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> |
52f5684c | 76 | #include <linux/compiler.h> |
1da177e4 | 77 | |
96f951ed | 78 | #include <asm/switch_to.h> |
5517d86b | 79 | #include <asm/tlb.h> |
838225b4 | 80 | #include <asm/irq_regs.h> |
db7e527d | 81 | #include <asm/mutex.h> |
e6e6685a GC |
82 | #ifdef CONFIG_PARAVIRT |
83 | #include <asm/paravirt.h> | |
84 | #endif | |
1da177e4 | 85 | |
029632fb | 86 | #include "sched.h" |
ea138446 | 87 | #include "../workqueue_internal.h" |
29d5e047 | 88 | #include "../smpboot.h" |
6e0534f2 | 89 | |
a8d154b0 | 90 | #define CREATE_TRACE_POINTS |
ad8d75ff | 91 | #include <trace/events/sched.h> |
a8d154b0 | 92 | |
febdbfe8 PZ |
93 | #ifdef smp_mb__before_atomic |
94 | void __smp_mb__before_atomic(void) | |
95 | { | |
96 | smp_mb__before_atomic(); | |
97 | } | |
98 | EXPORT_SYMBOL(__smp_mb__before_atomic); | |
99 | #endif | |
100 | ||
101 | #ifdef smp_mb__after_atomic | |
102 | void __smp_mb__after_atomic(void) | |
103 | { | |
104 | smp_mb__after_atomic(); | |
105 | } | |
106 | EXPORT_SYMBOL(__smp_mb__after_atomic); | |
107 | #endif | |
108 | ||
029632fb | 109 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 110 | { |
58088ad0 PT |
111 | unsigned long delta; |
112 | ktime_t soft, hard, now; | |
d0b27fa7 | 113 | |
58088ad0 PT |
114 | for (;;) { |
115 | if (hrtimer_active(period_timer)) | |
116 | break; | |
117 | ||
118 | now = hrtimer_cb_get_time(period_timer); | |
119 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 120 | |
58088ad0 PT |
121 | soft = hrtimer_get_softexpires(period_timer); |
122 | hard = hrtimer_get_expires(period_timer); | |
123 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
124 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
125 | HRTIMER_MODE_ABS_PINNED, 0); | |
126 | } | |
127 | } | |
128 | ||
029632fb PZ |
129 | DEFINE_MUTEX(sched_domains_mutex); |
130 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 131 | |
fe44d621 | 132 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 133 | |
029632fb | 134 | void update_rq_clock(struct rq *rq) |
3e51f33f | 135 | { |
fe44d621 | 136 | s64 delta; |
305e6835 | 137 | |
61eadef6 | 138 | if (rq->skip_clock_update > 0) |
f26f9aff | 139 | return; |
aa483808 | 140 | |
fe44d621 PZ |
141 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
142 | rq->clock += delta; | |
143 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
144 | } |
145 | ||
bf5c91ba IM |
146 | /* |
147 | * Debugging: various feature bits | |
148 | */ | |
f00b45c1 | 149 | |
f00b45c1 PZ |
150 | #define SCHED_FEAT(name, enabled) \ |
151 | (1UL << __SCHED_FEAT_##name) * enabled | | |
152 | ||
bf5c91ba | 153 | const_debug unsigned int sysctl_sched_features = |
391e43da | 154 | #include "features.h" |
f00b45c1 PZ |
155 | 0; |
156 | ||
157 | #undef SCHED_FEAT | |
158 | ||
159 | #ifdef CONFIG_SCHED_DEBUG | |
160 | #define SCHED_FEAT(name, enabled) \ | |
161 | #name , | |
162 | ||
1292531f | 163 | static const char * const sched_feat_names[] = { |
391e43da | 164 | #include "features.h" |
f00b45c1 PZ |
165 | }; |
166 | ||
167 | #undef SCHED_FEAT | |
168 | ||
34f3a814 | 169 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 170 | { |
f00b45c1 PZ |
171 | int i; |
172 | ||
f8b6d1cc | 173 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
174 | if (!(sysctl_sched_features & (1UL << i))) |
175 | seq_puts(m, "NO_"); | |
176 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 177 | } |
34f3a814 | 178 | seq_puts(m, "\n"); |
f00b45c1 | 179 | |
34f3a814 | 180 | return 0; |
f00b45c1 PZ |
181 | } |
182 | ||
f8b6d1cc PZ |
183 | #ifdef HAVE_JUMP_LABEL |
184 | ||
c5905afb IM |
185 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
186 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
187 | |
188 | #define SCHED_FEAT(name, enabled) \ | |
189 | jump_label_key__##enabled , | |
190 | ||
c5905afb | 191 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
192 | #include "features.h" |
193 | }; | |
194 | ||
195 | #undef SCHED_FEAT | |
196 | ||
197 | static void sched_feat_disable(int i) | |
198 | { | |
c5905afb IM |
199 | if (static_key_enabled(&sched_feat_keys[i])) |
200 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
201 | } |
202 | ||
203 | static void sched_feat_enable(int i) | |
204 | { | |
c5905afb IM |
205 | if (!static_key_enabled(&sched_feat_keys[i])) |
206 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
207 | } |
208 | #else | |
209 | static void sched_feat_disable(int i) { }; | |
210 | static void sched_feat_enable(int i) { }; | |
211 | #endif /* HAVE_JUMP_LABEL */ | |
212 | ||
1a687c2e | 213 | static int sched_feat_set(char *cmp) |
f00b45c1 | 214 | { |
f00b45c1 | 215 | int i; |
1a687c2e | 216 | int neg = 0; |
f00b45c1 | 217 | |
524429c3 | 218 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
219 | neg = 1; |
220 | cmp += 3; | |
221 | } | |
222 | ||
f8b6d1cc | 223 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 224 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 225 | if (neg) { |
f00b45c1 | 226 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
227 | sched_feat_disable(i); |
228 | } else { | |
f00b45c1 | 229 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
230 | sched_feat_enable(i); |
231 | } | |
f00b45c1 PZ |
232 | break; |
233 | } | |
234 | } | |
235 | ||
1a687c2e MG |
236 | return i; |
237 | } | |
238 | ||
239 | static ssize_t | |
240 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
241 | size_t cnt, loff_t *ppos) | |
242 | { | |
243 | char buf[64]; | |
244 | char *cmp; | |
245 | int i; | |
246 | ||
247 | if (cnt > 63) | |
248 | cnt = 63; | |
249 | ||
250 | if (copy_from_user(&buf, ubuf, cnt)) | |
251 | return -EFAULT; | |
252 | ||
253 | buf[cnt] = 0; | |
254 | cmp = strstrip(buf); | |
255 | ||
256 | i = sched_feat_set(cmp); | |
f8b6d1cc | 257 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
258 | return -EINVAL; |
259 | ||
42994724 | 260 | *ppos += cnt; |
f00b45c1 PZ |
261 | |
262 | return cnt; | |
263 | } | |
264 | ||
34f3a814 LZ |
265 | static int sched_feat_open(struct inode *inode, struct file *filp) |
266 | { | |
267 | return single_open(filp, sched_feat_show, NULL); | |
268 | } | |
269 | ||
828c0950 | 270 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
271 | .open = sched_feat_open, |
272 | .write = sched_feat_write, | |
273 | .read = seq_read, | |
274 | .llseek = seq_lseek, | |
275 | .release = single_release, | |
f00b45c1 PZ |
276 | }; |
277 | ||
278 | static __init int sched_init_debug(void) | |
279 | { | |
f00b45c1 PZ |
280 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
281 | &sched_feat_fops); | |
282 | ||
283 | return 0; | |
284 | } | |
285 | late_initcall(sched_init_debug); | |
f8b6d1cc | 286 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 287 | |
b82d9fdd PZ |
288 | /* |
289 | * Number of tasks to iterate in a single balance run. | |
290 | * Limited because this is done with IRQs disabled. | |
291 | */ | |
292 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
293 | ||
e9e9250b PZ |
294 | /* |
295 | * period over which we average the RT time consumption, measured | |
296 | * in ms. | |
297 | * | |
298 | * default: 1s | |
299 | */ | |
300 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
301 | ||
fa85ae24 | 302 | /* |
9f0c1e56 | 303 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
304 | * default: 1s |
305 | */ | |
9f0c1e56 | 306 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 307 | |
029632fb | 308 | __read_mostly int scheduler_running; |
6892b75e | 309 | |
9f0c1e56 PZ |
310 | /* |
311 | * part of the period that we allow rt tasks to run in us. | |
312 | * default: 0.95s | |
313 | */ | |
314 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 315 | |
0970d299 | 316 | /* |
0122ec5b | 317 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 318 | */ |
70b97a7f | 319 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
320 | __acquires(rq->lock) |
321 | { | |
0970d299 PZ |
322 | struct rq *rq; |
323 | ||
0122ec5b PZ |
324 | lockdep_assert_held(&p->pi_lock); |
325 | ||
3a5c359a | 326 | for (;;) { |
0970d299 | 327 | rq = task_rq(p); |
05fa785c | 328 | raw_spin_lock(&rq->lock); |
65cc8e48 | 329 | if (likely(rq == task_rq(p))) |
3a5c359a | 330 | return rq; |
05fa785c | 331 | raw_spin_unlock(&rq->lock); |
b29739f9 | 332 | } |
b29739f9 IM |
333 | } |
334 | ||
1da177e4 | 335 | /* |
0122ec5b | 336 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 337 | */ |
70b97a7f | 338 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 339 | __acquires(p->pi_lock) |
1da177e4 LT |
340 | __acquires(rq->lock) |
341 | { | |
70b97a7f | 342 | struct rq *rq; |
1da177e4 | 343 | |
3a5c359a | 344 | for (;;) { |
0122ec5b | 345 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 346 | rq = task_rq(p); |
05fa785c | 347 | raw_spin_lock(&rq->lock); |
65cc8e48 | 348 | if (likely(rq == task_rq(p))) |
3a5c359a | 349 | return rq; |
0122ec5b PZ |
350 | raw_spin_unlock(&rq->lock); |
351 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 352 | } |
1da177e4 LT |
353 | } |
354 | ||
a9957449 | 355 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
356 | __releases(rq->lock) |
357 | { | |
05fa785c | 358 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
359 | } |
360 | ||
0122ec5b PZ |
361 | static inline void |
362 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 363 | __releases(rq->lock) |
0122ec5b | 364 | __releases(p->pi_lock) |
1da177e4 | 365 | { |
0122ec5b PZ |
366 | raw_spin_unlock(&rq->lock); |
367 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
368 | } |
369 | ||
1da177e4 | 370 | /* |
cc2a73b5 | 371 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 372 | */ |
a9957449 | 373 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
374 | __acquires(rq->lock) |
375 | { | |
70b97a7f | 376 | struct rq *rq; |
1da177e4 LT |
377 | |
378 | local_irq_disable(); | |
379 | rq = this_rq(); | |
05fa785c | 380 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
381 | |
382 | return rq; | |
383 | } | |
384 | ||
8f4d37ec PZ |
385 | #ifdef CONFIG_SCHED_HRTICK |
386 | /* | |
387 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 388 | */ |
8f4d37ec | 389 | |
8f4d37ec PZ |
390 | static void hrtick_clear(struct rq *rq) |
391 | { | |
392 | if (hrtimer_active(&rq->hrtick_timer)) | |
393 | hrtimer_cancel(&rq->hrtick_timer); | |
394 | } | |
395 | ||
8f4d37ec PZ |
396 | /* |
397 | * High-resolution timer tick. | |
398 | * Runs from hardirq context with interrupts disabled. | |
399 | */ | |
400 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
401 | { | |
402 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
403 | ||
404 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
405 | ||
05fa785c | 406 | raw_spin_lock(&rq->lock); |
3e51f33f | 407 | update_rq_clock(rq); |
8f4d37ec | 408 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 409 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
410 | |
411 | return HRTIMER_NORESTART; | |
412 | } | |
413 | ||
95e904c7 | 414 | #ifdef CONFIG_SMP |
971ee28c PZ |
415 | |
416 | static int __hrtick_restart(struct rq *rq) | |
417 | { | |
418 | struct hrtimer *timer = &rq->hrtick_timer; | |
419 | ktime_t time = hrtimer_get_softexpires(timer); | |
420 | ||
421 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
422 | } | |
423 | ||
31656519 PZ |
424 | /* |
425 | * called from hardirq (IPI) context | |
426 | */ | |
427 | static void __hrtick_start(void *arg) | |
b328ca18 | 428 | { |
31656519 | 429 | struct rq *rq = arg; |
b328ca18 | 430 | |
05fa785c | 431 | raw_spin_lock(&rq->lock); |
971ee28c | 432 | __hrtick_restart(rq); |
31656519 | 433 | rq->hrtick_csd_pending = 0; |
05fa785c | 434 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
435 | } |
436 | ||
31656519 PZ |
437 | /* |
438 | * Called to set the hrtick timer state. | |
439 | * | |
440 | * called with rq->lock held and irqs disabled | |
441 | */ | |
029632fb | 442 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 443 | { |
31656519 PZ |
444 | struct hrtimer *timer = &rq->hrtick_timer; |
445 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 446 | |
cc584b21 | 447 | hrtimer_set_expires(timer, time); |
31656519 PZ |
448 | |
449 | if (rq == this_rq()) { | |
971ee28c | 450 | __hrtick_restart(rq); |
31656519 | 451 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 452 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
453 | rq->hrtick_csd_pending = 1; |
454 | } | |
b328ca18 PZ |
455 | } |
456 | ||
457 | static int | |
458 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
459 | { | |
460 | int cpu = (int)(long)hcpu; | |
461 | ||
462 | switch (action) { | |
463 | case CPU_UP_CANCELED: | |
464 | case CPU_UP_CANCELED_FROZEN: | |
465 | case CPU_DOWN_PREPARE: | |
466 | case CPU_DOWN_PREPARE_FROZEN: | |
467 | case CPU_DEAD: | |
468 | case CPU_DEAD_FROZEN: | |
31656519 | 469 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
470 | return NOTIFY_OK; |
471 | } | |
472 | ||
473 | return NOTIFY_DONE; | |
474 | } | |
475 | ||
fa748203 | 476 | static __init void init_hrtick(void) |
b328ca18 PZ |
477 | { |
478 | hotcpu_notifier(hotplug_hrtick, 0); | |
479 | } | |
31656519 PZ |
480 | #else |
481 | /* | |
482 | * Called to set the hrtick timer state. | |
483 | * | |
484 | * called with rq->lock held and irqs disabled | |
485 | */ | |
029632fb | 486 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 487 | { |
7f1e2ca9 | 488 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 489 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 490 | } |
b328ca18 | 491 | |
006c75f1 | 492 | static inline void init_hrtick(void) |
8f4d37ec | 493 | { |
8f4d37ec | 494 | } |
31656519 | 495 | #endif /* CONFIG_SMP */ |
8f4d37ec | 496 | |
31656519 | 497 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 498 | { |
31656519 PZ |
499 | #ifdef CONFIG_SMP |
500 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 501 | |
31656519 PZ |
502 | rq->hrtick_csd.flags = 0; |
503 | rq->hrtick_csd.func = __hrtick_start; | |
504 | rq->hrtick_csd.info = rq; | |
505 | #endif | |
8f4d37ec | 506 | |
31656519 PZ |
507 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
508 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 509 | } |
006c75f1 | 510 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
511 | static inline void hrtick_clear(struct rq *rq) |
512 | { | |
513 | } | |
514 | ||
8f4d37ec PZ |
515 | static inline void init_rq_hrtick(struct rq *rq) |
516 | { | |
517 | } | |
518 | ||
b328ca18 PZ |
519 | static inline void init_hrtick(void) |
520 | { | |
521 | } | |
006c75f1 | 522 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 523 | |
fd99f91a PZ |
524 | /* |
525 | * cmpxchg based fetch_or, macro so it works for different integer types | |
526 | */ | |
527 | #define fetch_or(ptr, val) \ | |
528 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
529 | for (;;) { \ | |
530 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
531 | if (__old == __val) \ | |
532 | break; \ | |
533 | __val = __old; \ | |
534 | } \ | |
535 | __old; \ | |
536 | }) | |
537 | ||
538 | #ifdef TIF_POLLING_NRFLAG | |
539 | /* | |
540 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
541 | * this avoids any races wrt polling state changes and thereby avoids | |
542 | * spurious IPIs. | |
543 | */ | |
544 | static bool set_nr_and_not_polling(struct task_struct *p) | |
545 | { | |
546 | struct thread_info *ti = task_thread_info(p); | |
547 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
548 | } | |
549 | #else | |
550 | static bool set_nr_and_not_polling(struct task_struct *p) | |
551 | { | |
552 | set_tsk_need_resched(p); | |
553 | return true; | |
554 | } | |
555 | #endif | |
556 | ||
c24d20db IM |
557 | /* |
558 | * resched_task - mark a task 'to be rescheduled now'. | |
559 | * | |
560 | * On UP this means the setting of the need_resched flag, on SMP it | |
561 | * might also involve a cross-CPU call to trigger the scheduler on | |
562 | * the target CPU. | |
563 | */ | |
029632fb | 564 | void resched_task(struct task_struct *p) |
c24d20db IM |
565 | { |
566 | int cpu; | |
567 | ||
b021fe3e | 568 | lockdep_assert_held(&task_rq(p)->lock); |
c24d20db | 569 | |
5ed0cec0 | 570 | if (test_tsk_need_resched(p)) |
c24d20db IM |
571 | return; |
572 | ||
c24d20db | 573 | cpu = task_cpu(p); |
fd99f91a | 574 | |
f27dde8d | 575 | if (cpu == smp_processor_id()) { |
fd99f91a | 576 | set_tsk_need_resched(p); |
f27dde8d | 577 | set_preempt_need_resched(); |
c24d20db | 578 | return; |
f27dde8d | 579 | } |
c24d20db | 580 | |
fd99f91a | 581 | if (set_nr_and_not_polling(p)) |
c24d20db IM |
582 | smp_send_reschedule(cpu); |
583 | } | |
584 | ||
029632fb | 585 | void resched_cpu(int cpu) |
c24d20db IM |
586 | { |
587 | struct rq *rq = cpu_rq(cpu); | |
588 | unsigned long flags; | |
589 | ||
05fa785c | 590 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
591 | return; |
592 | resched_task(cpu_curr(cpu)); | |
05fa785c | 593 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 594 | } |
06d8308c | 595 | |
b021fe3e | 596 | #ifdef CONFIG_SMP |
3451d024 | 597 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
598 | /* |
599 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
600 | * from an idle cpu. This is good for power-savings. | |
601 | * | |
602 | * We don't do similar optimization for completely idle system, as | |
603 | * selecting an idle cpu will add more delays to the timers than intended | |
604 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
605 | */ | |
6201b4d6 | 606 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
607 | { |
608 | int cpu = smp_processor_id(); | |
609 | int i; | |
610 | struct sched_domain *sd; | |
611 | ||
6201b4d6 VK |
612 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
613 | return cpu; | |
614 | ||
057f3fad | 615 | rcu_read_lock(); |
83cd4fe2 | 616 | for_each_domain(cpu, sd) { |
057f3fad PZ |
617 | for_each_cpu(i, sched_domain_span(sd)) { |
618 | if (!idle_cpu(i)) { | |
619 | cpu = i; | |
620 | goto unlock; | |
621 | } | |
622 | } | |
83cd4fe2 | 623 | } |
057f3fad PZ |
624 | unlock: |
625 | rcu_read_unlock(); | |
83cd4fe2 VP |
626 | return cpu; |
627 | } | |
06d8308c TG |
628 | /* |
629 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
630 | * idle CPU then this timer might expire before the next timer event | |
631 | * which is scheduled to wake up that CPU. In case of a completely | |
632 | * idle system the next event might even be infinite time into the | |
633 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
634 | * leaves the inner idle loop so the newly added timer is taken into | |
635 | * account when the CPU goes back to idle and evaluates the timer | |
636 | * wheel for the next timer event. | |
637 | */ | |
1c20091e | 638 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
639 | { |
640 | struct rq *rq = cpu_rq(cpu); | |
641 | ||
642 | if (cpu == smp_processor_id()) | |
643 | return; | |
644 | ||
645 | /* | |
646 | * This is safe, as this function is called with the timer | |
647 | * wheel base lock of (cpu) held. When the CPU is on the way | |
648 | * to idle and has not yet set rq->curr to idle then it will | |
649 | * be serialized on the timer wheel base lock and take the new | |
650 | * timer into account automatically. | |
651 | */ | |
652 | if (rq->curr != rq->idle) | |
653 | return; | |
45bf76df | 654 | |
45bf76df | 655 | /* |
06d8308c TG |
656 | * We can set TIF_RESCHED on the idle task of the other CPU |
657 | * lockless. The worst case is that the other CPU runs the | |
658 | * idle task through an additional NOOP schedule() | |
45bf76df | 659 | */ |
5ed0cec0 | 660 | set_tsk_need_resched(rq->idle); |
45bf76df | 661 | |
06d8308c TG |
662 | /* NEED_RESCHED must be visible before we test polling */ |
663 | smp_mb(); | |
664 | if (!tsk_is_polling(rq->idle)) | |
665 | smp_send_reschedule(cpu); | |
45bf76df IM |
666 | } |
667 | ||
c5bfece2 | 668 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 669 | { |
c5bfece2 | 670 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
671 | if (cpu != smp_processor_id() || |
672 | tick_nohz_tick_stopped()) | |
673 | smp_send_reschedule(cpu); | |
674 | return true; | |
675 | } | |
676 | ||
677 | return false; | |
678 | } | |
679 | ||
680 | void wake_up_nohz_cpu(int cpu) | |
681 | { | |
c5bfece2 | 682 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
683 | wake_up_idle_cpu(cpu); |
684 | } | |
685 | ||
ca38062e | 686 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 687 | { |
1c792db7 | 688 | int cpu = smp_processor_id(); |
873b4c65 VG |
689 | |
690 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
691 | return false; | |
692 | ||
693 | if (idle_cpu(cpu) && !need_resched()) | |
694 | return true; | |
695 | ||
696 | /* | |
697 | * We can't run Idle Load Balance on this CPU for this time so we | |
698 | * cancel it and clear NOHZ_BALANCE_KICK | |
699 | */ | |
700 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
701 | return false; | |
45bf76df IM |
702 | } |
703 | ||
3451d024 | 704 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 705 | |
ca38062e | 706 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 707 | { |
ca38062e | 708 | return false; |
2069dd75 PZ |
709 | } |
710 | ||
3451d024 | 711 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 712 | |
ce831b38 FW |
713 | #ifdef CONFIG_NO_HZ_FULL |
714 | bool sched_can_stop_tick(void) | |
715 | { | |
716 | struct rq *rq; | |
717 | ||
718 | rq = this_rq(); | |
719 | ||
720 | /* Make sure rq->nr_running update is visible after the IPI */ | |
721 | smp_rmb(); | |
722 | ||
723 | /* More than one running task need preemption */ | |
724 | if (rq->nr_running > 1) | |
725 | return false; | |
726 | ||
727 | return true; | |
728 | } | |
729 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 730 | |
029632fb | 731 | void sched_avg_update(struct rq *rq) |
18d95a28 | 732 | { |
e9e9250b PZ |
733 | s64 period = sched_avg_period(); |
734 | ||
78becc27 | 735 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
736 | /* |
737 | * Inline assembly required to prevent the compiler | |
738 | * optimising this loop into a divmod call. | |
739 | * See __iter_div_u64_rem() for another example of this. | |
740 | */ | |
741 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
742 | rq->age_stamp += period; |
743 | rq->rt_avg /= 2; | |
744 | } | |
18d95a28 PZ |
745 | } |
746 | ||
6d6bc0ad | 747 | #endif /* CONFIG_SMP */ |
18d95a28 | 748 | |
a790de99 PT |
749 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
750 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 751 | /* |
8277434e PT |
752 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
753 | * node and @up when leaving it for the final time. | |
754 | * | |
755 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 756 | */ |
029632fb | 757 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 758 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
759 | { |
760 | struct task_group *parent, *child; | |
eb755805 | 761 | int ret; |
c09595f6 | 762 | |
8277434e PT |
763 | parent = from; |
764 | ||
c09595f6 | 765 | down: |
eb755805 PZ |
766 | ret = (*down)(parent, data); |
767 | if (ret) | |
8277434e | 768 | goto out; |
c09595f6 PZ |
769 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
770 | parent = child; | |
771 | goto down; | |
772 | ||
773 | up: | |
774 | continue; | |
775 | } | |
eb755805 | 776 | ret = (*up)(parent, data); |
8277434e PT |
777 | if (ret || parent == from) |
778 | goto out; | |
c09595f6 PZ |
779 | |
780 | child = parent; | |
781 | parent = parent->parent; | |
782 | if (parent) | |
783 | goto up; | |
8277434e | 784 | out: |
eb755805 | 785 | return ret; |
c09595f6 PZ |
786 | } |
787 | ||
029632fb | 788 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 789 | { |
e2b245f8 | 790 | return 0; |
eb755805 | 791 | } |
18d95a28 PZ |
792 | #endif |
793 | ||
45bf76df IM |
794 | static void set_load_weight(struct task_struct *p) |
795 | { | |
f05998d4 NR |
796 | int prio = p->static_prio - MAX_RT_PRIO; |
797 | struct load_weight *load = &p->se.load; | |
798 | ||
dd41f596 IM |
799 | /* |
800 | * SCHED_IDLE tasks get minimal weight: | |
801 | */ | |
802 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 803 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 804 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
805 | return; |
806 | } | |
71f8bd46 | 807 | |
c8b28116 | 808 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 809 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
810 | } |
811 | ||
371fd7e7 | 812 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 813 | { |
a64692a3 | 814 | update_rq_clock(rq); |
43148951 | 815 | sched_info_queued(rq, p); |
371fd7e7 | 816 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
817 | } |
818 | ||
371fd7e7 | 819 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 820 | { |
a64692a3 | 821 | update_rq_clock(rq); |
43148951 | 822 | sched_info_dequeued(rq, p); |
371fd7e7 | 823 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
824 | } |
825 | ||
029632fb | 826 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
827 | { |
828 | if (task_contributes_to_load(p)) | |
829 | rq->nr_uninterruptible--; | |
830 | ||
371fd7e7 | 831 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
832 | } |
833 | ||
029632fb | 834 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
835 | { |
836 | if (task_contributes_to_load(p)) | |
837 | rq->nr_uninterruptible++; | |
838 | ||
371fd7e7 | 839 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
840 | } |
841 | ||
fe44d621 | 842 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 843 | { |
095c0aa8 GC |
844 | /* |
845 | * In theory, the compile should just see 0 here, and optimize out the call | |
846 | * to sched_rt_avg_update. But I don't trust it... | |
847 | */ | |
848 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
849 | s64 steal = 0, irq_delta = 0; | |
850 | #endif | |
851 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 852 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
853 | |
854 | /* | |
855 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
856 | * this case when a previous update_rq_clock() happened inside a | |
857 | * {soft,}irq region. | |
858 | * | |
859 | * When this happens, we stop ->clock_task and only update the | |
860 | * prev_irq_time stamp to account for the part that fit, so that a next | |
861 | * update will consume the rest. This ensures ->clock_task is | |
862 | * monotonic. | |
863 | * | |
864 | * It does however cause some slight miss-attribution of {soft,}irq | |
865 | * time, a more accurate solution would be to update the irq_time using | |
866 | * the current rq->clock timestamp, except that would require using | |
867 | * atomic ops. | |
868 | */ | |
869 | if (irq_delta > delta) | |
870 | irq_delta = delta; | |
871 | ||
872 | rq->prev_irq_time += irq_delta; | |
873 | delta -= irq_delta; | |
095c0aa8 GC |
874 | #endif |
875 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 876 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
877 | steal = paravirt_steal_clock(cpu_of(rq)); |
878 | steal -= rq->prev_steal_time_rq; | |
879 | ||
880 | if (unlikely(steal > delta)) | |
881 | steal = delta; | |
882 | ||
095c0aa8 | 883 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
884 | delta -= steal; |
885 | } | |
886 | #endif | |
887 | ||
fe44d621 PZ |
888 | rq->clock_task += delta; |
889 | ||
095c0aa8 GC |
890 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
891 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
892 | sched_rt_avg_update(rq, irq_delta + steal); | |
893 | #endif | |
aa483808 VP |
894 | } |
895 | ||
34f971f6 PZ |
896 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
897 | { | |
898 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
899 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
900 | ||
901 | if (stop) { | |
902 | /* | |
903 | * Make it appear like a SCHED_FIFO task, its something | |
904 | * userspace knows about and won't get confused about. | |
905 | * | |
906 | * Also, it will make PI more or less work without too | |
907 | * much confusion -- but then, stop work should not | |
908 | * rely on PI working anyway. | |
909 | */ | |
910 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
911 | ||
912 | stop->sched_class = &stop_sched_class; | |
913 | } | |
914 | ||
915 | cpu_rq(cpu)->stop = stop; | |
916 | ||
917 | if (old_stop) { | |
918 | /* | |
919 | * Reset it back to a normal scheduling class so that | |
920 | * it can die in pieces. | |
921 | */ | |
922 | old_stop->sched_class = &rt_sched_class; | |
923 | } | |
924 | } | |
925 | ||
14531189 | 926 | /* |
dd41f596 | 927 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 928 | */ |
14531189 IM |
929 | static inline int __normal_prio(struct task_struct *p) |
930 | { | |
dd41f596 | 931 | return p->static_prio; |
14531189 IM |
932 | } |
933 | ||
b29739f9 IM |
934 | /* |
935 | * Calculate the expected normal priority: i.e. priority | |
936 | * without taking RT-inheritance into account. Might be | |
937 | * boosted by interactivity modifiers. Changes upon fork, | |
938 | * setprio syscalls, and whenever the interactivity | |
939 | * estimator recalculates. | |
940 | */ | |
36c8b586 | 941 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
942 | { |
943 | int prio; | |
944 | ||
aab03e05 DF |
945 | if (task_has_dl_policy(p)) |
946 | prio = MAX_DL_PRIO-1; | |
947 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
948 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
949 | else | |
950 | prio = __normal_prio(p); | |
951 | return prio; | |
952 | } | |
953 | ||
954 | /* | |
955 | * Calculate the current priority, i.e. the priority | |
956 | * taken into account by the scheduler. This value might | |
957 | * be boosted by RT tasks, or might be boosted by | |
958 | * interactivity modifiers. Will be RT if the task got | |
959 | * RT-boosted. If not then it returns p->normal_prio. | |
960 | */ | |
36c8b586 | 961 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
962 | { |
963 | p->normal_prio = normal_prio(p); | |
964 | /* | |
965 | * If we are RT tasks or we were boosted to RT priority, | |
966 | * keep the priority unchanged. Otherwise, update priority | |
967 | * to the normal priority: | |
968 | */ | |
969 | if (!rt_prio(p->prio)) | |
970 | return p->normal_prio; | |
971 | return p->prio; | |
972 | } | |
973 | ||
1da177e4 LT |
974 | /** |
975 | * task_curr - is this task currently executing on a CPU? | |
976 | * @p: the task in question. | |
e69f6186 YB |
977 | * |
978 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 979 | */ |
36c8b586 | 980 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
981 | { |
982 | return cpu_curr(task_cpu(p)) == p; | |
983 | } | |
984 | ||
cb469845 SR |
985 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
986 | const struct sched_class *prev_class, | |
da7a735e | 987 | int oldprio) |
cb469845 SR |
988 | { |
989 | if (prev_class != p->sched_class) { | |
990 | if (prev_class->switched_from) | |
da7a735e PZ |
991 | prev_class->switched_from(rq, p); |
992 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 993 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 994 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
995 | } |
996 | ||
029632fb | 997 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
998 | { |
999 | const struct sched_class *class; | |
1000 | ||
1001 | if (p->sched_class == rq->curr->sched_class) { | |
1002 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1003 | } else { | |
1004 | for_each_class(class) { | |
1005 | if (class == rq->curr->sched_class) | |
1006 | break; | |
1007 | if (class == p->sched_class) { | |
1008 | resched_task(rq->curr); | |
1009 | break; | |
1010 | } | |
1011 | } | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * A queue event has occurred, and we're going to schedule. In | |
1016 | * this case, we can save a useless back to back clock update. | |
1017 | */ | |
fd2f4419 | 1018 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1019 | rq->skip_clock_update = 1; |
1020 | } | |
1021 | ||
1da177e4 | 1022 | #ifdef CONFIG_SMP |
dd41f596 | 1023 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1024 | { |
e2912009 PZ |
1025 | #ifdef CONFIG_SCHED_DEBUG |
1026 | /* | |
1027 | * We should never call set_task_cpu() on a blocked task, | |
1028 | * ttwu() will sort out the placement. | |
1029 | */ | |
077614ee | 1030 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1031 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1032 | |
1033 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1034 | /* |
1035 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1036 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1037 | * | |
1038 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1039 | * see task_group(). |
6c6c54e1 PZ |
1040 | * |
1041 | * Furthermore, all task_rq users should acquire both locks, see | |
1042 | * task_rq_lock(). | |
1043 | */ | |
0122ec5b PZ |
1044 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1045 | lockdep_is_held(&task_rq(p)->lock))); | |
1046 | #endif | |
e2912009 PZ |
1047 | #endif |
1048 | ||
de1d7286 | 1049 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1050 | |
0c69774e | 1051 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1052 | if (p->sched_class->migrate_task_rq) |
1053 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1054 | p->se.nr_migrations++; |
a8b0ca17 | 1055 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1056 | } |
dd41f596 IM |
1057 | |
1058 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1059 | } |
1060 | ||
ac66f547 PZ |
1061 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1062 | { | |
1063 | if (p->on_rq) { | |
1064 | struct rq *src_rq, *dst_rq; | |
1065 | ||
1066 | src_rq = task_rq(p); | |
1067 | dst_rq = cpu_rq(cpu); | |
1068 | ||
1069 | deactivate_task(src_rq, p, 0); | |
1070 | set_task_cpu(p, cpu); | |
1071 | activate_task(dst_rq, p, 0); | |
1072 | check_preempt_curr(dst_rq, p, 0); | |
1073 | } else { | |
1074 | /* | |
1075 | * Task isn't running anymore; make it appear like we migrated | |
1076 | * it before it went to sleep. This means on wakeup we make the | |
1077 | * previous cpu our targer instead of where it really is. | |
1078 | */ | |
1079 | p->wake_cpu = cpu; | |
1080 | } | |
1081 | } | |
1082 | ||
1083 | struct migration_swap_arg { | |
1084 | struct task_struct *src_task, *dst_task; | |
1085 | int src_cpu, dst_cpu; | |
1086 | }; | |
1087 | ||
1088 | static int migrate_swap_stop(void *data) | |
1089 | { | |
1090 | struct migration_swap_arg *arg = data; | |
1091 | struct rq *src_rq, *dst_rq; | |
1092 | int ret = -EAGAIN; | |
1093 | ||
1094 | src_rq = cpu_rq(arg->src_cpu); | |
1095 | dst_rq = cpu_rq(arg->dst_cpu); | |
1096 | ||
74602315 PZ |
1097 | double_raw_lock(&arg->src_task->pi_lock, |
1098 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1099 | double_rq_lock(src_rq, dst_rq); |
1100 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1101 | goto unlock; | |
1102 | ||
1103 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1104 | goto unlock; | |
1105 | ||
1106 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1107 | goto unlock; | |
1108 | ||
1109 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1110 | goto unlock; | |
1111 | ||
1112 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1113 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1114 | ||
1115 | ret = 0; | |
1116 | ||
1117 | unlock: | |
1118 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1119 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1120 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1121 | |
1122 | return ret; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Cross migrate two tasks | |
1127 | */ | |
1128 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1129 | { | |
1130 | struct migration_swap_arg arg; | |
1131 | int ret = -EINVAL; | |
1132 | ||
ac66f547 PZ |
1133 | arg = (struct migration_swap_arg){ |
1134 | .src_task = cur, | |
1135 | .src_cpu = task_cpu(cur), | |
1136 | .dst_task = p, | |
1137 | .dst_cpu = task_cpu(p), | |
1138 | }; | |
1139 | ||
1140 | if (arg.src_cpu == arg.dst_cpu) | |
1141 | goto out; | |
1142 | ||
6acce3ef PZ |
1143 | /* |
1144 | * These three tests are all lockless; this is OK since all of them | |
1145 | * will be re-checked with proper locks held further down the line. | |
1146 | */ | |
ac66f547 PZ |
1147 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1148 | goto out; | |
1149 | ||
1150 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1151 | goto out; | |
1152 | ||
1153 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1154 | goto out; | |
1155 | ||
286549dc | 1156 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1157 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1158 | ||
1159 | out: | |
ac66f547 PZ |
1160 | return ret; |
1161 | } | |
1162 | ||
969c7921 | 1163 | struct migration_arg { |
36c8b586 | 1164 | struct task_struct *task; |
1da177e4 | 1165 | int dest_cpu; |
70b97a7f | 1166 | }; |
1da177e4 | 1167 | |
969c7921 TH |
1168 | static int migration_cpu_stop(void *data); |
1169 | ||
1da177e4 LT |
1170 | /* |
1171 | * wait_task_inactive - wait for a thread to unschedule. | |
1172 | * | |
85ba2d86 RM |
1173 | * If @match_state is nonzero, it's the @p->state value just checked and |
1174 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1175 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1176 | * we return a positive number (its total switch count). If a second call | |
1177 | * a short while later returns the same number, the caller can be sure that | |
1178 | * @p has remained unscheduled the whole time. | |
1179 | * | |
1da177e4 LT |
1180 | * The caller must ensure that the task *will* unschedule sometime soon, |
1181 | * else this function might spin for a *long* time. This function can't | |
1182 | * be called with interrupts off, or it may introduce deadlock with | |
1183 | * smp_call_function() if an IPI is sent by the same process we are | |
1184 | * waiting to become inactive. | |
1185 | */ | |
85ba2d86 | 1186 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1187 | { |
1188 | unsigned long flags; | |
dd41f596 | 1189 | int running, on_rq; |
85ba2d86 | 1190 | unsigned long ncsw; |
70b97a7f | 1191 | struct rq *rq; |
1da177e4 | 1192 | |
3a5c359a AK |
1193 | for (;;) { |
1194 | /* | |
1195 | * We do the initial early heuristics without holding | |
1196 | * any task-queue locks at all. We'll only try to get | |
1197 | * the runqueue lock when things look like they will | |
1198 | * work out! | |
1199 | */ | |
1200 | rq = task_rq(p); | |
fa490cfd | 1201 | |
3a5c359a AK |
1202 | /* |
1203 | * If the task is actively running on another CPU | |
1204 | * still, just relax and busy-wait without holding | |
1205 | * any locks. | |
1206 | * | |
1207 | * NOTE! Since we don't hold any locks, it's not | |
1208 | * even sure that "rq" stays as the right runqueue! | |
1209 | * But we don't care, since "task_running()" will | |
1210 | * return false if the runqueue has changed and p | |
1211 | * is actually now running somewhere else! | |
1212 | */ | |
85ba2d86 RM |
1213 | while (task_running(rq, p)) { |
1214 | if (match_state && unlikely(p->state != match_state)) | |
1215 | return 0; | |
3a5c359a | 1216 | cpu_relax(); |
85ba2d86 | 1217 | } |
fa490cfd | 1218 | |
3a5c359a AK |
1219 | /* |
1220 | * Ok, time to look more closely! We need the rq | |
1221 | * lock now, to be *sure*. If we're wrong, we'll | |
1222 | * just go back and repeat. | |
1223 | */ | |
1224 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1225 | trace_sched_wait_task(p); |
3a5c359a | 1226 | running = task_running(rq, p); |
fd2f4419 | 1227 | on_rq = p->on_rq; |
85ba2d86 | 1228 | ncsw = 0; |
f31e11d8 | 1229 | if (!match_state || p->state == match_state) |
93dcf55f | 1230 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1231 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1232 | |
85ba2d86 RM |
1233 | /* |
1234 | * If it changed from the expected state, bail out now. | |
1235 | */ | |
1236 | if (unlikely(!ncsw)) | |
1237 | break; | |
1238 | ||
3a5c359a AK |
1239 | /* |
1240 | * Was it really running after all now that we | |
1241 | * checked with the proper locks actually held? | |
1242 | * | |
1243 | * Oops. Go back and try again.. | |
1244 | */ | |
1245 | if (unlikely(running)) { | |
1246 | cpu_relax(); | |
1247 | continue; | |
1248 | } | |
fa490cfd | 1249 | |
3a5c359a AK |
1250 | /* |
1251 | * It's not enough that it's not actively running, | |
1252 | * it must be off the runqueue _entirely_, and not | |
1253 | * preempted! | |
1254 | * | |
80dd99b3 | 1255 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1256 | * running right now), it's preempted, and we should |
1257 | * yield - it could be a while. | |
1258 | */ | |
1259 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1260 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1261 | ||
1262 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1263 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1264 | continue; |
1265 | } | |
fa490cfd | 1266 | |
3a5c359a AK |
1267 | /* |
1268 | * Ahh, all good. It wasn't running, and it wasn't | |
1269 | * runnable, which means that it will never become | |
1270 | * running in the future either. We're all done! | |
1271 | */ | |
1272 | break; | |
1273 | } | |
85ba2d86 RM |
1274 | |
1275 | return ncsw; | |
1da177e4 LT |
1276 | } |
1277 | ||
1278 | /*** | |
1279 | * kick_process - kick a running thread to enter/exit the kernel | |
1280 | * @p: the to-be-kicked thread | |
1281 | * | |
1282 | * Cause a process which is running on another CPU to enter | |
1283 | * kernel-mode, without any delay. (to get signals handled.) | |
1284 | * | |
25985edc | 1285 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1286 | * because all it wants to ensure is that the remote task enters |
1287 | * the kernel. If the IPI races and the task has been migrated | |
1288 | * to another CPU then no harm is done and the purpose has been | |
1289 | * achieved as well. | |
1290 | */ | |
36c8b586 | 1291 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1292 | { |
1293 | int cpu; | |
1294 | ||
1295 | preempt_disable(); | |
1296 | cpu = task_cpu(p); | |
1297 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1298 | smp_send_reschedule(cpu); | |
1299 | preempt_enable(); | |
1300 | } | |
b43e3521 | 1301 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1302 | #endif /* CONFIG_SMP */ |
1da177e4 | 1303 | |
970b13ba | 1304 | #ifdef CONFIG_SMP |
30da688e | 1305 | /* |
013fdb80 | 1306 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1307 | */ |
5da9a0fb PZ |
1308 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1309 | { | |
aa00d89c TC |
1310 | int nid = cpu_to_node(cpu); |
1311 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1312 | enum { cpuset, possible, fail } state = cpuset; |
1313 | int dest_cpu; | |
5da9a0fb | 1314 | |
aa00d89c TC |
1315 | /* |
1316 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1317 | * will return -1. There is no cpu on the node, and we should | |
1318 | * select the cpu on the other node. | |
1319 | */ | |
1320 | if (nid != -1) { | |
1321 | nodemask = cpumask_of_node(nid); | |
1322 | ||
1323 | /* Look for allowed, online CPU in same node. */ | |
1324 | for_each_cpu(dest_cpu, nodemask) { | |
1325 | if (!cpu_online(dest_cpu)) | |
1326 | continue; | |
1327 | if (!cpu_active(dest_cpu)) | |
1328 | continue; | |
1329 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1330 | return dest_cpu; | |
1331 | } | |
2baab4e9 | 1332 | } |
5da9a0fb | 1333 | |
2baab4e9 PZ |
1334 | for (;;) { |
1335 | /* Any allowed, online CPU? */ | |
e3831edd | 1336 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1337 | if (!cpu_online(dest_cpu)) |
1338 | continue; | |
1339 | if (!cpu_active(dest_cpu)) | |
1340 | continue; | |
1341 | goto out; | |
1342 | } | |
5da9a0fb | 1343 | |
2baab4e9 PZ |
1344 | switch (state) { |
1345 | case cpuset: | |
1346 | /* No more Mr. Nice Guy. */ | |
1347 | cpuset_cpus_allowed_fallback(p); | |
1348 | state = possible; | |
1349 | break; | |
1350 | ||
1351 | case possible: | |
1352 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1353 | state = fail; | |
1354 | break; | |
1355 | ||
1356 | case fail: | |
1357 | BUG(); | |
1358 | break; | |
1359 | } | |
1360 | } | |
1361 | ||
1362 | out: | |
1363 | if (state != cpuset) { | |
1364 | /* | |
1365 | * Don't tell them about moving exiting tasks or | |
1366 | * kernel threads (both mm NULL), since they never | |
1367 | * leave kernel. | |
1368 | */ | |
1369 | if (p->mm && printk_ratelimit()) { | |
1370 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1371 | task_pid_nr(p), p->comm, cpu); | |
1372 | } | |
5da9a0fb PZ |
1373 | } |
1374 | ||
1375 | return dest_cpu; | |
1376 | } | |
1377 | ||
e2912009 | 1378 | /* |
013fdb80 | 1379 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1380 | */ |
970b13ba | 1381 | static inline |
ac66f547 | 1382 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1383 | { |
ac66f547 | 1384 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1385 | |
1386 | /* | |
1387 | * In order not to call set_task_cpu() on a blocking task we need | |
1388 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1389 | * cpu. | |
1390 | * | |
1391 | * Since this is common to all placement strategies, this lives here. | |
1392 | * | |
1393 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1394 | * not worry about this generic constraint ] | |
1395 | */ | |
fa17b507 | 1396 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1397 | !cpu_online(cpu))) |
5da9a0fb | 1398 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1399 | |
1400 | return cpu; | |
970b13ba | 1401 | } |
09a40af5 MG |
1402 | |
1403 | static void update_avg(u64 *avg, u64 sample) | |
1404 | { | |
1405 | s64 diff = sample - *avg; | |
1406 | *avg += diff >> 3; | |
1407 | } | |
970b13ba PZ |
1408 | #endif |
1409 | ||
d7c01d27 | 1410 | static void |
b84cb5df | 1411 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1412 | { |
d7c01d27 | 1413 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1414 | struct rq *rq = this_rq(); |
1415 | ||
d7c01d27 PZ |
1416 | #ifdef CONFIG_SMP |
1417 | int this_cpu = smp_processor_id(); | |
1418 | ||
1419 | if (cpu == this_cpu) { | |
1420 | schedstat_inc(rq, ttwu_local); | |
1421 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1422 | } else { | |
1423 | struct sched_domain *sd; | |
1424 | ||
1425 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1426 | rcu_read_lock(); |
d7c01d27 PZ |
1427 | for_each_domain(this_cpu, sd) { |
1428 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1429 | schedstat_inc(sd, ttwu_wake_remote); | |
1430 | break; | |
1431 | } | |
1432 | } | |
057f3fad | 1433 | rcu_read_unlock(); |
d7c01d27 | 1434 | } |
f339b9dc PZ |
1435 | |
1436 | if (wake_flags & WF_MIGRATED) | |
1437 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1438 | ||
d7c01d27 PZ |
1439 | #endif /* CONFIG_SMP */ |
1440 | ||
1441 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1442 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1443 | |
1444 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1445 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1446 | |
d7c01d27 PZ |
1447 | #endif /* CONFIG_SCHEDSTATS */ |
1448 | } | |
1449 | ||
1450 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1451 | { | |
9ed3811a | 1452 | activate_task(rq, p, en_flags); |
fd2f4419 | 1453 | p->on_rq = 1; |
c2f7115e PZ |
1454 | |
1455 | /* if a worker is waking up, notify workqueue */ | |
1456 | if (p->flags & PF_WQ_WORKER) | |
1457 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1458 | } |
1459 | ||
23f41eeb PZ |
1460 | /* |
1461 | * Mark the task runnable and perform wakeup-preemption. | |
1462 | */ | |
89363381 | 1463 | static void |
23f41eeb | 1464 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1465 | { |
9ed3811a | 1466 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1467 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1468 | |
1469 | p->state = TASK_RUNNING; | |
1470 | #ifdef CONFIG_SMP | |
1471 | if (p->sched_class->task_woken) | |
1472 | p->sched_class->task_woken(rq, p); | |
1473 | ||
e69c6341 | 1474 | if (rq->idle_stamp) { |
78becc27 | 1475 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1476 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1477 | |
abfafa54 JL |
1478 | update_avg(&rq->avg_idle, delta); |
1479 | ||
1480 | if (rq->avg_idle > max) | |
9ed3811a | 1481 | rq->avg_idle = max; |
abfafa54 | 1482 | |
9ed3811a TH |
1483 | rq->idle_stamp = 0; |
1484 | } | |
1485 | #endif | |
1486 | } | |
1487 | ||
c05fbafb PZ |
1488 | static void |
1489 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1490 | { | |
1491 | #ifdef CONFIG_SMP | |
1492 | if (p->sched_contributes_to_load) | |
1493 | rq->nr_uninterruptible--; | |
1494 | #endif | |
1495 | ||
1496 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1497 | ttwu_do_wakeup(rq, p, wake_flags); | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1502 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1503 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1504 | * the task is still ->on_rq. | |
1505 | */ | |
1506 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1507 | { | |
1508 | struct rq *rq; | |
1509 | int ret = 0; | |
1510 | ||
1511 | rq = __task_rq_lock(p); | |
1512 | if (p->on_rq) { | |
1ad4ec0d FW |
1513 | /* check_preempt_curr() may use rq clock */ |
1514 | update_rq_clock(rq); | |
c05fbafb PZ |
1515 | ttwu_do_wakeup(rq, p, wake_flags); |
1516 | ret = 1; | |
1517 | } | |
1518 | __task_rq_unlock(rq); | |
1519 | ||
1520 | return ret; | |
1521 | } | |
1522 | ||
317f3941 | 1523 | #ifdef CONFIG_SMP |
fa14ff4a | 1524 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1525 | { |
1526 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1527 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1528 | struct task_struct *p; | |
317f3941 PZ |
1529 | |
1530 | raw_spin_lock(&rq->lock); | |
1531 | ||
fa14ff4a PZ |
1532 | while (llist) { |
1533 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1534 | llist = llist_next(llist); | |
317f3941 PZ |
1535 | ttwu_do_activate(rq, p, 0); |
1536 | } | |
1537 | ||
1538 | raw_spin_unlock(&rq->lock); | |
1539 | } | |
1540 | ||
1541 | void scheduler_ipi(void) | |
1542 | { | |
f27dde8d PZ |
1543 | /* |
1544 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1545 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1546 | * this IPI. | |
1547 | */ | |
8cb75e0c | 1548 | preempt_fold_need_resched(); |
f27dde8d | 1549 | |
873b4c65 VG |
1550 | if (llist_empty(&this_rq()->wake_list) |
1551 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1552 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1553 | return; |
1554 | ||
1555 | /* | |
1556 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1557 | * traditionally all their work was done from the interrupt return | |
1558 | * path. Now that we actually do some work, we need to make sure | |
1559 | * we do call them. | |
1560 | * | |
1561 | * Some archs already do call them, luckily irq_enter/exit nest | |
1562 | * properly. | |
1563 | * | |
1564 | * Arguably we should visit all archs and update all handlers, | |
1565 | * however a fair share of IPIs are still resched only so this would | |
1566 | * somewhat pessimize the simple resched case. | |
1567 | */ | |
1568 | irq_enter(); | |
ff442c51 | 1569 | tick_nohz_full_check(); |
fa14ff4a | 1570 | sched_ttwu_pending(); |
ca38062e SS |
1571 | |
1572 | /* | |
1573 | * Check if someone kicked us for doing the nohz idle load balance. | |
1574 | */ | |
873b4c65 | 1575 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1576 | this_rq()->idle_balance = 1; |
ca38062e | 1577 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1578 | } |
c5d753a5 | 1579 | irq_exit(); |
317f3941 PZ |
1580 | } |
1581 | ||
1582 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1583 | { | |
fa14ff4a | 1584 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1585 | smp_send_reschedule(cpu); |
1586 | } | |
d6aa8f85 | 1587 | |
39be3501 | 1588 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1589 | { |
1590 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1591 | } | |
d6aa8f85 | 1592 | #endif /* CONFIG_SMP */ |
317f3941 | 1593 | |
c05fbafb PZ |
1594 | static void ttwu_queue(struct task_struct *p, int cpu) |
1595 | { | |
1596 | struct rq *rq = cpu_rq(cpu); | |
1597 | ||
17d9f311 | 1598 | #if defined(CONFIG_SMP) |
39be3501 | 1599 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1600 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1601 | ttwu_queue_remote(p, cpu); |
1602 | return; | |
1603 | } | |
1604 | #endif | |
1605 | ||
c05fbafb PZ |
1606 | raw_spin_lock(&rq->lock); |
1607 | ttwu_do_activate(rq, p, 0); | |
1608 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1609 | } |
1610 | ||
1611 | /** | |
1da177e4 | 1612 | * try_to_wake_up - wake up a thread |
9ed3811a | 1613 | * @p: the thread to be awakened |
1da177e4 | 1614 | * @state: the mask of task states that can be woken |
9ed3811a | 1615 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1616 | * |
1617 | * Put it on the run-queue if it's not already there. The "current" | |
1618 | * thread is always on the run-queue (except when the actual | |
1619 | * re-schedule is in progress), and as such you're allowed to do | |
1620 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1621 | * runnable without the overhead of this. | |
1622 | * | |
e69f6186 | 1623 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1624 | * or @state didn't match @p's state. |
1da177e4 | 1625 | */ |
e4a52bcb PZ |
1626 | static int |
1627 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1628 | { |
1da177e4 | 1629 | unsigned long flags; |
c05fbafb | 1630 | int cpu, success = 0; |
2398f2c6 | 1631 | |
e0acd0a6 ON |
1632 | /* |
1633 | * If we are going to wake up a thread waiting for CONDITION we | |
1634 | * need to ensure that CONDITION=1 done by the caller can not be | |
1635 | * reordered with p->state check below. This pairs with mb() in | |
1636 | * set_current_state() the waiting thread does. | |
1637 | */ | |
1638 | smp_mb__before_spinlock(); | |
013fdb80 | 1639 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1640 | if (!(p->state & state)) |
1da177e4 LT |
1641 | goto out; |
1642 | ||
c05fbafb | 1643 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1644 | cpu = task_cpu(p); |
1da177e4 | 1645 | |
c05fbafb PZ |
1646 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1647 | goto stat; | |
1da177e4 | 1648 | |
1da177e4 | 1649 | #ifdef CONFIG_SMP |
e9c84311 | 1650 | /* |
c05fbafb PZ |
1651 | * If the owning (remote) cpu is still in the middle of schedule() with |
1652 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1653 | */ |
f3e94786 | 1654 | while (p->on_cpu) |
e4a52bcb | 1655 | cpu_relax(); |
0970d299 | 1656 | /* |
e4a52bcb | 1657 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1658 | */ |
e4a52bcb | 1659 | smp_rmb(); |
1da177e4 | 1660 | |
a8e4f2ea | 1661 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1662 | p->state = TASK_WAKING; |
e7693a36 | 1663 | |
e4a52bcb | 1664 | if (p->sched_class->task_waking) |
74f8e4b2 | 1665 | p->sched_class->task_waking(p); |
efbbd05a | 1666 | |
ac66f547 | 1667 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1668 | if (task_cpu(p) != cpu) { |
1669 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1670 | set_task_cpu(p, cpu); |
f339b9dc | 1671 | } |
1da177e4 | 1672 | #endif /* CONFIG_SMP */ |
1da177e4 | 1673 | |
c05fbafb PZ |
1674 | ttwu_queue(p, cpu); |
1675 | stat: | |
b84cb5df | 1676 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1677 | out: |
013fdb80 | 1678 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1679 | |
1680 | return success; | |
1681 | } | |
1682 | ||
21aa9af0 TH |
1683 | /** |
1684 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1685 | * @p: the thread to be awakened | |
1686 | * | |
2acca55e | 1687 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1688 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1689 | * the current task. |
21aa9af0 TH |
1690 | */ |
1691 | static void try_to_wake_up_local(struct task_struct *p) | |
1692 | { | |
1693 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1694 | |
383efcd0 TH |
1695 | if (WARN_ON_ONCE(rq != this_rq()) || |
1696 | WARN_ON_ONCE(p == current)) | |
1697 | return; | |
1698 | ||
21aa9af0 TH |
1699 | lockdep_assert_held(&rq->lock); |
1700 | ||
2acca55e PZ |
1701 | if (!raw_spin_trylock(&p->pi_lock)) { |
1702 | raw_spin_unlock(&rq->lock); | |
1703 | raw_spin_lock(&p->pi_lock); | |
1704 | raw_spin_lock(&rq->lock); | |
1705 | } | |
1706 | ||
21aa9af0 | 1707 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1708 | goto out; |
21aa9af0 | 1709 | |
fd2f4419 | 1710 | if (!p->on_rq) |
d7c01d27 PZ |
1711 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1712 | ||
23f41eeb | 1713 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1714 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1715 | out: |
1716 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1717 | } |
1718 | ||
50fa610a DH |
1719 | /** |
1720 | * wake_up_process - Wake up a specific process | |
1721 | * @p: The process to be woken up. | |
1722 | * | |
1723 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1724 | * processes. |
1725 | * | |
1726 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1727 | * |
1728 | * It may be assumed that this function implies a write memory barrier before | |
1729 | * changing the task state if and only if any tasks are woken up. | |
1730 | */ | |
7ad5b3a5 | 1731 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1732 | { |
9067ac85 ON |
1733 | WARN_ON(task_is_stopped_or_traced(p)); |
1734 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1735 | } |
1da177e4 LT |
1736 | EXPORT_SYMBOL(wake_up_process); |
1737 | ||
7ad5b3a5 | 1738 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1739 | { |
1740 | return try_to_wake_up(p, state, 0); | |
1741 | } | |
1742 | ||
1da177e4 LT |
1743 | /* |
1744 | * Perform scheduler related setup for a newly forked process p. | |
1745 | * p is forked by current. | |
dd41f596 IM |
1746 | * |
1747 | * __sched_fork() is basic setup used by init_idle() too: | |
1748 | */ | |
5e1576ed | 1749 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1750 | { |
fd2f4419 PZ |
1751 | p->on_rq = 0; |
1752 | ||
1753 | p->se.on_rq = 0; | |
dd41f596 IM |
1754 | p->se.exec_start = 0; |
1755 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1756 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1757 | p->se.nr_migrations = 0; |
da7a735e | 1758 | p->se.vruntime = 0; |
fd2f4419 | 1759 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1760 | |
1761 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1762 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1763 | #endif |
476d139c | 1764 | |
aab03e05 DF |
1765 | RB_CLEAR_NODE(&p->dl.rb_node); |
1766 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1767 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1768 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1769 | p->dl.dl_period = 0; |
aab03e05 DF |
1770 | p->dl.flags = 0; |
1771 | ||
fa717060 | 1772 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1773 | |
e107be36 AK |
1774 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1775 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1776 | #endif | |
cbee9f88 PZ |
1777 | |
1778 | #ifdef CONFIG_NUMA_BALANCING | |
1779 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1780 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1781 | p->mm->numa_scan_seq = 0; |
1782 | } | |
1783 | ||
5e1576ed RR |
1784 | if (clone_flags & CLONE_VM) |
1785 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1786 | else | |
1787 | p->numa_preferred_nid = -1; | |
1788 | ||
cbee9f88 PZ |
1789 | p->node_stamp = 0ULL; |
1790 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1791 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1792 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1793 | p->numa_faults_memory = NULL; |
1794 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1795 | p->last_task_numa_placement = 0; |
1796 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1797 | |
1798 | INIT_LIST_HEAD(&p->numa_entry); | |
1799 | p->numa_group = NULL; | |
cbee9f88 | 1800 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1801 | } |
1802 | ||
1a687c2e | 1803 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1804 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1805 | void set_numabalancing_state(bool enabled) |
1806 | { | |
1807 | if (enabled) | |
1808 | sched_feat_set("NUMA"); | |
1809 | else | |
1810 | sched_feat_set("NO_NUMA"); | |
1811 | } | |
3105b86a MG |
1812 | #else |
1813 | __read_mostly bool numabalancing_enabled; | |
1814 | ||
1815 | void set_numabalancing_state(bool enabled) | |
1816 | { | |
1817 | numabalancing_enabled = enabled; | |
dd41f596 | 1818 | } |
3105b86a | 1819 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1820 | |
1821 | #ifdef CONFIG_PROC_SYSCTL | |
1822 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1823 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1824 | { | |
1825 | struct ctl_table t; | |
1826 | int err; | |
1827 | int state = numabalancing_enabled; | |
1828 | ||
1829 | if (write && !capable(CAP_SYS_ADMIN)) | |
1830 | return -EPERM; | |
1831 | ||
1832 | t = *table; | |
1833 | t.data = &state; | |
1834 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1835 | if (err < 0) | |
1836 | return err; | |
1837 | if (write) | |
1838 | set_numabalancing_state(state); | |
1839 | return err; | |
1840 | } | |
1841 | #endif | |
1842 | #endif | |
dd41f596 IM |
1843 | |
1844 | /* | |
1845 | * fork()/clone()-time setup: | |
1846 | */ | |
aab03e05 | 1847 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1848 | { |
0122ec5b | 1849 | unsigned long flags; |
dd41f596 IM |
1850 | int cpu = get_cpu(); |
1851 | ||
5e1576ed | 1852 | __sched_fork(clone_flags, p); |
06b83b5f | 1853 | /* |
0017d735 | 1854 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1855 | * nobody will actually run it, and a signal or other external |
1856 | * event cannot wake it up and insert it on the runqueue either. | |
1857 | */ | |
0017d735 | 1858 | p->state = TASK_RUNNING; |
dd41f596 | 1859 | |
c350a04e MG |
1860 | /* |
1861 | * Make sure we do not leak PI boosting priority to the child. | |
1862 | */ | |
1863 | p->prio = current->normal_prio; | |
1864 | ||
b9dc29e7 MG |
1865 | /* |
1866 | * Revert to default priority/policy on fork if requested. | |
1867 | */ | |
1868 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1869 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1870 | p->policy = SCHED_NORMAL; |
6c697bdf | 1871 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1872 | p->rt_priority = 0; |
1873 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1874 | p->static_prio = NICE_TO_PRIO(0); | |
1875 | ||
1876 | p->prio = p->normal_prio = __normal_prio(p); | |
1877 | set_load_weight(p); | |
6c697bdf | 1878 | |
b9dc29e7 MG |
1879 | /* |
1880 | * We don't need the reset flag anymore after the fork. It has | |
1881 | * fulfilled its duty: | |
1882 | */ | |
1883 | p->sched_reset_on_fork = 0; | |
1884 | } | |
ca94c442 | 1885 | |
aab03e05 DF |
1886 | if (dl_prio(p->prio)) { |
1887 | put_cpu(); | |
1888 | return -EAGAIN; | |
1889 | } else if (rt_prio(p->prio)) { | |
1890 | p->sched_class = &rt_sched_class; | |
1891 | } else { | |
2ddbf952 | 1892 | p->sched_class = &fair_sched_class; |
aab03e05 | 1893 | } |
b29739f9 | 1894 | |
cd29fe6f PZ |
1895 | if (p->sched_class->task_fork) |
1896 | p->sched_class->task_fork(p); | |
1897 | ||
86951599 PZ |
1898 | /* |
1899 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1900 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1901 | * is ran before sched_fork(). | |
1902 | * | |
1903 | * Silence PROVE_RCU. | |
1904 | */ | |
0122ec5b | 1905 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1906 | set_task_cpu(p, cpu); |
0122ec5b | 1907 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1908 | |
52f17b6c | 1909 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1910 | if (likely(sched_info_on())) |
52f17b6c | 1911 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1912 | #endif |
3ca7a440 PZ |
1913 | #if defined(CONFIG_SMP) |
1914 | p->on_cpu = 0; | |
4866cde0 | 1915 | #endif |
01028747 | 1916 | init_task_preempt_count(p); |
806c09a7 | 1917 | #ifdef CONFIG_SMP |
917b627d | 1918 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1919 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1920 | #endif |
917b627d | 1921 | |
476d139c | 1922 | put_cpu(); |
aab03e05 | 1923 | return 0; |
1da177e4 LT |
1924 | } |
1925 | ||
332ac17e DF |
1926 | unsigned long to_ratio(u64 period, u64 runtime) |
1927 | { | |
1928 | if (runtime == RUNTIME_INF) | |
1929 | return 1ULL << 20; | |
1930 | ||
1931 | /* | |
1932 | * Doing this here saves a lot of checks in all | |
1933 | * the calling paths, and returning zero seems | |
1934 | * safe for them anyway. | |
1935 | */ | |
1936 | if (period == 0) | |
1937 | return 0; | |
1938 | ||
1939 | return div64_u64(runtime << 20, period); | |
1940 | } | |
1941 | ||
1942 | #ifdef CONFIG_SMP | |
1943 | inline struct dl_bw *dl_bw_of(int i) | |
1944 | { | |
1945 | return &cpu_rq(i)->rd->dl_bw; | |
1946 | } | |
1947 | ||
de212f18 | 1948 | static inline int dl_bw_cpus(int i) |
332ac17e | 1949 | { |
de212f18 PZ |
1950 | struct root_domain *rd = cpu_rq(i)->rd; |
1951 | int cpus = 0; | |
1952 | ||
1953 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1954 | cpus++; | |
1955 | ||
1956 | return cpus; | |
332ac17e DF |
1957 | } |
1958 | #else | |
1959 | inline struct dl_bw *dl_bw_of(int i) | |
1960 | { | |
1961 | return &cpu_rq(i)->dl.dl_bw; | |
1962 | } | |
1963 | ||
de212f18 | 1964 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
1965 | { |
1966 | return 1; | |
1967 | } | |
1968 | #endif | |
1969 | ||
1970 | static inline | |
1971 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
1972 | { | |
1973 | dl_b->total_bw -= tsk_bw; | |
1974 | } | |
1975 | ||
1976 | static inline | |
1977 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
1978 | { | |
1979 | dl_b->total_bw += tsk_bw; | |
1980 | } | |
1981 | ||
1982 | static inline | |
1983 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
1984 | { | |
1985 | return dl_b->bw != -1 && | |
1986 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
1987 | } | |
1988 | ||
1989 | /* | |
1990 | * We must be sure that accepting a new task (or allowing changing the | |
1991 | * parameters of an existing one) is consistent with the bandwidth | |
1992 | * constraints. If yes, this function also accordingly updates the currently | |
1993 | * allocated bandwidth to reflect the new situation. | |
1994 | * | |
1995 | * This function is called while holding p's rq->lock. | |
1996 | */ | |
1997 | static int dl_overflow(struct task_struct *p, int policy, | |
1998 | const struct sched_attr *attr) | |
1999 | { | |
2000 | ||
2001 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2002 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2003 | u64 runtime = attr->sched_runtime; |
2004 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2005 | int cpus, err = -1; |
332ac17e DF |
2006 | |
2007 | if (new_bw == p->dl.dl_bw) | |
2008 | return 0; | |
2009 | ||
2010 | /* | |
2011 | * Either if a task, enters, leave, or stays -deadline but changes | |
2012 | * its parameters, we may need to update accordingly the total | |
2013 | * allocated bandwidth of the container. | |
2014 | */ | |
2015 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2016 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2017 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2018 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2019 | __dl_add(dl_b, new_bw); | |
2020 | err = 0; | |
2021 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2022 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2023 | __dl_clear(dl_b, p->dl.dl_bw); | |
2024 | __dl_add(dl_b, new_bw); | |
2025 | err = 0; | |
2026 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2027 | __dl_clear(dl_b, p->dl.dl_bw); | |
2028 | err = 0; | |
2029 | } | |
2030 | raw_spin_unlock(&dl_b->lock); | |
2031 | ||
2032 | return err; | |
2033 | } | |
2034 | ||
2035 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2036 | ||
1da177e4 LT |
2037 | /* |
2038 | * wake_up_new_task - wake up a newly created task for the first time. | |
2039 | * | |
2040 | * This function will do some initial scheduler statistics housekeeping | |
2041 | * that must be done for every newly created context, then puts the task | |
2042 | * on the runqueue and wakes it. | |
2043 | */ | |
3e51e3ed | 2044 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2045 | { |
2046 | unsigned long flags; | |
dd41f596 | 2047 | struct rq *rq; |
fabf318e | 2048 | |
ab2515c4 | 2049 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2050 | #ifdef CONFIG_SMP |
2051 | /* | |
2052 | * Fork balancing, do it here and not earlier because: | |
2053 | * - cpus_allowed can change in the fork path | |
2054 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2055 | */ |
ac66f547 | 2056 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2057 | #endif |
2058 | ||
a75cdaa9 AS |
2059 | /* Initialize new task's runnable average */ |
2060 | init_task_runnable_average(p); | |
ab2515c4 | 2061 | rq = __task_rq_lock(p); |
cd29fe6f | 2062 | activate_task(rq, p, 0); |
fd2f4419 | 2063 | p->on_rq = 1; |
89363381 | 2064 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2065 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2066 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2067 | if (p->sched_class->task_woken) |
2068 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2069 | #endif |
0122ec5b | 2070 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2071 | } |
2072 | ||
e107be36 AK |
2073 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2074 | ||
2075 | /** | |
80dd99b3 | 2076 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2077 | * @notifier: notifier struct to register |
e107be36 AK |
2078 | */ |
2079 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2080 | { | |
2081 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2082 | } | |
2083 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2084 | ||
2085 | /** | |
2086 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2087 | * @notifier: notifier struct to unregister |
e107be36 AK |
2088 | * |
2089 | * This is safe to call from within a preemption notifier. | |
2090 | */ | |
2091 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2092 | { | |
2093 | hlist_del(¬ifier->link); | |
2094 | } | |
2095 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2096 | ||
2097 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2098 | { | |
2099 | struct preempt_notifier *notifier; | |
e107be36 | 2100 | |
b67bfe0d | 2101 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2102 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2103 | } | |
2104 | ||
2105 | static void | |
2106 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2107 | struct task_struct *next) | |
2108 | { | |
2109 | struct preempt_notifier *notifier; | |
e107be36 | 2110 | |
b67bfe0d | 2111 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2112 | notifier->ops->sched_out(notifier, next); |
2113 | } | |
2114 | ||
6d6bc0ad | 2115 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2116 | |
2117 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2118 | { | |
2119 | } | |
2120 | ||
2121 | static void | |
2122 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2123 | struct task_struct *next) | |
2124 | { | |
2125 | } | |
2126 | ||
6d6bc0ad | 2127 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2128 | |
4866cde0 NP |
2129 | /** |
2130 | * prepare_task_switch - prepare to switch tasks | |
2131 | * @rq: the runqueue preparing to switch | |
421cee29 | 2132 | * @prev: the current task that is being switched out |
4866cde0 NP |
2133 | * @next: the task we are going to switch to. |
2134 | * | |
2135 | * This is called with the rq lock held and interrupts off. It must | |
2136 | * be paired with a subsequent finish_task_switch after the context | |
2137 | * switch. | |
2138 | * | |
2139 | * prepare_task_switch sets up locking and calls architecture specific | |
2140 | * hooks. | |
2141 | */ | |
e107be36 AK |
2142 | static inline void |
2143 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2144 | struct task_struct *next) | |
4866cde0 | 2145 | { |
895dd92c | 2146 | trace_sched_switch(prev, next); |
43148951 | 2147 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2148 | perf_event_task_sched_out(prev, next); |
e107be36 | 2149 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2150 | prepare_lock_switch(rq, next); |
2151 | prepare_arch_switch(next); | |
2152 | } | |
2153 | ||
1da177e4 LT |
2154 | /** |
2155 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2156 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2157 | * @prev: the thread we just switched away from. |
2158 | * | |
4866cde0 NP |
2159 | * finish_task_switch must be called after the context switch, paired |
2160 | * with a prepare_task_switch call before the context switch. | |
2161 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2162 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2163 | * |
2164 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2165 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2166 | * with the lock held can cause deadlocks; see schedule() for |
2167 | * details.) | |
2168 | */ | |
a9957449 | 2169 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2170 | __releases(rq->lock) |
2171 | { | |
1da177e4 | 2172 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2173 | long prev_state; |
1da177e4 LT |
2174 | |
2175 | rq->prev_mm = NULL; | |
2176 | ||
2177 | /* | |
2178 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2179 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2180 | * schedule one last time. The schedule call will never return, and |
2181 | * the scheduled task must drop that reference. | |
c394cc9f | 2182 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2183 | * still held, otherwise prev could be scheduled on another cpu, die |
2184 | * there before we look at prev->state, and then the reference would | |
2185 | * be dropped twice. | |
2186 | * Manfred Spraul <manfred@colorfullife.com> | |
2187 | */ | |
55a101f8 | 2188 | prev_state = prev->state; |
bf9fae9f | 2189 | vtime_task_switch(prev); |
4866cde0 | 2190 | finish_arch_switch(prev); |
a8d757ef | 2191 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2192 | finish_lock_switch(rq, prev); |
01f23e16 | 2193 | finish_arch_post_lock_switch(); |
e8fa1362 | 2194 | |
e107be36 | 2195 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2196 | if (mm) |
2197 | mmdrop(mm); | |
c394cc9f | 2198 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2199 | if (prev->sched_class->task_dead) |
2200 | prev->sched_class->task_dead(prev); | |
2201 | ||
c6fd91f0 | 2202 | /* |
2203 | * Remove function-return probe instances associated with this | |
2204 | * task and put them back on the free list. | |
9761eea8 | 2205 | */ |
c6fd91f0 | 2206 | kprobe_flush_task(prev); |
1da177e4 | 2207 | put_task_struct(prev); |
c6fd91f0 | 2208 | } |
99e5ada9 FW |
2209 | |
2210 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2211 | } |
2212 | ||
3f029d3c GH |
2213 | #ifdef CONFIG_SMP |
2214 | ||
3f029d3c GH |
2215 | /* rq->lock is NOT held, but preemption is disabled */ |
2216 | static inline void post_schedule(struct rq *rq) | |
2217 | { | |
2218 | if (rq->post_schedule) { | |
2219 | unsigned long flags; | |
2220 | ||
05fa785c | 2221 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2222 | if (rq->curr->sched_class->post_schedule) |
2223 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2224 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2225 | |
2226 | rq->post_schedule = 0; | |
2227 | } | |
2228 | } | |
2229 | ||
2230 | #else | |
da19ab51 | 2231 | |
3f029d3c GH |
2232 | static inline void post_schedule(struct rq *rq) |
2233 | { | |
1da177e4 LT |
2234 | } |
2235 | ||
3f029d3c GH |
2236 | #endif |
2237 | ||
1da177e4 LT |
2238 | /** |
2239 | * schedule_tail - first thing a freshly forked thread must call. | |
2240 | * @prev: the thread we just switched away from. | |
2241 | */ | |
722a9f92 | 2242 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2243 | __releases(rq->lock) |
2244 | { | |
70b97a7f IM |
2245 | struct rq *rq = this_rq(); |
2246 | ||
4866cde0 | 2247 | finish_task_switch(rq, prev); |
da19ab51 | 2248 | |
3f029d3c GH |
2249 | /* |
2250 | * FIXME: do we need to worry about rq being invalidated by the | |
2251 | * task_switch? | |
2252 | */ | |
2253 | post_schedule(rq); | |
70b97a7f | 2254 | |
4866cde0 NP |
2255 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2256 | /* In this case, finish_task_switch does not reenable preemption */ | |
2257 | preempt_enable(); | |
2258 | #endif | |
1da177e4 | 2259 | if (current->set_child_tid) |
b488893a | 2260 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2261 | } |
2262 | ||
2263 | /* | |
2264 | * context_switch - switch to the new MM and the new | |
2265 | * thread's register state. | |
2266 | */ | |
dd41f596 | 2267 | static inline void |
70b97a7f | 2268 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2269 | struct task_struct *next) |
1da177e4 | 2270 | { |
dd41f596 | 2271 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2272 | |
e107be36 | 2273 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2274 | |
dd41f596 IM |
2275 | mm = next->mm; |
2276 | oldmm = prev->active_mm; | |
9226d125 ZA |
2277 | /* |
2278 | * For paravirt, this is coupled with an exit in switch_to to | |
2279 | * combine the page table reload and the switch backend into | |
2280 | * one hypercall. | |
2281 | */ | |
224101ed | 2282 | arch_start_context_switch(prev); |
9226d125 | 2283 | |
31915ab4 | 2284 | if (!mm) { |
1da177e4 LT |
2285 | next->active_mm = oldmm; |
2286 | atomic_inc(&oldmm->mm_count); | |
2287 | enter_lazy_tlb(oldmm, next); | |
2288 | } else | |
2289 | switch_mm(oldmm, mm, next); | |
2290 | ||
31915ab4 | 2291 | if (!prev->mm) { |
1da177e4 | 2292 | prev->active_mm = NULL; |
1da177e4 LT |
2293 | rq->prev_mm = oldmm; |
2294 | } | |
3a5f5e48 IM |
2295 | /* |
2296 | * Since the runqueue lock will be released by the next | |
2297 | * task (which is an invalid locking op but in the case | |
2298 | * of the scheduler it's an obvious special-case), so we | |
2299 | * do an early lockdep release here: | |
2300 | */ | |
2301 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2302 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2303 | #endif |
1da177e4 | 2304 | |
91d1aa43 | 2305 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2306 | /* Here we just switch the register state and the stack. */ |
2307 | switch_to(prev, next, prev); | |
2308 | ||
dd41f596 IM |
2309 | barrier(); |
2310 | /* | |
2311 | * this_rq must be evaluated again because prev may have moved | |
2312 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2313 | * frame will be invalid. | |
2314 | */ | |
2315 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2316 | } |
2317 | ||
2318 | /* | |
1c3e8264 | 2319 | * nr_running and nr_context_switches: |
1da177e4 LT |
2320 | * |
2321 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2322 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2323 | */ |
2324 | unsigned long nr_running(void) | |
2325 | { | |
2326 | unsigned long i, sum = 0; | |
2327 | ||
2328 | for_each_online_cpu(i) | |
2329 | sum += cpu_rq(i)->nr_running; | |
2330 | ||
2331 | return sum; | |
f711f609 | 2332 | } |
1da177e4 | 2333 | |
1da177e4 | 2334 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2335 | { |
cc94abfc SR |
2336 | int i; |
2337 | unsigned long long sum = 0; | |
46cb4b7c | 2338 | |
0a945022 | 2339 | for_each_possible_cpu(i) |
1da177e4 | 2340 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2341 | |
1da177e4 LT |
2342 | return sum; |
2343 | } | |
483b4ee6 | 2344 | |
1da177e4 LT |
2345 | unsigned long nr_iowait(void) |
2346 | { | |
2347 | unsigned long i, sum = 0; | |
483b4ee6 | 2348 | |
0a945022 | 2349 | for_each_possible_cpu(i) |
1da177e4 | 2350 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2351 | |
1da177e4 LT |
2352 | return sum; |
2353 | } | |
483b4ee6 | 2354 | |
8c215bd3 | 2355 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2356 | { |
8c215bd3 | 2357 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2358 | return atomic_read(&this->nr_iowait); |
2359 | } | |
46cb4b7c | 2360 | |
dd41f596 | 2361 | #ifdef CONFIG_SMP |
8a0be9ef | 2362 | |
46cb4b7c | 2363 | /* |
38022906 PZ |
2364 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2365 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2366 | */ |
38022906 | 2367 | void sched_exec(void) |
46cb4b7c | 2368 | { |
38022906 | 2369 | struct task_struct *p = current; |
1da177e4 | 2370 | unsigned long flags; |
0017d735 | 2371 | int dest_cpu; |
46cb4b7c | 2372 | |
8f42ced9 | 2373 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2374 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2375 | if (dest_cpu == smp_processor_id()) |
2376 | goto unlock; | |
38022906 | 2377 | |
8f42ced9 | 2378 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2379 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2380 | |
8f42ced9 PZ |
2381 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2382 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2383 | return; |
2384 | } | |
0017d735 | 2385 | unlock: |
8f42ced9 | 2386 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2387 | } |
dd41f596 | 2388 | |
1da177e4 LT |
2389 | #endif |
2390 | ||
1da177e4 | 2391 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2392 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2393 | |
2394 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2395 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2396 | |
2397 | /* | |
c5f8d995 | 2398 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2399 | * @p in case that task is currently running. |
c5f8d995 HS |
2400 | * |
2401 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2402 | */ |
c5f8d995 HS |
2403 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2404 | { | |
2405 | u64 ns = 0; | |
2406 | ||
2407 | if (task_current(rq, p)) { | |
2408 | update_rq_clock(rq); | |
78becc27 | 2409 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2410 | if ((s64)ns < 0) |
2411 | ns = 0; | |
2412 | } | |
2413 | ||
2414 | return ns; | |
2415 | } | |
2416 | ||
bb34d92f | 2417 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2418 | { |
1da177e4 | 2419 | unsigned long flags; |
41b86e9c | 2420 | struct rq *rq; |
bb34d92f | 2421 | u64 ns = 0; |
48f24c4d | 2422 | |
41b86e9c | 2423 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2424 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2425 | task_rq_unlock(rq, p, &flags); |
1508487e | 2426 | |
c5f8d995 HS |
2427 | return ns; |
2428 | } | |
f06febc9 | 2429 | |
c5f8d995 HS |
2430 | /* |
2431 | * Return accounted runtime for the task. | |
2432 | * In case the task is currently running, return the runtime plus current's | |
2433 | * pending runtime that have not been accounted yet. | |
2434 | */ | |
2435 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2436 | { | |
2437 | unsigned long flags; | |
2438 | struct rq *rq; | |
2439 | u64 ns = 0; | |
2440 | ||
911b2898 PZ |
2441 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2442 | /* | |
2443 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2444 | * So we have a optimization chance when the task's delta_exec is 0. | |
2445 | * Reading ->on_cpu is racy, but this is ok. | |
2446 | * | |
2447 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2448 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2449 | * indistinguishable from the read occurring a few cycles earlier. | |
2450 | */ | |
2451 | if (!p->on_cpu) | |
2452 | return p->se.sum_exec_runtime; | |
2453 | #endif | |
2454 | ||
c5f8d995 HS |
2455 | rq = task_rq_lock(p, &flags); |
2456 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2457 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2458 | |
2459 | return ns; | |
2460 | } | |
48f24c4d | 2461 | |
7835b98b CL |
2462 | /* |
2463 | * This function gets called by the timer code, with HZ frequency. | |
2464 | * We call it with interrupts disabled. | |
7835b98b CL |
2465 | */ |
2466 | void scheduler_tick(void) | |
2467 | { | |
7835b98b CL |
2468 | int cpu = smp_processor_id(); |
2469 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2470 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2471 | |
2472 | sched_clock_tick(); | |
dd41f596 | 2473 | |
05fa785c | 2474 | raw_spin_lock(&rq->lock); |
3e51f33f | 2475 | update_rq_clock(rq); |
fa85ae24 | 2476 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2477 | update_cpu_load_active(rq); |
05fa785c | 2478 | raw_spin_unlock(&rq->lock); |
7835b98b | 2479 | |
e9d2b064 | 2480 | perf_event_task_tick(); |
e220d2dc | 2481 | |
e418e1c2 | 2482 | #ifdef CONFIG_SMP |
6eb57e0d | 2483 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2484 | trigger_load_balance(rq); |
e418e1c2 | 2485 | #endif |
265f22a9 | 2486 | rq_last_tick_reset(rq); |
1da177e4 LT |
2487 | } |
2488 | ||
265f22a9 FW |
2489 | #ifdef CONFIG_NO_HZ_FULL |
2490 | /** | |
2491 | * scheduler_tick_max_deferment | |
2492 | * | |
2493 | * Keep at least one tick per second when a single | |
2494 | * active task is running because the scheduler doesn't | |
2495 | * yet completely support full dynticks environment. | |
2496 | * | |
2497 | * This makes sure that uptime, CFS vruntime, load | |
2498 | * balancing, etc... continue to move forward, even | |
2499 | * with a very low granularity. | |
e69f6186 YB |
2500 | * |
2501 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2502 | */ |
2503 | u64 scheduler_tick_max_deferment(void) | |
2504 | { | |
2505 | struct rq *rq = this_rq(); | |
2506 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2507 | ||
2508 | next = rq->last_sched_tick + HZ; | |
2509 | ||
2510 | if (time_before_eq(next, now)) | |
2511 | return 0; | |
2512 | ||
8fe8ff09 | 2513 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2514 | } |
265f22a9 | 2515 | #endif |
1da177e4 | 2516 | |
132380a0 | 2517 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2518 | { |
2519 | if (in_lock_functions(addr)) { | |
2520 | addr = CALLER_ADDR2; | |
2521 | if (in_lock_functions(addr)) | |
2522 | addr = CALLER_ADDR3; | |
2523 | } | |
2524 | return addr; | |
2525 | } | |
1da177e4 | 2526 | |
7e49fcce SR |
2527 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2528 | defined(CONFIG_PREEMPT_TRACER)) | |
2529 | ||
bdb43806 | 2530 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2531 | { |
6cd8a4bb | 2532 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2533 | /* |
2534 | * Underflow? | |
2535 | */ | |
9a11b49a IM |
2536 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2537 | return; | |
6cd8a4bb | 2538 | #endif |
bdb43806 | 2539 | __preempt_count_add(val); |
6cd8a4bb | 2540 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2541 | /* |
2542 | * Spinlock count overflowing soon? | |
2543 | */ | |
33859f7f MOS |
2544 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2545 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2546 | #endif |
8f47b187 TG |
2547 | if (preempt_count() == val) { |
2548 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2549 | #ifdef CONFIG_DEBUG_PREEMPT | |
2550 | current->preempt_disable_ip = ip; | |
2551 | #endif | |
2552 | trace_preempt_off(CALLER_ADDR0, ip); | |
2553 | } | |
1da177e4 | 2554 | } |
bdb43806 | 2555 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2556 | |
bdb43806 | 2557 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2558 | { |
6cd8a4bb | 2559 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2560 | /* |
2561 | * Underflow? | |
2562 | */ | |
01e3eb82 | 2563 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2564 | return; |
1da177e4 LT |
2565 | /* |
2566 | * Is the spinlock portion underflowing? | |
2567 | */ | |
9a11b49a IM |
2568 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2569 | !(preempt_count() & PREEMPT_MASK))) | |
2570 | return; | |
6cd8a4bb | 2571 | #endif |
9a11b49a | 2572 | |
6cd8a4bb SR |
2573 | if (preempt_count() == val) |
2574 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2575 | __preempt_count_sub(val); |
1da177e4 | 2576 | } |
bdb43806 | 2577 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2578 | |
2579 | #endif | |
2580 | ||
2581 | /* | |
dd41f596 | 2582 | * Print scheduling while atomic bug: |
1da177e4 | 2583 | */ |
dd41f596 | 2584 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2585 | { |
664dfa65 DJ |
2586 | if (oops_in_progress) |
2587 | return; | |
2588 | ||
3df0fc5b PZ |
2589 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2590 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2591 | |
dd41f596 | 2592 | debug_show_held_locks(prev); |
e21f5b15 | 2593 | print_modules(); |
dd41f596 IM |
2594 | if (irqs_disabled()) |
2595 | print_irqtrace_events(prev); | |
8f47b187 TG |
2596 | #ifdef CONFIG_DEBUG_PREEMPT |
2597 | if (in_atomic_preempt_off()) { | |
2598 | pr_err("Preemption disabled at:"); | |
2599 | print_ip_sym(current->preempt_disable_ip); | |
2600 | pr_cont("\n"); | |
2601 | } | |
2602 | #endif | |
6135fc1e | 2603 | dump_stack(); |
373d4d09 | 2604 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2605 | } |
1da177e4 | 2606 | |
dd41f596 IM |
2607 | /* |
2608 | * Various schedule()-time debugging checks and statistics: | |
2609 | */ | |
2610 | static inline void schedule_debug(struct task_struct *prev) | |
2611 | { | |
1da177e4 | 2612 | /* |
41a2d6cf | 2613 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2614 | * schedule() atomically, we ignore that path. Otherwise whine |
2615 | * if we are scheduling when we should not. | |
1da177e4 | 2616 | */ |
192301e7 | 2617 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2618 | __schedule_bug(prev); |
b3fbab05 | 2619 | rcu_sleep_check(); |
dd41f596 | 2620 | |
1da177e4 LT |
2621 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2622 | ||
2d72376b | 2623 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2624 | } |
2625 | ||
2626 | /* | |
2627 | * Pick up the highest-prio task: | |
2628 | */ | |
2629 | static inline struct task_struct * | |
606dba2e | 2630 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2631 | { |
37e117c0 | 2632 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2633 | struct task_struct *p; |
1da177e4 LT |
2634 | |
2635 | /* | |
dd41f596 IM |
2636 | * Optimization: we know that if all tasks are in |
2637 | * the fair class we can call that function directly: | |
1da177e4 | 2638 | */ |
37e117c0 | 2639 | if (likely(prev->sched_class == class && |
38033c37 | 2640 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2641 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2642 | if (unlikely(p == RETRY_TASK)) |
2643 | goto again; | |
2644 | ||
2645 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2646 | if (unlikely(!p)) | |
2647 | p = idle_sched_class.pick_next_task(rq, prev); | |
2648 | ||
2649 | return p; | |
1da177e4 LT |
2650 | } |
2651 | ||
37e117c0 | 2652 | again: |
34f971f6 | 2653 | for_each_class(class) { |
606dba2e | 2654 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2655 | if (p) { |
2656 | if (unlikely(p == RETRY_TASK)) | |
2657 | goto again; | |
dd41f596 | 2658 | return p; |
37e117c0 | 2659 | } |
dd41f596 | 2660 | } |
34f971f6 PZ |
2661 | |
2662 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2663 | } |
1da177e4 | 2664 | |
dd41f596 | 2665 | /* |
c259e01a | 2666 | * __schedule() is the main scheduler function. |
edde96ea PE |
2667 | * |
2668 | * The main means of driving the scheduler and thus entering this function are: | |
2669 | * | |
2670 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2671 | * | |
2672 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2673 | * paths. For example, see arch/x86/entry_64.S. | |
2674 | * | |
2675 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2676 | * interrupt handler scheduler_tick(). | |
2677 | * | |
2678 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2679 | * task to the run-queue and that's it. | |
2680 | * | |
2681 | * Now, if the new task added to the run-queue preempts the current | |
2682 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2683 | * called on the nearest possible occasion: | |
2684 | * | |
2685 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2686 | * | |
2687 | * - in syscall or exception context, at the next outmost | |
2688 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2689 | * spin_unlock()!) | |
2690 | * | |
2691 | * - in IRQ context, return from interrupt-handler to | |
2692 | * preemptible context | |
2693 | * | |
2694 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2695 | * then at the next: | |
2696 | * | |
2697 | * - cond_resched() call | |
2698 | * - explicit schedule() call | |
2699 | * - return from syscall or exception to user-space | |
2700 | * - return from interrupt-handler to user-space | |
dd41f596 | 2701 | */ |
c259e01a | 2702 | static void __sched __schedule(void) |
dd41f596 IM |
2703 | { |
2704 | struct task_struct *prev, *next; | |
67ca7bde | 2705 | unsigned long *switch_count; |
dd41f596 | 2706 | struct rq *rq; |
31656519 | 2707 | int cpu; |
dd41f596 | 2708 | |
ff743345 PZ |
2709 | need_resched: |
2710 | preempt_disable(); | |
dd41f596 IM |
2711 | cpu = smp_processor_id(); |
2712 | rq = cpu_rq(cpu); | |
25502a6c | 2713 | rcu_note_context_switch(cpu); |
dd41f596 | 2714 | prev = rq->curr; |
dd41f596 | 2715 | |
dd41f596 | 2716 | schedule_debug(prev); |
1da177e4 | 2717 | |
31656519 | 2718 | if (sched_feat(HRTICK)) |
f333fdc9 | 2719 | hrtick_clear(rq); |
8f4d37ec | 2720 | |
e0acd0a6 ON |
2721 | /* |
2722 | * Make sure that signal_pending_state()->signal_pending() below | |
2723 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2724 | * done by the caller to avoid the race with signal_wake_up(). | |
2725 | */ | |
2726 | smp_mb__before_spinlock(); | |
05fa785c | 2727 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2728 | |
246d86b5 | 2729 | switch_count = &prev->nivcsw; |
1da177e4 | 2730 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2731 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2732 | prev->state = TASK_RUNNING; |
21aa9af0 | 2733 | } else { |
2acca55e PZ |
2734 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2735 | prev->on_rq = 0; | |
2736 | ||
21aa9af0 | 2737 | /* |
2acca55e PZ |
2738 | * If a worker went to sleep, notify and ask workqueue |
2739 | * whether it wants to wake up a task to maintain | |
2740 | * concurrency. | |
21aa9af0 TH |
2741 | */ |
2742 | if (prev->flags & PF_WQ_WORKER) { | |
2743 | struct task_struct *to_wakeup; | |
2744 | ||
2745 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2746 | if (to_wakeup) | |
2747 | try_to_wake_up_local(to_wakeup); | |
2748 | } | |
21aa9af0 | 2749 | } |
dd41f596 | 2750 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2751 | } |
2752 | ||
606dba2e PZ |
2753 | if (prev->on_rq || rq->skip_clock_update < 0) |
2754 | update_rq_clock(rq); | |
2755 | ||
2756 | next = pick_next_task(rq, prev); | |
f26f9aff | 2757 | clear_tsk_need_resched(prev); |
f27dde8d | 2758 | clear_preempt_need_resched(); |
f26f9aff | 2759 | rq->skip_clock_update = 0; |
1da177e4 | 2760 | |
1da177e4 | 2761 | if (likely(prev != next)) { |
1da177e4 LT |
2762 | rq->nr_switches++; |
2763 | rq->curr = next; | |
2764 | ++*switch_count; | |
2765 | ||
dd41f596 | 2766 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2767 | /* |
246d86b5 ON |
2768 | * The context switch have flipped the stack from under us |
2769 | * and restored the local variables which were saved when | |
2770 | * this task called schedule() in the past. prev == current | |
2771 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2772 | */ |
2773 | cpu = smp_processor_id(); | |
2774 | rq = cpu_rq(cpu); | |
1da177e4 | 2775 | } else |
05fa785c | 2776 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2777 | |
3f029d3c | 2778 | post_schedule(rq); |
1da177e4 | 2779 | |
ba74c144 | 2780 | sched_preempt_enable_no_resched(); |
ff743345 | 2781 | if (need_resched()) |
1da177e4 LT |
2782 | goto need_resched; |
2783 | } | |
c259e01a | 2784 | |
9c40cef2 TG |
2785 | static inline void sched_submit_work(struct task_struct *tsk) |
2786 | { | |
3c7d5184 | 2787 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2788 | return; |
2789 | /* | |
2790 | * If we are going to sleep and we have plugged IO queued, | |
2791 | * make sure to submit it to avoid deadlocks. | |
2792 | */ | |
2793 | if (blk_needs_flush_plug(tsk)) | |
2794 | blk_schedule_flush_plug(tsk); | |
2795 | } | |
2796 | ||
722a9f92 | 2797 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2798 | { |
9c40cef2 TG |
2799 | struct task_struct *tsk = current; |
2800 | ||
2801 | sched_submit_work(tsk); | |
c259e01a TG |
2802 | __schedule(); |
2803 | } | |
1da177e4 LT |
2804 | EXPORT_SYMBOL(schedule); |
2805 | ||
91d1aa43 | 2806 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2807 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2808 | { |
2809 | /* | |
2810 | * If we come here after a random call to set_need_resched(), | |
2811 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2812 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2813 | * we find a better solution. | |
2814 | */ | |
91d1aa43 | 2815 | user_exit(); |
20ab65e3 | 2816 | schedule(); |
91d1aa43 | 2817 | user_enter(); |
20ab65e3 FW |
2818 | } |
2819 | #endif | |
2820 | ||
c5491ea7 TG |
2821 | /** |
2822 | * schedule_preempt_disabled - called with preemption disabled | |
2823 | * | |
2824 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2825 | */ | |
2826 | void __sched schedule_preempt_disabled(void) | |
2827 | { | |
ba74c144 | 2828 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2829 | schedule(); |
2830 | preempt_disable(); | |
2831 | } | |
2832 | ||
1da177e4 LT |
2833 | #ifdef CONFIG_PREEMPT |
2834 | /* | |
2ed6e34f | 2835 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2836 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2837 | * occur there and call schedule directly. |
2838 | */ | |
722a9f92 | 2839 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2840 | { |
1da177e4 LT |
2841 | /* |
2842 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2843 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2844 | */ |
fbb00b56 | 2845 | if (likely(!preemptible())) |
1da177e4 LT |
2846 | return; |
2847 | ||
3a5c359a | 2848 | do { |
bdb43806 | 2849 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2850 | __schedule(); |
bdb43806 | 2851 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2852 | |
3a5c359a AK |
2853 | /* |
2854 | * Check again in case we missed a preemption opportunity | |
2855 | * between schedule and now. | |
2856 | */ | |
2857 | barrier(); | |
5ed0cec0 | 2858 | } while (need_resched()); |
1da177e4 | 2859 | } |
1da177e4 | 2860 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2861 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2862 | |
2863 | /* | |
2ed6e34f | 2864 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2865 | * off of irq context. |
2866 | * Note, that this is called and return with irqs disabled. This will | |
2867 | * protect us against recursive calling from irq. | |
2868 | */ | |
722a9f92 | 2869 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2870 | { |
b22366cd | 2871 | enum ctx_state prev_state; |
6478d880 | 2872 | |
2ed6e34f | 2873 | /* Catch callers which need to be fixed */ |
f27dde8d | 2874 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2875 | |
b22366cd FW |
2876 | prev_state = exception_enter(); |
2877 | ||
3a5c359a | 2878 | do { |
bdb43806 | 2879 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2880 | local_irq_enable(); |
c259e01a | 2881 | __schedule(); |
3a5c359a | 2882 | local_irq_disable(); |
bdb43806 | 2883 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2884 | |
3a5c359a AK |
2885 | /* |
2886 | * Check again in case we missed a preemption opportunity | |
2887 | * between schedule and now. | |
2888 | */ | |
2889 | barrier(); | |
5ed0cec0 | 2890 | } while (need_resched()); |
b22366cd FW |
2891 | |
2892 | exception_exit(prev_state); | |
1da177e4 LT |
2893 | } |
2894 | ||
63859d4f | 2895 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2896 | void *key) |
1da177e4 | 2897 | { |
63859d4f | 2898 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2899 | } |
1da177e4 LT |
2900 | EXPORT_SYMBOL(default_wake_function); |
2901 | ||
b29739f9 IM |
2902 | #ifdef CONFIG_RT_MUTEXES |
2903 | ||
2904 | /* | |
2905 | * rt_mutex_setprio - set the current priority of a task | |
2906 | * @p: task | |
2907 | * @prio: prio value (kernel-internal form) | |
2908 | * | |
2909 | * This function changes the 'effective' priority of a task. It does | |
2910 | * not touch ->normal_prio like __setscheduler(). | |
2911 | * | |
c365c292 TG |
2912 | * Used by the rt_mutex code to implement priority inheritance |
2913 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2914 | */ |
36c8b586 | 2915 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2916 | { |
2d3d891d | 2917 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2918 | struct rq *rq; |
83ab0aa0 | 2919 | const struct sched_class *prev_class; |
b29739f9 | 2920 | |
aab03e05 | 2921 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2922 | |
0122ec5b | 2923 | rq = __task_rq_lock(p); |
b29739f9 | 2924 | |
1c4dd99b TG |
2925 | /* |
2926 | * Idle task boosting is a nono in general. There is one | |
2927 | * exception, when PREEMPT_RT and NOHZ is active: | |
2928 | * | |
2929 | * The idle task calls get_next_timer_interrupt() and holds | |
2930 | * the timer wheel base->lock on the CPU and another CPU wants | |
2931 | * to access the timer (probably to cancel it). We can safely | |
2932 | * ignore the boosting request, as the idle CPU runs this code | |
2933 | * with interrupts disabled and will complete the lock | |
2934 | * protected section without being interrupted. So there is no | |
2935 | * real need to boost. | |
2936 | */ | |
2937 | if (unlikely(p == rq->idle)) { | |
2938 | WARN_ON(p != rq->curr); | |
2939 | WARN_ON(p->pi_blocked_on); | |
2940 | goto out_unlock; | |
2941 | } | |
2942 | ||
a8027073 | 2943 | trace_sched_pi_setprio(p, prio); |
2d3d891d | 2944 | p->pi_top_task = rt_mutex_get_top_task(p); |
d5f9f942 | 2945 | oldprio = p->prio; |
83ab0aa0 | 2946 | prev_class = p->sched_class; |
fd2f4419 | 2947 | on_rq = p->on_rq; |
051a1d1a | 2948 | running = task_current(rq, p); |
0e1f3483 | 2949 | if (on_rq) |
69be72c1 | 2950 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2951 | if (running) |
2952 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2953 | |
2d3d891d DF |
2954 | /* |
2955 | * Boosting condition are: | |
2956 | * 1. -rt task is running and holds mutex A | |
2957 | * --> -dl task blocks on mutex A | |
2958 | * | |
2959 | * 2. -dl task is running and holds mutex A | |
2960 | * --> -dl task blocks on mutex A and could preempt the | |
2961 | * running task | |
2962 | */ | |
2963 | if (dl_prio(prio)) { | |
2964 | if (!dl_prio(p->normal_prio) || (p->pi_top_task && | |
2965 | dl_entity_preempt(&p->pi_top_task->dl, &p->dl))) { | |
2966 | p->dl.dl_boosted = 1; | |
2967 | p->dl.dl_throttled = 0; | |
2968 | enqueue_flag = ENQUEUE_REPLENISH; | |
2969 | } else | |
2970 | p->dl.dl_boosted = 0; | |
aab03e05 | 2971 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
2972 | } else if (rt_prio(prio)) { |
2973 | if (dl_prio(oldprio)) | |
2974 | p->dl.dl_boosted = 0; | |
2975 | if (oldprio < prio) | |
2976 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 2977 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
2978 | } else { |
2979 | if (dl_prio(oldprio)) | |
2980 | p->dl.dl_boosted = 0; | |
dd41f596 | 2981 | p->sched_class = &fair_sched_class; |
2d3d891d | 2982 | } |
dd41f596 | 2983 | |
b29739f9 IM |
2984 | p->prio = prio; |
2985 | ||
0e1f3483 HS |
2986 | if (running) |
2987 | p->sched_class->set_curr_task(rq); | |
da7a735e | 2988 | if (on_rq) |
2d3d891d | 2989 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 2990 | |
da7a735e | 2991 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 2992 | out_unlock: |
0122ec5b | 2993 | __task_rq_unlock(rq); |
b29739f9 | 2994 | } |
b29739f9 | 2995 | #endif |
d50dde5a | 2996 | |
36c8b586 | 2997 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 2998 | { |
dd41f596 | 2999 | int old_prio, delta, on_rq; |
1da177e4 | 3000 | unsigned long flags; |
70b97a7f | 3001 | struct rq *rq; |
1da177e4 | 3002 | |
75e45d51 | 3003 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3004 | return; |
3005 | /* | |
3006 | * We have to be careful, if called from sys_setpriority(), | |
3007 | * the task might be in the middle of scheduling on another CPU. | |
3008 | */ | |
3009 | rq = task_rq_lock(p, &flags); | |
3010 | /* | |
3011 | * The RT priorities are set via sched_setscheduler(), but we still | |
3012 | * allow the 'normal' nice value to be set - but as expected | |
3013 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3014 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3015 | */ |
aab03e05 | 3016 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3017 | p->static_prio = NICE_TO_PRIO(nice); |
3018 | goto out_unlock; | |
3019 | } | |
fd2f4419 | 3020 | on_rq = p->on_rq; |
c09595f6 | 3021 | if (on_rq) |
69be72c1 | 3022 | dequeue_task(rq, p, 0); |
1da177e4 | 3023 | |
1da177e4 | 3024 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3025 | set_load_weight(p); |
b29739f9 IM |
3026 | old_prio = p->prio; |
3027 | p->prio = effective_prio(p); | |
3028 | delta = p->prio - old_prio; | |
1da177e4 | 3029 | |
dd41f596 | 3030 | if (on_rq) { |
371fd7e7 | 3031 | enqueue_task(rq, p, 0); |
1da177e4 | 3032 | /* |
d5f9f942 AM |
3033 | * If the task increased its priority or is running and |
3034 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3035 | */ |
d5f9f942 | 3036 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3037 | resched_task(rq->curr); |
3038 | } | |
3039 | out_unlock: | |
0122ec5b | 3040 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3041 | } |
1da177e4 LT |
3042 | EXPORT_SYMBOL(set_user_nice); |
3043 | ||
e43379f1 MM |
3044 | /* |
3045 | * can_nice - check if a task can reduce its nice value | |
3046 | * @p: task | |
3047 | * @nice: nice value | |
3048 | */ | |
36c8b586 | 3049 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3050 | { |
024f4747 | 3051 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3052 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3053 | |
78d7d407 | 3054 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3055 | capable(CAP_SYS_NICE)); |
3056 | } | |
3057 | ||
1da177e4 LT |
3058 | #ifdef __ARCH_WANT_SYS_NICE |
3059 | ||
3060 | /* | |
3061 | * sys_nice - change the priority of the current process. | |
3062 | * @increment: priority increment | |
3063 | * | |
3064 | * sys_setpriority is a more generic, but much slower function that | |
3065 | * does similar things. | |
3066 | */ | |
5add95d4 | 3067 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3068 | { |
48f24c4d | 3069 | long nice, retval; |
1da177e4 LT |
3070 | |
3071 | /* | |
3072 | * Setpriority might change our priority at the same moment. | |
3073 | * We don't have to worry. Conceptually one call occurs first | |
3074 | * and we have a single winner. | |
3075 | */ | |
a9467fa3 | 3076 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3077 | nice = task_nice(current) + increment; |
1da177e4 | 3078 | |
a9467fa3 | 3079 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3080 | if (increment < 0 && !can_nice(current, nice)) |
3081 | return -EPERM; | |
3082 | ||
1da177e4 LT |
3083 | retval = security_task_setnice(current, nice); |
3084 | if (retval) | |
3085 | return retval; | |
3086 | ||
3087 | set_user_nice(current, nice); | |
3088 | return 0; | |
3089 | } | |
3090 | ||
3091 | #endif | |
3092 | ||
3093 | /** | |
3094 | * task_prio - return the priority value of a given task. | |
3095 | * @p: the task in question. | |
3096 | * | |
e69f6186 | 3097 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3098 | * RT tasks are offset by -200. Normal tasks are centered |
3099 | * around 0, value goes from -16 to +15. | |
3100 | */ | |
36c8b586 | 3101 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3102 | { |
3103 | return p->prio - MAX_RT_PRIO; | |
3104 | } | |
3105 | ||
1da177e4 LT |
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; | |
5bfd126e | 3173 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3174 | } |
3175 | ||
c365c292 TG |
3176 | static void __setscheduler_params(struct task_struct *p, |
3177 | const struct sched_attr *attr) | |
1da177e4 | 3178 | { |
d50dde5a DF |
3179 | int policy = attr->sched_policy; |
3180 | ||
39fd8fd2 PZ |
3181 | if (policy == -1) /* setparam */ |
3182 | policy = p->policy; | |
3183 | ||
1da177e4 | 3184 | p->policy = policy; |
d50dde5a | 3185 | |
aab03e05 DF |
3186 | if (dl_policy(policy)) |
3187 | __setparam_dl(p, attr); | |
39fd8fd2 | 3188 | else if (fair_policy(policy)) |
d50dde5a DF |
3189 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3190 | ||
39fd8fd2 PZ |
3191 | /* |
3192 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3193 | * !rt_policy. Always setting this ensures that things like | |
3194 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3195 | */ | |
3196 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3197 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3198 | set_load_weight(p); |
3199 | } | |
39fd8fd2 | 3200 | |
c365c292 TG |
3201 | /* Actually do priority change: must hold pi & rq lock. */ |
3202 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3203 | const struct sched_attr *attr) | |
3204 | { | |
3205 | __setscheduler_params(p, attr); | |
d50dde5a | 3206 | |
383afd09 SR |
3207 | /* |
3208 | * If we get here, there was no pi waiters boosting the | |
3209 | * task. It is safe to use the normal prio. | |
3210 | */ | |
3211 | p->prio = normal_prio(p); | |
3212 | ||
aab03e05 DF |
3213 | if (dl_prio(p->prio)) |
3214 | p->sched_class = &dl_sched_class; | |
3215 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3216 | p->sched_class = &rt_sched_class; |
3217 | else | |
3218 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3219 | } |
aab03e05 DF |
3220 | |
3221 | static void | |
3222 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3223 | { | |
3224 | struct sched_dl_entity *dl_se = &p->dl; | |
3225 | ||
3226 | attr->sched_priority = p->rt_priority; | |
3227 | attr->sched_runtime = dl_se->dl_runtime; | |
3228 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3229 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3230 | attr->sched_flags = dl_se->flags; |
3231 | } | |
3232 | ||
3233 | /* | |
3234 | * This function validates the new parameters of a -deadline task. | |
3235 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3236 | * than the runtime, as well as the period of being zero or |
332ac17e | 3237 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3238 | * user parameters are above the internal resolution of 1us (we |
3239 | * check sched_runtime only since it is always the smaller one) and | |
3240 | * below 2^63 ns (we have to check both sched_deadline and | |
3241 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3242 | */ |
3243 | static bool | |
3244 | __checkparam_dl(const struct sched_attr *attr) | |
3245 | { | |
b0827819 JL |
3246 | /* deadline != 0 */ |
3247 | if (attr->sched_deadline == 0) | |
3248 | return false; | |
3249 | ||
3250 | /* | |
3251 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3252 | * that big. | |
3253 | */ | |
3254 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3255 | return false; | |
3256 | ||
3257 | /* | |
3258 | * Since we use the MSB for wrap-around and sign issues, make | |
3259 | * sure it's not set (mind that period can be equal to zero). | |
3260 | */ | |
3261 | if (attr->sched_deadline & (1ULL << 63) || | |
3262 | attr->sched_period & (1ULL << 63)) | |
3263 | return false; | |
3264 | ||
3265 | /* runtime <= deadline <= period (if period != 0) */ | |
3266 | if ((attr->sched_period != 0 && | |
3267 | attr->sched_period < attr->sched_deadline) || | |
3268 | attr->sched_deadline < attr->sched_runtime) | |
3269 | return false; | |
3270 | ||
3271 | return true; | |
aab03e05 DF |
3272 | } |
3273 | ||
c69e8d9c DH |
3274 | /* |
3275 | * check the target process has a UID that matches the current process's | |
3276 | */ | |
3277 | static bool check_same_owner(struct task_struct *p) | |
3278 | { | |
3279 | const struct cred *cred = current_cred(), *pcred; | |
3280 | bool match; | |
3281 | ||
3282 | rcu_read_lock(); | |
3283 | pcred = __task_cred(p); | |
9c806aa0 EB |
3284 | match = (uid_eq(cred->euid, pcred->euid) || |
3285 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3286 | rcu_read_unlock(); |
3287 | return match; | |
3288 | } | |
3289 | ||
d50dde5a DF |
3290 | static int __sched_setscheduler(struct task_struct *p, |
3291 | const struct sched_attr *attr, | |
3292 | bool user) | |
1da177e4 | 3293 | { |
383afd09 SR |
3294 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3295 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
83b699ed | 3296 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3297 | int policy = attr->sched_policy; |
1da177e4 | 3298 | unsigned long flags; |
83ab0aa0 | 3299 | const struct sched_class *prev_class; |
70b97a7f | 3300 | struct rq *rq; |
ca94c442 | 3301 | int reset_on_fork; |
1da177e4 | 3302 | |
66e5393a SR |
3303 | /* may grab non-irq protected spin_locks */ |
3304 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3305 | recheck: |
3306 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3307 | if (policy < 0) { |
3308 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3309 | policy = oldpolicy = p->policy; |
ca94c442 | 3310 | } else { |
7479f3c9 | 3311 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3312 | |
aab03e05 DF |
3313 | if (policy != SCHED_DEADLINE && |
3314 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3315 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3316 | policy != SCHED_IDLE) | |
3317 | return -EINVAL; | |
3318 | } | |
3319 | ||
7479f3c9 PZ |
3320 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3321 | return -EINVAL; | |
3322 | ||
1da177e4 LT |
3323 | /* |
3324 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3325 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3326 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3327 | */ |
0bb040a4 | 3328 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3329 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3330 | return -EINVAL; |
aab03e05 DF |
3331 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3332 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3333 | return -EINVAL; |
3334 | ||
37e4ab3f OC |
3335 | /* |
3336 | * Allow unprivileged RT tasks to decrease priority: | |
3337 | */ | |
961ccddd | 3338 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3339 | if (fair_policy(policy)) { |
d0ea0268 | 3340 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3341 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3342 | return -EPERM; |
3343 | } | |
3344 | ||
e05606d3 | 3345 | if (rt_policy(policy)) { |
a44702e8 ON |
3346 | unsigned long rlim_rtprio = |
3347 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3348 | |
3349 | /* can't set/change the rt policy */ | |
3350 | if (policy != p->policy && !rlim_rtprio) | |
3351 | return -EPERM; | |
3352 | ||
3353 | /* can't increase priority */ | |
d50dde5a DF |
3354 | if (attr->sched_priority > p->rt_priority && |
3355 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3356 | return -EPERM; |
3357 | } | |
c02aa73b | 3358 | |
d44753b8 JL |
3359 | /* |
3360 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3361 | * (safest behavior); in the future we would like to allow | |
3362 | * unprivileged DL tasks to increase their relative deadline | |
3363 | * or reduce their runtime (both ways reducing utilization) | |
3364 | */ | |
3365 | if (dl_policy(policy)) | |
3366 | return -EPERM; | |
3367 | ||
dd41f596 | 3368 | /* |
c02aa73b DH |
3369 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3370 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3371 | */ |
c02aa73b | 3372 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3373 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3374 | return -EPERM; |
3375 | } | |
5fe1d75f | 3376 | |
37e4ab3f | 3377 | /* can't change other user's priorities */ |
c69e8d9c | 3378 | if (!check_same_owner(p)) |
37e4ab3f | 3379 | return -EPERM; |
ca94c442 LP |
3380 | |
3381 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3382 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3383 | return -EPERM; | |
37e4ab3f | 3384 | } |
1da177e4 | 3385 | |
725aad24 | 3386 | if (user) { |
b0ae1981 | 3387 | retval = security_task_setscheduler(p); |
725aad24 JF |
3388 | if (retval) |
3389 | return retval; | |
3390 | } | |
3391 | ||
b29739f9 IM |
3392 | /* |
3393 | * make sure no PI-waiters arrive (or leave) while we are | |
3394 | * changing the priority of the task: | |
0122ec5b | 3395 | * |
25985edc | 3396 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3397 | * runqueue lock must be held. |
3398 | */ | |
0122ec5b | 3399 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3400 | |
34f971f6 PZ |
3401 | /* |
3402 | * Changing the policy of the stop threads its a very bad idea | |
3403 | */ | |
3404 | if (p == rq->stop) { | |
0122ec5b | 3405 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3406 | return -EINVAL; |
3407 | } | |
3408 | ||
a51e9198 | 3409 | /* |
d6b1e911 TG |
3410 | * If not changing anything there's no need to proceed further, |
3411 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3412 | */ |
d50dde5a | 3413 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3414 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3415 | goto change; |
3416 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3417 | goto change; | |
aab03e05 DF |
3418 | if (dl_policy(policy)) |
3419 | goto change; | |
d50dde5a | 3420 | |
d6b1e911 | 3421 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3422 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3423 | return 0; |
3424 | } | |
d50dde5a | 3425 | change: |
a51e9198 | 3426 | |
dc61b1d6 | 3427 | if (user) { |
332ac17e | 3428 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3429 | /* |
3430 | * Do not allow realtime tasks into groups that have no runtime | |
3431 | * assigned. | |
3432 | */ | |
3433 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3434 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3435 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3436 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3437 | return -EPERM; |
3438 | } | |
dc61b1d6 | 3439 | #endif |
332ac17e DF |
3440 | #ifdef CONFIG_SMP |
3441 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3442 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3443 | |
3444 | /* | |
3445 | * Don't allow tasks with an affinity mask smaller than | |
3446 | * the entire root_domain to become SCHED_DEADLINE. We | |
3447 | * will also fail if there's no bandwidth available. | |
3448 | */ | |
e4099a5e PZ |
3449 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3450 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3451 | task_rq_unlock(rq, p, &flags); |
3452 | return -EPERM; | |
3453 | } | |
3454 | } | |
3455 | #endif | |
3456 | } | |
dc61b1d6 | 3457 | |
1da177e4 LT |
3458 | /* recheck policy now with rq lock held */ |
3459 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3460 | policy = oldpolicy = -1; | |
0122ec5b | 3461 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3462 | goto recheck; |
3463 | } | |
332ac17e DF |
3464 | |
3465 | /* | |
3466 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3467 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3468 | * is available. | |
3469 | */ | |
e4099a5e | 3470 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3471 | task_rq_unlock(rq, p, &flags); |
3472 | return -EBUSY; | |
3473 | } | |
3474 | ||
c365c292 TG |
3475 | p->sched_reset_on_fork = reset_on_fork; |
3476 | oldprio = p->prio; | |
3477 | ||
3478 | /* | |
3479 | * Special case for priority boosted tasks. | |
3480 | * | |
3481 | * If the new priority is lower or equal (user space view) | |
3482 | * than the current (boosted) priority, we just store the new | |
3483 | * normal parameters and do not touch the scheduler class and | |
3484 | * the runqueue. This will be done when the task deboost | |
3485 | * itself. | |
3486 | */ | |
3487 | if (rt_mutex_check_prio(p, newprio)) { | |
3488 | __setscheduler_params(p, attr); | |
3489 | task_rq_unlock(rq, p, &flags); | |
3490 | return 0; | |
3491 | } | |
3492 | ||
fd2f4419 | 3493 | on_rq = p->on_rq; |
051a1d1a | 3494 | running = task_current(rq, p); |
0e1f3483 | 3495 | if (on_rq) |
4ca9b72b | 3496 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3497 | if (running) |
3498 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3499 | |
83ab0aa0 | 3500 | prev_class = p->sched_class; |
d50dde5a | 3501 | __setscheduler(rq, p, attr); |
f6b53205 | 3502 | |
0e1f3483 HS |
3503 | if (running) |
3504 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3505 | if (on_rq) { |
3506 | /* | |
3507 | * We enqueue to tail when the priority of a task is | |
3508 | * increased (user space view). | |
3509 | */ | |
3510 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3511 | } | |
cb469845 | 3512 | |
da7a735e | 3513 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3514 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3515 | |
95e02ca9 TG |
3516 | rt_mutex_adjust_pi(p); |
3517 | ||
1da177e4 LT |
3518 | return 0; |
3519 | } | |
961ccddd | 3520 | |
7479f3c9 PZ |
3521 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3522 | const struct sched_param *param, bool check) | |
3523 | { | |
3524 | struct sched_attr attr = { | |
3525 | .sched_policy = policy, | |
3526 | .sched_priority = param->sched_priority, | |
3527 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3528 | }; | |
3529 | ||
3530 | /* | |
3531 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3532 | */ | |
3533 | if (policy & SCHED_RESET_ON_FORK) { | |
3534 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3535 | policy &= ~SCHED_RESET_ON_FORK; | |
3536 | attr.sched_policy = policy; | |
3537 | } | |
3538 | ||
3539 | return __sched_setscheduler(p, &attr, check); | |
3540 | } | |
961ccddd RR |
3541 | /** |
3542 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3543 | * @p: the task in question. | |
3544 | * @policy: new policy. | |
3545 | * @param: structure containing the new RT priority. | |
3546 | * | |
e69f6186 YB |
3547 | * Return: 0 on success. An error code otherwise. |
3548 | * | |
961ccddd RR |
3549 | * NOTE that the task may be already dead. |
3550 | */ | |
3551 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3552 | const struct sched_param *param) |
961ccddd | 3553 | { |
7479f3c9 | 3554 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3555 | } |
1da177e4 LT |
3556 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3557 | ||
d50dde5a DF |
3558 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3559 | { | |
3560 | return __sched_setscheduler(p, attr, true); | |
3561 | } | |
3562 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3563 | ||
961ccddd RR |
3564 | /** |
3565 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3566 | * @p: the task in question. | |
3567 | * @policy: new policy. | |
3568 | * @param: structure containing the new RT priority. | |
3569 | * | |
3570 | * Just like sched_setscheduler, only don't bother checking if the | |
3571 | * current context has permission. For example, this is needed in | |
3572 | * stop_machine(): we create temporary high priority worker threads, | |
3573 | * but our caller might not have that capability. | |
e69f6186 YB |
3574 | * |
3575 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3576 | */ |
3577 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3578 | const struct sched_param *param) |
961ccddd | 3579 | { |
7479f3c9 | 3580 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3581 | } |
3582 | ||
95cdf3b7 IM |
3583 | static int |
3584 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3585 | { |
1da177e4 LT |
3586 | struct sched_param lparam; |
3587 | struct task_struct *p; | |
36c8b586 | 3588 | int retval; |
1da177e4 LT |
3589 | |
3590 | if (!param || pid < 0) | |
3591 | return -EINVAL; | |
3592 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3593 | return -EFAULT; | |
5fe1d75f ON |
3594 | |
3595 | rcu_read_lock(); | |
3596 | retval = -ESRCH; | |
1da177e4 | 3597 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3598 | if (p != NULL) |
3599 | retval = sched_setscheduler(p, policy, &lparam); | |
3600 | rcu_read_unlock(); | |
36c8b586 | 3601 | |
1da177e4 LT |
3602 | return retval; |
3603 | } | |
3604 | ||
d50dde5a DF |
3605 | /* |
3606 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3607 | */ | |
3608 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3609 | struct sched_attr *attr) | |
3610 | { | |
3611 | u32 size; | |
3612 | int ret; | |
3613 | ||
3614 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3615 | return -EFAULT; | |
3616 | ||
3617 | /* | |
3618 | * zero the full structure, so that a short copy will be nice. | |
3619 | */ | |
3620 | memset(attr, 0, sizeof(*attr)); | |
3621 | ||
3622 | ret = get_user(size, &uattr->size); | |
3623 | if (ret) | |
3624 | return ret; | |
3625 | ||
3626 | if (size > PAGE_SIZE) /* silly large */ | |
3627 | goto err_size; | |
3628 | ||
3629 | if (!size) /* abi compat */ | |
3630 | size = SCHED_ATTR_SIZE_VER0; | |
3631 | ||
3632 | if (size < SCHED_ATTR_SIZE_VER0) | |
3633 | goto err_size; | |
3634 | ||
3635 | /* | |
3636 | * If we're handed a bigger struct than we know of, | |
3637 | * ensure all the unknown bits are 0 - i.e. new | |
3638 | * user-space does not rely on any kernel feature | |
3639 | * extensions we dont know about yet. | |
3640 | */ | |
3641 | if (size > sizeof(*attr)) { | |
3642 | unsigned char __user *addr; | |
3643 | unsigned char __user *end; | |
3644 | unsigned char val; | |
3645 | ||
3646 | addr = (void __user *)uattr + sizeof(*attr); | |
3647 | end = (void __user *)uattr + size; | |
3648 | ||
3649 | for (; addr < end; addr++) { | |
3650 | ret = get_user(val, addr); | |
3651 | if (ret) | |
3652 | return ret; | |
3653 | if (val) | |
3654 | goto err_size; | |
3655 | } | |
3656 | size = sizeof(*attr); | |
3657 | } | |
3658 | ||
3659 | ret = copy_from_user(attr, uattr, size); | |
3660 | if (ret) | |
3661 | return -EFAULT; | |
3662 | ||
3663 | /* | |
3664 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3665 | * to be strict and return an error on out-of-bounds values? | |
3666 | */ | |
75e45d51 | 3667 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3668 | |
e78c7bca | 3669 | return 0; |
d50dde5a DF |
3670 | |
3671 | err_size: | |
3672 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3673 | return -E2BIG; |
d50dde5a DF |
3674 | } |
3675 | ||
1da177e4 LT |
3676 | /** |
3677 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3678 | * @pid: the pid in question. | |
3679 | * @policy: new policy. | |
3680 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3681 | * |
3682 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3683 | */ |
5add95d4 HC |
3684 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3685 | struct sched_param __user *, param) | |
1da177e4 | 3686 | { |
c21761f1 JB |
3687 | /* negative values for policy are not valid */ |
3688 | if (policy < 0) | |
3689 | return -EINVAL; | |
3690 | ||
1da177e4 LT |
3691 | return do_sched_setscheduler(pid, policy, param); |
3692 | } | |
3693 | ||
3694 | /** | |
3695 | * sys_sched_setparam - set/change the RT priority of a thread | |
3696 | * @pid: the pid in question. | |
3697 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3698 | * |
3699 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3700 | */ |
5add95d4 | 3701 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3702 | { |
3703 | return do_sched_setscheduler(pid, -1, param); | |
3704 | } | |
3705 | ||
d50dde5a DF |
3706 | /** |
3707 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3708 | * @pid: the pid in question. | |
5778fccf | 3709 | * @uattr: structure containing the extended parameters. |
db66d756 | 3710 | * @flags: for future extension. |
d50dde5a | 3711 | */ |
6d35ab48 PZ |
3712 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3713 | unsigned int, flags) | |
d50dde5a DF |
3714 | { |
3715 | struct sched_attr attr; | |
3716 | struct task_struct *p; | |
3717 | int retval; | |
3718 | ||
6d35ab48 | 3719 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3720 | return -EINVAL; |
3721 | ||
143cf23d MK |
3722 | retval = sched_copy_attr(uattr, &attr); |
3723 | if (retval) | |
3724 | return retval; | |
d50dde5a | 3725 | |
dbdb2275 PZ |
3726 | if (attr.sched_policy < 0) |
3727 | return -EINVAL; | |
d50dde5a DF |
3728 | |
3729 | rcu_read_lock(); | |
3730 | retval = -ESRCH; | |
3731 | p = find_process_by_pid(pid); | |
3732 | if (p != NULL) | |
3733 | retval = sched_setattr(p, &attr); | |
3734 | rcu_read_unlock(); | |
3735 | ||
3736 | return retval; | |
3737 | } | |
3738 | ||
1da177e4 LT |
3739 | /** |
3740 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3741 | * @pid: the pid in question. | |
e69f6186 YB |
3742 | * |
3743 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3744 | * code. | |
1da177e4 | 3745 | */ |
5add95d4 | 3746 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3747 | { |
36c8b586 | 3748 | struct task_struct *p; |
3a5c359a | 3749 | int retval; |
1da177e4 LT |
3750 | |
3751 | if (pid < 0) | |
3a5c359a | 3752 | return -EINVAL; |
1da177e4 LT |
3753 | |
3754 | retval = -ESRCH; | |
5fe85be0 | 3755 | rcu_read_lock(); |
1da177e4 LT |
3756 | p = find_process_by_pid(pid); |
3757 | if (p) { | |
3758 | retval = security_task_getscheduler(p); | |
3759 | if (!retval) | |
ca94c442 LP |
3760 | retval = p->policy |
3761 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3762 | } |
5fe85be0 | 3763 | rcu_read_unlock(); |
1da177e4 LT |
3764 | return retval; |
3765 | } | |
3766 | ||
3767 | /** | |
ca94c442 | 3768 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3769 | * @pid: the pid in question. |
3770 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3771 | * |
3772 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3773 | * code. | |
1da177e4 | 3774 | */ |
5add95d4 | 3775 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3776 | { |
ce5f7f82 | 3777 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3778 | struct task_struct *p; |
3a5c359a | 3779 | int retval; |
1da177e4 LT |
3780 | |
3781 | if (!param || pid < 0) | |
3a5c359a | 3782 | return -EINVAL; |
1da177e4 | 3783 | |
5fe85be0 | 3784 | rcu_read_lock(); |
1da177e4 LT |
3785 | p = find_process_by_pid(pid); |
3786 | retval = -ESRCH; | |
3787 | if (!p) | |
3788 | goto out_unlock; | |
3789 | ||
3790 | retval = security_task_getscheduler(p); | |
3791 | if (retval) | |
3792 | goto out_unlock; | |
3793 | ||
ce5f7f82 PZ |
3794 | if (task_has_rt_policy(p)) |
3795 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3796 | rcu_read_unlock(); |
1da177e4 LT |
3797 | |
3798 | /* | |
3799 | * This one might sleep, we cannot do it with a spinlock held ... | |
3800 | */ | |
3801 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3802 | ||
1da177e4 LT |
3803 | return retval; |
3804 | ||
3805 | out_unlock: | |
5fe85be0 | 3806 | rcu_read_unlock(); |
1da177e4 LT |
3807 | return retval; |
3808 | } | |
3809 | ||
d50dde5a DF |
3810 | static int sched_read_attr(struct sched_attr __user *uattr, |
3811 | struct sched_attr *attr, | |
3812 | unsigned int usize) | |
3813 | { | |
3814 | int ret; | |
3815 | ||
3816 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3817 | return -EFAULT; | |
3818 | ||
3819 | /* | |
3820 | * If we're handed a smaller struct than we know of, | |
3821 | * ensure all the unknown bits are 0 - i.e. old | |
3822 | * user-space does not get uncomplete information. | |
3823 | */ | |
3824 | if (usize < sizeof(*attr)) { | |
3825 | unsigned char *addr; | |
3826 | unsigned char *end; | |
3827 | ||
3828 | addr = (void *)attr + usize; | |
3829 | end = (void *)attr + sizeof(*attr); | |
3830 | ||
3831 | for (; addr < end; addr++) { | |
3832 | if (*addr) | |
22400674 | 3833 | return -EFBIG; |
d50dde5a DF |
3834 | } |
3835 | ||
3836 | attr->size = usize; | |
3837 | } | |
3838 | ||
4efbc454 | 3839 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3840 | if (ret) |
3841 | return -EFAULT; | |
3842 | ||
22400674 | 3843 | return 0; |
d50dde5a DF |
3844 | } |
3845 | ||
3846 | /** | |
aab03e05 | 3847 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3848 | * @pid: the pid in question. |
5778fccf | 3849 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3850 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3851 | * @flags: for future extension. |
d50dde5a | 3852 | */ |
6d35ab48 PZ |
3853 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3854 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3855 | { |
3856 | struct sched_attr attr = { | |
3857 | .size = sizeof(struct sched_attr), | |
3858 | }; | |
3859 | struct task_struct *p; | |
3860 | int retval; | |
3861 | ||
3862 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3863 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3864 | return -EINVAL; |
3865 | ||
3866 | rcu_read_lock(); | |
3867 | p = find_process_by_pid(pid); | |
3868 | retval = -ESRCH; | |
3869 | if (!p) | |
3870 | goto out_unlock; | |
3871 | ||
3872 | retval = security_task_getscheduler(p); | |
3873 | if (retval) | |
3874 | goto out_unlock; | |
3875 | ||
3876 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3877 | if (p->sched_reset_on_fork) |
3878 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3879 | if (task_has_dl_policy(p)) |
3880 | __getparam_dl(p, &attr); | |
3881 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3882 | attr.sched_priority = p->rt_priority; |
3883 | else | |
d0ea0268 | 3884 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3885 | |
3886 | rcu_read_unlock(); | |
3887 | ||
3888 | retval = sched_read_attr(uattr, &attr, size); | |
3889 | return retval; | |
3890 | ||
3891 | out_unlock: | |
3892 | rcu_read_unlock(); | |
3893 | return retval; | |
3894 | } | |
3895 | ||
96f874e2 | 3896 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3897 | { |
5a16f3d3 | 3898 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3899 | struct task_struct *p; |
3900 | int retval; | |
1da177e4 | 3901 | |
23f5d142 | 3902 | rcu_read_lock(); |
1da177e4 LT |
3903 | |
3904 | p = find_process_by_pid(pid); | |
3905 | if (!p) { | |
23f5d142 | 3906 | rcu_read_unlock(); |
1da177e4 LT |
3907 | return -ESRCH; |
3908 | } | |
3909 | ||
23f5d142 | 3910 | /* Prevent p going away */ |
1da177e4 | 3911 | get_task_struct(p); |
23f5d142 | 3912 | rcu_read_unlock(); |
1da177e4 | 3913 | |
14a40ffc TH |
3914 | if (p->flags & PF_NO_SETAFFINITY) { |
3915 | retval = -EINVAL; | |
3916 | goto out_put_task; | |
3917 | } | |
5a16f3d3 RR |
3918 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3919 | retval = -ENOMEM; | |
3920 | goto out_put_task; | |
3921 | } | |
3922 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3923 | retval = -ENOMEM; | |
3924 | goto out_free_cpus_allowed; | |
3925 | } | |
1da177e4 | 3926 | retval = -EPERM; |
4c44aaaf EB |
3927 | if (!check_same_owner(p)) { |
3928 | rcu_read_lock(); | |
3929 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3930 | rcu_read_unlock(); | |
3931 | goto out_unlock; | |
3932 | } | |
3933 | rcu_read_unlock(); | |
3934 | } | |
1da177e4 | 3935 | |
b0ae1981 | 3936 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3937 | if (retval) |
3938 | goto out_unlock; | |
3939 | ||
e4099a5e PZ |
3940 | |
3941 | cpuset_cpus_allowed(p, cpus_allowed); | |
3942 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3943 | ||
332ac17e DF |
3944 | /* |
3945 | * Since bandwidth control happens on root_domain basis, | |
3946 | * if admission test is enabled, we only admit -deadline | |
3947 | * tasks allowed to run on all the CPUs in the task's | |
3948 | * root_domain. | |
3949 | */ | |
3950 | #ifdef CONFIG_SMP | |
3951 | if (task_has_dl_policy(p)) { | |
3952 | const struct cpumask *span = task_rq(p)->rd->span; | |
3953 | ||
e4099a5e | 3954 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
3955 | retval = -EBUSY; |
3956 | goto out_unlock; | |
3957 | } | |
3958 | } | |
3959 | #endif | |
49246274 | 3960 | again: |
5a16f3d3 | 3961 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3962 | |
8707d8b8 | 3963 | if (!retval) { |
5a16f3d3 RR |
3964 | cpuset_cpus_allowed(p, cpus_allowed); |
3965 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3966 | /* |
3967 | * We must have raced with a concurrent cpuset | |
3968 | * update. Just reset the cpus_allowed to the | |
3969 | * cpuset's cpus_allowed | |
3970 | */ | |
5a16f3d3 | 3971 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3972 | goto again; |
3973 | } | |
3974 | } | |
1da177e4 | 3975 | out_unlock: |
5a16f3d3 RR |
3976 | free_cpumask_var(new_mask); |
3977 | out_free_cpus_allowed: | |
3978 | free_cpumask_var(cpus_allowed); | |
3979 | out_put_task: | |
1da177e4 | 3980 | put_task_struct(p); |
1da177e4 LT |
3981 | return retval; |
3982 | } | |
3983 | ||
3984 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3985 | struct cpumask *new_mask) |
1da177e4 | 3986 | { |
96f874e2 RR |
3987 | if (len < cpumask_size()) |
3988 | cpumask_clear(new_mask); | |
3989 | else if (len > cpumask_size()) | |
3990 | len = cpumask_size(); | |
3991 | ||
1da177e4 LT |
3992 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3993 | } | |
3994 | ||
3995 | /** | |
3996 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3997 | * @pid: pid of the process | |
3998 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3999 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4000 | * |
4001 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4002 | */ |
5add95d4 HC |
4003 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4004 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4005 | { |
5a16f3d3 | 4006 | cpumask_var_t new_mask; |
1da177e4 LT |
4007 | int retval; |
4008 | ||
5a16f3d3 RR |
4009 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4010 | return -ENOMEM; | |
1da177e4 | 4011 | |
5a16f3d3 RR |
4012 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4013 | if (retval == 0) | |
4014 | retval = sched_setaffinity(pid, new_mask); | |
4015 | free_cpumask_var(new_mask); | |
4016 | return retval; | |
1da177e4 LT |
4017 | } |
4018 | ||
96f874e2 | 4019 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4020 | { |
36c8b586 | 4021 | struct task_struct *p; |
31605683 | 4022 | unsigned long flags; |
1da177e4 | 4023 | int retval; |
1da177e4 | 4024 | |
23f5d142 | 4025 | rcu_read_lock(); |
1da177e4 LT |
4026 | |
4027 | retval = -ESRCH; | |
4028 | p = find_process_by_pid(pid); | |
4029 | if (!p) | |
4030 | goto out_unlock; | |
4031 | ||
e7834f8f DQ |
4032 | retval = security_task_getscheduler(p); |
4033 | if (retval) | |
4034 | goto out_unlock; | |
4035 | ||
013fdb80 | 4036 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4037 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4038 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4039 | |
4040 | out_unlock: | |
23f5d142 | 4041 | rcu_read_unlock(); |
1da177e4 | 4042 | |
9531b62f | 4043 | return retval; |
1da177e4 LT |
4044 | } |
4045 | ||
4046 | /** | |
4047 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4048 | * @pid: pid of the process | |
4049 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4050 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4051 | * |
4052 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4053 | */ |
5add95d4 HC |
4054 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4055 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4056 | { |
4057 | int ret; | |
f17c8607 | 4058 | cpumask_var_t mask; |
1da177e4 | 4059 | |
84fba5ec | 4060 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4061 | return -EINVAL; |
4062 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4063 | return -EINVAL; |
4064 | ||
f17c8607 RR |
4065 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4066 | return -ENOMEM; | |
1da177e4 | 4067 | |
f17c8607 RR |
4068 | ret = sched_getaffinity(pid, mask); |
4069 | if (ret == 0) { | |
8bc037fb | 4070 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4071 | |
4072 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4073 | ret = -EFAULT; |
4074 | else | |
cd3d8031 | 4075 | ret = retlen; |
f17c8607 RR |
4076 | } |
4077 | free_cpumask_var(mask); | |
1da177e4 | 4078 | |
f17c8607 | 4079 | return ret; |
1da177e4 LT |
4080 | } |
4081 | ||
4082 | /** | |
4083 | * sys_sched_yield - yield the current processor to other threads. | |
4084 | * | |
dd41f596 IM |
4085 | * This function yields the current CPU to other tasks. If there are no |
4086 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4087 | * |
4088 | * Return: 0. | |
1da177e4 | 4089 | */ |
5add95d4 | 4090 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4091 | { |
70b97a7f | 4092 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4093 | |
2d72376b | 4094 | schedstat_inc(rq, yld_count); |
4530d7ab | 4095 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4096 | |
4097 | /* | |
4098 | * Since we are going to call schedule() anyway, there's | |
4099 | * no need to preempt or enable interrupts: | |
4100 | */ | |
4101 | __release(rq->lock); | |
8a25d5de | 4102 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4103 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4104 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4105 | |
4106 | schedule(); | |
4107 | ||
4108 | return 0; | |
4109 | } | |
4110 | ||
e7b38404 | 4111 | static void __cond_resched(void) |
1da177e4 | 4112 | { |
bdb43806 | 4113 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4114 | __schedule(); |
bdb43806 | 4115 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4116 | } |
4117 | ||
02b67cc3 | 4118 | int __sched _cond_resched(void) |
1da177e4 | 4119 | { |
ac1bea85 | 4120 | rcu_cond_resched(); |
d86ee480 | 4121 | if (should_resched()) { |
1da177e4 LT |
4122 | __cond_resched(); |
4123 | return 1; | |
4124 | } | |
4125 | return 0; | |
4126 | } | |
02b67cc3 | 4127 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4128 | |
4129 | /* | |
613afbf8 | 4130 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4131 | * call schedule, and on return reacquire the lock. |
4132 | * | |
41a2d6cf | 4133 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4134 | * operations here to prevent schedule() from being called twice (once via |
4135 | * spin_unlock(), once by hand). | |
4136 | */ | |
613afbf8 | 4137 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4138 | { |
ac1bea85 | 4139 | bool need_rcu_resched = rcu_should_resched(); |
d86ee480 | 4140 | int resched = should_resched(); |
6df3cecb JK |
4141 | int ret = 0; |
4142 | ||
f607c668 PZ |
4143 | lockdep_assert_held(lock); |
4144 | ||
ac1bea85 | 4145 | if (spin_needbreak(lock) || resched || need_rcu_resched) { |
1da177e4 | 4146 | spin_unlock(lock); |
d86ee480 | 4147 | if (resched) |
95c354fe | 4148 | __cond_resched(); |
ac1bea85 PM |
4149 | else if (unlikely(need_rcu_resched)) |
4150 | rcu_resched(); | |
95c354fe NP |
4151 | else |
4152 | cpu_relax(); | |
6df3cecb | 4153 | ret = 1; |
1da177e4 | 4154 | spin_lock(lock); |
1da177e4 | 4155 | } |
6df3cecb | 4156 | return ret; |
1da177e4 | 4157 | } |
613afbf8 | 4158 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4159 | |
613afbf8 | 4160 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4161 | { |
4162 | BUG_ON(!in_softirq()); | |
4163 | ||
ac1bea85 | 4164 | rcu_cond_resched(); /* BH disabled OK, just recording QSes. */ |
d86ee480 | 4165 | if (should_resched()) { |
98d82567 | 4166 | local_bh_enable(); |
1da177e4 LT |
4167 | __cond_resched(); |
4168 | local_bh_disable(); | |
4169 | return 1; | |
4170 | } | |
4171 | return 0; | |
4172 | } | |
613afbf8 | 4173 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4174 | |
1da177e4 LT |
4175 | /** |
4176 | * yield - yield the current processor to other threads. | |
4177 | * | |
8e3fabfd PZ |
4178 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4179 | * | |
4180 | * The scheduler is at all times free to pick the calling task as the most | |
4181 | * eligible task to run, if removing the yield() call from your code breaks | |
4182 | * it, its already broken. | |
4183 | * | |
4184 | * Typical broken usage is: | |
4185 | * | |
4186 | * while (!event) | |
4187 | * yield(); | |
4188 | * | |
4189 | * where one assumes that yield() will let 'the other' process run that will | |
4190 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4191 | * happen. Never use yield() as a progress guarantee!! | |
4192 | * | |
4193 | * If you want to use yield() to wait for something, use wait_event(). | |
4194 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4195 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4196 | */ |
4197 | void __sched yield(void) | |
4198 | { | |
4199 | set_current_state(TASK_RUNNING); | |
4200 | sys_sched_yield(); | |
4201 | } | |
1da177e4 LT |
4202 | EXPORT_SYMBOL(yield); |
4203 | ||
d95f4122 MG |
4204 | /** |
4205 | * yield_to - yield the current processor to another thread in | |
4206 | * your thread group, or accelerate that thread toward the | |
4207 | * processor it's on. | |
16addf95 RD |
4208 | * @p: target task |
4209 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4210 | * |
4211 | * It's the caller's job to ensure that the target task struct | |
4212 | * can't go away on us before we can do any checks. | |
4213 | * | |
e69f6186 | 4214 | * Return: |
7b270f60 PZ |
4215 | * true (>0) if we indeed boosted the target task. |
4216 | * false (0) if we failed to boost the target. | |
4217 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4218 | */ |
4219 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4220 | { | |
4221 | struct task_struct *curr = current; | |
4222 | struct rq *rq, *p_rq; | |
4223 | unsigned long flags; | |
c3c18640 | 4224 | int yielded = 0; |
d95f4122 MG |
4225 | |
4226 | local_irq_save(flags); | |
4227 | rq = this_rq(); | |
4228 | ||
4229 | again: | |
4230 | p_rq = task_rq(p); | |
7b270f60 PZ |
4231 | /* |
4232 | * If we're the only runnable task on the rq and target rq also | |
4233 | * has only one task, there's absolutely no point in yielding. | |
4234 | */ | |
4235 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4236 | yielded = -ESRCH; | |
4237 | goto out_irq; | |
4238 | } | |
4239 | ||
d95f4122 | 4240 | double_rq_lock(rq, p_rq); |
39e24d8f | 4241 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4242 | double_rq_unlock(rq, p_rq); |
4243 | goto again; | |
4244 | } | |
4245 | ||
4246 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4247 | goto out_unlock; |
d95f4122 MG |
4248 | |
4249 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4250 | goto out_unlock; |
d95f4122 MG |
4251 | |
4252 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4253 | goto out_unlock; |
d95f4122 MG |
4254 | |
4255 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4256 | if (yielded) { |
d95f4122 | 4257 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4258 | /* |
4259 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4260 | * fairness. | |
4261 | */ | |
4262 | if (preempt && rq != p_rq) | |
4263 | resched_task(p_rq->curr); | |
4264 | } | |
d95f4122 | 4265 | |
7b270f60 | 4266 | out_unlock: |
d95f4122 | 4267 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4268 | out_irq: |
d95f4122 MG |
4269 | local_irq_restore(flags); |
4270 | ||
7b270f60 | 4271 | if (yielded > 0) |
d95f4122 MG |
4272 | schedule(); |
4273 | ||
4274 | return yielded; | |
4275 | } | |
4276 | EXPORT_SYMBOL_GPL(yield_to); | |
4277 | ||
1da177e4 | 4278 | /* |
41a2d6cf | 4279 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4280 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4281 | */ |
4282 | void __sched io_schedule(void) | |
4283 | { | |
54d35f29 | 4284 | struct rq *rq = raw_rq(); |
1da177e4 | 4285 | |
0ff92245 | 4286 | delayacct_blkio_start(); |
1da177e4 | 4287 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4288 | blk_flush_plug(current); |
8f0dfc34 | 4289 | current->in_iowait = 1; |
1da177e4 | 4290 | schedule(); |
8f0dfc34 | 4291 | current->in_iowait = 0; |
1da177e4 | 4292 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4293 | delayacct_blkio_end(); |
1da177e4 | 4294 | } |
1da177e4 LT |
4295 | EXPORT_SYMBOL(io_schedule); |
4296 | ||
4297 | long __sched io_schedule_timeout(long timeout) | |
4298 | { | |
54d35f29 | 4299 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4300 | long ret; |
4301 | ||
0ff92245 | 4302 | delayacct_blkio_start(); |
1da177e4 | 4303 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4304 | blk_flush_plug(current); |
8f0dfc34 | 4305 | current->in_iowait = 1; |
1da177e4 | 4306 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4307 | current->in_iowait = 0; |
1da177e4 | 4308 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4309 | delayacct_blkio_end(); |
1da177e4 LT |
4310 | return ret; |
4311 | } | |
4312 | ||
4313 | /** | |
4314 | * sys_sched_get_priority_max - return maximum RT priority. | |
4315 | * @policy: scheduling class. | |
4316 | * | |
e69f6186 YB |
4317 | * Return: On success, this syscall returns the maximum |
4318 | * rt_priority that can be used by a given scheduling class. | |
4319 | * On failure, a negative error code is returned. | |
1da177e4 | 4320 | */ |
5add95d4 | 4321 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4322 | { |
4323 | int ret = -EINVAL; | |
4324 | ||
4325 | switch (policy) { | |
4326 | case SCHED_FIFO: | |
4327 | case SCHED_RR: | |
4328 | ret = MAX_USER_RT_PRIO-1; | |
4329 | break; | |
aab03e05 | 4330 | case SCHED_DEADLINE: |
1da177e4 | 4331 | case SCHED_NORMAL: |
b0a9499c | 4332 | case SCHED_BATCH: |
dd41f596 | 4333 | case SCHED_IDLE: |
1da177e4 LT |
4334 | ret = 0; |
4335 | break; | |
4336 | } | |
4337 | return ret; | |
4338 | } | |
4339 | ||
4340 | /** | |
4341 | * sys_sched_get_priority_min - return minimum RT priority. | |
4342 | * @policy: scheduling class. | |
4343 | * | |
e69f6186 YB |
4344 | * Return: On success, this syscall returns the minimum |
4345 | * rt_priority that can be used by a given scheduling class. | |
4346 | * On failure, a negative error code is returned. | |
1da177e4 | 4347 | */ |
5add95d4 | 4348 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4349 | { |
4350 | int ret = -EINVAL; | |
4351 | ||
4352 | switch (policy) { | |
4353 | case SCHED_FIFO: | |
4354 | case SCHED_RR: | |
4355 | ret = 1; | |
4356 | break; | |
aab03e05 | 4357 | case SCHED_DEADLINE: |
1da177e4 | 4358 | case SCHED_NORMAL: |
b0a9499c | 4359 | case SCHED_BATCH: |
dd41f596 | 4360 | case SCHED_IDLE: |
1da177e4 LT |
4361 | ret = 0; |
4362 | } | |
4363 | return ret; | |
4364 | } | |
4365 | ||
4366 | /** | |
4367 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4368 | * @pid: pid of the process. | |
4369 | * @interval: userspace pointer to the timeslice value. | |
4370 | * | |
4371 | * this syscall writes the default timeslice value of a given process | |
4372 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4373 | * |
4374 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4375 | * an error code. | |
1da177e4 | 4376 | */ |
17da2bd9 | 4377 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4378 | struct timespec __user *, interval) |
1da177e4 | 4379 | { |
36c8b586 | 4380 | struct task_struct *p; |
a4ec24b4 | 4381 | unsigned int time_slice; |
dba091b9 TG |
4382 | unsigned long flags; |
4383 | struct rq *rq; | |
3a5c359a | 4384 | int retval; |
1da177e4 | 4385 | struct timespec t; |
1da177e4 LT |
4386 | |
4387 | if (pid < 0) | |
3a5c359a | 4388 | return -EINVAL; |
1da177e4 LT |
4389 | |
4390 | retval = -ESRCH; | |
1a551ae7 | 4391 | rcu_read_lock(); |
1da177e4 LT |
4392 | p = find_process_by_pid(pid); |
4393 | if (!p) | |
4394 | goto out_unlock; | |
4395 | ||
4396 | retval = security_task_getscheduler(p); | |
4397 | if (retval) | |
4398 | goto out_unlock; | |
4399 | ||
dba091b9 | 4400 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4401 | time_slice = 0; |
4402 | if (p->sched_class->get_rr_interval) | |
4403 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4404 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4405 | |
1a551ae7 | 4406 | rcu_read_unlock(); |
a4ec24b4 | 4407 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4408 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4409 | return retval; |
3a5c359a | 4410 | |
1da177e4 | 4411 | out_unlock: |
1a551ae7 | 4412 | rcu_read_unlock(); |
1da177e4 LT |
4413 | return retval; |
4414 | } | |
4415 | ||
7c731e0a | 4416 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4417 | |
82a1fcb9 | 4418 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4419 | { |
1da177e4 | 4420 | unsigned long free = 0; |
4e79752c | 4421 | int ppid; |
36c8b586 | 4422 | unsigned state; |
1da177e4 | 4423 | |
1da177e4 | 4424 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4425 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4426 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4427 | #if BITS_PER_LONG == 32 |
1da177e4 | 4428 | if (state == TASK_RUNNING) |
3df0fc5b | 4429 | printk(KERN_CONT " running "); |
1da177e4 | 4430 | else |
3df0fc5b | 4431 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4432 | #else |
4433 | if (state == TASK_RUNNING) | |
3df0fc5b | 4434 | printk(KERN_CONT " running task "); |
1da177e4 | 4435 | else |
3df0fc5b | 4436 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4437 | #endif |
4438 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4439 | free = stack_not_used(p); |
1da177e4 | 4440 | #endif |
4e79752c PM |
4441 | rcu_read_lock(); |
4442 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4443 | rcu_read_unlock(); | |
3df0fc5b | 4444 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4445 | task_pid_nr(p), ppid, |
aa47b7e0 | 4446 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4447 | |
3d1cb205 | 4448 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4449 | show_stack(p, NULL); |
1da177e4 LT |
4450 | } |
4451 | ||
e59e2ae2 | 4452 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4453 | { |
36c8b586 | 4454 | struct task_struct *g, *p; |
1da177e4 | 4455 | |
4bd77321 | 4456 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4457 | printk(KERN_INFO |
4458 | " task PC stack pid father\n"); | |
1da177e4 | 4459 | #else |
3df0fc5b PZ |
4460 | printk(KERN_INFO |
4461 | " task PC stack pid father\n"); | |
1da177e4 | 4462 | #endif |
510f5acc | 4463 | rcu_read_lock(); |
1da177e4 LT |
4464 | do_each_thread(g, p) { |
4465 | /* | |
4466 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4467 | * console might take a lot of time: |
1da177e4 LT |
4468 | */ |
4469 | touch_nmi_watchdog(); | |
39bc89fd | 4470 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4471 | sched_show_task(p); |
1da177e4 LT |
4472 | } while_each_thread(g, p); |
4473 | ||
04c9167f JF |
4474 | touch_all_softlockup_watchdogs(); |
4475 | ||
dd41f596 IM |
4476 | #ifdef CONFIG_SCHED_DEBUG |
4477 | sysrq_sched_debug_show(); | |
4478 | #endif | |
510f5acc | 4479 | rcu_read_unlock(); |
e59e2ae2 IM |
4480 | /* |
4481 | * Only show locks if all tasks are dumped: | |
4482 | */ | |
93335a21 | 4483 | if (!state_filter) |
e59e2ae2 | 4484 | debug_show_all_locks(); |
1da177e4 LT |
4485 | } |
4486 | ||
0db0628d | 4487 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4488 | { |
dd41f596 | 4489 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4490 | } |
4491 | ||
f340c0d1 IM |
4492 | /** |
4493 | * init_idle - set up an idle thread for a given CPU | |
4494 | * @idle: task in question | |
4495 | * @cpu: cpu the idle task belongs to | |
4496 | * | |
4497 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4498 | * flag, to make booting more robust. | |
4499 | */ | |
0db0628d | 4500 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4501 | { |
70b97a7f | 4502 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4503 | unsigned long flags; |
4504 | ||
05fa785c | 4505 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4506 | |
5e1576ed | 4507 | __sched_fork(0, idle); |
06b83b5f | 4508 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4509 | idle->se.exec_start = sched_clock(); |
4510 | ||
1e1b6c51 | 4511 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4512 | /* |
4513 | * We're having a chicken and egg problem, even though we are | |
4514 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4515 | * lockdep check in task_group() will fail. | |
4516 | * | |
4517 | * Similar case to sched_fork(). / Alternatively we could | |
4518 | * use task_rq_lock() here and obtain the other rq->lock. | |
4519 | * | |
4520 | * Silence PROVE_RCU | |
4521 | */ | |
4522 | rcu_read_lock(); | |
dd41f596 | 4523 | __set_task_cpu(idle, cpu); |
6506cf6c | 4524 | rcu_read_unlock(); |
1da177e4 | 4525 | |
1da177e4 | 4526 | rq->curr = rq->idle = idle; |
77177856 | 4527 | idle->on_rq = 1; |
3ca7a440 PZ |
4528 | #if defined(CONFIG_SMP) |
4529 | idle->on_cpu = 1; | |
4866cde0 | 4530 | #endif |
05fa785c | 4531 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4532 | |
4533 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4534 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4535 | |
dd41f596 IM |
4536 | /* |
4537 | * The idle tasks have their own, simple scheduling class: | |
4538 | */ | |
4539 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4540 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4541 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4542 | #if defined(CONFIG_SMP) |
4543 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4544 | #endif | |
19978ca6 IM |
4545 | } |
4546 | ||
1da177e4 | 4547 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4548 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4549 | { | |
4550 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4551 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4552 | |
4553 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4554 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4555 | } |
4556 | ||
1da177e4 LT |
4557 | /* |
4558 | * This is how migration works: | |
4559 | * | |
969c7921 TH |
4560 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4561 | * stop_one_cpu(). | |
4562 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4563 | * off the CPU) | |
4564 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4565 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4566 | * it and puts it into the right queue. |
969c7921 TH |
4567 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4568 | * is done. | |
1da177e4 LT |
4569 | */ |
4570 | ||
4571 | /* | |
4572 | * Change a given task's CPU affinity. Migrate the thread to a | |
4573 | * proper CPU and schedule it away if the CPU it's executing on | |
4574 | * is removed from the allowed bitmask. | |
4575 | * | |
4576 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4577 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4578 | * call is not atomic; no spinlocks may be held. |
4579 | */ | |
96f874e2 | 4580 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4581 | { |
4582 | unsigned long flags; | |
70b97a7f | 4583 | struct rq *rq; |
969c7921 | 4584 | unsigned int dest_cpu; |
48f24c4d | 4585 | int ret = 0; |
1da177e4 LT |
4586 | |
4587 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4588 | |
db44fc01 YZ |
4589 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4590 | goto out; | |
4591 | ||
6ad4c188 | 4592 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4593 | ret = -EINVAL; |
4594 | goto out; | |
4595 | } | |
4596 | ||
1e1b6c51 | 4597 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4598 | |
1da177e4 | 4599 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4600 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4601 | goto out; |
4602 | ||
969c7921 | 4603 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4604 | if (p->on_rq) { |
969c7921 | 4605 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4606 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4607 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4608 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4609 | tlb_migrate_finish(p->mm); |
4610 | return 0; | |
4611 | } | |
4612 | out: | |
0122ec5b | 4613 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4614 | |
1da177e4 LT |
4615 | return ret; |
4616 | } | |
cd8ba7cd | 4617 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4618 | |
4619 | /* | |
41a2d6cf | 4620 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4621 | * this because either it can't run here any more (set_cpus_allowed() |
4622 | * away from this CPU, or CPU going down), or because we're | |
4623 | * attempting to rebalance this task on exec (sched_exec). | |
4624 | * | |
4625 | * So we race with normal scheduler movements, but that's OK, as long | |
4626 | * as the task is no longer on this CPU. | |
efc30814 KK |
4627 | * |
4628 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4629 | */ |
efc30814 | 4630 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4631 | { |
70b97a7f | 4632 | struct rq *rq_dest, *rq_src; |
e2912009 | 4633 | int ret = 0; |
1da177e4 | 4634 | |
e761b772 | 4635 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4636 | return ret; |
1da177e4 LT |
4637 | |
4638 | rq_src = cpu_rq(src_cpu); | |
4639 | rq_dest = cpu_rq(dest_cpu); | |
4640 | ||
0122ec5b | 4641 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4642 | double_rq_lock(rq_src, rq_dest); |
4643 | /* Already moved. */ | |
4644 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4645 | goto done; |
1da177e4 | 4646 | /* Affinity changed (again). */ |
fa17b507 | 4647 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4648 | goto fail; |
1da177e4 | 4649 | |
e2912009 PZ |
4650 | /* |
4651 | * If we're not on a rq, the next wake-up will ensure we're | |
4652 | * placed properly. | |
4653 | */ | |
fd2f4419 | 4654 | if (p->on_rq) { |
4ca9b72b | 4655 | dequeue_task(rq_src, p, 0); |
e2912009 | 4656 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4657 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4658 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4659 | } |
b1e38734 | 4660 | done: |
efc30814 | 4661 | ret = 1; |
b1e38734 | 4662 | fail: |
1da177e4 | 4663 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4664 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4665 | return ret; |
1da177e4 LT |
4666 | } |
4667 | ||
e6628d5b MG |
4668 | #ifdef CONFIG_NUMA_BALANCING |
4669 | /* Migrate current task p to target_cpu */ | |
4670 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4671 | { | |
4672 | struct migration_arg arg = { p, target_cpu }; | |
4673 | int curr_cpu = task_cpu(p); | |
4674 | ||
4675 | if (curr_cpu == target_cpu) | |
4676 | return 0; | |
4677 | ||
4678 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4679 | return -EINVAL; | |
4680 | ||
4681 | /* TODO: This is not properly updating schedstats */ | |
4682 | ||
286549dc | 4683 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4684 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4685 | } | |
0ec8aa00 PZ |
4686 | |
4687 | /* | |
4688 | * Requeue a task on a given node and accurately track the number of NUMA | |
4689 | * tasks on the runqueues | |
4690 | */ | |
4691 | void sched_setnuma(struct task_struct *p, int nid) | |
4692 | { | |
4693 | struct rq *rq; | |
4694 | unsigned long flags; | |
4695 | bool on_rq, running; | |
4696 | ||
4697 | rq = task_rq_lock(p, &flags); | |
4698 | on_rq = p->on_rq; | |
4699 | running = task_current(rq, p); | |
4700 | ||
4701 | if (on_rq) | |
4702 | dequeue_task(rq, p, 0); | |
4703 | if (running) | |
4704 | p->sched_class->put_prev_task(rq, p); | |
4705 | ||
4706 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4707 | |
4708 | if (running) | |
4709 | p->sched_class->set_curr_task(rq); | |
4710 | if (on_rq) | |
4711 | enqueue_task(rq, p, 0); | |
4712 | task_rq_unlock(rq, p, &flags); | |
4713 | } | |
e6628d5b MG |
4714 | #endif |
4715 | ||
1da177e4 | 4716 | /* |
969c7921 TH |
4717 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4718 | * and performs thread migration by bumping thread off CPU then | |
4719 | * 'pushing' onto another runqueue. | |
1da177e4 | 4720 | */ |
969c7921 | 4721 | static int migration_cpu_stop(void *data) |
1da177e4 | 4722 | { |
969c7921 | 4723 | struct migration_arg *arg = data; |
f7b4cddc | 4724 | |
969c7921 TH |
4725 | /* |
4726 | * The original target cpu might have gone down and we might | |
4727 | * be on another cpu but it doesn't matter. | |
4728 | */ | |
f7b4cddc | 4729 | local_irq_disable(); |
969c7921 | 4730 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4731 | local_irq_enable(); |
1da177e4 | 4732 | return 0; |
f7b4cddc ON |
4733 | } |
4734 | ||
1da177e4 | 4735 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4736 | |
054b9108 | 4737 | /* |
48c5ccae PZ |
4738 | * Ensures that the idle task is using init_mm right before its cpu goes |
4739 | * offline. | |
054b9108 | 4740 | */ |
48c5ccae | 4741 | void idle_task_exit(void) |
1da177e4 | 4742 | { |
48c5ccae | 4743 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4744 | |
48c5ccae | 4745 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4746 | |
a53efe5f | 4747 | if (mm != &init_mm) { |
48c5ccae | 4748 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4749 | finish_arch_post_lock_switch(); |
4750 | } | |
48c5ccae | 4751 | mmdrop(mm); |
1da177e4 LT |
4752 | } |
4753 | ||
4754 | /* | |
5d180232 PZ |
4755 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4756 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4757 | * nr_active count is stable. | |
4758 | * | |
4759 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4760 | */ |
5d180232 | 4761 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4762 | { |
5d180232 PZ |
4763 | long delta = calc_load_fold_active(rq); |
4764 | if (delta) | |
4765 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4766 | } |
4767 | ||
3f1d2a31 PZ |
4768 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4769 | { | |
4770 | } | |
4771 | ||
4772 | static const struct sched_class fake_sched_class = { | |
4773 | .put_prev_task = put_prev_task_fake, | |
4774 | }; | |
4775 | ||
4776 | static struct task_struct fake_task = { | |
4777 | /* | |
4778 | * Avoid pull_{rt,dl}_task() | |
4779 | */ | |
4780 | .prio = MAX_PRIO + 1, | |
4781 | .sched_class = &fake_sched_class, | |
4782 | }; | |
4783 | ||
48f24c4d | 4784 | /* |
48c5ccae PZ |
4785 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4786 | * try_to_wake_up()->select_task_rq(). | |
4787 | * | |
4788 | * Called with rq->lock held even though we'er in stop_machine() and | |
4789 | * there's no concurrency possible, we hold the required locks anyway | |
4790 | * because of lock validation efforts. | |
1da177e4 | 4791 | */ |
48c5ccae | 4792 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4793 | { |
70b97a7f | 4794 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4795 | struct task_struct *next, *stop = rq->stop; |
4796 | int dest_cpu; | |
1da177e4 LT |
4797 | |
4798 | /* | |
48c5ccae PZ |
4799 | * Fudge the rq selection such that the below task selection loop |
4800 | * doesn't get stuck on the currently eligible stop task. | |
4801 | * | |
4802 | * We're currently inside stop_machine() and the rq is either stuck | |
4803 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4804 | * either way we should never end up calling schedule() until we're | |
4805 | * done here. | |
1da177e4 | 4806 | */ |
48c5ccae | 4807 | rq->stop = NULL; |
48f24c4d | 4808 | |
77bd3970 FW |
4809 | /* |
4810 | * put_prev_task() and pick_next_task() sched | |
4811 | * class method both need to have an up-to-date | |
4812 | * value of rq->clock[_task] | |
4813 | */ | |
4814 | update_rq_clock(rq); | |
4815 | ||
dd41f596 | 4816 | for ( ; ; ) { |
48c5ccae PZ |
4817 | /* |
4818 | * There's this thread running, bail when that's the only | |
4819 | * remaining thread. | |
4820 | */ | |
4821 | if (rq->nr_running == 1) | |
dd41f596 | 4822 | break; |
48c5ccae | 4823 | |
3f1d2a31 | 4824 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4825 | BUG_ON(!next); |
79c53799 | 4826 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4827 | |
48c5ccae PZ |
4828 | /* Find suitable destination for @next, with force if needed. */ |
4829 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4830 | raw_spin_unlock(&rq->lock); | |
4831 | ||
4832 | __migrate_task(next, dead_cpu, dest_cpu); | |
4833 | ||
4834 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4835 | } |
dce48a84 | 4836 | |
48c5ccae | 4837 | rq->stop = stop; |
dce48a84 | 4838 | } |
48c5ccae | 4839 | |
1da177e4 LT |
4840 | #endif /* CONFIG_HOTPLUG_CPU */ |
4841 | ||
e692ab53 NP |
4842 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4843 | ||
4844 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4845 | { |
4846 | .procname = "sched_domain", | |
c57baf1e | 4847 | .mode = 0555, |
e0361851 | 4848 | }, |
56992309 | 4849 | {} |
e692ab53 NP |
4850 | }; |
4851 | ||
4852 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4853 | { |
4854 | .procname = "kernel", | |
c57baf1e | 4855 | .mode = 0555, |
e0361851 AD |
4856 | .child = sd_ctl_dir, |
4857 | }, | |
56992309 | 4858 | {} |
e692ab53 NP |
4859 | }; |
4860 | ||
4861 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4862 | { | |
4863 | struct ctl_table *entry = | |
5cf9f062 | 4864 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4865 | |
e692ab53 NP |
4866 | return entry; |
4867 | } | |
4868 | ||
6382bc90 MM |
4869 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4870 | { | |
cd790076 | 4871 | struct ctl_table *entry; |
6382bc90 | 4872 | |
cd790076 MM |
4873 | /* |
4874 | * In the intermediate directories, both the child directory and | |
4875 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4876 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4877 | * static strings and all have proc handlers. |
4878 | */ | |
4879 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4880 | if (entry->child) |
4881 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4882 | if (entry->proc_handler == NULL) |
4883 | kfree(entry->procname); | |
4884 | } | |
6382bc90 MM |
4885 | |
4886 | kfree(*tablep); | |
4887 | *tablep = NULL; | |
4888 | } | |
4889 | ||
201c373e | 4890 | static int min_load_idx = 0; |
fd9b86d3 | 4891 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4892 | |
e692ab53 | 4893 | static void |
e0361851 | 4894 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4895 | const char *procname, void *data, int maxlen, |
201c373e NK |
4896 | umode_t mode, proc_handler *proc_handler, |
4897 | bool load_idx) | |
e692ab53 | 4898 | { |
e692ab53 NP |
4899 | entry->procname = procname; |
4900 | entry->data = data; | |
4901 | entry->maxlen = maxlen; | |
4902 | entry->mode = mode; | |
4903 | entry->proc_handler = proc_handler; | |
201c373e NK |
4904 | |
4905 | if (load_idx) { | |
4906 | entry->extra1 = &min_load_idx; | |
4907 | entry->extra2 = &max_load_idx; | |
4908 | } | |
e692ab53 NP |
4909 | } |
4910 | ||
4911 | static struct ctl_table * | |
4912 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4913 | { | |
37e6bae8 | 4914 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4915 | |
ad1cdc1d MM |
4916 | if (table == NULL) |
4917 | return NULL; | |
4918 | ||
e0361851 | 4919 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4920 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4921 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4922 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4923 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4924 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4925 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4926 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4927 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4928 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4929 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4930 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4931 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4932 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4933 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4934 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4935 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4936 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4937 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4938 | &sd->cache_nice_tries, |
201c373e | 4939 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4940 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4941 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4942 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4943 | &sd->max_newidle_lb_cost, | |
4944 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4945 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4946 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4947 | /* &table[13] is terminator */ |
e692ab53 NP |
4948 | |
4949 | return table; | |
4950 | } | |
4951 | ||
be7002e6 | 4952 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4953 | { |
4954 | struct ctl_table *entry, *table; | |
4955 | struct sched_domain *sd; | |
4956 | int domain_num = 0, i; | |
4957 | char buf[32]; | |
4958 | ||
4959 | for_each_domain(cpu, sd) | |
4960 | domain_num++; | |
4961 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4962 | if (table == NULL) |
4963 | return NULL; | |
e692ab53 NP |
4964 | |
4965 | i = 0; | |
4966 | for_each_domain(cpu, sd) { | |
4967 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4968 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4969 | entry->mode = 0555; |
e692ab53 NP |
4970 | entry->child = sd_alloc_ctl_domain_table(sd); |
4971 | entry++; | |
4972 | i++; | |
4973 | } | |
4974 | return table; | |
4975 | } | |
4976 | ||
4977 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4978 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4979 | { |
6ad4c188 | 4980 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4981 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4982 | char buf[32]; | |
4983 | ||
7378547f MM |
4984 | WARN_ON(sd_ctl_dir[0].child); |
4985 | sd_ctl_dir[0].child = entry; | |
4986 | ||
ad1cdc1d MM |
4987 | if (entry == NULL) |
4988 | return; | |
4989 | ||
6ad4c188 | 4990 | for_each_possible_cpu(i) { |
e692ab53 | 4991 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4992 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4993 | entry->mode = 0555; |
e692ab53 | 4994 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4995 | entry++; |
e692ab53 | 4996 | } |
7378547f MM |
4997 | |
4998 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4999 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5000 | } | |
6382bc90 | 5001 | |
7378547f | 5002 | /* may be called multiple times per register */ |
6382bc90 MM |
5003 | static void unregister_sched_domain_sysctl(void) |
5004 | { | |
7378547f MM |
5005 | if (sd_sysctl_header) |
5006 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5007 | sd_sysctl_header = NULL; |
7378547f MM |
5008 | if (sd_ctl_dir[0].child) |
5009 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5010 | } |
e692ab53 | 5011 | #else |
6382bc90 MM |
5012 | static void register_sched_domain_sysctl(void) |
5013 | { | |
5014 | } | |
5015 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5016 | { |
5017 | } | |
5018 | #endif | |
5019 | ||
1f11eb6a GH |
5020 | static void set_rq_online(struct rq *rq) |
5021 | { | |
5022 | if (!rq->online) { | |
5023 | const struct sched_class *class; | |
5024 | ||
c6c4927b | 5025 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5026 | rq->online = 1; |
5027 | ||
5028 | for_each_class(class) { | |
5029 | if (class->rq_online) | |
5030 | class->rq_online(rq); | |
5031 | } | |
5032 | } | |
5033 | } | |
5034 | ||
5035 | static void set_rq_offline(struct rq *rq) | |
5036 | { | |
5037 | if (rq->online) { | |
5038 | const struct sched_class *class; | |
5039 | ||
5040 | for_each_class(class) { | |
5041 | if (class->rq_offline) | |
5042 | class->rq_offline(rq); | |
5043 | } | |
5044 | ||
c6c4927b | 5045 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5046 | rq->online = 0; |
5047 | } | |
5048 | } | |
5049 | ||
1da177e4 LT |
5050 | /* |
5051 | * migration_call - callback that gets triggered when a CPU is added. | |
5052 | * Here we can start up the necessary migration thread for the new CPU. | |
5053 | */ | |
0db0628d | 5054 | static int |
48f24c4d | 5055 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5056 | { |
48f24c4d | 5057 | int cpu = (long)hcpu; |
1da177e4 | 5058 | unsigned long flags; |
969c7921 | 5059 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5060 | |
48c5ccae | 5061 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5062 | |
1da177e4 | 5063 | case CPU_UP_PREPARE: |
a468d389 | 5064 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5065 | break; |
48f24c4d | 5066 | |
1da177e4 | 5067 | case CPU_ONLINE: |
1f94ef59 | 5068 | /* Update our root-domain */ |
05fa785c | 5069 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5070 | if (rq->rd) { |
c6c4927b | 5071 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5072 | |
5073 | set_rq_online(rq); | |
1f94ef59 | 5074 | } |
05fa785c | 5075 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5076 | break; |
48f24c4d | 5077 | |
1da177e4 | 5078 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5079 | case CPU_DYING: |
317f3941 | 5080 | sched_ttwu_pending(); |
57d885fe | 5081 | /* Update our root-domain */ |
05fa785c | 5082 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5083 | if (rq->rd) { |
c6c4927b | 5084 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5085 | set_rq_offline(rq); |
57d885fe | 5086 | } |
48c5ccae PZ |
5087 | migrate_tasks(cpu); |
5088 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5089 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5090 | break; |
48c5ccae | 5091 | |
5d180232 | 5092 | case CPU_DEAD: |
f319da0c | 5093 | calc_load_migrate(rq); |
57d885fe | 5094 | break; |
1da177e4 LT |
5095 | #endif |
5096 | } | |
49c022e6 PZ |
5097 | |
5098 | update_max_interval(); | |
5099 | ||
1da177e4 LT |
5100 | return NOTIFY_OK; |
5101 | } | |
5102 | ||
f38b0820 PM |
5103 | /* |
5104 | * Register at high priority so that task migration (migrate_all_tasks) | |
5105 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5106 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5107 | */ |
0db0628d | 5108 | static struct notifier_block migration_notifier = { |
1da177e4 | 5109 | .notifier_call = migration_call, |
50a323b7 | 5110 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5111 | }; |
5112 | ||
a803f026 CM |
5113 | static void __cpuinit set_cpu_rq_start_time(void) |
5114 | { | |
5115 | int cpu = smp_processor_id(); | |
5116 | struct rq *rq = cpu_rq(cpu); | |
5117 | rq->age_stamp = sched_clock_cpu(cpu); | |
5118 | } | |
5119 | ||
0db0628d | 5120 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5121 | unsigned long action, void *hcpu) |
5122 | { | |
5123 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5124 | case CPU_STARTING: |
5125 | set_cpu_rq_start_time(); | |
5126 | return NOTIFY_OK; | |
3a101d05 TH |
5127 | case CPU_DOWN_FAILED: |
5128 | set_cpu_active((long)hcpu, true); | |
5129 | return NOTIFY_OK; | |
5130 | default: | |
5131 | return NOTIFY_DONE; | |
5132 | } | |
5133 | } | |
5134 | ||
0db0628d | 5135 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5136 | unsigned long action, void *hcpu) |
5137 | { | |
de212f18 PZ |
5138 | unsigned long flags; |
5139 | long cpu = (long)hcpu; | |
5140 | ||
3a101d05 TH |
5141 | switch (action & ~CPU_TASKS_FROZEN) { |
5142 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5143 | set_cpu_active(cpu, false); |
5144 | ||
5145 | /* explicitly allow suspend */ | |
5146 | if (!(action & CPU_TASKS_FROZEN)) { | |
5147 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5148 | bool overflow; | |
5149 | int cpus; | |
5150 | ||
5151 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5152 | cpus = dl_bw_cpus(cpu); | |
5153 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5154 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5155 | ||
5156 | if (overflow) | |
5157 | return notifier_from_errno(-EBUSY); | |
5158 | } | |
3a101d05 | 5159 | return NOTIFY_OK; |
3a101d05 | 5160 | } |
de212f18 PZ |
5161 | |
5162 | return NOTIFY_DONE; | |
3a101d05 TH |
5163 | } |
5164 | ||
7babe8db | 5165 | static int __init migration_init(void) |
1da177e4 LT |
5166 | { |
5167 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5168 | int err; |
48f24c4d | 5169 | |
3a101d05 | 5170 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5171 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5172 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5173 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5174 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5175 | |
3a101d05 TH |
5176 | /* Register cpu active notifiers */ |
5177 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5178 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5179 | ||
a004cd42 | 5180 | return 0; |
1da177e4 | 5181 | } |
7babe8db | 5182 | early_initcall(migration_init); |
1da177e4 LT |
5183 | #endif |
5184 | ||
5185 | #ifdef CONFIG_SMP | |
476f3534 | 5186 | |
4cb98839 PZ |
5187 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5188 | ||
3e9830dc | 5189 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5190 | |
d039ac60 | 5191 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5192 | |
d039ac60 | 5193 | static int __init sched_debug_setup(char *str) |
f6630114 | 5194 | { |
d039ac60 | 5195 | sched_debug_enabled = 1; |
f6630114 MT |
5196 | |
5197 | return 0; | |
5198 | } | |
d039ac60 PZ |
5199 | early_param("sched_debug", sched_debug_setup); |
5200 | ||
5201 | static inline bool sched_debug(void) | |
5202 | { | |
5203 | return sched_debug_enabled; | |
5204 | } | |
f6630114 | 5205 | |
7c16ec58 | 5206 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5207 | struct cpumask *groupmask) |
1da177e4 | 5208 | { |
4dcf6aff | 5209 | struct sched_group *group = sd->groups; |
434d53b0 | 5210 | char str[256]; |
1da177e4 | 5211 | |
968ea6d8 | 5212 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5213 | cpumask_clear(groupmask); |
4dcf6aff IM |
5214 | |
5215 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5216 | ||
5217 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5218 | printk("does not load-balance\n"); |
4dcf6aff | 5219 | if (sd->parent) |
3df0fc5b PZ |
5220 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5221 | " has parent"); | |
4dcf6aff | 5222 | return -1; |
41c7ce9a NP |
5223 | } |
5224 | ||
3df0fc5b | 5225 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5226 | |
758b2cdc | 5227 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5228 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5229 | "CPU%d\n", cpu); | |
4dcf6aff | 5230 | } |
758b2cdc | 5231 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5232 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5233 | " CPU%d\n", cpu); | |
4dcf6aff | 5234 | } |
1da177e4 | 5235 | |
4dcf6aff | 5236 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5237 | do { |
4dcf6aff | 5238 | if (!group) { |
3df0fc5b PZ |
5239 | printk("\n"); |
5240 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5241 | break; |
5242 | } | |
5243 | ||
c3decf0d PZ |
5244 | /* |
5245 | * Even though we initialize ->power to something semi-sane, | |
5246 | * we leave power_orig unset. This allows us to detect if | |
5247 | * domain iteration is still funny without causing /0 traps. | |
5248 | */ | |
5249 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5250 | printk(KERN_CONT "\n"); |
5251 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5252 | "set\n"); | |
4dcf6aff IM |
5253 | break; |
5254 | } | |
1da177e4 | 5255 | |
758b2cdc | 5256 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5257 | printk(KERN_CONT "\n"); |
5258 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5259 | break; |
5260 | } | |
1da177e4 | 5261 | |
cb83b629 PZ |
5262 | if (!(sd->flags & SD_OVERLAP) && |
5263 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5264 | printk(KERN_CONT "\n"); |
5265 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5266 | break; |
5267 | } | |
1da177e4 | 5268 | |
758b2cdc | 5269 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5270 | |
968ea6d8 | 5271 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5272 | |
3df0fc5b | 5273 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5274 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5275 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5276 | group->sgp->power); |
381512cf | 5277 | } |
1da177e4 | 5278 | |
4dcf6aff IM |
5279 | group = group->next; |
5280 | } while (group != sd->groups); | |
3df0fc5b | 5281 | printk(KERN_CONT "\n"); |
1da177e4 | 5282 | |
758b2cdc | 5283 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5284 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5285 | |
758b2cdc RR |
5286 | if (sd->parent && |
5287 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5288 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5289 | "of domain->span\n"); | |
4dcf6aff IM |
5290 | return 0; |
5291 | } | |
1da177e4 | 5292 | |
4dcf6aff IM |
5293 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5294 | { | |
5295 | int level = 0; | |
1da177e4 | 5296 | |
d039ac60 | 5297 | if (!sched_debug_enabled) |
f6630114 MT |
5298 | return; |
5299 | ||
4dcf6aff IM |
5300 | if (!sd) { |
5301 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5302 | return; | |
5303 | } | |
1da177e4 | 5304 | |
4dcf6aff IM |
5305 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5306 | ||
5307 | for (;;) { | |
4cb98839 | 5308 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5309 | break; |
1da177e4 LT |
5310 | level++; |
5311 | sd = sd->parent; | |
33859f7f | 5312 | if (!sd) |
4dcf6aff IM |
5313 | break; |
5314 | } | |
1da177e4 | 5315 | } |
6d6bc0ad | 5316 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5317 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5318 | static inline bool sched_debug(void) |
5319 | { | |
5320 | return false; | |
5321 | } | |
6d6bc0ad | 5322 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5323 | |
1a20ff27 | 5324 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5325 | { |
758b2cdc | 5326 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5327 | return 1; |
5328 | ||
5329 | /* Following flags need at least 2 groups */ | |
5330 | if (sd->flags & (SD_LOAD_BALANCE | | |
5331 | SD_BALANCE_NEWIDLE | | |
5332 | SD_BALANCE_FORK | | |
89c4710e SS |
5333 | SD_BALANCE_EXEC | |
5334 | SD_SHARE_CPUPOWER | | |
d77b3ed5 VG |
5335 | SD_SHARE_PKG_RESOURCES | |
5336 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5337 | if (sd->groups != sd->groups->next) |
5338 | return 0; | |
5339 | } | |
5340 | ||
5341 | /* Following flags don't use groups */ | |
c88d5910 | 5342 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5343 | return 0; |
5344 | ||
5345 | return 1; | |
5346 | } | |
5347 | ||
48f24c4d IM |
5348 | static int |
5349 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5350 | { |
5351 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5352 | ||
5353 | if (sd_degenerate(parent)) | |
5354 | return 1; | |
5355 | ||
758b2cdc | 5356 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5357 | return 0; |
5358 | ||
245af2c7 SS |
5359 | /* Flags needing groups don't count if only 1 group in parent */ |
5360 | if (parent->groups == parent->groups->next) { | |
5361 | pflags &= ~(SD_LOAD_BALANCE | | |
5362 | SD_BALANCE_NEWIDLE | | |
5363 | SD_BALANCE_FORK | | |
89c4710e SS |
5364 | SD_BALANCE_EXEC | |
5365 | SD_SHARE_CPUPOWER | | |
10866e62 | 5366 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5367 | SD_PREFER_SIBLING | |
5368 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5369 | if (nr_node_ids == 1) |
5370 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5371 | } |
5372 | if (~cflags & pflags) | |
5373 | return 0; | |
5374 | ||
5375 | return 1; | |
5376 | } | |
5377 | ||
dce840a0 | 5378 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5379 | { |
dce840a0 | 5380 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5381 | |
68e74568 | 5382 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5383 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5384 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5385 | free_cpumask_var(rd->rto_mask); |
5386 | free_cpumask_var(rd->online); | |
5387 | free_cpumask_var(rd->span); | |
5388 | kfree(rd); | |
5389 | } | |
5390 | ||
57d885fe GH |
5391 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5392 | { | |
a0490fa3 | 5393 | struct root_domain *old_rd = NULL; |
57d885fe | 5394 | unsigned long flags; |
57d885fe | 5395 | |
05fa785c | 5396 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5397 | |
5398 | if (rq->rd) { | |
a0490fa3 | 5399 | old_rd = rq->rd; |
57d885fe | 5400 | |
c6c4927b | 5401 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5402 | set_rq_offline(rq); |
57d885fe | 5403 | |
c6c4927b | 5404 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5405 | |
a0490fa3 | 5406 | /* |
0515973f | 5407 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5408 | * set old_rd to NULL to skip the freeing later |
5409 | * in this function: | |
5410 | */ | |
5411 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5412 | old_rd = NULL; | |
57d885fe GH |
5413 | } |
5414 | ||
5415 | atomic_inc(&rd->refcount); | |
5416 | rq->rd = rd; | |
5417 | ||
c6c4927b | 5418 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5419 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5420 | set_rq_online(rq); |
57d885fe | 5421 | |
05fa785c | 5422 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5423 | |
5424 | if (old_rd) | |
dce840a0 | 5425 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5426 | } |
5427 | ||
68c38fc3 | 5428 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5429 | { |
5430 | memset(rd, 0, sizeof(*rd)); | |
5431 | ||
68c38fc3 | 5432 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5433 | goto out; |
68c38fc3 | 5434 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5435 | goto free_span; |
1baca4ce | 5436 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5437 | goto free_online; |
1baca4ce JL |
5438 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5439 | goto free_dlo_mask; | |
6e0534f2 | 5440 | |
332ac17e | 5441 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5442 | if (cpudl_init(&rd->cpudl) != 0) |
5443 | goto free_dlo_mask; | |
332ac17e | 5444 | |
68c38fc3 | 5445 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5446 | goto free_rto_mask; |
c6c4927b | 5447 | return 0; |
6e0534f2 | 5448 | |
68e74568 RR |
5449 | free_rto_mask: |
5450 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5451 | free_dlo_mask: |
5452 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5453 | free_online: |
5454 | free_cpumask_var(rd->online); | |
5455 | free_span: | |
5456 | free_cpumask_var(rd->span); | |
0c910d28 | 5457 | out: |
c6c4927b | 5458 | return -ENOMEM; |
57d885fe GH |
5459 | } |
5460 | ||
029632fb PZ |
5461 | /* |
5462 | * By default the system creates a single root-domain with all cpus as | |
5463 | * members (mimicking the global state we have today). | |
5464 | */ | |
5465 | struct root_domain def_root_domain; | |
5466 | ||
57d885fe GH |
5467 | static void init_defrootdomain(void) |
5468 | { | |
68c38fc3 | 5469 | init_rootdomain(&def_root_domain); |
c6c4927b | 5470 | |
57d885fe GH |
5471 | atomic_set(&def_root_domain.refcount, 1); |
5472 | } | |
5473 | ||
dc938520 | 5474 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5475 | { |
5476 | struct root_domain *rd; | |
5477 | ||
5478 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5479 | if (!rd) | |
5480 | return NULL; | |
5481 | ||
68c38fc3 | 5482 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5483 | kfree(rd); |
5484 | return NULL; | |
5485 | } | |
57d885fe GH |
5486 | |
5487 | return rd; | |
5488 | } | |
5489 | ||
e3589f6c PZ |
5490 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5491 | { | |
5492 | struct sched_group *tmp, *first; | |
5493 | ||
5494 | if (!sg) | |
5495 | return; | |
5496 | ||
5497 | first = sg; | |
5498 | do { | |
5499 | tmp = sg->next; | |
5500 | ||
5501 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5502 | kfree(sg->sgp); | |
5503 | ||
5504 | kfree(sg); | |
5505 | sg = tmp; | |
5506 | } while (sg != first); | |
5507 | } | |
5508 | ||
dce840a0 PZ |
5509 | static void free_sched_domain(struct rcu_head *rcu) |
5510 | { | |
5511 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5512 | |
5513 | /* | |
5514 | * If its an overlapping domain it has private groups, iterate and | |
5515 | * nuke them all. | |
5516 | */ | |
5517 | if (sd->flags & SD_OVERLAP) { | |
5518 | free_sched_groups(sd->groups, 1); | |
5519 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5520 | kfree(sd->groups->sgp); |
dce840a0 | 5521 | kfree(sd->groups); |
9c3f75cb | 5522 | } |
dce840a0 PZ |
5523 | kfree(sd); |
5524 | } | |
5525 | ||
5526 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5527 | { | |
5528 | call_rcu(&sd->rcu, free_sched_domain); | |
5529 | } | |
5530 | ||
5531 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5532 | { | |
5533 | for (; sd; sd = sd->parent) | |
5534 | destroy_sched_domain(sd, cpu); | |
5535 | } | |
5536 | ||
518cd623 PZ |
5537 | /* |
5538 | * Keep a special pointer to the highest sched_domain that has | |
5539 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5540 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5541 | * | |
5542 | * Also keep a unique ID per domain (we use the first cpu number in | |
5543 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5544 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5545 | */ |
5546 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5547 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5548 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5549 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5550 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5551 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5552 | |
5553 | static void update_top_cache_domain(int cpu) | |
5554 | { | |
5555 | struct sched_domain *sd; | |
5d4cf996 | 5556 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5557 | int id = cpu; |
7d9ffa89 | 5558 | int size = 1; |
518cd623 PZ |
5559 | |
5560 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5561 | if (sd) { |
518cd623 | 5562 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5563 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5564 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5565 | } |
5d4cf996 | 5566 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5567 | |
5568 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5569 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5570 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5571 | |
5572 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5573 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5574 | |
5575 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5576 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5577 | } |
5578 | ||
1da177e4 | 5579 | /* |
0eab9146 | 5580 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5581 | * hold the hotplug lock. |
5582 | */ | |
0eab9146 IM |
5583 | static void |
5584 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5585 | { |
70b97a7f | 5586 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5587 | struct sched_domain *tmp; |
5588 | ||
5589 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5590 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5591 | struct sched_domain *parent = tmp->parent; |
5592 | if (!parent) | |
5593 | break; | |
f29c9b1c | 5594 | |
1a848870 | 5595 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5596 | tmp->parent = parent->parent; |
1a848870 SS |
5597 | if (parent->parent) |
5598 | parent->parent->child = tmp; | |
10866e62 PZ |
5599 | /* |
5600 | * Transfer SD_PREFER_SIBLING down in case of a | |
5601 | * degenerate parent; the spans match for this | |
5602 | * so the property transfers. | |
5603 | */ | |
5604 | if (parent->flags & SD_PREFER_SIBLING) | |
5605 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5606 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5607 | } else |
5608 | tmp = tmp->parent; | |
245af2c7 SS |
5609 | } |
5610 | ||
1a848870 | 5611 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5612 | tmp = sd; |
245af2c7 | 5613 | sd = sd->parent; |
dce840a0 | 5614 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5615 | if (sd) |
5616 | sd->child = NULL; | |
5617 | } | |
1da177e4 | 5618 | |
4cb98839 | 5619 | sched_domain_debug(sd, cpu); |
1da177e4 | 5620 | |
57d885fe | 5621 | rq_attach_root(rq, rd); |
dce840a0 | 5622 | tmp = rq->sd; |
674311d5 | 5623 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5624 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5625 | |
5626 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5627 | } |
5628 | ||
5629 | /* cpus with isolated domains */ | |
dcc30a35 | 5630 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5631 | |
5632 | /* Setup the mask of cpus configured for isolated domains */ | |
5633 | static int __init isolated_cpu_setup(char *str) | |
5634 | { | |
bdddd296 | 5635 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5636 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5637 | return 1; |
5638 | } | |
5639 | ||
8927f494 | 5640 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5641 | |
49a02c51 | 5642 | struct s_data { |
21d42ccf | 5643 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5644 | struct root_domain *rd; |
5645 | }; | |
5646 | ||
2109b99e | 5647 | enum s_alloc { |
2109b99e | 5648 | sa_rootdomain, |
21d42ccf | 5649 | sa_sd, |
dce840a0 | 5650 | sa_sd_storage, |
2109b99e AH |
5651 | sa_none, |
5652 | }; | |
5653 | ||
c1174876 PZ |
5654 | /* |
5655 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5656 | * domain traversal. | |
5657 | * | |
5658 | * Asymmetric node setups can result in situations where the domain tree is of | |
5659 | * unequal depth, make sure to skip domains that already cover the entire | |
5660 | * range. | |
5661 | * | |
5662 | * In that case build_sched_domains() will have terminated the iteration early | |
5663 | * and our sibling sd spans will be empty. Domains should always include the | |
5664 | * cpu they're built on, so check that. | |
5665 | * | |
5666 | */ | |
5667 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5668 | { | |
5669 | const struct cpumask *span = sched_domain_span(sd); | |
5670 | struct sd_data *sdd = sd->private; | |
5671 | struct sched_domain *sibling; | |
5672 | int i; | |
5673 | ||
5674 | for_each_cpu(i, span) { | |
5675 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5676 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5677 | continue; | |
5678 | ||
5679 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5680 | } | |
5681 | } | |
5682 | ||
5683 | /* | |
5684 | * Return the canonical balance cpu for this group, this is the first cpu | |
5685 | * of this group that's also in the iteration mask. | |
5686 | */ | |
5687 | int group_balance_cpu(struct sched_group *sg) | |
5688 | { | |
5689 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5690 | } | |
5691 | ||
e3589f6c PZ |
5692 | static int |
5693 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5694 | { | |
5695 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5696 | const struct cpumask *span = sched_domain_span(sd); | |
5697 | struct cpumask *covered = sched_domains_tmpmask; | |
5698 | struct sd_data *sdd = sd->private; | |
5699 | struct sched_domain *child; | |
5700 | int i; | |
5701 | ||
5702 | cpumask_clear(covered); | |
5703 | ||
5704 | for_each_cpu(i, span) { | |
5705 | struct cpumask *sg_span; | |
5706 | ||
5707 | if (cpumask_test_cpu(i, covered)) | |
5708 | continue; | |
5709 | ||
c1174876 PZ |
5710 | child = *per_cpu_ptr(sdd->sd, i); |
5711 | ||
5712 | /* See the comment near build_group_mask(). */ | |
5713 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5714 | continue; | |
5715 | ||
e3589f6c | 5716 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5717 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5718 | |
5719 | if (!sg) | |
5720 | goto fail; | |
5721 | ||
5722 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5723 | if (child->child) { |
5724 | child = child->child; | |
5725 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5726 | } else | |
5727 | cpumask_set_cpu(i, sg_span); | |
5728 | ||
5729 | cpumask_or(covered, covered, sg_span); | |
5730 | ||
74a5ce20 | 5731 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5732 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5733 | build_group_mask(sd, sg); | |
5734 | ||
c3decf0d PZ |
5735 | /* |
5736 | * Initialize sgp->power such that even if we mess up the | |
5737 | * domains and no possible iteration will get us here, we won't | |
5738 | * die on a /0 trap. | |
5739 | */ | |
5740 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
8e8339a3 | 5741 | sg->sgp->power_orig = sg->sgp->power; |
e3589f6c | 5742 | |
c1174876 PZ |
5743 | /* |
5744 | * Make sure the first group of this domain contains the | |
5745 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5746 | * breaks. See update_sg_lb_stats(). | |
5747 | */ | |
74a5ce20 | 5748 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5749 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5750 | groups = sg; |
5751 | ||
5752 | if (!first) | |
5753 | first = sg; | |
5754 | if (last) | |
5755 | last->next = sg; | |
5756 | last = sg; | |
5757 | last->next = first; | |
5758 | } | |
5759 | sd->groups = groups; | |
5760 | ||
5761 | return 0; | |
5762 | ||
5763 | fail: | |
5764 | free_sched_groups(first, 0); | |
5765 | ||
5766 | return -ENOMEM; | |
5767 | } | |
5768 | ||
dce840a0 | 5769 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5770 | { |
dce840a0 PZ |
5771 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5772 | struct sched_domain *child = sd->child; | |
1da177e4 | 5773 | |
dce840a0 PZ |
5774 | if (child) |
5775 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5776 | |
9c3f75cb | 5777 | if (sg) { |
dce840a0 | 5778 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5779 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5780 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5781 | } |
dce840a0 PZ |
5782 | |
5783 | return cpu; | |
1e9f28fa | 5784 | } |
1e9f28fa | 5785 | |
01a08546 | 5786 | /* |
dce840a0 PZ |
5787 | * build_sched_groups will build a circular linked list of the groups |
5788 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5789 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5790 | * |
5791 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5792 | */ |
e3589f6c PZ |
5793 | static int |
5794 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5795 | { |
dce840a0 PZ |
5796 | struct sched_group *first = NULL, *last = NULL; |
5797 | struct sd_data *sdd = sd->private; | |
5798 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5799 | struct cpumask *covered; |
dce840a0 | 5800 | int i; |
9c1cfda2 | 5801 | |
e3589f6c PZ |
5802 | get_group(cpu, sdd, &sd->groups); |
5803 | atomic_inc(&sd->groups->ref); | |
5804 | ||
0936629f | 5805 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5806 | return 0; |
5807 | ||
f96225fd PZ |
5808 | lockdep_assert_held(&sched_domains_mutex); |
5809 | covered = sched_domains_tmpmask; | |
5810 | ||
dce840a0 | 5811 | cpumask_clear(covered); |
6711cab4 | 5812 | |
dce840a0 PZ |
5813 | for_each_cpu(i, span) { |
5814 | struct sched_group *sg; | |
cd08e923 | 5815 | int group, j; |
6711cab4 | 5816 | |
dce840a0 PZ |
5817 | if (cpumask_test_cpu(i, covered)) |
5818 | continue; | |
6711cab4 | 5819 | |
cd08e923 | 5820 | group = get_group(i, sdd, &sg); |
c1174876 | 5821 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5822 | |
dce840a0 PZ |
5823 | for_each_cpu(j, span) { |
5824 | if (get_group(j, sdd, NULL) != group) | |
5825 | continue; | |
0601a88d | 5826 | |
dce840a0 PZ |
5827 | cpumask_set_cpu(j, covered); |
5828 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5829 | } | |
0601a88d | 5830 | |
dce840a0 PZ |
5831 | if (!first) |
5832 | first = sg; | |
5833 | if (last) | |
5834 | last->next = sg; | |
5835 | last = sg; | |
5836 | } | |
5837 | last->next = first; | |
e3589f6c PZ |
5838 | |
5839 | return 0; | |
0601a88d | 5840 | } |
51888ca2 | 5841 | |
89c4710e SS |
5842 | /* |
5843 | * Initialize sched groups cpu_power. | |
5844 | * | |
5845 | * cpu_power indicates the capacity of sched group, which is used while | |
5846 | * distributing the load between different sched groups in a sched domain. | |
5847 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5848 | * there are asymmetries in the topology. If there are asymmetries, group | |
5849 | * having more cpu_power will pickup more load compared to the group having | |
5850 | * less cpu_power. | |
89c4710e SS |
5851 | */ |
5852 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5853 | { | |
e3589f6c | 5854 | struct sched_group *sg = sd->groups; |
89c4710e | 5855 | |
94c95ba6 | 5856 | WARN_ON(!sg); |
e3589f6c PZ |
5857 | |
5858 | do { | |
5859 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5860 | sg = sg->next; | |
5861 | } while (sg != sd->groups); | |
89c4710e | 5862 | |
c1174876 | 5863 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5864 | return; |
aae6d3dd | 5865 | |
d274cb30 | 5866 | update_group_power(sd, cpu); |
69e1e811 | 5867 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5868 | } |
5869 | ||
7c16ec58 MT |
5870 | /* |
5871 | * Initializers for schedule domains | |
5872 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5873 | */ | |
5874 | ||
1d3504fc | 5875 | static int default_relax_domain_level = -1; |
60495e77 | 5876 | int sched_domain_level_max; |
1d3504fc HS |
5877 | |
5878 | static int __init setup_relax_domain_level(char *str) | |
5879 | { | |
a841f8ce DS |
5880 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5881 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5882 | |
1d3504fc HS |
5883 | return 1; |
5884 | } | |
5885 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5886 | ||
5887 | static void set_domain_attribute(struct sched_domain *sd, | |
5888 | struct sched_domain_attr *attr) | |
5889 | { | |
5890 | int request; | |
5891 | ||
5892 | if (!attr || attr->relax_domain_level < 0) { | |
5893 | if (default_relax_domain_level < 0) | |
5894 | return; | |
5895 | else | |
5896 | request = default_relax_domain_level; | |
5897 | } else | |
5898 | request = attr->relax_domain_level; | |
5899 | if (request < sd->level) { | |
5900 | /* turn off idle balance on this domain */ | |
c88d5910 | 5901 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5902 | } else { |
5903 | /* turn on idle balance on this domain */ | |
c88d5910 | 5904 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5905 | } |
5906 | } | |
5907 | ||
54ab4ff4 PZ |
5908 | static void __sdt_free(const struct cpumask *cpu_map); |
5909 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5910 | ||
2109b99e AH |
5911 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5912 | const struct cpumask *cpu_map) | |
5913 | { | |
5914 | switch (what) { | |
2109b99e | 5915 | case sa_rootdomain: |
822ff793 PZ |
5916 | if (!atomic_read(&d->rd->refcount)) |
5917 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5918 | case sa_sd: |
5919 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5920 | case sa_sd_storage: |
54ab4ff4 | 5921 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5922 | case sa_none: |
5923 | break; | |
5924 | } | |
5925 | } | |
3404c8d9 | 5926 | |
2109b99e AH |
5927 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5928 | const struct cpumask *cpu_map) | |
5929 | { | |
dce840a0 PZ |
5930 | memset(d, 0, sizeof(*d)); |
5931 | ||
54ab4ff4 PZ |
5932 | if (__sdt_alloc(cpu_map)) |
5933 | return sa_sd_storage; | |
dce840a0 PZ |
5934 | d->sd = alloc_percpu(struct sched_domain *); |
5935 | if (!d->sd) | |
5936 | return sa_sd_storage; | |
2109b99e | 5937 | d->rd = alloc_rootdomain(); |
dce840a0 | 5938 | if (!d->rd) |
21d42ccf | 5939 | return sa_sd; |
2109b99e AH |
5940 | return sa_rootdomain; |
5941 | } | |
57d885fe | 5942 | |
dce840a0 PZ |
5943 | /* |
5944 | * NULL the sd_data elements we've used to build the sched_domain and | |
5945 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5946 | * will not free the data we're using. | |
5947 | */ | |
5948 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5949 | { | |
5950 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5951 | |
5952 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5953 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5954 | ||
e3589f6c | 5955 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5956 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5957 | |
5958 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5959 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5960 | } |
5961 | ||
cb83b629 | 5962 | #ifdef CONFIG_NUMA |
cb83b629 | 5963 | static int sched_domains_numa_levels; |
cb83b629 PZ |
5964 | static int *sched_domains_numa_distance; |
5965 | static struct cpumask ***sched_domains_numa_masks; | |
5966 | static int sched_domains_curr_level; | |
143e1e28 | 5967 | #endif |
cb83b629 | 5968 | |
143e1e28 VG |
5969 | /* |
5970 | * SD_flags allowed in topology descriptions. | |
5971 | * | |
5972 | * SD_SHARE_CPUPOWER - describes SMT topologies | |
5973 | * SD_SHARE_PKG_RESOURCES - describes shared caches | |
5974 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 5975 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
5976 | * |
5977 | * Odd one out: | |
5978 | * SD_ASYM_PACKING - describes SMT quirks | |
5979 | */ | |
5980 | #define TOPOLOGY_SD_FLAGS \ | |
5981 | (SD_SHARE_CPUPOWER | \ | |
5982 | SD_SHARE_PKG_RESOURCES | \ | |
5983 | SD_NUMA | \ | |
d77b3ed5 VG |
5984 | SD_ASYM_PACKING | \ |
5985 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
5986 | |
5987 | static struct sched_domain * | |
143e1e28 | 5988 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
5989 | { |
5990 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
5991 | int sd_weight, sd_flags = 0; |
5992 | ||
5993 | #ifdef CONFIG_NUMA | |
5994 | /* | |
5995 | * Ugly hack to pass state to sd_numa_mask()... | |
5996 | */ | |
5997 | sched_domains_curr_level = tl->numa_level; | |
5998 | #endif | |
5999 | ||
6000 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6001 | ||
6002 | if (tl->sd_flags) | |
6003 | sd_flags = (*tl->sd_flags)(); | |
6004 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6005 | "wrong sd_flags in topology description\n")) | |
6006 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6007 | |
6008 | *sd = (struct sched_domain){ | |
6009 | .min_interval = sd_weight, | |
6010 | .max_interval = 2*sd_weight, | |
6011 | .busy_factor = 32, | |
870a0bb5 | 6012 | .imbalance_pct = 125, |
143e1e28 VG |
6013 | |
6014 | .cache_nice_tries = 0, | |
6015 | .busy_idx = 0, | |
6016 | .idle_idx = 0, | |
cb83b629 PZ |
6017 | .newidle_idx = 0, |
6018 | .wake_idx = 0, | |
6019 | .forkexec_idx = 0, | |
6020 | ||
6021 | .flags = 1*SD_LOAD_BALANCE | |
6022 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6023 | | 1*SD_BALANCE_EXEC |
6024 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6025 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6026 | | 1*SD_WAKE_AFFINE |
cb83b629 | 6027 | | 0*SD_SHARE_CPUPOWER |
cb83b629 | 6028 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6029 | | 0*SD_SERIALIZE |
cb83b629 | 6030 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6031 | | 0*SD_NUMA |
6032 | | sd_flags | |
cb83b629 | 6033 | , |
143e1e28 | 6034 | |
cb83b629 PZ |
6035 | .last_balance = jiffies, |
6036 | .balance_interval = sd_weight, | |
143e1e28 | 6037 | .smt_gain = 0, |
2b4cfe64 JL |
6038 | .max_newidle_lb_cost = 0, |
6039 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6040 | #ifdef CONFIG_SCHED_DEBUG |
6041 | .name = tl->name, | |
6042 | #endif | |
cb83b629 | 6043 | }; |
cb83b629 PZ |
6044 | |
6045 | /* | |
143e1e28 | 6046 | * Convert topological properties into behaviour. |
cb83b629 | 6047 | */ |
143e1e28 VG |
6048 | |
6049 | if (sd->flags & SD_SHARE_CPUPOWER) { | |
6050 | sd->imbalance_pct = 110; | |
6051 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6052 | |
6053 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6054 | sd->imbalance_pct = 117; | |
6055 | sd->cache_nice_tries = 1; | |
6056 | sd->busy_idx = 2; | |
6057 | ||
6058 | #ifdef CONFIG_NUMA | |
6059 | } else if (sd->flags & SD_NUMA) { | |
6060 | sd->cache_nice_tries = 2; | |
6061 | sd->busy_idx = 3; | |
6062 | sd->idle_idx = 2; | |
6063 | ||
6064 | sd->flags |= SD_SERIALIZE; | |
6065 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6066 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6067 | SD_BALANCE_FORK | | |
6068 | SD_WAKE_AFFINE); | |
6069 | } | |
6070 | ||
6071 | #endif | |
6072 | } else { | |
6073 | sd->flags |= SD_PREFER_SIBLING; | |
6074 | sd->cache_nice_tries = 1; | |
6075 | sd->busy_idx = 2; | |
6076 | sd->idle_idx = 1; | |
6077 | } | |
6078 | ||
6079 | sd->private = &tl->data; | |
cb83b629 PZ |
6080 | |
6081 | return sd; | |
6082 | } | |
6083 | ||
143e1e28 VG |
6084 | /* |
6085 | * Topology list, bottom-up. | |
6086 | */ | |
6087 | static struct sched_domain_topology_level default_topology[] = { | |
6088 | #ifdef CONFIG_SCHED_SMT | |
6089 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6090 | #endif | |
6091 | #ifdef CONFIG_SCHED_MC | |
6092 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6093 | #endif |
6094 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6095 | { NULL, }, | |
6096 | }; | |
6097 | ||
6098 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6099 | ||
6100 | #define for_each_sd_topology(tl) \ | |
6101 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6102 | ||
6103 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6104 | { | |
6105 | sched_domain_topology = tl; | |
6106 | } | |
6107 | ||
6108 | #ifdef CONFIG_NUMA | |
6109 | ||
cb83b629 PZ |
6110 | static const struct cpumask *sd_numa_mask(int cpu) |
6111 | { | |
6112 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6113 | } | |
6114 | ||
d039ac60 PZ |
6115 | static void sched_numa_warn(const char *str) |
6116 | { | |
6117 | static int done = false; | |
6118 | int i,j; | |
6119 | ||
6120 | if (done) | |
6121 | return; | |
6122 | ||
6123 | done = true; | |
6124 | ||
6125 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6126 | ||
6127 | for (i = 0; i < nr_node_ids; i++) { | |
6128 | printk(KERN_WARNING " "); | |
6129 | for (j = 0; j < nr_node_ids; j++) | |
6130 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6131 | printk(KERN_CONT "\n"); | |
6132 | } | |
6133 | printk(KERN_WARNING "\n"); | |
6134 | } | |
6135 | ||
6136 | static bool find_numa_distance(int distance) | |
6137 | { | |
6138 | int i; | |
6139 | ||
6140 | if (distance == node_distance(0, 0)) | |
6141 | return true; | |
6142 | ||
6143 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6144 | if (sched_domains_numa_distance[i] == distance) | |
6145 | return true; | |
6146 | } | |
6147 | ||
6148 | return false; | |
6149 | } | |
6150 | ||
cb83b629 PZ |
6151 | static void sched_init_numa(void) |
6152 | { | |
6153 | int next_distance, curr_distance = node_distance(0, 0); | |
6154 | struct sched_domain_topology_level *tl; | |
6155 | int level = 0; | |
6156 | int i, j, k; | |
6157 | ||
cb83b629 PZ |
6158 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6159 | if (!sched_domains_numa_distance) | |
6160 | return; | |
6161 | ||
6162 | /* | |
6163 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6164 | * unique distances in the node_distance() table. | |
6165 | * | |
6166 | * Assumes node_distance(0,j) includes all distances in | |
6167 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6168 | */ |
6169 | next_distance = curr_distance; | |
6170 | for (i = 0; i < nr_node_ids; i++) { | |
6171 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6172 | for (k = 0; k < nr_node_ids; k++) { |
6173 | int distance = node_distance(i, k); | |
6174 | ||
6175 | if (distance > curr_distance && | |
6176 | (distance < next_distance || | |
6177 | next_distance == curr_distance)) | |
6178 | next_distance = distance; | |
6179 | ||
6180 | /* | |
6181 | * While not a strong assumption it would be nice to know | |
6182 | * about cases where if node A is connected to B, B is not | |
6183 | * equally connected to A. | |
6184 | */ | |
6185 | if (sched_debug() && node_distance(k, i) != distance) | |
6186 | sched_numa_warn("Node-distance not symmetric"); | |
6187 | ||
6188 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6189 | sched_numa_warn("Node-0 not representative"); | |
6190 | } | |
6191 | if (next_distance != curr_distance) { | |
6192 | sched_domains_numa_distance[level++] = next_distance; | |
6193 | sched_domains_numa_levels = level; | |
6194 | curr_distance = next_distance; | |
6195 | } else break; | |
cb83b629 | 6196 | } |
d039ac60 PZ |
6197 | |
6198 | /* | |
6199 | * In case of sched_debug() we verify the above assumption. | |
6200 | */ | |
6201 | if (!sched_debug()) | |
6202 | break; | |
cb83b629 PZ |
6203 | } |
6204 | /* | |
6205 | * 'level' contains the number of unique distances, excluding the | |
6206 | * identity distance node_distance(i,i). | |
6207 | * | |
28b4a521 | 6208 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6209 | * numbers. |
6210 | */ | |
6211 | ||
5f7865f3 TC |
6212 | /* |
6213 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6214 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6215 | * the array will contain less then 'level' members. This could be | |
6216 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6217 | * in other functions. | |
6218 | * | |
6219 | * We reset it to 'level' at the end of this function. | |
6220 | */ | |
6221 | sched_domains_numa_levels = 0; | |
6222 | ||
cb83b629 PZ |
6223 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6224 | if (!sched_domains_numa_masks) | |
6225 | return; | |
6226 | ||
6227 | /* | |
6228 | * Now for each level, construct a mask per node which contains all | |
6229 | * cpus of nodes that are that many hops away from us. | |
6230 | */ | |
6231 | for (i = 0; i < level; i++) { | |
6232 | sched_domains_numa_masks[i] = | |
6233 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6234 | if (!sched_domains_numa_masks[i]) | |
6235 | return; | |
6236 | ||
6237 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6238 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6239 | if (!mask) |
6240 | return; | |
6241 | ||
6242 | sched_domains_numa_masks[i][j] = mask; | |
6243 | ||
6244 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6245 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6246 | continue; |
6247 | ||
6248 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6249 | } | |
6250 | } | |
6251 | } | |
6252 | ||
143e1e28 VG |
6253 | /* Compute default topology size */ |
6254 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6255 | ||
c515db8c | 6256 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6257 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6258 | if (!tl) | |
6259 | return; | |
6260 | ||
6261 | /* | |
6262 | * Copy the default topology bits.. | |
6263 | */ | |
143e1e28 VG |
6264 | for (i = 0; sched_domain_topology[i].mask; i++) |
6265 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6266 | |
6267 | /* | |
6268 | * .. and append 'j' levels of NUMA goodness. | |
6269 | */ | |
6270 | for (j = 0; j < level; i++, j++) { | |
6271 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6272 | .mask = sd_numa_mask, |
143e1e28 | 6273 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6274 | .flags = SDTL_OVERLAP, |
6275 | .numa_level = j, | |
143e1e28 | 6276 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6277 | }; |
6278 | } | |
6279 | ||
6280 | sched_domain_topology = tl; | |
5f7865f3 TC |
6281 | |
6282 | sched_domains_numa_levels = level; | |
cb83b629 | 6283 | } |
301a5cba TC |
6284 | |
6285 | static void sched_domains_numa_masks_set(int cpu) | |
6286 | { | |
6287 | int i, j; | |
6288 | int node = cpu_to_node(cpu); | |
6289 | ||
6290 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6291 | for (j = 0; j < nr_node_ids; j++) { | |
6292 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6293 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6294 | } | |
6295 | } | |
6296 | } | |
6297 | ||
6298 | static void sched_domains_numa_masks_clear(int cpu) | |
6299 | { | |
6300 | int i, j; | |
6301 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6302 | for (j = 0; j < nr_node_ids; j++) | |
6303 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6304 | } | |
6305 | } | |
6306 | ||
6307 | /* | |
6308 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6309 | * are onlined. | |
6310 | */ | |
6311 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6312 | unsigned long action, | |
6313 | void *hcpu) | |
6314 | { | |
6315 | int cpu = (long)hcpu; | |
6316 | ||
6317 | switch (action & ~CPU_TASKS_FROZEN) { | |
6318 | case CPU_ONLINE: | |
6319 | sched_domains_numa_masks_set(cpu); | |
6320 | break; | |
6321 | ||
6322 | case CPU_DEAD: | |
6323 | sched_domains_numa_masks_clear(cpu); | |
6324 | break; | |
6325 | ||
6326 | default: | |
6327 | return NOTIFY_DONE; | |
6328 | } | |
6329 | ||
6330 | return NOTIFY_OK; | |
cb83b629 PZ |
6331 | } |
6332 | #else | |
6333 | static inline void sched_init_numa(void) | |
6334 | { | |
6335 | } | |
301a5cba TC |
6336 | |
6337 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6338 | unsigned long action, | |
6339 | void *hcpu) | |
6340 | { | |
6341 | return 0; | |
6342 | } | |
cb83b629 PZ |
6343 | #endif /* CONFIG_NUMA */ |
6344 | ||
54ab4ff4 PZ |
6345 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6346 | { | |
6347 | struct sched_domain_topology_level *tl; | |
6348 | int j; | |
6349 | ||
27723a68 | 6350 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6351 | struct sd_data *sdd = &tl->data; |
6352 | ||
6353 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6354 | if (!sdd->sd) | |
6355 | return -ENOMEM; | |
6356 | ||
6357 | sdd->sg = alloc_percpu(struct sched_group *); | |
6358 | if (!sdd->sg) | |
6359 | return -ENOMEM; | |
6360 | ||
9c3f75cb PZ |
6361 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6362 | if (!sdd->sgp) | |
6363 | return -ENOMEM; | |
6364 | ||
54ab4ff4 PZ |
6365 | for_each_cpu(j, cpu_map) { |
6366 | struct sched_domain *sd; | |
6367 | struct sched_group *sg; | |
9c3f75cb | 6368 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6369 | |
6370 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6371 | GFP_KERNEL, cpu_to_node(j)); | |
6372 | if (!sd) | |
6373 | return -ENOMEM; | |
6374 | ||
6375 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6376 | ||
6377 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6378 | GFP_KERNEL, cpu_to_node(j)); | |
6379 | if (!sg) | |
6380 | return -ENOMEM; | |
6381 | ||
30b4e9eb IM |
6382 | sg->next = sg; |
6383 | ||
54ab4ff4 | 6384 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6385 | |
c1174876 | 6386 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6387 | GFP_KERNEL, cpu_to_node(j)); |
6388 | if (!sgp) | |
6389 | return -ENOMEM; | |
6390 | ||
6391 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6392 | } |
6393 | } | |
6394 | ||
6395 | return 0; | |
6396 | } | |
6397 | ||
6398 | static void __sdt_free(const struct cpumask *cpu_map) | |
6399 | { | |
6400 | struct sched_domain_topology_level *tl; | |
6401 | int j; | |
6402 | ||
27723a68 | 6403 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6404 | struct sd_data *sdd = &tl->data; |
6405 | ||
6406 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6407 | struct sched_domain *sd; |
6408 | ||
6409 | if (sdd->sd) { | |
6410 | sd = *per_cpu_ptr(sdd->sd, j); | |
6411 | if (sd && (sd->flags & SD_OVERLAP)) | |
6412 | free_sched_groups(sd->groups, 0); | |
6413 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6414 | } | |
6415 | ||
6416 | if (sdd->sg) | |
6417 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6418 | if (sdd->sgp) | |
6419 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6420 | } |
6421 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6422 | sdd->sd = NULL; |
54ab4ff4 | 6423 | free_percpu(sdd->sg); |
fb2cf2c6 | 6424 | sdd->sg = NULL; |
9c3f75cb | 6425 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6426 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6427 | } |
6428 | } | |
6429 | ||
2c402dc3 | 6430 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6431 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6432 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6433 | { |
143e1e28 | 6434 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6435 | if (!sd) |
d069b916 | 6436 | return child; |
2c402dc3 | 6437 | |
2c402dc3 | 6438 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6439 | if (child) { |
6440 | sd->level = child->level + 1; | |
6441 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6442 | child->parent = sd; |
c75e0128 | 6443 | sd->child = child; |
60495e77 | 6444 | } |
a841f8ce | 6445 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6446 | |
6447 | return sd; | |
6448 | } | |
6449 | ||
2109b99e AH |
6450 | /* |
6451 | * Build sched domains for a given set of cpus and attach the sched domains | |
6452 | * to the individual cpus | |
6453 | */ | |
dce840a0 PZ |
6454 | static int build_sched_domains(const struct cpumask *cpu_map, |
6455 | struct sched_domain_attr *attr) | |
2109b99e | 6456 | { |
1c632169 | 6457 | enum s_alloc alloc_state; |
dce840a0 | 6458 | struct sched_domain *sd; |
2109b99e | 6459 | struct s_data d; |
822ff793 | 6460 | int i, ret = -ENOMEM; |
9c1cfda2 | 6461 | |
2109b99e AH |
6462 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6463 | if (alloc_state != sa_rootdomain) | |
6464 | goto error; | |
9c1cfda2 | 6465 | |
dce840a0 | 6466 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6467 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6468 | struct sched_domain_topology_level *tl; |
6469 | ||
3bd65a80 | 6470 | sd = NULL; |
27723a68 | 6471 | for_each_sd_topology(tl) { |
4a850cbe | 6472 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6473 | if (tl == sched_domain_topology) |
6474 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6475 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6476 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6477 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6478 | break; | |
e3589f6c | 6479 | } |
dce840a0 PZ |
6480 | } |
6481 | ||
6482 | /* Build the groups for the domains */ | |
6483 | for_each_cpu(i, cpu_map) { | |
6484 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6485 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6486 | if (sd->flags & SD_OVERLAP) { |
6487 | if (build_overlap_sched_groups(sd, i)) | |
6488 | goto error; | |
6489 | } else { | |
6490 | if (build_sched_groups(sd, i)) | |
6491 | goto error; | |
6492 | } | |
1cf51902 | 6493 | } |
a06dadbe | 6494 | } |
9c1cfda2 | 6495 | |
1da177e4 | 6496 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6497 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6498 | if (!cpumask_test_cpu(i, cpu_map)) | |
6499 | continue; | |
9c1cfda2 | 6500 | |
dce840a0 PZ |
6501 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6502 | claim_allocations(i, sd); | |
cd4ea6ae | 6503 | init_sched_groups_power(i, sd); |
dce840a0 | 6504 | } |
f712c0c7 | 6505 | } |
9c1cfda2 | 6506 | |
1da177e4 | 6507 | /* Attach the domains */ |
dce840a0 | 6508 | rcu_read_lock(); |
abcd083a | 6509 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6510 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6511 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6512 | } |
dce840a0 | 6513 | rcu_read_unlock(); |
51888ca2 | 6514 | |
822ff793 | 6515 | ret = 0; |
51888ca2 | 6516 | error: |
2109b99e | 6517 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6518 | return ret; |
1da177e4 | 6519 | } |
029190c5 | 6520 | |
acc3f5d7 | 6521 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6522 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6523 | static struct sched_domain_attr *dattr_cur; |
6524 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6525 | |
6526 | /* | |
6527 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6528 | * cpumask) fails, then fallback to a single sched domain, |
6529 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6530 | */ |
4212823f | 6531 | static cpumask_var_t fallback_doms; |
029190c5 | 6532 | |
ee79d1bd HC |
6533 | /* |
6534 | * arch_update_cpu_topology lets virtualized architectures update the | |
6535 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6536 | * or 0 if it stayed the same. | |
6537 | */ | |
52f5684c | 6538 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6539 | { |
ee79d1bd | 6540 | return 0; |
22e52b07 HC |
6541 | } |
6542 | ||
acc3f5d7 RR |
6543 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6544 | { | |
6545 | int i; | |
6546 | cpumask_var_t *doms; | |
6547 | ||
6548 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6549 | if (!doms) | |
6550 | return NULL; | |
6551 | for (i = 0; i < ndoms; i++) { | |
6552 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6553 | free_sched_domains(doms, i); | |
6554 | return NULL; | |
6555 | } | |
6556 | } | |
6557 | return doms; | |
6558 | } | |
6559 | ||
6560 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6561 | { | |
6562 | unsigned int i; | |
6563 | for (i = 0; i < ndoms; i++) | |
6564 | free_cpumask_var(doms[i]); | |
6565 | kfree(doms); | |
6566 | } | |
6567 | ||
1a20ff27 | 6568 | /* |
41a2d6cf | 6569 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6570 | * For now this just excludes isolated cpus, but could be used to |
6571 | * exclude other special cases in the future. | |
1a20ff27 | 6572 | */ |
c4a8849a | 6573 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6574 | { |
7378547f MM |
6575 | int err; |
6576 | ||
22e52b07 | 6577 | arch_update_cpu_topology(); |
029190c5 | 6578 | ndoms_cur = 1; |
acc3f5d7 | 6579 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6580 | if (!doms_cur) |
acc3f5d7 RR |
6581 | doms_cur = &fallback_doms; |
6582 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6583 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6584 | register_sched_domain_sysctl(); |
7378547f MM |
6585 | |
6586 | return err; | |
1a20ff27 DG |
6587 | } |
6588 | ||
1a20ff27 DG |
6589 | /* |
6590 | * Detach sched domains from a group of cpus specified in cpu_map | |
6591 | * These cpus will now be attached to the NULL domain | |
6592 | */ | |
96f874e2 | 6593 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6594 | { |
6595 | int i; | |
6596 | ||
dce840a0 | 6597 | rcu_read_lock(); |
abcd083a | 6598 | for_each_cpu(i, cpu_map) |
57d885fe | 6599 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6600 | rcu_read_unlock(); |
1a20ff27 DG |
6601 | } |
6602 | ||
1d3504fc HS |
6603 | /* handle null as "default" */ |
6604 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6605 | struct sched_domain_attr *new, int idx_new) | |
6606 | { | |
6607 | struct sched_domain_attr tmp; | |
6608 | ||
6609 | /* fast path */ | |
6610 | if (!new && !cur) | |
6611 | return 1; | |
6612 | ||
6613 | tmp = SD_ATTR_INIT; | |
6614 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6615 | new ? (new + idx_new) : &tmp, | |
6616 | sizeof(struct sched_domain_attr)); | |
6617 | } | |
6618 | ||
029190c5 PJ |
6619 | /* |
6620 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6621 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6622 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6623 | * It destroys each deleted domain and builds each new domain. | |
6624 | * | |
acc3f5d7 | 6625 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6626 | * The masks don't intersect (don't overlap.) We should setup one |
6627 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6628 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6629 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6630 | * it as it is. | |
6631 | * | |
acc3f5d7 RR |
6632 | * The passed in 'doms_new' should be allocated using |
6633 | * alloc_sched_domains. This routine takes ownership of it and will | |
6634 | * free_sched_domains it when done with it. If the caller failed the | |
6635 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6636 | * and partition_sched_domains() will fallback to the single partition | |
6637 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6638 | * |
96f874e2 | 6639 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6640 | * ndoms_new == 0 is a special case for destroying existing domains, |
6641 | * and it will not create the default domain. | |
dfb512ec | 6642 | * |
029190c5 PJ |
6643 | * Call with hotplug lock held |
6644 | */ | |
acc3f5d7 | 6645 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6646 | struct sched_domain_attr *dattr_new) |
029190c5 | 6647 | { |
dfb512ec | 6648 | int i, j, n; |
d65bd5ec | 6649 | int new_topology; |
029190c5 | 6650 | |
712555ee | 6651 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6652 | |
7378547f MM |
6653 | /* always unregister in case we don't destroy any domains */ |
6654 | unregister_sched_domain_sysctl(); | |
6655 | ||
d65bd5ec HC |
6656 | /* Let architecture update cpu core mappings. */ |
6657 | new_topology = arch_update_cpu_topology(); | |
6658 | ||
dfb512ec | 6659 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6660 | |
6661 | /* Destroy deleted domains */ | |
6662 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6663 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6664 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6665 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6666 | goto match1; |
6667 | } | |
6668 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6669 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6670 | match1: |
6671 | ; | |
6672 | } | |
6673 | ||
c8d2d47a | 6674 | n = ndoms_cur; |
e761b772 | 6675 | if (doms_new == NULL) { |
c8d2d47a | 6676 | n = 0; |
acc3f5d7 | 6677 | doms_new = &fallback_doms; |
6ad4c188 | 6678 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6679 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6680 | } |
6681 | ||
029190c5 PJ |
6682 | /* Build new domains */ |
6683 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6684 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6685 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6686 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6687 | goto match2; |
6688 | } | |
6689 | /* no match - add a new doms_new */ | |
dce840a0 | 6690 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6691 | match2: |
6692 | ; | |
6693 | } | |
6694 | ||
6695 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6696 | if (doms_cur != &fallback_doms) |
6697 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6698 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6699 | doms_cur = doms_new; |
1d3504fc | 6700 | dattr_cur = dattr_new; |
029190c5 | 6701 | ndoms_cur = ndoms_new; |
7378547f MM |
6702 | |
6703 | register_sched_domain_sysctl(); | |
a1835615 | 6704 | |
712555ee | 6705 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6706 | } |
6707 | ||
d35be8ba SB |
6708 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6709 | ||
1da177e4 | 6710 | /* |
3a101d05 TH |
6711 | * Update cpusets according to cpu_active mask. If cpusets are |
6712 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6713 | * around partition_sched_domains(). | |
d35be8ba SB |
6714 | * |
6715 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6716 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6717 | */ |
0b2e918a TH |
6718 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6719 | void *hcpu) | |
e761b772 | 6720 | { |
d35be8ba SB |
6721 | switch (action) { |
6722 | case CPU_ONLINE_FROZEN: | |
6723 | case CPU_DOWN_FAILED_FROZEN: | |
6724 | ||
6725 | /* | |
6726 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6727 | * resume sequence. As long as this is not the last online | |
6728 | * operation in the resume sequence, just build a single sched | |
6729 | * domain, ignoring cpusets. | |
6730 | */ | |
6731 | num_cpus_frozen--; | |
6732 | if (likely(num_cpus_frozen)) { | |
6733 | partition_sched_domains(1, NULL, NULL); | |
6734 | break; | |
6735 | } | |
6736 | ||
6737 | /* | |
6738 | * This is the last CPU online operation. So fall through and | |
6739 | * restore the original sched domains by considering the | |
6740 | * cpuset configurations. | |
6741 | */ | |
6742 | ||
e761b772 | 6743 | case CPU_ONLINE: |
6ad4c188 | 6744 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6745 | cpuset_update_active_cpus(true); |
d35be8ba | 6746 | break; |
3a101d05 TH |
6747 | default: |
6748 | return NOTIFY_DONE; | |
6749 | } | |
d35be8ba | 6750 | return NOTIFY_OK; |
3a101d05 | 6751 | } |
e761b772 | 6752 | |
0b2e918a TH |
6753 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6754 | void *hcpu) | |
3a101d05 | 6755 | { |
d35be8ba | 6756 | switch (action) { |
3a101d05 | 6757 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6758 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6759 | break; |
6760 | case CPU_DOWN_PREPARE_FROZEN: | |
6761 | num_cpus_frozen++; | |
6762 | partition_sched_domains(1, NULL, NULL); | |
6763 | break; | |
e761b772 MK |
6764 | default: |
6765 | return NOTIFY_DONE; | |
6766 | } | |
d35be8ba | 6767 | return NOTIFY_OK; |
e761b772 | 6768 | } |
e761b772 | 6769 | |
1da177e4 LT |
6770 | void __init sched_init_smp(void) |
6771 | { | |
dcc30a35 RR |
6772 | cpumask_var_t non_isolated_cpus; |
6773 | ||
6774 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6775 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6776 | |
cb83b629 PZ |
6777 | sched_init_numa(); |
6778 | ||
6acce3ef PZ |
6779 | /* |
6780 | * There's no userspace yet to cause hotplug operations; hence all the | |
6781 | * cpu masks are stable and all blatant races in the below code cannot | |
6782 | * happen. | |
6783 | */ | |
712555ee | 6784 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6785 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6786 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6787 | if (cpumask_empty(non_isolated_cpus)) | |
6788 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6789 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6790 | |
301a5cba | 6791 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6792 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6793 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6794 | |
b328ca18 | 6795 | init_hrtick(); |
5c1e1767 NP |
6796 | |
6797 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6798 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6799 | BUG(); |
19978ca6 | 6800 | sched_init_granularity(); |
dcc30a35 | 6801 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6802 | |
0e3900e6 | 6803 | init_sched_rt_class(); |
1baca4ce | 6804 | init_sched_dl_class(); |
1da177e4 LT |
6805 | } |
6806 | #else | |
6807 | void __init sched_init_smp(void) | |
6808 | { | |
19978ca6 | 6809 | sched_init_granularity(); |
1da177e4 LT |
6810 | } |
6811 | #endif /* CONFIG_SMP */ | |
6812 | ||
cd1bb94b AB |
6813 | const_debug unsigned int sysctl_timer_migration = 1; |
6814 | ||
1da177e4 LT |
6815 | int in_sched_functions(unsigned long addr) |
6816 | { | |
1da177e4 LT |
6817 | return in_lock_functions(addr) || |
6818 | (addr >= (unsigned long)__sched_text_start | |
6819 | && addr < (unsigned long)__sched_text_end); | |
6820 | } | |
6821 | ||
029632fb | 6822 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6823 | /* |
6824 | * Default task group. | |
6825 | * Every task in system belongs to this group at bootup. | |
6826 | */ | |
029632fb | 6827 | struct task_group root_task_group; |
35cf4e50 | 6828 | LIST_HEAD(task_groups); |
052f1dc7 | 6829 | #endif |
6f505b16 | 6830 | |
e6252c3e | 6831 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6832 | |
1da177e4 LT |
6833 | void __init sched_init(void) |
6834 | { | |
dd41f596 | 6835 | int i, j; |
434d53b0 MT |
6836 | unsigned long alloc_size = 0, ptr; |
6837 | ||
6838 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6839 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6840 | #endif | |
6841 | #ifdef CONFIG_RT_GROUP_SCHED | |
6842 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6843 | #endif |
df7c8e84 | 6844 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6845 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6846 | #endif |
434d53b0 | 6847 | if (alloc_size) { |
36b7b6d4 | 6848 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6849 | |
6850 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6851 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6852 | ptr += nr_cpu_ids * sizeof(void **); |
6853 | ||
07e06b01 | 6854 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6855 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6856 | |
6d6bc0ad | 6857 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6858 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6859 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6860 | ptr += nr_cpu_ids * sizeof(void **); |
6861 | ||
07e06b01 | 6862 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6863 | ptr += nr_cpu_ids * sizeof(void **); |
6864 | ||
6d6bc0ad | 6865 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6866 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6867 | for_each_possible_cpu(i) { | |
e6252c3e | 6868 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6869 | ptr += cpumask_size(); |
6870 | } | |
6871 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6872 | } |
dd41f596 | 6873 | |
332ac17e DF |
6874 | init_rt_bandwidth(&def_rt_bandwidth, |
6875 | global_rt_period(), global_rt_runtime()); | |
6876 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6877 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6878 | |
57d885fe GH |
6879 | #ifdef CONFIG_SMP |
6880 | init_defrootdomain(); | |
6881 | #endif | |
6882 | ||
d0b27fa7 | 6883 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6884 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6885 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6886 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6887 | |
7c941438 | 6888 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6889 | list_add(&root_task_group.list, &task_groups); |
6890 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6891 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6892 | autogroup_init(&init_task); |
54c707e9 | 6893 | |
7c941438 | 6894 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6895 | |
0a945022 | 6896 | for_each_possible_cpu(i) { |
70b97a7f | 6897 | struct rq *rq; |
1da177e4 LT |
6898 | |
6899 | rq = cpu_rq(i); | |
05fa785c | 6900 | raw_spin_lock_init(&rq->lock); |
7897986b | 6901 | rq->nr_running = 0; |
dce48a84 TG |
6902 | rq->calc_load_active = 0; |
6903 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6904 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6905 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6906 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6907 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6908 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6909 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6910 | /* |
07e06b01 | 6911 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6912 | * |
6913 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6914 | * gets 100% of the cpu resources in the system. This overall | |
6915 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6916 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6917 | * based on each entity's (task or task-group's) weight |
6918 | * (se->load.weight). | |
6919 | * | |
07e06b01 | 6920 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6921 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6922 | * then A0's share of the cpu resource is: | |
6923 | * | |
0d905bca | 6924 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6925 | * |
07e06b01 YZ |
6926 | * We achieve this by letting root_task_group's tasks sit |
6927 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6928 | */ |
ab84d31e | 6929 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6930 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6931 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6932 | ||
6933 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6934 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6935 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6936 | #endif |
1da177e4 | 6937 | |
dd41f596 IM |
6938 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6939 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6940 | |
6941 | rq->last_load_update_tick = jiffies; | |
6942 | ||
1da177e4 | 6943 | #ifdef CONFIG_SMP |
41c7ce9a | 6944 | rq->sd = NULL; |
57d885fe | 6945 | rq->rd = NULL; |
1399fa78 | 6946 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6947 | rq->post_schedule = 0; |
1da177e4 | 6948 | rq->active_balance = 0; |
dd41f596 | 6949 | rq->next_balance = jiffies; |
1da177e4 | 6950 | rq->push_cpu = 0; |
0a2966b4 | 6951 | rq->cpu = i; |
1f11eb6a | 6952 | rq->online = 0; |
eae0c9df MG |
6953 | rq->idle_stamp = 0; |
6954 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6955 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6956 | |
6957 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6958 | ||
dc938520 | 6959 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6960 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6961 | rq->nohz_flags = 0; |
83cd4fe2 | 6962 | #endif |
265f22a9 FW |
6963 | #ifdef CONFIG_NO_HZ_FULL |
6964 | rq->last_sched_tick = 0; | |
6965 | #endif | |
1da177e4 | 6966 | #endif |
8f4d37ec | 6967 | init_rq_hrtick(rq); |
1da177e4 | 6968 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6969 | } |
6970 | ||
2dd73a4f | 6971 | set_load_weight(&init_task); |
b50f60ce | 6972 | |
e107be36 AK |
6973 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6974 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6975 | #endif | |
6976 | ||
1da177e4 LT |
6977 | /* |
6978 | * The boot idle thread does lazy MMU switching as well: | |
6979 | */ | |
6980 | atomic_inc(&init_mm.mm_count); | |
6981 | enter_lazy_tlb(&init_mm, current); | |
6982 | ||
6983 | /* | |
6984 | * Make us the idle thread. Technically, schedule() should not be | |
6985 | * called from this thread, however somewhere below it might be, | |
6986 | * but because we are the idle thread, we just pick up running again | |
6987 | * when this runqueue becomes "idle". | |
6988 | */ | |
6989 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6990 | |
6991 | calc_load_update = jiffies + LOAD_FREQ; | |
6992 | ||
dd41f596 IM |
6993 | /* |
6994 | * During early bootup we pretend to be a normal task: | |
6995 | */ | |
6996 | current->sched_class = &fair_sched_class; | |
6892b75e | 6997 | |
bf4d83f6 | 6998 | #ifdef CONFIG_SMP |
4cb98839 | 6999 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7000 | /* May be allocated at isolcpus cmdline parse time */ |
7001 | if (cpu_isolated_map == NULL) | |
7002 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7003 | idle_thread_set_boot_cpu(); |
a803f026 | 7004 | set_cpu_rq_start_time(); |
029632fb PZ |
7005 | #endif |
7006 | init_sched_fair_class(); | |
6a7b3dc3 | 7007 | |
6892b75e | 7008 | scheduler_running = 1; |
1da177e4 LT |
7009 | } |
7010 | ||
d902db1e | 7011 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7012 | static inline int preempt_count_equals(int preempt_offset) |
7013 | { | |
234da7bc | 7014 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7015 | |
4ba8216c | 7016 | return (nested == preempt_offset); |
e4aafea2 FW |
7017 | } |
7018 | ||
d894837f | 7019 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7020 | { |
1da177e4 LT |
7021 | static unsigned long prev_jiffy; /* ratelimiting */ |
7022 | ||
b3fbab05 | 7023 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7024 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7025 | !is_idle_task(current)) || | |
e4aafea2 | 7026 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7027 | return; |
7028 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7029 | return; | |
7030 | prev_jiffy = jiffies; | |
7031 | ||
3df0fc5b PZ |
7032 | printk(KERN_ERR |
7033 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7034 | file, line); | |
7035 | printk(KERN_ERR | |
7036 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7037 | in_atomic(), irqs_disabled(), | |
7038 | current->pid, current->comm); | |
aef745fc IM |
7039 | |
7040 | debug_show_held_locks(current); | |
7041 | if (irqs_disabled()) | |
7042 | print_irqtrace_events(current); | |
8f47b187 TG |
7043 | #ifdef CONFIG_DEBUG_PREEMPT |
7044 | if (!preempt_count_equals(preempt_offset)) { | |
7045 | pr_err("Preemption disabled at:"); | |
7046 | print_ip_sym(current->preempt_disable_ip); | |
7047 | pr_cont("\n"); | |
7048 | } | |
7049 | #endif | |
aef745fc | 7050 | dump_stack(); |
1da177e4 LT |
7051 | } |
7052 | EXPORT_SYMBOL(__might_sleep); | |
7053 | #endif | |
7054 | ||
7055 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7056 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7057 | { | |
da7a735e | 7058 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7059 | struct sched_attr attr = { |
7060 | .sched_policy = SCHED_NORMAL, | |
7061 | }; | |
da7a735e | 7062 | int old_prio = p->prio; |
3a5e4dc1 | 7063 | int on_rq; |
3e51f33f | 7064 | |
fd2f4419 | 7065 | on_rq = p->on_rq; |
3a5e4dc1 | 7066 | if (on_rq) |
4ca9b72b | 7067 | dequeue_task(rq, p, 0); |
d50dde5a | 7068 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7069 | if (on_rq) { |
4ca9b72b | 7070 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7071 | resched_task(rq->curr); |
7072 | } | |
da7a735e PZ |
7073 | |
7074 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7075 | } |
7076 | ||
1da177e4 LT |
7077 | void normalize_rt_tasks(void) |
7078 | { | |
a0f98a1c | 7079 | struct task_struct *g, *p; |
1da177e4 | 7080 | unsigned long flags; |
70b97a7f | 7081 | struct rq *rq; |
1da177e4 | 7082 | |
4cf5d77a | 7083 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7084 | do_each_thread(g, p) { |
178be793 IM |
7085 | /* |
7086 | * Only normalize user tasks: | |
7087 | */ | |
7088 | if (!p->mm) | |
7089 | continue; | |
7090 | ||
6cfb0d5d | 7091 | p->se.exec_start = 0; |
6cfb0d5d | 7092 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7093 | p->se.statistics.wait_start = 0; |
7094 | p->se.statistics.sleep_start = 0; | |
7095 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7096 | #endif |
dd41f596 | 7097 | |
aab03e05 | 7098 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7099 | /* |
7100 | * Renice negative nice level userspace | |
7101 | * tasks back to 0: | |
7102 | */ | |
d0ea0268 | 7103 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7104 | set_user_nice(p, 0); |
1da177e4 | 7105 | continue; |
dd41f596 | 7106 | } |
1da177e4 | 7107 | |
1d615482 | 7108 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7109 | rq = __task_rq_lock(p); |
1da177e4 | 7110 | |
178be793 | 7111 | normalize_task(rq, p); |
3a5e4dc1 | 7112 | |
b29739f9 | 7113 | __task_rq_unlock(rq); |
1d615482 | 7114 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7115 | } while_each_thread(g, p); |
7116 | ||
4cf5d77a | 7117 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7118 | } |
7119 | ||
7120 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7121 | |
67fc4e0c | 7122 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7123 | /* |
67fc4e0c | 7124 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7125 | * |
7126 | * They can only be called when the whole system has been | |
7127 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7128 | * activity can take place. Using them for anything else would | |
7129 | * be a serious bug, and as a result, they aren't even visible | |
7130 | * under any other configuration. | |
7131 | */ | |
7132 | ||
7133 | /** | |
7134 | * curr_task - return the current task for a given cpu. | |
7135 | * @cpu: the processor in question. | |
7136 | * | |
7137 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7138 | * |
7139 | * Return: The current task for @cpu. | |
1df5c10a | 7140 | */ |
36c8b586 | 7141 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7142 | { |
7143 | return cpu_curr(cpu); | |
7144 | } | |
7145 | ||
67fc4e0c JW |
7146 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7147 | ||
7148 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7149 | /** |
7150 | * set_curr_task - set the current task for a given cpu. | |
7151 | * @cpu: the processor in question. | |
7152 | * @p: the task pointer to set. | |
7153 | * | |
7154 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7155 | * are serviced on a separate stack. It allows the architecture to switch the |
7156 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7157 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7158 | * and caller must save the original value of the current task (see | |
7159 | * curr_task() above) and restore that value before reenabling interrupts and | |
7160 | * re-starting the system. | |
7161 | * | |
7162 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7163 | */ | |
36c8b586 | 7164 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7165 | { |
7166 | cpu_curr(cpu) = p; | |
7167 | } | |
7168 | ||
7169 | #endif | |
29f59db3 | 7170 | |
7c941438 | 7171 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7172 | /* task_group_lock serializes the addition/removal of task groups */ |
7173 | static DEFINE_SPINLOCK(task_group_lock); | |
7174 | ||
bccbe08a PZ |
7175 | static void free_sched_group(struct task_group *tg) |
7176 | { | |
7177 | free_fair_sched_group(tg); | |
7178 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7179 | autogroup_free(tg); |
bccbe08a PZ |
7180 | kfree(tg); |
7181 | } | |
7182 | ||
7183 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7184 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7185 | { |
7186 | struct task_group *tg; | |
bccbe08a PZ |
7187 | |
7188 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7189 | if (!tg) | |
7190 | return ERR_PTR(-ENOMEM); | |
7191 | ||
ec7dc8ac | 7192 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7193 | goto err; |
7194 | ||
ec7dc8ac | 7195 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7196 | goto err; |
7197 | ||
ace783b9 LZ |
7198 | return tg; |
7199 | ||
7200 | err: | |
7201 | free_sched_group(tg); | |
7202 | return ERR_PTR(-ENOMEM); | |
7203 | } | |
7204 | ||
7205 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7206 | { | |
7207 | unsigned long flags; | |
7208 | ||
8ed36996 | 7209 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7210 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7211 | |
7212 | WARN_ON(!parent); /* root should already exist */ | |
7213 | ||
7214 | tg->parent = parent; | |
f473aa5e | 7215 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7216 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7217 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7218 | } |
7219 | ||
9b5b7751 | 7220 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7221 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7222 | { |
29f59db3 | 7223 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7224 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7225 | } |
7226 | ||
9b5b7751 | 7227 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7228 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7229 | { |
7230 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7231 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7232 | } | |
7233 | ||
7234 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7235 | { |
8ed36996 | 7236 | unsigned long flags; |
9b5b7751 | 7237 | int i; |
29f59db3 | 7238 | |
3d4b47b4 PZ |
7239 | /* end participation in shares distribution */ |
7240 | for_each_possible_cpu(i) | |
bccbe08a | 7241 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7242 | |
7243 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7244 | list_del_rcu(&tg->list); |
f473aa5e | 7245 | list_del_rcu(&tg->siblings); |
8ed36996 | 7246 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7247 | } |
7248 | ||
9b5b7751 | 7249 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7250 | * The caller of this function should have put the task in its new group |
7251 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7252 | * reflect its new group. | |
9b5b7751 SV |
7253 | */ |
7254 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7255 | { |
8323f26c | 7256 | struct task_group *tg; |
29f59db3 SV |
7257 | int on_rq, running; |
7258 | unsigned long flags; | |
7259 | struct rq *rq; | |
7260 | ||
7261 | rq = task_rq_lock(tsk, &flags); | |
7262 | ||
051a1d1a | 7263 | running = task_current(rq, tsk); |
fd2f4419 | 7264 | on_rq = tsk->on_rq; |
29f59db3 | 7265 | |
0e1f3483 | 7266 | if (on_rq) |
29f59db3 | 7267 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7268 | if (unlikely(running)) |
7269 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7270 | |
073219e9 | 7271 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7272 | lockdep_is_held(&tsk->sighand->siglock)), |
7273 | struct task_group, css); | |
7274 | tg = autogroup_task_group(tsk, tg); | |
7275 | tsk->sched_task_group = tg; | |
7276 | ||
810b3817 | 7277 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7278 | if (tsk->sched_class->task_move_group) |
7279 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7280 | else | |
810b3817 | 7281 | #endif |
b2b5ce02 | 7282 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7283 | |
0e1f3483 HS |
7284 | if (unlikely(running)) |
7285 | tsk->sched_class->set_curr_task(rq); | |
7286 | if (on_rq) | |
371fd7e7 | 7287 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7288 | |
0122ec5b | 7289 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7290 | } |
7c941438 | 7291 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7292 | |
a790de99 PT |
7293 | #ifdef CONFIG_RT_GROUP_SCHED |
7294 | /* | |
7295 | * Ensure that the real time constraints are schedulable. | |
7296 | */ | |
7297 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7298 | |
9a7e0b18 PZ |
7299 | /* Must be called with tasklist_lock held */ |
7300 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7301 | { |
9a7e0b18 | 7302 | struct task_struct *g, *p; |
b40b2e8e | 7303 | |
9a7e0b18 | 7304 | do_each_thread(g, p) { |
029632fb | 7305 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7306 | return 1; |
7307 | } while_each_thread(g, p); | |
b40b2e8e | 7308 | |
9a7e0b18 PZ |
7309 | return 0; |
7310 | } | |
b40b2e8e | 7311 | |
9a7e0b18 PZ |
7312 | struct rt_schedulable_data { |
7313 | struct task_group *tg; | |
7314 | u64 rt_period; | |
7315 | u64 rt_runtime; | |
7316 | }; | |
b40b2e8e | 7317 | |
a790de99 | 7318 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7319 | { |
7320 | struct rt_schedulable_data *d = data; | |
7321 | struct task_group *child; | |
7322 | unsigned long total, sum = 0; | |
7323 | u64 period, runtime; | |
b40b2e8e | 7324 | |
9a7e0b18 PZ |
7325 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7326 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7327 | |
9a7e0b18 PZ |
7328 | if (tg == d->tg) { |
7329 | period = d->rt_period; | |
7330 | runtime = d->rt_runtime; | |
b40b2e8e | 7331 | } |
b40b2e8e | 7332 | |
4653f803 PZ |
7333 | /* |
7334 | * Cannot have more runtime than the period. | |
7335 | */ | |
7336 | if (runtime > period && runtime != RUNTIME_INF) | |
7337 | return -EINVAL; | |
6f505b16 | 7338 | |
4653f803 PZ |
7339 | /* |
7340 | * Ensure we don't starve existing RT tasks. | |
7341 | */ | |
9a7e0b18 PZ |
7342 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7343 | return -EBUSY; | |
6f505b16 | 7344 | |
9a7e0b18 | 7345 | total = to_ratio(period, runtime); |
6f505b16 | 7346 | |
4653f803 PZ |
7347 | /* |
7348 | * Nobody can have more than the global setting allows. | |
7349 | */ | |
7350 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7351 | return -EINVAL; | |
6f505b16 | 7352 | |
4653f803 PZ |
7353 | /* |
7354 | * The sum of our children's runtime should not exceed our own. | |
7355 | */ | |
9a7e0b18 PZ |
7356 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7357 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7358 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7359 | |
9a7e0b18 PZ |
7360 | if (child == d->tg) { |
7361 | period = d->rt_period; | |
7362 | runtime = d->rt_runtime; | |
7363 | } | |
6f505b16 | 7364 | |
9a7e0b18 | 7365 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7366 | } |
6f505b16 | 7367 | |
9a7e0b18 PZ |
7368 | if (sum > total) |
7369 | return -EINVAL; | |
7370 | ||
7371 | return 0; | |
6f505b16 PZ |
7372 | } |
7373 | ||
9a7e0b18 | 7374 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7375 | { |
8277434e PT |
7376 | int ret; |
7377 | ||
9a7e0b18 PZ |
7378 | struct rt_schedulable_data data = { |
7379 | .tg = tg, | |
7380 | .rt_period = period, | |
7381 | .rt_runtime = runtime, | |
7382 | }; | |
7383 | ||
8277434e PT |
7384 | rcu_read_lock(); |
7385 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7386 | rcu_read_unlock(); | |
7387 | ||
7388 | return ret; | |
521f1a24 DG |
7389 | } |
7390 | ||
ab84d31e | 7391 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7392 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7393 | { |
ac086bc2 | 7394 | int i, err = 0; |
9f0c1e56 | 7395 | |
9f0c1e56 | 7396 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7397 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7398 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7399 | if (err) | |
9f0c1e56 | 7400 | goto unlock; |
ac086bc2 | 7401 | |
0986b11b | 7402 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7403 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7404 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7405 | |
7406 | for_each_possible_cpu(i) { | |
7407 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7408 | ||
0986b11b | 7409 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7410 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7411 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7412 | } |
0986b11b | 7413 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7414 | unlock: |
521f1a24 | 7415 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7416 | mutex_unlock(&rt_constraints_mutex); |
7417 | ||
7418 | return err; | |
6f505b16 PZ |
7419 | } |
7420 | ||
25cc7da7 | 7421 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7422 | { |
7423 | u64 rt_runtime, rt_period; | |
7424 | ||
7425 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7426 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7427 | if (rt_runtime_us < 0) | |
7428 | rt_runtime = RUNTIME_INF; | |
7429 | ||
ab84d31e | 7430 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7431 | } |
7432 | ||
25cc7da7 | 7433 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7434 | { |
7435 | u64 rt_runtime_us; | |
7436 | ||
d0b27fa7 | 7437 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7438 | return -1; |
7439 | ||
d0b27fa7 | 7440 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7441 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7442 | return rt_runtime_us; | |
7443 | } | |
d0b27fa7 | 7444 | |
25cc7da7 | 7445 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7446 | { |
7447 | u64 rt_runtime, rt_period; | |
7448 | ||
7449 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7450 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7451 | ||
619b0488 R |
7452 | if (rt_period == 0) |
7453 | return -EINVAL; | |
7454 | ||
ab84d31e | 7455 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7456 | } |
7457 | ||
25cc7da7 | 7458 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7459 | { |
7460 | u64 rt_period_us; | |
7461 | ||
7462 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7463 | do_div(rt_period_us, NSEC_PER_USEC); | |
7464 | return rt_period_us; | |
7465 | } | |
332ac17e | 7466 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7467 | |
332ac17e | 7468 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7469 | static int sched_rt_global_constraints(void) |
7470 | { | |
7471 | int ret = 0; | |
7472 | ||
7473 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7474 | read_lock(&tasklist_lock); |
4653f803 | 7475 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7476 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7477 | mutex_unlock(&rt_constraints_mutex); |
7478 | ||
7479 | return ret; | |
7480 | } | |
54e99124 | 7481 | |
25cc7da7 | 7482 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7483 | { |
7484 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7485 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7486 | return 0; | |
7487 | ||
7488 | return 1; | |
7489 | } | |
7490 | ||
6d6bc0ad | 7491 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7492 | static int sched_rt_global_constraints(void) |
7493 | { | |
ac086bc2 | 7494 | unsigned long flags; |
332ac17e | 7495 | int i, ret = 0; |
ec5d4989 | 7496 | |
0986b11b | 7497 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7498 | for_each_possible_cpu(i) { |
7499 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7500 | ||
0986b11b | 7501 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7502 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7503 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7504 | } |
0986b11b | 7505 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7506 | |
332ac17e | 7507 | return ret; |
d0b27fa7 | 7508 | } |
6d6bc0ad | 7509 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7510 | |
332ac17e DF |
7511 | static int sched_dl_global_constraints(void) |
7512 | { | |
1724813d PZ |
7513 | u64 runtime = global_rt_runtime(); |
7514 | u64 period = global_rt_period(); | |
332ac17e | 7515 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7516 | int cpu, ret = 0; |
49516342 | 7517 | unsigned long flags; |
332ac17e DF |
7518 | |
7519 | /* | |
7520 | * Here we want to check the bandwidth not being set to some | |
7521 | * value smaller than the currently allocated bandwidth in | |
7522 | * any of the root_domains. | |
7523 | * | |
7524 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7525 | * cycling on root_domains... Discussion on different/better | |
7526 | * solutions is welcome! | |
7527 | */ | |
1724813d PZ |
7528 | for_each_possible_cpu(cpu) { |
7529 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7530 | |
49516342 | 7531 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7532 | if (new_bw < dl_b->total_bw) |
7533 | ret = -EBUSY; | |
49516342 | 7534 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7535 | |
7536 | if (ret) | |
7537 | break; | |
332ac17e DF |
7538 | } |
7539 | ||
1724813d | 7540 | return ret; |
332ac17e DF |
7541 | } |
7542 | ||
1724813d | 7543 | static void sched_dl_do_global(void) |
ce0dbbbb | 7544 | { |
1724813d PZ |
7545 | u64 new_bw = -1; |
7546 | int cpu; | |
49516342 | 7547 | unsigned long flags; |
ce0dbbbb | 7548 | |
1724813d PZ |
7549 | def_dl_bandwidth.dl_period = global_rt_period(); |
7550 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7551 | ||
7552 | if (global_rt_runtime() != RUNTIME_INF) | |
7553 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7554 | ||
7555 | /* | |
7556 | * FIXME: As above... | |
7557 | */ | |
7558 | for_each_possible_cpu(cpu) { | |
7559 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7560 | ||
49516342 | 7561 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7562 | dl_b->bw = new_bw; |
49516342 | 7563 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7564 | } |
1724813d PZ |
7565 | } |
7566 | ||
7567 | static int sched_rt_global_validate(void) | |
7568 | { | |
7569 | if (sysctl_sched_rt_period <= 0) | |
7570 | return -EINVAL; | |
7571 | ||
e9e7cb38 JL |
7572 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7573 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7574 | return -EINVAL; |
7575 | ||
7576 | return 0; | |
7577 | } | |
7578 | ||
7579 | static void sched_rt_do_global(void) | |
7580 | { | |
7581 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7582 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7583 | } |
7584 | ||
d0b27fa7 | 7585 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7586 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7587 | loff_t *ppos) |
7588 | { | |
d0b27fa7 PZ |
7589 | int old_period, old_runtime; |
7590 | static DEFINE_MUTEX(mutex); | |
1724813d | 7591 | int ret; |
d0b27fa7 PZ |
7592 | |
7593 | mutex_lock(&mutex); | |
7594 | old_period = sysctl_sched_rt_period; | |
7595 | old_runtime = sysctl_sched_rt_runtime; | |
7596 | ||
8d65af78 | 7597 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7598 | |
7599 | if (!ret && write) { | |
1724813d PZ |
7600 | ret = sched_rt_global_validate(); |
7601 | if (ret) | |
7602 | goto undo; | |
7603 | ||
d0b27fa7 | 7604 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7605 | if (ret) |
7606 | goto undo; | |
7607 | ||
7608 | ret = sched_dl_global_constraints(); | |
7609 | if (ret) | |
7610 | goto undo; | |
7611 | ||
7612 | sched_rt_do_global(); | |
7613 | sched_dl_do_global(); | |
7614 | } | |
7615 | if (0) { | |
7616 | undo: | |
7617 | sysctl_sched_rt_period = old_period; | |
7618 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7619 | } |
7620 | mutex_unlock(&mutex); | |
7621 | ||
7622 | return ret; | |
7623 | } | |
68318b8e | 7624 | |
1724813d | 7625 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7626 | void __user *buffer, size_t *lenp, |
7627 | loff_t *ppos) | |
7628 | { | |
7629 | int ret; | |
332ac17e | 7630 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7631 | |
7632 | mutex_lock(&mutex); | |
332ac17e | 7633 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7634 | /* make sure that internally we keep jiffies */ |
7635 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7636 | if (!ret && write) { |
1724813d PZ |
7637 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7638 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7639 | } |
7640 | mutex_unlock(&mutex); | |
332ac17e DF |
7641 | return ret; |
7642 | } | |
7643 | ||
052f1dc7 | 7644 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7645 | |
a7c6d554 | 7646 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7647 | { |
a7c6d554 | 7648 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7649 | } |
7650 | ||
eb95419b TH |
7651 | static struct cgroup_subsys_state * |
7652 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7653 | { |
eb95419b TH |
7654 | struct task_group *parent = css_tg(parent_css); |
7655 | struct task_group *tg; | |
68318b8e | 7656 | |
eb95419b | 7657 | if (!parent) { |
68318b8e | 7658 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7659 | return &root_task_group.css; |
68318b8e SV |
7660 | } |
7661 | ||
ec7dc8ac | 7662 | tg = sched_create_group(parent); |
68318b8e SV |
7663 | if (IS_ERR(tg)) |
7664 | return ERR_PTR(-ENOMEM); | |
7665 | ||
68318b8e SV |
7666 | return &tg->css; |
7667 | } | |
7668 | ||
eb95419b | 7669 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7670 | { |
eb95419b TH |
7671 | struct task_group *tg = css_tg(css); |
7672 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7673 | |
63876986 TH |
7674 | if (parent) |
7675 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7676 | return 0; |
7677 | } | |
7678 | ||
eb95419b | 7679 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7680 | { |
eb95419b | 7681 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7682 | |
7683 | sched_destroy_group(tg); | |
7684 | } | |
7685 | ||
eb95419b | 7686 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7687 | { |
eb95419b | 7688 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7689 | |
7690 | sched_offline_group(tg); | |
7691 | } | |
7692 | ||
eb95419b | 7693 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7694 | struct cgroup_taskset *tset) |
68318b8e | 7695 | { |
bb9d97b6 TH |
7696 | struct task_struct *task; |
7697 | ||
924f0d9a | 7698 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7699 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7700 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7701 | return -EINVAL; |
b68aa230 | 7702 | #else |
bb9d97b6 TH |
7703 | /* We don't support RT-tasks being in separate groups */ |
7704 | if (task->sched_class != &fair_sched_class) | |
7705 | return -EINVAL; | |
b68aa230 | 7706 | #endif |
bb9d97b6 | 7707 | } |
be367d09 BB |
7708 | return 0; |
7709 | } | |
68318b8e | 7710 | |
eb95419b | 7711 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7712 | struct cgroup_taskset *tset) |
68318b8e | 7713 | { |
bb9d97b6 TH |
7714 | struct task_struct *task; |
7715 | ||
924f0d9a | 7716 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7717 | sched_move_task(task); |
68318b8e SV |
7718 | } |
7719 | ||
eb95419b TH |
7720 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7721 | struct cgroup_subsys_state *old_css, | |
7722 | struct task_struct *task) | |
068c5cc5 PZ |
7723 | { |
7724 | /* | |
7725 | * cgroup_exit() is called in the copy_process() failure path. | |
7726 | * Ignore this case since the task hasn't ran yet, this avoids | |
7727 | * trying to poke a half freed task state from generic code. | |
7728 | */ | |
7729 | if (!(task->flags & PF_EXITING)) | |
7730 | return; | |
7731 | ||
7732 | sched_move_task(task); | |
7733 | } | |
7734 | ||
052f1dc7 | 7735 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7736 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7737 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7738 | { |
182446d0 | 7739 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7740 | } |
7741 | ||
182446d0 TH |
7742 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7743 | struct cftype *cft) | |
68318b8e | 7744 | { |
182446d0 | 7745 | struct task_group *tg = css_tg(css); |
68318b8e | 7746 | |
c8b28116 | 7747 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7748 | } |
ab84d31e PT |
7749 | |
7750 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7751 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7752 | ||
ab84d31e PT |
7753 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7754 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7755 | ||
a790de99 PT |
7756 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7757 | ||
ab84d31e PT |
7758 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7759 | { | |
56f570e5 | 7760 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7761 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7762 | |
7763 | if (tg == &root_task_group) | |
7764 | return -EINVAL; | |
7765 | ||
7766 | /* | |
7767 | * Ensure we have at some amount of bandwidth every period. This is | |
7768 | * to prevent reaching a state of large arrears when throttled via | |
7769 | * entity_tick() resulting in prolonged exit starvation. | |
7770 | */ | |
7771 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7772 | return -EINVAL; | |
7773 | ||
7774 | /* | |
7775 | * Likewise, bound things on the otherside by preventing insane quota | |
7776 | * periods. This also allows us to normalize in computing quota | |
7777 | * feasibility. | |
7778 | */ | |
7779 | if (period > max_cfs_quota_period) | |
7780 | return -EINVAL; | |
7781 | ||
a790de99 PT |
7782 | mutex_lock(&cfs_constraints_mutex); |
7783 | ret = __cfs_schedulable(tg, period, quota); | |
7784 | if (ret) | |
7785 | goto out_unlock; | |
7786 | ||
58088ad0 | 7787 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7788 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7789 | /* |
7790 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7791 | * before making related changes, and on->off must occur afterwards | |
7792 | */ | |
7793 | if (runtime_enabled && !runtime_was_enabled) | |
7794 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7795 | raw_spin_lock_irq(&cfs_b->lock); |
7796 | cfs_b->period = ns_to_ktime(period); | |
7797 | cfs_b->quota = quota; | |
58088ad0 | 7798 | |
a9cf55b2 | 7799 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7800 | /* restart the period timer (if active) to handle new period expiry */ |
7801 | if (runtime_enabled && cfs_b->timer_active) { | |
7802 | /* force a reprogram */ | |
7803 | cfs_b->timer_active = 0; | |
7804 | __start_cfs_bandwidth(cfs_b); | |
7805 | } | |
ab84d31e PT |
7806 | raw_spin_unlock_irq(&cfs_b->lock); |
7807 | ||
7808 | for_each_possible_cpu(i) { | |
7809 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7810 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7811 | |
7812 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7813 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7814 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7815 | |
029632fb | 7816 | if (cfs_rq->throttled) |
671fd9da | 7817 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7818 | raw_spin_unlock_irq(&rq->lock); |
7819 | } | |
1ee14e6c BS |
7820 | if (runtime_was_enabled && !runtime_enabled) |
7821 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7822 | out_unlock: |
7823 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7824 | |
a790de99 | 7825 | return ret; |
ab84d31e PT |
7826 | } |
7827 | ||
7828 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7829 | { | |
7830 | u64 quota, period; | |
7831 | ||
029632fb | 7832 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7833 | if (cfs_quota_us < 0) |
7834 | quota = RUNTIME_INF; | |
7835 | else | |
7836 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7837 | ||
7838 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7839 | } | |
7840 | ||
7841 | long tg_get_cfs_quota(struct task_group *tg) | |
7842 | { | |
7843 | u64 quota_us; | |
7844 | ||
029632fb | 7845 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7846 | return -1; |
7847 | ||
029632fb | 7848 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7849 | do_div(quota_us, NSEC_PER_USEC); |
7850 | ||
7851 | return quota_us; | |
7852 | } | |
7853 | ||
7854 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7855 | { | |
7856 | u64 quota, period; | |
7857 | ||
7858 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7859 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7860 | |
ab84d31e PT |
7861 | return tg_set_cfs_bandwidth(tg, period, quota); |
7862 | } | |
7863 | ||
7864 | long tg_get_cfs_period(struct task_group *tg) | |
7865 | { | |
7866 | u64 cfs_period_us; | |
7867 | ||
029632fb | 7868 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7869 | do_div(cfs_period_us, NSEC_PER_USEC); |
7870 | ||
7871 | return cfs_period_us; | |
7872 | } | |
7873 | ||
182446d0 TH |
7874 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7875 | struct cftype *cft) | |
ab84d31e | 7876 | { |
182446d0 | 7877 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7878 | } |
7879 | ||
182446d0 TH |
7880 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7881 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7882 | { |
182446d0 | 7883 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7884 | } |
7885 | ||
182446d0 TH |
7886 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7887 | struct cftype *cft) | |
ab84d31e | 7888 | { |
182446d0 | 7889 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7890 | } |
7891 | ||
182446d0 TH |
7892 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7893 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7894 | { |
182446d0 | 7895 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7896 | } |
7897 | ||
a790de99 PT |
7898 | struct cfs_schedulable_data { |
7899 | struct task_group *tg; | |
7900 | u64 period, quota; | |
7901 | }; | |
7902 | ||
7903 | /* | |
7904 | * normalize group quota/period to be quota/max_period | |
7905 | * note: units are usecs | |
7906 | */ | |
7907 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7908 | struct cfs_schedulable_data *d) | |
7909 | { | |
7910 | u64 quota, period; | |
7911 | ||
7912 | if (tg == d->tg) { | |
7913 | period = d->period; | |
7914 | quota = d->quota; | |
7915 | } else { | |
7916 | period = tg_get_cfs_period(tg); | |
7917 | quota = tg_get_cfs_quota(tg); | |
7918 | } | |
7919 | ||
7920 | /* note: these should typically be equivalent */ | |
7921 | if (quota == RUNTIME_INF || quota == -1) | |
7922 | return RUNTIME_INF; | |
7923 | ||
7924 | return to_ratio(period, quota); | |
7925 | } | |
7926 | ||
7927 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7928 | { | |
7929 | struct cfs_schedulable_data *d = data; | |
029632fb | 7930 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7931 | s64 quota = 0, parent_quota = -1; |
7932 | ||
7933 | if (!tg->parent) { | |
7934 | quota = RUNTIME_INF; | |
7935 | } else { | |
029632fb | 7936 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7937 | |
7938 | quota = normalize_cfs_quota(tg, d); | |
7939 | parent_quota = parent_b->hierarchal_quota; | |
7940 | ||
7941 | /* | |
7942 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7943 | * limit is set | |
7944 | */ | |
7945 | if (quota == RUNTIME_INF) | |
7946 | quota = parent_quota; | |
7947 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7948 | return -EINVAL; | |
7949 | } | |
7950 | cfs_b->hierarchal_quota = quota; | |
7951 | ||
7952 | return 0; | |
7953 | } | |
7954 | ||
7955 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7956 | { | |
8277434e | 7957 | int ret; |
a790de99 PT |
7958 | struct cfs_schedulable_data data = { |
7959 | .tg = tg, | |
7960 | .period = period, | |
7961 | .quota = quota, | |
7962 | }; | |
7963 | ||
7964 | if (quota != RUNTIME_INF) { | |
7965 | do_div(data.period, NSEC_PER_USEC); | |
7966 | do_div(data.quota, NSEC_PER_USEC); | |
7967 | } | |
7968 | ||
8277434e PT |
7969 | rcu_read_lock(); |
7970 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7971 | rcu_read_unlock(); | |
7972 | ||
7973 | return ret; | |
a790de99 | 7974 | } |
e8da1b18 | 7975 | |
2da8ca82 | 7976 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7977 | { |
2da8ca82 | 7978 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7979 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7980 | |
44ffc75b TH |
7981 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7982 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7983 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7984 | |
7985 | return 0; | |
7986 | } | |
ab84d31e | 7987 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7988 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7989 | |
052f1dc7 | 7990 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7991 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7992 | struct cftype *cft, s64 val) | |
6f505b16 | 7993 | { |
182446d0 | 7994 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7995 | } |
7996 | ||
182446d0 TH |
7997 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7998 | struct cftype *cft) | |
6f505b16 | 7999 | { |
182446d0 | 8000 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8001 | } |
d0b27fa7 | 8002 | |
182446d0 TH |
8003 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8004 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8005 | { |
182446d0 | 8006 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8007 | } |
8008 | ||
182446d0 TH |
8009 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8010 | struct cftype *cft) | |
d0b27fa7 | 8011 | { |
182446d0 | 8012 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8013 | } |
6d6bc0ad | 8014 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8015 | |
fe5c7cc2 | 8016 | static struct cftype cpu_files[] = { |
052f1dc7 | 8017 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8018 | { |
8019 | .name = "shares", | |
f4c753b7 PM |
8020 | .read_u64 = cpu_shares_read_u64, |
8021 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8022 | }, |
052f1dc7 | 8023 | #endif |
ab84d31e PT |
8024 | #ifdef CONFIG_CFS_BANDWIDTH |
8025 | { | |
8026 | .name = "cfs_quota_us", | |
8027 | .read_s64 = cpu_cfs_quota_read_s64, | |
8028 | .write_s64 = cpu_cfs_quota_write_s64, | |
8029 | }, | |
8030 | { | |
8031 | .name = "cfs_period_us", | |
8032 | .read_u64 = cpu_cfs_period_read_u64, | |
8033 | .write_u64 = cpu_cfs_period_write_u64, | |
8034 | }, | |
e8da1b18 NR |
8035 | { |
8036 | .name = "stat", | |
2da8ca82 | 8037 | .seq_show = cpu_stats_show, |
e8da1b18 | 8038 | }, |
ab84d31e | 8039 | #endif |
052f1dc7 | 8040 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8041 | { |
9f0c1e56 | 8042 | .name = "rt_runtime_us", |
06ecb27c PM |
8043 | .read_s64 = cpu_rt_runtime_read, |
8044 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8045 | }, |
d0b27fa7 PZ |
8046 | { |
8047 | .name = "rt_period_us", | |
f4c753b7 PM |
8048 | .read_u64 = cpu_rt_period_read_uint, |
8049 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8050 | }, |
052f1dc7 | 8051 | #endif |
4baf6e33 | 8052 | { } /* terminate */ |
68318b8e SV |
8053 | }; |
8054 | ||
073219e9 | 8055 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8056 | .css_alloc = cpu_cgroup_css_alloc, |
8057 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8058 | .css_online = cpu_cgroup_css_online, |
8059 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8060 | .can_attach = cpu_cgroup_can_attach, |
8061 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8062 | .exit = cpu_cgroup_exit, |
4baf6e33 | 8063 | .base_cftypes = cpu_files, |
68318b8e SV |
8064 | .early_init = 1, |
8065 | }; | |
8066 | ||
052f1dc7 | 8067 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8068 | |
b637a328 PM |
8069 | void dump_cpu_task(int cpu) |
8070 | { | |
8071 | pr_info("Task dump for CPU %d:\n", cpu); | |
8072 | sched_show_task(cpu_curr(cpu)); | |
8073 | } |