus_allowed() |
4573 | * away from this CPU, or CPU going down), or because we're | |
4574 | * attempting to rebalance this task on exec (sched_exec). | |
4575 | * | |
4576 | * So we race with normal scheduler movements, but that's OK, as long | |
4577 | * as the task is no longer on this CPU. | |
efc30814 KK |
4578 | * |
4579 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4580 | */ |
efc30814 | 4581 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4582 | { |
70b97a7f | 4583 | struct rq *rq_dest, *rq_src; |
e2912009 | 4584 | int ret = 0; |
1da177e4 | 4585 | |
e761b772 | 4586 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4587 | return ret; |
1da177e4 LT |
4588 | |
4589 | rq_src = cpu_rq(src_cpu); | |
4590 | rq_dest = cpu_rq(dest_cpu); | |
4591 | ||
0122ec5b | 4592 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4593 | double_rq_lock(rq_src, rq_dest); |
4594 | /* Already moved. */ | |
4595 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4596 | goto done; |
1da177e4 | 4597 | /* Affinity changed (again). */ |
fa17b507 | 4598 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4599 | goto fail; |
1da177e4 | 4600 | |
332ac17e DF |
4601 | /* |
4602 | * If p is -deadline, proceed only if there is enough | |
4603 | * bandwidth available on dest_cpu | |
4604 | */ | |
4605 | if (unlikely(dl_task(p)) && !set_task_cpu_dl(p, dest_cpu)) | |
4606 | goto fail; | |
4607 | ||
e2912009 PZ |
4608 | /* |
4609 | * If we're not on a rq, the next wake-up will ensure we're | |
4610 | * placed properly. | |
4611 | */ | |
fd2f4419 | 4612 | if (p->on_rq) { |
4ca9b72b | 4613 | dequeue_task(rq_src, p, 0); |
e2912009 | 4614 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4615 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4616 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4617 | } |
b1e38734 | 4618 | done: |
efc30814 | 4619 | ret = 1; |
b1e38734 | 4620 | fail: |
1da177e4 | 4621 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4622 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4623 | return ret; |
1da177e4 LT |
4624 | } |
4625 | ||
e6628d5b MG |
4626 | #ifdef CONFIG_NUMA_BALANCING |
4627 | /* Migrate current task p to target_cpu */ | |
4628 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4629 | { | |
4630 | struct migration_arg arg = { p, target_cpu }; | |
4631 | int curr_cpu = task_cpu(p); | |
4632 | ||
4633 | if (curr_cpu == target_cpu) | |
4634 | return 0; | |
4635 | ||
4636 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4637 | return -EINVAL; | |
4638 | ||
4639 | /* TODO: This is not properly updating schedstats */ | |
4640 | ||
4641 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); | |
4642 | } | |
0ec8aa00 PZ |
4643 | |
4644 | /* | |
4645 | * Requeue a task on a given node and accurately track the number of NUMA | |
4646 | * tasks on the runqueues | |
4647 | */ | |
4648 | void sched_setnuma(struct task_struct *p, int nid) | |
4649 | { | |
4650 | struct rq *rq; | |
4651 | unsigned long flags; | |
4652 | bool on_rq, running; | |
4653 | ||
4654 | rq = task_rq_lock(p, &flags); | |
4655 | on_rq = p->on_rq; | |
4656 | running = task_current(rq, p); | |
4657 | ||
4658 | if (on_rq) | |
4659 | dequeue_task(rq, p, 0); | |
4660 | if (running) | |
4661 | p->sched_class->put_prev_task(rq, p); | |
4662 | ||
4663 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4664 | |
4665 | if (running) | |
4666 | p->sched_class->set_curr_task(rq); | |
4667 | if (on_rq) | |
4668 | enqueue_task(rq, p, 0); | |
4669 | task_rq_unlock(rq, p, &flags); | |
4670 | } | |
e6628d5b MG |
4671 | #endif |
4672 | ||
1da177e4 | 4673 | /* |
969c7921 TH |
4674 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4675 | * and performs thread migration by bumping thread off CPU then | |
4676 | * 'pushing' onto another runqueue. | |
1da177e4 | 4677 | */ |
969c7921 | 4678 | static int migration_cpu_stop(void *data) |
1da177e4 | 4679 | { |
969c7921 | 4680 | struct migration_arg *arg = data; |
f7b4cddc | 4681 | |
969c7921 TH |
4682 | /* |
4683 | * The original target cpu might have gone down and we might | |
4684 | * be on another cpu but it doesn't matter. | |
4685 | */ | |
f7b4cddc | 4686 | local_irq_disable(); |
969c7921 | 4687 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4688 | local_irq_enable(); |
1da177e4 | 4689 | return 0; |
f7b4cddc ON |
4690 | } |
4691 | ||
1da177e4 | 4692 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4693 | |
054b9108 | 4694 | /* |
48c5ccae PZ |
4695 | * Ensures that the idle task is using init_mm right before its cpu goes |
4696 | * offline. | |
054b9108 | 4697 | */ |
48c5ccae | 4698 | void idle_task_exit(void) |
1da177e4 | 4699 | { |
48c5ccae | 4700 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4701 | |
48c5ccae | 4702 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4703 | |
48c5ccae PZ |
4704 | if (mm != &init_mm) |
4705 | switch_mm(mm, &init_mm, current); | |
4706 | mmdrop(mm); | |
1da177e4 LT |
4707 | } |
4708 | ||
4709 | /* | |
5d180232 PZ |
4710 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4711 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4712 | * nr_active count is stable. | |
4713 | * | |
4714 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4715 | */ |
5d180232 | 4716 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4717 | { |
5d180232 PZ |
4718 | long delta = calc_load_fold_active(rq); |
4719 | if (delta) | |
4720 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4721 | } |
4722 | ||
48f24c4d | 4723 | /* |
48c5ccae PZ |
4724 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4725 | * try_to_wake_up()->select_task_rq(). | |
4726 | * | |
4727 | * Called with rq->lock held even though we'er in stop_machine() and | |
4728 | * there's no concurrency possible, we hold the required locks anyway | |
4729 | * because of lock validation efforts. | |
1da177e4 | 4730 | */ |
48c5ccae | 4731 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4732 | { |
70b97a7f | 4733 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4734 | struct task_struct *next, *stop = rq->stop; |
4735 | int dest_cpu; | |
1da177e4 LT |
4736 | |
4737 | /* | |
48c5ccae PZ |
4738 | * Fudge the rq selection such that the below task selection loop |
4739 | * doesn't get stuck on the currently eligible stop task. | |
4740 | * | |
4741 | * We're currently inside stop_machine() and the rq is either stuck | |
4742 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4743 | * either way we should never end up calling schedule() until we're | |
4744 | * done here. | |
1da177e4 | 4745 | */ |
48c5ccae | 4746 | rq->stop = NULL; |
48f24c4d | 4747 | |
77bd3970 FW |
4748 | /* |
4749 | * put_prev_task() and pick_next_task() sched | |
4750 | * class method both need to have an up-to-date | |
4751 | * value of rq->clock[_task] | |
4752 | */ | |
4753 | update_rq_clock(rq); | |
4754 | ||
dd41f596 | 4755 | for ( ; ; ) { |
48c5ccae PZ |
4756 | /* |
4757 | * There's this thread running, bail when that's the only | |
4758 | * remaining thread. | |
4759 | */ | |
4760 | if (rq->nr_running == 1) | |
dd41f596 | 4761 | break; |
48c5ccae | 4762 | |
b67802ea | 4763 | next = pick_next_task(rq); |
48c5ccae | 4764 | BUG_ON(!next); |
79c53799 | 4765 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4766 | |
48c5ccae PZ |
4767 | /* Find suitable destination for @next, with force if needed. */ |
4768 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4769 | raw_spin_unlock(&rq->lock); | |
4770 | ||
4771 | __migrate_task(next, dead_cpu, dest_cpu); | |
4772 | ||
4773 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4774 | } |
dce48a84 | 4775 | |
48c5ccae | 4776 | rq->stop = stop; |
dce48a84 | 4777 | } |
48c5ccae | 4778 | |
1da177e4 LT |
4779 | #endif /* CONFIG_HOTPLUG_CPU */ |
4780 | ||
e692ab53 NP |
4781 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4782 | ||
4783 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4784 | { |
4785 | .procname = "sched_domain", | |
c57baf1e | 4786 | .mode = 0555, |
e0361851 | 4787 | }, |
56992309 | 4788 | {} |
e692ab53 NP |
4789 | }; |
4790 | ||
4791 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4792 | { |
4793 | .procname = "kernel", | |
c57baf1e | 4794 | .mode = 0555, |
e0361851 AD |
4795 | .child = sd_ctl_dir, |
4796 | }, | |
56992309 | 4797 | {} |
e692ab53 NP |
4798 | }; |
4799 | ||
4800 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4801 | { | |
4802 | struct ctl_table *entry = | |
5cf9f062 | 4803 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4804 | |
e692ab53 NP |
4805 | return entry; |
4806 | } | |
4807 | ||
6382bc90 MM |
4808 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4809 | { | |
cd790076 | 4810 | struct ctl_table *entry; |
6382bc90 | 4811 | |
cd790076 MM |
4812 | /* |
4813 | * In the intermediate directories, both the child directory and | |
4814 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4815 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4816 | * static strings and all have proc handlers. |
4817 | */ | |
4818 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4819 | if (entry->child) |
4820 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4821 | if (entry->proc_handler == NULL) |
4822 | kfree(entry->procname); | |
4823 | } | |
6382bc90 MM |
4824 | |
4825 | kfree(*tablep); | |
4826 | *tablep = NULL; | |
4827 | } | |
4828 | ||
201c373e | 4829 | static int min_load_idx = 0; |
fd9b86d3 | 4830 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4831 | |
e692ab53 | 4832 | static void |
e0361851 | 4833 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4834 | const char *procname, void *data, int maxlen, |
201c373e NK |
4835 | umode_t mode, proc_handler *proc_handler, |
4836 | bool load_idx) | |
e692ab53 | 4837 | { |
e692ab53 NP |
4838 | entry->procname = procname; |
4839 | entry->data = data; | |
4840 | entry->maxlen = maxlen; | |
4841 | entry->mode = mode; | |
4842 | entry->proc_handler = proc_handler; | |
201c373e NK |
4843 | |
4844 | if (load_idx) { | |
4845 | entry->extra1 = &min_load_idx; | |
4846 | entry->extra2 = &max_load_idx; | |
4847 | } | |
e692ab53 NP |
4848 | } |
4849 | ||
4850 | static struct ctl_table * | |
4851 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4852 | { | |
a5d8c348 | 4853 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 4854 | |
ad1cdc1d MM |
4855 | if (table == NULL) |
4856 | return NULL; | |
4857 | ||
e0361851 | 4858 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4859 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4860 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4861 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4862 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4863 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4864 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4865 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4866 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4867 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4868 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4869 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4870 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4871 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4872 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4873 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4874 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4875 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4876 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4877 | &sd->cache_nice_tries, |
201c373e | 4878 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4879 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4880 | sizeof(int), 0644, proc_dointvec_minmax, false); |
a5d8c348 | 4881 | set_table_entry(&table[11], "name", sd->name, |
201c373e | 4882 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
a5d8c348 | 4883 | /* &table[12] is terminator */ |
e692ab53 NP |
4884 | |
4885 | return table; | |
4886 | } | |
4887 | ||
be7002e6 | 4888 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4889 | { |
4890 | struct ctl_table *entry, *table; | |
4891 | struct sched_domain *sd; | |
4892 | int domain_num = 0, i; | |
4893 | char buf[32]; | |
4894 | ||
4895 | for_each_domain(cpu, sd) | |
4896 | domain_num++; | |
4897 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4898 | if (table == NULL) |
4899 | return NULL; | |
e692ab53 NP |
4900 | |
4901 | i = 0; | |
4902 | for_each_domain(cpu, sd) { | |
4903 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4904 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4905 | entry->mode = 0555; |
e692ab53 NP |
4906 | entry->child = sd_alloc_ctl_domain_table(sd); |
4907 | entry++; | |
4908 | i++; | |
4909 | } | |
4910 | return table; | |
4911 | } | |
4912 | ||
4913 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4914 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4915 | { |
6ad4c188 | 4916 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4917 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4918 | char buf[32]; | |
4919 | ||
7378547f MM |
4920 | WARN_ON(sd_ctl_dir[0].child); |
4921 | sd_ctl_dir[0].child = entry; | |
4922 | ||
ad1cdc1d MM |
4923 | if (entry == NULL) |
4924 | return; | |
4925 | ||
6ad4c188 | 4926 | for_each_possible_cpu(i) { |
e692ab53 | 4927 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4928 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4929 | entry->mode = 0555; |
e692ab53 | 4930 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4931 | entry++; |
e692ab53 | 4932 | } |
7378547f MM |
4933 | |
4934 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4935 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4936 | } | |
6382bc90 | 4937 | |
7378547f | 4938 | /* may be called multiple times per register */ |
6382bc90 MM |
4939 | static void unregister_sched_domain_sysctl(void) |
4940 | { | |
7378547f MM |
4941 | if (sd_sysctl_header) |
4942 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4943 | sd_sysctl_header = NULL; |
7378547f MM |
4944 | if (sd_ctl_dir[0].child) |
4945 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4946 | } |
e692ab53 | 4947 | #else |
6382bc90 MM |
4948 | static void register_sched_domain_sysctl(void) |
4949 | { | |
4950 | } | |
4951 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4952 | { |
4953 | } | |
4954 | #endif | |
4955 | ||
1f11eb6a GH |
4956 | static void set_rq_online(struct rq *rq) |
4957 | { | |
4958 | if (!rq->online) { | |
4959 | const struct sched_class *class; | |
4960 | ||
c6c4927b | 4961 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4962 | rq->online = 1; |
4963 | ||
4964 | for_each_class(class) { | |
4965 | if (class->rq_online) | |
4966 | class->rq_online(rq); | |
4967 | } | |
4968 | } | |
4969 | } | |
4970 | ||
4971 | static void set_rq_offline(struct rq *rq) | |
4972 | { | |
4973 | if (rq->online) { | |
4974 | const struct sched_class *class; | |
4975 | ||
4976 | for_each_class(class) { | |
4977 | if (class->rq_offline) | |
4978 | class->rq_offline(rq); | |
4979 | } | |
4980 | ||
c6c4927b | 4981 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4982 | rq->online = 0; |
4983 | } | |
4984 | } | |
4985 | ||
1da177e4 LT |
4986 | /* |
4987 | * migration_call - callback that gets triggered when a CPU is added. | |
4988 | * Here we can start up the necessary migration thread for the new CPU. | |
4989 | */ | |
0db0628d | 4990 | static int |
48f24c4d | 4991 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 4992 | { |
48f24c4d | 4993 | int cpu = (long)hcpu; |
1da177e4 | 4994 | unsigned long flags; |
969c7921 | 4995 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 4996 | |
48c5ccae | 4997 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 4998 | |
1da177e4 | 4999 | case CPU_UP_PREPARE: |
a468d389 | 5000 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5001 | break; |
48f24c4d | 5002 | |
1da177e4 | 5003 | case CPU_ONLINE: |
1f94ef59 | 5004 | /* Update our root-domain */ |
05fa785c | 5005 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5006 | if (rq->rd) { |
c6c4927b | 5007 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5008 | |
5009 | set_rq_online(rq); | |
1f94ef59 | 5010 | } |
05fa785c | 5011 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5012 | break; |
48f24c4d | 5013 | |
1da177e4 | 5014 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5015 | case CPU_DYING: |
317f3941 | 5016 | sched_ttwu_pending(); |
57d885fe | 5017 | /* Update our root-domain */ |
05fa785c | 5018 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5019 | if (rq->rd) { |
c6c4927b | 5020 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5021 | set_rq_offline(rq); |
57d885fe | 5022 | } |
48c5ccae PZ |
5023 | migrate_tasks(cpu); |
5024 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5025 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5026 | break; |
48c5ccae | 5027 | |
5d180232 | 5028 | case CPU_DEAD: |
f319da0c | 5029 | calc_load_migrate(rq); |
57d885fe | 5030 | break; |
1da177e4 LT |
5031 | #endif |
5032 | } | |
49c022e6 PZ |
5033 | |
5034 | update_max_interval(); | |
5035 | ||
1da177e4 LT |
5036 | return NOTIFY_OK; |
5037 | } | |
5038 | ||
f38b0820 PM |
5039 | /* |
5040 | * Register at high priority so that task migration (migrate_all_tasks) | |
5041 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5042 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5043 | */ |
0db0628d | 5044 | static struct notifier_block migration_notifier = { |
1da177e4 | 5045 | .notifier_call = migration_call, |
50a323b7 | 5046 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5047 | }; |
5048 | ||
0db0628d | 5049 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5050 | unsigned long action, void *hcpu) |
5051 | { | |
5052 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5053 | case CPU_STARTING: |
3a101d05 TH |
5054 | case CPU_DOWN_FAILED: |
5055 | set_cpu_active((long)hcpu, true); | |
5056 | return NOTIFY_OK; | |
5057 | default: | |
5058 | return NOTIFY_DONE; | |
5059 | } | |
5060 | } | |
5061 | ||
0db0628d | 5062 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5063 | unsigned long action, void *hcpu) |
5064 | { | |
5065 | switch (action & ~CPU_TASKS_FROZEN) { | |
5066 | case CPU_DOWN_PREPARE: | |
5067 | set_cpu_active((long)hcpu, false); | |
5068 | return NOTIFY_OK; | |
5069 | default: | |
5070 | return NOTIFY_DONE; | |
5071 | } | |
5072 | } | |
5073 | ||
7babe8db | 5074 | static int __init migration_init(void) |
1da177e4 LT |
5075 | { |
5076 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5077 | int err; |
48f24c4d | 5078 | |
3a101d05 | 5079 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5080 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5081 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5082 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5083 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5084 | |
3a101d05 TH |
5085 | /* Register cpu active notifiers */ |
5086 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5087 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5088 | ||
a004cd42 | 5089 | return 0; |
1da177e4 | 5090 | } |
7babe8db | 5091 | early_initcall(migration_init); |
1da177e4 LT |
5092 | #endif |
5093 | ||
5094 | #ifdef CONFIG_SMP | |
476f3534 | 5095 | |
4cb98839 PZ |
5096 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5097 | ||
3e9830dc | 5098 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5099 | |
d039ac60 | 5100 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5101 | |
d039ac60 | 5102 | static int __init sched_debug_setup(char *str) |
f6630114 | 5103 | { |
d039ac60 | 5104 | sched_debug_enabled = 1; |
f6630114 MT |
5105 | |
5106 | return 0; | |
5107 | } | |
d039ac60 PZ |
5108 | early_param("sched_debug", sched_debug_setup); |
5109 | ||
5110 | static inline bool sched_debug(void) | |
5111 | { | |
5112 | return sched_debug_enabled; | |
5113 | } | |
f6630114 | 5114 | |
7c16ec58 | 5115 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5116 | struct cpumask *groupmask) |
1da177e4 | 5117 | { |
4dcf6aff | 5118 | struct sched_group *group = sd->groups; |
434d53b0 | 5119 | char str[256]; |
1da177e4 | 5120 | |
968ea6d8 | 5121 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5122 | cpumask_clear(groupmask); |
4dcf6aff IM |
5123 | |
5124 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5125 | ||
5126 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5127 | printk("does not load-balance\n"); |
4dcf6aff | 5128 | if (sd->parent) |
3df0fc5b PZ |
5129 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5130 | " has parent"); | |
4dcf6aff | 5131 | return -1; |
41c7ce9a NP |
5132 | } |
5133 | ||
3df0fc5b | 5134 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5135 | |
758b2cdc | 5136 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5137 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5138 | "CPU%d\n", cpu); | |
4dcf6aff | 5139 | } |
758b2cdc | 5140 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5141 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5142 | " CPU%d\n", cpu); | |
4dcf6aff | 5143 | } |
1da177e4 | 5144 | |
4dcf6aff | 5145 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5146 | do { |
4dcf6aff | 5147 | if (!group) { |
3df0fc5b PZ |
5148 | printk("\n"); |
5149 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5150 | break; |
5151 | } | |
5152 | ||
c3decf0d PZ |
5153 | /* |
5154 | * Even though we initialize ->power to something semi-sane, | |
5155 | * we leave power_orig unset. This allows us to detect if | |
5156 | * domain iteration is still funny without causing /0 traps. | |
5157 | */ | |
5158 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5159 | printk(KERN_CONT "\n"); |
5160 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5161 | "set\n"); | |
4dcf6aff IM |
5162 | break; |
5163 | } | |
1da177e4 | 5164 | |
758b2cdc | 5165 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5166 | printk(KERN_CONT "\n"); |
5167 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5168 | break; |
5169 | } | |
1da177e4 | 5170 | |
cb83b629 PZ |
5171 | if (!(sd->flags & SD_OVERLAP) && |
5172 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5173 | printk(KERN_CONT "\n"); |
5174 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5175 | break; |
5176 | } | |
1da177e4 | 5177 | |
758b2cdc | 5178 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5179 | |
968ea6d8 | 5180 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5181 | |
3df0fc5b | 5182 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5183 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5184 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5185 | group->sgp->power); |
381512cf | 5186 | } |
1da177e4 | 5187 | |
4dcf6aff IM |
5188 | group = group->next; |
5189 | } while (group != sd->groups); | |
3df0fc5b | 5190 | printk(KERN_CONT "\n"); |
1da177e4 | 5191 | |
758b2cdc | 5192 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5193 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5194 | |
758b2cdc RR |
5195 | if (sd->parent && |
5196 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5197 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5198 | "of domain->span\n"); | |
4dcf6aff IM |
5199 | return 0; |
5200 | } | |
1da177e4 | 5201 | |
4dcf6aff IM |
5202 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5203 | { | |
5204 | int level = 0; | |
1da177e4 | 5205 | |
d039ac60 | 5206 | if (!sched_debug_enabled) |
f6630114 MT |
5207 | return; |
5208 | ||
4dcf6aff IM |
5209 | if (!sd) { |
5210 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5211 | return; | |
5212 | } | |
1da177e4 | 5213 | |
4dcf6aff IM |
5214 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5215 | ||
5216 | for (;;) { | |
4cb98839 | 5217 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5218 | break; |
1da177e4 LT |
5219 | level++; |
5220 | sd = sd->parent; | |
33859f7f | 5221 | if (!sd) |
4dcf6aff IM |
5222 | break; |
5223 | } | |
1da177e4 | 5224 | } |
6d6bc0ad | 5225 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5226 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5227 | static inline bool sched_debug(void) |
5228 | { | |
5229 | return false; | |
5230 | } | |
6d6bc0ad | 5231 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5232 | |
1a20ff27 | 5233 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5234 | { |
758b2cdc | 5235 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5236 | return 1; |
5237 | ||
5238 | /* Following flags need at least 2 groups */ | |
5239 | if (sd->flags & (SD_LOAD_BALANCE | | |
5240 | SD_BALANCE_NEWIDLE | | |
5241 | SD_BALANCE_FORK | | |
89c4710e SS |
5242 | SD_BALANCE_EXEC | |
5243 | SD_SHARE_CPUPOWER | | |
5244 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5245 | if (sd->groups != sd->groups->next) |
5246 | return 0; | |
5247 | } | |
5248 | ||
5249 | /* Following flags don't use groups */ | |
c88d5910 | 5250 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5251 | return 0; |
5252 | ||
5253 | return 1; | |
5254 | } | |
5255 | ||
48f24c4d IM |
5256 | static int |
5257 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5258 | { |
5259 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5260 | ||
5261 | if (sd_degenerate(parent)) | |
5262 | return 1; | |
5263 | ||
758b2cdc | 5264 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5265 | return 0; |
5266 | ||
245af2c7 SS |
5267 | /* Flags needing groups don't count if only 1 group in parent */ |
5268 | if (parent->groups == parent->groups->next) { | |
5269 | pflags &= ~(SD_LOAD_BALANCE | | |
5270 | SD_BALANCE_NEWIDLE | | |
5271 | SD_BALANCE_FORK | | |
89c4710e SS |
5272 | SD_BALANCE_EXEC | |
5273 | SD_SHARE_CPUPOWER | | |
10866e62 PZ |
5274 | SD_SHARE_PKG_RESOURCES | |
5275 | SD_PREFER_SIBLING); | |
5436499e KC |
5276 | if (nr_node_ids == 1) |
5277 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5278 | } |
5279 | if (~cflags & pflags) | |
5280 | return 0; | |
5281 | ||
5282 | return 1; | |
5283 | } | |
5284 | ||
dce840a0 | 5285 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5286 | { |
dce840a0 | 5287 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5288 | |
68e74568 | 5289 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5290 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5291 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5292 | free_cpumask_var(rd->rto_mask); |
5293 | free_cpumask_var(rd->online); | |
5294 | free_cpumask_var(rd->span); | |
5295 | kfree(rd); | |
5296 | } | |
5297 | ||
57d885fe GH |
5298 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5299 | { | |
a0490fa3 | 5300 | struct root_domain *old_rd = NULL; |
57d885fe | 5301 | unsigned long flags; |
57d885fe | 5302 | |
05fa785c | 5303 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5304 | |
5305 | if (rq->rd) { | |
a0490fa3 | 5306 | old_rd = rq->rd; |
57d885fe | 5307 | |
c6c4927b | 5308 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5309 | set_rq_offline(rq); |
57d885fe | 5310 | |
c6c4927b | 5311 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5312 | |
a0490fa3 | 5313 | /* |
0515973f | 5314 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5315 | * set old_rd to NULL to skip the freeing later |
5316 | * in this function: | |
5317 | */ | |
5318 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5319 | old_rd = NULL; | |
57d885fe GH |
5320 | } |
5321 | ||
5322 | atomic_inc(&rd->refcount); | |
5323 | rq->rd = rd; | |
5324 | ||
c6c4927b | 5325 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5326 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5327 | set_rq_online(rq); |
57d885fe | 5328 | |
05fa785c | 5329 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5330 | |
5331 | if (old_rd) | |
dce840a0 | 5332 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5333 | } |
5334 | ||
68c38fc3 | 5335 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5336 | { |
5337 | memset(rd, 0, sizeof(*rd)); | |
5338 | ||
68c38fc3 | 5339 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5340 | goto out; |
68c38fc3 | 5341 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5342 | goto free_span; |
1baca4ce | 5343 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5344 | goto free_online; |
1baca4ce JL |
5345 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5346 | goto free_dlo_mask; | |
6e0534f2 | 5347 | |
332ac17e | 5348 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5349 | if (cpudl_init(&rd->cpudl) != 0) |
5350 | goto free_dlo_mask; | |
332ac17e | 5351 | |
68c38fc3 | 5352 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5353 | goto free_rto_mask; |
c6c4927b | 5354 | return 0; |
6e0534f2 | 5355 | |
68e74568 RR |
5356 | free_rto_mask: |
5357 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5358 | free_dlo_mask: |
5359 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5360 | free_online: |
5361 | free_cpumask_var(rd->online); | |
5362 | free_span: | |
5363 | free_cpumask_var(rd->span); | |
0c910d28 | 5364 | out: |
c6c4927b | 5365 | return -ENOMEM; |
57d885fe GH |
5366 | } |
5367 | ||
029632fb PZ |
5368 | /* |
5369 | * By default the system creates a single root-domain with all cpus as | |
5370 | * members (mimicking the global state we have today). | |
5371 | */ | |
5372 | struct root_domain def_root_domain; | |
5373 | ||
57d885fe GH |
5374 | static void init_defrootdomain(void) |
5375 | { | |
68c38fc3 | 5376 | init_rootdomain(&def_root_domain); |
c6c4927b | 5377 | |
57d885fe GH |
5378 | atomic_set(&def_root_domain.refcount, 1); |
5379 | } | |
5380 | ||
dc938520 | 5381 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5382 | { |
5383 | struct root_domain *rd; | |
5384 | ||
5385 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5386 | if (!rd) | |
5387 | return NULL; | |
5388 | ||
68c38fc3 | 5389 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5390 | kfree(rd); |
5391 | return NULL; | |
5392 | } | |
57d885fe GH |
5393 | |
5394 | return rd; | |
5395 | } | |
5396 | ||
e3589f6c PZ |
5397 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5398 | { | |
5399 | struct sched_group *tmp, *first; | |
5400 | ||
5401 | if (!sg) | |
5402 | return; | |
5403 | ||
5404 | first = sg; | |
5405 | do { | |
5406 | tmp = sg->next; | |
5407 | ||
5408 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5409 | kfree(sg->sgp); | |
5410 | ||
5411 | kfree(sg); | |
5412 | sg = tmp; | |
5413 | } while (sg != first); | |
5414 | } | |
5415 | ||
dce840a0 PZ |
5416 | static void free_sched_domain(struct rcu_head *rcu) |
5417 | { | |
5418 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5419 | |
5420 | /* | |
5421 | * If its an overlapping domain it has private groups, iterate and | |
5422 | * nuke them all. | |
5423 | */ | |
5424 | if (sd->flags & SD_OVERLAP) { | |
5425 | free_sched_groups(sd->groups, 1); | |
5426 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5427 | kfree(sd->groups->sgp); |
dce840a0 | 5428 | kfree(sd->groups); |
9c3f75cb | 5429 | } |
dce840a0 PZ |
5430 | kfree(sd); |
5431 | } | |
5432 | ||
5433 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5434 | { | |
5435 | call_rcu(&sd->rcu, free_sched_domain); | |
5436 | } | |
5437 | ||
5438 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5439 | { | |
5440 | for (; sd; sd = sd->parent) | |
5441 | destroy_sched_domain(sd, cpu); | |
5442 | } | |
5443 | ||
518cd623 PZ |
5444 | /* |
5445 | * Keep a special pointer to the highest sched_domain that has | |
5446 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5447 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5448 | * | |
5449 | * Also keep a unique ID per domain (we use the first cpu number in | |
5450 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5451 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5452 | */ |
5453 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5454 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5455 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5456 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5457 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5458 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5459 | |
5460 | static void update_top_cache_domain(int cpu) | |
5461 | { | |
5462 | struct sched_domain *sd; | |
5d4cf996 | 5463 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5464 | int id = cpu; |
7d9ffa89 | 5465 | int size = 1; |
518cd623 PZ |
5466 | |
5467 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5468 | if (sd) { |
518cd623 | 5469 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5470 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5471 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5472 | } |
5d4cf996 | 5473 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5474 | |
5475 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5476 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5477 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5478 | |
5479 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5480 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5481 | |
5482 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5483 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5484 | } |
5485 | ||
1da177e4 | 5486 | /* |
0eab9146 | 5487 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5488 | * hold the hotplug lock. |
5489 | */ | |
0eab9146 IM |
5490 | static void |
5491 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5492 | { |
70b97a7f | 5493 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5494 | struct sched_domain *tmp; |
5495 | ||
5496 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5497 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5498 | struct sched_domain *parent = tmp->parent; |
5499 | if (!parent) | |
5500 | break; | |
f29c9b1c | 5501 | |
1a848870 | 5502 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5503 | tmp->parent = parent->parent; |
1a848870 SS |
5504 | if (parent->parent) |
5505 | parent->parent->child = tmp; | |
10866e62 PZ |
5506 | /* |
5507 | * Transfer SD_PREFER_SIBLING down in case of a | |
5508 | * degenerate parent; the spans match for this | |
5509 | * so the property transfers. | |
5510 | */ | |
5511 | if (parent->flags & SD_PREFER_SIBLING) | |
5512 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5513 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5514 | } else |
5515 | tmp = tmp->parent; | |
245af2c7 SS |
5516 | } |
5517 | ||
1a848870 | 5518 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5519 | tmp = sd; |
245af2c7 | 5520 | sd = sd->parent; |
dce840a0 | 5521 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5522 | if (sd) |
5523 | sd->child = NULL; | |
5524 | } | |
1da177e4 | 5525 | |
4cb98839 | 5526 | sched_domain_debug(sd, cpu); |
1da177e4 | 5527 | |
57d885fe | 5528 | rq_attach_root(rq, rd); |
dce840a0 | 5529 | tmp = rq->sd; |
674311d5 | 5530 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5531 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5532 | |
5533 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5534 | } |
5535 | ||
5536 | /* cpus with isolated domains */ | |
dcc30a35 | 5537 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5538 | |
5539 | /* Setup the mask of cpus configured for isolated domains */ | |
5540 | static int __init isolated_cpu_setup(char *str) | |
5541 | { | |
bdddd296 | 5542 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5543 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5544 | return 1; |
5545 | } | |
5546 | ||
8927f494 | 5547 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5548 | |
d3081f52 PZ |
5549 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5550 | { | |
5551 | return cpumask_of_node(cpu_to_node(cpu)); | |
5552 | } | |
5553 | ||
dce840a0 PZ |
5554 | struct sd_data { |
5555 | struct sched_domain **__percpu sd; | |
5556 | struct sched_group **__percpu sg; | |
9c3f75cb | 5557 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5558 | }; |
5559 | ||
49a02c51 | 5560 | struct s_data { |
21d42ccf | 5561 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5562 | struct root_domain *rd; |
5563 | }; | |
5564 | ||
2109b99e | 5565 | enum s_alloc { |
2109b99e | 5566 | sa_rootdomain, |
21d42ccf | 5567 | sa_sd, |
dce840a0 | 5568 | sa_sd_storage, |
2109b99e AH |
5569 | sa_none, |
5570 | }; | |
5571 | ||
54ab4ff4 PZ |
5572 | struct sched_domain_topology_level; |
5573 | ||
5574 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5575 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5576 | ||
e3589f6c PZ |
5577 | #define SDTL_OVERLAP 0x01 |
5578 | ||
eb7a74e6 | 5579 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5580 | sched_domain_init_f init; |
5581 | sched_domain_mask_f mask; | |
e3589f6c | 5582 | int flags; |
cb83b629 | 5583 | int numa_level; |
54ab4ff4 | 5584 | struct sd_data data; |
eb7a74e6 PZ |
5585 | }; |
5586 | ||
c1174876 PZ |
5587 | /* |
5588 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5589 | * domain traversal. | |
5590 | * | |
5591 | * Asymmetric node setups can result in situations where the domain tree is of | |
5592 | * unequal depth, make sure to skip domains that already cover the entire | |
5593 | * range. | |
5594 | * | |
5595 | * In that case build_sched_domains() will have terminated the iteration early | |
5596 | * and our sibling sd spans will be empty. Domains should always include the | |
5597 | * cpu they're built on, so check that. | |
5598 | * | |
5599 | */ | |
5600 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5601 | { | |
5602 | const struct cpumask *span = sched_domain_span(sd); | |
5603 | struct sd_data *sdd = sd->private; | |
5604 | struct sched_domain *sibling; | |
5605 | int i; | |
5606 | ||
5607 | for_each_cpu(i, span) { | |
5608 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5609 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5610 | continue; | |
5611 | ||
5612 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5613 | } | |
5614 | } | |
5615 | ||
5616 | /* | |
5617 | * Return the canonical balance cpu for this group, this is the first cpu | |
5618 | * of this group that's also in the iteration mask. | |
5619 | */ | |
5620 | int group_balance_cpu(struct sched_group *sg) | |
5621 | { | |
5622 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5623 | } | |
5624 | ||
e3589f6c PZ |
5625 | static int |
5626 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5627 | { | |
5628 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5629 | const struct cpumask *span = sched_domain_span(sd); | |
5630 | struct cpumask *covered = sched_domains_tmpmask; | |
5631 | struct sd_data *sdd = sd->private; | |
5632 | struct sched_domain *child; | |
5633 | int i; | |
5634 | ||
5635 | cpumask_clear(covered); | |
5636 | ||
5637 | for_each_cpu(i, span) { | |
5638 | struct cpumask *sg_span; | |
5639 | ||
5640 | if (cpumask_test_cpu(i, covered)) | |
5641 | continue; | |
5642 | ||
c1174876 PZ |
5643 | child = *per_cpu_ptr(sdd->sd, i); |
5644 | ||
5645 | /* See the comment near build_group_mask(). */ | |
5646 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5647 | continue; | |
5648 | ||
e3589f6c | 5649 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5650 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5651 | |
5652 | if (!sg) | |
5653 | goto fail; | |
5654 | ||
5655 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5656 | if (child->child) { |
5657 | child = child->child; | |
5658 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5659 | } else | |
5660 | cpumask_set_cpu(i, sg_span); | |
5661 | ||
5662 | cpumask_or(covered, covered, sg_span); | |
5663 | ||
74a5ce20 | 5664 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5665 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5666 | build_group_mask(sd, sg); | |
5667 | ||
c3decf0d PZ |
5668 | /* |
5669 | * Initialize sgp->power such that even if we mess up the | |
5670 | * domains and no possible iteration will get us here, we won't | |
5671 | * die on a /0 trap. | |
5672 | */ | |
5673 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
8e8339a3 | 5674 | sg->sgp->power_orig = sg->sgp->power; |
e3589f6c | 5675 | |
c1174876 PZ |
5676 | /* |
5677 | * Make sure the first group of this domain contains the | |
5678 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5679 | * breaks. See update_sg_lb_stats(). | |
5680 | */ | |
74a5ce20 | 5681 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5682 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5683 | groups = sg; |
5684 | ||
5685 | if (!first) | |
5686 | first = sg; | |
5687 | if (last) | |
5688 | last->next = sg; | |
5689 | last = sg; | |
5690 | last->next = first; | |
5691 | } | |
5692 | sd->groups = groups; | |
5693 | ||
5694 | return 0; | |
5695 | ||
5696 | fail: | |
5697 | free_sched_groups(first, 0); | |
5698 | ||
5699 | return -ENOMEM; | |
5700 | } | |
5701 | ||
dce840a0 | 5702 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5703 | { |
dce840a0 PZ |
5704 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5705 | struct sched_domain *child = sd->child; | |
1da177e4 | 5706 | |
dce840a0 PZ |
5707 | if (child) |
5708 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5709 | |
9c3f75cb | 5710 | if (sg) { |
dce840a0 | 5711 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5712 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5713 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5714 | } |
dce840a0 PZ |
5715 | |
5716 | return cpu; | |
1e9f28fa | 5717 | } |
1e9f28fa | 5718 | |
01a08546 | 5719 | /* |
dce840a0 PZ |
5720 | * build_sched_groups will build a circular linked list of the groups |
5721 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5722 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5723 | * |
5724 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5725 | */ |
e3589f6c PZ |
5726 | static int |
5727 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5728 | { |
dce840a0 PZ |
5729 | struct sched_group *first = NULL, *last = NULL; |
5730 | struct sd_data *sdd = sd->private; | |
5731 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5732 | struct cpumask *covered; |
dce840a0 | 5733 | int i; |
9c1cfda2 | 5734 | |
e3589f6c PZ |
5735 | get_group(cpu, sdd, &sd->groups); |
5736 | atomic_inc(&sd->groups->ref); | |
5737 | ||
0936629f | 5738 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5739 | return 0; |
5740 | ||
f96225fd PZ |
5741 | lockdep_assert_held(&sched_domains_mutex); |
5742 | covered = sched_domains_tmpmask; | |
5743 | ||
dce840a0 | 5744 | cpumask_clear(covered); |
6711cab4 | 5745 | |
dce840a0 PZ |
5746 | for_each_cpu(i, span) { |
5747 | struct sched_group *sg; | |
cd08e923 | 5748 | int group, j; |
6711cab4 | 5749 | |
dce840a0 PZ |
5750 | if (cpumask_test_cpu(i, covered)) |
5751 | continue; | |
6711cab4 | 5752 | |
cd08e923 | 5753 | group = get_group(i, sdd, &sg); |
dce840a0 | 5754 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5755 | sg->sgp->power = 0; |
c1174876 | 5756 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5757 | |
dce840a0 PZ |
5758 | for_each_cpu(j, span) { |
5759 | if (get_group(j, sdd, NULL) != group) | |
5760 | continue; | |
0601a88d | 5761 | |
dce840a0 PZ |
5762 | cpumask_set_cpu(j, covered); |
5763 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5764 | } | |
0601a88d | 5765 | |
dce840a0 PZ |
5766 | if (!first) |
5767 | first = sg; | |
5768 | if (last) | |
5769 | last->next = sg; | |
5770 | last = sg; | |
5771 | } | |
5772 | last->next = first; | |
e3589f6c PZ |
5773 | |
5774 | return 0; | |
0601a88d | 5775 | } |
51888ca2 | 5776 | |
89c4710e SS |
5777 | /* |
5778 | * Initialize sched groups cpu_power. | |
5779 | * | |
5780 | * cpu_power indicates the capacity of sched group, which is used while | |
5781 | * distributing the load between different sched groups in a sched domain. | |
5782 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5783 | * there are asymmetries in the topology. If there are asymmetries, group | |
5784 | * having more cpu_power will pickup more load compared to the group having | |
5785 | * less cpu_power. | |
89c4710e SS |
5786 | */ |
5787 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5788 | { | |
e3589f6c | 5789 | struct sched_group *sg = sd->groups; |
89c4710e | 5790 | |
94c95ba6 | 5791 | WARN_ON(!sg); |
e3589f6c PZ |
5792 | |
5793 | do { | |
5794 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5795 | sg = sg->next; | |
5796 | } while (sg != sd->groups); | |
89c4710e | 5797 | |
c1174876 | 5798 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5799 | return; |
aae6d3dd | 5800 | |
d274cb30 | 5801 | update_group_power(sd, cpu); |
69e1e811 | 5802 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5803 | } |
5804 | ||
029632fb PZ |
5805 | int __weak arch_sd_sibling_asym_packing(void) |
5806 | { | |
5807 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5808 | } |
5809 | ||
7c16ec58 MT |
5810 | /* |
5811 | * Initializers for schedule domains | |
5812 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5813 | */ | |
5814 | ||
a5d8c348 IM |
5815 | #ifdef CONFIG_SCHED_DEBUG |
5816 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5817 | #else | |
5818 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5819 | #endif | |
5820 | ||
54ab4ff4 PZ |
5821 | #define SD_INIT_FUNC(type) \ |
5822 | static noinline struct sched_domain * \ | |
5823 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5824 | { \ | |
5825 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5826 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5827 | SD_INIT_NAME(sd, type); \ |
5828 | sd->private = &tl->data; \ | |
5829 | return sd; \ | |
7c16ec58 MT |
5830 | } |
5831 | ||
5832 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5833 | #ifdef CONFIG_SCHED_SMT |
5834 | SD_INIT_FUNC(SIBLING) | |
5835 | #endif | |
5836 | #ifdef CONFIG_SCHED_MC | |
5837 | SD_INIT_FUNC(MC) | |
5838 | #endif | |
01a08546 HC |
5839 | #ifdef CONFIG_SCHED_BOOK |
5840 | SD_INIT_FUNC(BOOK) | |
5841 | #endif | |
7c16ec58 | 5842 | |
1d3504fc | 5843 | static int default_relax_domain_level = -1; |
60495e77 | 5844 | int sched_domain_level_max; |
1d3504fc HS |
5845 | |
5846 | static int __init setup_relax_domain_level(char *str) | |
5847 | { | |
a841f8ce DS |
5848 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5849 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5850 | |
1d3504fc HS |
5851 | return 1; |
5852 | } | |
5853 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5854 | ||
5855 | static void set_domain_attribute(struct sched_domain *sd, | |
5856 | struct sched_domain_attr *attr) | |
5857 | { | |
5858 | int request; | |
5859 | ||
5860 | if (!attr || attr->relax_domain_level < 0) { | |
5861 | if (default_relax_domain_level < 0) | |
5862 | return; | |
5863 | else | |
5864 | request = default_relax_domain_level; | |
5865 | } else | |
5866 | request = attr->relax_domain_level; | |
5867 | if (request < sd->level) { | |
5868 | /* turn off idle balance on this domain */ | |
c88d5910 | 5869 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5870 | } else { |
5871 | /* turn on idle balance on this domain */ | |
c88d5910 | 5872 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5873 | } |
5874 | } | |
5875 | ||
54ab4ff4 PZ |
5876 | static void __sdt_free(const struct cpumask *cpu_map); |
5877 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5878 | ||
2109b99e AH |
5879 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5880 | const struct cpumask *cpu_map) | |
5881 | { | |
5882 | switch (what) { | |
2109b99e | 5883 | case sa_rootdomain: |
822ff793 PZ |
5884 | if (!atomic_read(&d->rd->refcount)) |
5885 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5886 | case sa_sd: |
5887 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5888 | case sa_sd_storage: |
54ab4ff4 | 5889 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5890 | case sa_none: |
5891 | break; | |
5892 | } | |
5893 | } | |
3404c8d9 | 5894 | |
2109b99e AH |
5895 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5896 | const struct cpumask *cpu_map) | |
5897 | { | |
dce840a0 PZ |
5898 | memset(d, 0, sizeof(*d)); |
5899 | ||
54ab4ff4 PZ |
5900 | if (__sdt_alloc(cpu_map)) |
5901 | return sa_sd_storage; | |
dce840a0 PZ |
5902 | d->sd = alloc_percpu(struct sched_domain *); |
5903 | if (!d->sd) | |
5904 | return sa_sd_storage; | |
2109b99e | 5905 | d->rd = alloc_rootdomain(); |
dce840a0 | 5906 | if (!d->rd) |
21d42ccf | 5907 | return sa_sd; |
2109b99e AH |
5908 | return sa_rootdomain; |
5909 | } | |
57d885fe | 5910 | |
dce840a0 PZ |
5911 | /* |
5912 | * NULL the sd_data elements we've used to build the sched_domain and | |
5913 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5914 | * will not free the data we're using. | |
5915 | */ | |
5916 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5917 | { | |
5918 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5919 | |
5920 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5921 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5922 | ||
e3589f6c | 5923 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5924 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5925 | |
5926 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5927 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5928 | } |
5929 | ||
2c402dc3 PZ |
5930 | #ifdef CONFIG_SCHED_SMT |
5931 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 5932 | { |
2c402dc3 | 5933 | return topology_thread_cpumask(cpu); |
3bd65a80 | 5934 | } |
2c402dc3 | 5935 | #endif |
7f4588f3 | 5936 | |
d069b916 PZ |
5937 | /* |
5938 | * Topology list, bottom-up. | |
5939 | */ | |
2c402dc3 | 5940 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
5941 | #ifdef CONFIG_SCHED_SMT |
5942 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 5943 | #endif |
1e9f28fa | 5944 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 5945 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 5946 | #endif |
d069b916 PZ |
5947 | #ifdef CONFIG_SCHED_BOOK |
5948 | { sd_init_BOOK, cpu_book_mask, }, | |
5949 | #endif | |
5950 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
5951 | { NULL, }, |
5952 | }; | |
5953 | ||
5954 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
5955 | ||
27723a68 VK |
5956 | #define for_each_sd_topology(tl) \ |
5957 | for (tl = sched_domain_topology; tl->init; tl++) | |
5958 | ||
cb83b629 PZ |
5959 | #ifdef CONFIG_NUMA |
5960 | ||
5961 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
5962 | static int *sched_domains_numa_distance; |
5963 | static struct cpumask ***sched_domains_numa_masks; | |
5964 | static int sched_domains_curr_level; | |
5965 | ||
cb83b629 PZ |
5966 | static inline int sd_local_flags(int level) |
5967 | { | |
10717dcd | 5968 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
5969 | return 0; |
5970 | ||
5971 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
5972 | } | |
5973 | ||
5974 | static struct sched_domain * | |
5975 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
5976 | { | |
5977 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
5978 | int level = tl->numa_level; | |
5979 | int sd_weight = cpumask_weight( | |
5980 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
5981 | ||
5982 | *sd = (struct sched_domain){ | |
5983 | .min_interval = sd_weight, | |
5984 | .max_interval = 2*sd_weight, | |
5985 | .busy_factor = 32, | |
870a0bb5 | 5986 | .imbalance_pct = 125, |
cb83b629 PZ |
5987 | .cache_nice_tries = 2, |
5988 | .busy_idx = 3, | |
5989 | .idle_idx = 2, | |
5990 | .newidle_idx = 0, | |
5991 | .wake_idx = 0, | |
5992 | .forkexec_idx = 0, | |
5993 | ||
5994 | .flags = 1*SD_LOAD_BALANCE | |
5995 | | 1*SD_BALANCE_NEWIDLE | |
5996 | | 0*SD_BALANCE_EXEC | |
5997 | | 0*SD_BALANCE_FORK | |
5998 | | 0*SD_BALANCE_WAKE | |
5999 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 6000 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
6001 | | 0*SD_SHARE_PKG_RESOURCES |
6002 | | 1*SD_SERIALIZE | |
6003 | | 0*SD_PREFER_SIBLING | |
3a7053b3 | 6004 | | 1*SD_NUMA |
cb83b629 PZ |
6005 | | sd_local_flags(level) |
6006 | , | |
6007 | .last_balance = jiffies, | |
6008 | .balance_interval = sd_weight, | |
6009 | }; | |
6010 | SD_INIT_NAME(sd, NUMA); | |
6011 | sd->private = &tl->data; | |
6012 | ||
6013 | /* | |
6014 | * Ugly hack to pass state to sd_numa_mask()... | |
6015 | */ | |
6016 | sched_domains_curr_level = tl->numa_level; | |
6017 | ||
6018 | return sd; | |
6019 | } | |
6020 | ||
6021 | static const struct cpumask *sd_numa_mask(int cpu) | |
6022 | { | |
6023 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6024 | } | |
6025 | ||
d039ac60 PZ |
6026 | static void sched_numa_warn(const char *str) |
6027 | { | |
6028 | static int done = false; | |
6029 | int i,j; | |
6030 | ||
6031 | if (done) | |
6032 | return; | |
6033 | ||
6034 | done = true; | |
6035 | ||
6036 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6037 | ||
6038 | for (i = 0; i < nr_node_ids; i++) { | |
6039 | printk(KERN_WARNING " "); | |
6040 | for (j = 0; j < nr_node_ids; j++) | |
6041 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6042 | printk(KERN_CONT "\n"); | |
6043 | } | |
6044 | printk(KERN_WARNING "\n"); | |
6045 | } | |
6046 | ||
6047 | static bool find_numa_distance(int distance) | |
6048 | { | |
6049 | int i; | |
6050 | ||
6051 | if (distance == node_distance(0, 0)) | |
6052 | return true; | |
6053 | ||
6054 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6055 | if (sched_domains_numa_distance[i] == distance) | |
6056 | return true; | |
6057 | } | |
6058 | ||
6059 | return false; | |
6060 | } | |
6061 | ||
cb83b629 PZ |
6062 | static void sched_init_numa(void) |
6063 | { | |
6064 | int next_distance, curr_distance = node_distance(0, 0); | |
6065 | struct sched_domain_topology_level *tl; | |
6066 | int level = 0; | |
6067 | int i, j, k; | |
6068 | ||
cb83b629 PZ |
6069 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6070 | if (!sched_domains_numa_distance) | |
6071 | return; | |
6072 | ||
6073 | /* | |
6074 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6075 | * unique distances in the node_distance() table. | |
6076 | * | |
6077 | * Assumes node_distance(0,j) includes all distances in | |
6078 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6079 | */ |
6080 | next_distance = curr_distance; | |
6081 | for (i = 0; i < nr_node_ids; i++) { | |
6082 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6083 | for (k = 0; k < nr_node_ids; k++) { |
6084 | int distance = node_distance(i, k); | |
6085 | ||
6086 | if (distance > curr_distance && | |
6087 | (distance < next_distance || | |
6088 | next_distance == curr_distance)) | |
6089 | next_distance = distance; | |
6090 | ||
6091 | /* | |
6092 | * While not a strong assumption it would be nice to know | |
6093 | * about cases where if node A is connected to B, B is not | |
6094 | * equally connected to A. | |
6095 | */ | |
6096 | if (sched_debug() && node_distance(k, i) != distance) | |
6097 | sched_numa_warn("Node-distance not symmetric"); | |
6098 | ||
6099 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6100 | sched_numa_warn("Node-0 not representative"); | |
6101 | } | |
6102 | if (next_distance != curr_distance) { | |
6103 | sched_domains_numa_distance[level++] = next_distance; | |
6104 | sched_domains_numa_levels = level; | |
6105 | curr_distance = next_distance; | |
6106 | } else break; | |
cb83b629 | 6107 | } |
d039ac60 PZ |
6108 | |
6109 | /* | |
6110 | * In case of sched_debug() we verify the above assumption. | |
6111 | */ | |
6112 | if (!sched_debug()) | |
6113 | break; | |
cb83b629 PZ |
6114 | } |
6115 | /* | |
6116 | * 'level' contains the number of unique distances, excluding the | |
6117 | * identity distance node_distance(i,i). | |
6118 | * | |
28b4a521 | 6119 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6120 | * numbers. |
6121 | */ | |
6122 | ||
5f7865f3 TC |
6123 | /* |
6124 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6125 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6126 | * the array will contain less then 'level' members. This could be | |
6127 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6128 | * in other functions. | |
6129 | * | |
6130 | * We reset it to 'level' at the end of this function. | |
6131 | */ | |
6132 | sched_domains_numa_levels = 0; | |
6133 | ||
cb83b629 PZ |
6134 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6135 | if (!sched_domains_numa_masks) | |
6136 | return; | |
6137 | ||
6138 | /* | |
6139 | * Now for each level, construct a mask per node which contains all | |
6140 | * cpus of nodes that are that many hops away from us. | |
6141 | */ | |
6142 | for (i = 0; i < level; i++) { | |
6143 | sched_domains_numa_masks[i] = | |
6144 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6145 | if (!sched_domains_numa_masks[i]) | |
6146 | return; | |
6147 | ||
6148 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6149 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6150 | if (!mask) |
6151 | return; | |
6152 | ||
6153 | sched_domains_numa_masks[i][j] = mask; | |
6154 | ||
6155 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6156 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6157 | continue; |
6158 | ||
6159 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6160 | } | |
6161 | } | |
6162 | } | |
6163 | ||
6164 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6165 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6166 | if (!tl) | |
6167 | return; | |
6168 | ||
6169 | /* | |
6170 | * Copy the default topology bits.. | |
6171 | */ | |
6172 | for (i = 0; default_topology[i].init; i++) | |
6173 | tl[i] = default_topology[i]; | |
6174 | ||
6175 | /* | |
6176 | * .. and append 'j' levels of NUMA goodness. | |
6177 | */ | |
6178 | for (j = 0; j < level; i++, j++) { | |
6179 | tl[i] = (struct sched_domain_topology_level){ | |
6180 | .init = sd_numa_init, | |
6181 | .mask = sd_numa_mask, | |
6182 | .flags = SDTL_OVERLAP, | |
6183 | .numa_level = j, | |
6184 | }; | |
6185 | } | |
6186 | ||
6187 | sched_domain_topology = tl; | |
5f7865f3 TC |
6188 | |
6189 | sched_domains_numa_levels = level; | |
cb83b629 | 6190 | } |
301a5cba TC |
6191 | |
6192 | static void sched_domains_numa_masks_set(int cpu) | |
6193 | { | |
6194 | int i, j; | |
6195 | int node = cpu_to_node(cpu); | |
6196 | ||
6197 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6198 | for (j = 0; j < nr_node_ids; j++) { | |
6199 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6200 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6201 | } | |
6202 | } | |
6203 | } | |
6204 | ||
6205 | static void sched_domains_numa_masks_clear(int cpu) | |
6206 | { | |
6207 | int i, j; | |
6208 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6209 | for (j = 0; j < nr_node_ids; j++) | |
6210 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6211 | } | |
6212 | } | |
6213 | ||
6214 | /* | |
6215 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6216 | * are onlined. | |
6217 | */ | |
6218 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6219 | unsigned long action, | |
6220 | void *hcpu) | |
6221 | { | |
6222 | int cpu = (long)hcpu; | |
6223 | ||
6224 | switch (action & ~CPU_TASKS_FROZEN) { | |
6225 | case CPU_ONLINE: | |
6226 | sched_domains_numa_masks_set(cpu); | |
6227 | break; | |
6228 | ||
6229 | case CPU_DEAD: | |
6230 | sched_domains_numa_masks_clear(cpu); | |
6231 | break; | |
6232 | ||
6233 | default: | |
6234 | return NOTIFY_DONE; | |
6235 | } | |
6236 | ||
6237 | return NOTIFY_OK; | |
cb83b629 PZ |
6238 | } |
6239 | #else | |
6240 | static inline void sched_init_numa(void) | |
6241 | { | |
6242 | } | |
301a5cba TC |
6243 | |
6244 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6245 | unsigned long action, | |
6246 | void *hcpu) | |
6247 | { | |
6248 | return 0; | |
6249 | } | |
cb83b629 PZ |
6250 | #endif /* CONFIG_NUMA */ |
6251 | ||
54ab4ff4 PZ |
6252 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6253 | { | |
6254 | struct sched_domain_topology_level *tl; | |
6255 | int j; | |
6256 | ||
27723a68 | 6257 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6258 | struct sd_data *sdd = &tl->data; |
6259 | ||
6260 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6261 | if (!sdd->sd) | |
6262 | return -ENOMEM; | |
6263 | ||
6264 | sdd->sg = alloc_percpu(struct sched_group *); | |
6265 | if (!sdd->sg) | |
6266 | return -ENOMEM; | |
6267 | ||
9c3f75cb PZ |
6268 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6269 | if (!sdd->sgp) | |
6270 | return -ENOMEM; | |
6271 | ||
54ab4ff4 PZ |
6272 | for_each_cpu(j, cpu_map) { |
6273 | struct sched_domain *sd; | |
6274 | struct sched_group *sg; | |
9c3f75cb | 6275 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6276 | |
6277 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6278 | GFP_KERNEL, cpu_to_node(j)); | |
6279 | if (!sd) | |
6280 | return -ENOMEM; | |
6281 | ||
6282 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6283 | ||
6284 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6285 | GFP_KERNEL, cpu_to_node(j)); | |
6286 | if (!sg) | |
6287 | return -ENOMEM; | |
6288 | ||
30b4e9eb IM |
6289 | sg->next = sg; |
6290 | ||
54ab4ff4 | 6291 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6292 | |
c1174876 | 6293 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6294 | GFP_KERNEL, cpu_to_node(j)); |
6295 | if (!sgp) | |
6296 | return -ENOMEM; | |
6297 | ||
6298 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6299 | } |
6300 | } | |
6301 | ||
6302 | return 0; | |
6303 | } | |
6304 | ||
6305 | static void __sdt_free(const struct cpumask *cpu_map) | |
6306 | { | |
6307 | struct sched_domain_topology_level *tl; | |
6308 | int j; | |
6309 | ||
27723a68 | 6310 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6311 | struct sd_data *sdd = &tl->data; |
6312 | ||
6313 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6314 | struct sched_domain *sd; |
6315 | ||
6316 | if (sdd->sd) { | |
6317 | sd = *per_cpu_ptr(sdd->sd, j); | |
6318 | if (sd && (sd->flags & SD_OVERLAP)) | |
6319 | free_sched_groups(sd->groups, 0); | |
6320 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6321 | } | |
6322 | ||
6323 | if (sdd->sg) | |
6324 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6325 | if (sdd->sgp) | |
6326 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6327 | } |
6328 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6329 | sdd->sd = NULL; |
54ab4ff4 | 6330 | free_percpu(sdd->sg); |
fb2cf2c6 | 6331 | sdd->sg = NULL; |
9c3f75cb | 6332 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6333 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6334 | } |
6335 | } | |
6336 | ||
2c402dc3 | 6337 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6338 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6339 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6340 | { |
54ab4ff4 | 6341 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6342 | if (!sd) |
d069b916 | 6343 | return child; |
2c402dc3 | 6344 | |
2c402dc3 | 6345 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6346 | if (child) { |
6347 | sd->level = child->level + 1; | |
6348 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6349 | child->parent = sd; |
c75e0128 | 6350 | sd->child = child; |
60495e77 | 6351 | } |
a841f8ce | 6352 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6353 | |
6354 | return sd; | |
6355 | } | |
6356 | ||
2109b99e AH |
6357 | /* |
6358 | * Build sched domains for a given set of cpus and attach the sched domains | |
6359 | * to the individual cpus | |
6360 | */ | |
dce840a0 PZ |
6361 | static int build_sched_domains(const struct cpumask *cpu_map, |
6362 | struct sched_domain_attr *attr) | |
2109b99e | 6363 | { |
1c632169 | 6364 | enum s_alloc alloc_state; |
dce840a0 | 6365 | struct sched_domain *sd; |
2109b99e | 6366 | struct s_data d; |
822ff793 | 6367 | int i, ret = -ENOMEM; |
9c1cfda2 | 6368 | |
2109b99e AH |
6369 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6370 | if (alloc_state != sa_rootdomain) | |
6371 | goto error; | |
9c1cfda2 | 6372 | |
dce840a0 | 6373 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6374 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6375 | struct sched_domain_topology_level *tl; |
6376 | ||
3bd65a80 | 6377 | sd = NULL; |
27723a68 | 6378 | for_each_sd_topology(tl) { |
4a850cbe | 6379 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6380 | if (tl == sched_domain_topology) |
6381 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6382 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6383 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6384 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6385 | break; | |
e3589f6c | 6386 | } |
dce840a0 PZ |
6387 | } |
6388 | ||
6389 | /* Build the groups for the domains */ | |
6390 | for_each_cpu(i, cpu_map) { | |
6391 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6392 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6393 | if (sd->flags & SD_OVERLAP) { |
6394 | if (build_overlap_sched_groups(sd, i)) | |
6395 | goto error; | |
6396 | } else { | |
6397 | if (build_sched_groups(sd, i)) | |
6398 | goto error; | |
6399 | } | |
1cf51902 | 6400 | } |
a06dadbe | 6401 | } |
9c1cfda2 | 6402 | |
1da177e4 | 6403 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6404 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6405 | if (!cpumask_test_cpu(i, cpu_map)) | |
6406 | continue; | |
9c1cfda2 | 6407 | |
dce840a0 PZ |
6408 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6409 | claim_allocations(i, sd); | |
cd4ea6ae | 6410 | init_sched_groups_power(i, sd); |
dce840a0 | 6411 | } |
f712c0c7 | 6412 | } |
9c1cfda2 | 6413 | |
1da177e4 | 6414 | /* Attach the domains */ |
dce840a0 | 6415 | rcu_read_lock(); |
abcd083a | 6416 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6417 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6418 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6419 | } |
dce840a0 | 6420 | rcu_read_unlock(); |
51888ca2 | 6421 | |
822ff793 | 6422 | ret = 0; |
51888ca2 | 6423 | error: |
2109b99e | 6424 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6425 | return ret; |
1da177e4 | 6426 | } |
029190c5 | 6427 | |
acc3f5d7 | 6428 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6429 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6430 | static struct sched_domain_attr *dattr_cur; |
6431 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6432 | |
6433 | /* | |
6434 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6435 | * cpumask) fails, then fallback to a single sched domain, |
6436 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6437 | */ |
4212823f | 6438 | static cpumask_var_t fallback_doms; |
029190c5 | 6439 | |
ee79d1bd HC |
6440 | /* |
6441 | * arch_update_cpu_topology lets virtualized architectures update the | |
6442 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6443 | * or 0 if it stayed the same. | |
6444 | */ | |
6445 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6446 | { |
ee79d1bd | 6447 | return 0; |
22e52b07 HC |
6448 | } |
6449 | ||
acc3f5d7 RR |
6450 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6451 | { | |
6452 | int i; | |
6453 | cpumask_var_t *doms; | |
6454 | ||
6455 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6456 | if (!doms) | |
6457 | return NULL; | |
6458 | for (i = 0; i < ndoms; i++) { | |
6459 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6460 | free_sched_domains(doms, i); | |
6461 | return NULL; | |
6462 | } | |
6463 | } | |
6464 | return doms; | |
6465 | } | |
6466 | ||
6467 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6468 | { | |
6469 | unsigned int i; | |
6470 | for (i = 0; i < ndoms; i++) | |
6471 | free_cpumask_var(doms[i]); | |
6472 | kfree(doms); | |
6473 | } | |
6474 | ||
1a20ff27 | 6475 | /* |
41a2d6cf | 6476 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6477 | * For now this just excludes isolated cpus, but could be used to |
6478 | * exclude other special cases in the future. | |
1a20ff27 | 6479 | */ |
c4a8849a | 6480 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6481 | { |
7378547f MM |
6482 | int err; |
6483 | ||
22e52b07 | 6484 | arch_update_cpu_topology(); |
029190c5 | 6485 | ndoms_cur = 1; |
acc3f5d7 | 6486 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6487 | if (!doms_cur) |
acc3f5d7 RR |
6488 | doms_cur = &fallback_doms; |
6489 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6490 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6491 | register_sched_domain_sysctl(); |
7378547f MM |
6492 | |
6493 | return err; | |
1a20ff27 DG |
6494 | } |
6495 | ||
1a20ff27 DG |
6496 | /* |
6497 | * Detach sched domains from a group of cpus specified in cpu_map | |
6498 | * These cpus will now be attached to the NULL domain | |
6499 | */ | |
96f874e2 | 6500 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6501 | { |
6502 | int i; | |
6503 | ||
dce840a0 | 6504 | rcu_read_lock(); |
abcd083a | 6505 | for_each_cpu(i, cpu_map) |
57d885fe | 6506 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6507 | rcu_read_unlock(); |
1a20ff27 DG |
6508 | } |
6509 | ||
1d3504fc HS |
6510 | /* handle null as "default" */ |
6511 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6512 | struct sched_domain_attr *new, int idx_new) | |
6513 | { | |
6514 | struct sched_domain_attr tmp; | |
6515 | ||
6516 | /* fast path */ | |
6517 | if (!new && !cur) | |
6518 | return 1; | |
6519 | ||
6520 | tmp = SD_ATTR_INIT; | |
6521 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6522 | new ? (new + idx_new) : &tmp, | |
6523 | sizeof(struct sched_domain_attr)); | |
6524 | } | |
6525 | ||
029190c5 PJ |
6526 | /* |
6527 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6528 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6529 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6530 | * It destroys each deleted domain and builds each new domain. | |
6531 | * | |
acc3f5d7 | 6532 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6533 | * The masks don't intersect (don't overlap.) We should setup one |
6534 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6535 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6536 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6537 | * it as it is. | |
6538 | * | |
acc3f5d7 RR |
6539 | * The passed in 'doms_new' should be allocated using |
6540 | * alloc_sched_domains. This routine takes ownership of it and will | |
6541 | * free_sched_domains it when done with it. If the caller failed the | |
6542 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6543 | * and partition_sched_domains() will fallback to the single partition | |
6544 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6545 | * |
96f874e2 | 6546 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6547 | * ndoms_new == 0 is a special case for destroying existing domains, |
6548 | * and it will not create the default domain. | |
dfb512ec | 6549 | * |
029190c5 PJ |
6550 | * Call with hotplug lock held |
6551 | */ | |
acc3f5d7 | 6552 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6553 | struct sched_domain_attr *dattr_new) |
029190c5 | 6554 | { |
dfb512ec | 6555 | int i, j, n; |
d65bd5ec | 6556 | int new_topology; |
029190c5 | 6557 | |
712555ee | 6558 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6559 | |
7378547f MM |
6560 | /* always unregister in case we don't destroy any domains */ |
6561 | unregister_sched_domain_sysctl(); | |
6562 | ||
d65bd5ec HC |
6563 | /* Let architecture update cpu core mappings. */ |
6564 | new_topology = arch_update_cpu_topology(); | |
6565 | ||
dfb512ec | 6566 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6567 | |
6568 | /* Destroy deleted domains */ | |
6569 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6570 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6571 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6572 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6573 | goto match1; |
6574 | } | |
6575 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6576 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6577 | match1: |
6578 | ; | |
6579 | } | |
6580 | ||
c8d2d47a | 6581 | n = ndoms_cur; |
e761b772 | 6582 | if (doms_new == NULL) { |
c8d2d47a | 6583 | n = 0; |
acc3f5d7 | 6584 | doms_new = &fallback_doms; |
6ad4c188 | 6585 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6586 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6587 | } |
6588 | ||
029190c5 PJ |
6589 | /* Build new domains */ |
6590 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6591 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6592 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6593 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6594 | goto match2; |
6595 | } | |
6596 | /* no match - add a new doms_new */ | |
dce840a0 | 6597 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6598 | match2: |
6599 | ; | |
6600 | } | |
6601 | ||
6602 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6603 | if (doms_cur != &fallback_doms) |
6604 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6605 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6606 | doms_cur = doms_new; |
1d3504fc | 6607 | dattr_cur = dattr_new; |
029190c5 | 6608 | ndoms_cur = ndoms_new; |
7378547f MM |
6609 | |
6610 | register_sched_domain_sysctl(); | |
a1835615 | 6611 | |
712555ee | 6612 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6613 | } |
6614 | ||
d35be8ba SB |
6615 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6616 | ||
1da177e4 | 6617 | /* |
3a101d05 TH |
6618 | * Update cpusets according to cpu_active mask. If cpusets are |
6619 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6620 | * around partition_sched_domains(). | |
d35be8ba SB |
6621 | * |
6622 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6623 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6624 | */ |
0b2e918a TH |
6625 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6626 | void *hcpu) | |
e761b772 | 6627 | { |
d35be8ba SB |
6628 | switch (action) { |
6629 | case CPU_ONLINE_FROZEN: | |
6630 | case CPU_DOWN_FAILED_FROZEN: | |
6631 | ||
6632 | /* | |
6633 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6634 | * resume sequence. As long as this is not the last online | |
6635 | * operation in the resume sequence, just build a single sched | |
6636 | * domain, ignoring cpusets. | |
6637 | */ | |
6638 | num_cpus_frozen--; | |
6639 | if (likely(num_cpus_frozen)) { | |
6640 | partition_sched_domains(1, NULL, NULL); | |
6641 | break; | |
6642 | } | |
6643 | ||
6644 | /* | |
6645 | * This is the last CPU online operation. So fall through and | |
6646 | * restore the original sched domains by considering the | |
6647 | * cpuset configurations. | |
6648 | */ | |
6649 | ||
e761b772 | 6650 | case CPU_ONLINE: |
6ad4c188 | 6651 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6652 | cpuset_update_active_cpus(true); |
d35be8ba | 6653 | break; |
3a101d05 TH |
6654 | default: |
6655 | return NOTIFY_DONE; | |
6656 | } | |
d35be8ba | 6657 | return NOTIFY_OK; |
3a101d05 | 6658 | } |
e761b772 | 6659 | |
0b2e918a TH |
6660 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6661 | void *hcpu) | |
3a101d05 | 6662 | { |
d35be8ba | 6663 | switch (action) { |
3a101d05 | 6664 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6665 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6666 | break; |
6667 | case CPU_DOWN_PREPARE_FROZEN: | |
6668 | num_cpus_frozen++; | |
6669 | partition_sched_domains(1, NULL, NULL); | |
6670 | break; | |
e761b772 MK |
6671 | default: |
6672 | return NOTIFY_DONE; | |
6673 | } | |
d35be8ba | 6674 | return NOTIFY_OK; |
e761b772 | 6675 | } |
e761b772 | 6676 | |
1da177e4 LT |
6677 | void __init sched_init_smp(void) |
6678 | { | |
dcc30a35 RR |
6679 | cpumask_var_t non_isolated_cpus; |
6680 | ||
6681 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6682 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6683 | |
cb83b629 PZ |
6684 | sched_init_numa(); |
6685 | ||
6acce3ef PZ |
6686 | /* |
6687 | * There's no userspace yet to cause hotplug operations; hence all the | |
6688 | * cpu masks are stable and all blatant races in the below code cannot | |
6689 | * happen. | |
6690 | */ | |
712555ee | 6691 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6692 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6693 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6694 | if (cpumask_empty(non_isolated_cpus)) | |
6695 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6696 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6697 | |
301a5cba | 6698 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6699 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6700 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6701 | |
b328ca18 | 6702 | init_hrtick(); |
5c1e1767 NP |
6703 | |
6704 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6705 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6706 | BUG(); |
19978ca6 | 6707 | sched_init_granularity(); |
dcc30a35 | 6708 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6709 | |
0e3900e6 | 6710 | init_sched_rt_class(); |
1baca4ce | 6711 | init_sched_dl_class(); |
1da177e4 LT |
6712 | } |
6713 | #else | |
6714 | void __init sched_init_smp(void) | |
6715 | { | |
19978ca6 | 6716 | sched_init_granularity(); |
1da177e4 LT |
6717 | } |
6718 | #endif /* CONFIG_SMP */ | |
6719 | ||
cd1bb94b AB |
6720 | const_debug unsigned int sysctl_timer_migration = 1; |
6721 | ||
1da177e4 LT |
6722 | int in_sched_functions(unsigned long addr) |
6723 | { | |
1da177e4 LT |
6724 | return in_lock_functions(addr) || |
6725 | (addr >= (unsigned long)__sched_text_start | |
6726 | && addr < (unsigned long)__sched_text_end); | |
6727 | } | |
6728 | ||
029632fb | 6729 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6730 | /* |
6731 | * Default task group. | |
6732 | * Every task in system belongs to this group at bootup. | |
6733 | */ | |
029632fb | 6734 | struct task_group root_task_group; |
35cf4e50 | 6735 | LIST_HEAD(task_groups); |
052f1dc7 | 6736 | #endif |
6f505b16 | 6737 | |
e6252c3e | 6738 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6739 | |
1da177e4 LT |
6740 | void __init sched_init(void) |
6741 | { | |
dd41f596 | 6742 | int i, j; |
434d53b0 MT |
6743 | unsigned long alloc_size = 0, ptr; |
6744 | ||
6745 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6746 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6747 | #endif | |
6748 | #ifdef CONFIG_RT_GROUP_SCHED | |
6749 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6750 | #endif |
df7c8e84 | 6751 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6752 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6753 | #endif |
434d53b0 | 6754 | if (alloc_size) { |
36b7b6d4 | 6755 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6756 | |
6757 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6758 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6759 | ptr += nr_cpu_ids * sizeof(void **); |
6760 | ||
07e06b01 | 6761 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6762 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6763 | |
6d6bc0ad | 6764 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6765 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6766 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6767 | ptr += nr_cpu_ids * sizeof(void **); |
6768 | ||
07e06b01 | 6769 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6770 | ptr += nr_cpu_ids * sizeof(void **); |
6771 | ||
6d6bc0ad | 6772 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6773 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6774 | for_each_possible_cpu(i) { | |
e6252c3e | 6775 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6776 | ptr += cpumask_size(); |
6777 | } | |
6778 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6779 | } |
dd41f596 | 6780 | |
332ac17e DF |
6781 | init_rt_bandwidth(&def_rt_bandwidth, |
6782 | global_rt_period(), global_rt_runtime()); | |
6783 | init_dl_bandwidth(&def_dl_bandwidth, | |
6784 | global_dl_period(), global_dl_runtime()); | |
6785 | ||
57d885fe GH |
6786 | #ifdef CONFIG_SMP |
6787 | init_defrootdomain(); | |
6788 | #endif | |
6789 | ||
d0b27fa7 | 6790 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6791 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6792 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6793 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6794 | |
7c941438 | 6795 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6796 | list_add(&root_task_group.list, &task_groups); |
6797 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6798 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6799 | autogroup_init(&init_task); |
54c707e9 | 6800 | |
7c941438 | 6801 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6802 | |
0a945022 | 6803 | for_each_possible_cpu(i) { |
70b97a7f | 6804 | struct rq *rq; |
1da177e4 LT |
6805 | |
6806 | rq = cpu_rq(i); | |
05fa785c | 6807 | raw_spin_lock_init(&rq->lock); |
7897986b | 6808 | rq->nr_running = 0; |
dce48a84 TG |
6809 | rq->calc_load_active = 0; |
6810 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6811 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6812 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6813 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6814 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6815 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6816 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6817 | /* |
07e06b01 | 6818 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6819 | * |
6820 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6821 | * gets 100% of the cpu resources in the system. This overall | |
6822 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6823 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6824 | * based on each entity's (task or task-group's) weight |
6825 | * (se->load.weight). | |
6826 | * | |
07e06b01 | 6827 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6828 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6829 | * then A0's share of the cpu resource is: | |
6830 | * | |
0d905bca | 6831 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6832 | * |
07e06b01 YZ |
6833 | * We achieve this by letting root_task_group's tasks sit |
6834 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6835 | */ |
ab84d31e | 6836 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6837 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6838 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6839 | ||
6840 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6841 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6842 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 6843 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6844 | #endif |
1da177e4 | 6845 | |
dd41f596 IM |
6846 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6847 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6848 | |
6849 | rq->last_load_update_tick = jiffies; | |
6850 | ||
1da177e4 | 6851 | #ifdef CONFIG_SMP |
41c7ce9a | 6852 | rq->sd = NULL; |
57d885fe | 6853 | rq->rd = NULL; |
1399fa78 | 6854 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6855 | rq->post_schedule = 0; |
1da177e4 | 6856 | rq->active_balance = 0; |
dd41f596 | 6857 | rq->next_balance = jiffies; |
1da177e4 | 6858 | rq->push_cpu = 0; |
0a2966b4 | 6859 | rq->cpu = i; |
1f11eb6a | 6860 | rq->online = 0; |
eae0c9df MG |
6861 | rq->idle_stamp = 0; |
6862 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6863 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6864 | |
6865 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6866 | ||
dc938520 | 6867 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6868 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6869 | rq->nohz_flags = 0; |
83cd4fe2 | 6870 | #endif |
265f22a9 FW |
6871 | #ifdef CONFIG_NO_HZ_FULL |
6872 | rq->last_sched_tick = 0; | |
6873 | #endif | |
1da177e4 | 6874 | #endif |
8f4d37ec | 6875 | init_rq_hrtick(rq); |
1da177e4 | 6876 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6877 | } |
6878 | ||
2dd73a4f | 6879 | set_load_weight(&init_task); |
b50f60ce | 6880 | |
e107be36 AK |
6881 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6882 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6883 | #endif | |
6884 | ||
1da177e4 LT |
6885 | /* |
6886 | * The boot idle thread does lazy MMU switching as well: | |
6887 | */ | |
6888 | atomic_inc(&init_mm.mm_count); | |
6889 | enter_lazy_tlb(&init_mm, current); | |
6890 | ||
6891 | /* | |
6892 | * Make us the idle thread. Technically, schedule() should not be | |
6893 | * called from this thread, however somewhere below it might be, | |
6894 | * but because we are the idle thread, we just pick up running again | |
6895 | * when this runqueue becomes "idle". | |
6896 | */ | |
6897 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6898 | |
6899 | calc_load_update = jiffies + LOAD_FREQ; | |
6900 | ||
dd41f596 IM |
6901 | /* |
6902 | * During early bootup we pretend to be a normal task: | |
6903 | */ | |
6904 | current->sched_class = &fair_sched_class; | |
6892b75e | 6905 | |
bf4d83f6 | 6906 | #ifdef CONFIG_SMP |
4cb98839 | 6907 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6908 | /* May be allocated at isolcpus cmdline parse time */ |
6909 | if (cpu_isolated_map == NULL) | |
6910 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6911 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6912 | #endif |
6913 | init_sched_fair_class(); | |
6a7b3dc3 | 6914 | |
6892b75e | 6915 | scheduler_running = 1; |
1da177e4 LT |
6916 | } |
6917 | ||
d902db1e | 6918 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6919 | static inline int preempt_count_equals(int preempt_offset) |
6920 | { | |
234da7bc | 6921 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6922 | |
4ba8216c | 6923 | return (nested == preempt_offset); |
e4aafea2 FW |
6924 | } |
6925 | ||
d894837f | 6926 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6927 | { |
1da177e4 LT |
6928 | static unsigned long prev_jiffy; /* ratelimiting */ |
6929 | ||
b3fbab05 | 6930 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
6931 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
6932 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
6933 | return; |
6934 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6935 | return; | |
6936 | prev_jiffy = jiffies; | |
6937 | ||
3df0fc5b PZ |
6938 | printk(KERN_ERR |
6939 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6940 | file, line); | |
6941 | printk(KERN_ERR | |
6942 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6943 | in_atomic(), irqs_disabled(), | |
6944 | current->pid, current->comm); | |
aef745fc IM |
6945 | |
6946 | debug_show_held_locks(current); | |
6947 | if (irqs_disabled()) | |
6948 | print_irqtrace_events(current); | |
6949 | dump_stack(); | |
1da177e4 LT |
6950 | } |
6951 | EXPORT_SYMBOL(__might_sleep); | |
6952 | #endif | |
6953 | ||
6954 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6955 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6956 | { | |
da7a735e | 6957 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
6958 | struct sched_attr attr = { |
6959 | .sched_policy = SCHED_NORMAL, | |
6960 | }; | |
da7a735e | 6961 | int old_prio = p->prio; |
3a5e4dc1 | 6962 | int on_rq; |
3e51f33f | 6963 | |
fd2f4419 | 6964 | on_rq = p->on_rq; |
3a5e4dc1 | 6965 | if (on_rq) |
4ca9b72b | 6966 | dequeue_task(rq, p, 0); |
d50dde5a | 6967 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 6968 | if (on_rq) { |
4ca9b72b | 6969 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
6970 | resched_task(rq->curr); |
6971 | } | |
da7a735e PZ |
6972 | |
6973 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
6974 | } |
6975 | ||
1da177e4 LT |
6976 | void normalize_rt_tasks(void) |
6977 | { | |
a0f98a1c | 6978 | struct task_struct *g, *p; |
1da177e4 | 6979 | unsigned long flags; |
70b97a7f | 6980 | struct rq *rq; |
1da177e4 | 6981 | |
4cf5d77a | 6982 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 6983 | do_each_thread(g, p) { |
178be793 IM |
6984 | /* |
6985 | * Only normalize user tasks: | |
6986 | */ | |
6987 | if (!p->mm) | |
6988 | continue; | |
6989 | ||
6cfb0d5d | 6990 | p->se.exec_start = 0; |
6cfb0d5d | 6991 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
6992 | p->se.statistics.wait_start = 0; |
6993 | p->se.statistics.sleep_start = 0; | |
6994 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 6995 | #endif |
dd41f596 | 6996 | |
aab03e05 | 6997 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6998 | /* |
6999 | * Renice negative nice level userspace | |
7000 | * tasks back to 0: | |
7001 | */ | |
7002 | if (TASK_NICE(p) < 0 && p->mm) | |
7003 | set_user_nice(p, 0); | |
1da177e4 | 7004 | continue; |
dd41f596 | 7005 | } |
1da177e4 | 7006 | |
1d615482 | 7007 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7008 | rq = __task_rq_lock(p); |
1da177e4 | 7009 | |
178be793 | 7010 | normalize_task(rq, p); |
3a5e4dc1 | 7011 | |
b29739f9 | 7012 | __task_rq_unlock(rq); |
1d615482 | 7013 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7014 | } while_each_thread(g, p); |
7015 | ||
4cf5d77a | 7016 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7017 | } |
7018 | ||
7019 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7020 | |
67fc4e0c | 7021 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7022 | /* |
67fc4e0c | 7023 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7024 | * |
7025 | * They can only be called when the whole system has been | |
7026 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7027 | * activity can take place. Using them for anything else would | |
7028 | * be a serious bug, and as a result, they aren't even visible | |
7029 | * under any other configuration. | |
7030 | */ | |
7031 | ||
7032 | /** | |
7033 | * curr_task - return the current task for a given cpu. | |
7034 | * @cpu: the processor in question. | |
7035 | * | |
7036 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7037 | * |
7038 | * Return: The current task for @cpu. | |
1df5c10a | 7039 | */ |
36c8b586 | 7040 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7041 | { |
7042 | return cpu_curr(cpu); | |
7043 | } | |
7044 | ||
67fc4e0c JW |
7045 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7046 | ||
7047 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7048 | /** |
7049 | * set_curr_task - set the current task for a given cpu. | |
7050 | * @cpu: the processor in question. | |
7051 | * @p: the task pointer to set. | |
7052 | * | |
7053 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7054 | * are serviced on a separate stack. It allows the architecture to switch the |
7055 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7056 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7057 | * and caller must save the original value of the current task (see | |
7058 | * curr_task() above) and restore that value before reenabling interrupts and | |
7059 | * re-starting the system. | |
7060 | * | |
7061 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7062 | */ | |
36c8b586 | 7063 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7064 | { |
7065 | cpu_curr(cpu) = p; | |
7066 | } | |
7067 | ||
7068 | #endif | |
29f59db3 | 7069 | |
7c941438 | 7070 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7071 | /* task_group_lock serializes the addition/removal of task groups */ |
7072 | static DEFINE_SPINLOCK(task_group_lock); | |
7073 | ||
bccbe08a PZ |
7074 | static void free_sched_group(struct task_group *tg) |
7075 | { | |
7076 | free_fair_sched_group(tg); | |
7077 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7078 | autogroup_free(tg); |
bccbe08a PZ |
7079 | kfree(tg); |
7080 | } | |
7081 | ||
7082 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7083 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7084 | { |
7085 | struct task_group *tg; | |
bccbe08a PZ |
7086 | |
7087 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7088 | if (!tg) | |
7089 | return ERR_PTR(-ENOMEM); | |
7090 | ||
ec7dc8ac | 7091 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7092 | goto err; |
7093 | ||
ec7dc8ac | 7094 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7095 | goto err; |
7096 | ||
ace783b9 LZ |
7097 | return tg; |
7098 | ||
7099 | err: | |
7100 | free_sched_group(tg); | |
7101 | return ERR_PTR(-ENOMEM); | |
7102 | } | |
7103 | ||
7104 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7105 | { | |
7106 | unsigned long flags; | |
7107 | ||
8ed36996 | 7108 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7109 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7110 | |
7111 | WARN_ON(!parent); /* root should already exist */ | |
7112 | ||
7113 | tg->parent = parent; | |
f473aa5e | 7114 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7115 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7116 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7117 | } |
7118 | ||
9b5b7751 | 7119 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7120 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7121 | { |
29f59db3 | 7122 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7123 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7124 | } |
7125 | ||
9b5b7751 | 7126 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7127 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7128 | { |
7129 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7130 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7131 | } | |
7132 | ||
7133 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7134 | { |
8ed36996 | 7135 | unsigned long flags; |
9b5b7751 | 7136 | int i; |
29f59db3 | 7137 | |
3d4b47b4 PZ |
7138 | /* end participation in shares distribution */ |
7139 | for_each_possible_cpu(i) | |
bccbe08a | 7140 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7141 | |
7142 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7143 | list_del_rcu(&tg->list); |
f473aa5e | 7144 | list_del_rcu(&tg->siblings); |
8ed36996 | 7145 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7146 | } |
7147 | ||
9b5b7751 | 7148 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7149 | * The caller of this function should have put the task in its new group |
7150 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7151 | * reflect its new group. | |
9b5b7751 SV |
7152 | */ |
7153 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7154 | { |
8323f26c | 7155 | struct task_group *tg; |
29f59db3 SV |
7156 | int on_rq, running; |
7157 | unsigned long flags; | |
7158 | struct rq *rq; | |
7159 | ||
7160 | rq = task_rq_lock(tsk, &flags); | |
7161 | ||
051a1d1a | 7162 | running = task_current(rq, tsk); |
fd2f4419 | 7163 | on_rq = tsk->on_rq; |
29f59db3 | 7164 | |
0e1f3483 | 7165 | if (on_rq) |
29f59db3 | 7166 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7167 | if (unlikely(running)) |
7168 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7169 | |
8af01f56 | 7170 | tg = container_of(task_css_check(tsk, cpu_cgroup_subsys_id, |
8323f26c PZ |
7171 | lockdep_is_held(&tsk->sighand->siglock)), |
7172 | struct task_group, css); | |
7173 | tg = autogroup_task_group(tsk, tg); | |
7174 | tsk->sched_task_group = tg; | |
7175 | ||
810b3817 | 7176 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7177 | if (tsk->sched_class->task_move_group) |
7178 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7179 | else | |
810b3817 | 7180 | #endif |
b2b5ce02 | 7181 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7182 | |
0e1f3483 HS |
7183 | if (unlikely(running)) |
7184 | tsk->sched_class->set_curr_task(rq); | |
7185 | if (on_rq) | |
371fd7e7 | 7186 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7187 | |
0122ec5b | 7188 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7189 | } |
7c941438 | 7190 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7191 | |
a790de99 PT |
7192 | #ifdef CONFIG_RT_GROUP_SCHED |
7193 | /* | |
7194 | * Ensure that the real time constraints are schedulable. | |
7195 | */ | |
7196 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7197 | |
9a7e0b18 PZ |
7198 | /* Must be called with tasklist_lock held */ |
7199 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7200 | { |
9a7e0b18 | 7201 | struct task_struct *g, *p; |
b40b2e8e | 7202 | |
9a7e0b18 | 7203 | do_each_thread(g, p) { |
029632fb | 7204 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7205 | return 1; |
7206 | } while_each_thread(g, p); | |
b40b2e8e | 7207 | |
9a7e0b18 PZ |
7208 | return 0; |
7209 | } | |
b40b2e8e | 7210 | |
9a7e0b18 PZ |
7211 | struct rt_schedulable_data { |
7212 | struct task_group *tg; | |
7213 | u64 rt_period; | |
7214 | u64 rt_runtime; | |
7215 | }; | |
b40b2e8e | 7216 | |
a790de99 | 7217 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7218 | { |
7219 | struct rt_schedulable_data *d = data; | |
7220 | struct task_group *child; | |
7221 | unsigned long total, sum = 0; | |
7222 | u64 period, runtime; | |
b40b2e8e | 7223 | |
9a7e0b18 PZ |
7224 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7225 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7226 | |
9a7e0b18 PZ |
7227 | if (tg == d->tg) { |
7228 | period = d->rt_period; | |
7229 | runtime = d->rt_runtime; | |
b40b2e8e | 7230 | } |
b40b2e8e | 7231 | |
4653f803 PZ |
7232 | /* |
7233 | * Cannot have more runtime than the period. | |
7234 | */ | |
7235 | if (runtime > period && runtime != RUNTIME_INF) | |
7236 | return -EINVAL; | |
6f505b16 | 7237 | |
4653f803 PZ |
7238 | /* |
7239 | * Ensure we don't starve existing RT tasks. | |
7240 | */ | |
9a7e0b18 PZ |
7241 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7242 | return -EBUSY; | |
6f505b16 | 7243 | |
9a7e0b18 | 7244 | total = to_ratio(period, runtime); |
6f505b16 | 7245 | |
4653f803 PZ |
7246 | /* |
7247 | * Nobody can have more than the global setting allows. | |
7248 | */ | |
7249 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7250 | return -EINVAL; | |
6f505b16 | 7251 | |
4653f803 PZ |
7252 | /* |
7253 | * The sum of our children's runtime should not exceed our own. | |
7254 | */ | |
9a7e0b18 PZ |
7255 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7256 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7257 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7258 | |
9a7e0b18 PZ |
7259 | if (child == d->tg) { |
7260 | period = d->rt_period; | |
7261 | runtime = d->rt_runtime; | |
7262 | } | |
6f505b16 | 7263 | |
9a7e0b18 | 7264 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7265 | } |
6f505b16 | 7266 | |
9a7e0b18 PZ |
7267 | if (sum > total) |
7268 | return -EINVAL; | |
7269 | ||
7270 | return 0; | |
6f505b16 PZ |
7271 | } |
7272 | ||
9a7e0b18 | 7273 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7274 | { |
8277434e PT |
7275 | int ret; |
7276 | ||
9a7e0b18 PZ |
7277 | struct rt_schedulable_data data = { |
7278 | .tg = tg, | |
7279 | .rt_period = period, | |
7280 | .rt_runtime = runtime, | |
7281 | }; | |
7282 | ||
8277434e PT |
7283 | rcu_read_lock(); |
7284 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7285 | rcu_read_unlock(); | |
7286 | ||
7287 | return ret; | |
521f1a24 DG |
7288 | } |
7289 | ||
ab84d31e | 7290 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7291 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7292 | { |
ac086bc2 | 7293 | int i, err = 0; |
9f0c1e56 | 7294 | |
9f0c1e56 | 7295 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7296 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7297 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7298 | if (err) | |
9f0c1e56 | 7299 | goto unlock; |
ac086bc2 | 7300 | |
0986b11b | 7301 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7302 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7303 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7304 | |
7305 | for_each_possible_cpu(i) { | |
7306 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7307 | ||
0986b11b | 7308 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7309 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7310 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7311 | } |
0986b11b | 7312 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7313 | unlock: |
521f1a24 | 7314 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7315 | mutex_unlock(&rt_constraints_mutex); |
7316 | ||
7317 | return err; | |
6f505b16 PZ |
7318 | } |
7319 | ||
25cc7da7 | 7320 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7321 | { |
7322 | u64 rt_runtime, rt_period; | |
7323 | ||
7324 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7325 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7326 | if (rt_runtime_us < 0) | |
7327 | rt_runtime = RUNTIME_INF; | |
7328 | ||
ab84d31e | 7329 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7330 | } |
7331 | ||
25cc7da7 | 7332 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7333 | { |
7334 | u64 rt_runtime_us; | |
7335 | ||
d0b27fa7 | 7336 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7337 | return -1; |
7338 | ||
d0b27fa7 | 7339 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7340 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7341 | return rt_runtime_us; | |
7342 | } | |
d0b27fa7 | 7343 | |
25cc7da7 | 7344 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7345 | { |
7346 | u64 rt_runtime, rt_period; | |
7347 | ||
7348 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7349 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7350 | ||
619b0488 R |
7351 | if (rt_period == 0) |
7352 | return -EINVAL; | |
7353 | ||
ab84d31e | 7354 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7355 | } |
7356 | ||
25cc7da7 | 7357 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7358 | { |
7359 | u64 rt_period_us; | |
7360 | ||
7361 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7362 | do_div(rt_period_us, NSEC_PER_USEC); | |
7363 | return rt_period_us; | |
7364 | } | |
332ac17e | 7365 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7366 | |
332ac17e DF |
7367 | /* |
7368 | * Coupling of -rt and -deadline bandwidth. | |
7369 | * | |
7370 | * Here we check if the new -rt bandwidth value is consistent | |
7371 | * with the system settings for the bandwidth available | |
7372 | * to -deadline tasks. | |
7373 | * | |
7374 | * IOW, we want to enforce that | |
7375 | * | |
7376 | * rt_bandwidth + dl_bandwidth <= 100% | |
7377 | * | |
7378 | * is always true. | |
7379 | */ | |
7380 | static bool __sched_rt_dl_global_constraints(u64 rt_bw) | |
7381 | { | |
7382 | unsigned long flags; | |
7383 | u64 dl_bw; | |
7384 | bool ret; | |
7385 | ||
7386 | raw_spin_lock_irqsave(&def_dl_bandwidth.dl_runtime_lock, flags); | |
7387 | if (global_rt_runtime() == RUNTIME_INF || | |
7388 | global_dl_runtime() == RUNTIME_INF) { | |
7389 | ret = true; | |
7390 | goto unlock; | |
7391 | } | |
7392 | ||
7393 | dl_bw = to_ratio(def_dl_bandwidth.dl_period, | |
7394 | def_dl_bandwidth.dl_runtime); | |
7395 | ||
7396 | ret = rt_bw + dl_bw <= to_ratio(RUNTIME_INF, RUNTIME_INF); | |
7397 | unlock: | |
7398 | raw_spin_unlock_irqrestore(&def_dl_bandwidth.dl_runtime_lock, flags); | |
7399 | ||
7400 | return ret; | |
7401 | } | |
7402 | ||
7403 | #ifdef CONFIG_RT_GROUP_SCHED | |
d0b27fa7 PZ |
7404 | static int sched_rt_global_constraints(void) |
7405 | { | |
332ac17e | 7406 | u64 runtime, period, bw; |
d0b27fa7 PZ |
7407 | int ret = 0; |
7408 | ||
ec5d4989 HS |
7409 | if (sysctl_sched_rt_period <= 0) |
7410 | return -EINVAL; | |
7411 | ||
4653f803 PZ |
7412 | runtime = global_rt_runtime(); |
7413 | period = global_rt_period(); | |
7414 | ||
7415 | /* | |
7416 | * Sanity check on the sysctl variables. | |
7417 | */ | |
7418 | if (runtime > period && runtime != RUNTIME_INF) | |
7419 | return -EINVAL; | |
10b612f4 | 7420 | |
332ac17e DF |
7421 | bw = to_ratio(period, runtime); |
7422 | if (!__sched_rt_dl_global_constraints(bw)) | |
7423 | return -EINVAL; | |
7424 | ||
d0b27fa7 | 7425 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7426 | read_lock(&tasklist_lock); |
4653f803 | 7427 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7428 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7429 | mutex_unlock(&rt_constraints_mutex); |
7430 | ||
7431 | return ret; | |
7432 | } | |
54e99124 | 7433 | |
25cc7da7 | 7434 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7435 | { |
7436 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7437 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7438 | return 0; | |
7439 | ||
7440 | return 1; | |
7441 | } | |
7442 | ||
6d6bc0ad | 7443 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7444 | static int sched_rt_global_constraints(void) |
7445 | { | |
ac086bc2 | 7446 | unsigned long flags; |
332ac17e DF |
7447 | int i, ret = 0; |
7448 | u64 bw; | |
ac086bc2 | 7449 | |
ec5d4989 HS |
7450 | if (sysctl_sched_rt_period <= 0) |
7451 | return -EINVAL; | |
7452 | ||
0986b11b | 7453 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
332ac17e DF |
7454 | bw = to_ratio(global_rt_period(), global_rt_runtime()); |
7455 | if (!__sched_rt_dl_global_constraints(bw)) { | |
7456 | ret = -EINVAL; | |
7457 | goto unlock; | |
7458 | } | |
7459 | ||
ac086bc2 PZ |
7460 | for_each_possible_cpu(i) { |
7461 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7462 | ||
0986b11b | 7463 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7464 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7465 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7466 | } |
332ac17e | 7467 | unlock: |
0986b11b | 7468 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7469 | |
332ac17e | 7470 | return ret; |
d0b27fa7 | 7471 | } |
6d6bc0ad | 7472 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7473 | |
332ac17e DF |
7474 | /* |
7475 | * Coupling of -dl and -rt bandwidth. | |
7476 | * | |
7477 | * Here we check, while setting the system wide bandwidth available | |
7478 | * for -dl tasks and groups, if the new values are consistent with | |
7479 | * the system settings for the bandwidth available to -rt entities. | |
7480 | * | |
7481 | * IOW, we want to enforce that | |
7482 | * | |
7483 | * rt_bandwidth + dl_bandwidth <= 100% | |
7484 | * | |
7485 | * is always true. | |
7486 | */ | |
7487 | static bool __sched_dl_rt_global_constraints(u64 dl_bw) | |
7488 | { | |
7489 | u64 rt_bw; | |
7490 | bool ret; | |
7491 | ||
7492 | raw_spin_lock(&def_rt_bandwidth.rt_runtime_lock); | |
7493 | if (global_dl_runtime() == RUNTIME_INF || | |
7494 | global_rt_runtime() == RUNTIME_INF) { | |
7495 | ret = true; | |
7496 | goto unlock; | |
7497 | } | |
7498 | ||
7499 | rt_bw = to_ratio(ktime_to_ns(def_rt_bandwidth.rt_period), | |
7500 | def_rt_bandwidth.rt_runtime); | |
7501 | ||
7502 | ret = rt_bw + dl_bw <= to_ratio(RUNTIME_INF, RUNTIME_INF); | |
7503 | unlock: | |
7504 | raw_spin_unlock(&def_rt_bandwidth.rt_runtime_lock); | |
7505 | ||
7506 | return ret; | |
7507 | } | |
7508 | ||
7509 | static bool __sched_dl_global_constraints(u64 runtime, u64 period) | |
7510 | { | |
7511 | if (!period || (runtime != RUNTIME_INF && runtime > period)) | |
7512 | return -EINVAL; | |
7513 | ||
7514 | return 0; | |
7515 | } | |
7516 | ||
7517 | static int sched_dl_global_constraints(void) | |
7518 | { | |
7519 | u64 runtime = global_dl_runtime(); | |
7520 | u64 period = global_dl_period(); | |
7521 | u64 new_bw = to_ratio(period, runtime); | |
7522 | int ret, i; | |
7523 | ||
7524 | ret = __sched_dl_global_constraints(runtime, period); | |
7525 | if (ret) | |
7526 | return ret; | |
7527 | ||
7528 | if (!__sched_dl_rt_global_constraints(new_bw)) | |
7529 | return -EINVAL; | |
7530 | ||
7531 | /* | |
7532 | * Here we want to check the bandwidth not being set to some | |
7533 | * value smaller than the currently allocated bandwidth in | |
7534 | * any of the root_domains. | |
7535 | * | |
7536 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7537 | * cycling on root_domains... Discussion on different/better | |
7538 | * solutions is welcome! | |
7539 | */ | |
7540 | for_each_possible_cpu(i) { | |
7541 | struct dl_bw *dl_b = dl_bw_of(i); | |
7542 | ||
7543 | raw_spin_lock(&dl_b->lock); | |
7544 | if (new_bw < dl_b->total_bw) { | |
7545 | raw_spin_unlock(&dl_b->lock); | |
7546 | return -EBUSY; | |
7547 | } | |
7548 | raw_spin_unlock(&dl_b->lock); | |
7549 | } | |
7550 | ||
7551 | return 0; | |
7552 | } | |
7553 | ||
ce0dbbbb CW |
7554 | int sched_rr_handler(struct ctl_table *table, int write, |
7555 | void __user *buffer, size_t *lenp, | |
7556 | loff_t *ppos) | |
7557 | { | |
7558 | int ret; | |
7559 | static DEFINE_MUTEX(mutex); | |
7560 | ||
7561 | mutex_lock(&mutex); | |
7562 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7563 | /* make sure that internally we keep jiffies */ | |
7564 | /* also, writing zero resets timeslice to default */ | |
7565 | if (!ret && write) { | |
7566 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? | |
7567 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
7568 | } | |
7569 | mutex_unlock(&mutex); | |
7570 | return ret; | |
7571 | } | |
7572 | ||
d0b27fa7 | 7573 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7574 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7575 | loff_t *ppos) |
7576 | { | |
7577 | int ret; | |
7578 | int old_period, old_runtime; | |
7579 | static DEFINE_MUTEX(mutex); | |
7580 | ||
7581 | mutex_lock(&mutex); | |
7582 | old_period = sysctl_sched_rt_period; | |
7583 | old_runtime = sysctl_sched_rt_runtime; | |
7584 | ||
8d65af78 | 7585 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7586 | |
7587 | if (!ret && write) { | |
7588 | ret = sched_rt_global_constraints(); | |
7589 | if (ret) { | |
7590 | sysctl_sched_rt_period = old_period; | |
7591 | sysctl_sched_rt_runtime = old_runtime; | |
7592 | } else { | |
7593 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7594 | def_rt_bandwidth.rt_period = | |
7595 | ns_to_ktime(global_rt_period()); | |
7596 | } | |
7597 | } | |
7598 | mutex_unlock(&mutex); | |
7599 | ||
7600 | return ret; | |
7601 | } | |
68318b8e | 7602 | |
332ac17e DF |
7603 | int sched_dl_handler(struct ctl_table *table, int write, |
7604 | void __user *buffer, size_t *lenp, | |
7605 | loff_t *ppos) | |
7606 | { | |
7607 | int ret; | |
7608 | int old_period, old_runtime; | |
7609 | static DEFINE_MUTEX(mutex); | |
7610 | unsigned long flags; | |
7611 | ||
7612 | mutex_lock(&mutex); | |
7613 | old_period = sysctl_sched_dl_period; | |
7614 | old_runtime = sysctl_sched_dl_runtime; | |
7615 | ||
7616 | ret = proc_dointvec(table, write, buffer, lenp, ppos); | |
7617 | ||
7618 | if (!ret && write) { | |
7619 | raw_spin_lock_irqsave(&def_dl_bandwidth.dl_runtime_lock, | |
7620 | flags); | |
7621 | ||
7622 | ret = sched_dl_global_constraints(); | |
7623 | if (ret) { | |
7624 | sysctl_sched_dl_period = old_period; | |
7625 | sysctl_sched_dl_runtime = old_runtime; | |
7626 | } else { | |
7627 | u64 new_bw; | |
7628 | int i; | |
7629 | ||
7630 | def_dl_bandwidth.dl_period = global_dl_period(); | |
7631 | def_dl_bandwidth.dl_runtime = global_dl_runtime(); | |
7632 | if (global_dl_runtime() == RUNTIME_INF) | |
7633 | new_bw = -1; | |
7634 | else | |
7635 | new_bw = to_ratio(global_dl_period(), | |
7636 | global_dl_runtime()); | |
7637 | /* | |
7638 | * FIXME: As above... | |
7639 | */ | |
7640 | for_each_possible_cpu(i) { | |
7641 | struct dl_bw *dl_b = dl_bw_of(i); | |
7642 | ||
7643 | raw_spin_lock(&dl_b->lock); | |
7644 | dl_b->bw = new_bw; | |
7645 | raw_spin_unlock(&dl_b->lock); | |
7646 | } | |
7647 | } | |
7648 | ||
7649 | raw_spin_unlock_irqrestore(&def_dl_bandwidth.dl_runtime_lock, | |
7650 | flags); | |
7651 | } | |
7652 | mutex_unlock(&mutex); | |
7653 | ||
7654 | return ret; | |
7655 | } | |
7656 | ||
052f1dc7 | 7657 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7658 | |
a7c6d554 | 7659 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7660 | { |
a7c6d554 | 7661 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7662 | } |
7663 | ||
eb95419b TH |
7664 | static struct cgroup_subsys_state * |
7665 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7666 | { |
eb95419b TH |
7667 | struct task_group *parent = css_tg(parent_css); |
7668 | struct task_group *tg; | |
68318b8e | 7669 | |
eb95419b | 7670 | if (!parent) { |
68318b8e | 7671 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7672 | return &root_task_group.css; |
68318b8e SV |
7673 | } |
7674 | ||
ec7dc8ac | 7675 | tg = sched_create_group(parent); |
68318b8e SV |
7676 | if (IS_ERR(tg)) |
7677 | return ERR_PTR(-ENOMEM); | |
7678 | ||
68318b8e SV |
7679 | return &tg->css; |
7680 | } | |
7681 | ||
eb95419b | 7682 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7683 | { |
eb95419b TH |
7684 | struct task_group *tg = css_tg(css); |
7685 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7686 | |
63876986 TH |
7687 | if (parent) |
7688 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7689 | return 0; |
7690 | } | |
7691 | ||
eb95419b | 7692 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7693 | { |
eb95419b | 7694 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7695 | |
7696 | sched_destroy_group(tg); | |
7697 | } | |
7698 | ||
eb95419b | 7699 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7700 | { |
eb95419b | 7701 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7702 | |
7703 | sched_offline_group(tg); | |
7704 | } | |
7705 | ||
eb95419b | 7706 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7707 | struct cgroup_taskset *tset) |
68318b8e | 7708 | { |
bb9d97b6 TH |
7709 | struct task_struct *task; |
7710 | ||
d99c8727 | 7711 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7712 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7713 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7714 | return -EINVAL; |
b68aa230 | 7715 | #else |
bb9d97b6 TH |
7716 | /* We don't support RT-tasks being in separate groups */ |
7717 | if (task->sched_class != &fair_sched_class) | |
7718 | return -EINVAL; | |
b68aa230 | 7719 | #endif |
bb9d97b6 | 7720 | } |
be367d09 BB |
7721 | return 0; |
7722 | } | |
68318b8e | 7723 | |
eb95419b | 7724 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7725 | struct cgroup_taskset *tset) |
68318b8e | 7726 | { |
bb9d97b6 TH |
7727 | struct task_struct *task; |
7728 | ||
d99c8727 | 7729 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7730 | sched_move_task(task); |
68318b8e SV |
7731 | } |
7732 | ||
eb95419b TH |
7733 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7734 | struct cgroup_subsys_state *old_css, | |
7735 | struct task_struct *task) | |
068c5cc5 PZ |
7736 | { |
7737 | /* | |
7738 | * cgroup_exit() is called in the copy_process() failure path. | |
7739 | * Ignore this case since the task hasn't ran yet, this avoids | |
7740 | * trying to poke a half freed task state from generic code. | |
7741 | */ | |
7742 | if (!(task->flags & PF_EXITING)) | |
7743 | return; | |
7744 | ||
7745 | sched_move_task(task); | |
7746 | } | |
7747 | ||
052f1dc7 | 7748 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7749 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7750 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7751 | { |
182446d0 | 7752 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7753 | } |
7754 | ||
182446d0 TH |
7755 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7756 | struct cftype *cft) | |
68318b8e | 7757 | { |
182446d0 | 7758 | struct task_group *tg = css_tg(css); |
68318b8e | 7759 | |
c8b28116 | 7760 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7761 | } |
ab84d31e PT |
7762 | |
7763 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7764 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7765 | ||
ab84d31e PT |
7766 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7767 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7768 | ||
a790de99 PT |
7769 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7770 | ||
ab84d31e PT |
7771 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7772 | { | |
56f570e5 | 7773 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7774 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7775 | |
7776 | if (tg == &root_task_group) | |
7777 | return -EINVAL; | |
7778 | ||
7779 | /* | |
7780 | * Ensure we have at some amount of bandwidth every period. This is | |
7781 | * to prevent reaching a state of large arrears when throttled via | |
7782 | * entity_tick() resulting in prolonged exit starvation. | |
7783 | */ | |
7784 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7785 | return -EINVAL; | |
7786 | ||
7787 | /* | |
7788 | * Likewise, bound things on the otherside by preventing insane quota | |
7789 | * periods. This also allows us to normalize in computing quota | |
7790 | * feasibility. | |
7791 | */ | |
7792 | if (period > max_cfs_quota_period) | |
7793 | return -EINVAL; | |
7794 | ||
a790de99 PT |
7795 | mutex_lock(&cfs_constraints_mutex); |
7796 | ret = __cfs_schedulable(tg, period, quota); | |
7797 | if (ret) | |
7798 | goto out_unlock; | |
7799 | ||
58088ad0 | 7800 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7801 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7802 | /* |
7803 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7804 | * before making related changes, and on->off must occur afterwards | |
7805 | */ | |
7806 | if (runtime_enabled && !runtime_was_enabled) | |
7807 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7808 | raw_spin_lock_irq(&cfs_b->lock); |
7809 | cfs_b->period = ns_to_ktime(period); | |
7810 | cfs_b->quota = quota; | |
58088ad0 | 7811 | |
a9cf55b2 | 7812 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7813 | /* restart the period timer (if active) to handle new period expiry */ |
7814 | if (runtime_enabled && cfs_b->timer_active) { | |
7815 | /* force a reprogram */ | |
7816 | cfs_b->timer_active = 0; | |
7817 | __start_cfs_bandwidth(cfs_b); | |
7818 | } | |
ab84d31e PT |
7819 | raw_spin_unlock_irq(&cfs_b->lock); |
7820 | ||
7821 | for_each_possible_cpu(i) { | |
7822 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7823 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7824 | |
7825 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7826 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7827 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7828 | |
029632fb | 7829 | if (cfs_rq->throttled) |
671fd9da | 7830 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7831 | raw_spin_unlock_irq(&rq->lock); |
7832 | } | |
1ee14e6c BS |
7833 | if (runtime_was_enabled && !runtime_enabled) |
7834 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7835 | out_unlock: |
7836 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7837 | |
a790de99 | 7838 | return ret; |
ab84d31e PT |
7839 | } |
7840 | ||
7841 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7842 | { | |
7843 | u64 quota, period; | |
7844 | ||
029632fb | 7845 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7846 | if (cfs_quota_us < 0) |
7847 | quota = RUNTIME_INF; | |
7848 | else | |
7849 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7850 | ||
7851 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7852 | } | |
7853 | ||
7854 | long tg_get_cfs_quota(struct task_group *tg) | |
7855 | { | |
7856 | u64 quota_us; | |
7857 | ||
029632fb | 7858 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7859 | return -1; |
7860 | ||
029632fb | 7861 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7862 | do_div(quota_us, NSEC_PER_USEC); |
7863 | ||
7864 | return quota_us; | |
7865 | } | |
7866 | ||
7867 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7868 | { | |
7869 | u64 quota, period; | |
7870 | ||
7871 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7872 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7873 | |
ab84d31e PT |
7874 | return tg_set_cfs_bandwidth(tg, period, quota); |
7875 | } | |
7876 | ||
7877 | long tg_get_cfs_period(struct task_group *tg) | |
7878 | { | |
7879 | u64 cfs_period_us; | |
7880 | ||
029632fb | 7881 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7882 | do_div(cfs_period_us, NSEC_PER_USEC); |
7883 | ||
7884 | return cfs_period_us; | |
7885 | } | |
7886 | ||
182446d0 TH |
7887 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7888 | struct cftype *cft) | |
ab84d31e | 7889 | { |
182446d0 | 7890 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7891 | } |
7892 | ||
182446d0 TH |
7893 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7894 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7895 | { |
182446d0 | 7896 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7897 | } |
7898 | ||
182446d0 TH |
7899 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7900 | struct cftype *cft) | |
ab84d31e | 7901 | { |
182446d0 | 7902 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7903 | } |
7904 | ||
182446d0 TH |
7905 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7906 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7907 | { |
182446d0 | 7908 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7909 | } |
7910 | ||
a790de99 PT |
7911 | struct cfs_schedulable_data { |
7912 | struct task_group *tg; | |
7913 | u64 period, quota; | |
7914 | }; | |
7915 | ||
7916 | /* | |
7917 | * normalize group quota/period to be quota/max_period | |
7918 | * note: units are usecs | |
7919 | */ | |
7920 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7921 | struct cfs_schedulable_data *d) | |
7922 | { | |
7923 | u64 quota, period; | |
7924 | ||
7925 | if (tg == d->tg) { | |
7926 | period = d->period; | |
7927 | quota = d->quota; | |
7928 | } else { | |
7929 | period = tg_get_cfs_period(tg); | |
7930 | quota = tg_get_cfs_quota(tg); | |
7931 | } | |
7932 | ||
7933 | /* note: these should typically be equivalent */ | |
7934 | if (quota == RUNTIME_INF || quota == -1) | |
7935 | return RUNTIME_INF; | |
7936 | ||
7937 | return to_ratio(period, quota); | |
7938 | } | |
7939 | ||
7940 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7941 | { | |
7942 | struct cfs_schedulable_data *d = data; | |
029632fb | 7943 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7944 | s64 quota = 0, parent_quota = -1; |
7945 | ||
7946 | if (!tg->parent) { | |
7947 | quota = RUNTIME_INF; | |
7948 | } else { | |
029632fb | 7949 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7950 | |
7951 | quota = normalize_cfs_quota(tg, d); | |
7952 | parent_quota = parent_b->hierarchal_quota; | |
7953 | ||
7954 | /* | |
7955 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7956 | * limit is set | |
7957 | */ | |
7958 | if (quota == RUNTIME_INF) | |
7959 | quota = parent_quota; | |
7960 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7961 | return -EINVAL; | |
7962 | } | |
7963 | cfs_b->hierarchal_quota = quota; | |
7964 | ||
7965 | return 0; | |
7966 | } | |
7967 | ||
7968 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7969 | { | |
8277434e | 7970 | int ret; |
a790de99 PT |
7971 | struct cfs_schedulable_data data = { |
7972 | .tg = tg, | |
7973 | .period = period, | |
7974 | .quota = quota, | |
7975 | }; | |
7976 | ||
7977 | if (quota != RUNTIME_INF) { | |
7978 | do_div(data.period, NSEC_PER_USEC); | |
7979 | do_div(data.quota, NSEC_PER_USEC); | |
7980 | } | |
7981 | ||
8277434e PT |
7982 | rcu_read_lock(); |
7983 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7984 | rcu_read_unlock(); | |
7985 | ||
7986 | return ret; | |
a790de99 | 7987 | } |
e8da1b18 | 7988 | |
182446d0 | 7989 | static int cpu_stats_show(struct cgroup_subsys_state *css, struct cftype *cft, |
e8da1b18 NR |
7990 | struct cgroup_map_cb *cb) |
7991 | { | |
182446d0 | 7992 | struct task_group *tg = css_tg(css); |
029632fb | 7993 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
7994 | |
7995 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
7996 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
7997 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
7998 | ||
7999 | return 0; | |
8000 | } | |
ab84d31e | 8001 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8002 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8003 | |
052f1dc7 | 8004 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8005 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8006 | struct cftype *cft, s64 val) | |
6f505b16 | 8007 | { |
182446d0 | 8008 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8009 | } |
8010 | ||
182446d0 TH |
8011 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8012 | struct cftype *cft) | |
6f505b16 | 8013 | { |
182446d0 | 8014 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8015 | } |
d0b27fa7 | 8016 | |
182446d0 TH |
8017 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8018 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8019 | { |
182446d0 | 8020 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8021 | } |
8022 | ||
182446d0 TH |
8023 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8024 | struct cftype *cft) | |
d0b27fa7 | 8025 | { |
182446d0 | 8026 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8027 | } |
6d6bc0ad | 8028 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8029 | |
fe5c7cc2 | 8030 | static struct cftype cpu_files[] = { |
052f1dc7 | 8031 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8032 | { |
8033 | .name = "shares", | |
f4c753b7 PM |
8034 | .read_u64 = cpu_shares_read_u64, |
8035 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8036 | }, |
052f1dc7 | 8037 | #endif |
ab84d31e PT |
8038 | #ifdef CONFIG_CFS_BANDWIDTH |
8039 | { | |
8040 | .name = "cfs_quota_us", | |
8041 | .read_s64 = cpu_cfs_quota_read_s64, | |
8042 | .write_s64 = cpu_cfs_quota_write_s64, | |
8043 | }, | |
8044 | { | |
8045 | .name = "cfs_period_us", | |
8046 | .read_u64 = cpu_cfs_period_read_u64, | |
8047 | .write_u64 = cpu_cfs_period_write_u64, | |
8048 | }, | |
e8da1b18 NR |
8049 | { |
8050 | .name = "stat", | |
8051 | .read_map = cpu_stats_show, | |
8052 | }, | |
ab84d31e | 8053 | #endif |
052f1dc7 | 8054 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8055 | { |
9f0c1e56 | 8056 | .name = "rt_runtime_us", |
06ecb27c PM |
8057 | .read_s64 = cpu_rt_runtime_read, |
8058 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8059 | }, |
d0b27fa7 PZ |
8060 | { |
8061 | .name = "rt_period_us", | |
f4c753b7 PM |
8062 | .read_u64 = cpu_rt_period_read_uint, |
8063 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8064 | }, |
052f1dc7 | 8065 | #endif |
4baf6e33 | 8066 | { } /* terminate */ |
68318b8e SV |
8067 | }; |
8068 | ||
68318b8e | 8069 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 8070 | .name = "cpu", |
92fb9748 TH |
8071 | .css_alloc = cpu_cgroup_css_alloc, |
8072 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8073 | .css_online = cpu_cgroup_css_online, |
8074 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8075 | .can_attach = cpu_cgroup_can_attach, |
8076 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8077 | .exit = cpu_cgroup_exit, |
38605cae | 8078 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 8079 | .base_cftypes = cpu_files, |
68318b8e SV |
8080 | .early_init = 1, |
8081 | }; | |
8082 | ||
052f1dc7 | 8083 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8084 | |
b637a328 PM |
8085 | void dump_cpu_task(int cpu) |
8086 | { | |
8087 | pr_info("Task dump for CPU %d:\n", cpu); | |
8088 | sched_show_task(cpu_curr(cpu)); | |
8089 | } |