]>
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 | 141 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
4036ac15 MG |
142 | if (delta < 0) |
143 | return; | |
fe44d621 PZ |
144 | rq->clock += delta; |
145 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
146 | } |
147 | ||
bf5c91ba IM |
148 | /* |
149 | * Debugging: various feature bits | |
150 | */ | |
f00b45c1 | 151 | |
f00b45c1 PZ |
152 | #define SCHED_FEAT(name, enabled) \ |
153 | (1UL << __SCHED_FEAT_##name) * enabled | | |
154 | ||
bf5c91ba | 155 | const_debug unsigned int sysctl_sched_features = |
391e43da | 156 | #include "features.h" |
f00b45c1 PZ |
157 | 0; |
158 | ||
159 | #undef SCHED_FEAT | |
160 | ||
161 | #ifdef CONFIG_SCHED_DEBUG | |
162 | #define SCHED_FEAT(name, enabled) \ | |
163 | #name , | |
164 | ||
1292531f | 165 | static const char * const sched_feat_names[] = { |
391e43da | 166 | #include "features.h" |
f00b45c1 PZ |
167 | }; |
168 | ||
169 | #undef SCHED_FEAT | |
170 | ||
34f3a814 | 171 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 172 | { |
f00b45c1 PZ |
173 | int i; |
174 | ||
f8b6d1cc | 175 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
176 | if (!(sysctl_sched_features & (1UL << i))) |
177 | seq_puts(m, "NO_"); | |
178 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 179 | } |
34f3a814 | 180 | seq_puts(m, "\n"); |
f00b45c1 | 181 | |
34f3a814 | 182 | return 0; |
f00b45c1 PZ |
183 | } |
184 | ||
f8b6d1cc PZ |
185 | #ifdef HAVE_JUMP_LABEL |
186 | ||
c5905afb IM |
187 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
188 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
189 | |
190 | #define SCHED_FEAT(name, enabled) \ | |
191 | jump_label_key__##enabled , | |
192 | ||
c5905afb | 193 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
194 | #include "features.h" |
195 | }; | |
196 | ||
197 | #undef SCHED_FEAT | |
198 | ||
199 | static void sched_feat_disable(int i) | |
200 | { | |
c5905afb IM |
201 | if (static_key_enabled(&sched_feat_keys[i])) |
202 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
203 | } |
204 | ||
205 | static void sched_feat_enable(int i) | |
206 | { | |
c5905afb IM |
207 | if (!static_key_enabled(&sched_feat_keys[i])) |
208 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
209 | } |
210 | #else | |
211 | static void sched_feat_disable(int i) { }; | |
212 | static void sched_feat_enable(int i) { }; | |
213 | #endif /* HAVE_JUMP_LABEL */ | |
214 | ||
1a687c2e | 215 | static int sched_feat_set(char *cmp) |
f00b45c1 | 216 | { |
f00b45c1 | 217 | int i; |
1a687c2e | 218 | int neg = 0; |
f00b45c1 | 219 | |
524429c3 | 220 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
221 | neg = 1; |
222 | cmp += 3; | |
223 | } | |
224 | ||
f8b6d1cc | 225 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 226 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 227 | if (neg) { |
f00b45c1 | 228 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
229 | sched_feat_disable(i); |
230 | } else { | |
f00b45c1 | 231 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
232 | sched_feat_enable(i); |
233 | } | |
f00b45c1 PZ |
234 | break; |
235 | } | |
236 | } | |
237 | ||
1a687c2e MG |
238 | return i; |
239 | } | |
240 | ||
241 | static ssize_t | |
242 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
243 | size_t cnt, loff_t *ppos) | |
244 | { | |
245 | char buf[64]; | |
246 | char *cmp; | |
247 | int i; | |
248 | ||
249 | if (cnt > 63) | |
250 | cnt = 63; | |
251 | ||
252 | if (copy_from_user(&buf, ubuf, cnt)) | |
253 | return -EFAULT; | |
254 | ||
255 | buf[cnt] = 0; | |
256 | cmp = strstrip(buf); | |
257 | ||
258 | i = sched_feat_set(cmp); | |
f8b6d1cc | 259 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
260 | return -EINVAL; |
261 | ||
42994724 | 262 | *ppos += cnt; |
f00b45c1 PZ |
263 | |
264 | return cnt; | |
265 | } | |
266 | ||
34f3a814 LZ |
267 | static int sched_feat_open(struct inode *inode, struct file *filp) |
268 | { | |
269 | return single_open(filp, sched_feat_show, NULL); | |
270 | } | |
271 | ||
828c0950 | 272 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
273 | .open = sched_feat_open, |
274 | .write = sched_feat_write, | |
275 | .read = seq_read, | |
276 | .llseek = seq_lseek, | |
277 | .release = single_release, | |
f00b45c1 PZ |
278 | }; |
279 | ||
280 | static __init int sched_init_debug(void) | |
281 | { | |
f00b45c1 PZ |
282 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
283 | &sched_feat_fops); | |
284 | ||
285 | return 0; | |
286 | } | |
287 | late_initcall(sched_init_debug); | |
f8b6d1cc | 288 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 289 | |
b82d9fdd PZ |
290 | /* |
291 | * Number of tasks to iterate in a single balance run. | |
292 | * Limited because this is done with IRQs disabled. | |
293 | */ | |
294 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
295 | ||
e9e9250b PZ |
296 | /* |
297 | * period over which we average the RT time consumption, measured | |
298 | * in ms. | |
299 | * | |
300 | * default: 1s | |
301 | */ | |
302 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
303 | ||
fa85ae24 | 304 | /* |
9f0c1e56 | 305 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
306 | * default: 1s |
307 | */ | |
9f0c1e56 | 308 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 309 | |
029632fb | 310 | __read_mostly int scheduler_running; |
6892b75e | 311 | |
9f0c1e56 PZ |
312 | /* |
313 | * part of the period that we allow rt tasks to run in us. | |
314 | * default: 0.95s | |
315 | */ | |
316 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 317 | |
0970d299 | 318 | /* |
0122ec5b | 319 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 320 | */ |
70b97a7f | 321 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
322 | __acquires(rq->lock) |
323 | { | |
0970d299 PZ |
324 | struct rq *rq; |
325 | ||
0122ec5b PZ |
326 | lockdep_assert_held(&p->pi_lock); |
327 | ||
3a5c359a | 328 | for (;;) { |
0970d299 | 329 | rq = task_rq(p); |
05fa785c | 330 | raw_spin_lock(&rq->lock); |
65cc8e48 | 331 | if (likely(rq == task_rq(p))) |
3a5c359a | 332 | return rq; |
05fa785c | 333 | raw_spin_unlock(&rq->lock); |
b29739f9 | 334 | } |
b29739f9 IM |
335 | } |
336 | ||
1da177e4 | 337 | /* |
0122ec5b | 338 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 339 | */ |
70b97a7f | 340 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 341 | __acquires(p->pi_lock) |
1da177e4 LT |
342 | __acquires(rq->lock) |
343 | { | |
70b97a7f | 344 | struct rq *rq; |
1da177e4 | 345 | |
3a5c359a | 346 | for (;;) { |
0122ec5b | 347 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 348 | rq = task_rq(p); |
05fa785c | 349 | raw_spin_lock(&rq->lock); |
65cc8e48 | 350 | if (likely(rq == task_rq(p))) |
3a5c359a | 351 | return rq; |
0122ec5b PZ |
352 | raw_spin_unlock(&rq->lock); |
353 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 354 | } |
1da177e4 LT |
355 | } |
356 | ||
a9957449 | 357 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
358 | __releases(rq->lock) |
359 | { | |
05fa785c | 360 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
361 | } |
362 | ||
0122ec5b PZ |
363 | static inline void |
364 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 365 | __releases(rq->lock) |
0122ec5b | 366 | __releases(p->pi_lock) |
1da177e4 | 367 | { |
0122ec5b PZ |
368 | raw_spin_unlock(&rq->lock); |
369 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
370 | } |
371 | ||
1da177e4 | 372 | /* |
cc2a73b5 | 373 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 374 | */ |
a9957449 | 375 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
376 | __acquires(rq->lock) |
377 | { | |
70b97a7f | 378 | struct rq *rq; |
1da177e4 LT |
379 | |
380 | local_irq_disable(); | |
381 | rq = this_rq(); | |
05fa785c | 382 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
383 | |
384 | return rq; | |
385 | } | |
386 | ||
8f4d37ec PZ |
387 | #ifdef CONFIG_SCHED_HRTICK |
388 | /* | |
389 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 390 | */ |
8f4d37ec | 391 | |
8f4d37ec PZ |
392 | static void hrtick_clear(struct rq *rq) |
393 | { | |
394 | if (hrtimer_active(&rq->hrtick_timer)) | |
395 | hrtimer_cancel(&rq->hrtick_timer); | |
396 | } | |
397 | ||
8f4d37ec PZ |
398 | /* |
399 | * High-resolution timer tick. | |
400 | * Runs from hardirq context with interrupts disabled. | |
401 | */ | |
402 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
403 | { | |
404 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
405 | ||
406 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
407 | ||
05fa785c | 408 | raw_spin_lock(&rq->lock); |
3e51f33f | 409 | update_rq_clock(rq); |
8f4d37ec | 410 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 411 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
412 | |
413 | return HRTIMER_NORESTART; | |
414 | } | |
415 | ||
95e904c7 | 416 | #ifdef CONFIG_SMP |
971ee28c PZ |
417 | |
418 | static int __hrtick_restart(struct rq *rq) | |
419 | { | |
420 | struct hrtimer *timer = &rq->hrtick_timer; | |
421 | ktime_t time = hrtimer_get_softexpires(timer); | |
422 | ||
423 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
424 | } | |
425 | ||
31656519 PZ |
426 | /* |
427 | * called from hardirq (IPI) context | |
428 | */ | |
429 | static void __hrtick_start(void *arg) | |
b328ca18 | 430 | { |
31656519 | 431 | struct rq *rq = arg; |
b328ca18 | 432 | |
05fa785c | 433 | raw_spin_lock(&rq->lock); |
971ee28c | 434 | __hrtick_restart(rq); |
31656519 | 435 | rq->hrtick_csd_pending = 0; |
05fa785c | 436 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
437 | } |
438 | ||
31656519 PZ |
439 | /* |
440 | * Called to set the hrtick timer state. | |
441 | * | |
442 | * called with rq->lock held and irqs disabled | |
443 | */ | |
029632fb | 444 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 445 | { |
31656519 PZ |
446 | struct hrtimer *timer = &rq->hrtick_timer; |
447 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 448 | |
cc584b21 | 449 | hrtimer_set_expires(timer, time); |
31656519 PZ |
450 | |
451 | if (rq == this_rq()) { | |
971ee28c | 452 | __hrtick_restart(rq); |
31656519 | 453 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 454 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
455 | rq->hrtick_csd_pending = 1; |
456 | } | |
b328ca18 PZ |
457 | } |
458 | ||
459 | static int | |
460 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
461 | { | |
462 | int cpu = (int)(long)hcpu; | |
463 | ||
464 | switch (action) { | |
465 | case CPU_UP_CANCELED: | |
466 | case CPU_UP_CANCELED_FROZEN: | |
467 | case CPU_DOWN_PREPARE: | |
468 | case CPU_DOWN_PREPARE_FROZEN: | |
469 | case CPU_DEAD: | |
470 | case CPU_DEAD_FROZEN: | |
31656519 | 471 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
472 | return NOTIFY_OK; |
473 | } | |
474 | ||
475 | return NOTIFY_DONE; | |
476 | } | |
477 | ||
fa748203 | 478 | static __init void init_hrtick(void) |
b328ca18 PZ |
479 | { |
480 | hotcpu_notifier(hotplug_hrtick, 0); | |
481 | } | |
31656519 PZ |
482 | #else |
483 | /* | |
484 | * Called to set the hrtick timer state. | |
485 | * | |
486 | * called with rq->lock held and irqs disabled | |
487 | */ | |
029632fb | 488 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 489 | { |
7f1e2ca9 | 490 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 491 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 492 | } |
b328ca18 | 493 | |
006c75f1 | 494 | static inline void init_hrtick(void) |
8f4d37ec | 495 | { |
8f4d37ec | 496 | } |
31656519 | 497 | #endif /* CONFIG_SMP */ |
8f4d37ec | 498 | |
31656519 | 499 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 500 | { |
31656519 PZ |
501 | #ifdef CONFIG_SMP |
502 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 503 | |
31656519 PZ |
504 | rq->hrtick_csd.flags = 0; |
505 | rq->hrtick_csd.func = __hrtick_start; | |
506 | rq->hrtick_csd.info = rq; | |
507 | #endif | |
8f4d37ec | 508 | |
31656519 PZ |
509 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
510 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 511 | } |
006c75f1 | 512 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
513 | static inline void hrtick_clear(struct rq *rq) |
514 | { | |
515 | } | |
516 | ||
8f4d37ec PZ |
517 | static inline void init_rq_hrtick(struct rq *rq) |
518 | { | |
519 | } | |
520 | ||
b328ca18 PZ |
521 | static inline void init_hrtick(void) |
522 | { | |
523 | } | |
006c75f1 | 524 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 525 | |
fd99f91a PZ |
526 | /* |
527 | * cmpxchg based fetch_or, macro so it works for different integer types | |
528 | */ | |
529 | #define fetch_or(ptr, val) \ | |
530 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
531 | for (;;) { \ | |
532 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
533 | if (__old == __val) \ | |
534 | break; \ | |
535 | __val = __old; \ | |
536 | } \ | |
537 | __old; \ | |
538 | }) | |
539 | ||
e3baac47 | 540 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
541 | /* |
542 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
543 | * this avoids any races wrt polling state changes and thereby avoids | |
544 | * spurious IPIs. | |
545 | */ | |
546 | static bool set_nr_and_not_polling(struct task_struct *p) | |
547 | { | |
548 | struct thread_info *ti = task_thread_info(p); | |
549 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
550 | } | |
e3baac47 PZ |
551 | |
552 | /* | |
553 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
554 | * | |
555 | * If this returns true, then the idle task promises to call | |
556 | * sched_ttwu_pending() and reschedule soon. | |
557 | */ | |
558 | static bool set_nr_if_polling(struct task_struct *p) | |
559 | { | |
560 | struct thread_info *ti = task_thread_info(p); | |
561 | typeof(ti->flags) old, val = ACCESS_ONCE(ti->flags); | |
562 | ||
563 | for (;;) { | |
564 | if (!(val & _TIF_POLLING_NRFLAG)) | |
565 | return false; | |
566 | if (val & _TIF_NEED_RESCHED) | |
567 | return true; | |
568 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
569 | if (old == val) | |
570 | break; | |
571 | val = old; | |
572 | } | |
573 | return true; | |
574 | } | |
575 | ||
fd99f91a PZ |
576 | #else |
577 | static bool set_nr_and_not_polling(struct task_struct *p) | |
578 | { | |
579 | set_tsk_need_resched(p); | |
580 | return true; | |
581 | } | |
e3baac47 PZ |
582 | |
583 | #ifdef CONFIG_SMP | |
584 | static bool set_nr_if_polling(struct task_struct *p) | |
585 | { | |
586 | return false; | |
587 | } | |
588 | #endif | |
fd99f91a PZ |
589 | #endif |
590 | ||
c24d20db | 591 | /* |
8875125e | 592 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
593 | * |
594 | * On UP this means the setting of the need_resched flag, on SMP it | |
595 | * might also involve a cross-CPU call to trigger the scheduler on | |
596 | * the target CPU. | |
597 | */ | |
8875125e | 598 | void resched_curr(struct rq *rq) |
c24d20db | 599 | { |
8875125e | 600 | struct task_struct *curr = rq->curr; |
c24d20db IM |
601 | int cpu; |
602 | ||
8875125e | 603 | lockdep_assert_held(&rq->lock); |
c24d20db | 604 | |
8875125e | 605 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
606 | return; |
607 | ||
8875125e | 608 | cpu = cpu_of(rq); |
fd99f91a | 609 | |
f27dde8d | 610 | if (cpu == smp_processor_id()) { |
8875125e | 611 | set_tsk_need_resched(curr); |
f27dde8d | 612 | set_preempt_need_resched(); |
c24d20db | 613 | return; |
f27dde8d | 614 | } |
c24d20db | 615 | |
8875125e | 616 | if (set_nr_and_not_polling(curr)) |
c24d20db | 617 | smp_send_reschedule(cpu); |
dfc68f29 AL |
618 | else |
619 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
620 | } |
621 | ||
029632fb | 622 | void resched_cpu(int cpu) |
c24d20db IM |
623 | { |
624 | struct rq *rq = cpu_rq(cpu); | |
625 | unsigned long flags; | |
626 | ||
05fa785c | 627 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 628 | return; |
8875125e | 629 | resched_curr(rq); |
05fa785c | 630 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 631 | } |
06d8308c | 632 | |
b021fe3e | 633 | #ifdef CONFIG_SMP |
3451d024 | 634 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
635 | /* |
636 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
637 | * from an idle cpu. This is good for power-savings. | |
638 | * | |
639 | * We don't do similar optimization for completely idle system, as | |
640 | * selecting an idle cpu will add more delays to the timers than intended | |
641 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
642 | */ | |
6201b4d6 | 643 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
644 | { |
645 | int cpu = smp_processor_id(); | |
646 | int i; | |
647 | struct sched_domain *sd; | |
648 | ||
6201b4d6 VK |
649 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
650 | return cpu; | |
651 | ||
057f3fad | 652 | rcu_read_lock(); |
83cd4fe2 | 653 | for_each_domain(cpu, sd) { |
057f3fad PZ |
654 | for_each_cpu(i, sched_domain_span(sd)) { |
655 | if (!idle_cpu(i)) { | |
656 | cpu = i; | |
657 | goto unlock; | |
658 | } | |
659 | } | |
83cd4fe2 | 660 | } |
057f3fad PZ |
661 | unlock: |
662 | rcu_read_unlock(); | |
83cd4fe2 VP |
663 | return cpu; |
664 | } | |
06d8308c TG |
665 | /* |
666 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
667 | * idle CPU then this timer might expire before the next timer event | |
668 | * which is scheduled to wake up that CPU. In case of a completely | |
669 | * idle system the next event might even be infinite time into the | |
670 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
671 | * leaves the inner idle loop so the newly added timer is taken into | |
672 | * account when the CPU goes back to idle and evaluates the timer | |
673 | * wheel for the next timer event. | |
674 | */ | |
1c20091e | 675 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
676 | { |
677 | struct rq *rq = cpu_rq(cpu); | |
678 | ||
679 | if (cpu == smp_processor_id()) | |
680 | return; | |
681 | ||
67b9ca70 | 682 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 683 | smp_send_reschedule(cpu); |
dfc68f29 AL |
684 | else |
685 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
686 | } |
687 | ||
c5bfece2 | 688 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 689 | { |
53c5fa16 FW |
690 | /* |
691 | * We just need the target to call irq_exit() and re-evaluate | |
692 | * the next tick. The nohz full kick at least implies that. | |
693 | * If needed we can still optimize that later with an | |
694 | * empty IRQ. | |
695 | */ | |
c5bfece2 | 696 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
697 | if (cpu != smp_processor_id() || |
698 | tick_nohz_tick_stopped()) | |
53c5fa16 | 699 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
700 | return true; |
701 | } | |
702 | ||
703 | return false; | |
704 | } | |
705 | ||
706 | void wake_up_nohz_cpu(int cpu) | |
707 | { | |
c5bfece2 | 708 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
709 | wake_up_idle_cpu(cpu); |
710 | } | |
711 | ||
ca38062e | 712 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 713 | { |
1c792db7 | 714 | int cpu = smp_processor_id(); |
873b4c65 VG |
715 | |
716 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
717 | return false; | |
718 | ||
719 | if (idle_cpu(cpu) && !need_resched()) | |
720 | return true; | |
721 | ||
722 | /* | |
723 | * We can't run Idle Load Balance on this CPU for this time so we | |
724 | * cancel it and clear NOHZ_BALANCE_KICK | |
725 | */ | |
726 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
727 | return false; | |
45bf76df IM |
728 | } |
729 | ||
3451d024 | 730 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 731 | |
ca38062e | 732 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 733 | { |
ca38062e | 734 | return false; |
2069dd75 PZ |
735 | } |
736 | ||
3451d024 | 737 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 738 | |
ce831b38 FW |
739 | #ifdef CONFIG_NO_HZ_FULL |
740 | bool sched_can_stop_tick(void) | |
741 | { | |
3882ec64 FW |
742 | /* |
743 | * More than one running task need preemption. | |
744 | * nr_running update is assumed to be visible | |
745 | * after IPI is sent from wakers. | |
746 | */ | |
541b8264 VK |
747 | if (this_rq()->nr_running > 1) |
748 | return false; | |
ce831b38 | 749 | |
541b8264 | 750 | return true; |
ce831b38 FW |
751 | } |
752 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 753 | |
029632fb | 754 | void sched_avg_update(struct rq *rq) |
18d95a28 | 755 | { |
e9e9250b PZ |
756 | s64 period = sched_avg_period(); |
757 | ||
78becc27 | 758 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
759 | /* |
760 | * Inline assembly required to prevent the compiler | |
761 | * optimising this loop into a divmod call. | |
762 | * See __iter_div_u64_rem() for another example of this. | |
763 | */ | |
764 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
765 | rq->age_stamp += period; |
766 | rq->rt_avg /= 2; | |
767 | } | |
18d95a28 PZ |
768 | } |
769 | ||
6d6bc0ad | 770 | #endif /* CONFIG_SMP */ |
18d95a28 | 771 | |
a790de99 PT |
772 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
773 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 774 | /* |
8277434e PT |
775 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
776 | * node and @up when leaving it for the final time. | |
777 | * | |
778 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 779 | */ |
029632fb | 780 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 781 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
782 | { |
783 | struct task_group *parent, *child; | |
eb755805 | 784 | int ret; |
c09595f6 | 785 | |
8277434e PT |
786 | parent = from; |
787 | ||
c09595f6 | 788 | down: |
eb755805 PZ |
789 | ret = (*down)(parent, data); |
790 | if (ret) | |
8277434e | 791 | goto out; |
c09595f6 PZ |
792 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
793 | parent = child; | |
794 | goto down; | |
795 | ||
796 | up: | |
797 | continue; | |
798 | } | |
eb755805 | 799 | ret = (*up)(parent, data); |
8277434e PT |
800 | if (ret || parent == from) |
801 | goto out; | |
c09595f6 PZ |
802 | |
803 | child = parent; | |
804 | parent = parent->parent; | |
805 | if (parent) | |
806 | goto up; | |
8277434e | 807 | out: |
eb755805 | 808 | return ret; |
c09595f6 PZ |
809 | } |
810 | ||
029632fb | 811 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 812 | { |
e2b245f8 | 813 | return 0; |
eb755805 | 814 | } |
18d95a28 PZ |
815 | #endif |
816 | ||
45bf76df IM |
817 | static void set_load_weight(struct task_struct *p) |
818 | { | |
f05998d4 NR |
819 | int prio = p->static_prio - MAX_RT_PRIO; |
820 | struct load_weight *load = &p->se.load; | |
821 | ||
dd41f596 IM |
822 | /* |
823 | * SCHED_IDLE tasks get minimal weight: | |
824 | */ | |
825 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 826 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 827 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
828 | return; |
829 | } | |
71f8bd46 | 830 | |
c8b28116 | 831 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 832 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
833 | } |
834 | ||
371fd7e7 | 835 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 836 | { |
a64692a3 | 837 | update_rq_clock(rq); |
43148951 | 838 | sched_info_queued(rq, p); |
371fd7e7 | 839 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
840 | } |
841 | ||
371fd7e7 | 842 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 843 | { |
a64692a3 | 844 | update_rq_clock(rq); |
43148951 | 845 | sched_info_dequeued(rq, p); |
371fd7e7 | 846 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
847 | } |
848 | ||
029632fb | 849 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
850 | { |
851 | if (task_contributes_to_load(p)) | |
852 | rq->nr_uninterruptible--; | |
853 | ||
371fd7e7 | 854 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
855 | } |
856 | ||
029632fb | 857 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
858 | { |
859 | if (task_contributes_to_load(p)) | |
860 | rq->nr_uninterruptible++; | |
861 | ||
371fd7e7 | 862 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
863 | } |
864 | ||
fe44d621 | 865 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 866 | { |
095c0aa8 GC |
867 | /* |
868 | * In theory, the compile should just see 0 here, and optimize out the call | |
869 | * to sched_rt_avg_update. But I don't trust it... | |
870 | */ | |
871 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
872 | s64 steal = 0, irq_delta = 0; | |
873 | #endif | |
874 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 875 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
876 | |
877 | /* | |
878 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
879 | * this case when a previous update_rq_clock() happened inside a | |
880 | * {soft,}irq region. | |
881 | * | |
882 | * When this happens, we stop ->clock_task and only update the | |
883 | * prev_irq_time stamp to account for the part that fit, so that a next | |
884 | * update will consume the rest. This ensures ->clock_task is | |
885 | * monotonic. | |
886 | * | |
887 | * It does however cause some slight miss-attribution of {soft,}irq | |
888 | * time, a more accurate solution would be to update the irq_time using | |
889 | * the current rq->clock timestamp, except that would require using | |
890 | * atomic ops. | |
891 | */ | |
892 | if (irq_delta > delta) | |
893 | irq_delta = delta; | |
894 | ||
895 | rq->prev_irq_time += irq_delta; | |
896 | delta -= irq_delta; | |
095c0aa8 GC |
897 | #endif |
898 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 899 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
900 | steal = paravirt_steal_clock(cpu_of(rq)); |
901 | steal -= rq->prev_steal_time_rq; | |
902 | ||
903 | if (unlikely(steal > delta)) | |
904 | steal = delta; | |
905 | ||
095c0aa8 | 906 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
907 | delta -= steal; |
908 | } | |
909 | #endif | |
910 | ||
fe44d621 PZ |
911 | rq->clock_task += delta; |
912 | ||
095c0aa8 | 913 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 914 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
915 | sched_rt_avg_update(rq, irq_delta + steal); |
916 | #endif | |
aa483808 VP |
917 | } |
918 | ||
34f971f6 PZ |
919 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
920 | { | |
921 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
922 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
923 | ||
924 | if (stop) { | |
925 | /* | |
926 | * Make it appear like a SCHED_FIFO task, its something | |
927 | * userspace knows about and won't get confused about. | |
928 | * | |
929 | * Also, it will make PI more or less work without too | |
930 | * much confusion -- but then, stop work should not | |
931 | * rely on PI working anyway. | |
932 | */ | |
933 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
934 | ||
935 | stop->sched_class = &stop_sched_class; | |
936 | } | |
937 | ||
938 | cpu_rq(cpu)->stop = stop; | |
939 | ||
940 | if (old_stop) { | |
941 | /* | |
942 | * Reset it back to a normal scheduling class so that | |
943 | * it can die in pieces. | |
944 | */ | |
945 | old_stop->sched_class = &rt_sched_class; | |
946 | } | |
947 | } | |
948 | ||
14531189 | 949 | /* |
dd41f596 | 950 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 951 | */ |
14531189 IM |
952 | static inline int __normal_prio(struct task_struct *p) |
953 | { | |
dd41f596 | 954 | return p->static_prio; |
14531189 IM |
955 | } |
956 | ||
b29739f9 IM |
957 | /* |
958 | * Calculate the expected normal priority: i.e. priority | |
959 | * without taking RT-inheritance into account. Might be | |
960 | * boosted by interactivity modifiers. Changes upon fork, | |
961 | * setprio syscalls, and whenever the interactivity | |
962 | * estimator recalculates. | |
963 | */ | |
36c8b586 | 964 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
965 | { |
966 | int prio; | |
967 | ||
aab03e05 DF |
968 | if (task_has_dl_policy(p)) |
969 | prio = MAX_DL_PRIO-1; | |
970 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
971 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
972 | else | |
973 | prio = __normal_prio(p); | |
974 | return prio; | |
975 | } | |
976 | ||
977 | /* | |
978 | * Calculate the current priority, i.e. the priority | |
979 | * taken into account by the scheduler. This value might | |
980 | * be boosted by RT tasks, or might be boosted by | |
981 | * interactivity modifiers. Will be RT if the task got | |
982 | * RT-boosted. If not then it returns p->normal_prio. | |
983 | */ | |
36c8b586 | 984 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
985 | { |
986 | p->normal_prio = normal_prio(p); | |
987 | /* | |
988 | * If we are RT tasks or we were boosted to RT priority, | |
989 | * keep the priority unchanged. Otherwise, update priority | |
990 | * to the normal priority: | |
991 | */ | |
992 | if (!rt_prio(p->prio)) | |
993 | return p->normal_prio; | |
994 | return p->prio; | |
995 | } | |
996 | ||
1da177e4 LT |
997 | /** |
998 | * task_curr - is this task currently executing on a CPU? | |
999 | * @p: the task in question. | |
e69f6186 YB |
1000 | * |
1001 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1002 | */ |
36c8b586 | 1003 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1004 | { |
1005 | return cpu_curr(task_cpu(p)) == p; | |
1006 | } | |
1007 | ||
cb469845 SR |
1008 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1009 | const struct sched_class *prev_class, | |
da7a735e | 1010 | int oldprio) |
cb469845 SR |
1011 | { |
1012 | if (prev_class != p->sched_class) { | |
1013 | if (prev_class->switched_from) | |
da7a735e PZ |
1014 | prev_class->switched_from(rq, p); |
1015 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 1016 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1017 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1018 | } |
1019 | ||
029632fb | 1020 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1021 | { |
1022 | const struct sched_class *class; | |
1023 | ||
1024 | if (p->sched_class == rq->curr->sched_class) { | |
1025 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1026 | } else { | |
1027 | for_each_class(class) { | |
1028 | if (class == rq->curr->sched_class) | |
1029 | break; | |
1030 | if (class == p->sched_class) { | |
8875125e | 1031 | resched_curr(rq); |
1e5a7405 PZ |
1032 | break; |
1033 | } | |
1034 | } | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * A queue event has occurred, and we're going to schedule. In | |
1039 | * this case, we can save a useless back to back clock update. | |
1040 | */ | |
fd2f4419 | 1041 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1042 | rq->skip_clock_update = 1; |
1043 | } | |
1044 | ||
1da177e4 | 1045 | #ifdef CONFIG_SMP |
dd41f596 | 1046 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1047 | { |
e2912009 PZ |
1048 | #ifdef CONFIG_SCHED_DEBUG |
1049 | /* | |
1050 | * We should never call set_task_cpu() on a blocked task, | |
1051 | * ttwu() will sort out the placement. | |
1052 | */ | |
077614ee | 1053 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1054 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1055 | |
1056 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1057 | /* |
1058 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1059 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1060 | * | |
1061 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1062 | * see task_group(). |
6c6c54e1 PZ |
1063 | * |
1064 | * Furthermore, all task_rq users should acquire both locks, see | |
1065 | * task_rq_lock(). | |
1066 | */ | |
0122ec5b PZ |
1067 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1068 | lockdep_is_held(&task_rq(p)->lock))); | |
1069 | #endif | |
e2912009 PZ |
1070 | #endif |
1071 | ||
de1d7286 | 1072 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1073 | |
0c69774e | 1074 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1075 | if (p->sched_class->migrate_task_rq) |
1076 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1077 | p->se.nr_migrations++; |
a8b0ca17 | 1078 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1079 | } |
dd41f596 IM |
1080 | |
1081 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1082 | } |
1083 | ||
ac66f547 PZ |
1084 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1085 | { | |
1086 | if (p->on_rq) { | |
1087 | struct rq *src_rq, *dst_rq; | |
1088 | ||
1089 | src_rq = task_rq(p); | |
1090 | dst_rq = cpu_rq(cpu); | |
1091 | ||
1092 | deactivate_task(src_rq, p, 0); | |
1093 | set_task_cpu(p, cpu); | |
1094 | activate_task(dst_rq, p, 0); | |
1095 | check_preempt_curr(dst_rq, p, 0); | |
1096 | } else { | |
1097 | /* | |
1098 | * Task isn't running anymore; make it appear like we migrated | |
1099 | * it before it went to sleep. This means on wakeup we make the | |
1100 | * previous cpu our targer instead of where it really is. | |
1101 | */ | |
1102 | p->wake_cpu = cpu; | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | struct migration_swap_arg { | |
1107 | struct task_struct *src_task, *dst_task; | |
1108 | int src_cpu, dst_cpu; | |
1109 | }; | |
1110 | ||
1111 | static int migrate_swap_stop(void *data) | |
1112 | { | |
1113 | struct migration_swap_arg *arg = data; | |
1114 | struct rq *src_rq, *dst_rq; | |
1115 | int ret = -EAGAIN; | |
1116 | ||
1117 | src_rq = cpu_rq(arg->src_cpu); | |
1118 | dst_rq = cpu_rq(arg->dst_cpu); | |
1119 | ||
74602315 PZ |
1120 | double_raw_lock(&arg->src_task->pi_lock, |
1121 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1122 | double_rq_lock(src_rq, dst_rq); |
1123 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1124 | goto unlock; | |
1125 | ||
1126 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1127 | goto unlock; | |
1128 | ||
1129 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1130 | goto unlock; | |
1131 | ||
1132 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1133 | goto unlock; | |
1134 | ||
1135 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1136 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1137 | ||
1138 | ret = 0; | |
1139 | ||
1140 | unlock: | |
1141 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1142 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1143 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1144 | |
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * Cross migrate two tasks | |
1150 | */ | |
1151 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1152 | { | |
1153 | struct migration_swap_arg arg; | |
1154 | int ret = -EINVAL; | |
1155 | ||
ac66f547 PZ |
1156 | arg = (struct migration_swap_arg){ |
1157 | .src_task = cur, | |
1158 | .src_cpu = task_cpu(cur), | |
1159 | .dst_task = p, | |
1160 | .dst_cpu = task_cpu(p), | |
1161 | }; | |
1162 | ||
1163 | if (arg.src_cpu == arg.dst_cpu) | |
1164 | goto out; | |
1165 | ||
6acce3ef PZ |
1166 | /* |
1167 | * These three tests are all lockless; this is OK since all of them | |
1168 | * will be re-checked with proper locks held further down the line. | |
1169 | */ | |
ac66f547 PZ |
1170 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1171 | goto out; | |
1172 | ||
1173 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1174 | goto out; | |
1175 | ||
1176 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1177 | goto out; | |
1178 | ||
286549dc | 1179 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1180 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1181 | ||
1182 | out: | |
ac66f547 PZ |
1183 | return ret; |
1184 | } | |
1185 | ||
969c7921 | 1186 | struct migration_arg { |
36c8b586 | 1187 | struct task_struct *task; |
1da177e4 | 1188 | int dest_cpu; |
70b97a7f | 1189 | }; |
1da177e4 | 1190 | |
969c7921 TH |
1191 | static int migration_cpu_stop(void *data); |
1192 | ||
1da177e4 LT |
1193 | /* |
1194 | * wait_task_inactive - wait for a thread to unschedule. | |
1195 | * | |
85ba2d86 RM |
1196 | * If @match_state is nonzero, it's the @p->state value just checked and |
1197 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1198 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1199 | * we return a positive number (its total switch count). If a second call | |
1200 | * a short while later returns the same number, the caller can be sure that | |
1201 | * @p has remained unscheduled the whole time. | |
1202 | * | |
1da177e4 LT |
1203 | * The caller must ensure that the task *will* unschedule sometime soon, |
1204 | * else this function might spin for a *long* time. This function can't | |
1205 | * be called with interrupts off, or it may introduce deadlock with | |
1206 | * smp_call_function() if an IPI is sent by the same process we are | |
1207 | * waiting to become inactive. | |
1208 | */ | |
85ba2d86 | 1209 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1210 | { |
1211 | unsigned long flags; | |
dd41f596 | 1212 | int running, on_rq; |
85ba2d86 | 1213 | unsigned long ncsw; |
70b97a7f | 1214 | struct rq *rq; |
1da177e4 | 1215 | |
3a5c359a AK |
1216 | for (;;) { |
1217 | /* | |
1218 | * We do the initial early heuristics without holding | |
1219 | * any task-queue locks at all. We'll only try to get | |
1220 | * the runqueue lock when things look like they will | |
1221 | * work out! | |
1222 | */ | |
1223 | rq = task_rq(p); | |
fa490cfd | 1224 | |
3a5c359a AK |
1225 | /* |
1226 | * If the task is actively running on another CPU | |
1227 | * still, just relax and busy-wait without holding | |
1228 | * any locks. | |
1229 | * | |
1230 | * NOTE! Since we don't hold any locks, it's not | |
1231 | * even sure that "rq" stays as the right runqueue! | |
1232 | * But we don't care, since "task_running()" will | |
1233 | * return false if the runqueue has changed and p | |
1234 | * is actually now running somewhere else! | |
1235 | */ | |
85ba2d86 RM |
1236 | while (task_running(rq, p)) { |
1237 | if (match_state && unlikely(p->state != match_state)) | |
1238 | return 0; | |
3a5c359a | 1239 | cpu_relax(); |
85ba2d86 | 1240 | } |
fa490cfd | 1241 | |
3a5c359a AK |
1242 | /* |
1243 | * Ok, time to look more closely! We need the rq | |
1244 | * lock now, to be *sure*. If we're wrong, we'll | |
1245 | * just go back and repeat. | |
1246 | */ | |
1247 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1248 | trace_sched_wait_task(p); |
3a5c359a | 1249 | running = task_running(rq, p); |
fd2f4419 | 1250 | on_rq = p->on_rq; |
85ba2d86 | 1251 | ncsw = 0; |
f31e11d8 | 1252 | if (!match_state || p->state == match_state) |
93dcf55f | 1253 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1254 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1255 | |
85ba2d86 RM |
1256 | /* |
1257 | * If it changed from the expected state, bail out now. | |
1258 | */ | |
1259 | if (unlikely(!ncsw)) | |
1260 | break; | |
1261 | ||
3a5c359a AK |
1262 | /* |
1263 | * Was it really running after all now that we | |
1264 | * checked with the proper locks actually held? | |
1265 | * | |
1266 | * Oops. Go back and try again.. | |
1267 | */ | |
1268 | if (unlikely(running)) { | |
1269 | cpu_relax(); | |
1270 | continue; | |
1271 | } | |
fa490cfd | 1272 | |
3a5c359a AK |
1273 | /* |
1274 | * It's not enough that it's not actively running, | |
1275 | * it must be off the runqueue _entirely_, and not | |
1276 | * preempted! | |
1277 | * | |
80dd99b3 | 1278 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1279 | * running right now), it's preempted, and we should |
1280 | * yield - it could be a while. | |
1281 | */ | |
1282 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1283 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1284 | ||
1285 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1286 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1287 | continue; |
1288 | } | |
fa490cfd | 1289 | |
3a5c359a AK |
1290 | /* |
1291 | * Ahh, all good. It wasn't running, and it wasn't | |
1292 | * runnable, which means that it will never become | |
1293 | * running in the future either. We're all done! | |
1294 | */ | |
1295 | break; | |
1296 | } | |
85ba2d86 RM |
1297 | |
1298 | return ncsw; | |
1da177e4 LT |
1299 | } |
1300 | ||
1301 | /*** | |
1302 | * kick_process - kick a running thread to enter/exit the kernel | |
1303 | * @p: the to-be-kicked thread | |
1304 | * | |
1305 | * Cause a process which is running on another CPU to enter | |
1306 | * kernel-mode, without any delay. (to get signals handled.) | |
1307 | * | |
25985edc | 1308 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1309 | * because all it wants to ensure is that the remote task enters |
1310 | * the kernel. If the IPI races and the task has been migrated | |
1311 | * to another CPU then no harm is done and the purpose has been | |
1312 | * achieved as well. | |
1313 | */ | |
36c8b586 | 1314 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1315 | { |
1316 | int cpu; | |
1317 | ||
1318 | preempt_disable(); | |
1319 | cpu = task_cpu(p); | |
1320 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1321 | smp_send_reschedule(cpu); | |
1322 | preempt_enable(); | |
1323 | } | |
b43e3521 | 1324 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1325 | #endif /* CONFIG_SMP */ |
1da177e4 | 1326 | |
970b13ba | 1327 | #ifdef CONFIG_SMP |
30da688e | 1328 | /* |
013fdb80 | 1329 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1330 | */ |
5da9a0fb PZ |
1331 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1332 | { | |
aa00d89c TC |
1333 | int nid = cpu_to_node(cpu); |
1334 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1335 | enum { cpuset, possible, fail } state = cpuset; |
1336 | int dest_cpu; | |
5da9a0fb | 1337 | |
aa00d89c TC |
1338 | /* |
1339 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1340 | * will return -1. There is no cpu on the node, and we should | |
1341 | * select the cpu on the other node. | |
1342 | */ | |
1343 | if (nid != -1) { | |
1344 | nodemask = cpumask_of_node(nid); | |
1345 | ||
1346 | /* Look for allowed, online CPU in same node. */ | |
1347 | for_each_cpu(dest_cpu, nodemask) { | |
1348 | if (!cpu_online(dest_cpu)) | |
1349 | continue; | |
1350 | if (!cpu_active(dest_cpu)) | |
1351 | continue; | |
1352 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1353 | return dest_cpu; | |
1354 | } | |
2baab4e9 | 1355 | } |
5da9a0fb | 1356 | |
2baab4e9 PZ |
1357 | for (;;) { |
1358 | /* Any allowed, online CPU? */ | |
e3831edd | 1359 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1360 | if (!cpu_online(dest_cpu)) |
1361 | continue; | |
1362 | if (!cpu_active(dest_cpu)) | |
1363 | continue; | |
1364 | goto out; | |
1365 | } | |
5da9a0fb | 1366 | |
2baab4e9 PZ |
1367 | switch (state) { |
1368 | case cpuset: | |
1369 | /* No more Mr. Nice Guy. */ | |
1370 | cpuset_cpus_allowed_fallback(p); | |
1371 | state = possible; | |
1372 | break; | |
1373 | ||
1374 | case possible: | |
1375 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1376 | state = fail; | |
1377 | break; | |
1378 | ||
1379 | case fail: | |
1380 | BUG(); | |
1381 | break; | |
1382 | } | |
1383 | } | |
1384 | ||
1385 | out: | |
1386 | if (state != cpuset) { | |
1387 | /* | |
1388 | * Don't tell them about moving exiting tasks or | |
1389 | * kernel threads (both mm NULL), since they never | |
1390 | * leave kernel. | |
1391 | */ | |
1392 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1393 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1394 | task_pid_nr(p), p->comm, cpu); |
1395 | } | |
5da9a0fb PZ |
1396 | } |
1397 | ||
1398 | return dest_cpu; | |
1399 | } | |
1400 | ||
e2912009 | 1401 | /* |
013fdb80 | 1402 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1403 | */ |
970b13ba | 1404 | static inline |
ac66f547 | 1405 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1406 | { |
ac66f547 | 1407 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1408 | |
1409 | /* | |
1410 | * In order not to call set_task_cpu() on a blocking task we need | |
1411 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1412 | * cpu. | |
1413 | * | |
1414 | * Since this is common to all placement strategies, this lives here. | |
1415 | * | |
1416 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1417 | * not worry about this generic constraint ] | |
1418 | */ | |
fa17b507 | 1419 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1420 | !cpu_online(cpu))) |
5da9a0fb | 1421 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1422 | |
1423 | return cpu; | |
970b13ba | 1424 | } |
09a40af5 MG |
1425 | |
1426 | static void update_avg(u64 *avg, u64 sample) | |
1427 | { | |
1428 | s64 diff = sample - *avg; | |
1429 | *avg += diff >> 3; | |
1430 | } | |
970b13ba PZ |
1431 | #endif |
1432 | ||
d7c01d27 | 1433 | static void |
b84cb5df | 1434 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1435 | { |
d7c01d27 | 1436 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1437 | struct rq *rq = this_rq(); |
1438 | ||
d7c01d27 PZ |
1439 | #ifdef CONFIG_SMP |
1440 | int this_cpu = smp_processor_id(); | |
1441 | ||
1442 | if (cpu == this_cpu) { | |
1443 | schedstat_inc(rq, ttwu_local); | |
1444 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1445 | } else { | |
1446 | struct sched_domain *sd; | |
1447 | ||
1448 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1449 | rcu_read_lock(); |
d7c01d27 PZ |
1450 | for_each_domain(this_cpu, sd) { |
1451 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1452 | schedstat_inc(sd, ttwu_wake_remote); | |
1453 | break; | |
1454 | } | |
1455 | } | |
057f3fad | 1456 | rcu_read_unlock(); |
d7c01d27 | 1457 | } |
f339b9dc PZ |
1458 | |
1459 | if (wake_flags & WF_MIGRATED) | |
1460 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1461 | ||
d7c01d27 PZ |
1462 | #endif /* CONFIG_SMP */ |
1463 | ||
1464 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1465 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1466 | |
1467 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1468 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1469 | |
d7c01d27 PZ |
1470 | #endif /* CONFIG_SCHEDSTATS */ |
1471 | } | |
1472 | ||
1473 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1474 | { | |
9ed3811a | 1475 | activate_task(rq, p, en_flags); |
fd2f4419 | 1476 | p->on_rq = 1; |
c2f7115e PZ |
1477 | |
1478 | /* if a worker is waking up, notify workqueue */ | |
1479 | if (p->flags & PF_WQ_WORKER) | |
1480 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1481 | } |
1482 | ||
23f41eeb PZ |
1483 | /* |
1484 | * Mark the task runnable and perform wakeup-preemption. | |
1485 | */ | |
89363381 | 1486 | static void |
23f41eeb | 1487 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1488 | { |
9ed3811a | 1489 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1490 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1491 | |
1492 | p->state = TASK_RUNNING; | |
1493 | #ifdef CONFIG_SMP | |
1494 | if (p->sched_class->task_woken) | |
1495 | p->sched_class->task_woken(rq, p); | |
1496 | ||
e69c6341 | 1497 | if (rq->idle_stamp) { |
78becc27 | 1498 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1499 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1500 | |
abfafa54 JL |
1501 | update_avg(&rq->avg_idle, delta); |
1502 | ||
1503 | if (rq->avg_idle > max) | |
9ed3811a | 1504 | rq->avg_idle = max; |
abfafa54 | 1505 | |
9ed3811a TH |
1506 | rq->idle_stamp = 0; |
1507 | } | |
1508 | #endif | |
1509 | } | |
1510 | ||
c05fbafb PZ |
1511 | static void |
1512 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1513 | { | |
1514 | #ifdef CONFIG_SMP | |
1515 | if (p->sched_contributes_to_load) | |
1516 | rq->nr_uninterruptible--; | |
1517 | #endif | |
1518 | ||
1519 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1520 | ttwu_do_wakeup(rq, p, wake_flags); | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1525 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1526 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1527 | * the task is still ->on_rq. | |
1528 | */ | |
1529 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1530 | { | |
1531 | struct rq *rq; | |
1532 | int ret = 0; | |
1533 | ||
1534 | rq = __task_rq_lock(p); | |
1535 | if (p->on_rq) { | |
1ad4ec0d FW |
1536 | /* check_preempt_curr() may use rq clock */ |
1537 | update_rq_clock(rq); | |
c05fbafb PZ |
1538 | ttwu_do_wakeup(rq, p, wake_flags); |
1539 | ret = 1; | |
1540 | } | |
1541 | __task_rq_unlock(rq); | |
1542 | ||
1543 | return ret; | |
1544 | } | |
1545 | ||
317f3941 | 1546 | #ifdef CONFIG_SMP |
e3baac47 | 1547 | void sched_ttwu_pending(void) |
317f3941 PZ |
1548 | { |
1549 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1550 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1551 | struct task_struct *p; | |
e3baac47 | 1552 | unsigned long flags; |
317f3941 | 1553 | |
e3baac47 PZ |
1554 | if (!llist) |
1555 | return; | |
1556 | ||
1557 | raw_spin_lock_irqsave(&rq->lock, flags); | |
317f3941 | 1558 | |
fa14ff4a PZ |
1559 | while (llist) { |
1560 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1561 | llist = llist_next(llist); | |
317f3941 PZ |
1562 | ttwu_do_activate(rq, p, 0); |
1563 | } | |
1564 | ||
e3baac47 | 1565 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1566 | } |
1567 | ||
1568 | void scheduler_ipi(void) | |
1569 | { | |
f27dde8d PZ |
1570 | /* |
1571 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1572 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1573 | * this IPI. | |
1574 | */ | |
8cb75e0c | 1575 | preempt_fold_need_resched(); |
f27dde8d | 1576 | |
fd2ac4f4 | 1577 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1578 | return; |
1579 | ||
1580 | /* | |
1581 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1582 | * traditionally all their work was done from the interrupt return | |
1583 | * path. Now that we actually do some work, we need to make sure | |
1584 | * we do call them. | |
1585 | * | |
1586 | * Some archs already do call them, luckily irq_enter/exit nest | |
1587 | * properly. | |
1588 | * | |
1589 | * Arguably we should visit all archs and update all handlers, | |
1590 | * however a fair share of IPIs are still resched only so this would | |
1591 | * somewhat pessimize the simple resched case. | |
1592 | */ | |
1593 | irq_enter(); | |
fa14ff4a | 1594 | sched_ttwu_pending(); |
ca38062e SS |
1595 | |
1596 | /* | |
1597 | * Check if someone kicked us for doing the nohz idle load balance. | |
1598 | */ | |
873b4c65 | 1599 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1600 | this_rq()->idle_balance = 1; |
ca38062e | 1601 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1602 | } |
c5d753a5 | 1603 | irq_exit(); |
317f3941 PZ |
1604 | } |
1605 | ||
1606 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1607 | { | |
e3baac47 PZ |
1608 | struct rq *rq = cpu_rq(cpu); |
1609 | ||
1610 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1611 | if (!set_nr_if_polling(rq->idle)) | |
1612 | smp_send_reschedule(cpu); | |
1613 | else | |
1614 | trace_sched_wake_idle_without_ipi(cpu); | |
1615 | } | |
317f3941 | 1616 | } |
d6aa8f85 | 1617 | |
39be3501 | 1618 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1619 | { |
1620 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1621 | } | |
d6aa8f85 | 1622 | #endif /* CONFIG_SMP */ |
317f3941 | 1623 | |
c05fbafb PZ |
1624 | static void ttwu_queue(struct task_struct *p, int cpu) |
1625 | { | |
1626 | struct rq *rq = cpu_rq(cpu); | |
1627 | ||
17d9f311 | 1628 | #if defined(CONFIG_SMP) |
39be3501 | 1629 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1630 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1631 | ttwu_queue_remote(p, cpu); |
1632 | return; | |
1633 | } | |
1634 | #endif | |
1635 | ||
c05fbafb PZ |
1636 | raw_spin_lock(&rq->lock); |
1637 | ttwu_do_activate(rq, p, 0); | |
1638 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1639 | } |
1640 | ||
1641 | /** | |
1da177e4 | 1642 | * try_to_wake_up - wake up a thread |
9ed3811a | 1643 | * @p: the thread to be awakened |
1da177e4 | 1644 | * @state: the mask of task states that can be woken |
9ed3811a | 1645 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1646 | * |
1647 | * Put it on the run-queue if it's not already there. The "current" | |
1648 | * thread is always on the run-queue (except when the actual | |
1649 | * re-schedule is in progress), and as such you're allowed to do | |
1650 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1651 | * runnable without the overhead of this. | |
1652 | * | |
e69f6186 | 1653 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1654 | * or @state didn't match @p's state. |
1da177e4 | 1655 | */ |
e4a52bcb PZ |
1656 | static int |
1657 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1658 | { |
1da177e4 | 1659 | unsigned long flags; |
c05fbafb | 1660 | int cpu, success = 0; |
2398f2c6 | 1661 | |
e0acd0a6 ON |
1662 | /* |
1663 | * If we are going to wake up a thread waiting for CONDITION we | |
1664 | * need to ensure that CONDITION=1 done by the caller can not be | |
1665 | * reordered with p->state check below. This pairs with mb() in | |
1666 | * set_current_state() the waiting thread does. | |
1667 | */ | |
1668 | smp_mb__before_spinlock(); | |
013fdb80 | 1669 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1670 | if (!(p->state & state)) |
1da177e4 LT |
1671 | goto out; |
1672 | ||
c05fbafb | 1673 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1674 | cpu = task_cpu(p); |
1da177e4 | 1675 | |
c05fbafb PZ |
1676 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1677 | goto stat; | |
1da177e4 | 1678 | |
1da177e4 | 1679 | #ifdef CONFIG_SMP |
e9c84311 | 1680 | /* |
c05fbafb PZ |
1681 | * If the owning (remote) cpu is still in the middle of schedule() with |
1682 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1683 | */ |
f3e94786 | 1684 | while (p->on_cpu) |
e4a52bcb | 1685 | cpu_relax(); |
0970d299 | 1686 | /* |
e4a52bcb | 1687 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1688 | */ |
e4a52bcb | 1689 | smp_rmb(); |
1da177e4 | 1690 | |
a8e4f2ea | 1691 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1692 | p->state = TASK_WAKING; |
e7693a36 | 1693 | |
e4a52bcb | 1694 | if (p->sched_class->task_waking) |
74f8e4b2 | 1695 | p->sched_class->task_waking(p); |
efbbd05a | 1696 | |
ac66f547 | 1697 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1698 | if (task_cpu(p) != cpu) { |
1699 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1700 | set_task_cpu(p, cpu); |
f339b9dc | 1701 | } |
1da177e4 | 1702 | #endif /* CONFIG_SMP */ |
1da177e4 | 1703 | |
c05fbafb PZ |
1704 | ttwu_queue(p, cpu); |
1705 | stat: | |
b84cb5df | 1706 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1707 | out: |
013fdb80 | 1708 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1709 | |
1710 | return success; | |
1711 | } | |
1712 | ||
21aa9af0 TH |
1713 | /** |
1714 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1715 | * @p: the thread to be awakened | |
1716 | * | |
2acca55e | 1717 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1718 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1719 | * the current task. |
21aa9af0 TH |
1720 | */ |
1721 | static void try_to_wake_up_local(struct task_struct *p) | |
1722 | { | |
1723 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1724 | |
383efcd0 TH |
1725 | if (WARN_ON_ONCE(rq != this_rq()) || |
1726 | WARN_ON_ONCE(p == current)) | |
1727 | return; | |
1728 | ||
21aa9af0 TH |
1729 | lockdep_assert_held(&rq->lock); |
1730 | ||
2acca55e PZ |
1731 | if (!raw_spin_trylock(&p->pi_lock)) { |
1732 | raw_spin_unlock(&rq->lock); | |
1733 | raw_spin_lock(&p->pi_lock); | |
1734 | raw_spin_lock(&rq->lock); | |
1735 | } | |
1736 | ||
21aa9af0 | 1737 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1738 | goto out; |
21aa9af0 | 1739 | |
fd2f4419 | 1740 | if (!p->on_rq) |
d7c01d27 PZ |
1741 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1742 | ||
23f41eeb | 1743 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1744 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1745 | out: |
1746 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1747 | } |
1748 | ||
50fa610a DH |
1749 | /** |
1750 | * wake_up_process - Wake up a specific process | |
1751 | * @p: The process to be woken up. | |
1752 | * | |
1753 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1754 | * processes. |
1755 | * | |
1756 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1757 | * |
1758 | * It may be assumed that this function implies a write memory barrier before | |
1759 | * changing the task state if and only if any tasks are woken up. | |
1760 | */ | |
7ad5b3a5 | 1761 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1762 | { |
9067ac85 ON |
1763 | WARN_ON(task_is_stopped_or_traced(p)); |
1764 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1765 | } |
1da177e4 LT |
1766 | EXPORT_SYMBOL(wake_up_process); |
1767 | ||
7ad5b3a5 | 1768 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1769 | { |
1770 | return try_to_wake_up(p, state, 0); | |
1771 | } | |
1772 | ||
1da177e4 LT |
1773 | /* |
1774 | * Perform scheduler related setup for a newly forked process p. | |
1775 | * p is forked by current. | |
dd41f596 IM |
1776 | * |
1777 | * __sched_fork() is basic setup used by init_idle() too: | |
1778 | */ | |
5e1576ed | 1779 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1780 | { |
fd2f4419 PZ |
1781 | p->on_rq = 0; |
1782 | ||
1783 | p->se.on_rq = 0; | |
dd41f596 IM |
1784 | p->se.exec_start = 0; |
1785 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1786 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1787 | p->se.nr_migrations = 0; |
da7a735e | 1788 | p->se.vruntime = 0; |
fd2f4419 | 1789 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1790 | |
1791 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1792 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1793 | #endif |
476d139c | 1794 | |
aab03e05 DF |
1795 | RB_CLEAR_NODE(&p->dl.rb_node); |
1796 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1797 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1798 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1799 | p->dl.dl_period = 0; |
aab03e05 DF |
1800 | p->dl.flags = 0; |
1801 | ||
fa717060 | 1802 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1803 | |
e107be36 AK |
1804 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1805 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1806 | #endif | |
cbee9f88 PZ |
1807 | |
1808 | #ifdef CONFIG_NUMA_BALANCING | |
1809 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1810 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1811 | p->mm->numa_scan_seq = 0; |
1812 | } | |
1813 | ||
5e1576ed RR |
1814 | if (clone_flags & CLONE_VM) |
1815 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1816 | else | |
1817 | p->numa_preferred_nid = -1; | |
1818 | ||
cbee9f88 PZ |
1819 | p->node_stamp = 0ULL; |
1820 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1821 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1822 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1823 | p->numa_faults_memory = NULL; |
1824 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1825 | p->last_task_numa_placement = 0; |
1826 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1827 | |
1828 | INIT_LIST_HEAD(&p->numa_entry); | |
1829 | p->numa_group = NULL; | |
cbee9f88 | 1830 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1831 | } |
1832 | ||
1a687c2e | 1833 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1834 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1835 | void set_numabalancing_state(bool enabled) |
1836 | { | |
1837 | if (enabled) | |
1838 | sched_feat_set("NUMA"); | |
1839 | else | |
1840 | sched_feat_set("NO_NUMA"); | |
1841 | } | |
3105b86a MG |
1842 | #else |
1843 | __read_mostly bool numabalancing_enabled; | |
1844 | ||
1845 | void set_numabalancing_state(bool enabled) | |
1846 | { | |
1847 | numabalancing_enabled = enabled; | |
dd41f596 | 1848 | } |
3105b86a | 1849 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1850 | |
1851 | #ifdef CONFIG_PROC_SYSCTL | |
1852 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1853 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1854 | { | |
1855 | struct ctl_table t; | |
1856 | int err; | |
1857 | int state = numabalancing_enabled; | |
1858 | ||
1859 | if (write && !capable(CAP_SYS_ADMIN)) | |
1860 | return -EPERM; | |
1861 | ||
1862 | t = *table; | |
1863 | t.data = &state; | |
1864 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1865 | if (err < 0) | |
1866 | return err; | |
1867 | if (write) | |
1868 | set_numabalancing_state(state); | |
1869 | return err; | |
1870 | } | |
1871 | #endif | |
1872 | #endif | |
dd41f596 IM |
1873 | |
1874 | /* | |
1875 | * fork()/clone()-time setup: | |
1876 | */ | |
aab03e05 | 1877 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1878 | { |
0122ec5b | 1879 | unsigned long flags; |
dd41f596 IM |
1880 | int cpu = get_cpu(); |
1881 | ||
5e1576ed | 1882 | __sched_fork(clone_flags, p); |
06b83b5f | 1883 | /* |
0017d735 | 1884 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1885 | * nobody will actually run it, and a signal or other external |
1886 | * event cannot wake it up and insert it on the runqueue either. | |
1887 | */ | |
0017d735 | 1888 | p->state = TASK_RUNNING; |
dd41f596 | 1889 | |
c350a04e MG |
1890 | /* |
1891 | * Make sure we do not leak PI boosting priority to the child. | |
1892 | */ | |
1893 | p->prio = current->normal_prio; | |
1894 | ||
b9dc29e7 MG |
1895 | /* |
1896 | * Revert to default priority/policy on fork if requested. | |
1897 | */ | |
1898 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1899 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1900 | p->policy = SCHED_NORMAL; |
6c697bdf | 1901 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1902 | p->rt_priority = 0; |
1903 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1904 | p->static_prio = NICE_TO_PRIO(0); | |
1905 | ||
1906 | p->prio = p->normal_prio = __normal_prio(p); | |
1907 | set_load_weight(p); | |
6c697bdf | 1908 | |
b9dc29e7 MG |
1909 | /* |
1910 | * We don't need the reset flag anymore after the fork. It has | |
1911 | * fulfilled its duty: | |
1912 | */ | |
1913 | p->sched_reset_on_fork = 0; | |
1914 | } | |
ca94c442 | 1915 | |
aab03e05 DF |
1916 | if (dl_prio(p->prio)) { |
1917 | put_cpu(); | |
1918 | return -EAGAIN; | |
1919 | } else if (rt_prio(p->prio)) { | |
1920 | p->sched_class = &rt_sched_class; | |
1921 | } else { | |
2ddbf952 | 1922 | p->sched_class = &fair_sched_class; |
aab03e05 | 1923 | } |
b29739f9 | 1924 | |
cd29fe6f PZ |
1925 | if (p->sched_class->task_fork) |
1926 | p->sched_class->task_fork(p); | |
1927 | ||
86951599 PZ |
1928 | /* |
1929 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1930 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1931 | * is ran before sched_fork(). | |
1932 | * | |
1933 | * Silence PROVE_RCU. | |
1934 | */ | |
0122ec5b | 1935 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1936 | set_task_cpu(p, cpu); |
0122ec5b | 1937 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1938 | |
52f17b6c | 1939 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1940 | if (likely(sched_info_on())) |
52f17b6c | 1941 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1942 | #endif |
3ca7a440 PZ |
1943 | #if defined(CONFIG_SMP) |
1944 | p->on_cpu = 0; | |
4866cde0 | 1945 | #endif |
01028747 | 1946 | init_task_preempt_count(p); |
806c09a7 | 1947 | #ifdef CONFIG_SMP |
917b627d | 1948 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1949 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1950 | #endif |
917b627d | 1951 | |
476d139c | 1952 | put_cpu(); |
aab03e05 | 1953 | return 0; |
1da177e4 LT |
1954 | } |
1955 | ||
332ac17e DF |
1956 | unsigned long to_ratio(u64 period, u64 runtime) |
1957 | { | |
1958 | if (runtime == RUNTIME_INF) | |
1959 | return 1ULL << 20; | |
1960 | ||
1961 | /* | |
1962 | * Doing this here saves a lot of checks in all | |
1963 | * the calling paths, and returning zero seems | |
1964 | * safe for them anyway. | |
1965 | */ | |
1966 | if (period == 0) | |
1967 | return 0; | |
1968 | ||
1969 | return div64_u64(runtime << 20, period); | |
1970 | } | |
1971 | ||
1972 | #ifdef CONFIG_SMP | |
1973 | inline struct dl_bw *dl_bw_of(int i) | |
1974 | { | |
1975 | return &cpu_rq(i)->rd->dl_bw; | |
1976 | } | |
1977 | ||
de212f18 | 1978 | static inline int dl_bw_cpus(int i) |
332ac17e | 1979 | { |
de212f18 PZ |
1980 | struct root_domain *rd = cpu_rq(i)->rd; |
1981 | int cpus = 0; | |
1982 | ||
1983 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1984 | cpus++; | |
1985 | ||
1986 | return cpus; | |
332ac17e DF |
1987 | } |
1988 | #else | |
1989 | inline struct dl_bw *dl_bw_of(int i) | |
1990 | { | |
1991 | return &cpu_rq(i)->dl.dl_bw; | |
1992 | } | |
1993 | ||
de212f18 | 1994 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
1995 | { |
1996 | return 1; | |
1997 | } | |
1998 | #endif | |
1999 | ||
2000 | static inline | |
2001 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
2002 | { | |
2003 | dl_b->total_bw -= tsk_bw; | |
2004 | } | |
2005 | ||
2006 | static inline | |
2007 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
2008 | { | |
2009 | dl_b->total_bw += tsk_bw; | |
2010 | } | |
2011 | ||
2012 | static inline | |
2013 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
2014 | { | |
2015 | return dl_b->bw != -1 && | |
2016 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
2017 | } | |
2018 | ||
2019 | /* | |
2020 | * We must be sure that accepting a new task (or allowing changing the | |
2021 | * parameters of an existing one) is consistent with the bandwidth | |
2022 | * constraints. If yes, this function also accordingly updates the currently | |
2023 | * allocated bandwidth to reflect the new situation. | |
2024 | * | |
2025 | * This function is called while holding p's rq->lock. | |
2026 | */ | |
2027 | static int dl_overflow(struct task_struct *p, int policy, | |
2028 | const struct sched_attr *attr) | |
2029 | { | |
2030 | ||
2031 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2032 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2033 | u64 runtime = attr->sched_runtime; |
2034 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2035 | int cpus, err = -1; |
332ac17e DF |
2036 | |
2037 | if (new_bw == p->dl.dl_bw) | |
2038 | return 0; | |
2039 | ||
2040 | /* | |
2041 | * Either if a task, enters, leave, or stays -deadline but changes | |
2042 | * its parameters, we may need to update accordingly the total | |
2043 | * allocated bandwidth of the container. | |
2044 | */ | |
2045 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2046 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2047 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2048 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2049 | __dl_add(dl_b, new_bw); | |
2050 | err = 0; | |
2051 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2052 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2053 | __dl_clear(dl_b, p->dl.dl_bw); | |
2054 | __dl_add(dl_b, new_bw); | |
2055 | err = 0; | |
2056 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2057 | __dl_clear(dl_b, p->dl.dl_bw); | |
2058 | err = 0; | |
2059 | } | |
2060 | raw_spin_unlock(&dl_b->lock); | |
2061 | ||
2062 | return err; | |
2063 | } | |
2064 | ||
2065 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2066 | ||
1da177e4 LT |
2067 | /* |
2068 | * wake_up_new_task - wake up a newly created task for the first time. | |
2069 | * | |
2070 | * This function will do some initial scheduler statistics housekeeping | |
2071 | * that must be done for every newly created context, then puts the task | |
2072 | * on the runqueue and wakes it. | |
2073 | */ | |
3e51e3ed | 2074 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2075 | { |
2076 | unsigned long flags; | |
dd41f596 | 2077 | struct rq *rq; |
fabf318e | 2078 | |
ab2515c4 | 2079 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2080 | #ifdef CONFIG_SMP |
2081 | /* | |
2082 | * Fork balancing, do it here and not earlier because: | |
2083 | * - cpus_allowed can change in the fork path | |
2084 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2085 | */ |
ac66f547 | 2086 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2087 | #endif |
2088 | ||
a75cdaa9 AS |
2089 | /* Initialize new task's runnable average */ |
2090 | init_task_runnable_average(p); | |
ab2515c4 | 2091 | rq = __task_rq_lock(p); |
cd29fe6f | 2092 | activate_task(rq, p, 0); |
fd2f4419 | 2093 | p->on_rq = 1; |
89363381 | 2094 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2095 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2096 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2097 | if (p->sched_class->task_woken) |
2098 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2099 | #endif |
0122ec5b | 2100 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2101 | } |
2102 | ||
e107be36 AK |
2103 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2104 | ||
2105 | /** | |
80dd99b3 | 2106 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2107 | * @notifier: notifier struct to register |
e107be36 AK |
2108 | */ |
2109 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2110 | { | |
2111 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2112 | } | |
2113 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2114 | ||
2115 | /** | |
2116 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2117 | * @notifier: notifier struct to unregister |
e107be36 AK |
2118 | * |
2119 | * This is safe to call from within a preemption notifier. | |
2120 | */ | |
2121 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2122 | { | |
2123 | hlist_del(¬ifier->link); | |
2124 | } | |
2125 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2126 | ||
2127 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2128 | { | |
2129 | struct preempt_notifier *notifier; | |
e107be36 | 2130 | |
b67bfe0d | 2131 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2132 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2133 | } | |
2134 | ||
2135 | static void | |
2136 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2137 | struct task_struct *next) | |
2138 | { | |
2139 | struct preempt_notifier *notifier; | |
e107be36 | 2140 | |
b67bfe0d | 2141 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2142 | notifier->ops->sched_out(notifier, next); |
2143 | } | |
2144 | ||
6d6bc0ad | 2145 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2146 | |
2147 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2148 | { | |
2149 | } | |
2150 | ||
2151 | static void | |
2152 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2153 | struct task_struct *next) | |
2154 | { | |
2155 | } | |
2156 | ||
6d6bc0ad | 2157 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2158 | |
4866cde0 NP |
2159 | /** |
2160 | * prepare_task_switch - prepare to switch tasks | |
2161 | * @rq: the runqueue preparing to switch | |
421cee29 | 2162 | * @prev: the current task that is being switched out |
4866cde0 NP |
2163 | * @next: the task we are going to switch to. |
2164 | * | |
2165 | * This is called with the rq lock held and interrupts off. It must | |
2166 | * be paired with a subsequent finish_task_switch after the context | |
2167 | * switch. | |
2168 | * | |
2169 | * prepare_task_switch sets up locking and calls architecture specific | |
2170 | * hooks. | |
2171 | */ | |
e107be36 AK |
2172 | static inline void |
2173 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2174 | struct task_struct *next) | |
4866cde0 | 2175 | { |
895dd92c | 2176 | trace_sched_switch(prev, next); |
43148951 | 2177 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2178 | perf_event_task_sched_out(prev, next); |
e107be36 | 2179 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2180 | prepare_lock_switch(rq, next); |
2181 | prepare_arch_switch(next); | |
2182 | } | |
2183 | ||
1da177e4 LT |
2184 | /** |
2185 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2186 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2187 | * @prev: the thread we just switched away from. |
2188 | * | |
4866cde0 NP |
2189 | * finish_task_switch must be called after the context switch, paired |
2190 | * with a prepare_task_switch call before the context switch. | |
2191 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2192 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2193 | * |
2194 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2195 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2196 | * with the lock held can cause deadlocks; see schedule() for |
2197 | * details.) | |
2198 | */ | |
a9957449 | 2199 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2200 | __releases(rq->lock) |
2201 | { | |
1da177e4 | 2202 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2203 | long prev_state; |
1da177e4 LT |
2204 | |
2205 | rq->prev_mm = NULL; | |
2206 | ||
2207 | /* | |
2208 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2209 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2210 | * schedule one last time. The schedule call will never return, and |
2211 | * the scheduled task must drop that reference. | |
c394cc9f | 2212 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2213 | * still held, otherwise prev could be scheduled on another cpu, die |
2214 | * there before we look at prev->state, and then the reference would | |
2215 | * be dropped twice. | |
2216 | * Manfred Spraul <manfred@colorfullife.com> | |
2217 | */ | |
55a101f8 | 2218 | prev_state = prev->state; |
bf9fae9f | 2219 | vtime_task_switch(prev); |
4866cde0 | 2220 | finish_arch_switch(prev); |
a8d757ef | 2221 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2222 | finish_lock_switch(rq, prev); |
01f23e16 | 2223 | finish_arch_post_lock_switch(); |
e8fa1362 | 2224 | |
e107be36 | 2225 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2226 | if (mm) |
2227 | mmdrop(mm); | |
c394cc9f | 2228 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2229 | if (prev->sched_class->task_dead) |
2230 | prev->sched_class->task_dead(prev); | |
2231 | ||
c6fd91f0 | 2232 | /* |
2233 | * Remove function-return probe instances associated with this | |
2234 | * task and put them back on the free list. | |
9761eea8 | 2235 | */ |
c6fd91f0 | 2236 | kprobe_flush_task(prev); |
1da177e4 | 2237 | put_task_struct(prev); |
c6fd91f0 | 2238 | } |
99e5ada9 FW |
2239 | |
2240 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2241 | } |
2242 | ||
3f029d3c GH |
2243 | #ifdef CONFIG_SMP |
2244 | ||
3f029d3c GH |
2245 | /* rq->lock is NOT held, but preemption is disabled */ |
2246 | static inline void post_schedule(struct rq *rq) | |
2247 | { | |
2248 | if (rq->post_schedule) { | |
2249 | unsigned long flags; | |
2250 | ||
05fa785c | 2251 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2252 | if (rq->curr->sched_class->post_schedule) |
2253 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2254 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2255 | |
2256 | rq->post_schedule = 0; | |
2257 | } | |
2258 | } | |
2259 | ||
2260 | #else | |
da19ab51 | 2261 | |
3f029d3c GH |
2262 | static inline void post_schedule(struct rq *rq) |
2263 | { | |
1da177e4 LT |
2264 | } |
2265 | ||
3f029d3c GH |
2266 | #endif |
2267 | ||
1da177e4 LT |
2268 | /** |
2269 | * schedule_tail - first thing a freshly forked thread must call. | |
2270 | * @prev: the thread we just switched away from. | |
2271 | */ | |
722a9f92 | 2272 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2273 | __releases(rq->lock) |
2274 | { | |
70b97a7f IM |
2275 | struct rq *rq = this_rq(); |
2276 | ||
4866cde0 | 2277 | finish_task_switch(rq, prev); |
da19ab51 | 2278 | |
3f029d3c GH |
2279 | /* |
2280 | * FIXME: do we need to worry about rq being invalidated by the | |
2281 | * task_switch? | |
2282 | */ | |
2283 | post_schedule(rq); | |
70b97a7f | 2284 | |
4866cde0 NP |
2285 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2286 | /* In this case, finish_task_switch does not reenable preemption */ | |
2287 | preempt_enable(); | |
2288 | #endif | |
1da177e4 | 2289 | if (current->set_child_tid) |
b488893a | 2290 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2291 | } |
2292 | ||
2293 | /* | |
2294 | * context_switch - switch to the new MM and the new | |
2295 | * thread's register state. | |
2296 | */ | |
dd41f596 | 2297 | static inline void |
70b97a7f | 2298 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2299 | struct task_struct *next) |
1da177e4 | 2300 | { |
dd41f596 | 2301 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2302 | |
e107be36 | 2303 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2304 | |
dd41f596 IM |
2305 | mm = next->mm; |
2306 | oldmm = prev->active_mm; | |
9226d125 ZA |
2307 | /* |
2308 | * For paravirt, this is coupled with an exit in switch_to to | |
2309 | * combine the page table reload and the switch backend into | |
2310 | * one hypercall. | |
2311 | */ | |
224101ed | 2312 | arch_start_context_switch(prev); |
9226d125 | 2313 | |
31915ab4 | 2314 | if (!mm) { |
1da177e4 LT |
2315 | next->active_mm = oldmm; |
2316 | atomic_inc(&oldmm->mm_count); | |
2317 | enter_lazy_tlb(oldmm, next); | |
2318 | } else | |
2319 | switch_mm(oldmm, mm, next); | |
2320 | ||
31915ab4 | 2321 | if (!prev->mm) { |
1da177e4 | 2322 | prev->active_mm = NULL; |
1da177e4 LT |
2323 | rq->prev_mm = oldmm; |
2324 | } | |
3a5f5e48 IM |
2325 | /* |
2326 | * Since the runqueue lock will be released by the next | |
2327 | * task (which is an invalid locking op but in the case | |
2328 | * of the scheduler it's an obvious special-case), so we | |
2329 | * do an early lockdep release here: | |
2330 | */ | |
2331 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2332 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2333 | #endif |
1da177e4 | 2334 | |
91d1aa43 | 2335 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2336 | /* Here we just switch the register state and the stack. */ |
2337 | switch_to(prev, next, prev); | |
2338 | ||
dd41f596 IM |
2339 | barrier(); |
2340 | /* | |
2341 | * this_rq must be evaluated again because prev may have moved | |
2342 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2343 | * frame will be invalid. | |
2344 | */ | |
2345 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2346 | } |
2347 | ||
2348 | /* | |
1c3e8264 | 2349 | * nr_running and nr_context_switches: |
1da177e4 LT |
2350 | * |
2351 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2352 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2353 | */ |
2354 | unsigned long nr_running(void) | |
2355 | { | |
2356 | unsigned long i, sum = 0; | |
2357 | ||
2358 | for_each_online_cpu(i) | |
2359 | sum += cpu_rq(i)->nr_running; | |
2360 | ||
2361 | return sum; | |
f711f609 | 2362 | } |
1da177e4 | 2363 | |
1da177e4 | 2364 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2365 | { |
cc94abfc SR |
2366 | int i; |
2367 | unsigned long long sum = 0; | |
46cb4b7c | 2368 | |
0a945022 | 2369 | for_each_possible_cpu(i) |
1da177e4 | 2370 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2371 | |
1da177e4 LT |
2372 | return sum; |
2373 | } | |
483b4ee6 | 2374 | |
1da177e4 LT |
2375 | unsigned long nr_iowait(void) |
2376 | { | |
2377 | unsigned long i, sum = 0; | |
483b4ee6 | 2378 | |
0a945022 | 2379 | for_each_possible_cpu(i) |
1da177e4 | 2380 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2381 | |
1da177e4 LT |
2382 | return sum; |
2383 | } | |
483b4ee6 | 2384 | |
8c215bd3 | 2385 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2386 | { |
8c215bd3 | 2387 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2388 | return atomic_read(&this->nr_iowait); |
2389 | } | |
46cb4b7c | 2390 | |
dd41f596 | 2391 | #ifdef CONFIG_SMP |
8a0be9ef | 2392 | |
46cb4b7c | 2393 | /* |
38022906 PZ |
2394 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2395 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2396 | */ |
38022906 | 2397 | void sched_exec(void) |
46cb4b7c | 2398 | { |
38022906 | 2399 | struct task_struct *p = current; |
1da177e4 | 2400 | unsigned long flags; |
0017d735 | 2401 | int dest_cpu; |
46cb4b7c | 2402 | |
8f42ced9 | 2403 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2404 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2405 | if (dest_cpu == smp_processor_id()) |
2406 | goto unlock; | |
38022906 | 2407 | |
8f42ced9 | 2408 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2409 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2410 | |
8f42ced9 PZ |
2411 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2412 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2413 | return; |
2414 | } | |
0017d735 | 2415 | unlock: |
8f42ced9 | 2416 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2417 | } |
dd41f596 | 2418 | |
1da177e4 LT |
2419 | #endif |
2420 | ||
1da177e4 | 2421 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2422 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2423 | |
2424 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2425 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2426 | |
2427 | /* | |
c5f8d995 | 2428 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2429 | * @p in case that task is currently running. |
c5f8d995 HS |
2430 | * |
2431 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2432 | */ |
c5f8d995 HS |
2433 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2434 | { | |
2435 | u64 ns = 0; | |
2436 | ||
4036ac15 MG |
2437 | /* |
2438 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2439 | * project cycles that may never be accounted to this | |
2440 | * thread, breaking clock_gettime(). | |
2441 | */ | |
2442 | if (task_current(rq, p) && p->on_rq) { | |
c5f8d995 | 2443 | update_rq_clock(rq); |
78becc27 | 2444 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2445 | if ((s64)ns < 0) |
2446 | ns = 0; | |
2447 | } | |
2448 | ||
2449 | return ns; | |
2450 | } | |
2451 | ||
bb34d92f | 2452 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2453 | { |
1da177e4 | 2454 | unsigned long flags; |
41b86e9c | 2455 | struct rq *rq; |
bb34d92f | 2456 | u64 ns = 0; |
48f24c4d | 2457 | |
41b86e9c | 2458 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2459 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2460 | task_rq_unlock(rq, p, &flags); |
1508487e | 2461 | |
c5f8d995 HS |
2462 | return ns; |
2463 | } | |
f06febc9 | 2464 | |
c5f8d995 HS |
2465 | /* |
2466 | * Return accounted runtime for the task. | |
2467 | * In case the task is currently running, return the runtime plus current's | |
2468 | * pending runtime that have not been accounted yet. | |
2469 | */ | |
2470 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2471 | { | |
2472 | unsigned long flags; | |
2473 | struct rq *rq; | |
2474 | u64 ns = 0; | |
2475 | ||
911b2898 PZ |
2476 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2477 | /* | |
2478 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2479 | * So we have a optimization chance when the task's delta_exec is 0. | |
2480 | * Reading ->on_cpu is racy, but this is ok. | |
2481 | * | |
2482 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2483 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2484 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2485 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2486 | * been accounted, so we're correct here as well. | |
911b2898 | 2487 | */ |
4036ac15 | 2488 | if (!p->on_cpu || !p->on_rq) |
911b2898 PZ |
2489 | return p->se.sum_exec_runtime; |
2490 | #endif | |
2491 | ||
c5f8d995 HS |
2492 | rq = task_rq_lock(p, &flags); |
2493 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2494 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2495 | |
2496 | return ns; | |
2497 | } | |
48f24c4d | 2498 | |
7835b98b CL |
2499 | /* |
2500 | * This function gets called by the timer code, with HZ frequency. | |
2501 | * We call it with interrupts disabled. | |
7835b98b CL |
2502 | */ |
2503 | void scheduler_tick(void) | |
2504 | { | |
7835b98b CL |
2505 | int cpu = smp_processor_id(); |
2506 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2507 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2508 | |
2509 | sched_clock_tick(); | |
dd41f596 | 2510 | |
05fa785c | 2511 | raw_spin_lock(&rq->lock); |
3e51f33f | 2512 | update_rq_clock(rq); |
fa85ae24 | 2513 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2514 | update_cpu_load_active(rq); |
05fa785c | 2515 | raw_spin_unlock(&rq->lock); |
7835b98b | 2516 | |
e9d2b064 | 2517 | perf_event_task_tick(); |
e220d2dc | 2518 | |
e418e1c2 | 2519 | #ifdef CONFIG_SMP |
6eb57e0d | 2520 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2521 | trigger_load_balance(rq); |
e418e1c2 | 2522 | #endif |
265f22a9 | 2523 | rq_last_tick_reset(rq); |
1da177e4 LT |
2524 | } |
2525 | ||
265f22a9 FW |
2526 | #ifdef CONFIG_NO_HZ_FULL |
2527 | /** | |
2528 | * scheduler_tick_max_deferment | |
2529 | * | |
2530 | * Keep at least one tick per second when a single | |
2531 | * active task is running because the scheduler doesn't | |
2532 | * yet completely support full dynticks environment. | |
2533 | * | |
2534 | * This makes sure that uptime, CFS vruntime, load | |
2535 | * balancing, etc... continue to move forward, even | |
2536 | * with a very low granularity. | |
e69f6186 YB |
2537 | * |
2538 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2539 | */ |
2540 | u64 scheduler_tick_max_deferment(void) | |
2541 | { | |
2542 | struct rq *rq = this_rq(); | |
2543 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2544 | ||
2545 | next = rq->last_sched_tick + HZ; | |
2546 | ||
2547 | if (time_before_eq(next, now)) | |
2548 | return 0; | |
2549 | ||
8fe8ff09 | 2550 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2551 | } |
265f22a9 | 2552 | #endif |
1da177e4 | 2553 | |
132380a0 | 2554 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2555 | { |
2556 | if (in_lock_functions(addr)) { | |
2557 | addr = CALLER_ADDR2; | |
2558 | if (in_lock_functions(addr)) | |
2559 | addr = CALLER_ADDR3; | |
2560 | } | |
2561 | return addr; | |
2562 | } | |
1da177e4 | 2563 | |
7e49fcce SR |
2564 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2565 | defined(CONFIG_PREEMPT_TRACER)) | |
2566 | ||
edafe3a5 | 2567 | void preempt_count_add(int val) |
1da177e4 | 2568 | { |
6cd8a4bb | 2569 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2570 | /* |
2571 | * Underflow? | |
2572 | */ | |
9a11b49a IM |
2573 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2574 | return; | |
6cd8a4bb | 2575 | #endif |
bdb43806 | 2576 | __preempt_count_add(val); |
6cd8a4bb | 2577 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2578 | /* |
2579 | * Spinlock count overflowing soon? | |
2580 | */ | |
33859f7f MOS |
2581 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2582 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2583 | #endif |
8f47b187 TG |
2584 | if (preempt_count() == val) { |
2585 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2586 | #ifdef CONFIG_DEBUG_PREEMPT | |
2587 | current->preempt_disable_ip = ip; | |
2588 | #endif | |
2589 | trace_preempt_off(CALLER_ADDR0, ip); | |
2590 | } | |
1da177e4 | 2591 | } |
bdb43806 | 2592 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2593 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2594 | |
edafe3a5 | 2595 | void preempt_count_sub(int val) |
1da177e4 | 2596 | { |
6cd8a4bb | 2597 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2598 | /* |
2599 | * Underflow? | |
2600 | */ | |
01e3eb82 | 2601 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2602 | return; |
1da177e4 LT |
2603 | /* |
2604 | * Is the spinlock portion underflowing? | |
2605 | */ | |
9a11b49a IM |
2606 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2607 | !(preempt_count() & PREEMPT_MASK))) | |
2608 | return; | |
6cd8a4bb | 2609 | #endif |
9a11b49a | 2610 | |
6cd8a4bb SR |
2611 | if (preempt_count() == val) |
2612 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2613 | __preempt_count_sub(val); |
1da177e4 | 2614 | } |
bdb43806 | 2615 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2616 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2617 | |
2618 | #endif | |
2619 | ||
2620 | /* | |
dd41f596 | 2621 | * Print scheduling while atomic bug: |
1da177e4 | 2622 | */ |
dd41f596 | 2623 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2624 | { |
664dfa65 DJ |
2625 | if (oops_in_progress) |
2626 | return; | |
2627 | ||
3df0fc5b PZ |
2628 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2629 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2630 | |
dd41f596 | 2631 | debug_show_held_locks(prev); |
e21f5b15 | 2632 | print_modules(); |
dd41f596 IM |
2633 | if (irqs_disabled()) |
2634 | print_irqtrace_events(prev); | |
8f47b187 TG |
2635 | #ifdef CONFIG_DEBUG_PREEMPT |
2636 | if (in_atomic_preempt_off()) { | |
2637 | pr_err("Preemption disabled at:"); | |
2638 | print_ip_sym(current->preempt_disable_ip); | |
2639 | pr_cont("\n"); | |
2640 | } | |
2641 | #endif | |
6135fc1e | 2642 | dump_stack(); |
373d4d09 | 2643 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2644 | } |
1da177e4 | 2645 | |
dd41f596 IM |
2646 | /* |
2647 | * Various schedule()-time debugging checks and statistics: | |
2648 | */ | |
2649 | static inline void schedule_debug(struct task_struct *prev) | |
2650 | { | |
1da177e4 | 2651 | /* |
41a2d6cf | 2652 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2653 | * schedule() atomically, we ignore that path. Otherwise whine |
2654 | * if we are scheduling when we should not. | |
1da177e4 | 2655 | */ |
192301e7 | 2656 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2657 | __schedule_bug(prev); |
b3fbab05 | 2658 | rcu_sleep_check(); |
dd41f596 | 2659 | |
1da177e4 LT |
2660 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2661 | ||
2d72376b | 2662 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2663 | } |
2664 | ||
2665 | /* | |
2666 | * Pick up the highest-prio task: | |
2667 | */ | |
2668 | static inline struct task_struct * | |
606dba2e | 2669 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2670 | { |
37e117c0 | 2671 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2672 | struct task_struct *p; |
1da177e4 LT |
2673 | |
2674 | /* | |
dd41f596 IM |
2675 | * Optimization: we know that if all tasks are in |
2676 | * the fair class we can call that function directly: | |
1da177e4 | 2677 | */ |
37e117c0 | 2678 | if (likely(prev->sched_class == class && |
38033c37 | 2679 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2680 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2681 | if (unlikely(p == RETRY_TASK)) |
2682 | goto again; | |
2683 | ||
2684 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2685 | if (unlikely(!p)) | |
2686 | p = idle_sched_class.pick_next_task(rq, prev); | |
2687 | ||
2688 | return p; | |
1da177e4 LT |
2689 | } |
2690 | ||
37e117c0 | 2691 | again: |
34f971f6 | 2692 | for_each_class(class) { |
606dba2e | 2693 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2694 | if (p) { |
2695 | if (unlikely(p == RETRY_TASK)) | |
2696 | goto again; | |
dd41f596 | 2697 | return p; |
37e117c0 | 2698 | } |
dd41f596 | 2699 | } |
34f971f6 PZ |
2700 | |
2701 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2702 | } |
1da177e4 | 2703 | |
dd41f596 | 2704 | /* |
c259e01a | 2705 | * __schedule() is the main scheduler function. |
edde96ea PE |
2706 | * |
2707 | * The main means of driving the scheduler and thus entering this function are: | |
2708 | * | |
2709 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2710 | * | |
2711 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2712 | * paths. For example, see arch/x86/entry_64.S. | |
2713 | * | |
2714 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2715 | * interrupt handler scheduler_tick(). | |
2716 | * | |
2717 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2718 | * task to the run-queue and that's it. | |
2719 | * | |
2720 | * Now, if the new task added to the run-queue preempts the current | |
2721 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2722 | * called on the nearest possible occasion: | |
2723 | * | |
2724 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2725 | * | |
2726 | * - in syscall or exception context, at the next outmost | |
2727 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2728 | * spin_unlock()!) | |
2729 | * | |
2730 | * - in IRQ context, return from interrupt-handler to | |
2731 | * preemptible context | |
2732 | * | |
2733 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2734 | * then at the next: | |
2735 | * | |
2736 | * - cond_resched() call | |
2737 | * - explicit schedule() call | |
2738 | * - return from syscall or exception to user-space | |
2739 | * - return from interrupt-handler to user-space | |
dd41f596 | 2740 | */ |
c259e01a | 2741 | static void __sched __schedule(void) |
dd41f596 IM |
2742 | { |
2743 | struct task_struct *prev, *next; | |
67ca7bde | 2744 | unsigned long *switch_count; |
dd41f596 | 2745 | struct rq *rq; |
31656519 | 2746 | int cpu; |
dd41f596 | 2747 | |
ff743345 PZ |
2748 | need_resched: |
2749 | preempt_disable(); | |
dd41f596 IM |
2750 | cpu = smp_processor_id(); |
2751 | rq = cpu_rq(cpu); | |
25502a6c | 2752 | rcu_note_context_switch(cpu); |
dd41f596 | 2753 | prev = rq->curr; |
dd41f596 | 2754 | |
dd41f596 | 2755 | schedule_debug(prev); |
1da177e4 | 2756 | |
31656519 | 2757 | if (sched_feat(HRTICK)) |
f333fdc9 | 2758 | hrtick_clear(rq); |
8f4d37ec | 2759 | |
e0acd0a6 ON |
2760 | /* |
2761 | * Make sure that signal_pending_state()->signal_pending() below | |
2762 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2763 | * done by the caller to avoid the race with signal_wake_up(). | |
2764 | */ | |
2765 | smp_mb__before_spinlock(); | |
05fa785c | 2766 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2767 | |
246d86b5 | 2768 | switch_count = &prev->nivcsw; |
1da177e4 | 2769 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2770 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2771 | prev->state = TASK_RUNNING; |
21aa9af0 | 2772 | } else { |
2acca55e PZ |
2773 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2774 | prev->on_rq = 0; | |
2775 | ||
21aa9af0 | 2776 | /* |
2acca55e PZ |
2777 | * If a worker went to sleep, notify and ask workqueue |
2778 | * whether it wants to wake up a task to maintain | |
2779 | * concurrency. | |
21aa9af0 TH |
2780 | */ |
2781 | if (prev->flags & PF_WQ_WORKER) { | |
2782 | struct task_struct *to_wakeup; | |
2783 | ||
2784 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2785 | if (to_wakeup) | |
2786 | try_to_wake_up_local(to_wakeup); | |
2787 | } | |
21aa9af0 | 2788 | } |
dd41f596 | 2789 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2790 | } |
2791 | ||
606dba2e PZ |
2792 | if (prev->on_rq || rq->skip_clock_update < 0) |
2793 | update_rq_clock(rq); | |
2794 | ||
2795 | next = pick_next_task(rq, prev); | |
f26f9aff | 2796 | clear_tsk_need_resched(prev); |
f27dde8d | 2797 | clear_preempt_need_resched(); |
f26f9aff | 2798 | rq->skip_clock_update = 0; |
1da177e4 | 2799 | |
1da177e4 | 2800 | if (likely(prev != next)) { |
1da177e4 LT |
2801 | rq->nr_switches++; |
2802 | rq->curr = next; | |
2803 | ++*switch_count; | |
2804 | ||
dd41f596 | 2805 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2806 | /* |
246d86b5 ON |
2807 | * The context switch have flipped the stack from under us |
2808 | * and restored the local variables which were saved when | |
2809 | * this task called schedule() in the past. prev == current | |
2810 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2811 | */ |
2812 | cpu = smp_processor_id(); | |
2813 | rq = cpu_rq(cpu); | |
1da177e4 | 2814 | } else |
05fa785c | 2815 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2816 | |
3f029d3c | 2817 | post_schedule(rq); |
1da177e4 | 2818 | |
ba74c144 | 2819 | sched_preempt_enable_no_resched(); |
ff743345 | 2820 | if (need_resched()) |
1da177e4 LT |
2821 | goto need_resched; |
2822 | } | |
c259e01a | 2823 | |
9c40cef2 TG |
2824 | static inline void sched_submit_work(struct task_struct *tsk) |
2825 | { | |
3c7d5184 | 2826 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2827 | return; |
2828 | /* | |
2829 | * If we are going to sleep and we have plugged IO queued, | |
2830 | * make sure to submit it to avoid deadlocks. | |
2831 | */ | |
2832 | if (blk_needs_flush_plug(tsk)) | |
2833 | blk_schedule_flush_plug(tsk); | |
2834 | } | |
2835 | ||
722a9f92 | 2836 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2837 | { |
9c40cef2 TG |
2838 | struct task_struct *tsk = current; |
2839 | ||
2840 | sched_submit_work(tsk); | |
c259e01a TG |
2841 | __schedule(); |
2842 | } | |
1da177e4 LT |
2843 | EXPORT_SYMBOL(schedule); |
2844 | ||
91d1aa43 | 2845 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2846 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2847 | { |
2848 | /* | |
2849 | * If we come here after a random call to set_need_resched(), | |
2850 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2851 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2852 | * we find a better solution. | |
2853 | */ | |
91d1aa43 | 2854 | user_exit(); |
20ab65e3 | 2855 | schedule(); |
91d1aa43 | 2856 | user_enter(); |
20ab65e3 FW |
2857 | } |
2858 | #endif | |
2859 | ||
c5491ea7 TG |
2860 | /** |
2861 | * schedule_preempt_disabled - called with preemption disabled | |
2862 | * | |
2863 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2864 | */ | |
2865 | void __sched schedule_preempt_disabled(void) | |
2866 | { | |
ba74c144 | 2867 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2868 | schedule(); |
2869 | preempt_disable(); | |
2870 | } | |
2871 | ||
1da177e4 LT |
2872 | #ifdef CONFIG_PREEMPT |
2873 | /* | |
2ed6e34f | 2874 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2875 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2876 | * occur there and call schedule directly. |
2877 | */ | |
722a9f92 | 2878 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2879 | { |
1da177e4 LT |
2880 | /* |
2881 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2882 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2883 | */ |
fbb00b56 | 2884 | if (likely(!preemptible())) |
1da177e4 LT |
2885 | return; |
2886 | ||
3a5c359a | 2887 | do { |
bdb43806 | 2888 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2889 | __schedule(); |
bdb43806 | 2890 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2891 | |
3a5c359a AK |
2892 | /* |
2893 | * Check again in case we missed a preemption opportunity | |
2894 | * between schedule and now. | |
2895 | */ | |
2896 | barrier(); | |
5ed0cec0 | 2897 | } while (need_resched()); |
1da177e4 | 2898 | } |
376e2424 | 2899 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 2900 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2901 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2902 | |
2903 | /* | |
2ed6e34f | 2904 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2905 | * off of irq context. |
2906 | * Note, that this is called and return with irqs disabled. This will | |
2907 | * protect us against recursive calling from irq. | |
2908 | */ | |
722a9f92 | 2909 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2910 | { |
b22366cd | 2911 | enum ctx_state prev_state; |
6478d880 | 2912 | |
2ed6e34f | 2913 | /* Catch callers which need to be fixed */ |
f27dde8d | 2914 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2915 | |
b22366cd FW |
2916 | prev_state = exception_enter(); |
2917 | ||
3a5c359a | 2918 | do { |
bdb43806 | 2919 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2920 | local_irq_enable(); |
c259e01a | 2921 | __schedule(); |
3a5c359a | 2922 | local_irq_disable(); |
bdb43806 | 2923 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2924 | |
3a5c359a AK |
2925 | /* |
2926 | * Check again in case we missed a preemption opportunity | |
2927 | * between schedule and now. | |
2928 | */ | |
2929 | barrier(); | |
5ed0cec0 | 2930 | } while (need_resched()); |
b22366cd FW |
2931 | |
2932 | exception_exit(prev_state); | |
1da177e4 LT |
2933 | } |
2934 | ||
63859d4f | 2935 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2936 | void *key) |
1da177e4 | 2937 | { |
63859d4f | 2938 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2939 | } |
1da177e4 LT |
2940 | EXPORT_SYMBOL(default_wake_function); |
2941 | ||
b29739f9 IM |
2942 | #ifdef CONFIG_RT_MUTEXES |
2943 | ||
2944 | /* | |
2945 | * rt_mutex_setprio - set the current priority of a task | |
2946 | * @p: task | |
2947 | * @prio: prio value (kernel-internal form) | |
2948 | * | |
2949 | * This function changes the 'effective' priority of a task. It does | |
2950 | * not touch ->normal_prio like __setscheduler(). | |
2951 | * | |
c365c292 TG |
2952 | * Used by the rt_mutex code to implement priority inheritance |
2953 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2954 | */ |
36c8b586 | 2955 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2956 | { |
2d3d891d | 2957 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2958 | struct rq *rq; |
83ab0aa0 | 2959 | const struct sched_class *prev_class; |
b29739f9 | 2960 | |
aab03e05 | 2961 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2962 | |
0122ec5b | 2963 | rq = __task_rq_lock(p); |
b29739f9 | 2964 | |
1c4dd99b TG |
2965 | /* |
2966 | * Idle task boosting is a nono in general. There is one | |
2967 | * exception, when PREEMPT_RT and NOHZ is active: | |
2968 | * | |
2969 | * The idle task calls get_next_timer_interrupt() and holds | |
2970 | * the timer wheel base->lock on the CPU and another CPU wants | |
2971 | * to access the timer (probably to cancel it). We can safely | |
2972 | * ignore the boosting request, as the idle CPU runs this code | |
2973 | * with interrupts disabled and will complete the lock | |
2974 | * protected section without being interrupted. So there is no | |
2975 | * real need to boost. | |
2976 | */ | |
2977 | if (unlikely(p == rq->idle)) { | |
2978 | WARN_ON(p != rq->curr); | |
2979 | WARN_ON(p->pi_blocked_on); | |
2980 | goto out_unlock; | |
2981 | } | |
2982 | ||
a8027073 | 2983 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 2984 | oldprio = p->prio; |
83ab0aa0 | 2985 | prev_class = p->sched_class; |
fd2f4419 | 2986 | on_rq = p->on_rq; |
051a1d1a | 2987 | running = task_current(rq, p); |
0e1f3483 | 2988 | if (on_rq) |
69be72c1 | 2989 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2990 | if (running) |
2991 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2992 | |
2d3d891d DF |
2993 | /* |
2994 | * Boosting condition are: | |
2995 | * 1. -rt task is running and holds mutex A | |
2996 | * --> -dl task blocks on mutex A | |
2997 | * | |
2998 | * 2. -dl task is running and holds mutex A | |
2999 | * --> -dl task blocks on mutex A and could preempt the | |
3000 | * running task | |
3001 | */ | |
3002 | if (dl_prio(prio)) { | |
466af29b ON |
3003 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3004 | if (!dl_prio(p->normal_prio) || | |
3005 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d DF |
3006 | p->dl.dl_boosted = 1; |
3007 | p->dl.dl_throttled = 0; | |
3008 | enqueue_flag = ENQUEUE_REPLENISH; | |
3009 | } else | |
3010 | p->dl.dl_boosted = 0; | |
aab03e05 | 3011 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3012 | } else if (rt_prio(prio)) { |
3013 | if (dl_prio(oldprio)) | |
3014 | p->dl.dl_boosted = 0; | |
3015 | if (oldprio < prio) | |
3016 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3017 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3018 | } else { |
3019 | if (dl_prio(oldprio)) | |
3020 | p->dl.dl_boosted = 0; | |
dd41f596 | 3021 | p->sched_class = &fair_sched_class; |
2d3d891d | 3022 | } |
dd41f596 | 3023 | |
b29739f9 IM |
3024 | p->prio = prio; |
3025 | ||
0e1f3483 HS |
3026 | if (running) |
3027 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3028 | if (on_rq) |
2d3d891d | 3029 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3030 | |
da7a735e | 3031 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3032 | out_unlock: |
0122ec5b | 3033 | __task_rq_unlock(rq); |
b29739f9 | 3034 | } |
b29739f9 | 3035 | #endif |
d50dde5a | 3036 | |
36c8b586 | 3037 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3038 | { |
dd41f596 | 3039 | int old_prio, delta, on_rq; |
1da177e4 | 3040 | unsigned long flags; |
70b97a7f | 3041 | struct rq *rq; |
1da177e4 | 3042 | |
75e45d51 | 3043 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3044 | return; |
3045 | /* | |
3046 | * We have to be careful, if called from sys_setpriority(), | |
3047 | * the task might be in the middle of scheduling on another CPU. | |
3048 | */ | |
3049 | rq = task_rq_lock(p, &flags); | |
3050 | /* | |
3051 | * The RT priorities are set via sched_setscheduler(), but we still | |
3052 | * allow the 'normal' nice value to be set - but as expected | |
3053 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3054 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3055 | */ |
aab03e05 | 3056 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3057 | p->static_prio = NICE_TO_PRIO(nice); |
3058 | goto out_unlock; | |
3059 | } | |
fd2f4419 | 3060 | on_rq = p->on_rq; |
c09595f6 | 3061 | if (on_rq) |
69be72c1 | 3062 | dequeue_task(rq, p, 0); |
1da177e4 | 3063 | |
1da177e4 | 3064 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3065 | set_load_weight(p); |
b29739f9 IM |
3066 | old_prio = p->prio; |
3067 | p->prio = effective_prio(p); | |
3068 | delta = p->prio - old_prio; | |
1da177e4 | 3069 | |
dd41f596 | 3070 | if (on_rq) { |
371fd7e7 | 3071 | enqueue_task(rq, p, 0); |
1da177e4 | 3072 | /* |
d5f9f942 AM |
3073 | * If the task increased its priority or is running and |
3074 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3075 | */ |
d5f9f942 | 3076 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3077 | resched_curr(rq); |
1da177e4 LT |
3078 | } |
3079 | out_unlock: | |
0122ec5b | 3080 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3081 | } |
1da177e4 LT |
3082 | EXPORT_SYMBOL(set_user_nice); |
3083 | ||
e43379f1 MM |
3084 | /* |
3085 | * can_nice - check if a task can reduce its nice value | |
3086 | * @p: task | |
3087 | * @nice: nice value | |
3088 | */ | |
36c8b586 | 3089 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3090 | { |
024f4747 | 3091 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3092 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3093 | |
78d7d407 | 3094 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3095 | capable(CAP_SYS_NICE)); |
3096 | } | |
3097 | ||
1da177e4 LT |
3098 | #ifdef __ARCH_WANT_SYS_NICE |
3099 | ||
3100 | /* | |
3101 | * sys_nice - change the priority of the current process. | |
3102 | * @increment: priority increment | |
3103 | * | |
3104 | * sys_setpriority is a more generic, but much slower function that | |
3105 | * does similar things. | |
3106 | */ | |
5add95d4 | 3107 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3108 | { |
48f24c4d | 3109 | long nice, retval; |
1da177e4 LT |
3110 | |
3111 | /* | |
3112 | * Setpriority might change our priority at the same moment. | |
3113 | * We don't have to worry. Conceptually one call occurs first | |
3114 | * and we have a single winner. | |
3115 | */ | |
a9467fa3 | 3116 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3117 | nice = task_nice(current) + increment; |
1da177e4 | 3118 | |
a9467fa3 | 3119 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3120 | if (increment < 0 && !can_nice(current, nice)) |
3121 | return -EPERM; | |
3122 | ||
1da177e4 LT |
3123 | retval = security_task_setnice(current, nice); |
3124 | if (retval) | |
3125 | return retval; | |
3126 | ||
3127 | set_user_nice(current, nice); | |
3128 | return 0; | |
3129 | } | |
3130 | ||
3131 | #endif | |
3132 | ||
3133 | /** | |
3134 | * task_prio - return the priority value of a given task. | |
3135 | * @p: the task in question. | |
3136 | * | |
e69f6186 | 3137 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3138 | * RT tasks are offset by -200. Normal tasks are centered |
3139 | * around 0, value goes from -16 to +15. | |
3140 | */ | |
36c8b586 | 3141 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3142 | { |
3143 | return p->prio - MAX_RT_PRIO; | |
3144 | } | |
3145 | ||
1da177e4 LT |
3146 | /** |
3147 | * idle_cpu - is a given cpu idle currently? | |
3148 | * @cpu: the processor in question. | |
e69f6186 YB |
3149 | * |
3150 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3151 | */ |
3152 | int idle_cpu(int cpu) | |
3153 | { | |
908a3283 TG |
3154 | struct rq *rq = cpu_rq(cpu); |
3155 | ||
3156 | if (rq->curr != rq->idle) | |
3157 | return 0; | |
3158 | ||
3159 | if (rq->nr_running) | |
3160 | return 0; | |
3161 | ||
3162 | #ifdef CONFIG_SMP | |
3163 | if (!llist_empty(&rq->wake_list)) | |
3164 | return 0; | |
3165 | #endif | |
3166 | ||
3167 | return 1; | |
1da177e4 LT |
3168 | } |
3169 | ||
1da177e4 LT |
3170 | /** |
3171 | * idle_task - return the idle task for a given cpu. | |
3172 | * @cpu: the processor in question. | |
e69f6186 YB |
3173 | * |
3174 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3175 | */ |
36c8b586 | 3176 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3177 | { |
3178 | return cpu_rq(cpu)->idle; | |
3179 | } | |
3180 | ||
3181 | /** | |
3182 | * find_process_by_pid - find a process with a matching PID value. | |
3183 | * @pid: the pid in question. | |
e69f6186 YB |
3184 | * |
3185 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3186 | */ |
a9957449 | 3187 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3188 | { |
228ebcbe | 3189 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3190 | } |
3191 | ||
aab03e05 DF |
3192 | /* |
3193 | * This function initializes the sched_dl_entity of a newly becoming | |
3194 | * SCHED_DEADLINE task. | |
3195 | * | |
3196 | * Only the static values are considered here, the actual runtime and the | |
3197 | * absolute deadline will be properly calculated when the task is enqueued | |
3198 | * for the first time with its new policy. | |
3199 | */ | |
3200 | static void | |
3201 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3202 | { | |
3203 | struct sched_dl_entity *dl_se = &p->dl; | |
3204 | ||
3205 | init_dl_task_timer(dl_se); | |
3206 | dl_se->dl_runtime = attr->sched_runtime; | |
3207 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3208 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3209 | dl_se->flags = attr->sched_flags; |
332ac17e | 3210 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3211 | dl_se->dl_throttled = 0; |
3212 | dl_se->dl_new = 1; | |
5bfd126e | 3213 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3214 | } |
3215 | ||
c365c292 TG |
3216 | static void __setscheduler_params(struct task_struct *p, |
3217 | const struct sched_attr *attr) | |
1da177e4 | 3218 | { |
d50dde5a DF |
3219 | int policy = attr->sched_policy; |
3220 | ||
39fd8fd2 PZ |
3221 | if (policy == -1) /* setparam */ |
3222 | policy = p->policy; | |
3223 | ||
1da177e4 | 3224 | p->policy = policy; |
d50dde5a | 3225 | |
aab03e05 DF |
3226 | if (dl_policy(policy)) |
3227 | __setparam_dl(p, attr); | |
39fd8fd2 | 3228 | else if (fair_policy(policy)) |
d50dde5a DF |
3229 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3230 | ||
39fd8fd2 PZ |
3231 | /* |
3232 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3233 | * !rt_policy. Always setting this ensures that things like | |
3234 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3235 | */ | |
3236 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3237 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3238 | set_load_weight(p); |
3239 | } | |
39fd8fd2 | 3240 | |
c365c292 TG |
3241 | /* Actually do priority change: must hold pi & rq lock. */ |
3242 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3243 | const struct sched_attr *attr) | |
3244 | { | |
3245 | __setscheduler_params(p, attr); | |
d50dde5a | 3246 | |
383afd09 SR |
3247 | /* |
3248 | * If we get here, there was no pi waiters boosting the | |
3249 | * task. It is safe to use the normal prio. | |
3250 | */ | |
3251 | p->prio = normal_prio(p); | |
3252 | ||
aab03e05 DF |
3253 | if (dl_prio(p->prio)) |
3254 | p->sched_class = &dl_sched_class; | |
3255 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3256 | p->sched_class = &rt_sched_class; |
3257 | else | |
3258 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3259 | } |
aab03e05 DF |
3260 | |
3261 | static void | |
3262 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3263 | { | |
3264 | struct sched_dl_entity *dl_se = &p->dl; | |
3265 | ||
3266 | attr->sched_priority = p->rt_priority; | |
3267 | attr->sched_runtime = dl_se->dl_runtime; | |
3268 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3269 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3270 | attr->sched_flags = dl_se->flags; |
3271 | } | |
3272 | ||
3273 | /* | |
3274 | * This function validates the new parameters of a -deadline task. | |
3275 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3276 | * than the runtime, as well as the period of being zero or |
332ac17e | 3277 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3278 | * user parameters are above the internal resolution of 1us (we |
3279 | * check sched_runtime only since it is always the smaller one) and | |
3280 | * below 2^63 ns (we have to check both sched_deadline and | |
3281 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3282 | */ |
3283 | static bool | |
3284 | __checkparam_dl(const struct sched_attr *attr) | |
3285 | { | |
b0827819 JL |
3286 | /* deadline != 0 */ |
3287 | if (attr->sched_deadline == 0) | |
3288 | return false; | |
3289 | ||
3290 | /* | |
3291 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3292 | * that big. | |
3293 | */ | |
3294 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3295 | return false; | |
3296 | ||
3297 | /* | |
3298 | * Since we use the MSB for wrap-around and sign issues, make | |
3299 | * sure it's not set (mind that period can be equal to zero). | |
3300 | */ | |
3301 | if (attr->sched_deadline & (1ULL << 63) || | |
3302 | attr->sched_period & (1ULL << 63)) | |
3303 | return false; | |
3304 | ||
3305 | /* runtime <= deadline <= period (if period != 0) */ | |
3306 | if ((attr->sched_period != 0 && | |
3307 | attr->sched_period < attr->sched_deadline) || | |
3308 | attr->sched_deadline < attr->sched_runtime) | |
3309 | return false; | |
3310 | ||
3311 | return true; | |
aab03e05 DF |
3312 | } |
3313 | ||
c69e8d9c DH |
3314 | /* |
3315 | * check the target process has a UID that matches the current process's | |
3316 | */ | |
3317 | static bool check_same_owner(struct task_struct *p) | |
3318 | { | |
3319 | const struct cred *cred = current_cred(), *pcred; | |
3320 | bool match; | |
3321 | ||
3322 | rcu_read_lock(); | |
3323 | pcred = __task_cred(p); | |
9c806aa0 EB |
3324 | match = (uid_eq(cred->euid, pcred->euid) || |
3325 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3326 | rcu_read_unlock(); |
3327 | return match; | |
3328 | } | |
3329 | ||
d50dde5a DF |
3330 | static int __sched_setscheduler(struct task_struct *p, |
3331 | const struct sched_attr *attr, | |
3332 | bool user) | |
1da177e4 | 3333 | { |
383afd09 SR |
3334 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3335 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
83b699ed | 3336 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3337 | int policy = attr->sched_policy; |
1da177e4 | 3338 | unsigned long flags; |
83ab0aa0 | 3339 | const struct sched_class *prev_class; |
70b97a7f | 3340 | struct rq *rq; |
ca94c442 | 3341 | int reset_on_fork; |
1da177e4 | 3342 | |
66e5393a SR |
3343 | /* may grab non-irq protected spin_locks */ |
3344 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3345 | recheck: |
3346 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3347 | if (policy < 0) { |
3348 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3349 | policy = oldpolicy = p->policy; |
ca94c442 | 3350 | } else { |
7479f3c9 | 3351 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3352 | |
aab03e05 DF |
3353 | if (policy != SCHED_DEADLINE && |
3354 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3355 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3356 | policy != SCHED_IDLE) | |
3357 | return -EINVAL; | |
3358 | } | |
3359 | ||
7479f3c9 PZ |
3360 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3361 | return -EINVAL; | |
3362 | ||
1da177e4 LT |
3363 | /* |
3364 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3365 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3366 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3367 | */ |
0bb040a4 | 3368 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3369 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3370 | return -EINVAL; |
aab03e05 DF |
3371 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3372 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3373 | return -EINVAL; |
3374 | ||
37e4ab3f OC |
3375 | /* |
3376 | * Allow unprivileged RT tasks to decrease priority: | |
3377 | */ | |
961ccddd | 3378 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3379 | if (fair_policy(policy)) { |
d0ea0268 | 3380 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3381 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3382 | return -EPERM; |
3383 | } | |
3384 | ||
e05606d3 | 3385 | if (rt_policy(policy)) { |
a44702e8 ON |
3386 | unsigned long rlim_rtprio = |
3387 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3388 | |
3389 | /* can't set/change the rt policy */ | |
3390 | if (policy != p->policy && !rlim_rtprio) | |
3391 | return -EPERM; | |
3392 | ||
3393 | /* can't increase priority */ | |
d50dde5a DF |
3394 | if (attr->sched_priority > p->rt_priority && |
3395 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3396 | return -EPERM; |
3397 | } | |
c02aa73b | 3398 | |
d44753b8 JL |
3399 | /* |
3400 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3401 | * (safest behavior); in the future we would like to allow | |
3402 | * unprivileged DL tasks to increase their relative deadline | |
3403 | * or reduce their runtime (both ways reducing utilization) | |
3404 | */ | |
3405 | if (dl_policy(policy)) | |
3406 | return -EPERM; | |
3407 | ||
dd41f596 | 3408 | /* |
c02aa73b DH |
3409 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3410 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3411 | */ |
c02aa73b | 3412 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3413 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3414 | return -EPERM; |
3415 | } | |
5fe1d75f | 3416 | |
37e4ab3f | 3417 | /* can't change other user's priorities */ |
c69e8d9c | 3418 | if (!check_same_owner(p)) |
37e4ab3f | 3419 | return -EPERM; |
ca94c442 LP |
3420 | |
3421 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3422 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3423 | return -EPERM; | |
37e4ab3f | 3424 | } |
1da177e4 | 3425 | |
725aad24 | 3426 | if (user) { |
b0ae1981 | 3427 | retval = security_task_setscheduler(p); |
725aad24 JF |
3428 | if (retval) |
3429 | return retval; | |
3430 | } | |
3431 | ||
b29739f9 IM |
3432 | /* |
3433 | * make sure no PI-waiters arrive (or leave) while we are | |
3434 | * changing the priority of the task: | |
0122ec5b | 3435 | * |
25985edc | 3436 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3437 | * runqueue lock must be held. |
3438 | */ | |
0122ec5b | 3439 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3440 | |
34f971f6 PZ |
3441 | /* |
3442 | * Changing the policy of the stop threads its a very bad idea | |
3443 | */ | |
3444 | if (p == rq->stop) { | |
0122ec5b | 3445 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3446 | return -EINVAL; |
3447 | } | |
3448 | ||
a51e9198 | 3449 | /* |
d6b1e911 TG |
3450 | * If not changing anything there's no need to proceed further, |
3451 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3452 | */ |
d50dde5a | 3453 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3454 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3455 | goto change; |
3456 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3457 | goto change; | |
aab03e05 DF |
3458 | if (dl_policy(policy)) |
3459 | goto change; | |
d50dde5a | 3460 | |
d6b1e911 | 3461 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3462 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3463 | return 0; |
3464 | } | |
d50dde5a | 3465 | change: |
a51e9198 | 3466 | |
dc61b1d6 | 3467 | if (user) { |
332ac17e | 3468 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3469 | /* |
3470 | * Do not allow realtime tasks into groups that have no runtime | |
3471 | * assigned. | |
3472 | */ | |
3473 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3474 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3475 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3476 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3477 | return -EPERM; |
3478 | } | |
dc61b1d6 | 3479 | #endif |
332ac17e DF |
3480 | #ifdef CONFIG_SMP |
3481 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3482 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3483 | |
3484 | /* | |
3485 | * Don't allow tasks with an affinity mask smaller than | |
3486 | * the entire root_domain to become SCHED_DEADLINE. We | |
3487 | * will also fail if there's no bandwidth available. | |
3488 | */ | |
e4099a5e PZ |
3489 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3490 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3491 | task_rq_unlock(rq, p, &flags); |
3492 | return -EPERM; | |
3493 | } | |
3494 | } | |
3495 | #endif | |
3496 | } | |
dc61b1d6 | 3497 | |
1da177e4 LT |
3498 | /* recheck policy now with rq lock held */ |
3499 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3500 | policy = oldpolicy = -1; | |
0122ec5b | 3501 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3502 | goto recheck; |
3503 | } | |
332ac17e DF |
3504 | |
3505 | /* | |
3506 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3507 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3508 | * is available. | |
3509 | */ | |
e4099a5e | 3510 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3511 | task_rq_unlock(rq, p, &flags); |
3512 | return -EBUSY; | |
3513 | } | |
3514 | ||
c365c292 TG |
3515 | p->sched_reset_on_fork = reset_on_fork; |
3516 | oldprio = p->prio; | |
3517 | ||
3518 | /* | |
3519 | * Special case for priority boosted tasks. | |
3520 | * | |
3521 | * If the new priority is lower or equal (user space view) | |
3522 | * than the current (boosted) priority, we just store the new | |
3523 | * normal parameters and do not touch the scheduler class and | |
3524 | * the runqueue. This will be done when the task deboost | |
3525 | * itself. | |
3526 | */ | |
3527 | if (rt_mutex_check_prio(p, newprio)) { | |
3528 | __setscheduler_params(p, attr); | |
3529 | task_rq_unlock(rq, p, &flags); | |
3530 | return 0; | |
3531 | } | |
3532 | ||
fd2f4419 | 3533 | on_rq = p->on_rq; |
051a1d1a | 3534 | running = task_current(rq, p); |
0e1f3483 | 3535 | if (on_rq) |
4ca9b72b | 3536 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3537 | if (running) |
3538 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3539 | |
83ab0aa0 | 3540 | prev_class = p->sched_class; |
d50dde5a | 3541 | __setscheduler(rq, p, attr); |
f6b53205 | 3542 | |
0e1f3483 HS |
3543 | if (running) |
3544 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3545 | if (on_rq) { |
3546 | /* | |
3547 | * We enqueue to tail when the priority of a task is | |
3548 | * increased (user space view). | |
3549 | */ | |
3550 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3551 | } | |
cb469845 | 3552 | |
da7a735e | 3553 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3554 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3555 | |
95e02ca9 TG |
3556 | rt_mutex_adjust_pi(p); |
3557 | ||
1da177e4 LT |
3558 | return 0; |
3559 | } | |
961ccddd | 3560 | |
7479f3c9 PZ |
3561 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3562 | const struct sched_param *param, bool check) | |
3563 | { | |
3564 | struct sched_attr attr = { | |
3565 | .sched_policy = policy, | |
3566 | .sched_priority = param->sched_priority, | |
3567 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3568 | }; | |
3569 | ||
3570 | /* | |
3571 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3572 | */ | |
3573 | if (policy & SCHED_RESET_ON_FORK) { | |
3574 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3575 | policy &= ~SCHED_RESET_ON_FORK; | |
3576 | attr.sched_policy = policy; | |
3577 | } | |
3578 | ||
3579 | return __sched_setscheduler(p, &attr, check); | |
3580 | } | |
961ccddd RR |
3581 | /** |
3582 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3583 | * @p: the task in question. | |
3584 | * @policy: new policy. | |
3585 | * @param: structure containing the new RT priority. | |
3586 | * | |
e69f6186 YB |
3587 | * Return: 0 on success. An error code otherwise. |
3588 | * | |
961ccddd RR |
3589 | * NOTE that the task may be already dead. |
3590 | */ | |
3591 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3592 | const struct sched_param *param) |
961ccddd | 3593 | { |
7479f3c9 | 3594 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3595 | } |
1da177e4 LT |
3596 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3597 | ||
d50dde5a DF |
3598 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3599 | { | |
3600 | return __sched_setscheduler(p, attr, true); | |
3601 | } | |
3602 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3603 | ||
961ccddd RR |
3604 | /** |
3605 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3606 | * @p: the task in question. | |
3607 | * @policy: new policy. | |
3608 | * @param: structure containing the new RT priority. | |
3609 | * | |
3610 | * Just like sched_setscheduler, only don't bother checking if the | |
3611 | * current context has permission. For example, this is needed in | |
3612 | * stop_machine(): we create temporary high priority worker threads, | |
3613 | * but our caller might not have that capability. | |
e69f6186 YB |
3614 | * |
3615 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3616 | */ |
3617 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3618 | const struct sched_param *param) |
961ccddd | 3619 | { |
7479f3c9 | 3620 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3621 | } |
3622 | ||
95cdf3b7 IM |
3623 | static int |
3624 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3625 | { |
1da177e4 LT |
3626 | struct sched_param lparam; |
3627 | struct task_struct *p; | |
36c8b586 | 3628 | int retval; |
1da177e4 LT |
3629 | |
3630 | if (!param || pid < 0) | |
3631 | return -EINVAL; | |
3632 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3633 | return -EFAULT; | |
5fe1d75f ON |
3634 | |
3635 | rcu_read_lock(); | |
3636 | retval = -ESRCH; | |
1da177e4 | 3637 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3638 | if (p != NULL) |
3639 | retval = sched_setscheduler(p, policy, &lparam); | |
3640 | rcu_read_unlock(); | |
36c8b586 | 3641 | |
1da177e4 LT |
3642 | return retval; |
3643 | } | |
3644 | ||
d50dde5a DF |
3645 | /* |
3646 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3647 | */ | |
3648 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3649 | struct sched_attr *attr) | |
3650 | { | |
3651 | u32 size; | |
3652 | int ret; | |
3653 | ||
3654 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3655 | return -EFAULT; | |
3656 | ||
3657 | /* | |
3658 | * zero the full structure, so that a short copy will be nice. | |
3659 | */ | |
3660 | memset(attr, 0, sizeof(*attr)); | |
3661 | ||
3662 | ret = get_user(size, &uattr->size); | |
3663 | if (ret) | |
3664 | return ret; | |
3665 | ||
3666 | if (size > PAGE_SIZE) /* silly large */ | |
3667 | goto err_size; | |
3668 | ||
3669 | if (!size) /* abi compat */ | |
3670 | size = SCHED_ATTR_SIZE_VER0; | |
3671 | ||
3672 | if (size < SCHED_ATTR_SIZE_VER0) | |
3673 | goto err_size; | |
3674 | ||
3675 | /* | |
3676 | * If we're handed a bigger struct than we know of, | |
3677 | * ensure all the unknown bits are 0 - i.e. new | |
3678 | * user-space does not rely on any kernel feature | |
3679 | * extensions we dont know about yet. | |
3680 | */ | |
3681 | if (size > sizeof(*attr)) { | |
3682 | unsigned char __user *addr; | |
3683 | unsigned char __user *end; | |
3684 | unsigned char val; | |
3685 | ||
3686 | addr = (void __user *)uattr + sizeof(*attr); | |
3687 | end = (void __user *)uattr + size; | |
3688 | ||
3689 | for (; addr < end; addr++) { | |
3690 | ret = get_user(val, addr); | |
3691 | if (ret) | |
3692 | return ret; | |
3693 | if (val) | |
3694 | goto err_size; | |
3695 | } | |
3696 | size = sizeof(*attr); | |
3697 | } | |
3698 | ||
3699 | ret = copy_from_user(attr, uattr, size); | |
3700 | if (ret) | |
3701 | return -EFAULT; | |
3702 | ||
3703 | /* | |
3704 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3705 | * to be strict and return an error on out-of-bounds values? | |
3706 | */ | |
75e45d51 | 3707 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3708 | |
e78c7bca | 3709 | return 0; |
d50dde5a DF |
3710 | |
3711 | err_size: | |
3712 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3713 | return -E2BIG; |
d50dde5a DF |
3714 | } |
3715 | ||
1da177e4 LT |
3716 | /** |
3717 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3718 | * @pid: the pid in question. | |
3719 | * @policy: new policy. | |
3720 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3721 | * |
3722 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3723 | */ |
5add95d4 HC |
3724 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3725 | struct sched_param __user *, param) | |
1da177e4 | 3726 | { |
c21761f1 JB |
3727 | /* negative values for policy are not valid */ |
3728 | if (policy < 0) | |
3729 | return -EINVAL; | |
3730 | ||
1da177e4 LT |
3731 | return do_sched_setscheduler(pid, policy, param); |
3732 | } | |
3733 | ||
3734 | /** | |
3735 | * sys_sched_setparam - set/change the RT priority of a thread | |
3736 | * @pid: the pid in question. | |
3737 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3738 | * |
3739 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3740 | */ |
5add95d4 | 3741 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3742 | { |
3743 | return do_sched_setscheduler(pid, -1, param); | |
3744 | } | |
3745 | ||
d50dde5a DF |
3746 | /** |
3747 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3748 | * @pid: the pid in question. | |
5778fccf | 3749 | * @uattr: structure containing the extended parameters. |
db66d756 | 3750 | * @flags: for future extension. |
d50dde5a | 3751 | */ |
6d35ab48 PZ |
3752 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3753 | unsigned int, flags) | |
d50dde5a DF |
3754 | { |
3755 | struct sched_attr attr; | |
3756 | struct task_struct *p; | |
3757 | int retval; | |
3758 | ||
6d35ab48 | 3759 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3760 | return -EINVAL; |
3761 | ||
143cf23d MK |
3762 | retval = sched_copy_attr(uattr, &attr); |
3763 | if (retval) | |
3764 | return retval; | |
d50dde5a | 3765 | |
b14ed2c2 | 3766 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 3767 | return -EINVAL; |
d50dde5a DF |
3768 | |
3769 | rcu_read_lock(); | |
3770 | retval = -ESRCH; | |
3771 | p = find_process_by_pid(pid); | |
3772 | if (p != NULL) | |
3773 | retval = sched_setattr(p, &attr); | |
3774 | rcu_read_unlock(); | |
3775 | ||
3776 | return retval; | |
3777 | } | |
3778 | ||
1da177e4 LT |
3779 | /** |
3780 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3781 | * @pid: the pid in question. | |
e69f6186 YB |
3782 | * |
3783 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3784 | * code. | |
1da177e4 | 3785 | */ |
5add95d4 | 3786 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3787 | { |
36c8b586 | 3788 | struct task_struct *p; |
3a5c359a | 3789 | int retval; |
1da177e4 LT |
3790 | |
3791 | if (pid < 0) | |
3a5c359a | 3792 | return -EINVAL; |
1da177e4 LT |
3793 | |
3794 | retval = -ESRCH; | |
5fe85be0 | 3795 | rcu_read_lock(); |
1da177e4 LT |
3796 | p = find_process_by_pid(pid); |
3797 | if (p) { | |
3798 | retval = security_task_getscheduler(p); | |
3799 | if (!retval) | |
ca94c442 LP |
3800 | retval = p->policy |
3801 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3802 | } |
5fe85be0 | 3803 | rcu_read_unlock(); |
1da177e4 LT |
3804 | return retval; |
3805 | } | |
3806 | ||
3807 | /** | |
ca94c442 | 3808 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3809 | * @pid: the pid in question. |
3810 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3811 | * |
3812 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3813 | * code. | |
1da177e4 | 3814 | */ |
5add95d4 | 3815 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3816 | { |
ce5f7f82 | 3817 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3818 | struct task_struct *p; |
3a5c359a | 3819 | int retval; |
1da177e4 LT |
3820 | |
3821 | if (!param || pid < 0) | |
3a5c359a | 3822 | return -EINVAL; |
1da177e4 | 3823 | |
5fe85be0 | 3824 | rcu_read_lock(); |
1da177e4 LT |
3825 | p = find_process_by_pid(pid); |
3826 | retval = -ESRCH; | |
3827 | if (!p) | |
3828 | goto out_unlock; | |
3829 | ||
3830 | retval = security_task_getscheduler(p); | |
3831 | if (retval) | |
3832 | goto out_unlock; | |
3833 | ||
ce5f7f82 PZ |
3834 | if (task_has_rt_policy(p)) |
3835 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3836 | rcu_read_unlock(); |
1da177e4 LT |
3837 | |
3838 | /* | |
3839 | * This one might sleep, we cannot do it with a spinlock held ... | |
3840 | */ | |
3841 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3842 | ||
1da177e4 LT |
3843 | return retval; |
3844 | ||
3845 | out_unlock: | |
5fe85be0 | 3846 | rcu_read_unlock(); |
1da177e4 LT |
3847 | return retval; |
3848 | } | |
3849 | ||
d50dde5a DF |
3850 | static int sched_read_attr(struct sched_attr __user *uattr, |
3851 | struct sched_attr *attr, | |
3852 | unsigned int usize) | |
3853 | { | |
3854 | int ret; | |
3855 | ||
3856 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3857 | return -EFAULT; | |
3858 | ||
3859 | /* | |
3860 | * If we're handed a smaller struct than we know of, | |
3861 | * ensure all the unknown bits are 0 - i.e. old | |
3862 | * user-space does not get uncomplete information. | |
3863 | */ | |
3864 | if (usize < sizeof(*attr)) { | |
3865 | unsigned char *addr; | |
3866 | unsigned char *end; | |
3867 | ||
3868 | addr = (void *)attr + usize; | |
3869 | end = (void *)attr + sizeof(*attr); | |
3870 | ||
3871 | for (; addr < end; addr++) { | |
3872 | if (*addr) | |
22400674 | 3873 | return -EFBIG; |
d50dde5a DF |
3874 | } |
3875 | ||
3876 | attr->size = usize; | |
3877 | } | |
3878 | ||
4efbc454 | 3879 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3880 | if (ret) |
3881 | return -EFAULT; | |
3882 | ||
22400674 | 3883 | return 0; |
d50dde5a DF |
3884 | } |
3885 | ||
3886 | /** | |
aab03e05 | 3887 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3888 | * @pid: the pid in question. |
5778fccf | 3889 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3890 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3891 | * @flags: for future extension. |
d50dde5a | 3892 | */ |
6d35ab48 PZ |
3893 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3894 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3895 | { |
3896 | struct sched_attr attr = { | |
3897 | .size = sizeof(struct sched_attr), | |
3898 | }; | |
3899 | struct task_struct *p; | |
3900 | int retval; | |
3901 | ||
3902 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3903 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3904 | return -EINVAL; |
3905 | ||
3906 | rcu_read_lock(); | |
3907 | p = find_process_by_pid(pid); | |
3908 | retval = -ESRCH; | |
3909 | if (!p) | |
3910 | goto out_unlock; | |
3911 | ||
3912 | retval = security_task_getscheduler(p); | |
3913 | if (retval) | |
3914 | goto out_unlock; | |
3915 | ||
3916 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3917 | if (p->sched_reset_on_fork) |
3918 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3919 | if (task_has_dl_policy(p)) |
3920 | __getparam_dl(p, &attr); | |
3921 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3922 | attr.sched_priority = p->rt_priority; |
3923 | else | |
d0ea0268 | 3924 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3925 | |
3926 | rcu_read_unlock(); | |
3927 | ||
3928 | retval = sched_read_attr(uattr, &attr, size); | |
3929 | return retval; | |
3930 | ||
3931 | out_unlock: | |
3932 | rcu_read_unlock(); | |
3933 | return retval; | |
3934 | } | |
3935 | ||
96f874e2 | 3936 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3937 | { |
5a16f3d3 | 3938 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3939 | struct task_struct *p; |
3940 | int retval; | |
1da177e4 | 3941 | |
23f5d142 | 3942 | rcu_read_lock(); |
1da177e4 LT |
3943 | |
3944 | p = find_process_by_pid(pid); | |
3945 | if (!p) { | |
23f5d142 | 3946 | rcu_read_unlock(); |
1da177e4 LT |
3947 | return -ESRCH; |
3948 | } | |
3949 | ||
23f5d142 | 3950 | /* Prevent p going away */ |
1da177e4 | 3951 | get_task_struct(p); |
23f5d142 | 3952 | rcu_read_unlock(); |
1da177e4 | 3953 | |
14a40ffc TH |
3954 | if (p->flags & PF_NO_SETAFFINITY) { |
3955 | retval = -EINVAL; | |
3956 | goto out_put_task; | |
3957 | } | |
5a16f3d3 RR |
3958 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3959 | retval = -ENOMEM; | |
3960 | goto out_put_task; | |
3961 | } | |
3962 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3963 | retval = -ENOMEM; | |
3964 | goto out_free_cpus_allowed; | |
3965 | } | |
1da177e4 | 3966 | retval = -EPERM; |
4c44aaaf EB |
3967 | if (!check_same_owner(p)) { |
3968 | rcu_read_lock(); | |
3969 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3970 | rcu_read_unlock(); | |
3971 | goto out_unlock; | |
3972 | } | |
3973 | rcu_read_unlock(); | |
3974 | } | |
1da177e4 | 3975 | |
b0ae1981 | 3976 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3977 | if (retval) |
3978 | goto out_unlock; | |
3979 | ||
e4099a5e PZ |
3980 | |
3981 | cpuset_cpus_allowed(p, cpus_allowed); | |
3982 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3983 | ||
332ac17e DF |
3984 | /* |
3985 | * Since bandwidth control happens on root_domain basis, | |
3986 | * if admission test is enabled, we only admit -deadline | |
3987 | * tasks allowed to run on all the CPUs in the task's | |
3988 | * root_domain. | |
3989 | */ | |
3990 | #ifdef CONFIG_SMP | |
3991 | if (task_has_dl_policy(p)) { | |
3992 | const struct cpumask *span = task_rq(p)->rd->span; | |
3993 | ||
e4099a5e | 3994 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
3995 | retval = -EBUSY; |
3996 | goto out_unlock; | |
3997 | } | |
3998 | } | |
3999 | #endif | |
49246274 | 4000 | again: |
5a16f3d3 | 4001 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4002 | |
8707d8b8 | 4003 | if (!retval) { |
5a16f3d3 RR |
4004 | cpuset_cpus_allowed(p, cpus_allowed); |
4005 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4006 | /* |
4007 | * We must have raced with a concurrent cpuset | |
4008 | * update. Just reset the cpus_allowed to the | |
4009 | * cpuset's cpus_allowed | |
4010 | */ | |
5a16f3d3 | 4011 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4012 | goto again; |
4013 | } | |
4014 | } | |
1da177e4 | 4015 | out_unlock: |
5a16f3d3 RR |
4016 | free_cpumask_var(new_mask); |
4017 | out_free_cpus_allowed: | |
4018 | free_cpumask_var(cpus_allowed); | |
4019 | out_put_task: | |
1da177e4 | 4020 | put_task_struct(p); |
1da177e4 LT |
4021 | return retval; |
4022 | } | |
4023 | ||
4024 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4025 | struct cpumask *new_mask) |
1da177e4 | 4026 | { |
96f874e2 RR |
4027 | if (len < cpumask_size()) |
4028 | cpumask_clear(new_mask); | |
4029 | else if (len > cpumask_size()) | |
4030 | len = cpumask_size(); | |
4031 | ||
1da177e4 LT |
4032 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4033 | } | |
4034 | ||
4035 | /** | |
4036 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4037 | * @pid: pid of the process | |
4038 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4039 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4040 | * |
4041 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4042 | */ |
5add95d4 HC |
4043 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4044 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4045 | { |
5a16f3d3 | 4046 | cpumask_var_t new_mask; |
1da177e4 LT |
4047 | int retval; |
4048 | ||
5a16f3d3 RR |
4049 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4050 | return -ENOMEM; | |
1da177e4 | 4051 | |
5a16f3d3 RR |
4052 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4053 | if (retval == 0) | |
4054 | retval = sched_setaffinity(pid, new_mask); | |
4055 | free_cpumask_var(new_mask); | |
4056 | return retval; | |
1da177e4 LT |
4057 | } |
4058 | ||
96f874e2 | 4059 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4060 | { |
36c8b586 | 4061 | struct task_struct *p; |
31605683 | 4062 | unsigned long flags; |
1da177e4 | 4063 | int retval; |
1da177e4 | 4064 | |
23f5d142 | 4065 | rcu_read_lock(); |
1da177e4 LT |
4066 | |
4067 | retval = -ESRCH; | |
4068 | p = find_process_by_pid(pid); | |
4069 | if (!p) | |
4070 | goto out_unlock; | |
4071 | ||
e7834f8f DQ |
4072 | retval = security_task_getscheduler(p); |
4073 | if (retval) | |
4074 | goto out_unlock; | |
4075 | ||
013fdb80 | 4076 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4077 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4078 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4079 | |
4080 | out_unlock: | |
23f5d142 | 4081 | rcu_read_unlock(); |
1da177e4 | 4082 | |
9531b62f | 4083 | return retval; |
1da177e4 LT |
4084 | } |
4085 | ||
4086 | /** | |
4087 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4088 | * @pid: pid of the process | |
4089 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4090 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4091 | * |
4092 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4093 | */ |
5add95d4 HC |
4094 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4095 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4096 | { |
4097 | int ret; | |
f17c8607 | 4098 | cpumask_var_t mask; |
1da177e4 | 4099 | |
84fba5ec | 4100 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4101 | return -EINVAL; |
4102 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4103 | return -EINVAL; |
4104 | ||
f17c8607 RR |
4105 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4106 | return -ENOMEM; | |
1da177e4 | 4107 | |
f17c8607 RR |
4108 | ret = sched_getaffinity(pid, mask); |
4109 | if (ret == 0) { | |
8bc037fb | 4110 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4111 | |
4112 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4113 | ret = -EFAULT; |
4114 | else | |
cd3d8031 | 4115 | ret = retlen; |
f17c8607 RR |
4116 | } |
4117 | free_cpumask_var(mask); | |
1da177e4 | 4118 | |
f17c8607 | 4119 | return ret; |
1da177e4 LT |
4120 | } |
4121 | ||
4122 | /** | |
4123 | * sys_sched_yield - yield the current processor to other threads. | |
4124 | * | |
dd41f596 IM |
4125 | * This function yields the current CPU to other tasks. If there are no |
4126 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4127 | * |
4128 | * Return: 0. | |
1da177e4 | 4129 | */ |
5add95d4 | 4130 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4131 | { |
70b97a7f | 4132 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4133 | |
2d72376b | 4134 | schedstat_inc(rq, yld_count); |
4530d7ab | 4135 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4136 | |
4137 | /* | |
4138 | * Since we are going to call schedule() anyway, there's | |
4139 | * no need to preempt or enable interrupts: | |
4140 | */ | |
4141 | __release(rq->lock); | |
8a25d5de | 4142 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4143 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4144 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4145 | |
4146 | schedule(); | |
4147 | ||
4148 | return 0; | |
4149 | } | |
4150 | ||
e7b38404 | 4151 | static void __cond_resched(void) |
1da177e4 | 4152 | { |
bdb43806 | 4153 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4154 | __schedule(); |
bdb43806 | 4155 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4156 | } |
4157 | ||
02b67cc3 | 4158 | int __sched _cond_resched(void) |
1da177e4 | 4159 | { |
ac1bea85 | 4160 | rcu_cond_resched(); |
d86ee480 | 4161 | if (should_resched()) { |
1da177e4 LT |
4162 | __cond_resched(); |
4163 | return 1; | |
4164 | } | |
4165 | return 0; | |
4166 | } | |
02b67cc3 | 4167 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4168 | |
4169 | /* | |
613afbf8 | 4170 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4171 | * call schedule, and on return reacquire the lock. |
4172 | * | |
41a2d6cf | 4173 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4174 | * operations here to prevent schedule() from being called twice (once via |
4175 | * spin_unlock(), once by hand). | |
4176 | */ | |
613afbf8 | 4177 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4178 | { |
ac1bea85 | 4179 | bool need_rcu_resched = rcu_should_resched(); |
d86ee480 | 4180 | int resched = should_resched(); |
6df3cecb JK |
4181 | int ret = 0; |
4182 | ||
f607c668 PZ |
4183 | lockdep_assert_held(lock); |
4184 | ||
ac1bea85 | 4185 | if (spin_needbreak(lock) || resched || need_rcu_resched) { |
1da177e4 | 4186 | spin_unlock(lock); |
d86ee480 | 4187 | if (resched) |
95c354fe | 4188 | __cond_resched(); |
ac1bea85 PM |
4189 | else if (unlikely(need_rcu_resched)) |
4190 | rcu_resched(); | |
95c354fe NP |
4191 | else |
4192 | cpu_relax(); | |
6df3cecb | 4193 | ret = 1; |
1da177e4 | 4194 | spin_lock(lock); |
1da177e4 | 4195 | } |
6df3cecb | 4196 | return ret; |
1da177e4 | 4197 | } |
613afbf8 | 4198 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4199 | |
613afbf8 | 4200 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4201 | { |
4202 | BUG_ON(!in_softirq()); | |
4203 | ||
ac1bea85 | 4204 | rcu_cond_resched(); /* BH disabled OK, just recording QSes. */ |
d86ee480 | 4205 | if (should_resched()) { |
98d82567 | 4206 | local_bh_enable(); |
1da177e4 LT |
4207 | __cond_resched(); |
4208 | local_bh_disable(); | |
4209 | return 1; | |
4210 | } | |
4211 | return 0; | |
4212 | } | |
613afbf8 | 4213 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4214 | |
1da177e4 LT |
4215 | /** |
4216 | * yield - yield the current processor to other threads. | |
4217 | * | |
8e3fabfd PZ |
4218 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4219 | * | |
4220 | * The scheduler is at all times free to pick the calling task as the most | |
4221 | * eligible task to run, if removing the yield() call from your code breaks | |
4222 | * it, its already broken. | |
4223 | * | |
4224 | * Typical broken usage is: | |
4225 | * | |
4226 | * while (!event) | |
4227 | * yield(); | |
4228 | * | |
4229 | * where one assumes that yield() will let 'the other' process run that will | |
4230 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4231 | * happen. Never use yield() as a progress guarantee!! | |
4232 | * | |
4233 | * If you want to use yield() to wait for something, use wait_event(). | |
4234 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4235 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4236 | */ |
4237 | void __sched yield(void) | |
4238 | { | |
4239 | set_current_state(TASK_RUNNING); | |
4240 | sys_sched_yield(); | |
4241 | } | |
1da177e4 LT |
4242 | EXPORT_SYMBOL(yield); |
4243 | ||
d95f4122 MG |
4244 | /** |
4245 | * yield_to - yield the current processor to another thread in | |
4246 | * your thread group, or accelerate that thread toward the | |
4247 | * processor it's on. | |
16addf95 RD |
4248 | * @p: target task |
4249 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4250 | * |
4251 | * It's the caller's job to ensure that the target task struct | |
4252 | * can't go away on us before we can do any checks. | |
4253 | * | |
e69f6186 | 4254 | * Return: |
7b270f60 PZ |
4255 | * true (>0) if we indeed boosted the target task. |
4256 | * false (0) if we failed to boost the target. | |
4257 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4258 | */ |
fa93384f | 4259 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4260 | { |
4261 | struct task_struct *curr = current; | |
4262 | struct rq *rq, *p_rq; | |
4263 | unsigned long flags; | |
c3c18640 | 4264 | int yielded = 0; |
d95f4122 MG |
4265 | |
4266 | local_irq_save(flags); | |
4267 | rq = this_rq(); | |
4268 | ||
4269 | again: | |
4270 | p_rq = task_rq(p); | |
7b270f60 PZ |
4271 | /* |
4272 | * If we're the only runnable task on the rq and target rq also | |
4273 | * has only one task, there's absolutely no point in yielding. | |
4274 | */ | |
4275 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4276 | yielded = -ESRCH; | |
4277 | goto out_irq; | |
4278 | } | |
4279 | ||
d95f4122 | 4280 | double_rq_lock(rq, p_rq); |
39e24d8f | 4281 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4282 | double_rq_unlock(rq, p_rq); |
4283 | goto again; | |
4284 | } | |
4285 | ||
4286 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4287 | goto out_unlock; |
d95f4122 MG |
4288 | |
4289 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4290 | goto out_unlock; |
d95f4122 MG |
4291 | |
4292 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4293 | goto out_unlock; |
d95f4122 MG |
4294 | |
4295 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4296 | if (yielded) { |
d95f4122 | 4297 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4298 | /* |
4299 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4300 | * fairness. | |
4301 | */ | |
4302 | if (preempt && rq != p_rq) | |
8875125e | 4303 | resched_curr(p_rq); |
6d1cafd8 | 4304 | } |
d95f4122 | 4305 | |
7b270f60 | 4306 | out_unlock: |
d95f4122 | 4307 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4308 | out_irq: |
d95f4122 MG |
4309 | local_irq_restore(flags); |
4310 | ||
7b270f60 | 4311 | if (yielded > 0) |
d95f4122 MG |
4312 | schedule(); |
4313 | ||
4314 | return yielded; | |
4315 | } | |
4316 | EXPORT_SYMBOL_GPL(yield_to); | |
4317 | ||
1da177e4 | 4318 | /* |
41a2d6cf | 4319 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4320 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4321 | */ |
4322 | void __sched io_schedule(void) | |
4323 | { | |
54d35f29 | 4324 | struct rq *rq = raw_rq(); |
1da177e4 | 4325 | |
0ff92245 | 4326 | delayacct_blkio_start(); |
1da177e4 | 4327 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4328 | blk_flush_plug(current); |
8f0dfc34 | 4329 | current->in_iowait = 1; |
1da177e4 | 4330 | schedule(); |
8f0dfc34 | 4331 | current->in_iowait = 0; |
1da177e4 | 4332 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4333 | delayacct_blkio_end(); |
1da177e4 | 4334 | } |
1da177e4 LT |
4335 | EXPORT_SYMBOL(io_schedule); |
4336 | ||
4337 | long __sched io_schedule_timeout(long timeout) | |
4338 | { | |
54d35f29 | 4339 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4340 | long ret; |
4341 | ||
0ff92245 | 4342 | delayacct_blkio_start(); |
1da177e4 | 4343 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4344 | blk_flush_plug(current); |
8f0dfc34 | 4345 | current->in_iowait = 1; |
1da177e4 | 4346 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4347 | current->in_iowait = 0; |
1da177e4 | 4348 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4349 | delayacct_blkio_end(); |
1da177e4 LT |
4350 | return ret; |
4351 | } | |
4352 | ||
4353 | /** | |
4354 | * sys_sched_get_priority_max - return maximum RT priority. | |
4355 | * @policy: scheduling class. | |
4356 | * | |
e69f6186 YB |
4357 | * Return: On success, this syscall returns the maximum |
4358 | * rt_priority that can be used by a given scheduling class. | |
4359 | * On failure, a negative error code is returned. | |
1da177e4 | 4360 | */ |
5add95d4 | 4361 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4362 | { |
4363 | int ret = -EINVAL; | |
4364 | ||
4365 | switch (policy) { | |
4366 | case SCHED_FIFO: | |
4367 | case SCHED_RR: | |
4368 | ret = MAX_USER_RT_PRIO-1; | |
4369 | break; | |
aab03e05 | 4370 | case SCHED_DEADLINE: |
1da177e4 | 4371 | case SCHED_NORMAL: |
b0a9499c | 4372 | case SCHED_BATCH: |
dd41f596 | 4373 | case SCHED_IDLE: |
1da177e4 LT |
4374 | ret = 0; |
4375 | break; | |
4376 | } | |
4377 | return ret; | |
4378 | } | |
4379 | ||
4380 | /** | |
4381 | * sys_sched_get_priority_min - return minimum RT priority. | |
4382 | * @policy: scheduling class. | |
4383 | * | |
e69f6186 YB |
4384 | * Return: On success, this syscall returns the minimum |
4385 | * rt_priority that can be used by a given scheduling class. | |
4386 | * On failure, a negative error code is returned. | |
1da177e4 | 4387 | */ |
5add95d4 | 4388 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4389 | { |
4390 | int ret = -EINVAL; | |
4391 | ||
4392 | switch (policy) { | |
4393 | case SCHED_FIFO: | |
4394 | case SCHED_RR: | |
4395 | ret = 1; | |
4396 | break; | |
aab03e05 | 4397 | case SCHED_DEADLINE: |
1da177e4 | 4398 | case SCHED_NORMAL: |
b0a9499c | 4399 | case SCHED_BATCH: |
dd41f596 | 4400 | case SCHED_IDLE: |
1da177e4 LT |
4401 | ret = 0; |
4402 | } | |
4403 | return ret; | |
4404 | } | |
4405 | ||
4406 | /** | |
4407 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4408 | * @pid: pid of the process. | |
4409 | * @interval: userspace pointer to the timeslice value. | |
4410 | * | |
4411 | * this syscall writes the default timeslice value of a given process | |
4412 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4413 | * |
4414 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4415 | * an error code. | |
1da177e4 | 4416 | */ |
17da2bd9 | 4417 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4418 | struct timespec __user *, interval) |
1da177e4 | 4419 | { |
36c8b586 | 4420 | struct task_struct *p; |
a4ec24b4 | 4421 | unsigned int time_slice; |
dba091b9 TG |
4422 | unsigned long flags; |
4423 | struct rq *rq; | |
3a5c359a | 4424 | int retval; |
1da177e4 | 4425 | struct timespec t; |
1da177e4 LT |
4426 | |
4427 | if (pid < 0) | |
3a5c359a | 4428 | return -EINVAL; |
1da177e4 LT |
4429 | |
4430 | retval = -ESRCH; | |
1a551ae7 | 4431 | rcu_read_lock(); |
1da177e4 LT |
4432 | p = find_process_by_pid(pid); |
4433 | if (!p) | |
4434 | goto out_unlock; | |
4435 | ||
4436 | retval = security_task_getscheduler(p); | |
4437 | if (retval) | |
4438 | goto out_unlock; | |
4439 | ||
dba091b9 | 4440 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4441 | time_slice = 0; |
4442 | if (p->sched_class->get_rr_interval) | |
4443 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4444 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4445 | |
1a551ae7 | 4446 | rcu_read_unlock(); |
a4ec24b4 | 4447 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4448 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4449 | return retval; |
3a5c359a | 4450 | |
1da177e4 | 4451 | out_unlock: |
1a551ae7 | 4452 | rcu_read_unlock(); |
1da177e4 LT |
4453 | return retval; |
4454 | } | |
4455 | ||
7c731e0a | 4456 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4457 | |
82a1fcb9 | 4458 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4459 | { |
1da177e4 | 4460 | unsigned long free = 0; |
4e79752c | 4461 | int ppid; |
36c8b586 | 4462 | unsigned state; |
1da177e4 | 4463 | |
1da177e4 | 4464 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4465 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4466 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4467 | #if BITS_PER_LONG == 32 |
1da177e4 | 4468 | if (state == TASK_RUNNING) |
3df0fc5b | 4469 | printk(KERN_CONT " running "); |
1da177e4 | 4470 | else |
3df0fc5b | 4471 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4472 | #else |
4473 | if (state == TASK_RUNNING) | |
3df0fc5b | 4474 | printk(KERN_CONT " running task "); |
1da177e4 | 4475 | else |
3df0fc5b | 4476 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4477 | #endif |
4478 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4479 | free = stack_not_used(p); |
1da177e4 | 4480 | #endif |
4e79752c PM |
4481 | rcu_read_lock(); |
4482 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4483 | rcu_read_unlock(); | |
3df0fc5b | 4484 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4485 | task_pid_nr(p), ppid, |
aa47b7e0 | 4486 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4487 | |
3d1cb205 | 4488 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4489 | show_stack(p, NULL); |
1da177e4 LT |
4490 | } |
4491 | ||
e59e2ae2 | 4492 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4493 | { |
36c8b586 | 4494 | struct task_struct *g, *p; |
1da177e4 | 4495 | |
4bd77321 | 4496 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4497 | printk(KERN_INFO |
4498 | " task PC stack pid father\n"); | |
1da177e4 | 4499 | #else |
3df0fc5b PZ |
4500 | printk(KERN_INFO |
4501 | " task PC stack pid father\n"); | |
1da177e4 | 4502 | #endif |
510f5acc | 4503 | rcu_read_lock(); |
1da177e4 LT |
4504 | do_each_thread(g, p) { |
4505 | /* | |
4506 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4507 | * console might take a lot of time: |
1da177e4 LT |
4508 | */ |
4509 | touch_nmi_watchdog(); | |
39bc89fd | 4510 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4511 | sched_show_task(p); |
1da177e4 LT |
4512 | } while_each_thread(g, p); |
4513 | ||
04c9167f JF |
4514 | touch_all_softlockup_watchdogs(); |
4515 | ||
dd41f596 IM |
4516 | #ifdef CONFIG_SCHED_DEBUG |
4517 | sysrq_sched_debug_show(); | |
4518 | #endif | |
510f5acc | 4519 | rcu_read_unlock(); |
e59e2ae2 IM |
4520 | /* |
4521 | * Only show locks if all tasks are dumped: | |
4522 | */ | |
93335a21 | 4523 | if (!state_filter) |
e59e2ae2 | 4524 | debug_show_all_locks(); |
1da177e4 LT |
4525 | } |
4526 | ||
0db0628d | 4527 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4528 | { |
dd41f596 | 4529 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4530 | } |
4531 | ||
f340c0d1 IM |
4532 | /** |
4533 | * init_idle - set up an idle thread for a given CPU | |
4534 | * @idle: task in question | |
4535 | * @cpu: cpu the idle task belongs to | |
4536 | * | |
4537 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4538 | * flag, to make booting more robust. | |
4539 | */ | |
0db0628d | 4540 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4541 | { |
70b97a7f | 4542 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4543 | unsigned long flags; |
4544 | ||
05fa785c | 4545 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4546 | |
5e1576ed | 4547 | __sched_fork(0, idle); |
06b83b5f | 4548 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4549 | idle->se.exec_start = sched_clock(); |
4550 | ||
1e1b6c51 | 4551 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4552 | /* |
4553 | * We're having a chicken and egg problem, even though we are | |
4554 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4555 | * lockdep check in task_group() will fail. | |
4556 | * | |
4557 | * Similar case to sched_fork(). / Alternatively we could | |
4558 | * use task_rq_lock() here and obtain the other rq->lock. | |
4559 | * | |
4560 | * Silence PROVE_RCU | |
4561 | */ | |
4562 | rcu_read_lock(); | |
dd41f596 | 4563 | __set_task_cpu(idle, cpu); |
6506cf6c | 4564 | rcu_read_unlock(); |
1da177e4 | 4565 | |
1da177e4 | 4566 | rq->curr = rq->idle = idle; |
77177856 | 4567 | idle->on_rq = 1; |
3ca7a440 PZ |
4568 | #if defined(CONFIG_SMP) |
4569 | idle->on_cpu = 1; | |
4866cde0 | 4570 | #endif |
05fa785c | 4571 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4572 | |
4573 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4574 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4575 | |
dd41f596 IM |
4576 | /* |
4577 | * The idle tasks have their own, simple scheduling class: | |
4578 | */ | |
4579 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4580 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4581 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4582 | #if defined(CONFIG_SMP) |
4583 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4584 | #endif | |
19978ca6 IM |
4585 | } |
4586 | ||
1da177e4 | 4587 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4588 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4589 | { | |
4590 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4591 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4592 | |
4593 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4594 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4595 | } |
4596 | ||
1da177e4 LT |
4597 | /* |
4598 | * This is how migration works: | |
4599 | * | |
969c7921 TH |
4600 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4601 | * stop_one_cpu(). | |
4602 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4603 | * off the CPU) | |
4604 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4605 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4606 | * it and puts it into the right queue. |
969c7921 TH |
4607 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4608 | * is done. | |
1da177e4 LT |
4609 | */ |
4610 | ||
4611 | /* | |
4612 | * Change a given task's CPU affinity. Migrate the thread to a | |
4613 | * proper CPU and schedule it away if the CPU it's executing on | |
4614 | * is removed from the allowed bitmask. | |
4615 | * | |
4616 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4617 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4618 | * call is not atomic; no spinlocks may be held. |
4619 | */ | |
96f874e2 | 4620 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4621 | { |
4622 | unsigned long flags; | |
70b97a7f | 4623 | struct rq *rq; |
969c7921 | 4624 | unsigned int dest_cpu; |
48f24c4d | 4625 | int ret = 0; |
1da177e4 LT |
4626 | |
4627 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4628 | |
db44fc01 YZ |
4629 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4630 | goto out; | |
4631 | ||
6ad4c188 | 4632 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4633 | ret = -EINVAL; |
4634 | goto out; | |
4635 | } | |
4636 | ||
1e1b6c51 | 4637 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4638 | |
1da177e4 | 4639 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4640 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4641 | goto out; |
4642 | ||
969c7921 | 4643 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4644 | if (p->on_rq) { |
969c7921 | 4645 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4646 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4647 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4648 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4649 | tlb_migrate_finish(p->mm); |
4650 | return 0; | |
4651 | } | |
4652 | out: | |
0122ec5b | 4653 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4654 | |
1da177e4 LT |
4655 | return ret; |
4656 | } | |
cd8ba7cd | 4657 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4658 | |
4659 | /* | |
41a2d6cf | 4660 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4661 | * this because either it can't run here any more (set_cpus_allowed() |
4662 | * away from this CPU, or CPU going down), or because we're | |
4663 | * attempting to rebalance this task on exec (sched_exec). | |
4664 | * | |
4665 | * So we race with normal scheduler movements, but that's OK, as long | |
4666 | * as the task is no longer on this CPU. | |
efc30814 KK |
4667 | * |
4668 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4669 | */ |
efc30814 | 4670 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4671 | { |
70b97a7f | 4672 | struct rq *rq_dest, *rq_src; |
e2912009 | 4673 | int ret = 0; |
1da177e4 | 4674 | |
e761b772 | 4675 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4676 | return ret; |
1da177e4 LT |
4677 | |
4678 | rq_src = cpu_rq(src_cpu); | |
4679 | rq_dest = cpu_rq(dest_cpu); | |
4680 | ||
0122ec5b | 4681 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4682 | double_rq_lock(rq_src, rq_dest); |
4683 | /* Already moved. */ | |
4684 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4685 | goto done; |
1da177e4 | 4686 | /* Affinity changed (again). */ |
fa17b507 | 4687 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4688 | goto fail; |
1da177e4 | 4689 | |
e2912009 PZ |
4690 | /* |
4691 | * If we're not on a rq, the next wake-up will ensure we're | |
4692 | * placed properly. | |
4693 | */ | |
fd2f4419 | 4694 | if (p->on_rq) { |
4ca9b72b | 4695 | dequeue_task(rq_src, p, 0); |
e2912009 | 4696 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4697 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4698 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4699 | } |
b1e38734 | 4700 | done: |
efc30814 | 4701 | ret = 1; |
b1e38734 | 4702 | fail: |
1da177e4 | 4703 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4704 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4705 | return ret; |
1da177e4 LT |
4706 | } |
4707 | ||
e6628d5b MG |
4708 | #ifdef CONFIG_NUMA_BALANCING |
4709 | /* Migrate current task p to target_cpu */ | |
4710 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4711 | { | |
4712 | struct migration_arg arg = { p, target_cpu }; | |
4713 | int curr_cpu = task_cpu(p); | |
4714 | ||
4715 | if (curr_cpu == target_cpu) | |
4716 | return 0; | |
4717 | ||
4718 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4719 | return -EINVAL; | |
4720 | ||
4721 | /* TODO: This is not properly updating schedstats */ | |
4722 | ||
286549dc | 4723 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4724 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4725 | } | |
0ec8aa00 PZ |
4726 | |
4727 | /* | |
4728 | * Requeue a task on a given node and accurately track the number of NUMA | |
4729 | * tasks on the runqueues | |
4730 | */ | |
4731 | void sched_setnuma(struct task_struct *p, int nid) | |
4732 | { | |
4733 | struct rq *rq; | |
4734 | unsigned long flags; | |
4735 | bool on_rq, running; | |
4736 | ||
4737 | rq = task_rq_lock(p, &flags); | |
4738 | on_rq = p->on_rq; | |
4739 | running = task_current(rq, p); | |
4740 | ||
4741 | if (on_rq) | |
4742 | dequeue_task(rq, p, 0); | |
4743 | if (running) | |
4744 | p->sched_class->put_prev_task(rq, p); | |
4745 | ||
4746 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4747 | |
4748 | if (running) | |
4749 | p->sched_class->set_curr_task(rq); | |
4750 | if (on_rq) | |
4751 | enqueue_task(rq, p, 0); | |
4752 | task_rq_unlock(rq, p, &flags); | |
4753 | } | |
e6628d5b MG |
4754 | #endif |
4755 | ||
1da177e4 | 4756 | /* |
969c7921 TH |
4757 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4758 | * and performs thread migration by bumping thread off CPU then | |
4759 | * 'pushing' onto another runqueue. | |
1da177e4 | 4760 | */ |
969c7921 | 4761 | static int migration_cpu_stop(void *data) |
1da177e4 | 4762 | { |
969c7921 | 4763 | struct migration_arg *arg = data; |
f7b4cddc | 4764 | |
969c7921 TH |
4765 | /* |
4766 | * The original target cpu might have gone down and we might | |
4767 | * be on another cpu but it doesn't matter. | |
4768 | */ | |
f7b4cddc | 4769 | local_irq_disable(); |
969c7921 | 4770 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4771 | local_irq_enable(); |
1da177e4 | 4772 | return 0; |
f7b4cddc ON |
4773 | } |
4774 | ||
1da177e4 | 4775 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4776 | |
054b9108 | 4777 | /* |
48c5ccae PZ |
4778 | * Ensures that the idle task is using init_mm right before its cpu goes |
4779 | * offline. | |
054b9108 | 4780 | */ |
48c5ccae | 4781 | void idle_task_exit(void) |
1da177e4 | 4782 | { |
48c5ccae | 4783 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4784 | |
48c5ccae | 4785 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4786 | |
a53efe5f | 4787 | if (mm != &init_mm) { |
48c5ccae | 4788 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4789 | finish_arch_post_lock_switch(); |
4790 | } | |
48c5ccae | 4791 | mmdrop(mm); |
1da177e4 LT |
4792 | } |
4793 | ||
4794 | /* | |
5d180232 PZ |
4795 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4796 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4797 | * nr_active count is stable. | |
4798 | * | |
4799 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4800 | */ |
5d180232 | 4801 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4802 | { |
5d180232 PZ |
4803 | long delta = calc_load_fold_active(rq); |
4804 | if (delta) | |
4805 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4806 | } |
4807 | ||
3f1d2a31 PZ |
4808 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4809 | { | |
4810 | } | |
4811 | ||
4812 | static const struct sched_class fake_sched_class = { | |
4813 | .put_prev_task = put_prev_task_fake, | |
4814 | }; | |
4815 | ||
4816 | static struct task_struct fake_task = { | |
4817 | /* | |
4818 | * Avoid pull_{rt,dl}_task() | |
4819 | */ | |
4820 | .prio = MAX_PRIO + 1, | |
4821 | .sched_class = &fake_sched_class, | |
4822 | }; | |
4823 | ||
48f24c4d | 4824 | /* |
48c5ccae PZ |
4825 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4826 | * try_to_wake_up()->select_task_rq(). | |
4827 | * | |
4828 | * Called with rq->lock held even though we'er in stop_machine() and | |
4829 | * there's no concurrency possible, we hold the required locks anyway | |
4830 | * because of lock validation efforts. | |
1da177e4 | 4831 | */ |
48c5ccae | 4832 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4833 | { |
70b97a7f | 4834 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4835 | struct task_struct *next, *stop = rq->stop; |
4836 | int dest_cpu; | |
1da177e4 LT |
4837 | |
4838 | /* | |
48c5ccae PZ |
4839 | * Fudge the rq selection such that the below task selection loop |
4840 | * doesn't get stuck on the currently eligible stop task. | |
4841 | * | |
4842 | * We're currently inside stop_machine() and the rq is either stuck | |
4843 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4844 | * either way we should never end up calling schedule() until we're | |
4845 | * done here. | |
1da177e4 | 4846 | */ |
48c5ccae | 4847 | rq->stop = NULL; |
48f24c4d | 4848 | |
77bd3970 FW |
4849 | /* |
4850 | * put_prev_task() and pick_next_task() sched | |
4851 | * class method both need to have an up-to-date | |
4852 | * value of rq->clock[_task] | |
4853 | */ | |
4854 | update_rq_clock(rq); | |
4855 | ||
dd41f596 | 4856 | for ( ; ; ) { |
48c5ccae PZ |
4857 | /* |
4858 | * There's this thread running, bail when that's the only | |
4859 | * remaining thread. | |
4860 | */ | |
4861 | if (rq->nr_running == 1) | |
dd41f596 | 4862 | break; |
48c5ccae | 4863 | |
3f1d2a31 | 4864 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4865 | BUG_ON(!next); |
79c53799 | 4866 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4867 | |
48c5ccae PZ |
4868 | /* Find suitable destination for @next, with force if needed. */ |
4869 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4870 | raw_spin_unlock(&rq->lock); | |
4871 | ||
4872 | __migrate_task(next, dead_cpu, dest_cpu); | |
4873 | ||
4874 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4875 | } |
dce48a84 | 4876 | |
48c5ccae | 4877 | rq->stop = stop; |
dce48a84 | 4878 | } |
48c5ccae | 4879 | |
1da177e4 LT |
4880 | #endif /* CONFIG_HOTPLUG_CPU */ |
4881 | ||
e692ab53 NP |
4882 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4883 | ||
4884 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4885 | { |
4886 | .procname = "sched_domain", | |
c57baf1e | 4887 | .mode = 0555, |
e0361851 | 4888 | }, |
56992309 | 4889 | {} |
e692ab53 NP |
4890 | }; |
4891 | ||
4892 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4893 | { |
4894 | .procname = "kernel", | |
c57baf1e | 4895 | .mode = 0555, |
e0361851 AD |
4896 | .child = sd_ctl_dir, |
4897 | }, | |
56992309 | 4898 | {} |
e692ab53 NP |
4899 | }; |
4900 | ||
4901 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4902 | { | |
4903 | struct ctl_table *entry = | |
5cf9f062 | 4904 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4905 | |
e692ab53 NP |
4906 | return entry; |
4907 | } | |
4908 | ||
6382bc90 MM |
4909 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4910 | { | |
cd790076 | 4911 | struct ctl_table *entry; |
6382bc90 | 4912 | |
cd790076 MM |
4913 | /* |
4914 | * In the intermediate directories, both the child directory and | |
4915 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4916 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4917 | * static strings and all have proc handlers. |
4918 | */ | |
4919 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4920 | if (entry->child) |
4921 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4922 | if (entry->proc_handler == NULL) |
4923 | kfree(entry->procname); | |
4924 | } | |
6382bc90 MM |
4925 | |
4926 | kfree(*tablep); | |
4927 | *tablep = NULL; | |
4928 | } | |
4929 | ||
201c373e | 4930 | static int min_load_idx = 0; |
fd9b86d3 | 4931 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4932 | |
e692ab53 | 4933 | static void |
e0361851 | 4934 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4935 | const char *procname, void *data, int maxlen, |
201c373e NK |
4936 | umode_t mode, proc_handler *proc_handler, |
4937 | bool load_idx) | |
e692ab53 | 4938 | { |
e692ab53 NP |
4939 | entry->procname = procname; |
4940 | entry->data = data; | |
4941 | entry->maxlen = maxlen; | |
4942 | entry->mode = mode; | |
4943 | entry->proc_handler = proc_handler; | |
201c373e NK |
4944 | |
4945 | if (load_idx) { | |
4946 | entry->extra1 = &min_load_idx; | |
4947 | entry->extra2 = &max_load_idx; | |
4948 | } | |
e692ab53 NP |
4949 | } |
4950 | ||
4951 | static struct ctl_table * | |
4952 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4953 | { | |
37e6bae8 | 4954 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4955 | |
ad1cdc1d MM |
4956 | if (table == NULL) |
4957 | return NULL; | |
4958 | ||
e0361851 | 4959 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4960 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4961 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4962 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4963 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4964 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4965 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4966 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4967 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4968 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4969 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4970 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4971 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4972 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4973 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4974 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4975 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4976 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4977 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4978 | &sd->cache_nice_tries, |
201c373e | 4979 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4980 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4981 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4982 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4983 | &sd->max_newidle_lb_cost, | |
4984 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4985 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4986 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4987 | /* &table[13] is terminator */ |
e692ab53 NP |
4988 | |
4989 | return table; | |
4990 | } | |
4991 | ||
be7002e6 | 4992 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4993 | { |
4994 | struct ctl_table *entry, *table; | |
4995 | struct sched_domain *sd; | |
4996 | int domain_num = 0, i; | |
4997 | char buf[32]; | |
4998 | ||
4999 | for_each_domain(cpu, sd) | |
5000 | domain_num++; | |
5001 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5002 | if (table == NULL) |
5003 | return NULL; | |
e692ab53 NP |
5004 | |
5005 | i = 0; | |
5006 | for_each_domain(cpu, sd) { | |
5007 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5008 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5009 | entry->mode = 0555; |
e692ab53 NP |
5010 | entry->child = sd_alloc_ctl_domain_table(sd); |
5011 | entry++; | |
5012 | i++; | |
5013 | } | |
5014 | return table; | |
5015 | } | |
5016 | ||
5017 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5018 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5019 | { |
6ad4c188 | 5020 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5021 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5022 | char buf[32]; | |
5023 | ||
7378547f MM |
5024 | WARN_ON(sd_ctl_dir[0].child); |
5025 | sd_ctl_dir[0].child = entry; | |
5026 | ||
ad1cdc1d MM |
5027 | if (entry == NULL) |
5028 | return; | |
5029 | ||
6ad4c188 | 5030 | for_each_possible_cpu(i) { |
e692ab53 | 5031 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5032 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5033 | entry->mode = 0555; |
e692ab53 | 5034 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5035 | entry++; |
e692ab53 | 5036 | } |
7378547f MM |
5037 | |
5038 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5039 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5040 | } | |
6382bc90 | 5041 | |
7378547f | 5042 | /* may be called multiple times per register */ |
6382bc90 MM |
5043 | static void unregister_sched_domain_sysctl(void) |
5044 | { | |
7378547f MM |
5045 | if (sd_sysctl_header) |
5046 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5047 | sd_sysctl_header = NULL; |
7378547f MM |
5048 | if (sd_ctl_dir[0].child) |
5049 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5050 | } |
e692ab53 | 5051 | #else |
6382bc90 MM |
5052 | static void register_sched_domain_sysctl(void) |
5053 | { | |
5054 | } | |
5055 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5056 | { |
5057 | } | |
5058 | #endif | |
5059 | ||
1f11eb6a GH |
5060 | static void set_rq_online(struct rq *rq) |
5061 | { | |
5062 | if (!rq->online) { | |
5063 | const struct sched_class *class; | |
5064 | ||
c6c4927b | 5065 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5066 | rq->online = 1; |
5067 | ||
5068 | for_each_class(class) { | |
5069 | if (class->rq_online) | |
5070 | class->rq_online(rq); | |
5071 | } | |
5072 | } | |
5073 | } | |
5074 | ||
5075 | static void set_rq_offline(struct rq *rq) | |
5076 | { | |
5077 | if (rq->online) { | |
5078 | const struct sched_class *class; | |
5079 | ||
5080 | for_each_class(class) { | |
5081 | if (class->rq_offline) | |
5082 | class->rq_offline(rq); | |
5083 | } | |
5084 | ||
c6c4927b | 5085 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5086 | rq->online = 0; |
5087 | } | |
5088 | } | |
5089 | ||
1da177e4 LT |
5090 | /* |
5091 | * migration_call - callback that gets triggered when a CPU is added. | |
5092 | * Here we can start up the necessary migration thread for the new CPU. | |
5093 | */ | |
0db0628d | 5094 | static int |
48f24c4d | 5095 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5096 | { |
48f24c4d | 5097 | int cpu = (long)hcpu; |
1da177e4 | 5098 | unsigned long flags; |
969c7921 | 5099 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5100 | |
48c5ccae | 5101 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5102 | |
1da177e4 | 5103 | case CPU_UP_PREPARE: |
a468d389 | 5104 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5105 | break; |
48f24c4d | 5106 | |
1da177e4 | 5107 | case CPU_ONLINE: |
1f94ef59 | 5108 | /* Update our root-domain */ |
05fa785c | 5109 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5110 | if (rq->rd) { |
c6c4927b | 5111 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5112 | |
5113 | set_rq_online(rq); | |
1f94ef59 | 5114 | } |
05fa785c | 5115 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5116 | break; |
48f24c4d | 5117 | |
1da177e4 | 5118 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5119 | case CPU_DYING: |
317f3941 | 5120 | sched_ttwu_pending(); |
57d885fe | 5121 | /* Update our root-domain */ |
05fa785c | 5122 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5123 | if (rq->rd) { |
c6c4927b | 5124 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5125 | set_rq_offline(rq); |
57d885fe | 5126 | } |
48c5ccae PZ |
5127 | migrate_tasks(cpu); |
5128 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5129 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5130 | break; |
48c5ccae | 5131 | |
5d180232 | 5132 | case CPU_DEAD: |
f319da0c | 5133 | calc_load_migrate(rq); |
57d885fe | 5134 | break; |
1da177e4 LT |
5135 | #endif |
5136 | } | |
49c022e6 PZ |
5137 | |
5138 | update_max_interval(); | |
5139 | ||
1da177e4 LT |
5140 | return NOTIFY_OK; |
5141 | } | |
5142 | ||
f38b0820 PM |
5143 | /* |
5144 | * Register at high priority so that task migration (migrate_all_tasks) | |
5145 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5146 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5147 | */ |
0db0628d | 5148 | static struct notifier_block migration_notifier = { |
1da177e4 | 5149 | .notifier_call = migration_call, |
50a323b7 | 5150 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5151 | }; |
5152 | ||
a803f026 CM |
5153 | static void __cpuinit set_cpu_rq_start_time(void) |
5154 | { | |
5155 | int cpu = smp_processor_id(); | |
5156 | struct rq *rq = cpu_rq(cpu); | |
5157 | rq->age_stamp = sched_clock_cpu(cpu); | |
5158 | } | |
5159 | ||
0db0628d | 5160 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5161 | unsigned long action, void *hcpu) |
5162 | { | |
5163 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5164 | case CPU_STARTING: |
5165 | set_cpu_rq_start_time(); | |
5166 | return NOTIFY_OK; | |
3a101d05 TH |
5167 | case CPU_DOWN_FAILED: |
5168 | set_cpu_active((long)hcpu, true); | |
5169 | return NOTIFY_OK; | |
5170 | default: | |
5171 | return NOTIFY_DONE; | |
5172 | } | |
5173 | } | |
5174 | ||
0db0628d | 5175 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5176 | unsigned long action, void *hcpu) |
5177 | { | |
de212f18 PZ |
5178 | unsigned long flags; |
5179 | long cpu = (long)hcpu; | |
5180 | ||
3a101d05 TH |
5181 | switch (action & ~CPU_TASKS_FROZEN) { |
5182 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5183 | set_cpu_active(cpu, false); |
5184 | ||
5185 | /* explicitly allow suspend */ | |
5186 | if (!(action & CPU_TASKS_FROZEN)) { | |
5187 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5188 | bool overflow; | |
5189 | int cpus; | |
5190 | ||
5191 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5192 | cpus = dl_bw_cpus(cpu); | |
5193 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5194 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5195 | ||
5196 | if (overflow) | |
5197 | return notifier_from_errno(-EBUSY); | |
5198 | } | |
3a101d05 | 5199 | return NOTIFY_OK; |
3a101d05 | 5200 | } |
de212f18 PZ |
5201 | |
5202 | return NOTIFY_DONE; | |
3a101d05 TH |
5203 | } |
5204 | ||
7babe8db | 5205 | static int __init migration_init(void) |
1da177e4 LT |
5206 | { |
5207 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5208 | int err; |
48f24c4d | 5209 | |
3a101d05 | 5210 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5211 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5212 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5213 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5214 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5215 | |
3a101d05 TH |
5216 | /* Register cpu active notifiers */ |
5217 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5218 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5219 | ||
a004cd42 | 5220 | return 0; |
1da177e4 | 5221 | } |
7babe8db | 5222 | early_initcall(migration_init); |
1da177e4 LT |
5223 | #endif |
5224 | ||
5225 | #ifdef CONFIG_SMP | |
476f3534 | 5226 | |
4cb98839 PZ |
5227 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5228 | ||
3e9830dc | 5229 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5230 | |
d039ac60 | 5231 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5232 | |
d039ac60 | 5233 | static int __init sched_debug_setup(char *str) |
f6630114 | 5234 | { |
d039ac60 | 5235 | sched_debug_enabled = 1; |
f6630114 MT |
5236 | |
5237 | return 0; | |
5238 | } | |
d039ac60 PZ |
5239 | early_param("sched_debug", sched_debug_setup); |
5240 | ||
5241 | static inline bool sched_debug(void) | |
5242 | { | |
5243 | return sched_debug_enabled; | |
5244 | } | |
f6630114 | 5245 | |
7c16ec58 | 5246 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5247 | struct cpumask *groupmask) |
1da177e4 | 5248 | { |
4dcf6aff | 5249 | struct sched_group *group = sd->groups; |
434d53b0 | 5250 | char str[256]; |
1da177e4 | 5251 | |
968ea6d8 | 5252 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5253 | cpumask_clear(groupmask); |
4dcf6aff IM |
5254 | |
5255 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5256 | ||
5257 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5258 | printk("does not load-balance\n"); |
4dcf6aff | 5259 | if (sd->parent) |
3df0fc5b PZ |
5260 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5261 | " has parent"); | |
4dcf6aff | 5262 | return -1; |
41c7ce9a NP |
5263 | } |
5264 | ||
3df0fc5b | 5265 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5266 | |
758b2cdc | 5267 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5268 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5269 | "CPU%d\n", cpu); | |
4dcf6aff | 5270 | } |
758b2cdc | 5271 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5272 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5273 | " CPU%d\n", cpu); | |
4dcf6aff | 5274 | } |
1da177e4 | 5275 | |
4dcf6aff | 5276 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5277 | do { |
4dcf6aff | 5278 | if (!group) { |
3df0fc5b PZ |
5279 | printk("\n"); |
5280 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5281 | break; |
5282 | } | |
5283 | ||
c3decf0d | 5284 | /* |
63b2ca30 NP |
5285 | * Even though we initialize ->capacity to something semi-sane, |
5286 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5287 | * domain iteration is still funny without causing /0 traps. |
5288 | */ | |
63b2ca30 | 5289 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5290 | printk(KERN_CONT "\n"); |
63b2ca30 | 5291 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5292 | break; |
5293 | } | |
1da177e4 | 5294 | |
758b2cdc | 5295 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5296 | printk(KERN_CONT "\n"); |
5297 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5298 | break; |
5299 | } | |
1da177e4 | 5300 | |
cb83b629 PZ |
5301 | if (!(sd->flags & SD_OVERLAP) && |
5302 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5303 | printk(KERN_CONT "\n"); |
5304 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5305 | break; |
5306 | } | |
1da177e4 | 5307 | |
758b2cdc | 5308 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5309 | |
968ea6d8 | 5310 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5311 | |
3df0fc5b | 5312 | printk(KERN_CONT " %s", str); |
ca8ce3d0 | 5313 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5314 | printk(KERN_CONT " (cpu_capacity = %d)", |
5315 | group->sgc->capacity); | |
381512cf | 5316 | } |
1da177e4 | 5317 | |
4dcf6aff IM |
5318 | group = group->next; |
5319 | } while (group != sd->groups); | |
3df0fc5b | 5320 | printk(KERN_CONT "\n"); |
1da177e4 | 5321 | |
758b2cdc | 5322 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5323 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5324 | |
758b2cdc RR |
5325 | if (sd->parent && |
5326 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5327 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5328 | "of domain->span\n"); | |
4dcf6aff IM |
5329 | return 0; |
5330 | } | |
1da177e4 | 5331 | |
4dcf6aff IM |
5332 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5333 | { | |
5334 | int level = 0; | |
1da177e4 | 5335 | |
d039ac60 | 5336 | if (!sched_debug_enabled) |
f6630114 MT |
5337 | return; |
5338 | ||
4dcf6aff IM |
5339 | if (!sd) { |
5340 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5341 | return; | |
5342 | } | |
1da177e4 | 5343 | |
4dcf6aff IM |
5344 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5345 | ||
5346 | for (;;) { | |
4cb98839 | 5347 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5348 | break; |
1da177e4 LT |
5349 | level++; |
5350 | sd = sd->parent; | |
33859f7f | 5351 | if (!sd) |
4dcf6aff IM |
5352 | break; |
5353 | } | |
1da177e4 | 5354 | } |
6d6bc0ad | 5355 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5356 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5357 | static inline bool sched_debug(void) |
5358 | { | |
5359 | return false; | |
5360 | } | |
6d6bc0ad | 5361 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5362 | |
1a20ff27 | 5363 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5364 | { |
758b2cdc | 5365 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5366 | return 1; |
5367 | ||
5368 | /* Following flags need at least 2 groups */ | |
5369 | if (sd->flags & (SD_LOAD_BALANCE | | |
5370 | SD_BALANCE_NEWIDLE | | |
5371 | SD_BALANCE_FORK | | |
89c4710e | 5372 | SD_BALANCE_EXEC | |
5d4dfddd | 5373 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5374 | SD_SHARE_PKG_RESOURCES | |
5375 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5376 | if (sd->groups != sd->groups->next) |
5377 | return 0; | |
5378 | } | |
5379 | ||
5380 | /* Following flags don't use groups */ | |
c88d5910 | 5381 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5382 | return 0; |
5383 | ||
5384 | return 1; | |
5385 | } | |
5386 | ||
48f24c4d IM |
5387 | static int |
5388 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5389 | { |
5390 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5391 | ||
5392 | if (sd_degenerate(parent)) | |
5393 | return 1; | |
5394 | ||
758b2cdc | 5395 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5396 | return 0; |
5397 | ||
245af2c7 SS |
5398 | /* Flags needing groups don't count if only 1 group in parent */ |
5399 | if (parent->groups == parent->groups->next) { | |
5400 | pflags &= ~(SD_LOAD_BALANCE | | |
5401 | SD_BALANCE_NEWIDLE | | |
5402 | SD_BALANCE_FORK | | |
89c4710e | 5403 | SD_BALANCE_EXEC | |
5d4dfddd | 5404 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5405 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5406 | SD_PREFER_SIBLING | |
5407 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5408 | if (nr_node_ids == 1) |
5409 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5410 | } |
5411 | if (~cflags & pflags) | |
5412 | return 0; | |
5413 | ||
5414 | return 1; | |
5415 | } | |
5416 | ||
dce840a0 | 5417 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5418 | { |
dce840a0 | 5419 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5420 | |
68e74568 | 5421 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5422 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5423 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5424 | free_cpumask_var(rd->rto_mask); |
5425 | free_cpumask_var(rd->online); | |
5426 | free_cpumask_var(rd->span); | |
5427 | kfree(rd); | |
5428 | } | |
5429 | ||
57d885fe GH |
5430 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5431 | { | |
a0490fa3 | 5432 | struct root_domain *old_rd = NULL; |
57d885fe | 5433 | unsigned long flags; |
57d885fe | 5434 | |
05fa785c | 5435 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5436 | |
5437 | if (rq->rd) { | |
a0490fa3 | 5438 | old_rd = rq->rd; |
57d885fe | 5439 | |
c6c4927b | 5440 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5441 | set_rq_offline(rq); |
57d885fe | 5442 | |
c6c4927b | 5443 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5444 | |
a0490fa3 | 5445 | /* |
0515973f | 5446 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5447 | * set old_rd to NULL to skip the freeing later |
5448 | * in this function: | |
5449 | */ | |
5450 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5451 | old_rd = NULL; | |
57d885fe GH |
5452 | } |
5453 | ||
5454 | atomic_inc(&rd->refcount); | |
5455 | rq->rd = rd; | |
5456 | ||
c6c4927b | 5457 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5458 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5459 | set_rq_online(rq); |
57d885fe | 5460 | |
05fa785c | 5461 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5462 | |
5463 | if (old_rd) | |
dce840a0 | 5464 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5465 | } |
5466 | ||
68c38fc3 | 5467 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5468 | { |
5469 | memset(rd, 0, sizeof(*rd)); | |
5470 | ||
68c38fc3 | 5471 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5472 | goto out; |
68c38fc3 | 5473 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5474 | goto free_span; |
1baca4ce | 5475 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5476 | goto free_online; |
1baca4ce JL |
5477 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5478 | goto free_dlo_mask; | |
6e0534f2 | 5479 | |
332ac17e | 5480 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5481 | if (cpudl_init(&rd->cpudl) != 0) |
5482 | goto free_dlo_mask; | |
332ac17e | 5483 | |
68c38fc3 | 5484 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5485 | goto free_rto_mask; |
c6c4927b | 5486 | return 0; |
6e0534f2 | 5487 | |
68e74568 RR |
5488 | free_rto_mask: |
5489 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5490 | free_dlo_mask: |
5491 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5492 | free_online: |
5493 | free_cpumask_var(rd->online); | |
5494 | free_span: | |
5495 | free_cpumask_var(rd->span); | |
0c910d28 | 5496 | out: |
c6c4927b | 5497 | return -ENOMEM; |
57d885fe GH |
5498 | } |
5499 | ||
029632fb PZ |
5500 | /* |
5501 | * By default the system creates a single root-domain with all cpus as | |
5502 | * members (mimicking the global state we have today). | |
5503 | */ | |
5504 | struct root_domain def_root_domain; | |
5505 | ||
57d885fe GH |
5506 | static void init_defrootdomain(void) |
5507 | { | |
68c38fc3 | 5508 | init_rootdomain(&def_root_domain); |
c6c4927b | 5509 | |
57d885fe GH |
5510 | atomic_set(&def_root_domain.refcount, 1); |
5511 | } | |
5512 | ||
dc938520 | 5513 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5514 | { |
5515 | struct root_domain *rd; | |
5516 | ||
5517 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5518 | if (!rd) | |
5519 | return NULL; | |
5520 | ||
68c38fc3 | 5521 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5522 | kfree(rd); |
5523 | return NULL; | |
5524 | } | |
57d885fe GH |
5525 | |
5526 | return rd; | |
5527 | } | |
5528 | ||
63b2ca30 | 5529 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5530 | { |
5531 | struct sched_group *tmp, *first; | |
5532 | ||
5533 | if (!sg) | |
5534 | return; | |
5535 | ||
5536 | first = sg; | |
5537 | do { | |
5538 | tmp = sg->next; | |
5539 | ||
63b2ca30 NP |
5540 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5541 | kfree(sg->sgc); | |
e3589f6c PZ |
5542 | |
5543 | kfree(sg); | |
5544 | sg = tmp; | |
5545 | } while (sg != first); | |
5546 | } | |
5547 | ||
dce840a0 PZ |
5548 | static void free_sched_domain(struct rcu_head *rcu) |
5549 | { | |
5550 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5551 | |
5552 | /* | |
5553 | * If its an overlapping domain it has private groups, iterate and | |
5554 | * nuke them all. | |
5555 | */ | |
5556 | if (sd->flags & SD_OVERLAP) { | |
5557 | free_sched_groups(sd->groups, 1); | |
5558 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5559 | kfree(sd->groups->sgc); |
dce840a0 | 5560 | kfree(sd->groups); |
9c3f75cb | 5561 | } |
dce840a0 PZ |
5562 | kfree(sd); |
5563 | } | |
5564 | ||
5565 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5566 | { | |
5567 | call_rcu(&sd->rcu, free_sched_domain); | |
5568 | } | |
5569 | ||
5570 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5571 | { | |
5572 | for (; sd; sd = sd->parent) | |
5573 | destroy_sched_domain(sd, cpu); | |
5574 | } | |
5575 | ||
518cd623 PZ |
5576 | /* |
5577 | * Keep a special pointer to the highest sched_domain that has | |
5578 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5579 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5580 | * | |
5581 | * Also keep a unique ID per domain (we use the first cpu number in | |
5582 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5583 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5584 | */ |
5585 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5586 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5587 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5588 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5589 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5590 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5591 | |
5592 | static void update_top_cache_domain(int cpu) | |
5593 | { | |
5594 | struct sched_domain *sd; | |
5d4cf996 | 5595 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5596 | int id = cpu; |
7d9ffa89 | 5597 | int size = 1; |
518cd623 PZ |
5598 | |
5599 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5600 | if (sd) { |
518cd623 | 5601 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5602 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5603 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5604 | } |
5d4cf996 | 5605 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5606 | |
5607 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5608 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5609 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5610 | |
5611 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5612 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5613 | |
5614 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5615 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5616 | } |
5617 | ||
1da177e4 | 5618 | /* |
0eab9146 | 5619 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5620 | * hold the hotplug lock. |
5621 | */ | |
0eab9146 IM |
5622 | static void |
5623 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5624 | { |
70b97a7f | 5625 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5626 | struct sched_domain *tmp; |
5627 | ||
5628 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5629 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5630 | struct sched_domain *parent = tmp->parent; |
5631 | if (!parent) | |
5632 | break; | |
f29c9b1c | 5633 | |
1a848870 | 5634 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5635 | tmp->parent = parent->parent; |
1a848870 SS |
5636 | if (parent->parent) |
5637 | parent->parent->child = tmp; | |
10866e62 PZ |
5638 | /* |
5639 | * Transfer SD_PREFER_SIBLING down in case of a | |
5640 | * degenerate parent; the spans match for this | |
5641 | * so the property transfers. | |
5642 | */ | |
5643 | if (parent->flags & SD_PREFER_SIBLING) | |
5644 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5645 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5646 | } else |
5647 | tmp = tmp->parent; | |
245af2c7 SS |
5648 | } |
5649 | ||
1a848870 | 5650 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5651 | tmp = sd; |
245af2c7 | 5652 | sd = sd->parent; |
dce840a0 | 5653 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5654 | if (sd) |
5655 | sd->child = NULL; | |
5656 | } | |
1da177e4 | 5657 | |
4cb98839 | 5658 | sched_domain_debug(sd, cpu); |
1da177e4 | 5659 | |
57d885fe | 5660 | rq_attach_root(rq, rd); |
dce840a0 | 5661 | tmp = rq->sd; |
674311d5 | 5662 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5663 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5664 | |
5665 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5666 | } |
5667 | ||
5668 | /* cpus with isolated domains */ | |
dcc30a35 | 5669 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5670 | |
5671 | /* Setup the mask of cpus configured for isolated domains */ | |
5672 | static int __init isolated_cpu_setup(char *str) | |
5673 | { | |
bdddd296 | 5674 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5675 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5676 | return 1; |
5677 | } | |
5678 | ||
8927f494 | 5679 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5680 | |
49a02c51 | 5681 | struct s_data { |
21d42ccf | 5682 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5683 | struct root_domain *rd; |
5684 | }; | |
5685 | ||
2109b99e | 5686 | enum s_alloc { |
2109b99e | 5687 | sa_rootdomain, |
21d42ccf | 5688 | sa_sd, |
dce840a0 | 5689 | sa_sd_storage, |
2109b99e AH |
5690 | sa_none, |
5691 | }; | |
5692 | ||
c1174876 PZ |
5693 | /* |
5694 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5695 | * domain traversal. | |
5696 | * | |
5697 | * Asymmetric node setups can result in situations where the domain tree is of | |
5698 | * unequal depth, make sure to skip domains that already cover the entire | |
5699 | * range. | |
5700 | * | |
5701 | * In that case build_sched_domains() will have terminated the iteration early | |
5702 | * and our sibling sd spans will be empty. Domains should always include the | |
5703 | * cpu they're built on, so check that. | |
5704 | * | |
5705 | */ | |
5706 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5707 | { | |
5708 | const struct cpumask *span = sched_domain_span(sd); | |
5709 | struct sd_data *sdd = sd->private; | |
5710 | struct sched_domain *sibling; | |
5711 | int i; | |
5712 | ||
5713 | for_each_cpu(i, span) { | |
5714 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5715 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5716 | continue; | |
5717 | ||
5718 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5719 | } | |
5720 | } | |
5721 | ||
5722 | /* | |
5723 | * Return the canonical balance cpu for this group, this is the first cpu | |
5724 | * of this group that's also in the iteration mask. | |
5725 | */ | |
5726 | int group_balance_cpu(struct sched_group *sg) | |
5727 | { | |
5728 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5729 | } | |
5730 | ||
e3589f6c PZ |
5731 | static int |
5732 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5733 | { | |
5734 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5735 | const struct cpumask *span = sched_domain_span(sd); | |
5736 | struct cpumask *covered = sched_domains_tmpmask; | |
5737 | struct sd_data *sdd = sd->private; | |
5738 | struct sched_domain *child; | |
5739 | int i; | |
5740 | ||
5741 | cpumask_clear(covered); | |
5742 | ||
5743 | for_each_cpu(i, span) { | |
5744 | struct cpumask *sg_span; | |
5745 | ||
5746 | if (cpumask_test_cpu(i, covered)) | |
5747 | continue; | |
5748 | ||
c1174876 PZ |
5749 | child = *per_cpu_ptr(sdd->sd, i); |
5750 | ||
5751 | /* See the comment near build_group_mask(). */ | |
5752 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5753 | continue; | |
5754 | ||
e3589f6c | 5755 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5756 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5757 | |
5758 | if (!sg) | |
5759 | goto fail; | |
5760 | ||
5761 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5762 | if (child->child) { |
5763 | child = child->child; | |
5764 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5765 | } else | |
5766 | cpumask_set_cpu(i, sg_span); | |
5767 | ||
5768 | cpumask_or(covered, covered, sg_span); | |
5769 | ||
63b2ca30 NP |
5770 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5771 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5772 | build_group_mask(sd, sg); |
5773 | ||
c3decf0d | 5774 | /* |
63b2ca30 | 5775 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5776 | * domains and no possible iteration will get us here, we won't |
5777 | * die on a /0 trap. | |
5778 | */ | |
ca8ce3d0 | 5779 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
63b2ca30 | 5780 | sg->sgc->capacity_orig = sg->sgc->capacity; |
e3589f6c | 5781 | |
c1174876 PZ |
5782 | /* |
5783 | * Make sure the first group of this domain contains the | |
5784 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5785 | * breaks. See update_sg_lb_stats(). | |
5786 | */ | |
74a5ce20 | 5787 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5788 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5789 | groups = sg; |
5790 | ||
5791 | if (!first) | |
5792 | first = sg; | |
5793 | if (last) | |
5794 | last->next = sg; | |
5795 | last = sg; | |
5796 | last->next = first; | |
5797 | } | |
5798 | sd->groups = groups; | |
5799 | ||
5800 | return 0; | |
5801 | ||
5802 | fail: | |
5803 | free_sched_groups(first, 0); | |
5804 | ||
5805 | return -ENOMEM; | |
5806 | } | |
5807 | ||
dce840a0 | 5808 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5809 | { |
dce840a0 PZ |
5810 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5811 | struct sched_domain *child = sd->child; | |
1da177e4 | 5812 | |
dce840a0 PZ |
5813 | if (child) |
5814 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5815 | |
9c3f75cb | 5816 | if (sg) { |
dce840a0 | 5817 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5818 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5819 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5820 | } |
dce840a0 PZ |
5821 | |
5822 | return cpu; | |
1e9f28fa | 5823 | } |
1e9f28fa | 5824 | |
01a08546 | 5825 | /* |
dce840a0 PZ |
5826 | * build_sched_groups will build a circular linked list of the groups |
5827 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5828 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5829 | * |
5830 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5831 | */ |
e3589f6c PZ |
5832 | static int |
5833 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5834 | { |
dce840a0 PZ |
5835 | struct sched_group *first = NULL, *last = NULL; |
5836 | struct sd_data *sdd = sd->private; | |
5837 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5838 | struct cpumask *covered; |
dce840a0 | 5839 | int i; |
9c1cfda2 | 5840 | |
e3589f6c PZ |
5841 | get_group(cpu, sdd, &sd->groups); |
5842 | atomic_inc(&sd->groups->ref); | |
5843 | ||
0936629f | 5844 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5845 | return 0; |
5846 | ||
f96225fd PZ |
5847 | lockdep_assert_held(&sched_domains_mutex); |
5848 | covered = sched_domains_tmpmask; | |
5849 | ||
dce840a0 | 5850 | cpumask_clear(covered); |
6711cab4 | 5851 | |
dce840a0 PZ |
5852 | for_each_cpu(i, span) { |
5853 | struct sched_group *sg; | |
cd08e923 | 5854 | int group, j; |
6711cab4 | 5855 | |
dce840a0 PZ |
5856 | if (cpumask_test_cpu(i, covered)) |
5857 | continue; | |
6711cab4 | 5858 | |
cd08e923 | 5859 | group = get_group(i, sdd, &sg); |
c1174876 | 5860 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5861 | |
dce840a0 PZ |
5862 | for_each_cpu(j, span) { |
5863 | if (get_group(j, sdd, NULL) != group) | |
5864 | continue; | |
0601a88d | 5865 | |
dce840a0 PZ |
5866 | cpumask_set_cpu(j, covered); |
5867 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5868 | } | |
0601a88d | 5869 | |
dce840a0 PZ |
5870 | if (!first) |
5871 | first = sg; | |
5872 | if (last) | |
5873 | last->next = sg; | |
5874 | last = sg; | |
5875 | } | |
5876 | last->next = first; | |
e3589f6c PZ |
5877 | |
5878 | return 0; | |
0601a88d | 5879 | } |
51888ca2 | 5880 | |
89c4710e | 5881 | /* |
63b2ca30 | 5882 | * Initialize sched groups cpu_capacity. |
89c4710e | 5883 | * |
63b2ca30 | 5884 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 5885 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
5886 | * Typically cpu_capacity for all the groups in a sched domain will be same |
5887 | * unless there are asymmetries in the topology. If there are asymmetries, | |
5888 | * group having more cpu_capacity will pickup more load compared to the | |
5889 | * group having less cpu_capacity. | |
89c4710e | 5890 | */ |
63b2ca30 | 5891 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 5892 | { |
e3589f6c | 5893 | struct sched_group *sg = sd->groups; |
89c4710e | 5894 | |
94c95ba6 | 5895 | WARN_ON(!sg); |
e3589f6c PZ |
5896 | |
5897 | do { | |
5898 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5899 | sg = sg->next; | |
5900 | } while (sg != sd->groups); | |
89c4710e | 5901 | |
c1174876 | 5902 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5903 | return; |
aae6d3dd | 5904 | |
63b2ca30 NP |
5905 | update_group_capacity(sd, cpu); |
5906 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
5907 | } |
5908 | ||
7c16ec58 MT |
5909 | /* |
5910 | * Initializers for schedule domains | |
5911 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5912 | */ | |
5913 | ||
1d3504fc | 5914 | static int default_relax_domain_level = -1; |
60495e77 | 5915 | int sched_domain_level_max; |
1d3504fc HS |
5916 | |
5917 | static int __init setup_relax_domain_level(char *str) | |
5918 | { | |
a841f8ce DS |
5919 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5920 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5921 | |
1d3504fc HS |
5922 | return 1; |
5923 | } | |
5924 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5925 | ||
5926 | static void set_domain_attribute(struct sched_domain *sd, | |
5927 | struct sched_domain_attr *attr) | |
5928 | { | |
5929 | int request; | |
5930 | ||
5931 | if (!attr || attr->relax_domain_level < 0) { | |
5932 | if (default_relax_domain_level < 0) | |
5933 | return; | |
5934 | else | |
5935 | request = default_relax_domain_level; | |
5936 | } else | |
5937 | request = attr->relax_domain_level; | |
5938 | if (request < sd->level) { | |
5939 | /* turn off idle balance on this domain */ | |
c88d5910 | 5940 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5941 | } else { |
5942 | /* turn on idle balance on this domain */ | |
c88d5910 | 5943 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5944 | } |
5945 | } | |
5946 | ||
54ab4ff4 PZ |
5947 | static void __sdt_free(const struct cpumask *cpu_map); |
5948 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5949 | ||
2109b99e AH |
5950 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5951 | const struct cpumask *cpu_map) | |
5952 | { | |
5953 | switch (what) { | |
2109b99e | 5954 | case sa_rootdomain: |
822ff793 PZ |
5955 | if (!atomic_read(&d->rd->refcount)) |
5956 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5957 | case sa_sd: |
5958 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5959 | case sa_sd_storage: |
54ab4ff4 | 5960 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5961 | case sa_none: |
5962 | break; | |
5963 | } | |
5964 | } | |
3404c8d9 | 5965 | |
2109b99e AH |
5966 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5967 | const struct cpumask *cpu_map) | |
5968 | { | |
dce840a0 PZ |
5969 | memset(d, 0, sizeof(*d)); |
5970 | ||
54ab4ff4 PZ |
5971 | if (__sdt_alloc(cpu_map)) |
5972 | return sa_sd_storage; | |
dce840a0 PZ |
5973 | d->sd = alloc_percpu(struct sched_domain *); |
5974 | if (!d->sd) | |
5975 | return sa_sd_storage; | |
2109b99e | 5976 | d->rd = alloc_rootdomain(); |
dce840a0 | 5977 | if (!d->rd) |
21d42ccf | 5978 | return sa_sd; |
2109b99e AH |
5979 | return sa_rootdomain; |
5980 | } | |
57d885fe | 5981 | |
dce840a0 PZ |
5982 | /* |
5983 | * NULL the sd_data elements we've used to build the sched_domain and | |
5984 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5985 | * will not free the data we're using. | |
5986 | */ | |
5987 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5988 | { | |
5989 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5990 | |
5991 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5992 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5993 | ||
e3589f6c | 5994 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5995 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 5996 | |
63b2ca30 NP |
5997 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
5998 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
5999 | } |
6000 | ||
cb83b629 | 6001 | #ifdef CONFIG_NUMA |
cb83b629 | 6002 | static int sched_domains_numa_levels; |
cb83b629 PZ |
6003 | static int *sched_domains_numa_distance; |
6004 | static struct cpumask ***sched_domains_numa_masks; | |
6005 | static int sched_domains_curr_level; | |
143e1e28 | 6006 | #endif |
cb83b629 | 6007 | |
143e1e28 VG |
6008 | /* |
6009 | * SD_flags allowed in topology descriptions. | |
6010 | * | |
5d4dfddd | 6011 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6012 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6013 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6014 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6015 | * |
6016 | * Odd one out: | |
6017 | * SD_ASYM_PACKING - describes SMT quirks | |
6018 | */ | |
6019 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6020 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6021 | SD_SHARE_PKG_RESOURCES | \ |
6022 | SD_NUMA | \ | |
d77b3ed5 VG |
6023 | SD_ASYM_PACKING | \ |
6024 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6025 | |
6026 | static struct sched_domain * | |
143e1e28 | 6027 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6028 | { |
6029 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6030 | int sd_weight, sd_flags = 0; |
6031 | ||
6032 | #ifdef CONFIG_NUMA | |
6033 | /* | |
6034 | * Ugly hack to pass state to sd_numa_mask()... | |
6035 | */ | |
6036 | sched_domains_curr_level = tl->numa_level; | |
6037 | #endif | |
6038 | ||
6039 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6040 | ||
6041 | if (tl->sd_flags) | |
6042 | sd_flags = (*tl->sd_flags)(); | |
6043 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6044 | "wrong sd_flags in topology description\n")) | |
6045 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6046 | |
6047 | *sd = (struct sched_domain){ | |
6048 | .min_interval = sd_weight, | |
6049 | .max_interval = 2*sd_weight, | |
6050 | .busy_factor = 32, | |
870a0bb5 | 6051 | .imbalance_pct = 125, |
143e1e28 VG |
6052 | |
6053 | .cache_nice_tries = 0, | |
6054 | .busy_idx = 0, | |
6055 | .idle_idx = 0, | |
cb83b629 PZ |
6056 | .newidle_idx = 0, |
6057 | .wake_idx = 0, | |
6058 | .forkexec_idx = 0, | |
6059 | ||
6060 | .flags = 1*SD_LOAD_BALANCE | |
6061 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6062 | | 1*SD_BALANCE_EXEC |
6063 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6064 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6065 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6066 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6067 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6068 | | 0*SD_SERIALIZE |
cb83b629 | 6069 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6070 | | 0*SD_NUMA |
6071 | | sd_flags | |
cb83b629 | 6072 | , |
143e1e28 | 6073 | |
cb83b629 PZ |
6074 | .last_balance = jiffies, |
6075 | .balance_interval = sd_weight, | |
143e1e28 | 6076 | .smt_gain = 0, |
2b4cfe64 JL |
6077 | .max_newidle_lb_cost = 0, |
6078 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6079 | #ifdef CONFIG_SCHED_DEBUG |
6080 | .name = tl->name, | |
6081 | #endif | |
cb83b629 | 6082 | }; |
cb83b629 PZ |
6083 | |
6084 | /* | |
143e1e28 | 6085 | * Convert topological properties into behaviour. |
cb83b629 | 6086 | */ |
143e1e28 | 6087 | |
5d4dfddd | 6088 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
143e1e28 VG |
6089 | sd->imbalance_pct = 110; |
6090 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6091 | |
6092 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6093 | sd->imbalance_pct = 117; | |
6094 | sd->cache_nice_tries = 1; | |
6095 | sd->busy_idx = 2; | |
6096 | ||
6097 | #ifdef CONFIG_NUMA | |
6098 | } else if (sd->flags & SD_NUMA) { | |
6099 | sd->cache_nice_tries = 2; | |
6100 | sd->busy_idx = 3; | |
6101 | sd->idle_idx = 2; | |
6102 | ||
6103 | sd->flags |= SD_SERIALIZE; | |
6104 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6105 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6106 | SD_BALANCE_FORK | | |
6107 | SD_WAKE_AFFINE); | |
6108 | } | |
6109 | ||
6110 | #endif | |
6111 | } else { | |
6112 | sd->flags |= SD_PREFER_SIBLING; | |
6113 | sd->cache_nice_tries = 1; | |
6114 | sd->busy_idx = 2; | |
6115 | sd->idle_idx = 1; | |
6116 | } | |
6117 | ||
6118 | sd->private = &tl->data; | |
cb83b629 PZ |
6119 | |
6120 | return sd; | |
6121 | } | |
6122 | ||
143e1e28 VG |
6123 | /* |
6124 | * Topology list, bottom-up. | |
6125 | */ | |
6126 | static struct sched_domain_topology_level default_topology[] = { | |
6127 | #ifdef CONFIG_SCHED_SMT | |
6128 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6129 | #endif | |
6130 | #ifdef CONFIG_SCHED_MC | |
6131 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6132 | #endif |
6133 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6134 | { NULL, }, | |
6135 | }; | |
6136 | ||
6137 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6138 | ||
6139 | #define for_each_sd_topology(tl) \ | |
6140 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6141 | ||
6142 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6143 | { | |
6144 | sched_domain_topology = tl; | |
6145 | } | |
6146 | ||
6147 | #ifdef CONFIG_NUMA | |
6148 | ||
cb83b629 PZ |
6149 | static const struct cpumask *sd_numa_mask(int cpu) |
6150 | { | |
6151 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6152 | } | |
6153 | ||
d039ac60 PZ |
6154 | static void sched_numa_warn(const char *str) |
6155 | { | |
6156 | static int done = false; | |
6157 | int i,j; | |
6158 | ||
6159 | if (done) | |
6160 | return; | |
6161 | ||
6162 | done = true; | |
6163 | ||
6164 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6165 | ||
6166 | for (i = 0; i < nr_node_ids; i++) { | |
6167 | printk(KERN_WARNING " "); | |
6168 | for (j = 0; j < nr_node_ids; j++) | |
6169 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6170 | printk(KERN_CONT "\n"); | |
6171 | } | |
6172 | printk(KERN_WARNING "\n"); | |
6173 | } | |
6174 | ||
6175 | static bool find_numa_distance(int distance) | |
6176 | { | |
6177 | int i; | |
6178 | ||
6179 | if (distance == node_distance(0, 0)) | |
6180 | return true; | |
6181 | ||
6182 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6183 | if (sched_domains_numa_distance[i] == distance) | |
6184 | return true; | |
6185 | } | |
6186 | ||
6187 | return false; | |
6188 | } | |
6189 | ||
cb83b629 PZ |
6190 | static void sched_init_numa(void) |
6191 | { | |
6192 | int next_distance, curr_distance = node_distance(0, 0); | |
6193 | struct sched_domain_topology_level *tl; | |
6194 | int level = 0; | |
6195 | int i, j, k; | |
6196 | ||
cb83b629 PZ |
6197 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6198 | if (!sched_domains_numa_distance) | |
6199 | return; | |
6200 | ||
6201 | /* | |
6202 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6203 | * unique distances in the node_distance() table. | |
6204 | * | |
6205 | * Assumes node_distance(0,j) includes all distances in | |
6206 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6207 | */ |
6208 | next_distance = curr_distance; | |
6209 | for (i = 0; i < nr_node_ids; i++) { | |
6210 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6211 | for (k = 0; k < nr_node_ids; k++) { |
6212 | int distance = node_distance(i, k); | |
6213 | ||
6214 | if (distance > curr_distance && | |
6215 | (distance < next_distance || | |
6216 | next_distance == curr_distance)) | |
6217 | next_distance = distance; | |
6218 | ||
6219 | /* | |
6220 | * While not a strong assumption it would be nice to know | |
6221 | * about cases where if node A is connected to B, B is not | |
6222 | * equally connected to A. | |
6223 | */ | |
6224 | if (sched_debug() && node_distance(k, i) != distance) | |
6225 | sched_numa_warn("Node-distance not symmetric"); | |
6226 | ||
6227 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6228 | sched_numa_warn("Node-0 not representative"); | |
6229 | } | |
6230 | if (next_distance != curr_distance) { | |
6231 | sched_domains_numa_distance[level++] = next_distance; | |
6232 | sched_domains_numa_levels = level; | |
6233 | curr_distance = next_distance; | |
6234 | } else break; | |
cb83b629 | 6235 | } |
d039ac60 PZ |
6236 | |
6237 | /* | |
6238 | * In case of sched_debug() we verify the above assumption. | |
6239 | */ | |
6240 | if (!sched_debug()) | |
6241 | break; | |
cb83b629 PZ |
6242 | } |
6243 | /* | |
6244 | * 'level' contains the number of unique distances, excluding the | |
6245 | * identity distance node_distance(i,i). | |
6246 | * | |
28b4a521 | 6247 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6248 | * numbers. |
6249 | */ | |
6250 | ||
5f7865f3 TC |
6251 | /* |
6252 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6253 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6254 | * the array will contain less then 'level' members. This could be | |
6255 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6256 | * in other functions. | |
6257 | * | |
6258 | * We reset it to 'level' at the end of this function. | |
6259 | */ | |
6260 | sched_domains_numa_levels = 0; | |
6261 | ||
cb83b629 PZ |
6262 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6263 | if (!sched_domains_numa_masks) | |
6264 | return; | |
6265 | ||
6266 | /* | |
6267 | * Now for each level, construct a mask per node which contains all | |
6268 | * cpus of nodes that are that many hops away from us. | |
6269 | */ | |
6270 | for (i = 0; i < level; i++) { | |
6271 | sched_domains_numa_masks[i] = | |
6272 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6273 | if (!sched_domains_numa_masks[i]) | |
6274 | return; | |
6275 | ||
6276 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6277 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6278 | if (!mask) |
6279 | return; | |
6280 | ||
6281 | sched_domains_numa_masks[i][j] = mask; | |
6282 | ||
6283 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6284 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6285 | continue; |
6286 | ||
6287 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6288 | } | |
6289 | } | |
6290 | } | |
6291 | ||
143e1e28 VG |
6292 | /* Compute default topology size */ |
6293 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6294 | ||
c515db8c | 6295 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6296 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6297 | if (!tl) | |
6298 | return; | |
6299 | ||
6300 | /* | |
6301 | * Copy the default topology bits.. | |
6302 | */ | |
143e1e28 VG |
6303 | for (i = 0; sched_domain_topology[i].mask; i++) |
6304 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6305 | |
6306 | /* | |
6307 | * .. and append 'j' levels of NUMA goodness. | |
6308 | */ | |
6309 | for (j = 0; j < level; i++, j++) { | |
6310 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6311 | .mask = sd_numa_mask, |
143e1e28 | 6312 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6313 | .flags = SDTL_OVERLAP, |
6314 | .numa_level = j, | |
143e1e28 | 6315 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6316 | }; |
6317 | } | |
6318 | ||
6319 | sched_domain_topology = tl; | |
5f7865f3 TC |
6320 | |
6321 | sched_domains_numa_levels = level; | |
cb83b629 | 6322 | } |
301a5cba TC |
6323 | |
6324 | static void sched_domains_numa_masks_set(int cpu) | |
6325 | { | |
6326 | int i, j; | |
6327 | int node = cpu_to_node(cpu); | |
6328 | ||
6329 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6330 | for (j = 0; j < nr_node_ids; j++) { | |
6331 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6332 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6333 | } | |
6334 | } | |
6335 | } | |
6336 | ||
6337 | static void sched_domains_numa_masks_clear(int cpu) | |
6338 | { | |
6339 | int i, j; | |
6340 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6341 | for (j = 0; j < nr_node_ids; j++) | |
6342 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6343 | } | |
6344 | } | |
6345 | ||
6346 | /* | |
6347 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6348 | * are onlined. | |
6349 | */ | |
6350 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6351 | unsigned long action, | |
6352 | void *hcpu) | |
6353 | { | |
6354 | int cpu = (long)hcpu; | |
6355 | ||
6356 | switch (action & ~CPU_TASKS_FROZEN) { | |
6357 | case CPU_ONLINE: | |
6358 | sched_domains_numa_masks_set(cpu); | |
6359 | break; | |
6360 | ||
6361 | case CPU_DEAD: | |
6362 | sched_domains_numa_masks_clear(cpu); | |
6363 | break; | |
6364 | ||
6365 | default: | |
6366 | return NOTIFY_DONE; | |
6367 | } | |
6368 | ||
6369 | return NOTIFY_OK; | |
cb83b629 PZ |
6370 | } |
6371 | #else | |
6372 | static inline void sched_init_numa(void) | |
6373 | { | |
6374 | } | |
301a5cba TC |
6375 | |
6376 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6377 | unsigned long action, | |
6378 | void *hcpu) | |
6379 | { | |
6380 | return 0; | |
6381 | } | |
cb83b629 PZ |
6382 | #endif /* CONFIG_NUMA */ |
6383 | ||
54ab4ff4 PZ |
6384 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6385 | { | |
6386 | struct sched_domain_topology_level *tl; | |
6387 | int j; | |
6388 | ||
27723a68 | 6389 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6390 | struct sd_data *sdd = &tl->data; |
6391 | ||
6392 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6393 | if (!sdd->sd) | |
6394 | return -ENOMEM; | |
6395 | ||
6396 | sdd->sg = alloc_percpu(struct sched_group *); | |
6397 | if (!sdd->sg) | |
6398 | return -ENOMEM; | |
6399 | ||
63b2ca30 NP |
6400 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6401 | if (!sdd->sgc) | |
9c3f75cb PZ |
6402 | return -ENOMEM; |
6403 | ||
54ab4ff4 PZ |
6404 | for_each_cpu(j, cpu_map) { |
6405 | struct sched_domain *sd; | |
6406 | struct sched_group *sg; | |
63b2ca30 | 6407 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6408 | |
6409 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6410 | GFP_KERNEL, cpu_to_node(j)); | |
6411 | if (!sd) | |
6412 | return -ENOMEM; | |
6413 | ||
6414 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6415 | ||
6416 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6417 | GFP_KERNEL, cpu_to_node(j)); | |
6418 | if (!sg) | |
6419 | return -ENOMEM; | |
6420 | ||
30b4e9eb IM |
6421 | sg->next = sg; |
6422 | ||
54ab4ff4 | 6423 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6424 | |
63b2ca30 | 6425 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6426 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6427 | if (!sgc) |
9c3f75cb PZ |
6428 | return -ENOMEM; |
6429 | ||
63b2ca30 | 6430 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6431 | } |
6432 | } | |
6433 | ||
6434 | return 0; | |
6435 | } | |
6436 | ||
6437 | static void __sdt_free(const struct cpumask *cpu_map) | |
6438 | { | |
6439 | struct sched_domain_topology_level *tl; | |
6440 | int j; | |
6441 | ||
27723a68 | 6442 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6443 | struct sd_data *sdd = &tl->data; |
6444 | ||
6445 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6446 | struct sched_domain *sd; |
6447 | ||
6448 | if (sdd->sd) { | |
6449 | sd = *per_cpu_ptr(sdd->sd, j); | |
6450 | if (sd && (sd->flags & SD_OVERLAP)) | |
6451 | free_sched_groups(sd->groups, 0); | |
6452 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6453 | } | |
6454 | ||
6455 | if (sdd->sg) | |
6456 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6457 | if (sdd->sgc) |
6458 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6459 | } |
6460 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6461 | sdd->sd = NULL; |
54ab4ff4 | 6462 | free_percpu(sdd->sg); |
fb2cf2c6 | 6463 | sdd->sg = NULL; |
63b2ca30 NP |
6464 | free_percpu(sdd->sgc); |
6465 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6466 | } |
6467 | } | |
6468 | ||
2c402dc3 | 6469 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6470 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6471 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6472 | { |
143e1e28 | 6473 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6474 | if (!sd) |
d069b916 | 6475 | return child; |
2c402dc3 | 6476 | |
2c402dc3 | 6477 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6478 | if (child) { |
6479 | sd->level = child->level + 1; | |
6480 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6481 | child->parent = sd; |
c75e0128 | 6482 | sd->child = child; |
60495e77 | 6483 | } |
a841f8ce | 6484 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6485 | |
6486 | return sd; | |
6487 | } | |
6488 | ||
2109b99e AH |
6489 | /* |
6490 | * Build sched domains for a given set of cpus and attach the sched domains | |
6491 | * to the individual cpus | |
6492 | */ | |
dce840a0 PZ |
6493 | static int build_sched_domains(const struct cpumask *cpu_map, |
6494 | struct sched_domain_attr *attr) | |
2109b99e | 6495 | { |
1c632169 | 6496 | enum s_alloc alloc_state; |
dce840a0 | 6497 | struct sched_domain *sd; |
2109b99e | 6498 | struct s_data d; |
822ff793 | 6499 | int i, ret = -ENOMEM; |
9c1cfda2 | 6500 | |
2109b99e AH |
6501 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6502 | if (alloc_state != sa_rootdomain) | |
6503 | goto error; | |
9c1cfda2 | 6504 | |
dce840a0 | 6505 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6506 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6507 | struct sched_domain_topology_level *tl; |
6508 | ||
3bd65a80 | 6509 | sd = NULL; |
27723a68 | 6510 | for_each_sd_topology(tl) { |
4a850cbe | 6511 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6512 | if (tl == sched_domain_topology) |
6513 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6514 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6515 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6516 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6517 | break; | |
e3589f6c | 6518 | } |
dce840a0 PZ |
6519 | } |
6520 | ||
6521 | /* Build the groups for the domains */ | |
6522 | for_each_cpu(i, cpu_map) { | |
6523 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6524 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6525 | if (sd->flags & SD_OVERLAP) { |
6526 | if (build_overlap_sched_groups(sd, i)) | |
6527 | goto error; | |
6528 | } else { | |
6529 | if (build_sched_groups(sd, i)) | |
6530 | goto error; | |
6531 | } | |
1cf51902 | 6532 | } |
a06dadbe | 6533 | } |
9c1cfda2 | 6534 | |
ced549fa | 6535 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6536 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6537 | if (!cpumask_test_cpu(i, cpu_map)) | |
6538 | continue; | |
9c1cfda2 | 6539 | |
dce840a0 PZ |
6540 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6541 | claim_allocations(i, sd); | |
63b2ca30 | 6542 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6543 | } |
f712c0c7 | 6544 | } |
9c1cfda2 | 6545 | |
1da177e4 | 6546 | /* Attach the domains */ |
dce840a0 | 6547 | rcu_read_lock(); |
abcd083a | 6548 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6549 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6550 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6551 | } |
dce840a0 | 6552 | rcu_read_unlock(); |
51888ca2 | 6553 | |
822ff793 | 6554 | ret = 0; |
51888ca2 | 6555 | error: |
2109b99e | 6556 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6557 | return ret; |
1da177e4 | 6558 | } |
029190c5 | 6559 | |
acc3f5d7 | 6560 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6561 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6562 | static struct sched_domain_attr *dattr_cur; |
6563 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6564 | |
6565 | /* | |
6566 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6567 | * cpumask) fails, then fallback to a single sched domain, |
6568 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6569 | */ |
4212823f | 6570 | static cpumask_var_t fallback_doms; |
029190c5 | 6571 | |
ee79d1bd HC |
6572 | /* |
6573 | * arch_update_cpu_topology lets virtualized architectures update the | |
6574 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6575 | * or 0 if it stayed the same. | |
6576 | */ | |
52f5684c | 6577 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6578 | { |
ee79d1bd | 6579 | return 0; |
22e52b07 HC |
6580 | } |
6581 | ||
acc3f5d7 RR |
6582 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6583 | { | |
6584 | int i; | |
6585 | cpumask_var_t *doms; | |
6586 | ||
6587 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6588 | if (!doms) | |
6589 | return NULL; | |
6590 | for (i = 0; i < ndoms; i++) { | |
6591 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6592 | free_sched_domains(doms, i); | |
6593 | return NULL; | |
6594 | } | |
6595 | } | |
6596 | return doms; | |
6597 | } | |
6598 | ||
6599 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6600 | { | |
6601 | unsigned int i; | |
6602 | for (i = 0; i < ndoms; i++) | |
6603 | free_cpumask_var(doms[i]); | |
6604 | kfree(doms); | |
6605 | } | |
6606 | ||
1a20ff27 | 6607 | /* |
41a2d6cf | 6608 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6609 | * For now this just excludes isolated cpus, but could be used to |
6610 | * exclude other special cases in the future. | |
1a20ff27 | 6611 | */ |
c4a8849a | 6612 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6613 | { |
7378547f MM |
6614 | int err; |
6615 | ||
22e52b07 | 6616 | arch_update_cpu_topology(); |
029190c5 | 6617 | ndoms_cur = 1; |
acc3f5d7 | 6618 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6619 | if (!doms_cur) |
acc3f5d7 RR |
6620 | doms_cur = &fallback_doms; |
6621 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6622 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6623 | register_sched_domain_sysctl(); |
7378547f MM |
6624 | |
6625 | return err; | |
1a20ff27 DG |
6626 | } |
6627 | ||
1a20ff27 DG |
6628 | /* |
6629 | * Detach sched domains from a group of cpus specified in cpu_map | |
6630 | * These cpus will now be attached to the NULL domain | |
6631 | */ | |
96f874e2 | 6632 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6633 | { |
6634 | int i; | |
6635 | ||
dce840a0 | 6636 | rcu_read_lock(); |
abcd083a | 6637 | for_each_cpu(i, cpu_map) |
57d885fe | 6638 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6639 | rcu_read_unlock(); |
1a20ff27 DG |
6640 | } |
6641 | ||
1d3504fc HS |
6642 | /* handle null as "default" */ |
6643 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6644 | struct sched_domain_attr *new, int idx_new) | |
6645 | { | |
6646 | struct sched_domain_attr tmp; | |
6647 | ||
6648 | /* fast path */ | |
6649 | if (!new && !cur) | |
6650 | return 1; | |
6651 | ||
6652 | tmp = SD_ATTR_INIT; | |
6653 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6654 | new ? (new + idx_new) : &tmp, | |
6655 | sizeof(struct sched_domain_attr)); | |
6656 | } | |
6657 | ||
029190c5 PJ |
6658 | /* |
6659 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6660 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6661 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6662 | * It destroys each deleted domain and builds each new domain. | |
6663 | * | |
acc3f5d7 | 6664 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6665 | * The masks don't intersect (don't overlap.) We should setup one |
6666 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6667 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6668 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6669 | * it as it is. | |
6670 | * | |
acc3f5d7 RR |
6671 | * The passed in 'doms_new' should be allocated using |
6672 | * alloc_sched_domains. This routine takes ownership of it and will | |
6673 | * free_sched_domains it when done with it. If the caller failed the | |
6674 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6675 | * and partition_sched_domains() will fallback to the single partition | |
6676 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6677 | * |
96f874e2 | 6678 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6679 | * ndoms_new == 0 is a special case for destroying existing domains, |
6680 | * and it will not create the default domain. | |
dfb512ec | 6681 | * |
029190c5 PJ |
6682 | * Call with hotplug lock held |
6683 | */ | |
acc3f5d7 | 6684 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6685 | struct sched_domain_attr *dattr_new) |
029190c5 | 6686 | { |
dfb512ec | 6687 | int i, j, n; |
d65bd5ec | 6688 | int new_topology; |
029190c5 | 6689 | |
712555ee | 6690 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6691 | |
7378547f MM |
6692 | /* always unregister in case we don't destroy any domains */ |
6693 | unregister_sched_domain_sysctl(); | |
6694 | ||
d65bd5ec HC |
6695 | /* Let architecture update cpu core mappings. */ |
6696 | new_topology = arch_update_cpu_topology(); | |
6697 | ||
dfb512ec | 6698 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6699 | |
6700 | /* Destroy deleted domains */ | |
6701 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6702 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6703 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6704 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6705 | goto match1; |
6706 | } | |
6707 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6708 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6709 | match1: |
6710 | ; | |
6711 | } | |
6712 | ||
c8d2d47a | 6713 | n = ndoms_cur; |
e761b772 | 6714 | if (doms_new == NULL) { |
c8d2d47a | 6715 | n = 0; |
acc3f5d7 | 6716 | doms_new = &fallback_doms; |
6ad4c188 | 6717 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6718 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6719 | } |
6720 | ||
029190c5 PJ |
6721 | /* Build new domains */ |
6722 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6723 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6724 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6725 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6726 | goto match2; |
6727 | } | |
6728 | /* no match - add a new doms_new */ | |
dce840a0 | 6729 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6730 | match2: |
6731 | ; | |
6732 | } | |
6733 | ||
6734 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6735 | if (doms_cur != &fallback_doms) |
6736 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6737 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6738 | doms_cur = doms_new; |
1d3504fc | 6739 | dattr_cur = dattr_new; |
029190c5 | 6740 | ndoms_cur = ndoms_new; |
7378547f MM |
6741 | |
6742 | register_sched_domain_sysctl(); | |
a1835615 | 6743 | |
712555ee | 6744 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6745 | } |
6746 | ||
d35be8ba SB |
6747 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6748 | ||
1da177e4 | 6749 | /* |
3a101d05 TH |
6750 | * Update cpusets according to cpu_active mask. If cpusets are |
6751 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6752 | * around partition_sched_domains(). | |
d35be8ba SB |
6753 | * |
6754 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6755 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6756 | */ |
0b2e918a TH |
6757 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6758 | void *hcpu) | |
e761b772 | 6759 | { |
d35be8ba SB |
6760 | switch (action) { |
6761 | case CPU_ONLINE_FROZEN: | |
6762 | case CPU_DOWN_FAILED_FROZEN: | |
6763 | ||
6764 | /* | |
6765 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6766 | * resume sequence. As long as this is not the last online | |
6767 | * operation in the resume sequence, just build a single sched | |
6768 | * domain, ignoring cpusets. | |
6769 | */ | |
6770 | num_cpus_frozen--; | |
6771 | if (likely(num_cpus_frozen)) { | |
6772 | partition_sched_domains(1, NULL, NULL); | |
6773 | break; | |
6774 | } | |
6775 | ||
6776 | /* | |
6777 | * This is the last CPU online operation. So fall through and | |
6778 | * restore the original sched domains by considering the | |
6779 | * cpuset configurations. | |
6780 | */ | |
6781 | ||
e761b772 | 6782 | case CPU_ONLINE: |
6ad4c188 | 6783 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6784 | cpuset_update_active_cpus(true); |
d35be8ba | 6785 | break; |
3a101d05 TH |
6786 | default: |
6787 | return NOTIFY_DONE; | |
6788 | } | |
d35be8ba | 6789 | return NOTIFY_OK; |
3a101d05 | 6790 | } |
e761b772 | 6791 | |
0b2e918a TH |
6792 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6793 | void *hcpu) | |
3a101d05 | 6794 | { |
d35be8ba | 6795 | switch (action) { |
3a101d05 | 6796 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6797 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6798 | break; |
6799 | case CPU_DOWN_PREPARE_FROZEN: | |
6800 | num_cpus_frozen++; | |
6801 | partition_sched_domains(1, NULL, NULL); | |
6802 | break; | |
e761b772 MK |
6803 | default: |
6804 | return NOTIFY_DONE; | |
6805 | } | |
d35be8ba | 6806 | return NOTIFY_OK; |
e761b772 | 6807 | } |
e761b772 | 6808 | |
1da177e4 LT |
6809 | void __init sched_init_smp(void) |
6810 | { | |
dcc30a35 RR |
6811 | cpumask_var_t non_isolated_cpus; |
6812 | ||
6813 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6814 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6815 | |
cb83b629 PZ |
6816 | sched_init_numa(); |
6817 | ||
6acce3ef PZ |
6818 | /* |
6819 | * There's no userspace yet to cause hotplug operations; hence all the | |
6820 | * cpu masks are stable and all blatant races in the below code cannot | |
6821 | * happen. | |
6822 | */ | |
712555ee | 6823 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6824 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6825 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6826 | if (cpumask_empty(non_isolated_cpus)) | |
6827 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6828 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6829 | |
301a5cba | 6830 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6831 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6832 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6833 | |
b328ca18 | 6834 | init_hrtick(); |
5c1e1767 NP |
6835 | |
6836 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6837 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6838 | BUG(); |
19978ca6 | 6839 | sched_init_granularity(); |
dcc30a35 | 6840 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6841 | |
0e3900e6 | 6842 | init_sched_rt_class(); |
1baca4ce | 6843 | init_sched_dl_class(); |
1da177e4 LT |
6844 | } |
6845 | #else | |
6846 | void __init sched_init_smp(void) | |
6847 | { | |
19978ca6 | 6848 | sched_init_granularity(); |
1da177e4 LT |
6849 | } |
6850 | #endif /* CONFIG_SMP */ | |
6851 | ||
cd1bb94b AB |
6852 | const_debug unsigned int sysctl_timer_migration = 1; |
6853 | ||
1da177e4 LT |
6854 | int in_sched_functions(unsigned long addr) |
6855 | { | |
1da177e4 LT |
6856 | return in_lock_functions(addr) || |
6857 | (addr >= (unsigned long)__sched_text_start | |
6858 | && addr < (unsigned long)__sched_text_end); | |
6859 | } | |
6860 | ||
029632fb | 6861 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6862 | /* |
6863 | * Default task group. | |
6864 | * Every task in system belongs to this group at bootup. | |
6865 | */ | |
029632fb | 6866 | struct task_group root_task_group; |
35cf4e50 | 6867 | LIST_HEAD(task_groups); |
052f1dc7 | 6868 | #endif |
6f505b16 | 6869 | |
e6252c3e | 6870 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6871 | |
1da177e4 LT |
6872 | void __init sched_init(void) |
6873 | { | |
dd41f596 | 6874 | int i, j; |
434d53b0 MT |
6875 | unsigned long alloc_size = 0, ptr; |
6876 | ||
6877 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6878 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6879 | #endif | |
6880 | #ifdef CONFIG_RT_GROUP_SCHED | |
6881 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6882 | #endif |
df7c8e84 | 6883 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6884 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6885 | #endif |
434d53b0 | 6886 | if (alloc_size) { |
36b7b6d4 | 6887 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6888 | |
6889 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6890 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6891 | ptr += nr_cpu_ids * sizeof(void **); |
6892 | ||
07e06b01 | 6893 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6894 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6895 | |
6d6bc0ad | 6896 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6897 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6898 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6899 | ptr += nr_cpu_ids * sizeof(void **); |
6900 | ||
07e06b01 | 6901 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6902 | ptr += nr_cpu_ids * sizeof(void **); |
6903 | ||
6d6bc0ad | 6904 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6905 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6906 | for_each_possible_cpu(i) { | |
e6252c3e | 6907 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6908 | ptr += cpumask_size(); |
6909 | } | |
6910 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6911 | } |
dd41f596 | 6912 | |
332ac17e DF |
6913 | init_rt_bandwidth(&def_rt_bandwidth, |
6914 | global_rt_period(), global_rt_runtime()); | |
6915 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6916 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6917 | |
57d885fe GH |
6918 | #ifdef CONFIG_SMP |
6919 | init_defrootdomain(); | |
6920 | #endif | |
6921 | ||
d0b27fa7 | 6922 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6923 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6924 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6925 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6926 | |
7c941438 | 6927 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6928 | list_add(&root_task_group.list, &task_groups); |
6929 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6930 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6931 | autogroup_init(&init_task); |
54c707e9 | 6932 | |
7c941438 | 6933 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6934 | |
0a945022 | 6935 | for_each_possible_cpu(i) { |
70b97a7f | 6936 | struct rq *rq; |
1da177e4 LT |
6937 | |
6938 | rq = cpu_rq(i); | |
05fa785c | 6939 | raw_spin_lock_init(&rq->lock); |
7897986b | 6940 | rq->nr_running = 0; |
dce48a84 TG |
6941 | rq->calc_load_active = 0; |
6942 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6943 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6944 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6945 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6946 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6947 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6948 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6949 | /* |
07e06b01 | 6950 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6951 | * |
6952 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6953 | * gets 100% of the cpu resources in the system. This overall | |
6954 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6955 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6956 | * based on each entity's (task or task-group's) weight |
6957 | * (se->load.weight). | |
6958 | * | |
07e06b01 | 6959 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6960 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6961 | * then A0's share of the cpu resource is: | |
6962 | * | |
0d905bca | 6963 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6964 | * |
07e06b01 YZ |
6965 | * We achieve this by letting root_task_group's tasks sit |
6966 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6967 | */ |
ab84d31e | 6968 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6969 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6970 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6971 | ||
6972 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6973 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6974 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6975 | #endif |
1da177e4 | 6976 | |
dd41f596 IM |
6977 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6978 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6979 | |
6980 | rq->last_load_update_tick = jiffies; | |
6981 | ||
1da177e4 | 6982 | #ifdef CONFIG_SMP |
41c7ce9a | 6983 | rq->sd = NULL; |
57d885fe | 6984 | rq->rd = NULL; |
ca8ce3d0 | 6985 | rq->cpu_capacity = SCHED_CAPACITY_SCALE; |
3f029d3c | 6986 | rq->post_schedule = 0; |
1da177e4 | 6987 | rq->active_balance = 0; |
dd41f596 | 6988 | rq->next_balance = jiffies; |
1da177e4 | 6989 | rq->push_cpu = 0; |
0a2966b4 | 6990 | rq->cpu = i; |
1f11eb6a | 6991 | rq->online = 0; |
eae0c9df MG |
6992 | rq->idle_stamp = 0; |
6993 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6994 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6995 | |
6996 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6997 | ||
dc938520 | 6998 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6999 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7000 | rq->nohz_flags = 0; |
83cd4fe2 | 7001 | #endif |
265f22a9 FW |
7002 | #ifdef CONFIG_NO_HZ_FULL |
7003 | rq->last_sched_tick = 0; | |
7004 | #endif | |
1da177e4 | 7005 | #endif |
8f4d37ec | 7006 | init_rq_hrtick(rq); |
1da177e4 | 7007 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7008 | } |
7009 | ||
2dd73a4f | 7010 | set_load_weight(&init_task); |
b50f60ce | 7011 | |
e107be36 AK |
7012 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7013 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7014 | #endif | |
7015 | ||
1da177e4 LT |
7016 | /* |
7017 | * The boot idle thread does lazy MMU switching as well: | |
7018 | */ | |
7019 | atomic_inc(&init_mm.mm_count); | |
7020 | enter_lazy_tlb(&init_mm, current); | |
7021 | ||
7022 | /* | |
7023 | * Make us the idle thread. Technically, schedule() should not be | |
7024 | * called from this thread, however somewhere below it might be, | |
7025 | * but because we are the idle thread, we just pick up running again | |
7026 | * when this runqueue becomes "idle". | |
7027 | */ | |
7028 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7029 | |
7030 | calc_load_update = jiffies + LOAD_FREQ; | |
7031 | ||
dd41f596 IM |
7032 | /* |
7033 | * During early bootup we pretend to be a normal task: | |
7034 | */ | |
7035 | current->sched_class = &fair_sched_class; | |
6892b75e | 7036 | |
bf4d83f6 | 7037 | #ifdef CONFIG_SMP |
4cb98839 | 7038 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7039 | /* May be allocated at isolcpus cmdline parse time */ |
7040 | if (cpu_isolated_map == NULL) | |
7041 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7042 | idle_thread_set_boot_cpu(); |
a803f026 | 7043 | set_cpu_rq_start_time(); |
029632fb PZ |
7044 | #endif |
7045 | init_sched_fair_class(); | |
6a7b3dc3 | 7046 | |
6892b75e | 7047 | scheduler_running = 1; |
1da177e4 LT |
7048 | } |
7049 | ||
d902db1e | 7050 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7051 | static inline int preempt_count_equals(int preempt_offset) |
7052 | { | |
234da7bc | 7053 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7054 | |
4ba8216c | 7055 | return (nested == preempt_offset); |
e4aafea2 FW |
7056 | } |
7057 | ||
d894837f | 7058 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7059 | { |
1da177e4 LT |
7060 | static unsigned long prev_jiffy; /* ratelimiting */ |
7061 | ||
b3fbab05 | 7062 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7063 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7064 | !is_idle_task(current)) || | |
e4aafea2 | 7065 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7066 | return; |
7067 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7068 | return; | |
7069 | prev_jiffy = jiffies; | |
7070 | ||
3df0fc5b PZ |
7071 | printk(KERN_ERR |
7072 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7073 | file, line); | |
7074 | printk(KERN_ERR | |
7075 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7076 | in_atomic(), irqs_disabled(), | |
7077 | current->pid, current->comm); | |
aef745fc IM |
7078 | |
7079 | debug_show_held_locks(current); | |
7080 | if (irqs_disabled()) | |
7081 | print_irqtrace_events(current); | |
8f47b187 TG |
7082 | #ifdef CONFIG_DEBUG_PREEMPT |
7083 | if (!preempt_count_equals(preempt_offset)) { | |
7084 | pr_err("Preemption disabled at:"); | |
7085 | print_ip_sym(current->preempt_disable_ip); | |
7086 | pr_cont("\n"); | |
7087 | } | |
7088 | #endif | |
aef745fc | 7089 | dump_stack(); |
1da177e4 LT |
7090 | } |
7091 | EXPORT_SYMBOL(__might_sleep); | |
7092 | #endif | |
7093 | ||
7094 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7095 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7096 | { | |
da7a735e | 7097 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7098 | struct sched_attr attr = { |
7099 | .sched_policy = SCHED_NORMAL, | |
7100 | }; | |
da7a735e | 7101 | int old_prio = p->prio; |
3a5e4dc1 | 7102 | int on_rq; |
3e51f33f | 7103 | |
fd2f4419 | 7104 | on_rq = p->on_rq; |
3a5e4dc1 | 7105 | if (on_rq) |
4ca9b72b | 7106 | dequeue_task(rq, p, 0); |
d50dde5a | 7107 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7108 | if (on_rq) { |
4ca9b72b | 7109 | enqueue_task(rq, p, 0); |
8875125e | 7110 | resched_curr(rq); |
3a5e4dc1 | 7111 | } |
da7a735e PZ |
7112 | |
7113 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7114 | } |
7115 | ||
1da177e4 LT |
7116 | void normalize_rt_tasks(void) |
7117 | { | |
a0f98a1c | 7118 | struct task_struct *g, *p; |
1da177e4 | 7119 | unsigned long flags; |
70b97a7f | 7120 | struct rq *rq; |
1da177e4 | 7121 | |
4cf5d77a | 7122 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7123 | do_each_thread(g, p) { |
178be793 IM |
7124 | /* |
7125 | * Only normalize user tasks: | |
7126 | */ | |
7127 | if (!p->mm) | |
7128 | continue; | |
7129 | ||
6cfb0d5d | 7130 | p->se.exec_start = 0; |
6cfb0d5d | 7131 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7132 | p->se.statistics.wait_start = 0; |
7133 | p->se.statistics.sleep_start = 0; | |
7134 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7135 | #endif |
dd41f596 | 7136 | |
aab03e05 | 7137 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7138 | /* |
7139 | * Renice negative nice level userspace | |
7140 | * tasks back to 0: | |
7141 | */ | |
d0ea0268 | 7142 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7143 | set_user_nice(p, 0); |
1da177e4 | 7144 | continue; |
dd41f596 | 7145 | } |
1da177e4 | 7146 | |
1d615482 | 7147 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7148 | rq = __task_rq_lock(p); |
1da177e4 | 7149 | |
178be793 | 7150 | normalize_task(rq, p); |
3a5e4dc1 | 7151 | |
b29739f9 | 7152 | __task_rq_unlock(rq); |
1d615482 | 7153 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7154 | } while_each_thread(g, p); |
7155 | ||
4cf5d77a | 7156 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7157 | } |
7158 | ||
7159 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7160 | |
67fc4e0c | 7161 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7162 | /* |
67fc4e0c | 7163 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7164 | * |
7165 | * They can only be called when the whole system has been | |
7166 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7167 | * activity can take place. Using them for anything else would | |
7168 | * be a serious bug, and as a result, they aren't even visible | |
7169 | * under any other configuration. | |
7170 | */ | |
7171 | ||
7172 | /** | |
7173 | * curr_task - return the current task for a given cpu. | |
7174 | * @cpu: the processor in question. | |
7175 | * | |
7176 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7177 | * |
7178 | * Return: The current task for @cpu. | |
1df5c10a | 7179 | */ |
36c8b586 | 7180 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7181 | { |
7182 | return cpu_curr(cpu); | |
7183 | } | |
7184 | ||
67fc4e0c JW |
7185 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7186 | ||
7187 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7188 | /** |
7189 | * set_curr_task - set the current task for a given cpu. | |
7190 | * @cpu: the processor in question. | |
7191 | * @p: the task pointer to set. | |
7192 | * | |
7193 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7194 | * are serviced on a separate stack. It allows the architecture to switch the |
7195 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7196 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7197 | * and caller must save the original value of the current task (see | |
7198 | * curr_task() above) and restore that value before reenabling interrupts and | |
7199 | * re-starting the system. | |
7200 | * | |
7201 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7202 | */ | |
36c8b586 | 7203 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7204 | { |
7205 | cpu_curr(cpu) = p; | |
7206 | } | |
7207 | ||
7208 | #endif | |
29f59db3 | 7209 | |
7c941438 | 7210 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7211 | /* task_group_lock serializes the addition/removal of task groups */ |
7212 | static DEFINE_SPINLOCK(task_group_lock); | |
7213 | ||
bccbe08a PZ |
7214 | static void free_sched_group(struct task_group *tg) |
7215 | { | |
7216 | free_fair_sched_group(tg); | |
7217 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7218 | autogroup_free(tg); |
bccbe08a PZ |
7219 | kfree(tg); |
7220 | } | |
7221 | ||
7222 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7223 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7224 | { |
7225 | struct task_group *tg; | |
bccbe08a PZ |
7226 | |
7227 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7228 | if (!tg) | |
7229 | return ERR_PTR(-ENOMEM); | |
7230 | ||
ec7dc8ac | 7231 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7232 | goto err; |
7233 | ||
ec7dc8ac | 7234 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7235 | goto err; |
7236 | ||
ace783b9 LZ |
7237 | return tg; |
7238 | ||
7239 | err: | |
7240 | free_sched_group(tg); | |
7241 | return ERR_PTR(-ENOMEM); | |
7242 | } | |
7243 | ||
7244 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7245 | { | |
7246 | unsigned long flags; | |
7247 | ||
8ed36996 | 7248 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7249 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7250 | |
7251 | WARN_ON(!parent); /* root should already exist */ | |
7252 | ||
7253 | tg->parent = parent; | |
f473aa5e | 7254 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7255 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7256 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7257 | } |
7258 | ||
9b5b7751 | 7259 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7260 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7261 | { |
29f59db3 | 7262 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7263 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7264 | } |
7265 | ||
9b5b7751 | 7266 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7267 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7268 | { |
7269 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7270 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7271 | } | |
7272 | ||
7273 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7274 | { |
8ed36996 | 7275 | unsigned long flags; |
9b5b7751 | 7276 | int i; |
29f59db3 | 7277 | |
3d4b47b4 PZ |
7278 | /* end participation in shares distribution */ |
7279 | for_each_possible_cpu(i) | |
bccbe08a | 7280 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7281 | |
7282 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7283 | list_del_rcu(&tg->list); |
f473aa5e | 7284 | list_del_rcu(&tg->siblings); |
8ed36996 | 7285 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7286 | } |
7287 | ||
9b5b7751 | 7288 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7289 | * The caller of this function should have put the task in its new group |
7290 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7291 | * reflect its new group. | |
9b5b7751 SV |
7292 | */ |
7293 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7294 | { |
8323f26c | 7295 | struct task_group *tg; |
29f59db3 SV |
7296 | int on_rq, running; |
7297 | unsigned long flags; | |
7298 | struct rq *rq; | |
7299 | ||
7300 | rq = task_rq_lock(tsk, &flags); | |
7301 | ||
051a1d1a | 7302 | running = task_current(rq, tsk); |
fd2f4419 | 7303 | on_rq = tsk->on_rq; |
29f59db3 | 7304 | |
0e1f3483 | 7305 | if (on_rq) |
29f59db3 | 7306 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7307 | if (unlikely(running)) |
7308 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7309 | |
073219e9 | 7310 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7311 | lockdep_is_held(&tsk->sighand->siglock)), |
7312 | struct task_group, css); | |
7313 | tg = autogroup_task_group(tsk, tg); | |
7314 | tsk->sched_task_group = tg; | |
7315 | ||
810b3817 | 7316 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7317 | if (tsk->sched_class->task_move_group) |
7318 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7319 | else | |
810b3817 | 7320 | #endif |
b2b5ce02 | 7321 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7322 | |
0e1f3483 HS |
7323 | if (unlikely(running)) |
7324 | tsk->sched_class->set_curr_task(rq); | |
7325 | if (on_rq) | |
371fd7e7 | 7326 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7327 | |
0122ec5b | 7328 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7329 | } |
7c941438 | 7330 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7331 | |
a790de99 PT |
7332 | #ifdef CONFIG_RT_GROUP_SCHED |
7333 | /* | |
7334 | * Ensure that the real time constraints are schedulable. | |
7335 | */ | |
7336 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7337 | |
9a7e0b18 PZ |
7338 | /* Must be called with tasklist_lock held */ |
7339 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7340 | { |
9a7e0b18 | 7341 | struct task_struct *g, *p; |
b40b2e8e | 7342 | |
9a7e0b18 | 7343 | do_each_thread(g, p) { |
029632fb | 7344 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7345 | return 1; |
7346 | } while_each_thread(g, p); | |
b40b2e8e | 7347 | |
9a7e0b18 PZ |
7348 | return 0; |
7349 | } | |
b40b2e8e | 7350 | |
9a7e0b18 PZ |
7351 | struct rt_schedulable_data { |
7352 | struct task_group *tg; | |
7353 | u64 rt_period; | |
7354 | u64 rt_runtime; | |
7355 | }; | |
b40b2e8e | 7356 | |
a790de99 | 7357 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7358 | { |
7359 | struct rt_schedulable_data *d = data; | |
7360 | struct task_group *child; | |
7361 | unsigned long total, sum = 0; | |
7362 | u64 period, runtime; | |
b40b2e8e | 7363 | |
9a7e0b18 PZ |
7364 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7365 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7366 | |
9a7e0b18 PZ |
7367 | if (tg == d->tg) { |
7368 | period = d->rt_period; | |
7369 | runtime = d->rt_runtime; | |
b40b2e8e | 7370 | } |
b40b2e8e | 7371 | |
4653f803 PZ |
7372 | /* |
7373 | * Cannot have more runtime than the period. | |
7374 | */ | |
7375 | if (runtime > period && runtime != RUNTIME_INF) | |
7376 | return -EINVAL; | |
6f505b16 | 7377 | |
4653f803 PZ |
7378 | /* |
7379 | * Ensure we don't starve existing RT tasks. | |
7380 | */ | |
9a7e0b18 PZ |
7381 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7382 | return -EBUSY; | |
6f505b16 | 7383 | |
9a7e0b18 | 7384 | total = to_ratio(period, runtime); |
6f505b16 | 7385 | |
4653f803 PZ |
7386 | /* |
7387 | * Nobody can have more than the global setting allows. | |
7388 | */ | |
7389 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7390 | return -EINVAL; | |
6f505b16 | 7391 | |
4653f803 PZ |
7392 | /* |
7393 | * The sum of our children's runtime should not exceed our own. | |
7394 | */ | |
9a7e0b18 PZ |
7395 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7396 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7397 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7398 | |
9a7e0b18 PZ |
7399 | if (child == d->tg) { |
7400 | period = d->rt_period; | |
7401 | runtime = d->rt_runtime; | |
7402 | } | |
6f505b16 | 7403 | |
9a7e0b18 | 7404 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7405 | } |
6f505b16 | 7406 | |
9a7e0b18 PZ |
7407 | if (sum > total) |
7408 | return -EINVAL; | |
7409 | ||
7410 | return 0; | |
6f505b16 PZ |
7411 | } |
7412 | ||
9a7e0b18 | 7413 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7414 | { |
8277434e PT |
7415 | int ret; |
7416 | ||
9a7e0b18 PZ |
7417 | struct rt_schedulable_data data = { |
7418 | .tg = tg, | |
7419 | .rt_period = period, | |
7420 | .rt_runtime = runtime, | |
7421 | }; | |
7422 | ||
8277434e PT |
7423 | rcu_read_lock(); |
7424 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7425 | rcu_read_unlock(); | |
7426 | ||
7427 | return ret; | |
521f1a24 DG |
7428 | } |
7429 | ||
ab84d31e | 7430 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7431 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7432 | { |
ac086bc2 | 7433 | int i, err = 0; |
9f0c1e56 | 7434 | |
9f0c1e56 | 7435 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7436 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7437 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7438 | if (err) | |
9f0c1e56 | 7439 | goto unlock; |
ac086bc2 | 7440 | |
0986b11b | 7441 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7442 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7443 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7444 | |
7445 | for_each_possible_cpu(i) { | |
7446 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7447 | ||
0986b11b | 7448 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7449 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7450 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7451 | } |
0986b11b | 7452 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7453 | unlock: |
521f1a24 | 7454 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7455 | mutex_unlock(&rt_constraints_mutex); |
7456 | ||
7457 | return err; | |
6f505b16 PZ |
7458 | } |
7459 | ||
25cc7da7 | 7460 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7461 | { |
7462 | u64 rt_runtime, rt_period; | |
7463 | ||
7464 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7465 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7466 | if (rt_runtime_us < 0) | |
7467 | rt_runtime = RUNTIME_INF; | |
7468 | ||
ab84d31e | 7469 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7470 | } |
7471 | ||
25cc7da7 | 7472 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7473 | { |
7474 | u64 rt_runtime_us; | |
7475 | ||
d0b27fa7 | 7476 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7477 | return -1; |
7478 | ||
d0b27fa7 | 7479 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7480 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7481 | return rt_runtime_us; | |
7482 | } | |
d0b27fa7 | 7483 | |
25cc7da7 | 7484 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7485 | { |
7486 | u64 rt_runtime, rt_period; | |
7487 | ||
7488 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7489 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7490 | ||
619b0488 R |
7491 | if (rt_period == 0) |
7492 | return -EINVAL; | |
7493 | ||
ab84d31e | 7494 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7495 | } |
7496 | ||
25cc7da7 | 7497 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7498 | { |
7499 | u64 rt_period_us; | |
7500 | ||
7501 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7502 | do_div(rt_period_us, NSEC_PER_USEC); | |
7503 | return rt_period_us; | |
7504 | } | |
332ac17e | 7505 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7506 | |
332ac17e | 7507 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7508 | static int sched_rt_global_constraints(void) |
7509 | { | |
7510 | int ret = 0; | |
7511 | ||
7512 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7513 | read_lock(&tasklist_lock); |
4653f803 | 7514 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7515 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7516 | mutex_unlock(&rt_constraints_mutex); |
7517 | ||
7518 | return ret; | |
7519 | } | |
54e99124 | 7520 | |
25cc7da7 | 7521 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7522 | { |
7523 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7524 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7525 | return 0; | |
7526 | ||
7527 | return 1; | |
7528 | } | |
7529 | ||
6d6bc0ad | 7530 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7531 | static int sched_rt_global_constraints(void) |
7532 | { | |
ac086bc2 | 7533 | unsigned long flags; |
332ac17e | 7534 | int i, ret = 0; |
ec5d4989 | 7535 | |
0986b11b | 7536 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7537 | for_each_possible_cpu(i) { |
7538 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7539 | ||
0986b11b | 7540 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7541 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7542 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7543 | } |
0986b11b | 7544 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7545 | |
332ac17e | 7546 | return ret; |
d0b27fa7 | 7547 | } |
6d6bc0ad | 7548 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7549 | |
332ac17e DF |
7550 | static int sched_dl_global_constraints(void) |
7551 | { | |
1724813d PZ |
7552 | u64 runtime = global_rt_runtime(); |
7553 | u64 period = global_rt_period(); | |
332ac17e | 7554 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7555 | int cpu, ret = 0; |
49516342 | 7556 | unsigned long flags; |
332ac17e DF |
7557 | |
7558 | /* | |
7559 | * Here we want to check the bandwidth not being set to some | |
7560 | * value smaller than the currently allocated bandwidth in | |
7561 | * any of the root_domains. | |
7562 | * | |
7563 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7564 | * cycling on root_domains... Discussion on different/better | |
7565 | * solutions is welcome! | |
7566 | */ | |
1724813d PZ |
7567 | for_each_possible_cpu(cpu) { |
7568 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7569 | |
49516342 | 7570 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7571 | if (new_bw < dl_b->total_bw) |
7572 | ret = -EBUSY; | |
49516342 | 7573 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7574 | |
7575 | if (ret) | |
7576 | break; | |
332ac17e DF |
7577 | } |
7578 | ||
1724813d | 7579 | return ret; |
332ac17e DF |
7580 | } |
7581 | ||
1724813d | 7582 | static void sched_dl_do_global(void) |
ce0dbbbb | 7583 | { |
1724813d PZ |
7584 | u64 new_bw = -1; |
7585 | int cpu; | |
49516342 | 7586 | unsigned long flags; |
ce0dbbbb | 7587 | |
1724813d PZ |
7588 | def_dl_bandwidth.dl_period = global_rt_period(); |
7589 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7590 | ||
7591 | if (global_rt_runtime() != RUNTIME_INF) | |
7592 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7593 | ||
7594 | /* | |
7595 | * FIXME: As above... | |
7596 | */ | |
7597 | for_each_possible_cpu(cpu) { | |
7598 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7599 | ||
49516342 | 7600 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7601 | dl_b->bw = new_bw; |
49516342 | 7602 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7603 | } |
1724813d PZ |
7604 | } |
7605 | ||
7606 | static int sched_rt_global_validate(void) | |
7607 | { | |
7608 | if (sysctl_sched_rt_period <= 0) | |
7609 | return -EINVAL; | |
7610 | ||
e9e7cb38 JL |
7611 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7612 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7613 | return -EINVAL; |
7614 | ||
7615 | return 0; | |
7616 | } | |
7617 | ||
7618 | static void sched_rt_do_global(void) | |
7619 | { | |
7620 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7621 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7622 | } |
7623 | ||
d0b27fa7 | 7624 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7625 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7626 | loff_t *ppos) |
7627 | { | |
d0b27fa7 PZ |
7628 | int old_period, old_runtime; |
7629 | static DEFINE_MUTEX(mutex); | |
1724813d | 7630 | int ret; |
d0b27fa7 PZ |
7631 | |
7632 | mutex_lock(&mutex); | |
7633 | old_period = sysctl_sched_rt_period; | |
7634 | old_runtime = sysctl_sched_rt_runtime; | |
7635 | ||
8d65af78 | 7636 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7637 | |
7638 | if (!ret && write) { | |
1724813d PZ |
7639 | ret = sched_rt_global_validate(); |
7640 | if (ret) | |
7641 | goto undo; | |
7642 | ||
d0b27fa7 | 7643 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7644 | if (ret) |
7645 | goto undo; | |
7646 | ||
7647 | ret = sched_dl_global_constraints(); | |
7648 | if (ret) | |
7649 | goto undo; | |
7650 | ||
7651 | sched_rt_do_global(); | |
7652 | sched_dl_do_global(); | |
7653 | } | |
7654 | if (0) { | |
7655 | undo: | |
7656 | sysctl_sched_rt_period = old_period; | |
7657 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7658 | } |
7659 | mutex_unlock(&mutex); | |
7660 | ||
7661 | return ret; | |
7662 | } | |
68318b8e | 7663 | |
1724813d | 7664 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7665 | void __user *buffer, size_t *lenp, |
7666 | loff_t *ppos) | |
7667 | { | |
7668 | int ret; | |
332ac17e | 7669 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7670 | |
7671 | mutex_lock(&mutex); | |
332ac17e | 7672 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7673 | /* make sure that internally we keep jiffies */ |
7674 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7675 | if (!ret && write) { |
1724813d PZ |
7676 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7677 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7678 | } |
7679 | mutex_unlock(&mutex); | |
332ac17e DF |
7680 | return ret; |
7681 | } | |
7682 | ||
052f1dc7 | 7683 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7684 | |
a7c6d554 | 7685 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7686 | { |
a7c6d554 | 7687 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7688 | } |
7689 | ||
eb95419b TH |
7690 | static struct cgroup_subsys_state * |
7691 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7692 | { |
eb95419b TH |
7693 | struct task_group *parent = css_tg(parent_css); |
7694 | struct task_group *tg; | |
68318b8e | 7695 | |
eb95419b | 7696 | if (!parent) { |
68318b8e | 7697 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7698 | return &root_task_group.css; |
68318b8e SV |
7699 | } |
7700 | ||
ec7dc8ac | 7701 | tg = sched_create_group(parent); |
68318b8e SV |
7702 | if (IS_ERR(tg)) |
7703 | return ERR_PTR(-ENOMEM); | |
7704 | ||
68318b8e SV |
7705 | return &tg->css; |
7706 | } | |
7707 | ||
eb95419b | 7708 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7709 | { |
eb95419b | 7710 | struct task_group *tg = css_tg(css); |
5c9d535b | 7711 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 7712 | |
63876986 TH |
7713 | if (parent) |
7714 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7715 | return 0; |
7716 | } | |
7717 | ||
eb95419b | 7718 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7719 | { |
eb95419b | 7720 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7721 | |
7722 | sched_destroy_group(tg); | |
7723 | } | |
7724 | ||
eb95419b | 7725 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7726 | { |
eb95419b | 7727 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7728 | |
7729 | sched_offline_group(tg); | |
7730 | } | |
7731 | ||
eb95419b | 7732 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7733 | struct cgroup_taskset *tset) |
68318b8e | 7734 | { |
bb9d97b6 TH |
7735 | struct task_struct *task; |
7736 | ||
924f0d9a | 7737 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7738 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7739 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7740 | return -EINVAL; |
b68aa230 | 7741 | #else |
bb9d97b6 TH |
7742 | /* We don't support RT-tasks being in separate groups */ |
7743 | if (task->sched_class != &fair_sched_class) | |
7744 | return -EINVAL; | |
b68aa230 | 7745 | #endif |
bb9d97b6 | 7746 | } |
be367d09 BB |
7747 | return 0; |
7748 | } | |
68318b8e | 7749 | |
eb95419b | 7750 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7751 | struct cgroup_taskset *tset) |
68318b8e | 7752 | { |
bb9d97b6 TH |
7753 | struct task_struct *task; |
7754 | ||
924f0d9a | 7755 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7756 | sched_move_task(task); |
68318b8e SV |
7757 | } |
7758 | ||
eb95419b TH |
7759 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7760 | struct cgroup_subsys_state *old_css, | |
7761 | struct task_struct *task) | |
068c5cc5 PZ |
7762 | { |
7763 | /* | |
7764 | * cgroup_exit() is called in the copy_process() failure path. | |
7765 | * Ignore this case since the task hasn't ran yet, this avoids | |
7766 | * trying to poke a half freed task state from generic code. | |
7767 | */ | |
7768 | if (!(task->flags & PF_EXITING)) | |
7769 | return; | |
7770 | ||
7771 | sched_move_task(task); | |
7772 | } | |
7773 | ||
052f1dc7 | 7774 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7775 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7776 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7777 | { |
182446d0 | 7778 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7779 | } |
7780 | ||
182446d0 TH |
7781 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7782 | struct cftype *cft) | |
68318b8e | 7783 | { |
182446d0 | 7784 | struct task_group *tg = css_tg(css); |
68318b8e | 7785 | |
c8b28116 | 7786 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7787 | } |
ab84d31e PT |
7788 | |
7789 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7790 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7791 | ||
ab84d31e PT |
7792 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7793 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7794 | ||
a790de99 PT |
7795 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7796 | ||
ab84d31e PT |
7797 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7798 | { | |
56f570e5 | 7799 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7800 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7801 | |
7802 | if (tg == &root_task_group) | |
7803 | return -EINVAL; | |
7804 | ||
7805 | /* | |
7806 | * Ensure we have at some amount of bandwidth every period. This is | |
7807 | * to prevent reaching a state of large arrears when throttled via | |
7808 | * entity_tick() resulting in prolonged exit starvation. | |
7809 | */ | |
7810 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7811 | return -EINVAL; | |
7812 | ||
7813 | /* | |
7814 | * Likewise, bound things on the otherside by preventing insane quota | |
7815 | * periods. This also allows us to normalize in computing quota | |
7816 | * feasibility. | |
7817 | */ | |
7818 | if (period > max_cfs_quota_period) | |
7819 | return -EINVAL; | |
7820 | ||
0e59bdae KT |
7821 | /* |
7822 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
7823 | * unthrottle_offline_cfs_rqs(). | |
7824 | */ | |
7825 | get_online_cpus(); | |
a790de99 PT |
7826 | mutex_lock(&cfs_constraints_mutex); |
7827 | ret = __cfs_schedulable(tg, period, quota); | |
7828 | if (ret) | |
7829 | goto out_unlock; | |
7830 | ||
58088ad0 | 7831 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7832 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7833 | /* |
7834 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7835 | * before making related changes, and on->off must occur afterwards | |
7836 | */ | |
7837 | if (runtime_enabled && !runtime_was_enabled) | |
7838 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7839 | raw_spin_lock_irq(&cfs_b->lock); |
7840 | cfs_b->period = ns_to_ktime(period); | |
7841 | cfs_b->quota = quota; | |
58088ad0 | 7842 | |
a9cf55b2 | 7843 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7844 | /* restart the period timer (if active) to handle new period expiry */ |
7845 | if (runtime_enabled && cfs_b->timer_active) { | |
7846 | /* force a reprogram */ | |
09dc4ab0 | 7847 | __start_cfs_bandwidth(cfs_b, true); |
58088ad0 | 7848 | } |
ab84d31e PT |
7849 | raw_spin_unlock_irq(&cfs_b->lock); |
7850 | ||
0e59bdae | 7851 | for_each_online_cpu(i) { |
ab84d31e | 7852 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7853 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7854 | |
7855 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7856 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7857 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7858 | |
029632fb | 7859 | if (cfs_rq->throttled) |
671fd9da | 7860 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7861 | raw_spin_unlock_irq(&rq->lock); |
7862 | } | |
1ee14e6c BS |
7863 | if (runtime_was_enabled && !runtime_enabled) |
7864 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7865 | out_unlock: |
7866 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7867 | put_online_cpus(); |
ab84d31e | 7868 | |
a790de99 | 7869 | return ret; |
ab84d31e PT |
7870 | } |
7871 | ||
7872 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7873 | { | |
7874 | u64 quota, period; | |
7875 | ||
029632fb | 7876 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7877 | if (cfs_quota_us < 0) |
7878 | quota = RUNTIME_INF; | |
7879 | else | |
7880 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7881 | ||
7882 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7883 | } | |
7884 | ||
7885 | long tg_get_cfs_quota(struct task_group *tg) | |
7886 | { | |
7887 | u64 quota_us; | |
7888 | ||
029632fb | 7889 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7890 | return -1; |
7891 | ||
029632fb | 7892 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7893 | do_div(quota_us, NSEC_PER_USEC); |
7894 | ||
7895 | return quota_us; | |
7896 | } | |
7897 | ||
7898 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7899 | { | |
7900 | u64 quota, period; | |
7901 | ||
7902 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7903 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7904 | |
ab84d31e PT |
7905 | return tg_set_cfs_bandwidth(tg, period, quota); |
7906 | } | |
7907 | ||
7908 | long tg_get_cfs_period(struct task_group *tg) | |
7909 | { | |
7910 | u64 cfs_period_us; | |
7911 | ||
029632fb | 7912 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7913 | do_div(cfs_period_us, NSEC_PER_USEC); |
7914 | ||
7915 | return cfs_period_us; | |
7916 | } | |
7917 | ||
182446d0 TH |
7918 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7919 | struct cftype *cft) | |
ab84d31e | 7920 | { |
182446d0 | 7921 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7922 | } |
7923 | ||
182446d0 TH |
7924 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7925 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7926 | { |
182446d0 | 7927 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7928 | } |
7929 | ||
182446d0 TH |
7930 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7931 | struct cftype *cft) | |
ab84d31e | 7932 | { |
182446d0 | 7933 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7934 | } |
7935 | ||
182446d0 TH |
7936 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7937 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7938 | { |
182446d0 | 7939 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7940 | } |
7941 | ||
a790de99 PT |
7942 | struct cfs_schedulable_data { |
7943 | struct task_group *tg; | |
7944 | u64 period, quota; | |
7945 | }; | |
7946 | ||
7947 | /* | |
7948 | * normalize group quota/period to be quota/max_period | |
7949 | * note: units are usecs | |
7950 | */ | |
7951 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7952 | struct cfs_schedulable_data *d) | |
7953 | { | |
7954 | u64 quota, period; | |
7955 | ||
7956 | if (tg == d->tg) { | |
7957 | period = d->period; | |
7958 | quota = d->quota; | |
7959 | } else { | |
7960 | period = tg_get_cfs_period(tg); | |
7961 | quota = tg_get_cfs_quota(tg); | |
7962 | } | |
7963 | ||
7964 | /* note: these should typically be equivalent */ | |
7965 | if (quota == RUNTIME_INF || quota == -1) | |
7966 | return RUNTIME_INF; | |
7967 | ||
7968 | return to_ratio(period, quota); | |
7969 | } | |
7970 | ||
7971 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7972 | { | |
7973 | struct cfs_schedulable_data *d = data; | |
029632fb | 7974 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7975 | s64 quota = 0, parent_quota = -1; |
7976 | ||
7977 | if (!tg->parent) { | |
7978 | quota = RUNTIME_INF; | |
7979 | } else { | |
029632fb | 7980 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7981 | |
7982 | quota = normalize_cfs_quota(tg, d); | |
7983 | parent_quota = parent_b->hierarchal_quota; | |
7984 | ||
7985 | /* | |
7986 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7987 | * limit is set | |
7988 | */ | |
7989 | if (quota == RUNTIME_INF) | |
7990 | quota = parent_quota; | |
7991 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7992 | return -EINVAL; | |
7993 | } | |
7994 | cfs_b->hierarchal_quota = quota; | |
7995 | ||
7996 | return 0; | |
7997 | } | |
7998 | ||
7999 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8000 | { | |
8277434e | 8001 | int ret; |
a790de99 PT |
8002 | struct cfs_schedulable_data data = { |
8003 | .tg = tg, | |
8004 | .period = period, | |
8005 | .quota = quota, | |
8006 | }; | |
8007 | ||
8008 | if (quota != RUNTIME_INF) { | |
8009 | do_div(data.period, NSEC_PER_USEC); | |
8010 | do_div(data.quota, NSEC_PER_USEC); | |
8011 | } | |
8012 | ||
8277434e PT |
8013 | rcu_read_lock(); |
8014 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8015 | rcu_read_unlock(); | |
8016 | ||
8017 | return ret; | |
a790de99 | 8018 | } |
e8da1b18 | 8019 | |
2da8ca82 | 8020 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8021 | { |
2da8ca82 | 8022 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8023 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8024 | |
44ffc75b TH |
8025 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8026 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8027 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8028 | |
8029 | return 0; | |
8030 | } | |
ab84d31e | 8031 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8032 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8033 | |
052f1dc7 | 8034 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8035 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8036 | struct cftype *cft, s64 val) | |
6f505b16 | 8037 | { |
182446d0 | 8038 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8039 | } |
8040 | ||
182446d0 TH |
8041 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8042 | struct cftype *cft) | |
6f505b16 | 8043 | { |
182446d0 | 8044 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8045 | } |
d0b27fa7 | 8046 | |
182446d0 TH |
8047 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8048 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8049 | { |
182446d0 | 8050 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8051 | } |
8052 | ||
182446d0 TH |
8053 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8054 | struct cftype *cft) | |
d0b27fa7 | 8055 | { |
182446d0 | 8056 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8057 | } |
6d6bc0ad | 8058 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8059 | |
fe5c7cc2 | 8060 | static struct cftype cpu_files[] = { |
052f1dc7 | 8061 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8062 | { |
8063 | .name = "shares", | |
f4c753b7 PM |
8064 | .read_u64 = cpu_shares_read_u64, |
8065 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8066 | }, |
052f1dc7 | 8067 | #endif |
ab84d31e PT |
8068 | #ifdef CONFIG_CFS_BANDWIDTH |
8069 | { | |
8070 | .name = "cfs_quota_us", | |
8071 | .read_s64 = cpu_cfs_quota_read_s64, | |
8072 | .write_s64 = cpu_cfs_quota_write_s64, | |
8073 | }, | |
8074 | { | |
8075 | .name = "cfs_period_us", | |
8076 | .read_u64 = cpu_cfs_period_read_u64, | |
8077 | .write_u64 = cpu_cfs_period_write_u64, | |
8078 | }, | |
e8da1b18 NR |
8079 | { |
8080 | .name = "stat", | |
2da8ca82 | 8081 | .seq_show = cpu_stats_show, |
e8da1b18 | 8082 | }, |
ab84d31e | 8083 | #endif |
052f1dc7 | 8084 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8085 | { |
9f0c1e56 | 8086 | .name = "rt_runtime_us", |
06ecb27c PM |
8087 | .read_s64 = cpu_rt_runtime_read, |
8088 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8089 | }, |
d0b27fa7 PZ |
8090 | { |
8091 | .name = "rt_period_us", | |
f4c753b7 PM |
8092 | .read_u64 = cpu_rt_period_read_uint, |
8093 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8094 | }, |
052f1dc7 | 8095 | #endif |
4baf6e33 | 8096 | { } /* terminate */ |
68318b8e SV |
8097 | }; |
8098 | ||
073219e9 | 8099 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8100 | .css_alloc = cpu_cgroup_css_alloc, |
8101 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8102 | .css_online = cpu_cgroup_css_online, |
8103 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8104 | .can_attach = cpu_cgroup_can_attach, |
8105 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8106 | .exit = cpu_cgroup_exit, |
4baf6e33 | 8107 | .base_cftypes = cpu_files, |
68318b8e SV |
8108 | .early_init = 1, |
8109 | }; | |
8110 | ||
052f1dc7 | 8111 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8112 | |
b637a328 PM |
8113 | void dump_cpu_task(int cpu) |
8114 | { | |
8115 | pr_info("Task dump for CPU %d:\n", cpu); | |
8116 | sched_show_task(cpu_curr(cpu)); | |
8117 | } |