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; | |
5cd08fbf | 248 | struct inode *inode; |
1a687c2e MG |
249 | |
250 | if (cnt > 63) | |
251 | cnt = 63; | |
252 | ||
253 | if (copy_from_user(&buf, ubuf, cnt)) | |
254 | return -EFAULT; | |
255 | ||
256 | buf[cnt] = 0; | |
257 | cmp = strstrip(buf); | |
258 | ||
5cd08fbf JB |
259 | /* Ensure the static_key remains in a consistent state */ |
260 | inode = file_inode(filp); | |
261 | mutex_lock(&inode->i_mutex); | |
1a687c2e | 262 | i = sched_feat_set(cmp); |
5cd08fbf | 263 | mutex_unlock(&inode->i_mutex); |
f8b6d1cc | 264 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
265 | return -EINVAL; |
266 | ||
42994724 | 267 | *ppos += cnt; |
f00b45c1 PZ |
268 | |
269 | return cnt; | |
270 | } | |
271 | ||
34f3a814 LZ |
272 | static int sched_feat_open(struct inode *inode, struct file *filp) |
273 | { | |
274 | return single_open(filp, sched_feat_show, NULL); | |
275 | } | |
276 | ||
828c0950 | 277 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
278 | .open = sched_feat_open, |
279 | .write = sched_feat_write, | |
280 | .read = seq_read, | |
281 | .llseek = seq_lseek, | |
282 | .release = single_release, | |
f00b45c1 PZ |
283 | }; |
284 | ||
285 | static __init int sched_init_debug(void) | |
286 | { | |
f00b45c1 PZ |
287 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
288 | &sched_feat_fops); | |
289 | ||
290 | return 0; | |
291 | } | |
292 | late_initcall(sched_init_debug); | |
f8b6d1cc | 293 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 294 | |
b82d9fdd PZ |
295 | /* |
296 | * Number of tasks to iterate in a single balance run. | |
297 | * Limited because this is done with IRQs disabled. | |
298 | */ | |
299 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
300 | ||
e9e9250b PZ |
301 | /* |
302 | * period over which we average the RT time consumption, measured | |
303 | * in ms. | |
304 | * | |
305 | * default: 1s | |
306 | */ | |
307 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
308 | ||
fa85ae24 | 309 | /* |
9f0c1e56 | 310 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
311 | * default: 1s |
312 | */ | |
9f0c1e56 | 313 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 314 | |
029632fb | 315 | __read_mostly int scheduler_running; |
6892b75e | 316 | |
9f0c1e56 PZ |
317 | /* |
318 | * part of the period that we allow rt tasks to run in us. | |
319 | * default: 0.95s | |
320 | */ | |
321 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 322 | |
0970d299 | 323 | /* |
0122ec5b | 324 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 325 | */ |
70b97a7f | 326 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
327 | __acquires(rq->lock) |
328 | { | |
0970d299 PZ |
329 | struct rq *rq; |
330 | ||
0122ec5b PZ |
331 | lockdep_assert_held(&p->pi_lock); |
332 | ||
3a5c359a | 333 | for (;;) { |
0970d299 | 334 | rq = task_rq(p); |
05fa785c | 335 | raw_spin_lock(&rq->lock); |
cca26e80 | 336 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) |
3a5c359a | 337 | return rq; |
05fa785c | 338 | raw_spin_unlock(&rq->lock); |
cca26e80 KT |
339 | |
340 | while (unlikely(task_on_rq_migrating(p))) | |
341 | cpu_relax(); | |
b29739f9 | 342 | } |
b29739f9 IM |
343 | } |
344 | ||
1da177e4 | 345 | /* |
0122ec5b | 346 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 347 | */ |
70b97a7f | 348 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 349 | __acquires(p->pi_lock) |
1da177e4 LT |
350 | __acquires(rq->lock) |
351 | { | |
70b97a7f | 352 | struct rq *rq; |
1da177e4 | 353 | |
3a5c359a | 354 | for (;;) { |
0122ec5b | 355 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 356 | rq = task_rq(p); |
05fa785c | 357 | raw_spin_lock(&rq->lock); |
cca26e80 | 358 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) |
3a5c359a | 359 | return rq; |
0122ec5b PZ |
360 | raw_spin_unlock(&rq->lock); |
361 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
cca26e80 KT |
362 | |
363 | while (unlikely(task_on_rq_migrating(p))) | |
364 | cpu_relax(); | |
1da177e4 | 365 | } |
1da177e4 LT |
366 | } |
367 | ||
a9957449 | 368 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
369 | __releases(rq->lock) |
370 | { | |
05fa785c | 371 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
372 | } |
373 | ||
0122ec5b PZ |
374 | static inline void |
375 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 376 | __releases(rq->lock) |
0122ec5b | 377 | __releases(p->pi_lock) |
1da177e4 | 378 | { |
0122ec5b PZ |
379 | raw_spin_unlock(&rq->lock); |
380 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
381 | } |
382 | ||
1da177e4 | 383 | /* |
cc2a73b5 | 384 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 385 | */ |
a9957449 | 386 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
387 | __acquires(rq->lock) |
388 | { | |
70b97a7f | 389 | struct rq *rq; |
1da177e4 LT |
390 | |
391 | local_irq_disable(); | |
392 | rq = this_rq(); | |
05fa785c | 393 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
394 | |
395 | return rq; | |
396 | } | |
397 | ||
8f4d37ec PZ |
398 | #ifdef CONFIG_SCHED_HRTICK |
399 | /* | |
400 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 401 | */ |
8f4d37ec | 402 | |
8f4d37ec PZ |
403 | static void hrtick_clear(struct rq *rq) |
404 | { | |
405 | if (hrtimer_active(&rq->hrtick_timer)) | |
406 | hrtimer_cancel(&rq->hrtick_timer); | |
407 | } | |
408 | ||
8f4d37ec PZ |
409 | /* |
410 | * High-resolution timer tick. | |
411 | * Runs from hardirq context with interrupts disabled. | |
412 | */ | |
413 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
414 | { | |
415 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
416 | ||
417 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
418 | ||
05fa785c | 419 | raw_spin_lock(&rq->lock); |
3e51f33f | 420 | update_rq_clock(rq); |
8f4d37ec | 421 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 422 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
423 | |
424 | return HRTIMER_NORESTART; | |
425 | } | |
426 | ||
95e904c7 | 427 | #ifdef CONFIG_SMP |
971ee28c PZ |
428 | |
429 | static int __hrtick_restart(struct rq *rq) | |
430 | { | |
431 | struct hrtimer *timer = &rq->hrtick_timer; | |
432 | ktime_t time = hrtimer_get_softexpires(timer); | |
433 | ||
434 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
435 | } | |
436 | ||
31656519 PZ |
437 | /* |
438 | * called from hardirq (IPI) context | |
439 | */ | |
440 | static void __hrtick_start(void *arg) | |
b328ca18 | 441 | { |
31656519 | 442 | struct rq *rq = arg; |
b328ca18 | 443 | |
05fa785c | 444 | raw_spin_lock(&rq->lock); |
971ee28c | 445 | __hrtick_restart(rq); |
31656519 | 446 | rq->hrtick_csd_pending = 0; |
05fa785c | 447 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
448 | } |
449 | ||
31656519 PZ |
450 | /* |
451 | * Called to set the hrtick timer state. | |
452 | * | |
453 | * called with rq->lock held and irqs disabled | |
454 | */ | |
029632fb | 455 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 456 | { |
31656519 PZ |
457 | struct hrtimer *timer = &rq->hrtick_timer; |
458 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 459 | |
cc584b21 | 460 | hrtimer_set_expires(timer, time); |
31656519 PZ |
461 | |
462 | if (rq == this_rq()) { | |
971ee28c | 463 | __hrtick_restart(rq); |
31656519 | 464 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 465 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
466 | rq->hrtick_csd_pending = 1; |
467 | } | |
b328ca18 PZ |
468 | } |
469 | ||
470 | static int | |
471 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
472 | { | |
473 | int cpu = (int)(long)hcpu; | |
474 | ||
475 | switch (action) { | |
476 | case CPU_UP_CANCELED: | |
477 | case CPU_UP_CANCELED_FROZEN: | |
478 | case CPU_DOWN_PREPARE: | |
479 | case CPU_DOWN_PREPARE_FROZEN: | |
480 | case CPU_DEAD: | |
481 | case CPU_DEAD_FROZEN: | |
31656519 | 482 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
483 | return NOTIFY_OK; |
484 | } | |
485 | ||
486 | return NOTIFY_DONE; | |
487 | } | |
488 | ||
fa748203 | 489 | static __init void init_hrtick(void) |
b328ca18 PZ |
490 | { |
491 | hotcpu_notifier(hotplug_hrtick, 0); | |
492 | } | |
31656519 PZ |
493 | #else |
494 | /* | |
495 | * Called to set the hrtick timer state. | |
496 | * | |
497 | * called with rq->lock held and irqs disabled | |
498 | */ | |
029632fb | 499 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 500 | { |
7f1e2ca9 | 501 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 502 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 503 | } |
b328ca18 | 504 | |
006c75f1 | 505 | static inline void init_hrtick(void) |
8f4d37ec | 506 | { |
8f4d37ec | 507 | } |
31656519 | 508 | #endif /* CONFIG_SMP */ |
8f4d37ec | 509 | |
31656519 | 510 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 511 | { |
31656519 PZ |
512 | #ifdef CONFIG_SMP |
513 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 514 | |
31656519 PZ |
515 | rq->hrtick_csd.flags = 0; |
516 | rq->hrtick_csd.func = __hrtick_start; | |
517 | rq->hrtick_csd.info = rq; | |
518 | #endif | |
8f4d37ec | 519 | |
31656519 PZ |
520 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
521 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 522 | } |
006c75f1 | 523 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
524 | static inline void hrtick_clear(struct rq *rq) |
525 | { | |
526 | } | |
527 | ||
8f4d37ec PZ |
528 | static inline void init_rq_hrtick(struct rq *rq) |
529 | { | |
530 | } | |
531 | ||
b328ca18 PZ |
532 | static inline void init_hrtick(void) |
533 | { | |
534 | } | |
006c75f1 | 535 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 536 | |
fd99f91a PZ |
537 | /* |
538 | * cmpxchg based fetch_or, macro so it works for different integer types | |
539 | */ | |
540 | #define fetch_or(ptr, val) \ | |
541 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
542 | for (;;) { \ | |
543 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
544 | if (__old == __val) \ | |
545 | break; \ | |
546 | __val = __old; \ | |
547 | } \ | |
548 | __old; \ | |
549 | }) | |
550 | ||
e3baac47 | 551 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
552 | /* |
553 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
554 | * this avoids any races wrt polling state changes and thereby avoids | |
555 | * spurious IPIs. | |
556 | */ | |
557 | static bool set_nr_and_not_polling(struct task_struct *p) | |
558 | { | |
559 | struct thread_info *ti = task_thread_info(p); | |
560 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
561 | } | |
e3baac47 PZ |
562 | |
563 | /* | |
564 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
565 | * | |
566 | * If this returns true, then the idle task promises to call | |
567 | * sched_ttwu_pending() and reschedule soon. | |
568 | */ | |
569 | static bool set_nr_if_polling(struct task_struct *p) | |
570 | { | |
571 | struct thread_info *ti = task_thread_info(p); | |
572 | typeof(ti->flags) old, val = ACCESS_ONCE(ti->flags); | |
573 | ||
574 | for (;;) { | |
575 | if (!(val & _TIF_POLLING_NRFLAG)) | |
576 | return false; | |
577 | if (val & _TIF_NEED_RESCHED) | |
578 | return true; | |
579 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
580 | if (old == val) | |
581 | break; | |
582 | val = old; | |
583 | } | |
584 | return true; | |
585 | } | |
586 | ||
fd99f91a PZ |
587 | #else |
588 | static bool set_nr_and_not_polling(struct task_struct *p) | |
589 | { | |
590 | set_tsk_need_resched(p); | |
591 | return true; | |
592 | } | |
e3baac47 PZ |
593 | |
594 | #ifdef CONFIG_SMP | |
595 | static bool set_nr_if_polling(struct task_struct *p) | |
596 | { | |
597 | return false; | |
598 | } | |
599 | #endif | |
fd99f91a PZ |
600 | #endif |
601 | ||
c24d20db | 602 | /* |
8875125e | 603 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
604 | * |
605 | * On UP this means the setting of the need_resched flag, on SMP it | |
606 | * might also involve a cross-CPU call to trigger the scheduler on | |
607 | * the target CPU. | |
608 | */ | |
8875125e | 609 | void resched_curr(struct rq *rq) |
c24d20db | 610 | { |
8875125e | 611 | struct task_struct *curr = rq->curr; |
c24d20db IM |
612 | int cpu; |
613 | ||
8875125e | 614 | lockdep_assert_held(&rq->lock); |
c24d20db | 615 | |
8875125e | 616 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
617 | return; |
618 | ||
8875125e | 619 | cpu = cpu_of(rq); |
fd99f91a | 620 | |
f27dde8d | 621 | if (cpu == smp_processor_id()) { |
8875125e | 622 | set_tsk_need_resched(curr); |
f27dde8d | 623 | set_preempt_need_resched(); |
c24d20db | 624 | return; |
f27dde8d | 625 | } |
c24d20db | 626 | |
8875125e | 627 | if (set_nr_and_not_polling(curr)) |
c24d20db | 628 | smp_send_reschedule(cpu); |
dfc68f29 AL |
629 | else |
630 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
631 | } |
632 | ||
029632fb | 633 | void resched_cpu(int cpu) |
c24d20db IM |
634 | { |
635 | struct rq *rq = cpu_rq(cpu); | |
636 | unsigned long flags; | |
637 | ||
05fa785c | 638 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 639 | return; |
8875125e | 640 | resched_curr(rq); |
05fa785c | 641 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 642 | } |
06d8308c | 643 | |
b021fe3e | 644 | #ifdef CONFIG_SMP |
3451d024 | 645 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
646 | /* |
647 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
648 | * from an idle cpu. This is good for power-savings. | |
649 | * | |
650 | * We don't do similar optimization for completely idle system, as | |
651 | * selecting an idle cpu will add more delays to the timers than intended | |
652 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
653 | */ | |
6201b4d6 | 654 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
655 | { |
656 | int cpu = smp_processor_id(); | |
657 | int i; | |
658 | struct sched_domain *sd; | |
659 | ||
6201b4d6 VK |
660 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
661 | return cpu; | |
662 | ||
057f3fad | 663 | rcu_read_lock(); |
83cd4fe2 | 664 | for_each_domain(cpu, sd) { |
057f3fad PZ |
665 | for_each_cpu(i, sched_domain_span(sd)) { |
666 | if (!idle_cpu(i)) { | |
667 | cpu = i; | |
668 | goto unlock; | |
669 | } | |
670 | } | |
83cd4fe2 | 671 | } |
057f3fad PZ |
672 | unlock: |
673 | rcu_read_unlock(); | |
83cd4fe2 VP |
674 | return cpu; |
675 | } | |
06d8308c TG |
676 | /* |
677 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
678 | * idle CPU then this timer might expire before the next timer event | |
679 | * which is scheduled to wake up that CPU. In case of a completely | |
680 | * idle system the next event might even be infinite time into the | |
681 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
682 | * leaves the inner idle loop so the newly added timer is taken into | |
683 | * account when the CPU goes back to idle and evaluates the timer | |
684 | * wheel for the next timer event. | |
685 | */ | |
1c20091e | 686 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
687 | { |
688 | struct rq *rq = cpu_rq(cpu); | |
689 | ||
690 | if (cpu == smp_processor_id()) | |
691 | return; | |
692 | ||
67b9ca70 | 693 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 694 | smp_send_reschedule(cpu); |
dfc68f29 AL |
695 | else |
696 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
697 | } |
698 | ||
c5bfece2 | 699 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 700 | { |
53c5fa16 FW |
701 | /* |
702 | * We just need the target to call irq_exit() and re-evaluate | |
703 | * the next tick. The nohz full kick at least implies that. | |
704 | * If needed we can still optimize that later with an | |
705 | * empty IRQ. | |
706 | */ | |
c5bfece2 | 707 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
708 | if (cpu != smp_processor_id() || |
709 | tick_nohz_tick_stopped()) | |
53c5fa16 | 710 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
711 | return true; |
712 | } | |
713 | ||
714 | return false; | |
715 | } | |
716 | ||
717 | void wake_up_nohz_cpu(int cpu) | |
718 | { | |
c5bfece2 | 719 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
720 | wake_up_idle_cpu(cpu); |
721 | } | |
722 | ||
ca38062e | 723 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 724 | { |
1c792db7 | 725 | int cpu = smp_processor_id(); |
873b4c65 VG |
726 | |
727 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
728 | return false; | |
729 | ||
730 | if (idle_cpu(cpu) && !need_resched()) | |
731 | return true; | |
732 | ||
733 | /* | |
734 | * We can't run Idle Load Balance on this CPU for this time so we | |
735 | * cancel it and clear NOHZ_BALANCE_KICK | |
736 | */ | |
737 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
738 | return false; | |
45bf76df IM |
739 | } |
740 | ||
3451d024 | 741 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 742 | |
ca38062e | 743 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 744 | { |
ca38062e | 745 | return false; |
2069dd75 PZ |
746 | } |
747 | ||
3451d024 | 748 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 749 | |
ce831b38 FW |
750 | #ifdef CONFIG_NO_HZ_FULL |
751 | bool sched_can_stop_tick(void) | |
752 | { | |
3882ec64 FW |
753 | /* |
754 | * More than one running task need preemption. | |
755 | * nr_running update is assumed to be visible | |
756 | * after IPI is sent from wakers. | |
757 | */ | |
541b8264 VK |
758 | if (this_rq()->nr_running > 1) |
759 | return false; | |
ce831b38 | 760 | |
541b8264 | 761 | return true; |
ce831b38 FW |
762 | } |
763 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 764 | |
029632fb | 765 | void sched_avg_update(struct rq *rq) |
18d95a28 | 766 | { |
e9e9250b PZ |
767 | s64 period = sched_avg_period(); |
768 | ||
78becc27 | 769 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
770 | /* |
771 | * Inline assembly required to prevent the compiler | |
772 | * optimising this loop into a divmod call. | |
773 | * See __iter_div_u64_rem() for another example of this. | |
774 | */ | |
775 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
776 | rq->age_stamp += period; |
777 | rq->rt_avg /= 2; | |
778 | } | |
18d95a28 PZ |
779 | } |
780 | ||
6d6bc0ad | 781 | #endif /* CONFIG_SMP */ |
18d95a28 | 782 | |
a790de99 PT |
783 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
784 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 785 | /* |
8277434e PT |
786 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
787 | * node and @up when leaving it for the final time. | |
788 | * | |
789 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 790 | */ |
029632fb | 791 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 792 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
793 | { |
794 | struct task_group *parent, *child; | |
eb755805 | 795 | int ret; |
c09595f6 | 796 | |
8277434e PT |
797 | parent = from; |
798 | ||
c09595f6 | 799 | down: |
eb755805 PZ |
800 | ret = (*down)(parent, data); |
801 | if (ret) | |
8277434e | 802 | goto out; |
c09595f6 PZ |
803 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
804 | parent = child; | |
805 | goto down; | |
806 | ||
807 | up: | |
808 | continue; | |
809 | } | |
eb755805 | 810 | ret = (*up)(parent, data); |
8277434e PT |
811 | if (ret || parent == from) |
812 | goto out; | |
c09595f6 PZ |
813 | |
814 | child = parent; | |
815 | parent = parent->parent; | |
816 | if (parent) | |
817 | goto up; | |
8277434e | 818 | out: |
eb755805 | 819 | return ret; |
c09595f6 PZ |
820 | } |
821 | ||
029632fb | 822 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 823 | { |
e2b245f8 | 824 | return 0; |
eb755805 | 825 | } |
18d95a28 PZ |
826 | #endif |
827 | ||
45bf76df IM |
828 | static void set_load_weight(struct task_struct *p) |
829 | { | |
f05998d4 NR |
830 | int prio = p->static_prio - MAX_RT_PRIO; |
831 | struct load_weight *load = &p->se.load; | |
832 | ||
dd41f596 IM |
833 | /* |
834 | * SCHED_IDLE tasks get minimal weight: | |
835 | */ | |
836 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 837 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 838 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
839 | return; |
840 | } | |
71f8bd46 | 841 | |
c8b28116 | 842 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 843 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
844 | } |
845 | ||
371fd7e7 | 846 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 847 | { |
a64692a3 | 848 | update_rq_clock(rq); |
43148951 | 849 | sched_info_queued(rq, p); |
371fd7e7 | 850 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
851 | } |
852 | ||
371fd7e7 | 853 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 854 | { |
a64692a3 | 855 | update_rq_clock(rq); |
43148951 | 856 | sched_info_dequeued(rq, p); |
371fd7e7 | 857 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
858 | } |
859 | ||
029632fb | 860 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
861 | { |
862 | if (task_contributes_to_load(p)) | |
863 | rq->nr_uninterruptible--; | |
864 | ||
371fd7e7 | 865 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
866 | } |
867 | ||
029632fb | 868 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
869 | { |
870 | if (task_contributes_to_load(p)) | |
871 | rq->nr_uninterruptible++; | |
872 | ||
371fd7e7 | 873 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
874 | } |
875 | ||
fe44d621 | 876 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 877 | { |
095c0aa8 GC |
878 | /* |
879 | * In theory, the compile should just see 0 here, and optimize out the call | |
880 | * to sched_rt_avg_update. But I don't trust it... | |
881 | */ | |
882 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
883 | s64 steal = 0, irq_delta = 0; | |
884 | #endif | |
885 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 886 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
887 | |
888 | /* | |
889 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
890 | * this case when a previous update_rq_clock() happened inside a | |
891 | * {soft,}irq region. | |
892 | * | |
893 | * When this happens, we stop ->clock_task and only update the | |
894 | * prev_irq_time stamp to account for the part that fit, so that a next | |
895 | * update will consume the rest. This ensures ->clock_task is | |
896 | * monotonic. | |
897 | * | |
898 | * It does however cause some slight miss-attribution of {soft,}irq | |
899 | * time, a more accurate solution would be to update the irq_time using | |
900 | * the current rq->clock timestamp, except that would require using | |
901 | * atomic ops. | |
902 | */ | |
903 | if (irq_delta > delta) | |
904 | irq_delta = delta; | |
905 | ||
906 | rq->prev_irq_time += irq_delta; | |
907 | delta -= irq_delta; | |
095c0aa8 GC |
908 | #endif |
909 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 910 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
911 | steal = paravirt_steal_clock(cpu_of(rq)); |
912 | steal -= rq->prev_steal_time_rq; | |
913 | ||
914 | if (unlikely(steal > delta)) | |
915 | steal = delta; | |
916 | ||
095c0aa8 | 917 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
918 | delta -= steal; |
919 | } | |
920 | #endif | |
921 | ||
fe44d621 PZ |
922 | rq->clock_task += delta; |
923 | ||
095c0aa8 | 924 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 925 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
926 | sched_rt_avg_update(rq, irq_delta + steal); |
927 | #endif | |
aa483808 VP |
928 | } |
929 | ||
34f971f6 PZ |
930 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
931 | { | |
932 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
933 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
934 | ||
935 | if (stop) { | |
936 | /* | |
937 | * Make it appear like a SCHED_FIFO task, its something | |
938 | * userspace knows about and won't get confused about. | |
939 | * | |
940 | * Also, it will make PI more or less work without too | |
941 | * much confusion -- but then, stop work should not | |
942 | * rely on PI working anyway. | |
943 | */ | |
944 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
945 | ||
946 | stop->sched_class = &stop_sched_class; | |
947 | } | |
948 | ||
949 | cpu_rq(cpu)->stop = stop; | |
950 | ||
951 | if (old_stop) { | |
952 | /* | |
953 | * Reset it back to a normal scheduling class so that | |
954 | * it can die in pieces. | |
955 | */ | |
956 | old_stop->sched_class = &rt_sched_class; | |
957 | } | |
958 | } | |
959 | ||
14531189 | 960 | /* |
dd41f596 | 961 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 962 | */ |
14531189 IM |
963 | static inline int __normal_prio(struct task_struct *p) |
964 | { | |
dd41f596 | 965 | return p->static_prio; |
14531189 IM |
966 | } |
967 | ||
b29739f9 IM |
968 | /* |
969 | * Calculate the expected normal priority: i.e. priority | |
970 | * without taking RT-inheritance into account. Might be | |
971 | * boosted by interactivity modifiers. Changes upon fork, | |
972 | * setprio syscalls, and whenever the interactivity | |
973 | * estimator recalculates. | |
974 | */ | |
36c8b586 | 975 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
976 | { |
977 | int prio; | |
978 | ||
aab03e05 DF |
979 | if (task_has_dl_policy(p)) |
980 | prio = MAX_DL_PRIO-1; | |
981 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
982 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
983 | else | |
984 | prio = __normal_prio(p); | |
985 | return prio; | |
986 | } | |
987 | ||
988 | /* | |
989 | * Calculate the current priority, i.e. the priority | |
990 | * taken into account by the scheduler. This value might | |
991 | * be boosted by RT tasks, or might be boosted by | |
992 | * interactivity modifiers. Will be RT if the task got | |
993 | * RT-boosted. If not then it returns p->normal_prio. | |
994 | */ | |
36c8b586 | 995 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
996 | { |
997 | p->normal_prio = normal_prio(p); | |
998 | /* | |
999 | * If we are RT tasks or we were boosted to RT priority, | |
1000 | * keep the priority unchanged. Otherwise, update priority | |
1001 | * to the normal priority: | |
1002 | */ | |
1003 | if (!rt_prio(p->prio)) | |
1004 | return p->normal_prio; | |
1005 | return p->prio; | |
1006 | } | |
1007 | ||
1da177e4 LT |
1008 | /** |
1009 | * task_curr - is this task currently executing on a CPU? | |
1010 | * @p: the task in question. | |
e69f6186 YB |
1011 | * |
1012 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1013 | */ |
36c8b586 | 1014 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1015 | { |
1016 | return cpu_curr(task_cpu(p)) == p; | |
1017 | } | |
1018 | ||
cb469845 SR |
1019 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1020 | const struct sched_class *prev_class, | |
da7a735e | 1021 | int oldprio) |
cb469845 SR |
1022 | { |
1023 | if (prev_class != p->sched_class) { | |
1024 | if (prev_class->switched_from) | |
da7a735e PZ |
1025 | prev_class->switched_from(rq, p); |
1026 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 1027 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1028 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1029 | } |
1030 | ||
029632fb | 1031 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1032 | { |
1033 | const struct sched_class *class; | |
1034 | ||
1035 | if (p->sched_class == rq->curr->sched_class) { | |
1036 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1037 | } else { | |
1038 | for_each_class(class) { | |
1039 | if (class == rq->curr->sched_class) | |
1040 | break; | |
1041 | if (class == p->sched_class) { | |
8875125e | 1042 | resched_curr(rq); |
1e5a7405 PZ |
1043 | break; |
1044 | } | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * A queue event has occurred, and we're going to schedule. In | |
1050 | * this case, we can save a useless back to back clock update. | |
1051 | */ | |
da0c1e65 | 1052 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1053 | rq->skip_clock_update = 1; |
1054 | } | |
1055 | ||
1da177e4 | 1056 | #ifdef CONFIG_SMP |
dd41f596 | 1057 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1058 | { |
e2912009 PZ |
1059 | #ifdef CONFIG_SCHED_DEBUG |
1060 | /* | |
1061 | * We should never call set_task_cpu() on a blocked task, | |
1062 | * ttwu() will sort out the placement. | |
1063 | */ | |
077614ee | 1064 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1065 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1066 | |
1067 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1068 | /* |
1069 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1070 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1071 | * | |
1072 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1073 | * see task_group(). |
6c6c54e1 PZ |
1074 | * |
1075 | * Furthermore, all task_rq users should acquire both locks, see | |
1076 | * task_rq_lock(). | |
1077 | */ | |
0122ec5b PZ |
1078 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1079 | lockdep_is_held(&task_rq(p)->lock))); | |
1080 | #endif | |
e2912009 PZ |
1081 | #endif |
1082 | ||
de1d7286 | 1083 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1084 | |
0c69774e | 1085 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1086 | if (p->sched_class->migrate_task_rq) |
1087 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1088 | p->se.nr_migrations++; |
a8b0ca17 | 1089 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1090 | } |
dd41f596 IM |
1091 | |
1092 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1093 | } |
1094 | ||
ac66f547 PZ |
1095 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1096 | { | |
da0c1e65 | 1097 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1098 | struct rq *src_rq, *dst_rq; |
1099 | ||
1100 | src_rq = task_rq(p); | |
1101 | dst_rq = cpu_rq(cpu); | |
1102 | ||
1103 | deactivate_task(src_rq, p, 0); | |
1104 | set_task_cpu(p, cpu); | |
1105 | activate_task(dst_rq, p, 0); | |
1106 | check_preempt_curr(dst_rq, p, 0); | |
1107 | } else { | |
1108 | /* | |
1109 | * Task isn't running anymore; make it appear like we migrated | |
1110 | * it before it went to sleep. This means on wakeup we make the | |
1111 | * previous cpu our targer instead of where it really is. | |
1112 | */ | |
1113 | p->wake_cpu = cpu; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | struct migration_swap_arg { | |
1118 | struct task_struct *src_task, *dst_task; | |
1119 | int src_cpu, dst_cpu; | |
1120 | }; | |
1121 | ||
1122 | static int migrate_swap_stop(void *data) | |
1123 | { | |
1124 | struct migration_swap_arg *arg = data; | |
1125 | struct rq *src_rq, *dst_rq; | |
1126 | int ret = -EAGAIN; | |
1127 | ||
1128 | src_rq = cpu_rq(arg->src_cpu); | |
1129 | dst_rq = cpu_rq(arg->dst_cpu); | |
1130 | ||
74602315 PZ |
1131 | double_raw_lock(&arg->src_task->pi_lock, |
1132 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1133 | double_rq_lock(src_rq, dst_rq); |
1134 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1135 | goto unlock; | |
1136 | ||
1137 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1138 | goto unlock; | |
1139 | ||
1140 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1141 | goto unlock; | |
1142 | ||
1143 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1144 | goto unlock; | |
1145 | ||
1146 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1147 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1148 | ||
1149 | ret = 0; | |
1150 | ||
1151 | unlock: | |
1152 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1153 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1154 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1155 | |
1156 | return ret; | |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * Cross migrate two tasks | |
1161 | */ | |
1162 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1163 | { | |
1164 | struct migration_swap_arg arg; | |
1165 | int ret = -EINVAL; | |
1166 | ||
ac66f547 PZ |
1167 | arg = (struct migration_swap_arg){ |
1168 | .src_task = cur, | |
1169 | .src_cpu = task_cpu(cur), | |
1170 | .dst_task = p, | |
1171 | .dst_cpu = task_cpu(p), | |
1172 | }; | |
1173 | ||
1174 | if (arg.src_cpu == arg.dst_cpu) | |
1175 | goto out; | |
1176 | ||
6acce3ef PZ |
1177 | /* |
1178 | * These three tests are all lockless; this is OK since all of them | |
1179 | * will be re-checked with proper locks held further down the line. | |
1180 | */ | |
ac66f547 PZ |
1181 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1182 | goto out; | |
1183 | ||
1184 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1185 | goto out; | |
1186 | ||
1187 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1188 | goto out; | |
1189 | ||
286549dc | 1190 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1191 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1192 | ||
1193 | out: | |
ac66f547 PZ |
1194 | return ret; |
1195 | } | |
1196 | ||
969c7921 | 1197 | struct migration_arg { |
36c8b586 | 1198 | struct task_struct *task; |
1da177e4 | 1199 | int dest_cpu; |
70b97a7f | 1200 | }; |
1da177e4 | 1201 | |
969c7921 TH |
1202 | static int migration_cpu_stop(void *data); |
1203 | ||
1da177e4 LT |
1204 | /* |
1205 | * wait_task_inactive - wait for a thread to unschedule. | |
1206 | * | |
85ba2d86 RM |
1207 | * If @match_state is nonzero, it's the @p->state value just checked and |
1208 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1209 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1210 | * we return a positive number (its total switch count). If a second call | |
1211 | * a short while later returns the same number, the caller can be sure that | |
1212 | * @p has remained unscheduled the whole time. | |
1213 | * | |
1da177e4 LT |
1214 | * The caller must ensure that the task *will* unschedule sometime soon, |
1215 | * else this function might spin for a *long* time. This function can't | |
1216 | * be called with interrupts off, or it may introduce deadlock with | |
1217 | * smp_call_function() if an IPI is sent by the same process we are | |
1218 | * waiting to become inactive. | |
1219 | */ | |
85ba2d86 | 1220 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1221 | { |
1222 | unsigned long flags; | |
da0c1e65 | 1223 | int running, queued; |
85ba2d86 | 1224 | unsigned long ncsw; |
70b97a7f | 1225 | struct rq *rq; |
1da177e4 | 1226 | |
3a5c359a AK |
1227 | for (;;) { |
1228 | /* | |
1229 | * We do the initial early heuristics without holding | |
1230 | * any task-queue locks at all. We'll only try to get | |
1231 | * the runqueue lock when things look like they will | |
1232 | * work out! | |
1233 | */ | |
1234 | rq = task_rq(p); | |
fa490cfd | 1235 | |
3a5c359a AK |
1236 | /* |
1237 | * If the task is actively running on another CPU | |
1238 | * still, just relax and busy-wait without holding | |
1239 | * any locks. | |
1240 | * | |
1241 | * NOTE! Since we don't hold any locks, it's not | |
1242 | * even sure that "rq" stays as the right runqueue! | |
1243 | * But we don't care, since "task_running()" will | |
1244 | * return false if the runqueue has changed and p | |
1245 | * is actually now running somewhere else! | |
1246 | */ | |
85ba2d86 RM |
1247 | while (task_running(rq, p)) { |
1248 | if (match_state && unlikely(p->state != match_state)) | |
1249 | return 0; | |
3a5c359a | 1250 | cpu_relax(); |
85ba2d86 | 1251 | } |
fa490cfd | 1252 | |
3a5c359a AK |
1253 | /* |
1254 | * Ok, time to look more closely! We need the rq | |
1255 | * lock now, to be *sure*. If we're wrong, we'll | |
1256 | * just go back and repeat. | |
1257 | */ | |
1258 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1259 | trace_sched_wait_task(p); |
3a5c359a | 1260 | running = task_running(rq, p); |
da0c1e65 | 1261 | queued = task_on_rq_queued(p); |
85ba2d86 | 1262 | ncsw = 0; |
f31e11d8 | 1263 | if (!match_state || p->state == match_state) |
93dcf55f | 1264 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1265 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1266 | |
85ba2d86 RM |
1267 | /* |
1268 | * If it changed from the expected state, bail out now. | |
1269 | */ | |
1270 | if (unlikely(!ncsw)) | |
1271 | break; | |
1272 | ||
3a5c359a AK |
1273 | /* |
1274 | * Was it really running after all now that we | |
1275 | * checked with the proper locks actually held? | |
1276 | * | |
1277 | * Oops. Go back and try again.. | |
1278 | */ | |
1279 | if (unlikely(running)) { | |
1280 | cpu_relax(); | |
1281 | continue; | |
1282 | } | |
fa490cfd | 1283 | |
3a5c359a AK |
1284 | /* |
1285 | * It's not enough that it's not actively running, | |
1286 | * it must be off the runqueue _entirely_, and not | |
1287 | * preempted! | |
1288 | * | |
80dd99b3 | 1289 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1290 | * running right now), it's preempted, and we should |
1291 | * yield - it could be a while. | |
1292 | */ | |
da0c1e65 | 1293 | if (unlikely(queued)) { |
8eb90c30 TG |
1294 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1295 | ||
1296 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1297 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1298 | continue; |
1299 | } | |
fa490cfd | 1300 | |
3a5c359a AK |
1301 | /* |
1302 | * Ahh, all good. It wasn't running, and it wasn't | |
1303 | * runnable, which means that it will never become | |
1304 | * running in the future either. We're all done! | |
1305 | */ | |
1306 | break; | |
1307 | } | |
85ba2d86 RM |
1308 | |
1309 | return ncsw; | |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | /*** | |
1313 | * kick_process - kick a running thread to enter/exit the kernel | |
1314 | * @p: the to-be-kicked thread | |
1315 | * | |
1316 | * Cause a process which is running on another CPU to enter | |
1317 | * kernel-mode, without any delay. (to get signals handled.) | |
1318 | * | |
25985edc | 1319 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1320 | * because all it wants to ensure is that the remote task enters |
1321 | * the kernel. If the IPI races and the task has been migrated | |
1322 | * to another CPU then no harm is done and the purpose has been | |
1323 | * achieved as well. | |
1324 | */ | |
36c8b586 | 1325 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1326 | { |
1327 | int cpu; | |
1328 | ||
1329 | preempt_disable(); | |
1330 | cpu = task_cpu(p); | |
1331 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1332 | smp_send_reschedule(cpu); | |
1333 | preempt_enable(); | |
1334 | } | |
b43e3521 | 1335 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1336 | #endif /* CONFIG_SMP */ |
1da177e4 | 1337 | |
970b13ba | 1338 | #ifdef CONFIG_SMP |
30da688e | 1339 | /* |
013fdb80 | 1340 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1341 | */ |
5da9a0fb PZ |
1342 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1343 | { | |
aa00d89c TC |
1344 | int nid = cpu_to_node(cpu); |
1345 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1346 | enum { cpuset, possible, fail } state = cpuset; |
1347 | int dest_cpu; | |
5da9a0fb | 1348 | |
aa00d89c TC |
1349 | /* |
1350 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1351 | * will return -1. There is no cpu on the node, and we should | |
1352 | * select the cpu on the other node. | |
1353 | */ | |
1354 | if (nid != -1) { | |
1355 | nodemask = cpumask_of_node(nid); | |
1356 | ||
1357 | /* Look for allowed, online CPU in same node. */ | |
1358 | for_each_cpu(dest_cpu, nodemask) { | |
1359 | if (!cpu_online(dest_cpu)) | |
1360 | continue; | |
1361 | if (!cpu_active(dest_cpu)) | |
1362 | continue; | |
1363 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1364 | return dest_cpu; | |
1365 | } | |
2baab4e9 | 1366 | } |
5da9a0fb | 1367 | |
2baab4e9 PZ |
1368 | for (;;) { |
1369 | /* Any allowed, online CPU? */ | |
e3831edd | 1370 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1371 | if (!cpu_online(dest_cpu)) |
1372 | continue; | |
1373 | if (!cpu_active(dest_cpu)) | |
1374 | continue; | |
1375 | goto out; | |
1376 | } | |
5da9a0fb | 1377 | |
2baab4e9 PZ |
1378 | switch (state) { |
1379 | case cpuset: | |
1380 | /* No more Mr. Nice Guy. */ | |
1381 | cpuset_cpus_allowed_fallback(p); | |
1382 | state = possible; | |
1383 | break; | |
1384 | ||
1385 | case possible: | |
1386 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1387 | state = fail; | |
1388 | break; | |
1389 | ||
1390 | case fail: | |
1391 | BUG(); | |
1392 | break; | |
1393 | } | |
1394 | } | |
1395 | ||
1396 | out: | |
1397 | if (state != cpuset) { | |
1398 | /* | |
1399 | * Don't tell them about moving exiting tasks or | |
1400 | * kernel threads (both mm NULL), since they never | |
1401 | * leave kernel. | |
1402 | */ | |
1403 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1404 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1405 | task_pid_nr(p), p->comm, cpu); |
1406 | } | |
5da9a0fb PZ |
1407 | } |
1408 | ||
1409 | return dest_cpu; | |
1410 | } | |
1411 | ||
e2912009 | 1412 | /* |
013fdb80 | 1413 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1414 | */ |
970b13ba | 1415 | static inline |
ac66f547 | 1416 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1417 | { |
ac66f547 | 1418 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1419 | |
1420 | /* | |
1421 | * In order not to call set_task_cpu() on a blocking task we need | |
1422 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1423 | * cpu. | |
1424 | * | |
1425 | * Since this is common to all placement strategies, this lives here. | |
1426 | * | |
1427 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1428 | * not worry about this generic constraint ] | |
1429 | */ | |
fa17b507 | 1430 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1431 | !cpu_online(cpu))) |
5da9a0fb | 1432 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1433 | |
1434 | return cpu; | |
970b13ba | 1435 | } |
09a40af5 MG |
1436 | |
1437 | static void update_avg(u64 *avg, u64 sample) | |
1438 | { | |
1439 | s64 diff = sample - *avg; | |
1440 | *avg += diff >> 3; | |
1441 | } | |
970b13ba PZ |
1442 | #endif |
1443 | ||
d7c01d27 | 1444 | static void |
b84cb5df | 1445 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1446 | { |
d7c01d27 | 1447 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1448 | struct rq *rq = this_rq(); |
1449 | ||
d7c01d27 PZ |
1450 | #ifdef CONFIG_SMP |
1451 | int this_cpu = smp_processor_id(); | |
1452 | ||
1453 | if (cpu == this_cpu) { | |
1454 | schedstat_inc(rq, ttwu_local); | |
1455 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1456 | } else { | |
1457 | struct sched_domain *sd; | |
1458 | ||
1459 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1460 | rcu_read_lock(); |
d7c01d27 PZ |
1461 | for_each_domain(this_cpu, sd) { |
1462 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1463 | schedstat_inc(sd, ttwu_wake_remote); | |
1464 | break; | |
1465 | } | |
1466 | } | |
057f3fad | 1467 | rcu_read_unlock(); |
d7c01d27 | 1468 | } |
f339b9dc PZ |
1469 | |
1470 | if (wake_flags & WF_MIGRATED) | |
1471 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1472 | ||
d7c01d27 PZ |
1473 | #endif /* CONFIG_SMP */ |
1474 | ||
1475 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1476 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1477 | |
1478 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1479 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1480 | |
d7c01d27 PZ |
1481 | #endif /* CONFIG_SCHEDSTATS */ |
1482 | } | |
1483 | ||
1484 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1485 | { | |
9ed3811a | 1486 | activate_task(rq, p, en_flags); |
da0c1e65 | 1487 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e PZ |
1488 | |
1489 | /* if a worker is waking up, notify workqueue */ | |
1490 | if (p->flags & PF_WQ_WORKER) | |
1491 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1492 | } |
1493 | ||
23f41eeb PZ |
1494 | /* |
1495 | * Mark the task runnable and perform wakeup-preemption. | |
1496 | */ | |
89363381 | 1497 | static void |
23f41eeb | 1498 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1499 | { |
9ed3811a | 1500 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1501 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1502 | |
1503 | p->state = TASK_RUNNING; | |
1504 | #ifdef CONFIG_SMP | |
1505 | if (p->sched_class->task_woken) | |
1506 | p->sched_class->task_woken(rq, p); | |
1507 | ||
e69c6341 | 1508 | if (rq->idle_stamp) { |
78becc27 | 1509 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1510 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1511 | |
abfafa54 JL |
1512 | update_avg(&rq->avg_idle, delta); |
1513 | ||
1514 | if (rq->avg_idle > max) | |
9ed3811a | 1515 | rq->avg_idle = max; |
abfafa54 | 1516 | |
9ed3811a TH |
1517 | rq->idle_stamp = 0; |
1518 | } | |
1519 | #endif | |
1520 | } | |
1521 | ||
c05fbafb PZ |
1522 | static void |
1523 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1524 | { | |
1525 | #ifdef CONFIG_SMP | |
1526 | if (p->sched_contributes_to_load) | |
1527 | rq->nr_uninterruptible--; | |
1528 | #endif | |
1529 | ||
1530 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1531 | ttwu_do_wakeup(rq, p, wake_flags); | |
1532 | } | |
1533 | ||
1534 | /* | |
1535 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1536 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1537 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1538 | * the task is still ->on_rq. | |
1539 | */ | |
1540 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1541 | { | |
1542 | struct rq *rq; | |
1543 | int ret = 0; | |
1544 | ||
1545 | rq = __task_rq_lock(p); | |
da0c1e65 | 1546 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1547 | /* check_preempt_curr() may use rq clock */ |
1548 | update_rq_clock(rq); | |
c05fbafb PZ |
1549 | ttwu_do_wakeup(rq, p, wake_flags); |
1550 | ret = 1; | |
1551 | } | |
1552 | __task_rq_unlock(rq); | |
1553 | ||
1554 | return ret; | |
1555 | } | |
1556 | ||
317f3941 | 1557 | #ifdef CONFIG_SMP |
e3baac47 | 1558 | void sched_ttwu_pending(void) |
317f3941 PZ |
1559 | { |
1560 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1561 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1562 | struct task_struct *p; | |
e3baac47 | 1563 | unsigned long flags; |
317f3941 | 1564 | |
e3baac47 PZ |
1565 | if (!llist) |
1566 | return; | |
1567 | ||
1568 | raw_spin_lock_irqsave(&rq->lock, flags); | |
317f3941 | 1569 | |
fa14ff4a PZ |
1570 | while (llist) { |
1571 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1572 | llist = llist_next(llist); | |
317f3941 PZ |
1573 | ttwu_do_activate(rq, p, 0); |
1574 | } | |
1575 | ||
e3baac47 | 1576 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1577 | } |
1578 | ||
1579 | void scheduler_ipi(void) | |
1580 | { | |
f27dde8d PZ |
1581 | /* |
1582 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1583 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1584 | * this IPI. | |
1585 | */ | |
8cb75e0c | 1586 | preempt_fold_need_resched(); |
f27dde8d | 1587 | |
fd2ac4f4 | 1588 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1589 | return; |
1590 | ||
1591 | /* | |
1592 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1593 | * traditionally all their work was done from the interrupt return | |
1594 | * path. Now that we actually do some work, we need to make sure | |
1595 | * we do call them. | |
1596 | * | |
1597 | * Some archs already do call them, luckily irq_enter/exit nest | |
1598 | * properly. | |
1599 | * | |
1600 | * Arguably we should visit all archs and update all handlers, | |
1601 | * however a fair share of IPIs are still resched only so this would | |
1602 | * somewhat pessimize the simple resched case. | |
1603 | */ | |
1604 | irq_enter(); | |
fa14ff4a | 1605 | sched_ttwu_pending(); |
ca38062e SS |
1606 | |
1607 | /* | |
1608 | * Check if someone kicked us for doing the nohz idle load balance. | |
1609 | */ | |
873b4c65 | 1610 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1611 | this_rq()->idle_balance = 1; |
ca38062e | 1612 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1613 | } |
c5d753a5 | 1614 | irq_exit(); |
317f3941 PZ |
1615 | } |
1616 | ||
1617 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1618 | { | |
e3baac47 PZ |
1619 | struct rq *rq = cpu_rq(cpu); |
1620 | ||
1621 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1622 | if (!set_nr_if_polling(rq->idle)) | |
1623 | smp_send_reschedule(cpu); | |
1624 | else | |
1625 | trace_sched_wake_idle_without_ipi(cpu); | |
1626 | } | |
317f3941 | 1627 | } |
d6aa8f85 | 1628 | |
39be3501 | 1629 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1630 | { |
1631 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1632 | } | |
d6aa8f85 | 1633 | #endif /* CONFIG_SMP */ |
317f3941 | 1634 | |
c05fbafb PZ |
1635 | static void ttwu_queue(struct task_struct *p, int cpu) |
1636 | { | |
1637 | struct rq *rq = cpu_rq(cpu); | |
1638 | ||
17d9f311 | 1639 | #if defined(CONFIG_SMP) |
39be3501 | 1640 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1641 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1642 | ttwu_queue_remote(p, cpu); |
1643 | return; | |
1644 | } | |
1645 | #endif | |
1646 | ||
c05fbafb PZ |
1647 | raw_spin_lock(&rq->lock); |
1648 | ttwu_do_activate(rq, p, 0); | |
1649 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1650 | } |
1651 | ||
1652 | /** | |
1da177e4 | 1653 | * try_to_wake_up - wake up a thread |
9ed3811a | 1654 | * @p: the thread to be awakened |
1da177e4 | 1655 | * @state: the mask of task states that can be woken |
9ed3811a | 1656 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1657 | * |
1658 | * Put it on the run-queue if it's not already there. The "current" | |
1659 | * thread is always on the run-queue (except when the actual | |
1660 | * re-schedule is in progress), and as such you're allowed to do | |
1661 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1662 | * runnable without the overhead of this. | |
1663 | * | |
e69f6186 | 1664 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1665 | * or @state didn't match @p's state. |
1da177e4 | 1666 | */ |
e4a52bcb PZ |
1667 | static int |
1668 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1669 | { |
1da177e4 | 1670 | unsigned long flags; |
c05fbafb | 1671 | int cpu, success = 0; |
2398f2c6 | 1672 | |
e0acd0a6 ON |
1673 | /* |
1674 | * If we are going to wake up a thread waiting for CONDITION we | |
1675 | * need to ensure that CONDITION=1 done by the caller can not be | |
1676 | * reordered with p->state check below. This pairs with mb() in | |
1677 | * set_current_state() the waiting thread does. | |
1678 | */ | |
1679 | smp_mb__before_spinlock(); | |
013fdb80 | 1680 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1681 | if (!(p->state & state)) |
1da177e4 LT |
1682 | goto out; |
1683 | ||
c05fbafb | 1684 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1685 | cpu = task_cpu(p); |
1da177e4 | 1686 | |
cca26e80 | 1687 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
c05fbafb | 1688 | goto stat; |
1da177e4 | 1689 | |
1da177e4 | 1690 | #ifdef CONFIG_SMP |
e9c84311 | 1691 | /* |
c05fbafb PZ |
1692 | * If the owning (remote) cpu is still in the middle of schedule() with |
1693 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1694 | */ |
f3e94786 | 1695 | while (p->on_cpu) |
e4a52bcb | 1696 | cpu_relax(); |
0970d299 | 1697 | /* |
e4a52bcb | 1698 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1699 | */ |
e4a52bcb | 1700 | smp_rmb(); |
1da177e4 | 1701 | |
a8e4f2ea | 1702 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1703 | p->state = TASK_WAKING; |
e7693a36 | 1704 | |
e4a52bcb | 1705 | if (p->sched_class->task_waking) |
74f8e4b2 | 1706 | p->sched_class->task_waking(p); |
efbbd05a | 1707 | |
ac66f547 | 1708 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1709 | if (task_cpu(p) != cpu) { |
1710 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1711 | set_task_cpu(p, cpu); |
f339b9dc | 1712 | } |
1da177e4 | 1713 | #endif /* CONFIG_SMP */ |
1da177e4 | 1714 | |
c05fbafb PZ |
1715 | ttwu_queue(p, cpu); |
1716 | stat: | |
b84cb5df | 1717 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1718 | out: |
013fdb80 | 1719 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1720 | |
1721 | return success; | |
1722 | } | |
1723 | ||
21aa9af0 TH |
1724 | /** |
1725 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1726 | * @p: the thread to be awakened | |
1727 | * | |
2acca55e | 1728 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1729 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1730 | * the current task. |
21aa9af0 TH |
1731 | */ |
1732 | static void try_to_wake_up_local(struct task_struct *p) | |
1733 | { | |
1734 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1735 | |
383efcd0 TH |
1736 | if (WARN_ON_ONCE(rq != this_rq()) || |
1737 | WARN_ON_ONCE(p == current)) | |
1738 | return; | |
1739 | ||
21aa9af0 TH |
1740 | lockdep_assert_held(&rq->lock); |
1741 | ||
2acca55e PZ |
1742 | if (!raw_spin_trylock(&p->pi_lock)) { |
1743 | raw_spin_unlock(&rq->lock); | |
1744 | raw_spin_lock(&p->pi_lock); | |
1745 | raw_spin_lock(&rq->lock); | |
1746 | } | |
1747 | ||
21aa9af0 | 1748 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1749 | goto out; |
21aa9af0 | 1750 | |
da0c1e65 | 1751 | if (!task_on_rq_queued(p)) |
d7c01d27 PZ |
1752 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1753 | ||
23f41eeb | 1754 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1755 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1756 | out: |
1757 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1758 | } |
1759 | ||
50fa610a DH |
1760 | /** |
1761 | * wake_up_process - Wake up a specific process | |
1762 | * @p: The process to be woken up. | |
1763 | * | |
1764 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1765 | * processes. |
1766 | * | |
1767 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1768 | * |
1769 | * It may be assumed that this function implies a write memory barrier before | |
1770 | * changing the task state if and only if any tasks are woken up. | |
1771 | */ | |
7ad5b3a5 | 1772 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1773 | { |
9067ac85 ON |
1774 | WARN_ON(task_is_stopped_or_traced(p)); |
1775 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1776 | } |
1da177e4 LT |
1777 | EXPORT_SYMBOL(wake_up_process); |
1778 | ||
7ad5b3a5 | 1779 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1780 | { |
1781 | return try_to_wake_up(p, state, 0); | |
1782 | } | |
1783 | ||
1da177e4 LT |
1784 | /* |
1785 | * Perform scheduler related setup for a newly forked process p. | |
1786 | * p is forked by current. | |
dd41f596 IM |
1787 | * |
1788 | * __sched_fork() is basic setup used by init_idle() too: | |
1789 | */ | |
5e1576ed | 1790 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1791 | { |
fd2f4419 PZ |
1792 | p->on_rq = 0; |
1793 | ||
1794 | p->se.on_rq = 0; | |
dd41f596 IM |
1795 | p->se.exec_start = 0; |
1796 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1797 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1798 | p->se.nr_migrations = 0; |
da7a735e | 1799 | p->se.vruntime = 0; |
fd2f4419 | 1800 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1801 | |
1802 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1803 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1804 | #endif |
476d139c | 1805 | |
aab03e05 DF |
1806 | RB_CLEAR_NODE(&p->dl.rb_node); |
1807 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1808 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1809 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1810 | p->dl.dl_period = 0; |
aab03e05 DF |
1811 | p->dl.flags = 0; |
1812 | ||
fa717060 | 1813 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1814 | |
e107be36 AK |
1815 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1816 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1817 | #endif | |
cbee9f88 PZ |
1818 | |
1819 | #ifdef CONFIG_NUMA_BALANCING | |
1820 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1821 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1822 | p->mm->numa_scan_seq = 0; |
1823 | } | |
1824 | ||
5e1576ed RR |
1825 | if (clone_flags & CLONE_VM) |
1826 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1827 | else | |
1828 | p->numa_preferred_nid = -1; | |
1829 | ||
cbee9f88 PZ |
1830 | p->node_stamp = 0ULL; |
1831 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1832 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1833 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1834 | p->numa_faults_memory = NULL; |
1835 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1836 | p->last_task_numa_placement = 0; |
1837 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1838 | |
1839 | INIT_LIST_HEAD(&p->numa_entry); | |
1840 | p->numa_group = NULL; | |
cbee9f88 | 1841 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1842 | } |
1843 | ||
1a687c2e | 1844 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1845 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1846 | void set_numabalancing_state(bool enabled) |
1847 | { | |
1848 | if (enabled) | |
1849 | sched_feat_set("NUMA"); | |
1850 | else | |
1851 | sched_feat_set("NO_NUMA"); | |
1852 | } | |
3105b86a MG |
1853 | #else |
1854 | __read_mostly bool numabalancing_enabled; | |
1855 | ||
1856 | void set_numabalancing_state(bool enabled) | |
1857 | { | |
1858 | numabalancing_enabled = enabled; | |
dd41f596 | 1859 | } |
3105b86a | 1860 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1861 | |
1862 | #ifdef CONFIG_PROC_SYSCTL | |
1863 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1864 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1865 | { | |
1866 | struct ctl_table t; | |
1867 | int err; | |
1868 | int state = numabalancing_enabled; | |
1869 | ||
1870 | if (write && !capable(CAP_SYS_ADMIN)) | |
1871 | return -EPERM; | |
1872 | ||
1873 | t = *table; | |
1874 | t.data = &state; | |
1875 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1876 | if (err < 0) | |
1877 | return err; | |
1878 | if (write) | |
1879 | set_numabalancing_state(state); | |
1880 | return err; | |
1881 | } | |
1882 | #endif | |
1883 | #endif | |
dd41f596 IM |
1884 | |
1885 | /* | |
1886 | * fork()/clone()-time setup: | |
1887 | */ | |
aab03e05 | 1888 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1889 | { |
0122ec5b | 1890 | unsigned long flags; |
dd41f596 IM |
1891 | int cpu = get_cpu(); |
1892 | ||
5e1576ed | 1893 | __sched_fork(clone_flags, p); |
06b83b5f | 1894 | /* |
0017d735 | 1895 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1896 | * nobody will actually run it, and a signal or other external |
1897 | * event cannot wake it up and insert it on the runqueue either. | |
1898 | */ | |
0017d735 | 1899 | p->state = TASK_RUNNING; |
dd41f596 | 1900 | |
c350a04e MG |
1901 | /* |
1902 | * Make sure we do not leak PI boosting priority to the child. | |
1903 | */ | |
1904 | p->prio = current->normal_prio; | |
1905 | ||
b9dc29e7 MG |
1906 | /* |
1907 | * Revert to default priority/policy on fork if requested. | |
1908 | */ | |
1909 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1910 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1911 | p->policy = SCHED_NORMAL; |
6c697bdf | 1912 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1913 | p->rt_priority = 0; |
1914 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1915 | p->static_prio = NICE_TO_PRIO(0); | |
1916 | ||
1917 | p->prio = p->normal_prio = __normal_prio(p); | |
1918 | set_load_weight(p); | |
6c697bdf | 1919 | |
b9dc29e7 MG |
1920 | /* |
1921 | * We don't need the reset flag anymore after the fork. It has | |
1922 | * fulfilled its duty: | |
1923 | */ | |
1924 | p->sched_reset_on_fork = 0; | |
1925 | } | |
ca94c442 | 1926 | |
aab03e05 DF |
1927 | if (dl_prio(p->prio)) { |
1928 | put_cpu(); | |
1929 | return -EAGAIN; | |
1930 | } else if (rt_prio(p->prio)) { | |
1931 | p->sched_class = &rt_sched_class; | |
1932 | } else { | |
2ddbf952 | 1933 | p->sched_class = &fair_sched_class; |
aab03e05 | 1934 | } |
b29739f9 | 1935 | |
cd29fe6f PZ |
1936 | if (p->sched_class->task_fork) |
1937 | p->sched_class->task_fork(p); | |
1938 | ||
86951599 PZ |
1939 | /* |
1940 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1941 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1942 | * is ran before sched_fork(). | |
1943 | * | |
1944 | * Silence PROVE_RCU. | |
1945 | */ | |
0122ec5b | 1946 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1947 | set_task_cpu(p, cpu); |
0122ec5b | 1948 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1949 | |
52f17b6c | 1950 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1951 | if (likely(sched_info_on())) |
52f17b6c | 1952 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1953 | #endif |
3ca7a440 PZ |
1954 | #if defined(CONFIG_SMP) |
1955 | p->on_cpu = 0; | |
4866cde0 | 1956 | #endif |
01028747 | 1957 | init_task_preempt_count(p); |
806c09a7 | 1958 | #ifdef CONFIG_SMP |
917b627d | 1959 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1960 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1961 | #endif |
917b627d | 1962 | |
476d139c | 1963 | put_cpu(); |
aab03e05 | 1964 | return 0; |
1da177e4 LT |
1965 | } |
1966 | ||
332ac17e DF |
1967 | unsigned long to_ratio(u64 period, u64 runtime) |
1968 | { | |
1969 | if (runtime == RUNTIME_INF) | |
1970 | return 1ULL << 20; | |
1971 | ||
1972 | /* | |
1973 | * Doing this here saves a lot of checks in all | |
1974 | * the calling paths, and returning zero seems | |
1975 | * safe for them anyway. | |
1976 | */ | |
1977 | if (period == 0) | |
1978 | return 0; | |
1979 | ||
1980 | return div64_u64(runtime << 20, period); | |
1981 | } | |
1982 | ||
1983 | #ifdef CONFIG_SMP | |
1984 | inline struct dl_bw *dl_bw_of(int i) | |
1985 | { | |
1986 | return &cpu_rq(i)->rd->dl_bw; | |
1987 | } | |
1988 | ||
de212f18 | 1989 | static inline int dl_bw_cpus(int i) |
332ac17e | 1990 | { |
de212f18 PZ |
1991 | struct root_domain *rd = cpu_rq(i)->rd; |
1992 | int cpus = 0; | |
1993 | ||
1994 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1995 | cpus++; | |
1996 | ||
1997 | return cpus; | |
332ac17e DF |
1998 | } |
1999 | #else | |
2000 | inline struct dl_bw *dl_bw_of(int i) | |
2001 | { | |
2002 | return &cpu_rq(i)->dl.dl_bw; | |
2003 | } | |
2004 | ||
de212f18 | 2005 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2006 | { |
2007 | return 1; | |
2008 | } | |
2009 | #endif | |
2010 | ||
2011 | static inline | |
2012 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
2013 | { | |
2014 | dl_b->total_bw -= tsk_bw; | |
2015 | } | |
2016 | ||
2017 | static inline | |
2018 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
2019 | { | |
2020 | dl_b->total_bw += tsk_bw; | |
2021 | } | |
2022 | ||
2023 | static inline | |
2024 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
2025 | { | |
2026 | return dl_b->bw != -1 && | |
2027 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * We must be sure that accepting a new task (or allowing changing the | |
2032 | * parameters of an existing one) is consistent with the bandwidth | |
2033 | * constraints. If yes, this function also accordingly updates the currently | |
2034 | * allocated bandwidth to reflect the new situation. | |
2035 | * | |
2036 | * This function is called while holding p's rq->lock. | |
2037 | */ | |
2038 | static int dl_overflow(struct task_struct *p, int policy, | |
2039 | const struct sched_attr *attr) | |
2040 | { | |
2041 | ||
2042 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2043 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2044 | u64 runtime = attr->sched_runtime; |
2045 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2046 | int cpus, err = -1; |
332ac17e DF |
2047 | |
2048 | if (new_bw == p->dl.dl_bw) | |
2049 | return 0; | |
2050 | ||
2051 | /* | |
2052 | * Either if a task, enters, leave, or stays -deadline but changes | |
2053 | * its parameters, we may need to update accordingly the total | |
2054 | * allocated bandwidth of the container. | |
2055 | */ | |
2056 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2057 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2058 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2059 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2060 | __dl_add(dl_b, new_bw); | |
2061 | err = 0; | |
2062 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2063 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2064 | __dl_clear(dl_b, p->dl.dl_bw); | |
2065 | __dl_add(dl_b, new_bw); | |
2066 | err = 0; | |
2067 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2068 | __dl_clear(dl_b, p->dl.dl_bw); | |
2069 | err = 0; | |
2070 | } | |
2071 | raw_spin_unlock(&dl_b->lock); | |
2072 | ||
2073 | return err; | |
2074 | } | |
2075 | ||
2076 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2077 | ||
1da177e4 LT |
2078 | /* |
2079 | * wake_up_new_task - wake up a newly created task for the first time. | |
2080 | * | |
2081 | * This function will do some initial scheduler statistics housekeeping | |
2082 | * that must be done for every newly created context, then puts the task | |
2083 | * on the runqueue and wakes it. | |
2084 | */ | |
3e51e3ed | 2085 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2086 | { |
2087 | unsigned long flags; | |
dd41f596 | 2088 | struct rq *rq; |
fabf318e | 2089 | |
ab2515c4 | 2090 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2091 | #ifdef CONFIG_SMP |
2092 | /* | |
2093 | * Fork balancing, do it here and not earlier because: | |
2094 | * - cpus_allowed can change in the fork path | |
2095 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2096 | */ |
ac66f547 | 2097 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2098 | #endif |
2099 | ||
a75cdaa9 AS |
2100 | /* Initialize new task's runnable average */ |
2101 | init_task_runnable_average(p); | |
ab2515c4 | 2102 | rq = __task_rq_lock(p); |
cd29fe6f | 2103 | activate_task(rq, p, 0); |
da0c1e65 | 2104 | p->on_rq = TASK_ON_RQ_QUEUED; |
89363381 | 2105 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2106 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2107 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2108 | if (p->sched_class->task_woken) |
2109 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2110 | #endif |
0122ec5b | 2111 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2112 | } |
2113 | ||
e107be36 AK |
2114 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2115 | ||
2116 | /** | |
80dd99b3 | 2117 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2118 | * @notifier: notifier struct to register |
e107be36 AK |
2119 | */ |
2120 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2121 | { | |
2122 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2123 | } | |
2124 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2125 | ||
2126 | /** | |
2127 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2128 | * @notifier: notifier struct to unregister |
e107be36 AK |
2129 | * |
2130 | * This is safe to call from within a preemption notifier. | |
2131 | */ | |
2132 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2133 | { | |
2134 | hlist_del(¬ifier->link); | |
2135 | } | |
2136 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2137 | ||
2138 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2139 | { | |
2140 | struct preempt_notifier *notifier; | |
e107be36 | 2141 | |
b67bfe0d | 2142 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2143 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2144 | } | |
2145 | ||
2146 | static void | |
2147 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2148 | struct task_struct *next) | |
2149 | { | |
2150 | struct preempt_notifier *notifier; | |
e107be36 | 2151 | |
b67bfe0d | 2152 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2153 | notifier->ops->sched_out(notifier, next); |
2154 | } | |
2155 | ||
6d6bc0ad | 2156 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2157 | |
2158 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2159 | { | |
2160 | } | |
2161 | ||
2162 | static void | |
2163 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2164 | struct task_struct *next) | |
2165 | { | |
2166 | } | |
2167 | ||
6d6bc0ad | 2168 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2169 | |
4866cde0 NP |
2170 | /** |
2171 | * prepare_task_switch - prepare to switch tasks | |
2172 | * @rq: the runqueue preparing to switch | |
421cee29 | 2173 | * @prev: the current task that is being switched out |
4866cde0 NP |
2174 | * @next: the task we are going to switch to. |
2175 | * | |
2176 | * This is called with the rq lock held and interrupts off. It must | |
2177 | * be paired with a subsequent finish_task_switch after the context | |
2178 | * switch. | |
2179 | * | |
2180 | * prepare_task_switch sets up locking and calls architecture specific | |
2181 | * hooks. | |
2182 | */ | |
e107be36 AK |
2183 | static inline void |
2184 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2185 | struct task_struct *next) | |
4866cde0 | 2186 | { |
895dd92c | 2187 | trace_sched_switch(prev, next); |
43148951 | 2188 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2189 | perf_event_task_sched_out(prev, next); |
e107be36 | 2190 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2191 | prepare_lock_switch(rq, next); |
2192 | prepare_arch_switch(next); | |
2193 | } | |
2194 | ||
1da177e4 LT |
2195 | /** |
2196 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2197 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2198 | * @prev: the thread we just switched away from. |
2199 | * | |
4866cde0 NP |
2200 | * finish_task_switch must be called after the context switch, paired |
2201 | * with a prepare_task_switch call before the context switch. | |
2202 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2203 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2204 | * |
2205 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2206 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2207 | * with the lock held can cause deadlocks; see schedule() for |
2208 | * details.) | |
2209 | */ | |
a9957449 | 2210 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2211 | __releases(rq->lock) |
2212 | { | |
1da177e4 | 2213 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2214 | long prev_state; |
1da177e4 LT |
2215 | |
2216 | rq->prev_mm = NULL; | |
2217 | ||
2218 | /* | |
2219 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2220 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2221 | * schedule one last time. The schedule call will never return, and |
2222 | * the scheduled task must drop that reference. | |
c394cc9f | 2223 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2224 | * still held, otherwise prev could be scheduled on another cpu, die |
2225 | * there before we look at prev->state, and then the reference would | |
2226 | * be dropped twice. | |
2227 | * Manfred Spraul <manfred@colorfullife.com> | |
2228 | */ | |
55a101f8 | 2229 | prev_state = prev->state; |
bf9fae9f | 2230 | vtime_task_switch(prev); |
4866cde0 | 2231 | finish_arch_switch(prev); |
a8d757ef | 2232 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2233 | finish_lock_switch(rq, prev); |
01f23e16 | 2234 | finish_arch_post_lock_switch(); |
e8fa1362 | 2235 | |
e107be36 | 2236 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2237 | if (mm) |
2238 | mmdrop(mm); | |
c394cc9f | 2239 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2240 | if (prev->sched_class->task_dead) |
2241 | prev->sched_class->task_dead(prev); | |
2242 | ||
c6fd91f0 | 2243 | /* |
2244 | * Remove function-return probe instances associated with this | |
2245 | * task and put them back on the free list. | |
9761eea8 | 2246 | */ |
c6fd91f0 | 2247 | kprobe_flush_task(prev); |
1da177e4 | 2248 | put_task_struct(prev); |
c6fd91f0 | 2249 | } |
99e5ada9 FW |
2250 | |
2251 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2252 | } |
2253 | ||
3f029d3c GH |
2254 | #ifdef CONFIG_SMP |
2255 | ||
3f029d3c GH |
2256 | /* rq->lock is NOT held, but preemption is disabled */ |
2257 | static inline void post_schedule(struct rq *rq) | |
2258 | { | |
2259 | if (rq->post_schedule) { | |
2260 | unsigned long flags; | |
2261 | ||
05fa785c | 2262 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2263 | if (rq->curr->sched_class->post_schedule) |
2264 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2265 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2266 | |
2267 | rq->post_schedule = 0; | |
2268 | } | |
2269 | } | |
2270 | ||
2271 | #else | |
da19ab51 | 2272 | |
3f029d3c GH |
2273 | static inline void post_schedule(struct rq *rq) |
2274 | { | |
1da177e4 LT |
2275 | } |
2276 | ||
3f029d3c GH |
2277 | #endif |
2278 | ||
1da177e4 LT |
2279 | /** |
2280 | * schedule_tail - first thing a freshly forked thread must call. | |
2281 | * @prev: the thread we just switched away from. | |
2282 | */ | |
722a9f92 | 2283 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2284 | __releases(rq->lock) |
2285 | { | |
70b97a7f IM |
2286 | struct rq *rq = this_rq(); |
2287 | ||
4866cde0 | 2288 | finish_task_switch(rq, prev); |
da19ab51 | 2289 | |
3f029d3c GH |
2290 | /* |
2291 | * FIXME: do we need to worry about rq being invalidated by the | |
2292 | * task_switch? | |
2293 | */ | |
2294 | post_schedule(rq); | |
70b97a7f | 2295 | |
4866cde0 NP |
2296 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2297 | /* In this case, finish_task_switch does not reenable preemption */ | |
2298 | preempt_enable(); | |
2299 | #endif | |
1da177e4 | 2300 | if (current->set_child_tid) |
b488893a | 2301 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2302 | } |
2303 | ||
2304 | /* | |
2305 | * context_switch - switch to the new MM and the new | |
2306 | * thread's register state. | |
2307 | */ | |
dd41f596 | 2308 | static inline void |
70b97a7f | 2309 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2310 | struct task_struct *next) |
1da177e4 | 2311 | { |
dd41f596 | 2312 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2313 | |
e107be36 | 2314 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2315 | |
dd41f596 IM |
2316 | mm = next->mm; |
2317 | oldmm = prev->active_mm; | |
9226d125 ZA |
2318 | /* |
2319 | * For paravirt, this is coupled with an exit in switch_to to | |
2320 | * combine the page table reload and the switch backend into | |
2321 | * one hypercall. | |
2322 | */ | |
224101ed | 2323 | arch_start_context_switch(prev); |
9226d125 | 2324 | |
31915ab4 | 2325 | if (!mm) { |
1da177e4 LT |
2326 | next->active_mm = oldmm; |
2327 | atomic_inc(&oldmm->mm_count); | |
2328 | enter_lazy_tlb(oldmm, next); | |
2329 | } else | |
2330 | switch_mm(oldmm, mm, next); | |
2331 | ||
31915ab4 | 2332 | if (!prev->mm) { |
1da177e4 | 2333 | prev->active_mm = NULL; |
1da177e4 LT |
2334 | rq->prev_mm = oldmm; |
2335 | } | |
3a5f5e48 IM |
2336 | /* |
2337 | * Since the runqueue lock will be released by the next | |
2338 | * task (which is an invalid locking op but in the case | |
2339 | * of the scheduler it's an obvious special-case), so we | |
2340 | * do an early lockdep release here: | |
2341 | */ | |
2342 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2343 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2344 | #endif |
1da177e4 | 2345 | |
91d1aa43 | 2346 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2347 | /* Here we just switch the register state and the stack. */ |
2348 | switch_to(prev, next, prev); | |
2349 | ||
dd41f596 IM |
2350 | barrier(); |
2351 | /* | |
2352 | * this_rq must be evaluated again because prev may have moved | |
2353 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2354 | * frame will be invalid. | |
2355 | */ | |
2356 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2357 | } |
2358 | ||
2359 | /* | |
1c3e8264 | 2360 | * nr_running and nr_context_switches: |
1da177e4 LT |
2361 | * |
2362 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2363 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2364 | */ |
2365 | unsigned long nr_running(void) | |
2366 | { | |
2367 | unsigned long i, sum = 0; | |
2368 | ||
2369 | for_each_online_cpu(i) | |
2370 | sum += cpu_rq(i)->nr_running; | |
2371 | ||
2372 | return sum; | |
f711f609 | 2373 | } |
1da177e4 | 2374 | |
1da177e4 | 2375 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2376 | { |
cc94abfc SR |
2377 | int i; |
2378 | unsigned long long sum = 0; | |
46cb4b7c | 2379 | |
0a945022 | 2380 | for_each_possible_cpu(i) |
1da177e4 | 2381 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2382 | |
1da177e4 LT |
2383 | return sum; |
2384 | } | |
483b4ee6 | 2385 | |
1da177e4 LT |
2386 | unsigned long nr_iowait(void) |
2387 | { | |
2388 | unsigned long i, sum = 0; | |
483b4ee6 | 2389 | |
0a945022 | 2390 | for_each_possible_cpu(i) |
1da177e4 | 2391 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2392 | |
1da177e4 LT |
2393 | return sum; |
2394 | } | |
483b4ee6 | 2395 | |
8c215bd3 | 2396 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2397 | { |
8c215bd3 | 2398 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2399 | return atomic_read(&this->nr_iowait); |
2400 | } | |
46cb4b7c | 2401 | |
dd41f596 | 2402 | #ifdef CONFIG_SMP |
8a0be9ef | 2403 | |
46cb4b7c | 2404 | /* |
38022906 PZ |
2405 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2406 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2407 | */ |
38022906 | 2408 | void sched_exec(void) |
46cb4b7c | 2409 | { |
38022906 | 2410 | struct task_struct *p = current; |
1da177e4 | 2411 | unsigned long flags; |
0017d735 | 2412 | int dest_cpu; |
46cb4b7c | 2413 | |
8f42ced9 | 2414 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2415 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2416 | if (dest_cpu == smp_processor_id()) |
2417 | goto unlock; | |
38022906 | 2418 | |
8f42ced9 | 2419 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2420 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2421 | |
8f42ced9 PZ |
2422 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2423 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2424 | return; |
2425 | } | |
0017d735 | 2426 | unlock: |
8f42ced9 | 2427 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2428 | } |
dd41f596 | 2429 | |
1da177e4 LT |
2430 | #endif |
2431 | ||
1da177e4 | 2432 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2433 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2434 | |
2435 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2436 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2437 | |
2438 | /* | |
c5f8d995 | 2439 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2440 | * @p in case that task is currently running. |
c5f8d995 HS |
2441 | * |
2442 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2443 | */ |
c5f8d995 HS |
2444 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2445 | { | |
2446 | u64 ns = 0; | |
2447 | ||
4036ac15 MG |
2448 | /* |
2449 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2450 | * project cycles that may never be accounted to this | |
2451 | * thread, breaking clock_gettime(). | |
2452 | */ | |
da0c1e65 | 2453 | if (task_current(rq, p) && task_on_rq_queued(p)) { |
c5f8d995 | 2454 | update_rq_clock(rq); |
78becc27 | 2455 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2456 | if ((s64)ns < 0) |
2457 | ns = 0; | |
2458 | } | |
2459 | ||
2460 | return ns; | |
2461 | } | |
2462 | ||
bb34d92f | 2463 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2464 | { |
1da177e4 | 2465 | unsigned long flags; |
41b86e9c | 2466 | struct rq *rq; |
bb34d92f | 2467 | u64 ns = 0; |
48f24c4d | 2468 | |
41b86e9c | 2469 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2470 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2471 | task_rq_unlock(rq, p, &flags); |
1508487e | 2472 | |
c5f8d995 HS |
2473 | return ns; |
2474 | } | |
f06febc9 | 2475 | |
c5f8d995 HS |
2476 | /* |
2477 | * Return accounted runtime for the task. | |
2478 | * In case the task is currently running, return the runtime plus current's | |
2479 | * pending runtime that have not been accounted yet. | |
2480 | */ | |
2481 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2482 | { | |
2483 | unsigned long flags; | |
2484 | struct rq *rq; | |
2485 | u64 ns = 0; | |
2486 | ||
911b2898 PZ |
2487 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2488 | /* | |
2489 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2490 | * So we have a optimization chance when the task's delta_exec is 0. | |
2491 | * Reading ->on_cpu is racy, but this is ok. | |
2492 | * | |
2493 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2494 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2495 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2496 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2497 | * been accounted, so we're correct here as well. | |
911b2898 | 2498 | */ |
da0c1e65 | 2499 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
2500 | return p->se.sum_exec_runtime; |
2501 | #endif | |
2502 | ||
c5f8d995 HS |
2503 | rq = task_rq_lock(p, &flags); |
2504 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2505 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2506 | |
2507 | return ns; | |
2508 | } | |
48f24c4d | 2509 | |
7835b98b CL |
2510 | /* |
2511 | * This function gets called by the timer code, with HZ frequency. | |
2512 | * We call it with interrupts disabled. | |
7835b98b CL |
2513 | */ |
2514 | void scheduler_tick(void) | |
2515 | { | |
7835b98b CL |
2516 | int cpu = smp_processor_id(); |
2517 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2518 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2519 | |
2520 | sched_clock_tick(); | |
dd41f596 | 2521 | |
05fa785c | 2522 | raw_spin_lock(&rq->lock); |
3e51f33f | 2523 | update_rq_clock(rq); |
fa85ae24 | 2524 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2525 | update_cpu_load_active(rq); |
05fa785c | 2526 | raw_spin_unlock(&rq->lock); |
7835b98b | 2527 | |
e9d2b064 | 2528 | perf_event_task_tick(); |
e220d2dc | 2529 | |
e418e1c2 | 2530 | #ifdef CONFIG_SMP |
6eb57e0d | 2531 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2532 | trigger_load_balance(rq); |
e418e1c2 | 2533 | #endif |
265f22a9 | 2534 | rq_last_tick_reset(rq); |
1da177e4 LT |
2535 | } |
2536 | ||
265f22a9 FW |
2537 | #ifdef CONFIG_NO_HZ_FULL |
2538 | /** | |
2539 | * scheduler_tick_max_deferment | |
2540 | * | |
2541 | * Keep at least one tick per second when a single | |
2542 | * active task is running because the scheduler doesn't | |
2543 | * yet completely support full dynticks environment. | |
2544 | * | |
2545 | * This makes sure that uptime, CFS vruntime, load | |
2546 | * balancing, etc... continue to move forward, even | |
2547 | * with a very low granularity. | |
e69f6186 YB |
2548 | * |
2549 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2550 | */ |
2551 | u64 scheduler_tick_max_deferment(void) | |
2552 | { | |
2553 | struct rq *rq = this_rq(); | |
2554 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2555 | ||
2556 | next = rq->last_sched_tick + HZ; | |
2557 | ||
2558 | if (time_before_eq(next, now)) | |
2559 | return 0; | |
2560 | ||
8fe8ff09 | 2561 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2562 | } |
265f22a9 | 2563 | #endif |
1da177e4 | 2564 | |
132380a0 | 2565 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2566 | { |
2567 | if (in_lock_functions(addr)) { | |
2568 | addr = CALLER_ADDR2; | |
2569 | if (in_lock_functions(addr)) | |
2570 | addr = CALLER_ADDR3; | |
2571 | } | |
2572 | return addr; | |
2573 | } | |
1da177e4 | 2574 | |
7e49fcce SR |
2575 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2576 | defined(CONFIG_PREEMPT_TRACER)) | |
2577 | ||
edafe3a5 | 2578 | void preempt_count_add(int val) |
1da177e4 | 2579 | { |
6cd8a4bb | 2580 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2581 | /* |
2582 | * Underflow? | |
2583 | */ | |
9a11b49a IM |
2584 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2585 | return; | |
6cd8a4bb | 2586 | #endif |
bdb43806 | 2587 | __preempt_count_add(val); |
6cd8a4bb | 2588 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2589 | /* |
2590 | * Spinlock count overflowing soon? | |
2591 | */ | |
33859f7f MOS |
2592 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2593 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2594 | #endif |
8f47b187 TG |
2595 | if (preempt_count() == val) { |
2596 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2597 | #ifdef CONFIG_DEBUG_PREEMPT | |
2598 | current->preempt_disable_ip = ip; | |
2599 | #endif | |
2600 | trace_preempt_off(CALLER_ADDR0, ip); | |
2601 | } | |
1da177e4 | 2602 | } |
bdb43806 | 2603 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2604 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2605 | |
edafe3a5 | 2606 | void preempt_count_sub(int val) |
1da177e4 | 2607 | { |
6cd8a4bb | 2608 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2609 | /* |
2610 | * Underflow? | |
2611 | */ | |
01e3eb82 | 2612 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2613 | return; |
1da177e4 LT |
2614 | /* |
2615 | * Is the spinlock portion underflowing? | |
2616 | */ | |
9a11b49a IM |
2617 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2618 | !(preempt_count() & PREEMPT_MASK))) | |
2619 | return; | |
6cd8a4bb | 2620 | #endif |
9a11b49a | 2621 | |
6cd8a4bb SR |
2622 | if (preempt_count() == val) |
2623 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2624 | __preempt_count_sub(val); |
1da177e4 | 2625 | } |
bdb43806 | 2626 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2627 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2628 | |
2629 | #endif | |
2630 | ||
2631 | /* | |
dd41f596 | 2632 | * Print scheduling while atomic bug: |
1da177e4 | 2633 | */ |
dd41f596 | 2634 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2635 | { |
664dfa65 DJ |
2636 | if (oops_in_progress) |
2637 | return; | |
2638 | ||
3df0fc5b PZ |
2639 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2640 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2641 | |
dd41f596 | 2642 | debug_show_held_locks(prev); |
e21f5b15 | 2643 | print_modules(); |
dd41f596 IM |
2644 | if (irqs_disabled()) |
2645 | print_irqtrace_events(prev); | |
8f47b187 TG |
2646 | #ifdef CONFIG_DEBUG_PREEMPT |
2647 | if (in_atomic_preempt_off()) { | |
2648 | pr_err("Preemption disabled at:"); | |
2649 | print_ip_sym(current->preempt_disable_ip); | |
2650 | pr_cont("\n"); | |
2651 | } | |
2652 | #endif | |
6135fc1e | 2653 | dump_stack(); |
373d4d09 | 2654 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2655 | } |
1da177e4 | 2656 | |
dd41f596 IM |
2657 | /* |
2658 | * Various schedule()-time debugging checks and statistics: | |
2659 | */ | |
2660 | static inline void schedule_debug(struct task_struct *prev) | |
2661 | { | |
1da177e4 | 2662 | /* |
41a2d6cf | 2663 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2664 | * schedule() atomically, we ignore that path. Otherwise whine |
2665 | * if we are scheduling when we should not. | |
1da177e4 | 2666 | */ |
192301e7 | 2667 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2668 | __schedule_bug(prev); |
b3fbab05 | 2669 | rcu_sleep_check(); |
dd41f596 | 2670 | |
1da177e4 LT |
2671 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2672 | ||
2d72376b | 2673 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2674 | } |
2675 | ||
2676 | /* | |
2677 | * Pick up the highest-prio task: | |
2678 | */ | |
2679 | static inline struct task_struct * | |
606dba2e | 2680 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2681 | { |
37e117c0 | 2682 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2683 | struct task_struct *p; |
1da177e4 LT |
2684 | |
2685 | /* | |
dd41f596 IM |
2686 | * Optimization: we know that if all tasks are in |
2687 | * the fair class we can call that function directly: | |
1da177e4 | 2688 | */ |
37e117c0 | 2689 | if (likely(prev->sched_class == class && |
38033c37 | 2690 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2691 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2692 | if (unlikely(p == RETRY_TASK)) |
2693 | goto again; | |
2694 | ||
2695 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2696 | if (unlikely(!p)) | |
2697 | p = idle_sched_class.pick_next_task(rq, prev); | |
2698 | ||
2699 | return p; | |
1da177e4 LT |
2700 | } |
2701 | ||
37e117c0 | 2702 | again: |
34f971f6 | 2703 | for_each_class(class) { |
606dba2e | 2704 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2705 | if (p) { |
2706 | if (unlikely(p == RETRY_TASK)) | |
2707 | goto again; | |
dd41f596 | 2708 | return p; |
37e117c0 | 2709 | } |
dd41f596 | 2710 | } |
34f971f6 PZ |
2711 | |
2712 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2713 | } |
1da177e4 | 2714 | |
dd41f596 | 2715 | /* |
c259e01a | 2716 | * __schedule() is the main scheduler function. |
edde96ea PE |
2717 | * |
2718 | * The main means of driving the scheduler and thus entering this function are: | |
2719 | * | |
2720 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2721 | * | |
2722 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2723 | * paths. For example, see arch/x86/entry_64.S. | |
2724 | * | |
2725 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2726 | * interrupt handler scheduler_tick(). | |
2727 | * | |
2728 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2729 | * task to the run-queue and that's it. | |
2730 | * | |
2731 | * Now, if the new task added to the run-queue preempts the current | |
2732 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2733 | * called on the nearest possible occasion: | |
2734 | * | |
2735 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2736 | * | |
2737 | * - in syscall or exception context, at the next outmost | |
2738 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2739 | * spin_unlock()!) | |
2740 | * | |
2741 | * - in IRQ context, return from interrupt-handler to | |
2742 | * preemptible context | |
2743 | * | |
2744 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2745 | * then at the next: | |
2746 | * | |
2747 | * - cond_resched() call | |
2748 | * - explicit schedule() call | |
2749 | * - return from syscall or exception to user-space | |
2750 | * - return from interrupt-handler to user-space | |
dd41f596 | 2751 | */ |
c259e01a | 2752 | static void __sched __schedule(void) |
dd41f596 IM |
2753 | { |
2754 | struct task_struct *prev, *next; | |
67ca7bde | 2755 | unsigned long *switch_count; |
dd41f596 | 2756 | struct rq *rq; |
31656519 | 2757 | int cpu; |
dd41f596 | 2758 | |
ff743345 PZ |
2759 | need_resched: |
2760 | preempt_disable(); | |
dd41f596 IM |
2761 | cpu = smp_processor_id(); |
2762 | rq = cpu_rq(cpu); | |
25502a6c | 2763 | rcu_note_context_switch(cpu); |
dd41f596 | 2764 | prev = rq->curr; |
dd41f596 | 2765 | |
dd41f596 | 2766 | schedule_debug(prev); |
1da177e4 | 2767 | |
31656519 | 2768 | if (sched_feat(HRTICK)) |
f333fdc9 | 2769 | hrtick_clear(rq); |
8f4d37ec | 2770 | |
e0acd0a6 ON |
2771 | /* |
2772 | * Make sure that signal_pending_state()->signal_pending() below | |
2773 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2774 | * done by the caller to avoid the race with signal_wake_up(). | |
2775 | */ | |
2776 | smp_mb__before_spinlock(); | |
05fa785c | 2777 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2778 | |
246d86b5 | 2779 | switch_count = &prev->nivcsw; |
1da177e4 | 2780 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2781 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2782 | prev->state = TASK_RUNNING; |
21aa9af0 | 2783 | } else { |
2acca55e PZ |
2784 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2785 | prev->on_rq = 0; | |
2786 | ||
21aa9af0 | 2787 | /* |
2acca55e PZ |
2788 | * If a worker went to sleep, notify and ask workqueue |
2789 | * whether it wants to wake up a task to maintain | |
2790 | * concurrency. | |
21aa9af0 TH |
2791 | */ |
2792 | if (prev->flags & PF_WQ_WORKER) { | |
2793 | struct task_struct *to_wakeup; | |
2794 | ||
2795 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2796 | if (to_wakeup) | |
2797 | try_to_wake_up_local(to_wakeup); | |
2798 | } | |
21aa9af0 | 2799 | } |
dd41f596 | 2800 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2801 | } |
2802 | ||
da0c1e65 | 2803 | if (task_on_rq_queued(prev) || rq->skip_clock_update < 0) |
606dba2e PZ |
2804 | update_rq_clock(rq); |
2805 | ||
2806 | next = pick_next_task(rq, prev); | |
f26f9aff | 2807 | clear_tsk_need_resched(prev); |
f27dde8d | 2808 | clear_preempt_need_resched(); |
f26f9aff | 2809 | rq->skip_clock_update = 0; |
1da177e4 | 2810 | |
1da177e4 | 2811 | if (likely(prev != next)) { |
1da177e4 LT |
2812 | rq->nr_switches++; |
2813 | rq->curr = next; | |
2814 | ++*switch_count; | |
2815 | ||
dd41f596 | 2816 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2817 | /* |
246d86b5 ON |
2818 | * The context switch have flipped the stack from under us |
2819 | * and restored the local variables which were saved when | |
2820 | * this task called schedule() in the past. prev == current | |
2821 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2822 | */ |
2823 | cpu = smp_processor_id(); | |
2824 | rq = cpu_rq(cpu); | |
1da177e4 | 2825 | } else |
05fa785c | 2826 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2827 | |
3f029d3c | 2828 | post_schedule(rq); |
1da177e4 | 2829 | |
ba74c144 | 2830 | sched_preempt_enable_no_resched(); |
ff743345 | 2831 | if (need_resched()) |
1da177e4 LT |
2832 | goto need_resched; |
2833 | } | |
c259e01a | 2834 | |
9c40cef2 TG |
2835 | static inline void sched_submit_work(struct task_struct *tsk) |
2836 | { | |
3c7d5184 | 2837 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2838 | return; |
2839 | /* | |
2840 | * If we are going to sleep and we have plugged IO queued, | |
2841 | * make sure to submit it to avoid deadlocks. | |
2842 | */ | |
2843 | if (blk_needs_flush_plug(tsk)) | |
2844 | blk_schedule_flush_plug(tsk); | |
2845 | } | |
2846 | ||
722a9f92 | 2847 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2848 | { |
9c40cef2 TG |
2849 | struct task_struct *tsk = current; |
2850 | ||
2851 | sched_submit_work(tsk); | |
c259e01a TG |
2852 | __schedule(); |
2853 | } | |
1da177e4 LT |
2854 | EXPORT_SYMBOL(schedule); |
2855 | ||
91d1aa43 | 2856 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2857 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2858 | { |
2859 | /* | |
2860 | * If we come here after a random call to set_need_resched(), | |
2861 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2862 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2863 | * we find a better solution. | |
2864 | */ | |
91d1aa43 | 2865 | user_exit(); |
20ab65e3 | 2866 | schedule(); |
91d1aa43 | 2867 | user_enter(); |
20ab65e3 FW |
2868 | } |
2869 | #endif | |
2870 | ||
c5491ea7 TG |
2871 | /** |
2872 | * schedule_preempt_disabled - called with preemption disabled | |
2873 | * | |
2874 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2875 | */ | |
2876 | void __sched schedule_preempt_disabled(void) | |
2877 | { | |
ba74c144 | 2878 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2879 | schedule(); |
2880 | preempt_disable(); | |
2881 | } | |
2882 | ||
1da177e4 LT |
2883 | #ifdef CONFIG_PREEMPT |
2884 | /* | |
2ed6e34f | 2885 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2886 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2887 | * occur there and call schedule directly. |
2888 | */ | |
722a9f92 | 2889 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2890 | { |
1da177e4 LT |
2891 | /* |
2892 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2893 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2894 | */ |
fbb00b56 | 2895 | if (likely(!preemptible())) |
1da177e4 LT |
2896 | return; |
2897 | ||
3a5c359a | 2898 | do { |
bdb43806 | 2899 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2900 | __schedule(); |
bdb43806 | 2901 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2902 | |
3a5c359a AK |
2903 | /* |
2904 | * Check again in case we missed a preemption opportunity | |
2905 | * between schedule and now. | |
2906 | */ | |
2907 | barrier(); | |
5ed0cec0 | 2908 | } while (need_resched()); |
1da177e4 | 2909 | } |
376e2424 | 2910 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 2911 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2912 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2913 | |
2914 | /* | |
2ed6e34f | 2915 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2916 | * off of irq context. |
2917 | * Note, that this is called and return with irqs disabled. This will | |
2918 | * protect us against recursive calling from irq. | |
2919 | */ | |
722a9f92 | 2920 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2921 | { |
b22366cd | 2922 | enum ctx_state prev_state; |
6478d880 | 2923 | |
2ed6e34f | 2924 | /* Catch callers which need to be fixed */ |
f27dde8d | 2925 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2926 | |
b22366cd FW |
2927 | prev_state = exception_enter(); |
2928 | ||
3a5c359a | 2929 | do { |
bdb43806 | 2930 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2931 | local_irq_enable(); |
c259e01a | 2932 | __schedule(); |
3a5c359a | 2933 | local_irq_disable(); |
bdb43806 | 2934 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2935 | |
3a5c359a AK |
2936 | /* |
2937 | * Check again in case we missed a preemption opportunity | |
2938 | * between schedule and now. | |
2939 | */ | |
2940 | barrier(); | |
5ed0cec0 | 2941 | } while (need_resched()); |
b22366cd FW |
2942 | |
2943 | exception_exit(prev_state); | |
1da177e4 LT |
2944 | } |
2945 | ||
63859d4f | 2946 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2947 | void *key) |
1da177e4 | 2948 | { |
63859d4f | 2949 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2950 | } |
1da177e4 LT |
2951 | EXPORT_SYMBOL(default_wake_function); |
2952 | ||
b29739f9 IM |
2953 | #ifdef CONFIG_RT_MUTEXES |
2954 | ||
2955 | /* | |
2956 | * rt_mutex_setprio - set the current priority of a task | |
2957 | * @p: task | |
2958 | * @prio: prio value (kernel-internal form) | |
2959 | * | |
2960 | * This function changes the 'effective' priority of a task. It does | |
2961 | * not touch ->normal_prio like __setscheduler(). | |
2962 | * | |
c365c292 TG |
2963 | * Used by the rt_mutex code to implement priority inheritance |
2964 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2965 | */ |
36c8b586 | 2966 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2967 | { |
da0c1e65 | 2968 | int oldprio, queued, running, enqueue_flag = 0; |
70b97a7f | 2969 | struct rq *rq; |
83ab0aa0 | 2970 | const struct sched_class *prev_class; |
b29739f9 | 2971 | |
aab03e05 | 2972 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2973 | |
0122ec5b | 2974 | rq = __task_rq_lock(p); |
b29739f9 | 2975 | |
1c4dd99b TG |
2976 | /* |
2977 | * Idle task boosting is a nono in general. There is one | |
2978 | * exception, when PREEMPT_RT and NOHZ is active: | |
2979 | * | |
2980 | * The idle task calls get_next_timer_interrupt() and holds | |
2981 | * the timer wheel base->lock on the CPU and another CPU wants | |
2982 | * to access the timer (probably to cancel it). We can safely | |
2983 | * ignore the boosting request, as the idle CPU runs this code | |
2984 | * with interrupts disabled and will complete the lock | |
2985 | * protected section without being interrupted. So there is no | |
2986 | * real need to boost. | |
2987 | */ | |
2988 | if (unlikely(p == rq->idle)) { | |
2989 | WARN_ON(p != rq->curr); | |
2990 | WARN_ON(p->pi_blocked_on); | |
2991 | goto out_unlock; | |
2992 | } | |
2993 | ||
a8027073 | 2994 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 2995 | oldprio = p->prio; |
83ab0aa0 | 2996 | prev_class = p->sched_class; |
da0c1e65 | 2997 | queued = task_on_rq_queued(p); |
051a1d1a | 2998 | running = task_current(rq, p); |
da0c1e65 | 2999 | if (queued) |
69be72c1 | 3000 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3001 | if (running) |
3002 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 3003 | |
2d3d891d DF |
3004 | /* |
3005 | * Boosting condition are: | |
3006 | * 1. -rt task is running and holds mutex A | |
3007 | * --> -dl task blocks on mutex A | |
3008 | * | |
3009 | * 2. -dl task is running and holds mutex A | |
3010 | * --> -dl task blocks on mutex A and could preempt the | |
3011 | * running task | |
3012 | */ | |
3013 | if (dl_prio(prio)) { | |
466af29b ON |
3014 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3015 | if (!dl_prio(p->normal_prio) || | |
3016 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d DF |
3017 | p->dl.dl_boosted = 1; |
3018 | p->dl.dl_throttled = 0; | |
3019 | enqueue_flag = ENQUEUE_REPLENISH; | |
3020 | } else | |
3021 | p->dl.dl_boosted = 0; | |
aab03e05 | 3022 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3023 | } else if (rt_prio(prio)) { |
3024 | if (dl_prio(oldprio)) | |
3025 | p->dl.dl_boosted = 0; | |
3026 | if (oldprio < prio) | |
3027 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3028 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3029 | } else { |
3030 | if (dl_prio(oldprio)) | |
3031 | p->dl.dl_boosted = 0; | |
dd41f596 | 3032 | p->sched_class = &fair_sched_class; |
2d3d891d | 3033 | } |
dd41f596 | 3034 | |
b29739f9 IM |
3035 | p->prio = prio; |
3036 | ||
0e1f3483 HS |
3037 | if (running) |
3038 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3039 | if (queued) |
2d3d891d | 3040 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3041 | |
da7a735e | 3042 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3043 | out_unlock: |
0122ec5b | 3044 | __task_rq_unlock(rq); |
b29739f9 | 3045 | } |
b29739f9 | 3046 | #endif |
d50dde5a | 3047 | |
36c8b586 | 3048 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3049 | { |
da0c1e65 | 3050 | int old_prio, delta, queued; |
1da177e4 | 3051 | unsigned long flags; |
70b97a7f | 3052 | struct rq *rq; |
1da177e4 | 3053 | |
75e45d51 | 3054 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3055 | return; |
3056 | /* | |
3057 | * We have to be careful, if called from sys_setpriority(), | |
3058 | * the task might be in the middle of scheduling on another CPU. | |
3059 | */ | |
3060 | rq = task_rq_lock(p, &flags); | |
3061 | /* | |
3062 | * The RT priorities are set via sched_setscheduler(), but we still | |
3063 | * allow the 'normal' nice value to be set - but as expected | |
3064 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3065 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3066 | */ |
aab03e05 | 3067 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3068 | p->static_prio = NICE_TO_PRIO(nice); |
3069 | goto out_unlock; | |
3070 | } | |
da0c1e65 KT |
3071 | queued = task_on_rq_queued(p); |
3072 | if (queued) | |
69be72c1 | 3073 | dequeue_task(rq, p, 0); |
1da177e4 | 3074 | |
1da177e4 | 3075 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3076 | set_load_weight(p); |
b29739f9 IM |
3077 | old_prio = p->prio; |
3078 | p->prio = effective_prio(p); | |
3079 | delta = p->prio - old_prio; | |
1da177e4 | 3080 | |
da0c1e65 | 3081 | if (queued) { |
371fd7e7 | 3082 | enqueue_task(rq, p, 0); |
1da177e4 | 3083 | /* |
d5f9f942 AM |
3084 | * If the task increased its priority or is running and |
3085 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3086 | */ |
d5f9f942 | 3087 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3088 | resched_curr(rq); |
1da177e4 LT |
3089 | } |
3090 | out_unlock: | |
0122ec5b | 3091 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3092 | } |
1da177e4 LT |
3093 | EXPORT_SYMBOL(set_user_nice); |
3094 | ||
e43379f1 MM |
3095 | /* |
3096 | * can_nice - check if a task can reduce its nice value | |
3097 | * @p: task | |
3098 | * @nice: nice value | |
3099 | */ | |
36c8b586 | 3100 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3101 | { |
024f4747 | 3102 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3103 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3104 | |
78d7d407 | 3105 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3106 | capable(CAP_SYS_NICE)); |
3107 | } | |
3108 | ||
1da177e4 LT |
3109 | #ifdef __ARCH_WANT_SYS_NICE |
3110 | ||
3111 | /* | |
3112 | * sys_nice - change the priority of the current process. | |
3113 | * @increment: priority increment | |
3114 | * | |
3115 | * sys_setpriority is a more generic, but much slower function that | |
3116 | * does similar things. | |
3117 | */ | |
5add95d4 | 3118 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3119 | { |
48f24c4d | 3120 | long nice, retval; |
1da177e4 LT |
3121 | |
3122 | /* | |
3123 | * Setpriority might change our priority at the same moment. | |
3124 | * We don't have to worry. Conceptually one call occurs first | |
3125 | * and we have a single winner. | |
3126 | */ | |
a9467fa3 | 3127 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3128 | nice = task_nice(current) + increment; |
1da177e4 | 3129 | |
a9467fa3 | 3130 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3131 | if (increment < 0 && !can_nice(current, nice)) |
3132 | return -EPERM; | |
3133 | ||
1da177e4 LT |
3134 | retval = security_task_setnice(current, nice); |
3135 | if (retval) | |
3136 | return retval; | |
3137 | ||
3138 | set_user_nice(current, nice); | |
3139 | return 0; | |
3140 | } | |
3141 | ||
3142 | #endif | |
3143 | ||
3144 | /** | |
3145 | * task_prio - return the priority value of a given task. | |
3146 | * @p: the task in question. | |
3147 | * | |
e69f6186 | 3148 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3149 | * RT tasks are offset by -200. Normal tasks are centered |
3150 | * around 0, value goes from -16 to +15. | |
3151 | */ | |
36c8b586 | 3152 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3153 | { |
3154 | return p->prio - MAX_RT_PRIO; | |
3155 | } | |
3156 | ||
1da177e4 LT |
3157 | /** |
3158 | * idle_cpu - is a given cpu idle currently? | |
3159 | * @cpu: the processor in question. | |
e69f6186 YB |
3160 | * |
3161 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3162 | */ |
3163 | int idle_cpu(int cpu) | |
3164 | { | |
908a3283 TG |
3165 | struct rq *rq = cpu_rq(cpu); |
3166 | ||
3167 | if (rq->curr != rq->idle) | |
3168 | return 0; | |
3169 | ||
3170 | if (rq->nr_running) | |
3171 | return 0; | |
3172 | ||
3173 | #ifdef CONFIG_SMP | |
3174 | if (!llist_empty(&rq->wake_list)) | |
3175 | return 0; | |
3176 | #endif | |
3177 | ||
3178 | return 1; | |
1da177e4 LT |
3179 | } |
3180 | ||
1da177e4 LT |
3181 | /** |
3182 | * idle_task - return the idle task for a given cpu. | |
3183 | * @cpu: the processor in question. | |
e69f6186 YB |
3184 | * |
3185 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3186 | */ |
36c8b586 | 3187 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3188 | { |
3189 | return cpu_rq(cpu)->idle; | |
3190 | } | |
3191 | ||
3192 | /** | |
3193 | * find_process_by_pid - find a process with a matching PID value. | |
3194 | * @pid: the pid in question. | |
e69f6186 YB |
3195 | * |
3196 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3197 | */ |
a9957449 | 3198 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3199 | { |
228ebcbe | 3200 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3201 | } |
3202 | ||
aab03e05 DF |
3203 | /* |
3204 | * This function initializes the sched_dl_entity of a newly becoming | |
3205 | * SCHED_DEADLINE task. | |
3206 | * | |
3207 | * Only the static values are considered here, the actual runtime and the | |
3208 | * absolute deadline will be properly calculated when the task is enqueued | |
3209 | * for the first time with its new policy. | |
3210 | */ | |
3211 | static void | |
3212 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3213 | { | |
3214 | struct sched_dl_entity *dl_se = &p->dl; | |
3215 | ||
3216 | init_dl_task_timer(dl_se); | |
3217 | dl_se->dl_runtime = attr->sched_runtime; | |
3218 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3219 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3220 | dl_se->flags = attr->sched_flags; |
332ac17e | 3221 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3222 | dl_se->dl_throttled = 0; |
3223 | dl_se->dl_new = 1; | |
5bfd126e | 3224 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3225 | } |
3226 | ||
c13db6b1 SR |
3227 | /* |
3228 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3229 | * it calls know not to change it. | |
3230 | */ | |
3231 | #define SETPARAM_POLICY -1 | |
3232 | ||
c365c292 TG |
3233 | static void __setscheduler_params(struct task_struct *p, |
3234 | const struct sched_attr *attr) | |
1da177e4 | 3235 | { |
d50dde5a DF |
3236 | int policy = attr->sched_policy; |
3237 | ||
c13db6b1 | 3238 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3239 | policy = p->policy; |
3240 | ||
1da177e4 | 3241 | p->policy = policy; |
d50dde5a | 3242 | |
aab03e05 DF |
3243 | if (dl_policy(policy)) |
3244 | __setparam_dl(p, attr); | |
39fd8fd2 | 3245 | else if (fair_policy(policy)) |
d50dde5a DF |
3246 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3247 | ||
39fd8fd2 PZ |
3248 | /* |
3249 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3250 | * !rt_policy. Always setting this ensures that things like | |
3251 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3252 | */ | |
3253 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3254 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3255 | set_load_weight(p); |
3256 | } | |
39fd8fd2 | 3257 | |
c365c292 TG |
3258 | /* Actually do priority change: must hold pi & rq lock. */ |
3259 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3260 | const struct sched_attr *attr) | |
3261 | { | |
3262 | __setscheduler_params(p, attr); | |
d50dde5a | 3263 | |
383afd09 SR |
3264 | /* |
3265 | * If we get here, there was no pi waiters boosting the | |
3266 | * task. It is safe to use the normal prio. | |
3267 | */ | |
3268 | p->prio = normal_prio(p); | |
3269 | ||
aab03e05 DF |
3270 | if (dl_prio(p->prio)) |
3271 | p->sched_class = &dl_sched_class; | |
3272 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3273 | p->sched_class = &rt_sched_class; |
3274 | else | |
3275 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3276 | } |
aab03e05 DF |
3277 | |
3278 | static void | |
3279 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3280 | { | |
3281 | struct sched_dl_entity *dl_se = &p->dl; | |
3282 | ||
3283 | attr->sched_priority = p->rt_priority; | |
3284 | attr->sched_runtime = dl_se->dl_runtime; | |
3285 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3286 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3287 | attr->sched_flags = dl_se->flags; |
3288 | } | |
3289 | ||
3290 | /* | |
3291 | * This function validates the new parameters of a -deadline task. | |
3292 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3293 | * than the runtime, as well as the period of being zero or |
332ac17e | 3294 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3295 | * user parameters are above the internal resolution of 1us (we |
3296 | * check sched_runtime only since it is always the smaller one) and | |
3297 | * below 2^63 ns (we have to check both sched_deadline and | |
3298 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3299 | */ |
3300 | static bool | |
3301 | __checkparam_dl(const struct sched_attr *attr) | |
3302 | { | |
b0827819 JL |
3303 | /* deadline != 0 */ |
3304 | if (attr->sched_deadline == 0) | |
3305 | return false; | |
3306 | ||
3307 | /* | |
3308 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3309 | * that big. | |
3310 | */ | |
3311 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3312 | return false; | |
3313 | ||
3314 | /* | |
3315 | * Since we use the MSB for wrap-around and sign issues, make | |
3316 | * sure it's not set (mind that period can be equal to zero). | |
3317 | */ | |
3318 | if (attr->sched_deadline & (1ULL << 63) || | |
3319 | attr->sched_period & (1ULL << 63)) | |
3320 | return false; | |
3321 | ||
3322 | /* runtime <= deadline <= period (if period != 0) */ | |
3323 | if ((attr->sched_period != 0 && | |
3324 | attr->sched_period < attr->sched_deadline) || | |
3325 | attr->sched_deadline < attr->sched_runtime) | |
3326 | return false; | |
3327 | ||
3328 | return true; | |
aab03e05 DF |
3329 | } |
3330 | ||
c69e8d9c DH |
3331 | /* |
3332 | * check the target process has a UID that matches the current process's | |
3333 | */ | |
3334 | static bool check_same_owner(struct task_struct *p) | |
3335 | { | |
3336 | const struct cred *cred = current_cred(), *pcred; | |
3337 | bool match; | |
3338 | ||
3339 | rcu_read_lock(); | |
3340 | pcred = __task_cred(p); | |
9c806aa0 EB |
3341 | match = (uid_eq(cred->euid, pcred->euid) || |
3342 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3343 | rcu_read_unlock(); |
3344 | return match; | |
3345 | } | |
3346 | ||
d50dde5a DF |
3347 | static int __sched_setscheduler(struct task_struct *p, |
3348 | const struct sched_attr *attr, | |
3349 | bool user) | |
1da177e4 | 3350 | { |
383afd09 SR |
3351 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3352 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 3353 | int retval, oldprio, oldpolicy = -1, queued, running; |
d50dde5a | 3354 | int policy = attr->sched_policy; |
1da177e4 | 3355 | unsigned long flags; |
83ab0aa0 | 3356 | const struct sched_class *prev_class; |
70b97a7f | 3357 | struct rq *rq; |
ca94c442 | 3358 | int reset_on_fork; |
1da177e4 | 3359 | |
66e5393a SR |
3360 | /* may grab non-irq protected spin_locks */ |
3361 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3362 | recheck: |
3363 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3364 | if (policy < 0) { |
3365 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3366 | policy = oldpolicy = p->policy; |
ca94c442 | 3367 | } else { |
7479f3c9 | 3368 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3369 | |
aab03e05 DF |
3370 | if (policy != SCHED_DEADLINE && |
3371 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3372 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3373 | policy != SCHED_IDLE) | |
3374 | return -EINVAL; | |
3375 | } | |
3376 | ||
7479f3c9 PZ |
3377 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3378 | return -EINVAL; | |
3379 | ||
1da177e4 LT |
3380 | /* |
3381 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3382 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3383 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3384 | */ |
0bb040a4 | 3385 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3386 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3387 | return -EINVAL; |
aab03e05 DF |
3388 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3389 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3390 | return -EINVAL; |
3391 | ||
37e4ab3f OC |
3392 | /* |
3393 | * Allow unprivileged RT tasks to decrease priority: | |
3394 | */ | |
961ccddd | 3395 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3396 | if (fair_policy(policy)) { |
d0ea0268 | 3397 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3398 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3399 | return -EPERM; |
3400 | } | |
3401 | ||
e05606d3 | 3402 | if (rt_policy(policy)) { |
a44702e8 ON |
3403 | unsigned long rlim_rtprio = |
3404 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3405 | |
3406 | /* can't set/change the rt policy */ | |
3407 | if (policy != p->policy && !rlim_rtprio) | |
3408 | return -EPERM; | |
3409 | ||
3410 | /* can't increase priority */ | |
d50dde5a DF |
3411 | if (attr->sched_priority > p->rt_priority && |
3412 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3413 | return -EPERM; |
3414 | } | |
c02aa73b | 3415 | |
d44753b8 JL |
3416 | /* |
3417 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3418 | * (safest behavior); in the future we would like to allow | |
3419 | * unprivileged DL tasks to increase their relative deadline | |
3420 | * or reduce their runtime (both ways reducing utilization) | |
3421 | */ | |
3422 | if (dl_policy(policy)) | |
3423 | return -EPERM; | |
3424 | ||
dd41f596 | 3425 | /* |
c02aa73b DH |
3426 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3427 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3428 | */ |
c02aa73b | 3429 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3430 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3431 | return -EPERM; |
3432 | } | |
5fe1d75f | 3433 | |
37e4ab3f | 3434 | /* can't change other user's priorities */ |
c69e8d9c | 3435 | if (!check_same_owner(p)) |
37e4ab3f | 3436 | return -EPERM; |
ca94c442 LP |
3437 | |
3438 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3439 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3440 | return -EPERM; | |
37e4ab3f | 3441 | } |
1da177e4 | 3442 | |
725aad24 | 3443 | if (user) { |
b0ae1981 | 3444 | retval = security_task_setscheduler(p); |
725aad24 JF |
3445 | if (retval) |
3446 | return retval; | |
3447 | } | |
3448 | ||
b29739f9 IM |
3449 | /* |
3450 | * make sure no PI-waiters arrive (or leave) while we are | |
3451 | * changing the priority of the task: | |
0122ec5b | 3452 | * |
25985edc | 3453 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3454 | * runqueue lock must be held. |
3455 | */ | |
0122ec5b | 3456 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3457 | |
34f971f6 PZ |
3458 | /* |
3459 | * Changing the policy of the stop threads its a very bad idea | |
3460 | */ | |
3461 | if (p == rq->stop) { | |
0122ec5b | 3462 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3463 | return -EINVAL; |
3464 | } | |
3465 | ||
a51e9198 | 3466 | /* |
d6b1e911 TG |
3467 | * If not changing anything there's no need to proceed further, |
3468 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3469 | */ |
d50dde5a | 3470 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3471 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3472 | goto change; |
3473 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3474 | goto change; | |
aab03e05 DF |
3475 | if (dl_policy(policy)) |
3476 | goto change; | |
d50dde5a | 3477 | |
d6b1e911 | 3478 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3479 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3480 | return 0; |
3481 | } | |
d50dde5a | 3482 | change: |
a51e9198 | 3483 | |
dc61b1d6 | 3484 | if (user) { |
332ac17e | 3485 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3486 | /* |
3487 | * Do not allow realtime tasks into groups that have no runtime | |
3488 | * assigned. | |
3489 | */ | |
3490 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3491 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3492 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3493 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3494 | return -EPERM; |
3495 | } | |
dc61b1d6 | 3496 | #endif |
332ac17e DF |
3497 | #ifdef CONFIG_SMP |
3498 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3499 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3500 | |
3501 | /* | |
3502 | * Don't allow tasks with an affinity mask smaller than | |
3503 | * the entire root_domain to become SCHED_DEADLINE. We | |
3504 | * will also fail if there's no bandwidth available. | |
3505 | */ | |
e4099a5e PZ |
3506 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3507 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3508 | task_rq_unlock(rq, p, &flags); |
3509 | return -EPERM; | |
3510 | } | |
3511 | } | |
3512 | #endif | |
3513 | } | |
dc61b1d6 | 3514 | |
1da177e4 LT |
3515 | /* recheck policy now with rq lock held */ |
3516 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3517 | policy = oldpolicy = -1; | |
0122ec5b | 3518 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3519 | goto recheck; |
3520 | } | |
332ac17e DF |
3521 | |
3522 | /* | |
3523 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3524 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3525 | * is available. | |
3526 | */ | |
e4099a5e | 3527 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3528 | task_rq_unlock(rq, p, &flags); |
3529 | return -EBUSY; | |
3530 | } | |
3531 | ||
c365c292 TG |
3532 | p->sched_reset_on_fork = reset_on_fork; |
3533 | oldprio = p->prio; | |
3534 | ||
3535 | /* | |
3536 | * Special case for priority boosted tasks. | |
3537 | * | |
3538 | * If the new priority is lower or equal (user space view) | |
3539 | * than the current (boosted) priority, we just store the new | |
3540 | * normal parameters and do not touch the scheduler class and | |
3541 | * the runqueue. This will be done when the task deboost | |
3542 | * itself. | |
3543 | */ | |
3544 | if (rt_mutex_check_prio(p, newprio)) { | |
3545 | __setscheduler_params(p, attr); | |
3546 | task_rq_unlock(rq, p, &flags); | |
3547 | return 0; | |
3548 | } | |
3549 | ||
da0c1e65 | 3550 | queued = task_on_rq_queued(p); |
051a1d1a | 3551 | running = task_current(rq, p); |
da0c1e65 | 3552 | if (queued) |
4ca9b72b | 3553 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3554 | if (running) |
3555 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3556 | |
83ab0aa0 | 3557 | prev_class = p->sched_class; |
d50dde5a | 3558 | __setscheduler(rq, p, attr); |
f6b53205 | 3559 | |
0e1f3483 HS |
3560 | if (running) |
3561 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 3562 | if (queued) { |
81a44c54 TG |
3563 | /* |
3564 | * We enqueue to tail when the priority of a task is | |
3565 | * increased (user space view). | |
3566 | */ | |
3567 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3568 | } | |
cb469845 | 3569 | |
da7a735e | 3570 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3571 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3572 | |
95e02ca9 TG |
3573 | rt_mutex_adjust_pi(p); |
3574 | ||
1da177e4 LT |
3575 | return 0; |
3576 | } | |
961ccddd | 3577 | |
7479f3c9 PZ |
3578 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3579 | const struct sched_param *param, bool check) | |
3580 | { | |
3581 | struct sched_attr attr = { | |
3582 | .sched_policy = policy, | |
3583 | .sched_priority = param->sched_priority, | |
3584 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3585 | }; | |
3586 | ||
c13db6b1 SR |
3587 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
3588 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
3589 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
3590 | policy &= ~SCHED_RESET_ON_FORK; | |
3591 | attr.sched_policy = policy; | |
3592 | } | |
3593 | ||
3594 | return __sched_setscheduler(p, &attr, check); | |
3595 | } | |
961ccddd RR |
3596 | /** |
3597 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3598 | * @p: the task in question. | |
3599 | * @policy: new policy. | |
3600 | * @param: structure containing the new RT priority. | |
3601 | * | |
e69f6186 YB |
3602 | * Return: 0 on success. An error code otherwise. |
3603 | * | |
961ccddd RR |
3604 | * NOTE that the task may be already dead. |
3605 | */ | |
3606 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3607 | const struct sched_param *param) |
961ccddd | 3608 | { |
7479f3c9 | 3609 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3610 | } |
1da177e4 LT |
3611 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3612 | ||
d50dde5a DF |
3613 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3614 | { | |
3615 | return __sched_setscheduler(p, attr, true); | |
3616 | } | |
3617 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3618 | ||
961ccddd RR |
3619 | /** |
3620 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3621 | * @p: the task in question. | |
3622 | * @policy: new policy. | |
3623 | * @param: structure containing the new RT priority. | |
3624 | * | |
3625 | * Just like sched_setscheduler, only don't bother checking if the | |
3626 | * current context has permission. For example, this is needed in | |
3627 | * stop_machine(): we create temporary high priority worker threads, | |
3628 | * but our caller might not have that capability. | |
e69f6186 YB |
3629 | * |
3630 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3631 | */ |
3632 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3633 | const struct sched_param *param) |
961ccddd | 3634 | { |
7479f3c9 | 3635 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3636 | } |
3637 | ||
95cdf3b7 IM |
3638 | static int |
3639 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3640 | { |
1da177e4 LT |
3641 | struct sched_param lparam; |
3642 | struct task_struct *p; | |
36c8b586 | 3643 | int retval; |
1da177e4 LT |
3644 | |
3645 | if (!param || pid < 0) | |
3646 | return -EINVAL; | |
3647 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3648 | return -EFAULT; | |
5fe1d75f ON |
3649 | |
3650 | rcu_read_lock(); | |
3651 | retval = -ESRCH; | |
1da177e4 | 3652 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3653 | if (p != NULL) |
3654 | retval = sched_setscheduler(p, policy, &lparam); | |
3655 | rcu_read_unlock(); | |
36c8b586 | 3656 | |
1da177e4 LT |
3657 | return retval; |
3658 | } | |
3659 | ||
d50dde5a DF |
3660 | /* |
3661 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3662 | */ | |
3663 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3664 | struct sched_attr *attr) | |
3665 | { | |
3666 | u32 size; | |
3667 | int ret; | |
3668 | ||
3669 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3670 | return -EFAULT; | |
3671 | ||
3672 | /* | |
3673 | * zero the full structure, so that a short copy will be nice. | |
3674 | */ | |
3675 | memset(attr, 0, sizeof(*attr)); | |
3676 | ||
3677 | ret = get_user(size, &uattr->size); | |
3678 | if (ret) | |
3679 | return ret; | |
3680 | ||
3681 | if (size > PAGE_SIZE) /* silly large */ | |
3682 | goto err_size; | |
3683 | ||
3684 | if (!size) /* abi compat */ | |
3685 | size = SCHED_ATTR_SIZE_VER0; | |
3686 | ||
3687 | if (size < SCHED_ATTR_SIZE_VER0) | |
3688 | goto err_size; | |
3689 | ||
3690 | /* | |
3691 | * If we're handed a bigger struct than we know of, | |
3692 | * ensure all the unknown bits are 0 - i.e. new | |
3693 | * user-space does not rely on any kernel feature | |
3694 | * extensions we dont know about yet. | |
3695 | */ | |
3696 | if (size > sizeof(*attr)) { | |
3697 | unsigned char __user *addr; | |
3698 | unsigned char __user *end; | |
3699 | unsigned char val; | |
3700 | ||
3701 | addr = (void __user *)uattr + sizeof(*attr); | |
3702 | end = (void __user *)uattr + size; | |
3703 | ||
3704 | for (; addr < end; addr++) { | |
3705 | ret = get_user(val, addr); | |
3706 | if (ret) | |
3707 | return ret; | |
3708 | if (val) | |
3709 | goto err_size; | |
3710 | } | |
3711 | size = sizeof(*attr); | |
3712 | } | |
3713 | ||
3714 | ret = copy_from_user(attr, uattr, size); | |
3715 | if (ret) | |
3716 | return -EFAULT; | |
3717 | ||
3718 | /* | |
3719 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3720 | * to be strict and return an error on out-of-bounds values? | |
3721 | */ | |
75e45d51 | 3722 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3723 | |
e78c7bca | 3724 | return 0; |
d50dde5a DF |
3725 | |
3726 | err_size: | |
3727 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3728 | return -E2BIG; |
d50dde5a DF |
3729 | } |
3730 | ||
1da177e4 LT |
3731 | /** |
3732 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3733 | * @pid: the pid in question. | |
3734 | * @policy: new policy. | |
3735 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3736 | * |
3737 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3738 | */ |
5add95d4 HC |
3739 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3740 | struct sched_param __user *, param) | |
1da177e4 | 3741 | { |
c21761f1 JB |
3742 | /* negative values for policy are not valid */ |
3743 | if (policy < 0) | |
3744 | return -EINVAL; | |
3745 | ||
1da177e4 LT |
3746 | return do_sched_setscheduler(pid, policy, param); |
3747 | } | |
3748 | ||
3749 | /** | |
3750 | * sys_sched_setparam - set/change the RT priority of a thread | |
3751 | * @pid: the pid in question. | |
3752 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3753 | * |
3754 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3755 | */ |
5add95d4 | 3756 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3757 | { |
c13db6b1 | 3758 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
3759 | } |
3760 | ||
d50dde5a DF |
3761 | /** |
3762 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3763 | * @pid: the pid in question. | |
5778fccf | 3764 | * @uattr: structure containing the extended parameters. |
db66d756 | 3765 | * @flags: for future extension. |
d50dde5a | 3766 | */ |
6d35ab48 PZ |
3767 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3768 | unsigned int, flags) | |
d50dde5a DF |
3769 | { |
3770 | struct sched_attr attr; | |
3771 | struct task_struct *p; | |
3772 | int retval; | |
3773 | ||
6d35ab48 | 3774 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3775 | return -EINVAL; |
3776 | ||
143cf23d MK |
3777 | retval = sched_copy_attr(uattr, &attr); |
3778 | if (retval) | |
3779 | return retval; | |
d50dde5a | 3780 | |
b14ed2c2 | 3781 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 3782 | return -EINVAL; |
d50dde5a DF |
3783 | |
3784 | rcu_read_lock(); | |
3785 | retval = -ESRCH; | |
3786 | p = find_process_by_pid(pid); | |
3787 | if (p != NULL) | |
3788 | retval = sched_setattr(p, &attr); | |
3789 | rcu_read_unlock(); | |
3790 | ||
3791 | return retval; | |
3792 | } | |
3793 | ||
1da177e4 LT |
3794 | /** |
3795 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3796 | * @pid: the pid in question. | |
e69f6186 YB |
3797 | * |
3798 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3799 | * code. | |
1da177e4 | 3800 | */ |
5add95d4 | 3801 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3802 | { |
36c8b586 | 3803 | struct task_struct *p; |
3a5c359a | 3804 | int retval; |
1da177e4 LT |
3805 | |
3806 | if (pid < 0) | |
3a5c359a | 3807 | return -EINVAL; |
1da177e4 LT |
3808 | |
3809 | retval = -ESRCH; | |
5fe85be0 | 3810 | rcu_read_lock(); |
1da177e4 LT |
3811 | p = find_process_by_pid(pid); |
3812 | if (p) { | |
3813 | retval = security_task_getscheduler(p); | |
3814 | if (!retval) | |
ca94c442 LP |
3815 | retval = p->policy |
3816 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3817 | } |
5fe85be0 | 3818 | rcu_read_unlock(); |
1da177e4 LT |
3819 | return retval; |
3820 | } | |
3821 | ||
3822 | /** | |
ca94c442 | 3823 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3824 | * @pid: the pid in question. |
3825 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3826 | * |
3827 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3828 | * code. | |
1da177e4 | 3829 | */ |
5add95d4 | 3830 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3831 | { |
ce5f7f82 | 3832 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3833 | struct task_struct *p; |
3a5c359a | 3834 | int retval; |
1da177e4 LT |
3835 | |
3836 | if (!param || pid < 0) | |
3a5c359a | 3837 | return -EINVAL; |
1da177e4 | 3838 | |
5fe85be0 | 3839 | rcu_read_lock(); |
1da177e4 LT |
3840 | p = find_process_by_pid(pid); |
3841 | retval = -ESRCH; | |
3842 | if (!p) | |
3843 | goto out_unlock; | |
3844 | ||
3845 | retval = security_task_getscheduler(p); | |
3846 | if (retval) | |
3847 | goto out_unlock; | |
3848 | ||
ce5f7f82 PZ |
3849 | if (task_has_rt_policy(p)) |
3850 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3851 | rcu_read_unlock(); |
1da177e4 LT |
3852 | |
3853 | /* | |
3854 | * This one might sleep, we cannot do it with a spinlock held ... | |
3855 | */ | |
3856 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3857 | ||
1da177e4 LT |
3858 | return retval; |
3859 | ||
3860 | out_unlock: | |
5fe85be0 | 3861 | rcu_read_unlock(); |
1da177e4 LT |
3862 | return retval; |
3863 | } | |
3864 | ||
d50dde5a DF |
3865 | static int sched_read_attr(struct sched_attr __user *uattr, |
3866 | struct sched_attr *attr, | |
3867 | unsigned int usize) | |
3868 | { | |
3869 | int ret; | |
3870 | ||
3871 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3872 | return -EFAULT; | |
3873 | ||
3874 | /* | |
3875 | * If we're handed a smaller struct than we know of, | |
3876 | * ensure all the unknown bits are 0 - i.e. old | |
3877 | * user-space does not get uncomplete information. | |
3878 | */ | |
3879 | if (usize < sizeof(*attr)) { | |
3880 | unsigned char *addr; | |
3881 | unsigned char *end; | |
3882 | ||
3883 | addr = (void *)attr + usize; | |
3884 | end = (void *)attr + sizeof(*attr); | |
3885 | ||
3886 | for (; addr < end; addr++) { | |
3887 | if (*addr) | |
22400674 | 3888 | return -EFBIG; |
d50dde5a DF |
3889 | } |
3890 | ||
3891 | attr->size = usize; | |
3892 | } | |
3893 | ||
4efbc454 | 3894 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3895 | if (ret) |
3896 | return -EFAULT; | |
3897 | ||
22400674 | 3898 | return 0; |
d50dde5a DF |
3899 | } |
3900 | ||
3901 | /** | |
aab03e05 | 3902 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3903 | * @pid: the pid in question. |
5778fccf | 3904 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3905 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3906 | * @flags: for future extension. |
d50dde5a | 3907 | */ |
6d35ab48 PZ |
3908 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3909 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3910 | { |
3911 | struct sched_attr attr = { | |
3912 | .size = sizeof(struct sched_attr), | |
3913 | }; | |
3914 | struct task_struct *p; | |
3915 | int retval; | |
3916 | ||
3917 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3918 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3919 | return -EINVAL; |
3920 | ||
3921 | rcu_read_lock(); | |
3922 | p = find_process_by_pid(pid); | |
3923 | retval = -ESRCH; | |
3924 | if (!p) | |
3925 | goto out_unlock; | |
3926 | ||
3927 | retval = security_task_getscheduler(p); | |
3928 | if (retval) | |
3929 | goto out_unlock; | |
3930 | ||
3931 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3932 | if (p->sched_reset_on_fork) |
3933 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3934 | if (task_has_dl_policy(p)) |
3935 | __getparam_dl(p, &attr); | |
3936 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3937 | attr.sched_priority = p->rt_priority; |
3938 | else | |
d0ea0268 | 3939 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3940 | |
3941 | rcu_read_unlock(); | |
3942 | ||
3943 | retval = sched_read_attr(uattr, &attr, size); | |
3944 | return retval; | |
3945 | ||
3946 | out_unlock: | |
3947 | rcu_read_unlock(); | |
3948 | return retval; | |
3949 | } | |
3950 | ||
96f874e2 | 3951 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3952 | { |
5a16f3d3 | 3953 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3954 | struct task_struct *p; |
3955 | int retval; | |
1da177e4 | 3956 | |
23f5d142 | 3957 | rcu_read_lock(); |
1da177e4 LT |
3958 | |
3959 | p = find_process_by_pid(pid); | |
3960 | if (!p) { | |
23f5d142 | 3961 | rcu_read_unlock(); |
1da177e4 LT |
3962 | return -ESRCH; |
3963 | } | |
3964 | ||
23f5d142 | 3965 | /* Prevent p going away */ |
1da177e4 | 3966 | get_task_struct(p); |
23f5d142 | 3967 | rcu_read_unlock(); |
1da177e4 | 3968 | |
14a40ffc TH |
3969 | if (p->flags & PF_NO_SETAFFINITY) { |
3970 | retval = -EINVAL; | |
3971 | goto out_put_task; | |
3972 | } | |
5a16f3d3 RR |
3973 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3974 | retval = -ENOMEM; | |
3975 | goto out_put_task; | |
3976 | } | |
3977 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3978 | retval = -ENOMEM; | |
3979 | goto out_free_cpus_allowed; | |
3980 | } | |
1da177e4 | 3981 | retval = -EPERM; |
4c44aaaf EB |
3982 | if (!check_same_owner(p)) { |
3983 | rcu_read_lock(); | |
3984 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3985 | rcu_read_unlock(); | |
3986 | goto out_unlock; | |
3987 | } | |
3988 | rcu_read_unlock(); | |
3989 | } | |
1da177e4 | 3990 | |
b0ae1981 | 3991 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3992 | if (retval) |
3993 | goto out_unlock; | |
3994 | ||
e4099a5e PZ |
3995 | |
3996 | cpuset_cpus_allowed(p, cpus_allowed); | |
3997 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3998 | ||
332ac17e DF |
3999 | /* |
4000 | * Since bandwidth control happens on root_domain basis, | |
4001 | * if admission test is enabled, we only admit -deadline | |
4002 | * tasks allowed to run on all the CPUs in the task's | |
4003 | * root_domain. | |
4004 | */ | |
4005 | #ifdef CONFIG_SMP | |
4006 | if (task_has_dl_policy(p)) { | |
4007 | const struct cpumask *span = task_rq(p)->rd->span; | |
4008 | ||
e4099a5e | 4009 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
4010 | retval = -EBUSY; |
4011 | goto out_unlock; | |
4012 | } | |
4013 | } | |
4014 | #endif | |
49246274 | 4015 | again: |
5a16f3d3 | 4016 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4017 | |
8707d8b8 | 4018 | if (!retval) { |
5a16f3d3 RR |
4019 | cpuset_cpus_allowed(p, cpus_allowed); |
4020 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4021 | /* |
4022 | * We must have raced with a concurrent cpuset | |
4023 | * update. Just reset the cpus_allowed to the | |
4024 | * cpuset's cpus_allowed | |
4025 | */ | |
5a16f3d3 | 4026 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4027 | goto again; |
4028 | } | |
4029 | } | |
1da177e4 | 4030 | out_unlock: |
5a16f3d3 RR |
4031 | free_cpumask_var(new_mask); |
4032 | out_free_cpus_allowed: | |
4033 | free_cpumask_var(cpus_allowed); | |
4034 | out_put_task: | |
1da177e4 | 4035 | put_task_struct(p); |
1da177e4 LT |
4036 | return retval; |
4037 | } | |
4038 | ||
4039 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4040 | struct cpumask *new_mask) |
1da177e4 | 4041 | { |
96f874e2 RR |
4042 | if (len < cpumask_size()) |
4043 | cpumask_clear(new_mask); | |
4044 | else if (len > cpumask_size()) | |
4045 | len = cpumask_size(); | |
4046 | ||
1da177e4 LT |
4047 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4048 | } | |
4049 | ||
4050 | /** | |
4051 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4052 | * @pid: pid of the process | |
4053 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4054 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4055 | * |
4056 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4057 | */ |
5add95d4 HC |
4058 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4059 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4060 | { |
5a16f3d3 | 4061 | cpumask_var_t new_mask; |
1da177e4 LT |
4062 | int retval; |
4063 | ||
5a16f3d3 RR |
4064 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4065 | return -ENOMEM; | |
1da177e4 | 4066 | |
5a16f3d3 RR |
4067 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4068 | if (retval == 0) | |
4069 | retval = sched_setaffinity(pid, new_mask); | |
4070 | free_cpumask_var(new_mask); | |
4071 | return retval; | |
1da177e4 LT |
4072 | } |
4073 | ||
96f874e2 | 4074 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4075 | { |
36c8b586 | 4076 | struct task_struct *p; |
31605683 | 4077 | unsigned long flags; |
1da177e4 | 4078 | int retval; |
1da177e4 | 4079 | |
23f5d142 | 4080 | rcu_read_lock(); |
1da177e4 LT |
4081 | |
4082 | retval = -ESRCH; | |
4083 | p = find_process_by_pid(pid); | |
4084 | if (!p) | |
4085 | goto out_unlock; | |
4086 | ||
e7834f8f DQ |
4087 | retval = security_task_getscheduler(p); |
4088 | if (retval) | |
4089 | goto out_unlock; | |
4090 | ||
013fdb80 | 4091 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4092 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4093 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4094 | |
4095 | out_unlock: | |
23f5d142 | 4096 | rcu_read_unlock(); |
1da177e4 | 4097 | |
9531b62f | 4098 | return retval; |
1da177e4 LT |
4099 | } |
4100 | ||
4101 | /** | |
4102 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4103 | * @pid: pid of the process | |
4104 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4105 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4106 | * |
4107 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4108 | */ |
5add95d4 HC |
4109 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4110 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4111 | { |
4112 | int ret; | |
f17c8607 | 4113 | cpumask_var_t mask; |
1da177e4 | 4114 | |
84fba5ec | 4115 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4116 | return -EINVAL; |
4117 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4118 | return -EINVAL; |
4119 | ||
f17c8607 RR |
4120 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4121 | return -ENOMEM; | |
1da177e4 | 4122 | |
f17c8607 RR |
4123 | ret = sched_getaffinity(pid, mask); |
4124 | if (ret == 0) { | |
8bc037fb | 4125 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4126 | |
4127 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4128 | ret = -EFAULT; |
4129 | else | |
cd3d8031 | 4130 | ret = retlen; |
f17c8607 RR |
4131 | } |
4132 | free_cpumask_var(mask); | |
1da177e4 | 4133 | |
f17c8607 | 4134 | return ret; |
1da177e4 LT |
4135 | } |
4136 | ||
4137 | /** | |
4138 | * sys_sched_yield - yield the current processor to other threads. | |
4139 | * | |
dd41f596 IM |
4140 | * This function yields the current CPU to other tasks. If there are no |
4141 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4142 | * |
4143 | * Return: 0. | |
1da177e4 | 4144 | */ |
5add95d4 | 4145 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4146 | { |
70b97a7f | 4147 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4148 | |
2d72376b | 4149 | schedstat_inc(rq, yld_count); |
4530d7ab | 4150 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4151 | |
4152 | /* | |
4153 | * Since we are going to call schedule() anyway, there's | |
4154 | * no need to preempt or enable interrupts: | |
4155 | */ | |
4156 | __release(rq->lock); | |
8a25d5de | 4157 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4158 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4159 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4160 | |
4161 | schedule(); | |
4162 | ||
4163 | return 0; | |
4164 | } | |
4165 | ||
e7b38404 | 4166 | static void __cond_resched(void) |
1da177e4 | 4167 | { |
bdb43806 | 4168 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4169 | __schedule(); |
bdb43806 | 4170 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4171 | } |
4172 | ||
02b67cc3 | 4173 | int __sched _cond_resched(void) |
1da177e4 | 4174 | { |
d86ee480 | 4175 | if (should_resched()) { |
1da177e4 LT |
4176 | __cond_resched(); |
4177 | return 1; | |
4178 | } | |
4179 | return 0; | |
4180 | } | |
02b67cc3 | 4181 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4182 | |
4183 | /* | |
613afbf8 | 4184 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4185 | * call schedule, and on return reacquire the lock. |
4186 | * | |
41a2d6cf | 4187 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4188 | * operations here to prevent schedule() from being called twice (once via |
4189 | * spin_unlock(), once by hand). | |
4190 | */ | |
613afbf8 | 4191 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4192 | { |
d86ee480 | 4193 | int resched = should_resched(); |
6df3cecb JK |
4194 | int ret = 0; |
4195 | ||
f607c668 PZ |
4196 | lockdep_assert_held(lock); |
4197 | ||
4a81e832 | 4198 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4199 | spin_unlock(lock); |
d86ee480 | 4200 | if (resched) |
95c354fe NP |
4201 | __cond_resched(); |
4202 | else | |
4203 | cpu_relax(); | |
6df3cecb | 4204 | ret = 1; |
1da177e4 | 4205 | spin_lock(lock); |
1da177e4 | 4206 | } |
6df3cecb | 4207 | return ret; |
1da177e4 | 4208 | } |
613afbf8 | 4209 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4210 | |
613afbf8 | 4211 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4212 | { |
4213 | BUG_ON(!in_softirq()); | |
4214 | ||
d86ee480 | 4215 | if (should_resched()) { |
98d82567 | 4216 | local_bh_enable(); |
1da177e4 LT |
4217 | __cond_resched(); |
4218 | local_bh_disable(); | |
4219 | return 1; | |
4220 | } | |
4221 | return 0; | |
4222 | } | |
613afbf8 | 4223 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4224 | |
1da177e4 LT |
4225 | /** |
4226 | * yield - yield the current processor to other threads. | |
4227 | * | |
8e3fabfd PZ |
4228 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4229 | * | |
4230 | * The scheduler is at all times free to pick the calling task as the most | |
4231 | * eligible task to run, if removing the yield() call from your code breaks | |
4232 | * it, its already broken. | |
4233 | * | |
4234 | * Typical broken usage is: | |
4235 | * | |
4236 | * while (!event) | |
4237 | * yield(); | |
4238 | * | |
4239 | * where one assumes that yield() will let 'the other' process run that will | |
4240 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4241 | * happen. Never use yield() as a progress guarantee!! | |
4242 | * | |
4243 | * If you want to use yield() to wait for something, use wait_event(). | |
4244 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4245 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4246 | */ |
4247 | void __sched yield(void) | |
4248 | { | |
4249 | set_current_state(TASK_RUNNING); | |
4250 | sys_sched_yield(); | |
4251 | } | |
1da177e4 LT |
4252 | EXPORT_SYMBOL(yield); |
4253 | ||
d95f4122 MG |
4254 | /** |
4255 | * yield_to - yield the current processor to another thread in | |
4256 | * your thread group, or accelerate that thread toward the | |
4257 | * processor it's on. | |
16addf95 RD |
4258 | * @p: target task |
4259 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4260 | * |
4261 | * It's the caller's job to ensure that the target task struct | |
4262 | * can't go away on us before we can do any checks. | |
4263 | * | |
e69f6186 | 4264 | * Return: |
7b270f60 PZ |
4265 | * true (>0) if we indeed boosted the target task. |
4266 | * false (0) if we failed to boost the target. | |
4267 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4268 | */ |
fa93384f | 4269 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4270 | { |
4271 | struct task_struct *curr = current; | |
4272 | struct rq *rq, *p_rq; | |
4273 | unsigned long flags; | |
c3c18640 | 4274 | int yielded = 0; |
d95f4122 MG |
4275 | |
4276 | local_irq_save(flags); | |
4277 | rq = this_rq(); | |
4278 | ||
4279 | again: | |
4280 | p_rq = task_rq(p); | |
7b270f60 PZ |
4281 | /* |
4282 | * If we're the only runnable task on the rq and target rq also | |
4283 | * has only one task, there's absolutely no point in yielding. | |
4284 | */ | |
4285 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4286 | yielded = -ESRCH; | |
4287 | goto out_irq; | |
4288 | } | |
4289 | ||
d95f4122 | 4290 | double_rq_lock(rq, p_rq); |
39e24d8f | 4291 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4292 | double_rq_unlock(rq, p_rq); |
4293 | goto again; | |
4294 | } | |
4295 | ||
4296 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4297 | goto out_unlock; |
d95f4122 MG |
4298 | |
4299 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4300 | goto out_unlock; |
d95f4122 MG |
4301 | |
4302 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4303 | goto out_unlock; |
d95f4122 MG |
4304 | |
4305 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4306 | if (yielded) { |
d95f4122 | 4307 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4308 | /* |
4309 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4310 | * fairness. | |
4311 | */ | |
4312 | if (preempt && rq != p_rq) | |
8875125e | 4313 | resched_curr(p_rq); |
6d1cafd8 | 4314 | } |
d95f4122 | 4315 | |
7b270f60 | 4316 | out_unlock: |
d95f4122 | 4317 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4318 | out_irq: |
d95f4122 MG |
4319 | local_irq_restore(flags); |
4320 | ||
7b270f60 | 4321 | if (yielded > 0) |
d95f4122 MG |
4322 | schedule(); |
4323 | ||
4324 | return yielded; | |
4325 | } | |
4326 | EXPORT_SYMBOL_GPL(yield_to); | |
4327 | ||
1da177e4 | 4328 | /* |
41a2d6cf | 4329 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4330 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4331 | */ |
4332 | void __sched io_schedule(void) | |
4333 | { | |
54d35f29 | 4334 | struct rq *rq = raw_rq(); |
1da177e4 | 4335 | |
0ff92245 | 4336 | delayacct_blkio_start(); |
1da177e4 | 4337 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4338 | blk_flush_plug(current); |
8f0dfc34 | 4339 | current->in_iowait = 1; |
1da177e4 | 4340 | schedule(); |
8f0dfc34 | 4341 | current->in_iowait = 0; |
1da177e4 | 4342 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4343 | delayacct_blkio_end(); |
1da177e4 | 4344 | } |
1da177e4 LT |
4345 | EXPORT_SYMBOL(io_schedule); |
4346 | ||
4347 | long __sched io_schedule_timeout(long timeout) | |
4348 | { | |
54d35f29 | 4349 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4350 | long ret; |
4351 | ||
0ff92245 | 4352 | delayacct_blkio_start(); |
1da177e4 | 4353 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4354 | blk_flush_plug(current); |
8f0dfc34 | 4355 | current->in_iowait = 1; |
1da177e4 | 4356 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4357 | current->in_iowait = 0; |
1da177e4 | 4358 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4359 | delayacct_blkio_end(); |
1da177e4 LT |
4360 | return ret; |
4361 | } | |
4362 | ||
4363 | /** | |
4364 | * sys_sched_get_priority_max - return maximum RT priority. | |
4365 | * @policy: scheduling class. | |
4366 | * | |
e69f6186 YB |
4367 | * Return: On success, this syscall returns the maximum |
4368 | * rt_priority that can be used by a given scheduling class. | |
4369 | * On failure, a negative error code is returned. | |
1da177e4 | 4370 | */ |
5add95d4 | 4371 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4372 | { |
4373 | int ret = -EINVAL; | |
4374 | ||
4375 | switch (policy) { | |
4376 | case SCHED_FIFO: | |
4377 | case SCHED_RR: | |
4378 | ret = MAX_USER_RT_PRIO-1; | |
4379 | break; | |
aab03e05 | 4380 | case SCHED_DEADLINE: |
1da177e4 | 4381 | case SCHED_NORMAL: |
b0a9499c | 4382 | case SCHED_BATCH: |
dd41f596 | 4383 | case SCHED_IDLE: |
1da177e4 LT |
4384 | ret = 0; |
4385 | break; | |
4386 | } | |
4387 | return ret; | |
4388 | } | |
4389 | ||
4390 | /** | |
4391 | * sys_sched_get_priority_min - return minimum RT priority. | |
4392 | * @policy: scheduling class. | |
4393 | * | |
e69f6186 YB |
4394 | * Return: On success, this syscall returns the minimum |
4395 | * rt_priority that can be used by a given scheduling class. | |
4396 | * On failure, a negative error code is returned. | |
1da177e4 | 4397 | */ |
5add95d4 | 4398 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4399 | { |
4400 | int ret = -EINVAL; | |
4401 | ||
4402 | switch (policy) { | |
4403 | case SCHED_FIFO: | |
4404 | case SCHED_RR: | |
4405 | ret = 1; | |
4406 | break; | |
aab03e05 | 4407 | case SCHED_DEADLINE: |
1da177e4 | 4408 | case SCHED_NORMAL: |
b0a9499c | 4409 | case SCHED_BATCH: |
dd41f596 | 4410 | case SCHED_IDLE: |
1da177e4 LT |
4411 | ret = 0; |
4412 | } | |
4413 | return ret; | |
4414 | } | |
4415 | ||
4416 | /** | |
4417 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4418 | * @pid: pid of the process. | |
4419 | * @interval: userspace pointer to the timeslice value. | |
4420 | * | |
4421 | * this syscall writes the default timeslice value of a given process | |
4422 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4423 | * |
4424 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4425 | * an error code. | |
1da177e4 | 4426 | */ |
17da2bd9 | 4427 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4428 | struct timespec __user *, interval) |
1da177e4 | 4429 | { |
36c8b586 | 4430 | struct task_struct *p; |
a4ec24b4 | 4431 | unsigned int time_slice; |
dba091b9 TG |
4432 | unsigned long flags; |
4433 | struct rq *rq; | |
3a5c359a | 4434 | int retval; |
1da177e4 | 4435 | struct timespec t; |
1da177e4 LT |
4436 | |
4437 | if (pid < 0) | |
3a5c359a | 4438 | return -EINVAL; |
1da177e4 LT |
4439 | |
4440 | retval = -ESRCH; | |
1a551ae7 | 4441 | rcu_read_lock(); |
1da177e4 LT |
4442 | p = find_process_by_pid(pid); |
4443 | if (!p) | |
4444 | goto out_unlock; | |
4445 | ||
4446 | retval = security_task_getscheduler(p); | |
4447 | if (retval) | |
4448 | goto out_unlock; | |
4449 | ||
dba091b9 | 4450 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4451 | time_slice = 0; |
4452 | if (p->sched_class->get_rr_interval) | |
4453 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4454 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4455 | |
1a551ae7 | 4456 | rcu_read_unlock(); |
a4ec24b4 | 4457 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4458 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4459 | return retval; |
3a5c359a | 4460 | |
1da177e4 | 4461 | out_unlock: |
1a551ae7 | 4462 | rcu_read_unlock(); |
1da177e4 LT |
4463 | return retval; |
4464 | } | |
4465 | ||
7c731e0a | 4466 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4467 | |
82a1fcb9 | 4468 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4469 | { |
1da177e4 | 4470 | unsigned long free = 0; |
4e79752c | 4471 | int ppid; |
36c8b586 | 4472 | unsigned state; |
1da177e4 | 4473 | |
1da177e4 | 4474 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4475 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4476 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4477 | #if BITS_PER_LONG == 32 |
1da177e4 | 4478 | if (state == TASK_RUNNING) |
3df0fc5b | 4479 | printk(KERN_CONT " running "); |
1da177e4 | 4480 | else |
3df0fc5b | 4481 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4482 | #else |
4483 | if (state == TASK_RUNNING) | |
3df0fc5b | 4484 | printk(KERN_CONT " running task "); |
1da177e4 | 4485 | else |
3df0fc5b | 4486 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4487 | #endif |
4488 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4489 | free = stack_not_used(p); |
1da177e4 | 4490 | #endif |
4e79752c PM |
4491 | rcu_read_lock(); |
4492 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4493 | rcu_read_unlock(); | |
3df0fc5b | 4494 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4495 | task_pid_nr(p), ppid, |
aa47b7e0 | 4496 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4497 | |
3d1cb205 | 4498 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4499 | show_stack(p, NULL); |
1da177e4 LT |
4500 | } |
4501 | ||
e59e2ae2 | 4502 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4503 | { |
36c8b586 | 4504 | struct task_struct *g, *p; |
1da177e4 | 4505 | |
4bd77321 | 4506 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4507 | printk(KERN_INFO |
4508 | " task PC stack pid father\n"); | |
1da177e4 | 4509 | #else |
3df0fc5b PZ |
4510 | printk(KERN_INFO |
4511 | " task PC stack pid father\n"); | |
1da177e4 | 4512 | #endif |
510f5acc | 4513 | rcu_read_lock(); |
5d07f420 | 4514 | for_each_process_thread(g, p) { |
1da177e4 LT |
4515 | /* |
4516 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4517 | * console might take a lot of time: |
1da177e4 LT |
4518 | */ |
4519 | touch_nmi_watchdog(); | |
39bc89fd | 4520 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4521 | sched_show_task(p); |
5d07f420 | 4522 | } |
1da177e4 | 4523 | |
04c9167f JF |
4524 | touch_all_softlockup_watchdogs(); |
4525 | ||
dd41f596 IM |
4526 | #ifdef CONFIG_SCHED_DEBUG |
4527 | sysrq_sched_debug_show(); | |
4528 | #endif | |
510f5acc | 4529 | rcu_read_unlock(); |
e59e2ae2 IM |
4530 | /* |
4531 | * Only show locks if all tasks are dumped: | |
4532 | */ | |
93335a21 | 4533 | if (!state_filter) |
e59e2ae2 | 4534 | debug_show_all_locks(); |
1da177e4 LT |
4535 | } |
4536 | ||
0db0628d | 4537 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4538 | { |
dd41f596 | 4539 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4540 | } |
4541 | ||
f340c0d1 IM |
4542 | /** |
4543 | * init_idle - set up an idle thread for a given CPU | |
4544 | * @idle: task in question | |
4545 | * @cpu: cpu the idle task belongs to | |
4546 | * | |
4547 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4548 | * flag, to make booting more robust. | |
4549 | */ | |
0db0628d | 4550 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4551 | { |
70b97a7f | 4552 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4553 | unsigned long flags; |
4554 | ||
05fa785c | 4555 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4556 | |
5e1576ed | 4557 | __sched_fork(0, idle); |
06b83b5f | 4558 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4559 | idle->se.exec_start = sched_clock(); |
4560 | ||
1e1b6c51 | 4561 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4562 | /* |
4563 | * We're having a chicken and egg problem, even though we are | |
4564 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4565 | * lockdep check in task_group() will fail. | |
4566 | * | |
4567 | * Similar case to sched_fork(). / Alternatively we could | |
4568 | * use task_rq_lock() here and obtain the other rq->lock. | |
4569 | * | |
4570 | * Silence PROVE_RCU | |
4571 | */ | |
4572 | rcu_read_lock(); | |
dd41f596 | 4573 | __set_task_cpu(idle, cpu); |
6506cf6c | 4574 | rcu_read_unlock(); |
1da177e4 | 4575 | |
1da177e4 | 4576 | rq->curr = rq->idle = idle; |
da0c1e65 | 4577 | idle->on_rq = TASK_ON_RQ_QUEUED; |
3ca7a440 PZ |
4578 | #if defined(CONFIG_SMP) |
4579 | idle->on_cpu = 1; | |
4866cde0 | 4580 | #endif |
05fa785c | 4581 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4582 | |
4583 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4584 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4585 | |
dd41f596 IM |
4586 | /* |
4587 | * The idle tasks have their own, simple scheduling class: | |
4588 | */ | |
4589 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4590 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4591 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4592 | #if defined(CONFIG_SMP) |
4593 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4594 | #endif | |
19978ca6 IM |
4595 | } |
4596 | ||
1da177e4 | 4597 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4598 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4599 | { | |
4600 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4601 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4602 | |
4603 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4604 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4605 | } |
4606 | ||
1da177e4 LT |
4607 | /* |
4608 | * This is how migration works: | |
4609 | * | |
969c7921 TH |
4610 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4611 | * stop_one_cpu(). | |
4612 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4613 | * off the CPU) | |
4614 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4615 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4616 | * it and puts it into the right queue. |
969c7921 TH |
4617 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4618 | * is done. | |
1da177e4 LT |
4619 | */ |
4620 | ||
4621 | /* | |
4622 | * Change a given task's CPU affinity. Migrate the thread to a | |
4623 | * proper CPU and schedule it away if the CPU it's executing on | |
4624 | * is removed from the allowed bitmask. | |
4625 | * | |
4626 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4627 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4628 | * call is not atomic; no spinlocks may be held. |
4629 | */ | |
96f874e2 | 4630 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4631 | { |
4632 | unsigned long flags; | |
70b97a7f | 4633 | struct rq *rq; |
969c7921 | 4634 | unsigned int dest_cpu; |
48f24c4d | 4635 | int ret = 0; |
1da177e4 LT |
4636 | |
4637 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4638 | |
db44fc01 YZ |
4639 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4640 | goto out; | |
4641 | ||
6ad4c188 | 4642 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4643 | ret = -EINVAL; |
4644 | goto out; | |
4645 | } | |
4646 | ||
1e1b6c51 | 4647 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4648 | |
1da177e4 | 4649 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4650 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4651 | goto out; |
4652 | ||
969c7921 | 4653 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
da0c1e65 | 4654 | if (task_on_rq_queued(p)) { |
969c7921 | 4655 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4656 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4657 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4658 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4659 | tlb_migrate_finish(p->mm); |
4660 | return 0; | |
4661 | } | |
4662 | out: | |
0122ec5b | 4663 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4664 | |
1da177e4 LT |
4665 | return ret; |
4666 | } | |
cd8ba7cd | 4667 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4668 | |
4669 | /* | |
41a2d6cf | 4670 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4671 | * this because either it can't run here any more (set_cpus_allowed() |
4672 | * away from this CPU, or CPU going down), or because we're | |
4673 | * attempting to rebalance this task on exec (sched_exec). | |
4674 | * | |
4675 | * So we race with normal scheduler movements, but that's OK, as long | |
4676 | * as the task is no longer on this CPU. | |
efc30814 KK |
4677 | * |
4678 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4679 | */ |
efc30814 | 4680 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4681 | { |
a1e01829 | 4682 | struct rq *rq; |
e2912009 | 4683 | int ret = 0; |
1da177e4 | 4684 | |
e761b772 | 4685 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4686 | return ret; |
1da177e4 | 4687 | |
a1e01829 | 4688 | rq = cpu_rq(src_cpu); |
1da177e4 | 4689 | |
0122ec5b | 4690 | raw_spin_lock(&p->pi_lock); |
a1e01829 | 4691 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
4692 | /* Already moved. */ |
4693 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4694 | goto done; |
a1e01829 | 4695 | |
1da177e4 | 4696 | /* Affinity changed (again). */ |
fa17b507 | 4697 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4698 | goto fail; |
1da177e4 | 4699 | |
e2912009 PZ |
4700 | /* |
4701 | * If we're not on a rq, the next wake-up will ensure we're | |
4702 | * placed properly. | |
4703 | */ | |
da0c1e65 | 4704 | if (task_on_rq_queued(p)) { |
a1e01829 KT |
4705 | dequeue_task(rq, p, 0); |
4706 | p->on_rq = TASK_ON_RQ_MIGRATING; | |
e2912009 | 4707 | set_task_cpu(p, dest_cpu); |
a1e01829 KT |
4708 | raw_spin_unlock(&rq->lock); |
4709 | ||
4710 | rq = cpu_rq(dest_cpu); | |
4711 | raw_spin_lock(&rq->lock); | |
4712 | BUG_ON(task_rq(p) != rq); | |
4713 | p->on_rq = TASK_ON_RQ_QUEUED; | |
4714 | enqueue_task(rq, p, 0); | |
4715 | check_preempt_curr(rq, p, 0); | |
1da177e4 | 4716 | } |
b1e38734 | 4717 | done: |
efc30814 | 4718 | ret = 1; |
b1e38734 | 4719 | fail: |
a1e01829 | 4720 | raw_spin_unlock(&rq->lock); |
0122ec5b | 4721 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4722 | return ret; |
1da177e4 LT |
4723 | } |
4724 | ||
e6628d5b MG |
4725 | #ifdef CONFIG_NUMA_BALANCING |
4726 | /* Migrate current task p to target_cpu */ | |
4727 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4728 | { | |
4729 | struct migration_arg arg = { p, target_cpu }; | |
4730 | int curr_cpu = task_cpu(p); | |
4731 | ||
4732 | if (curr_cpu == target_cpu) | |
4733 | return 0; | |
4734 | ||
4735 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4736 | return -EINVAL; | |
4737 | ||
4738 | /* TODO: This is not properly updating schedstats */ | |
4739 | ||
286549dc | 4740 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4741 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4742 | } | |
0ec8aa00 PZ |
4743 | |
4744 | /* | |
4745 | * Requeue a task on a given node and accurately track the number of NUMA | |
4746 | * tasks on the runqueues | |
4747 | */ | |
4748 | void sched_setnuma(struct task_struct *p, int nid) | |
4749 | { | |
4750 | struct rq *rq; | |
4751 | unsigned long flags; | |
da0c1e65 | 4752 | bool queued, running; |
0ec8aa00 PZ |
4753 | |
4754 | rq = task_rq_lock(p, &flags); | |
da0c1e65 | 4755 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
4756 | running = task_current(rq, p); |
4757 | ||
da0c1e65 | 4758 | if (queued) |
0ec8aa00 PZ |
4759 | dequeue_task(rq, p, 0); |
4760 | if (running) | |
4761 | p->sched_class->put_prev_task(rq, p); | |
4762 | ||
4763 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4764 | |
4765 | if (running) | |
4766 | p->sched_class->set_curr_task(rq); | |
da0c1e65 | 4767 | if (queued) |
0ec8aa00 PZ |
4768 | enqueue_task(rq, p, 0); |
4769 | task_rq_unlock(rq, p, &flags); | |
4770 | } | |
e6628d5b MG |
4771 | #endif |
4772 | ||
1da177e4 | 4773 | /* |
969c7921 TH |
4774 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4775 | * and performs thread migration by bumping thread off CPU then | |
4776 | * 'pushing' onto another runqueue. | |
1da177e4 | 4777 | */ |
969c7921 | 4778 | static int migration_cpu_stop(void *data) |
1da177e4 | 4779 | { |
969c7921 | 4780 | struct migration_arg *arg = data; |
f7b4cddc | 4781 | |
969c7921 TH |
4782 | /* |
4783 | * The original target cpu might have gone down and we might | |
4784 | * be on another cpu but it doesn't matter. | |
4785 | */ | |
f7b4cddc | 4786 | local_irq_disable(); |
969c7921 | 4787 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4788 | local_irq_enable(); |
1da177e4 | 4789 | return 0; |
f7b4cddc ON |
4790 | } |
4791 | ||
1da177e4 | 4792 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4793 | |
054b9108 | 4794 | /* |
48c5ccae PZ |
4795 | * Ensures that the idle task is using init_mm right before its cpu goes |
4796 | * offline. | |
054b9108 | 4797 | */ |
48c5ccae | 4798 | void idle_task_exit(void) |
1da177e4 | 4799 | { |
48c5ccae | 4800 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4801 | |
48c5ccae | 4802 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4803 | |
a53efe5f | 4804 | if (mm != &init_mm) { |
48c5ccae | 4805 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4806 | finish_arch_post_lock_switch(); |
4807 | } | |
48c5ccae | 4808 | mmdrop(mm); |
1da177e4 LT |
4809 | } |
4810 | ||
4811 | /* | |
5d180232 PZ |
4812 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4813 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4814 | * nr_active count is stable. | |
4815 | * | |
4816 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4817 | */ |
5d180232 | 4818 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4819 | { |
5d180232 PZ |
4820 | long delta = calc_load_fold_active(rq); |
4821 | if (delta) | |
4822 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4823 | } |
4824 | ||
3f1d2a31 PZ |
4825 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4826 | { | |
4827 | } | |
4828 | ||
4829 | static const struct sched_class fake_sched_class = { | |
4830 | .put_prev_task = put_prev_task_fake, | |
4831 | }; | |
4832 | ||
4833 | static struct task_struct fake_task = { | |
4834 | /* | |
4835 | * Avoid pull_{rt,dl}_task() | |
4836 | */ | |
4837 | .prio = MAX_PRIO + 1, | |
4838 | .sched_class = &fake_sched_class, | |
4839 | }; | |
4840 | ||
48f24c4d | 4841 | /* |
48c5ccae PZ |
4842 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4843 | * try_to_wake_up()->select_task_rq(). | |
4844 | * | |
4845 | * Called with rq->lock held even though we'er in stop_machine() and | |
4846 | * there's no concurrency possible, we hold the required locks anyway | |
4847 | * because of lock validation efforts. | |
1da177e4 | 4848 | */ |
48c5ccae | 4849 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4850 | { |
70b97a7f | 4851 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4852 | struct task_struct *next, *stop = rq->stop; |
4853 | int dest_cpu; | |
1da177e4 LT |
4854 | |
4855 | /* | |
48c5ccae PZ |
4856 | * Fudge the rq selection such that the below task selection loop |
4857 | * doesn't get stuck on the currently eligible stop task. | |
4858 | * | |
4859 | * We're currently inside stop_machine() and the rq is either stuck | |
4860 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4861 | * either way we should never end up calling schedule() until we're | |
4862 | * done here. | |
1da177e4 | 4863 | */ |
48c5ccae | 4864 | rq->stop = NULL; |
48f24c4d | 4865 | |
77bd3970 FW |
4866 | /* |
4867 | * put_prev_task() and pick_next_task() sched | |
4868 | * class method both need to have an up-to-date | |
4869 | * value of rq->clock[_task] | |
4870 | */ | |
4871 | update_rq_clock(rq); | |
4872 | ||
dd41f596 | 4873 | for ( ; ; ) { |
48c5ccae PZ |
4874 | /* |
4875 | * There's this thread running, bail when that's the only | |
4876 | * remaining thread. | |
4877 | */ | |
4878 | if (rq->nr_running == 1) | |
dd41f596 | 4879 | break; |
48c5ccae | 4880 | |
3f1d2a31 | 4881 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4882 | BUG_ON(!next); |
79c53799 | 4883 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4884 | |
48c5ccae PZ |
4885 | /* Find suitable destination for @next, with force if needed. */ |
4886 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4887 | raw_spin_unlock(&rq->lock); | |
4888 | ||
4889 | __migrate_task(next, dead_cpu, dest_cpu); | |
4890 | ||
4891 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4892 | } |
dce48a84 | 4893 | |
48c5ccae | 4894 | rq->stop = stop; |
dce48a84 | 4895 | } |
48c5ccae | 4896 | |
1da177e4 LT |
4897 | #endif /* CONFIG_HOTPLUG_CPU */ |
4898 | ||
e692ab53 NP |
4899 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4900 | ||
4901 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4902 | { |
4903 | .procname = "sched_domain", | |
c57baf1e | 4904 | .mode = 0555, |
e0361851 | 4905 | }, |
56992309 | 4906 | {} |
e692ab53 NP |
4907 | }; |
4908 | ||
4909 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4910 | { |
4911 | .procname = "kernel", | |
c57baf1e | 4912 | .mode = 0555, |
e0361851 AD |
4913 | .child = sd_ctl_dir, |
4914 | }, | |
56992309 | 4915 | {} |
e692ab53 NP |
4916 | }; |
4917 | ||
4918 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4919 | { | |
4920 | struct ctl_table *entry = | |
5cf9f062 | 4921 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4922 | |
e692ab53 NP |
4923 | return entry; |
4924 | } | |
4925 | ||
6382bc90 MM |
4926 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4927 | { | |
cd790076 | 4928 | struct ctl_table *entry; |
6382bc90 | 4929 | |
cd790076 MM |
4930 | /* |
4931 | * In the intermediate directories, both the child directory and | |
4932 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4933 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4934 | * static strings and all have proc handlers. |
4935 | */ | |
4936 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4937 | if (entry->child) |
4938 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4939 | if (entry->proc_handler == NULL) |
4940 | kfree(entry->procname); | |
4941 | } | |
6382bc90 MM |
4942 | |
4943 | kfree(*tablep); | |
4944 | *tablep = NULL; | |
4945 | } | |
4946 | ||
201c373e | 4947 | static int min_load_idx = 0; |
fd9b86d3 | 4948 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4949 | |
e692ab53 | 4950 | static void |
e0361851 | 4951 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4952 | const char *procname, void *data, int maxlen, |
201c373e NK |
4953 | umode_t mode, proc_handler *proc_handler, |
4954 | bool load_idx) | |
e692ab53 | 4955 | { |
e692ab53 NP |
4956 | entry->procname = procname; |
4957 | entry->data = data; | |
4958 | entry->maxlen = maxlen; | |
4959 | entry->mode = mode; | |
4960 | entry->proc_handler = proc_handler; | |
201c373e NK |
4961 | |
4962 | if (load_idx) { | |
4963 | entry->extra1 = &min_load_idx; | |
4964 | entry->extra2 = &max_load_idx; | |
4965 | } | |
e692ab53 NP |
4966 | } |
4967 | ||
4968 | static struct ctl_table * | |
4969 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4970 | { | |
37e6bae8 | 4971 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4972 | |
ad1cdc1d MM |
4973 | if (table == NULL) |
4974 | return NULL; | |
4975 | ||
e0361851 | 4976 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4977 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4978 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4979 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4980 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4981 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4982 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4983 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4984 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4985 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4986 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4987 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4988 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4989 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4990 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4991 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4992 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4993 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4994 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4995 | &sd->cache_nice_tries, |
201c373e | 4996 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4997 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4998 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4999 | set_table_entry(&table[11], "max_newidle_lb_cost", |
5000 | &sd->max_newidle_lb_cost, | |
5001 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
5002 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 5003 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 5004 | /* &table[13] is terminator */ |
e692ab53 NP |
5005 | |
5006 | return table; | |
5007 | } | |
5008 | ||
be7002e6 | 5009 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5010 | { |
5011 | struct ctl_table *entry, *table; | |
5012 | struct sched_domain *sd; | |
5013 | int domain_num = 0, i; | |
5014 | char buf[32]; | |
5015 | ||
5016 | for_each_domain(cpu, sd) | |
5017 | domain_num++; | |
5018 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5019 | if (table == NULL) |
5020 | return NULL; | |
e692ab53 NP |
5021 | |
5022 | i = 0; | |
5023 | for_each_domain(cpu, sd) { | |
5024 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5025 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5026 | entry->mode = 0555; |
e692ab53 NP |
5027 | entry->child = sd_alloc_ctl_domain_table(sd); |
5028 | entry++; | |
5029 | i++; | |
5030 | } | |
5031 | return table; | |
5032 | } | |
5033 | ||
5034 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5035 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5036 | { |
6ad4c188 | 5037 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5038 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5039 | char buf[32]; | |
5040 | ||
7378547f MM |
5041 | WARN_ON(sd_ctl_dir[0].child); |
5042 | sd_ctl_dir[0].child = entry; | |
5043 | ||
ad1cdc1d MM |
5044 | if (entry == NULL) |
5045 | return; | |
5046 | ||
6ad4c188 | 5047 | for_each_possible_cpu(i) { |
e692ab53 | 5048 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5049 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5050 | entry->mode = 0555; |
e692ab53 | 5051 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5052 | entry++; |
e692ab53 | 5053 | } |
7378547f MM |
5054 | |
5055 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5056 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5057 | } | |
6382bc90 | 5058 | |
7378547f | 5059 | /* may be called multiple times per register */ |
6382bc90 MM |
5060 | static void unregister_sched_domain_sysctl(void) |
5061 | { | |
7378547f MM |
5062 | if (sd_sysctl_header) |
5063 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5064 | sd_sysctl_header = NULL; |
7378547f MM |
5065 | if (sd_ctl_dir[0].child) |
5066 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5067 | } |
e692ab53 | 5068 | #else |
6382bc90 MM |
5069 | static void register_sched_domain_sysctl(void) |
5070 | { | |
5071 | } | |
5072 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5073 | { |
5074 | } | |
5075 | #endif | |
5076 | ||
1f11eb6a GH |
5077 | static void set_rq_online(struct rq *rq) |
5078 | { | |
5079 | if (!rq->online) { | |
5080 | const struct sched_class *class; | |
5081 | ||
c6c4927b | 5082 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5083 | rq->online = 1; |
5084 | ||
5085 | for_each_class(class) { | |
5086 | if (class->rq_online) | |
5087 | class->rq_online(rq); | |
5088 | } | |
5089 | } | |
5090 | } | |
5091 | ||
5092 | static void set_rq_offline(struct rq *rq) | |
5093 | { | |
5094 | if (rq->online) { | |
5095 | const struct sched_class *class; | |
5096 | ||
5097 | for_each_class(class) { | |
5098 | if (class->rq_offline) | |
5099 | class->rq_offline(rq); | |
5100 | } | |
5101 | ||
c6c4927b | 5102 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5103 | rq->online = 0; |
5104 | } | |
5105 | } | |
5106 | ||
1da177e4 LT |
5107 | /* |
5108 | * migration_call - callback that gets triggered when a CPU is added. | |
5109 | * Here we can start up the necessary migration thread for the new CPU. | |
5110 | */ | |
0db0628d | 5111 | static int |
48f24c4d | 5112 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5113 | { |
48f24c4d | 5114 | int cpu = (long)hcpu; |
1da177e4 | 5115 | unsigned long flags; |
969c7921 | 5116 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5117 | |
48c5ccae | 5118 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5119 | |
1da177e4 | 5120 | case CPU_UP_PREPARE: |
a468d389 | 5121 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5122 | break; |
48f24c4d | 5123 | |
1da177e4 | 5124 | case CPU_ONLINE: |
1f94ef59 | 5125 | /* Update our root-domain */ |
05fa785c | 5126 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5127 | if (rq->rd) { |
c6c4927b | 5128 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5129 | |
5130 | set_rq_online(rq); | |
1f94ef59 | 5131 | } |
05fa785c | 5132 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5133 | break; |
48f24c4d | 5134 | |
1da177e4 | 5135 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5136 | case CPU_DYING: |
317f3941 | 5137 | sched_ttwu_pending(); |
57d885fe | 5138 | /* Update our root-domain */ |
05fa785c | 5139 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5140 | if (rq->rd) { |
c6c4927b | 5141 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5142 | set_rq_offline(rq); |
57d885fe | 5143 | } |
48c5ccae PZ |
5144 | migrate_tasks(cpu); |
5145 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5146 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5147 | break; |
48c5ccae | 5148 | |
5d180232 | 5149 | case CPU_DEAD: |
f319da0c | 5150 | calc_load_migrate(rq); |
57d885fe | 5151 | break; |
1da177e4 LT |
5152 | #endif |
5153 | } | |
49c022e6 PZ |
5154 | |
5155 | update_max_interval(); | |
5156 | ||
1da177e4 LT |
5157 | return NOTIFY_OK; |
5158 | } | |
5159 | ||
f38b0820 PM |
5160 | /* |
5161 | * Register at high priority so that task migration (migrate_all_tasks) | |
5162 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5163 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5164 | */ |
0db0628d | 5165 | static struct notifier_block migration_notifier = { |
1da177e4 | 5166 | .notifier_call = migration_call, |
50a323b7 | 5167 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5168 | }; |
5169 | ||
a803f026 CM |
5170 | static void __cpuinit set_cpu_rq_start_time(void) |
5171 | { | |
5172 | int cpu = smp_processor_id(); | |
5173 | struct rq *rq = cpu_rq(cpu); | |
5174 | rq->age_stamp = sched_clock_cpu(cpu); | |
5175 | } | |
5176 | ||
0db0628d | 5177 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5178 | unsigned long action, void *hcpu) |
5179 | { | |
5180 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5181 | case CPU_STARTING: |
5182 | set_cpu_rq_start_time(); | |
5183 | return NOTIFY_OK; | |
3a101d05 TH |
5184 | case CPU_DOWN_FAILED: |
5185 | set_cpu_active((long)hcpu, true); | |
5186 | return NOTIFY_OK; | |
5187 | default: | |
5188 | return NOTIFY_DONE; | |
5189 | } | |
5190 | } | |
5191 | ||
0db0628d | 5192 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5193 | unsigned long action, void *hcpu) |
5194 | { | |
de212f18 PZ |
5195 | unsigned long flags; |
5196 | long cpu = (long)hcpu; | |
5197 | ||
3a101d05 TH |
5198 | switch (action & ~CPU_TASKS_FROZEN) { |
5199 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5200 | set_cpu_active(cpu, false); |
5201 | ||
5202 | /* explicitly allow suspend */ | |
5203 | if (!(action & CPU_TASKS_FROZEN)) { | |
5204 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5205 | bool overflow; | |
5206 | int cpus; | |
5207 | ||
5208 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5209 | cpus = dl_bw_cpus(cpu); | |
5210 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5211 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5212 | ||
5213 | if (overflow) | |
5214 | return notifier_from_errno(-EBUSY); | |
5215 | } | |
3a101d05 | 5216 | return NOTIFY_OK; |
3a101d05 | 5217 | } |
de212f18 PZ |
5218 | |
5219 | return NOTIFY_DONE; | |
3a101d05 TH |
5220 | } |
5221 | ||
7babe8db | 5222 | static int __init migration_init(void) |
1da177e4 LT |
5223 | { |
5224 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5225 | int err; |
48f24c4d | 5226 | |
3a101d05 | 5227 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5228 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5229 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5230 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5231 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5232 | |
3a101d05 TH |
5233 | /* Register cpu active notifiers */ |
5234 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5235 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5236 | ||
a004cd42 | 5237 | return 0; |
1da177e4 | 5238 | } |
7babe8db | 5239 | early_initcall(migration_init); |
1da177e4 LT |
5240 | #endif |
5241 | ||
5242 | #ifdef CONFIG_SMP | |
476f3534 | 5243 | |
4cb98839 PZ |
5244 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5245 | ||
3e9830dc | 5246 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5247 | |
d039ac60 | 5248 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5249 | |
d039ac60 | 5250 | static int __init sched_debug_setup(char *str) |
f6630114 | 5251 | { |
d039ac60 | 5252 | sched_debug_enabled = 1; |
f6630114 MT |
5253 | |
5254 | return 0; | |
5255 | } | |
d039ac60 PZ |
5256 | early_param("sched_debug", sched_debug_setup); |
5257 | ||
5258 | static inline bool sched_debug(void) | |
5259 | { | |
5260 | return sched_debug_enabled; | |
5261 | } | |
f6630114 | 5262 | |
7c16ec58 | 5263 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5264 | struct cpumask *groupmask) |
1da177e4 | 5265 | { |
4dcf6aff | 5266 | struct sched_group *group = sd->groups; |
434d53b0 | 5267 | char str[256]; |
1da177e4 | 5268 | |
968ea6d8 | 5269 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5270 | cpumask_clear(groupmask); |
4dcf6aff IM |
5271 | |
5272 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5273 | ||
5274 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5275 | printk("does not load-balance\n"); |
4dcf6aff | 5276 | if (sd->parent) |
3df0fc5b PZ |
5277 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5278 | " has parent"); | |
4dcf6aff | 5279 | return -1; |
41c7ce9a NP |
5280 | } |
5281 | ||
3df0fc5b | 5282 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5283 | |
758b2cdc | 5284 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5285 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5286 | "CPU%d\n", cpu); | |
4dcf6aff | 5287 | } |
758b2cdc | 5288 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5289 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5290 | " CPU%d\n", cpu); | |
4dcf6aff | 5291 | } |
1da177e4 | 5292 | |
4dcf6aff | 5293 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5294 | do { |
4dcf6aff | 5295 | if (!group) { |
3df0fc5b PZ |
5296 | printk("\n"); |
5297 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5298 | break; |
5299 | } | |
5300 | ||
c3decf0d | 5301 | /* |
63b2ca30 NP |
5302 | * Even though we initialize ->capacity to something semi-sane, |
5303 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5304 | * domain iteration is still funny without causing /0 traps. |
5305 | */ | |
63b2ca30 | 5306 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5307 | printk(KERN_CONT "\n"); |
63b2ca30 | 5308 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5309 | break; |
5310 | } | |
1da177e4 | 5311 | |
758b2cdc | 5312 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5313 | printk(KERN_CONT "\n"); |
5314 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5315 | break; |
5316 | } | |
1da177e4 | 5317 | |
cb83b629 PZ |
5318 | if (!(sd->flags & SD_OVERLAP) && |
5319 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5320 | printk(KERN_CONT "\n"); |
5321 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5322 | break; |
5323 | } | |
1da177e4 | 5324 | |
758b2cdc | 5325 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5326 | |
968ea6d8 | 5327 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5328 | |
3df0fc5b | 5329 | printk(KERN_CONT " %s", str); |
ca8ce3d0 | 5330 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5331 | printk(KERN_CONT " (cpu_capacity = %d)", |
5332 | group->sgc->capacity); | |
381512cf | 5333 | } |
1da177e4 | 5334 | |
4dcf6aff IM |
5335 | group = group->next; |
5336 | } while (group != sd->groups); | |
3df0fc5b | 5337 | printk(KERN_CONT "\n"); |
1da177e4 | 5338 | |
758b2cdc | 5339 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5340 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5341 | |
758b2cdc RR |
5342 | if (sd->parent && |
5343 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5344 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5345 | "of domain->span\n"); | |
4dcf6aff IM |
5346 | return 0; |
5347 | } | |
1da177e4 | 5348 | |
4dcf6aff IM |
5349 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5350 | { | |
5351 | int level = 0; | |
1da177e4 | 5352 | |
d039ac60 | 5353 | if (!sched_debug_enabled) |
f6630114 MT |
5354 | return; |
5355 | ||
4dcf6aff IM |
5356 | if (!sd) { |
5357 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5358 | return; | |
5359 | } | |
1da177e4 | 5360 | |
4dcf6aff IM |
5361 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5362 | ||
5363 | for (;;) { | |
4cb98839 | 5364 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5365 | break; |
1da177e4 LT |
5366 | level++; |
5367 | sd = sd->parent; | |
33859f7f | 5368 | if (!sd) |
4dcf6aff IM |
5369 | break; |
5370 | } | |
1da177e4 | 5371 | } |
6d6bc0ad | 5372 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5373 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5374 | static inline bool sched_debug(void) |
5375 | { | |
5376 | return false; | |
5377 | } | |
6d6bc0ad | 5378 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5379 | |
1a20ff27 | 5380 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5381 | { |
758b2cdc | 5382 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5383 | return 1; |
5384 | ||
5385 | /* Following flags need at least 2 groups */ | |
5386 | if (sd->flags & (SD_LOAD_BALANCE | | |
5387 | SD_BALANCE_NEWIDLE | | |
5388 | SD_BALANCE_FORK | | |
89c4710e | 5389 | SD_BALANCE_EXEC | |
5d4dfddd | 5390 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5391 | SD_SHARE_PKG_RESOURCES | |
5392 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5393 | if (sd->groups != sd->groups->next) |
5394 | return 0; | |
5395 | } | |
5396 | ||
5397 | /* Following flags don't use groups */ | |
c88d5910 | 5398 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5399 | return 0; |
5400 | ||
5401 | return 1; | |
5402 | } | |
5403 | ||
48f24c4d IM |
5404 | static int |
5405 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5406 | { |
5407 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5408 | ||
5409 | if (sd_degenerate(parent)) | |
5410 | return 1; | |
5411 | ||
758b2cdc | 5412 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5413 | return 0; |
5414 | ||
245af2c7 SS |
5415 | /* Flags needing groups don't count if only 1 group in parent */ |
5416 | if (parent->groups == parent->groups->next) { | |
5417 | pflags &= ~(SD_LOAD_BALANCE | | |
5418 | SD_BALANCE_NEWIDLE | | |
5419 | SD_BALANCE_FORK | | |
89c4710e | 5420 | SD_BALANCE_EXEC | |
5d4dfddd | 5421 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5422 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5423 | SD_PREFER_SIBLING | |
5424 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5425 | if (nr_node_ids == 1) |
5426 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5427 | } |
5428 | if (~cflags & pflags) | |
5429 | return 0; | |
5430 | ||
5431 | return 1; | |
5432 | } | |
5433 | ||
dce840a0 | 5434 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5435 | { |
dce840a0 | 5436 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5437 | |
68e74568 | 5438 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5439 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5440 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5441 | free_cpumask_var(rd->rto_mask); |
5442 | free_cpumask_var(rd->online); | |
5443 | free_cpumask_var(rd->span); | |
5444 | kfree(rd); | |
5445 | } | |
5446 | ||
57d885fe GH |
5447 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5448 | { | |
a0490fa3 | 5449 | struct root_domain *old_rd = NULL; |
57d885fe | 5450 | unsigned long flags; |
57d885fe | 5451 | |
05fa785c | 5452 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5453 | |
5454 | if (rq->rd) { | |
a0490fa3 | 5455 | old_rd = rq->rd; |
57d885fe | 5456 | |
c6c4927b | 5457 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5458 | set_rq_offline(rq); |
57d885fe | 5459 | |
c6c4927b | 5460 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5461 | |
a0490fa3 | 5462 | /* |
0515973f | 5463 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5464 | * set old_rd to NULL to skip the freeing later |
5465 | * in this function: | |
5466 | */ | |
5467 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5468 | old_rd = NULL; | |
57d885fe GH |
5469 | } |
5470 | ||
5471 | atomic_inc(&rd->refcount); | |
5472 | rq->rd = rd; | |
5473 | ||
c6c4927b | 5474 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5475 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5476 | set_rq_online(rq); |
57d885fe | 5477 | |
05fa785c | 5478 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5479 | |
5480 | if (old_rd) | |
dce840a0 | 5481 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5482 | } |
5483 | ||
68c38fc3 | 5484 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5485 | { |
5486 | memset(rd, 0, sizeof(*rd)); | |
5487 | ||
68c38fc3 | 5488 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5489 | goto out; |
68c38fc3 | 5490 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5491 | goto free_span; |
1baca4ce | 5492 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5493 | goto free_online; |
1baca4ce JL |
5494 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5495 | goto free_dlo_mask; | |
6e0534f2 | 5496 | |
332ac17e | 5497 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5498 | if (cpudl_init(&rd->cpudl) != 0) |
5499 | goto free_dlo_mask; | |
332ac17e | 5500 | |
68c38fc3 | 5501 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5502 | goto free_rto_mask; |
c6c4927b | 5503 | return 0; |
6e0534f2 | 5504 | |
68e74568 RR |
5505 | free_rto_mask: |
5506 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5507 | free_dlo_mask: |
5508 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5509 | free_online: |
5510 | free_cpumask_var(rd->online); | |
5511 | free_span: | |
5512 | free_cpumask_var(rd->span); | |
0c910d28 | 5513 | out: |
c6c4927b | 5514 | return -ENOMEM; |
57d885fe GH |
5515 | } |
5516 | ||
029632fb PZ |
5517 | /* |
5518 | * By default the system creates a single root-domain with all cpus as | |
5519 | * members (mimicking the global state we have today). | |
5520 | */ | |
5521 | struct root_domain def_root_domain; | |
5522 | ||
57d885fe GH |
5523 | static void init_defrootdomain(void) |
5524 | { | |
68c38fc3 | 5525 | init_rootdomain(&def_root_domain); |
c6c4927b | 5526 | |
57d885fe GH |
5527 | atomic_set(&def_root_domain.refcount, 1); |
5528 | } | |
5529 | ||
dc938520 | 5530 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5531 | { |
5532 | struct root_domain *rd; | |
5533 | ||
5534 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5535 | if (!rd) | |
5536 | return NULL; | |
5537 | ||
68c38fc3 | 5538 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5539 | kfree(rd); |
5540 | return NULL; | |
5541 | } | |
57d885fe GH |
5542 | |
5543 | return rd; | |
5544 | } | |
5545 | ||
63b2ca30 | 5546 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5547 | { |
5548 | struct sched_group *tmp, *first; | |
5549 | ||
5550 | if (!sg) | |
5551 | return; | |
5552 | ||
5553 | first = sg; | |
5554 | do { | |
5555 | tmp = sg->next; | |
5556 | ||
63b2ca30 NP |
5557 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5558 | kfree(sg->sgc); | |
e3589f6c PZ |
5559 | |
5560 | kfree(sg); | |
5561 | sg = tmp; | |
5562 | } while (sg != first); | |
5563 | } | |
5564 | ||
dce840a0 PZ |
5565 | static void free_sched_domain(struct rcu_head *rcu) |
5566 | { | |
5567 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5568 | |
5569 | /* | |
5570 | * If its an overlapping domain it has private groups, iterate and | |
5571 | * nuke them all. | |
5572 | */ | |
5573 | if (sd->flags & SD_OVERLAP) { | |
5574 | free_sched_groups(sd->groups, 1); | |
5575 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5576 | kfree(sd->groups->sgc); |
dce840a0 | 5577 | kfree(sd->groups); |
9c3f75cb | 5578 | } |
dce840a0 PZ |
5579 | kfree(sd); |
5580 | } | |
5581 | ||
5582 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5583 | { | |
5584 | call_rcu(&sd->rcu, free_sched_domain); | |
5585 | } | |
5586 | ||
5587 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5588 | { | |
5589 | for (; sd; sd = sd->parent) | |
5590 | destroy_sched_domain(sd, cpu); | |
5591 | } | |
5592 | ||
518cd623 PZ |
5593 | /* |
5594 | * Keep a special pointer to the highest sched_domain that has | |
5595 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5596 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5597 | * | |
5598 | * Also keep a unique ID per domain (we use the first cpu number in | |
5599 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5600 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5601 | */ |
5602 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5603 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5604 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5605 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5606 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5607 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5608 | |
5609 | static void update_top_cache_domain(int cpu) | |
5610 | { | |
5611 | struct sched_domain *sd; | |
5d4cf996 | 5612 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5613 | int id = cpu; |
7d9ffa89 | 5614 | int size = 1; |
518cd623 PZ |
5615 | |
5616 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5617 | if (sd) { |
518cd623 | 5618 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5619 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5620 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5621 | } |
5d4cf996 | 5622 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5623 | |
5624 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5625 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5626 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5627 | |
5628 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5629 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5630 | |
5631 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5632 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5633 | } |
5634 | ||
1da177e4 | 5635 | /* |
0eab9146 | 5636 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5637 | * hold the hotplug lock. |
5638 | */ | |
0eab9146 IM |
5639 | static void |
5640 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5641 | { |
70b97a7f | 5642 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5643 | struct sched_domain *tmp; |
5644 | ||
5645 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5646 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5647 | struct sched_domain *parent = tmp->parent; |
5648 | if (!parent) | |
5649 | break; | |
f29c9b1c | 5650 | |
1a848870 | 5651 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5652 | tmp->parent = parent->parent; |
1a848870 SS |
5653 | if (parent->parent) |
5654 | parent->parent->child = tmp; | |
10866e62 PZ |
5655 | /* |
5656 | * Transfer SD_PREFER_SIBLING down in case of a | |
5657 | * degenerate parent; the spans match for this | |
5658 | * so the property transfers. | |
5659 | */ | |
5660 | if (parent->flags & SD_PREFER_SIBLING) | |
5661 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5662 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5663 | } else |
5664 | tmp = tmp->parent; | |
245af2c7 SS |
5665 | } |
5666 | ||
1a848870 | 5667 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5668 | tmp = sd; |
245af2c7 | 5669 | sd = sd->parent; |
dce840a0 | 5670 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5671 | if (sd) |
5672 | sd->child = NULL; | |
5673 | } | |
1da177e4 | 5674 | |
4cb98839 | 5675 | sched_domain_debug(sd, cpu); |
1da177e4 | 5676 | |
57d885fe | 5677 | rq_attach_root(rq, rd); |
dce840a0 | 5678 | tmp = rq->sd; |
674311d5 | 5679 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5680 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5681 | |
5682 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5683 | } |
5684 | ||
5685 | /* cpus with isolated domains */ | |
dcc30a35 | 5686 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5687 | |
5688 | /* Setup the mask of cpus configured for isolated domains */ | |
5689 | static int __init isolated_cpu_setup(char *str) | |
5690 | { | |
bdddd296 | 5691 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5692 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5693 | return 1; |
5694 | } | |
5695 | ||
8927f494 | 5696 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5697 | |
49a02c51 | 5698 | struct s_data { |
21d42ccf | 5699 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5700 | struct root_domain *rd; |
5701 | }; | |
5702 | ||
2109b99e | 5703 | enum s_alloc { |
2109b99e | 5704 | sa_rootdomain, |
21d42ccf | 5705 | sa_sd, |
dce840a0 | 5706 | sa_sd_storage, |
2109b99e AH |
5707 | sa_none, |
5708 | }; | |
5709 | ||
c1174876 PZ |
5710 | /* |
5711 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5712 | * domain traversal. | |
5713 | * | |
5714 | * Asymmetric node setups can result in situations where the domain tree is of | |
5715 | * unequal depth, make sure to skip domains that already cover the entire | |
5716 | * range. | |
5717 | * | |
5718 | * In that case build_sched_domains() will have terminated the iteration early | |
5719 | * and our sibling sd spans will be empty. Domains should always include the | |
5720 | * cpu they're built on, so check that. | |
5721 | * | |
5722 | */ | |
5723 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5724 | { | |
5725 | const struct cpumask *span = sched_domain_span(sd); | |
5726 | struct sd_data *sdd = sd->private; | |
5727 | struct sched_domain *sibling; | |
5728 | int i; | |
5729 | ||
5730 | for_each_cpu(i, span) { | |
5731 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5732 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5733 | continue; | |
5734 | ||
5735 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5736 | } | |
5737 | } | |
5738 | ||
5739 | /* | |
5740 | * Return the canonical balance cpu for this group, this is the first cpu | |
5741 | * of this group that's also in the iteration mask. | |
5742 | */ | |
5743 | int group_balance_cpu(struct sched_group *sg) | |
5744 | { | |
5745 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5746 | } | |
5747 | ||
e3589f6c PZ |
5748 | static int |
5749 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5750 | { | |
5751 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5752 | const struct cpumask *span = sched_domain_span(sd); | |
5753 | struct cpumask *covered = sched_domains_tmpmask; | |
5754 | struct sd_data *sdd = sd->private; | |
aaecac4a | 5755 | struct sched_domain *sibling; |
e3589f6c PZ |
5756 | int i; |
5757 | ||
5758 | cpumask_clear(covered); | |
5759 | ||
5760 | for_each_cpu(i, span) { | |
5761 | struct cpumask *sg_span; | |
5762 | ||
5763 | if (cpumask_test_cpu(i, covered)) | |
5764 | continue; | |
5765 | ||
aaecac4a | 5766 | sibling = *per_cpu_ptr(sdd->sd, i); |
c1174876 PZ |
5767 | |
5768 | /* See the comment near build_group_mask(). */ | |
aaecac4a | 5769 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
c1174876 PZ |
5770 | continue; |
5771 | ||
e3589f6c | 5772 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5773 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5774 | |
5775 | if (!sg) | |
5776 | goto fail; | |
5777 | ||
5778 | sg_span = sched_group_cpus(sg); | |
aaecac4a ZZ |
5779 | if (sibling->child) |
5780 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); | |
5781 | else | |
e3589f6c PZ |
5782 | cpumask_set_cpu(i, sg_span); |
5783 | ||
5784 | cpumask_or(covered, covered, sg_span); | |
5785 | ||
63b2ca30 NP |
5786 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5787 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5788 | build_group_mask(sd, sg); |
5789 | ||
c3decf0d | 5790 | /* |
63b2ca30 | 5791 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5792 | * domains and no possible iteration will get us here, we won't |
5793 | * die on a /0 trap. | |
5794 | */ | |
ca8ce3d0 | 5795 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
63b2ca30 | 5796 | sg->sgc->capacity_orig = sg->sgc->capacity; |
e3589f6c | 5797 | |
c1174876 PZ |
5798 | /* |
5799 | * Make sure the first group of this domain contains the | |
5800 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5801 | * breaks. See update_sg_lb_stats(). | |
5802 | */ | |
74a5ce20 | 5803 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5804 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5805 | groups = sg; |
5806 | ||
5807 | if (!first) | |
5808 | first = sg; | |
5809 | if (last) | |
5810 | last->next = sg; | |
5811 | last = sg; | |
5812 | last->next = first; | |
5813 | } | |
5814 | sd->groups = groups; | |
5815 | ||
5816 | return 0; | |
5817 | ||
5818 | fail: | |
5819 | free_sched_groups(first, 0); | |
5820 | ||
5821 | return -ENOMEM; | |
5822 | } | |
5823 | ||
dce840a0 | 5824 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5825 | { |
dce840a0 PZ |
5826 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5827 | struct sched_domain *child = sd->child; | |
1da177e4 | 5828 | |
dce840a0 PZ |
5829 | if (child) |
5830 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5831 | |
9c3f75cb | 5832 | if (sg) { |
dce840a0 | 5833 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5834 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5835 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5836 | } |
dce840a0 PZ |
5837 | |
5838 | return cpu; | |
1e9f28fa | 5839 | } |
1e9f28fa | 5840 | |
01a08546 | 5841 | /* |
dce840a0 PZ |
5842 | * build_sched_groups will build a circular linked list of the groups |
5843 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5844 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5845 | * |
5846 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5847 | */ |
e3589f6c PZ |
5848 | static int |
5849 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5850 | { |
dce840a0 PZ |
5851 | struct sched_group *first = NULL, *last = NULL; |
5852 | struct sd_data *sdd = sd->private; | |
5853 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5854 | struct cpumask *covered; |
dce840a0 | 5855 | int i; |
9c1cfda2 | 5856 | |
e3589f6c PZ |
5857 | get_group(cpu, sdd, &sd->groups); |
5858 | atomic_inc(&sd->groups->ref); | |
5859 | ||
0936629f | 5860 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5861 | return 0; |
5862 | ||
f96225fd PZ |
5863 | lockdep_assert_held(&sched_domains_mutex); |
5864 | covered = sched_domains_tmpmask; | |
5865 | ||
dce840a0 | 5866 | cpumask_clear(covered); |
6711cab4 | 5867 | |
dce840a0 PZ |
5868 | for_each_cpu(i, span) { |
5869 | struct sched_group *sg; | |
cd08e923 | 5870 | int group, j; |
6711cab4 | 5871 | |
dce840a0 PZ |
5872 | if (cpumask_test_cpu(i, covered)) |
5873 | continue; | |
6711cab4 | 5874 | |
cd08e923 | 5875 | group = get_group(i, sdd, &sg); |
c1174876 | 5876 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5877 | |
dce840a0 PZ |
5878 | for_each_cpu(j, span) { |
5879 | if (get_group(j, sdd, NULL) != group) | |
5880 | continue; | |
0601a88d | 5881 | |
dce840a0 PZ |
5882 | cpumask_set_cpu(j, covered); |
5883 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5884 | } | |
0601a88d | 5885 | |
dce840a0 PZ |
5886 | if (!first) |
5887 | first = sg; | |
5888 | if (last) | |
5889 | last->next = sg; | |
5890 | last = sg; | |
5891 | } | |
5892 | last->next = first; | |
e3589f6c PZ |
5893 | |
5894 | return 0; | |
0601a88d | 5895 | } |
51888ca2 | 5896 | |
89c4710e | 5897 | /* |
63b2ca30 | 5898 | * Initialize sched groups cpu_capacity. |
89c4710e | 5899 | * |
63b2ca30 | 5900 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 5901 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
5902 | * Typically cpu_capacity for all the groups in a sched domain will be same |
5903 | * unless there are asymmetries in the topology. If there are asymmetries, | |
5904 | * group having more cpu_capacity will pickup more load compared to the | |
5905 | * group having less cpu_capacity. | |
89c4710e | 5906 | */ |
63b2ca30 | 5907 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 5908 | { |
e3589f6c | 5909 | struct sched_group *sg = sd->groups; |
89c4710e | 5910 | |
94c95ba6 | 5911 | WARN_ON(!sg); |
e3589f6c PZ |
5912 | |
5913 | do { | |
5914 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5915 | sg = sg->next; | |
5916 | } while (sg != sd->groups); | |
89c4710e | 5917 | |
c1174876 | 5918 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5919 | return; |
aae6d3dd | 5920 | |
63b2ca30 NP |
5921 | update_group_capacity(sd, cpu); |
5922 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
5923 | } |
5924 | ||
7c16ec58 MT |
5925 | /* |
5926 | * Initializers for schedule domains | |
5927 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5928 | */ | |
5929 | ||
1d3504fc | 5930 | static int default_relax_domain_level = -1; |
60495e77 | 5931 | int sched_domain_level_max; |
1d3504fc HS |
5932 | |
5933 | static int __init setup_relax_domain_level(char *str) | |
5934 | { | |
a841f8ce DS |
5935 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5936 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5937 | |
1d3504fc HS |
5938 | return 1; |
5939 | } | |
5940 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5941 | ||
5942 | static void set_domain_attribute(struct sched_domain *sd, | |
5943 | struct sched_domain_attr *attr) | |
5944 | { | |
5945 | int request; | |
5946 | ||
5947 | if (!attr || attr->relax_domain_level < 0) { | |
5948 | if (default_relax_domain_level < 0) | |
5949 | return; | |
5950 | else | |
5951 | request = default_relax_domain_level; | |
5952 | } else | |
5953 | request = attr->relax_domain_level; | |
5954 | if (request < sd->level) { | |
5955 | /* turn off idle balance on this domain */ | |
c88d5910 | 5956 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5957 | } else { |
5958 | /* turn on idle balance on this domain */ | |
c88d5910 | 5959 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5960 | } |
5961 | } | |
5962 | ||
54ab4ff4 PZ |
5963 | static void __sdt_free(const struct cpumask *cpu_map); |
5964 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5965 | ||
2109b99e AH |
5966 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5967 | const struct cpumask *cpu_map) | |
5968 | { | |
5969 | switch (what) { | |
2109b99e | 5970 | case sa_rootdomain: |
822ff793 PZ |
5971 | if (!atomic_read(&d->rd->refcount)) |
5972 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5973 | case sa_sd: |
5974 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5975 | case sa_sd_storage: |
54ab4ff4 | 5976 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5977 | case sa_none: |
5978 | break; | |
5979 | } | |
5980 | } | |
3404c8d9 | 5981 | |
2109b99e AH |
5982 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5983 | const struct cpumask *cpu_map) | |
5984 | { | |
dce840a0 PZ |
5985 | memset(d, 0, sizeof(*d)); |
5986 | ||
54ab4ff4 PZ |
5987 | if (__sdt_alloc(cpu_map)) |
5988 | return sa_sd_storage; | |
dce840a0 PZ |
5989 | d->sd = alloc_percpu(struct sched_domain *); |
5990 | if (!d->sd) | |
5991 | return sa_sd_storage; | |
2109b99e | 5992 | d->rd = alloc_rootdomain(); |
dce840a0 | 5993 | if (!d->rd) |
21d42ccf | 5994 | return sa_sd; |
2109b99e AH |
5995 | return sa_rootdomain; |
5996 | } | |
57d885fe | 5997 | |
dce840a0 PZ |
5998 | /* |
5999 | * NULL the sd_data elements we've used to build the sched_domain and | |
6000 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6001 | * will not free the data we're using. | |
6002 | */ | |
6003 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6004 | { | |
6005 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6006 | |
6007 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6008 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6009 | ||
e3589f6c | 6010 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6011 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6012 | |
63b2ca30 NP |
6013 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6014 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6015 | } |
6016 | ||
cb83b629 | 6017 | #ifdef CONFIG_NUMA |
cb83b629 | 6018 | static int sched_domains_numa_levels; |
cb83b629 PZ |
6019 | static int *sched_domains_numa_distance; |
6020 | static struct cpumask ***sched_domains_numa_masks; | |
6021 | static int sched_domains_curr_level; | |
143e1e28 | 6022 | #endif |
cb83b629 | 6023 | |
143e1e28 VG |
6024 | /* |
6025 | * SD_flags allowed in topology descriptions. | |
6026 | * | |
5d4dfddd | 6027 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6028 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6029 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6030 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6031 | * |
6032 | * Odd one out: | |
6033 | * SD_ASYM_PACKING - describes SMT quirks | |
6034 | */ | |
6035 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6036 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6037 | SD_SHARE_PKG_RESOURCES | \ |
6038 | SD_NUMA | \ | |
d77b3ed5 VG |
6039 | SD_ASYM_PACKING | \ |
6040 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6041 | |
6042 | static struct sched_domain * | |
143e1e28 | 6043 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6044 | { |
6045 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6046 | int sd_weight, sd_flags = 0; |
6047 | ||
6048 | #ifdef CONFIG_NUMA | |
6049 | /* | |
6050 | * Ugly hack to pass state to sd_numa_mask()... | |
6051 | */ | |
6052 | sched_domains_curr_level = tl->numa_level; | |
6053 | #endif | |
6054 | ||
6055 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6056 | ||
6057 | if (tl->sd_flags) | |
6058 | sd_flags = (*tl->sd_flags)(); | |
6059 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6060 | "wrong sd_flags in topology description\n")) | |
6061 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6062 | |
6063 | *sd = (struct sched_domain){ | |
6064 | .min_interval = sd_weight, | |
6065 | .max_interval = 2*sd_weight, | |
6066 | .busy_factor = 32, | |
870a0bb5 | 6067 | .imbalance_pct = 125, |
143e1e28 VG |
6068 | |
6069 | .cache_nice_tries = 0, | |
6070 | .busy_idx = 0, | |
6071 | .idle_idx = 0, | |
cb83b629 PZ |
6072 | .newidle_idx = 0, |
6073 | .wake_idx = 0, | |
6074 | .forkexec_idx = 0, | |
6075 | ||
6076 | .flags = 1*SD_LOAD_BALANCE | |
6077 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6078 | | 1*SD_BALANCE_EXEC |
6079 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6080 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6081 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6082 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6083 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6084 | | 0*SD_SERIALIZE |
cb83b629 | 6085 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6086 | | 0*SD_NUMA |
6087 | | sd_flags | |
cb83b629 | 6088 | , |
143e1e28 | 6089 | |
cb83b629 PZ |
6090 | .last_balance = jiffies, |
6091 | .balance_interval = sd_weight, | |
143e1e28 | 6092 | .smt_gain = 0, |
2b4cfe64 JL |
6093 | .max_newidle_lb_cost = 0, |
6094 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6095 | #ifdef CONFIG_SCHED_DEBUG |
6096 | .name = tl->name, | |
6097 | #endif | |
cb83b629 | 6098 | }; |
cb83b629 PZ |
6099 | |
6100 | /* | |
143e1e28 | 6101 | * Convert topological properties into behaviour. |
cb83b629 | 6102 | */ |
143e1e28 | 6103 | |
5d4dfddd | 6104 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
143e1e28 VG |
6105 | sd->imbalance_pct = 110; |
6106 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6107 | |
6108 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6109 | sd->imbalance_pct = 117; | |
6110 | sd->cache_nice_tries = 1; | |
6111 | sd->busy_idx = 2; | |
6112 | ||
6113 | #ifdef CONFIG_NUMA | |
6114 | } else if (sd->flags & SD_NUMA) { | |
6115 | sd->cache_nice_tries = 2; | |
6116 | sd->busy_idx = 3; | |
6117 | sd->idle_idx = 2; | |
6118 | ||
6119 | sd->flags |= SD_SERIALIZE; | |
6120 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6121 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6122 | SD_BALANCE_FORK | | |
6123 | SD_WAKE_AFFINE); | |
6124 | } | |
6125 | ||
6126 | #endif | |
6127 | } else { | |
6128 | sd->flags |= SD_PREFER_SIBLING; | |
6129 | sd->cache_nice_tries = 1; | |
6130 | sd->busy_idx = 2; | |
6131 | sd->idle_idx = 1; | |
6132 | } | |
6133 | ||
6134 | sd->private = &tl->data; | |
cb83b629 PZ |
6135 | |
6136 | return sd; | |
6137 | } | |
6138 | ||
143e1e28 VG |
6139 | /* |
6140 | * Topology list, bottom-up. | |
6141 | */ | |
6142 | static struct sched_domain_topology_level default_topology[] = { | |
6143 | #ifdef CONFIG_SCHED_SMT | |
6144 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6145 | #endif | |
6146 | #ifdef CONFIG_SCHED_MC | |
6147 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6148 | #endif |
6149 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6150 | { NULL, }, | |
6151 | }; | |
6152 | ||
6153 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6154 | ||
6155 | #define for_each_sd_topology(tl) \ | |
6156 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6157 | ||
6158 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6159 | { | |
6160 | sched_domain_topology = tl; | |
6161 | } | |
6162 | ||
6163 | #ifdef CONFIG_NUMA | |
6164 | ||
cb83b629 PZ |
6165 | static const struct cpumask *sd_numa_mask(int cpu) |
6166 | { | |
6167 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6168 | } | |
6169 | ||
d039ac60 PZ |
6170 | static void sched_numa_warn(const char *str) |
6171 | { | |
6172 | static int done = false; | |
6173 | int i,j; | |
6174 | ||
6175 | if (done) | |
6176 | return; | |
6177 | ||
6178 | done = true; | |
6179 | ||
6180 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6181 | ||
6182 | for (i = 0; i < nr_node_ids; i++) { | |
6183 | printk(KERN_WARNING " "); | |
6184 | for (j = 0; j < nr_node_ids; j++) | |
6185 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6186 | printk(KERN_CONT "\n"); | |
6187 | } | |
6188 | printk(KERN_WARNING "\n"); | |
6189 | } | |
6190 | ||
6191 | static bool find_numa_distance(int distance) | |
6192 | { | |
6193 | int i; | |
6194 | ||
6195 | if (distance == node_distance(0, 0)) | |
6196 | return true; | |
6197 | ||
6198 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6199 | if (sched_domains_numa_distance[i] == distance) | |
6200 | return true; | |
6201 | } | |
6202 | ||
6203 | return false; | |
6204 | } | |
6205 | ||
cb83b629 PZ |
6206 | static void sched_init_numa(void) |
6207 | { | |
6208 | int next_distance, curr_distance = node_distance(0, 0); | |
6209 | struct sched_domain_topology_level *tl; | |
6210 | int level = 0; | |
6211 | int i, j, k; | |
6212 | ||
cb83b629 PZ |
6213 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6214 | if (!sched_domains_numa_distance) | |
6215 | return; | |
6216 | ||
6217 | /* | |
6218 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6219 | * unique distances in the node_distance() table. | |
6220 | * | |
6221 | * Assumes node_distance(0,j) includes all distances in | |
6222 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6223 | */ |
6224 | next_distance = curr_distance; | |
6225 | for (i = 0; i < nr_node_ids; i++) { | |
6226 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6227 | for (k = 0; k < nr_node_ids; k++) { |
6228 | int distance = node_distance(i, k); | |
6229 | ||
6230 | if (distance > curr_distance && | |
6231 | (distance < next_distance || | |
6232 | next_distance == curr_distance)) | |
6233 | next_distance = distance; | |
6234 | ||
6235 | /* | |
6236 | * While not a strong assumption it would be nice to know | |
6237 | * about cases where if node A is connected to B, B is not | |
6238 | * equally connected to A. | |
6239 | */ | |
6240 | if (sched_debug() && node_distance(k, i) != distance) | |
6241 | sched_numa_warn("Node-distance not symmetric"); | |
6242 | ||
6243 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6244 | sched_numa_warn("Node-0 not representative"); | |
6245 | } | |
6246 | if (next_distance != curr_distance) { | |
6247 | sched_domains_numa_distance[level++] = next_distance; | |
6248 | sched_domains_numa_levels = level; | |
6249 | curr_distance = next_distance; | |
6250 | } else break; | |
cb83b629 | 6251 | } |
d039ac60 PZ |
6252 | |
6253 | /* | |
6254 | * In case of sched_debug() we verify the above assumption. | |
6255 | */ | |
6256 | if (!sched_debug()) | |
6257 | break; | |
cb83b629 PZ |
6258 | } |
6259 | /* | |
6260 | * 'level' contains the number of unique distances, excluding the | |
6261 | * identity distance node_distance(i,i). | |
6262 | * | |
28b4a521 | 6263 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6264 | * numbers. |
6265 | */ | |
6266 | ||
5f7865f3 TC |
6267 | /* |
6268 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6269 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6270 | * the array will contain less then 'level' members. This could be | |
6271 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6272 | * in other functions. | |
6273 | * | |
6274 | * We reset it to 'level' at the end of this function. | |
6275 | */ | |
6276 | sched_domains_numa_levels = 0; | |
6277 | ||
cb83b629 PZ |
6278 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6279 | if (!sched_domains_numa_masks) | |
6280 | return; | |
6281 | ||
6282 | /* | |
6283 | * Now for each level, construct a mask per node which contains all | |
6284 | * cpus of nodes that are that many hops away from us. | |
6285 | */ | |
6286 | for (i = 0; i < level; i++) { | |
6287 | sched_domains_numa_masks[i] = | |
6288 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6289 | if (!sched_domains_numa_masks[i]) | |
6290 | return; | |
6291 | ||
6292 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6293 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6294 | if (!mask) |
6295 | return; | |
6296 | ||
6297 | sched_domains_numa_masks[i][j] = mask; | |
6298 | ||
6299 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6300 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6301 | continue; |
6302 | ||
6303 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6304 | } | |
6305 | } | |
6306 | } | |
6307 | ||
143e1e28 VG |
6308 | /* Compute default topology size */ |
6309 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6310 | ||
c515db8c | 6311 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6312 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6313 | if (!tl) | |
6314 | return; | |
6315 | ||
6316 | /* | |
6317 | * Copy the default topology bits.. | |
6318 | */ | |
143e1e28 VG |
6319 | for (i = 0; sched_domain_topology[i].mask; i++) |
6320 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6321 | |
6322 | /* | |
6323 | * .. and append 'j' levels of NUMA goodness. | |
6324 | */ | |
6325 | for (j = 0; j < level; i++, j++) { | |
6326 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6327 | .mask = sd_numa_mask, |
143e1e28 | 6328 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6329 | .flags = SDTL_OVERLAP, |
6330 | .numa_level = j, | |
143e1e28 | 6331 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6332 | }; |
6333 | } | |
6334 | ||
6335 | sched_domain_topology = tl; | |
5f7865f3 TC |
6336 | |
6337 | sched_domains_numa_levels = level; | |
cb83b629 | 6338 | } |
301a5cba TC |
6339 | |
6340 | static void sched_domains_numa_masks_set(int cpu) | |
6341 | { | |
6342 | int i, j; | |
6343 | int node = cpu_to_node(cpu); | |
6344 | ||
6345 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6346 | for (j = 0; j < nr_node_ids; j++) { | |
6347 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6348 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6349 | } | |
6350 | } | |
6351 | } | |
6352 | ||
6353 | static void sched_domains_numa_masks_clear(int cpu) | |
6354 | { | |
6355 | int i, j; | |
6356 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6357 | for (j = 0; j < nr_node_ids; j++) | |
6358 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6359 | } | |
6360 | } | |
6361 | ||
6362 | /* | |
6363 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6364 | * are onlined. | |
6365 | */ | |
6366 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6367 | unsigned long action, | |
6368 | void *hcpu) | |
6369 | { | |
6370 | int cpu = (long)hcpu; | |
6371 | ||
6372 | switch (action & ~CPU_TASKS_FROZEN) { | |
6373 | case CPU_ONLINE: | |
6374 | sched_domains_numa_masks_set(cpu); | |
6375 | break; | |
6376 | ||
6377 | case CPU_DEAD: | |
6378 | sched_domains_numa_masks_clear(cpu); | |
6379 | break; | |
6380 | ||
6381 | default: | |
6382 | return NOTIFY_DONE; | |
6383 | } | |
6384 | ||
6385 | return NOTIFY_OK; | |
cb83b629 PZ |
6386 | } |
6387 | #else | |
6388 | static inline void sched_init_numa(void) | |
6389 | { | |
6390 | } | |
301a5cba TC |
6391 | |
6392 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6393 | unsigned long action, | |
6394 | void *hcpu) | |
6395 | { | |
6396 | return 0; | |
6397 | } | |
cb83b629 PZ |
6398 | #endif /* CONFIG_NUMA */ |
6399 | ||
54ab4ff4 PZ |
6400 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6401 | { | |
6402 | struct sched_domain_topology_level *tl; | |
6403 | int j; | |
6404 | ||
27723a68 | 6405 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6406 | struct sd_data *sdd = &tl->data; |
6407 | ||
6408 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6409 | if (!sdd->sd) | |
6410 | return -ENOMEM; | |
6411 | ||
6412 | sdd->sg = alloc_percpu(struct sched_group *); | |
6413 | if (!sdd->sg) | |
6414 | return -ENOMEM; | |
6415 | ||
63b2ca30 NP |
6416 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6417 | if (!sdd->sgc) | |
9c3f75cb PZ |
6418 | return -ENOMEM; |
6419 | ||
54ab4ff4 PZ |
6420 | for_each_cpu(j, cpu_map) { |
6421 | struct sched_domain *sd; | |
6422 | struct sched_group *sg; | |
63b2ca30 | 6423 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6424 | |
6425 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6426 | GFP_KERNEL, cpu_to_node(j)); | |
6427 | if (!sd) | |
6428 | return -ENOMEM; | |
6429 | ||
6430 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6431 | ||
6432 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6433 | GFP_KERNEL, cpu_to_node(j)); | |
6434 | if (!sg) | |
6435 | return -ENOMEM; | |
6436 | ||
30b4e9eb IM |
6437 | sg->next = sg; |
6438 | ||
54ab4ff4 | 6439 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6440 | |
63b2ca30 | 6441 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6442 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6443 | if (!sgc) |
9c3f75cb PZ |
6444 | return -ENOMEM; |
6445 | ||
63b2ca30 | 6446 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6447 | } |
6448 | } | |
6449 | ||
6450 | return 0; | |
6451 | } | |
6452 | ||
6453 | static void __sdt_free(const struct cpumask *cpu_map) | |
6454 | { | |
6455 | struct sched_domain_topology_level *tl; | |
6456 | int j; | |
6457 | ||
27723a68 | 6458 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6459 | struct sd_data *sdd = &tl->data; |
6460 | ||
6461 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6462 | struct sched_domain *sd; |
6463 | ||
6464 | if (sdd->sd) { | |
6465 | sd = *per_cpu_ptr(sdd->sd, j); | |
6466 | if (sd && (sd->flags & SD_OVERLAP)) | |
6467 | free_sched_groups(sd->groups, 0); | |
6468 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6469 | } | |
6470 | ||
6471 | if (sdd->sg) | |
6472 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6473 | if (sdd->sgc) |
6474 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6475 | } |
6476 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6477 | sdd->sd = NULL; |
54ab4ff4 | 6478 | free_percpu(sdd->sg); |
fb2cf2c6 | 6479 | sdd->sg = NULL; |
63b2ca30 NP |
6480 | free_percpu(sdd->sgc); |
6481 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6482 | } |
6483 | } | |
6484 | ||
2c402dc3 | 6485 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6486 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6487 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6488 | { |
143e1e28 | 6489 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6490 | if (!sd) |
d069b916 | 6491 | return child; |
2c402dc3 | 6492 | |
2c402dc3 | 6493 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6494 | if (child) { |
6495 | sd->level = child->level + 1; | |
6496 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6497 | child->parent = sd; |
c75e0128 | 6498 | sd->child = child; |
6ae72dff PZ |
6499 | |
6500 | if (!cpumask_subset(sched_domain_span(child), | |
6501 | sched_domain_span(sd))) { | |
6502 | pr_err("BUG: arch topology borken\n"); | |
6503 | #ifdef CONFIG_SCHED_DEBUG | |
6504 | pr_err(" the %s domain not a subset of the %s domain\n", | |
6505 | child->name, sd->name); | |
6506 | #endif | |
6507 | /* Fixup, ensure @sd has at least @child cpus. */ | |
6508 | cpumask_or(sched_domain_span(sd), | |
6509 | sched_domain_span(sd), | |
6510 | sched_domain_span(child)); | |
6511 | } | |
6512 | ||
60495e77 | 6513 | } |
a841f8ce | 6514 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6515 | |
6516 | return sd; | |
6517 | } | |
6518 | ||
2109b99e AH |
6519 | /* |
6520 | * Build sched domains for a given set of cpus and attach the sched domains | |
6521 | * to the individual cpus | |
6522 | */ | |
dce840a0 PZ |
6523 | static int build_sched_domains(const struct cpumask *cpu_map, |
6524 | struct sched_domain_attr *attr) | |
2109b99e | 6525 | { |
1c632169 | 6526 | enum s_alloc alloc_state; |
dce840a0 | 6527 | struct sched_domain *sd; |
2109b99e | 6528 | struct s_data d; |
822ff793 | 6529 | int i, ret = -ENOMEM; |
9c1cfda2 | 6530 | |
2109b99e AH |
6531 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6532 | if (alloc_state != sa_rootdomain) | |
6533 | goto error; | |
9c1cfda2 | 6534 | |
dce840a0 | 6535 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6536 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6537 | struct sched_domain_topology_level *tl; |
6538 | ||
3bd65a80 | 6539 | sd = NULL; |
27723a68 | 6540 | for_each_sd_topology(tl) { |
4a850cbe | 6541 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6542 | if (tl == sched_domain_topology) |
6543 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6544 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6545 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6546 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6547 | break; | |
e3589f6c | 6548 | } |
dce840a0 PZ |
6549 | } |
6550 | ||
6551 | /* Build the groups for the domains */ | |
6552 | for_each_cpu(i, cpu_map) { | |
6553 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6554 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6555 | if (sd->flags & SD_OVERLAP) { |
6556 | if (build_overlap_sched_groups(sd, i)) | |
6557 | goto error; | |
6558 | } else { | |
6559 | if (build_sched_groups(sd, i)) | |
6560 | goto error; | |
6561 | } | |
1cf51902 | 6562 | } |
a06dadbe | 6563 | } |
9c1cfda2 | 6564 | |
ced549fa | 6565 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6566 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6567 | if (!cpumask_test_cpu(i, cpu_map)) | |
6568 | continue; | |
9c1cfda2 | 6569 | |
dce840a0 PZ |
6570 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6571 | claim_allocations(i, sd); | |
63b2ca30 | 6572 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6573 | } |
f712c0c7 | 6574 | } |
9c1cfda2 | 6575 | |
1da177e4 | 6576 | /* Attach the domains */ |
dce840a0 | 6577 | rcu_read_lock(); |
abcd083a | 6578 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6579 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6580 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6581 | } |
dce840a0 | 6582 | rcu_read_unlock(); |
51888ca2 | 6583 | |
822ff793 | 6584 | ret = 0; |
51888ca2 | 6585 | error: |
2109b99e | 6586 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6587 | return ret; |
1da177e4 | 6588 | } |
029190c5 | 6589 | |
acc3f5d7 | 6590 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6591 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6592 | static struct sched_domain_attr *dattr_cur; |
6593 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6594 | |
6595 | /* | |
6596 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6597 | * cpumask) fails, then fallback to a single sched domain, |
6598 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6599 | */ |
4212823f | 6600 | static cpumask_var_t fallback_doms; |
029190c5 | 6601 | |
ee79d1bd HC |
6602 | /* |
6603 | * arch_update_cpu_topology lets virtualized architectures update the | |
6604 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6605 | * or 0 if it stayed the same. | |
6606 | */ | |
52f5684c | 6607 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6608 | { |
ee79d1bd | 6609 | return 0; |
22e52b07 HC |
6610 | } |
6611 | ||
acc3f5d7 RR |
6612 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6613 | { | |
6614 | int i; | |
6615 | cpumask_var_t *doms; | |
6616 | ||
6617 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6618 | if (!doms) | |
6619 | return NULL; | |
6620 | for (i = 0; i < ndoms; i++) { | |
6621 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6622 | free_sched_domains(doms, i); | |
6623 | return NULL; | |
6624 | } | |
6625 | } | |
6626 | return doms; | |
6627 | } | |
6628 | ||
6629 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6630 | { | |
6631 | unsigned int i; | |
6632 | for (i = 0; i < ndoms; i++) | |
6633 | free_cpumask_var(doms[i]); | |
6634 | kfree(doms); | |
6635 | } | |
6636 | ||
1a20ff27 | 6637 | /* |
41a2d6cf | 6638 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6639 | * For now this just excludes isolated cpus, but could be used to |
6640 | * exclude other special cases in the future. | |
1a20ff27 | 6641 | */ |
c4a8849a | 6642 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6643 | { |
7378547f MM |
6644 | int err; |
6645 | ||
22e52b07 | 6646 | arch_update_cpu_topology(); |
029190c5 | 6647 | ndoms_cur = 1; |
acc3f5d7 | 6648 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6649 | if (!doms_cur) |
acc3f5d7 RR |
6650 | doms_cur = &fallback_doms; |
6651 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6652 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6653 | register_sched_domain_sysctl(); |
7378547f MM |
6654 | |
6655 | return err; | |
1a20ff27 DG |
6656 | } |
6657 | ||
1a20ff27 DG |
6658 | /* |
6659 | * Detach sched domains from a group of cpus specified in cpu_map | |
6660 | * These cpus will now be attached to the NULL domain | |
6661 | */ | |
96f874e2 | 6662 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6663 | { |
6664 | int i; | |
6665 | ||
dce840a0 | 6666 | rcu_read_lock(); |
abcd083a | 6667 | for_each_cpu(i, cpu_map) |
57d885fe | 6668 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6669 | rcu_read_unlock(); |
1a20ff27 DG |
6670 | } |
6671 | ||
1d3504fc HS |
6672 | /* handle null as "default" */ |
6673 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6674 | struct sched_domain_attr *new, int idx_new) | |
6675 | { | |
6676 | struct sched_domain_attr tmp; | |
6677 | ||
6678 | /* fast path */ | |
6679 | if (!new && !cur) | |
6680 | return 1; | |
6681 | ||
6682 | tmp = SD_ATTR_INIT; | |
6683 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6684 | new ? (new + idx_new) : &tmp, | |
6685 | sizeof(struct sched_domain_attr)); | |
6686 | } | |
6687 | ||
029190c5 PJ |
6688 | /* |
6689 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6690 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6691 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6692 | * It destroys each deleted domain and builds each new domain. | |
6693 | * | |
acc3f5d7 | 6694 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6695 | * The masks don't intersect (don't overlap.) We should setup one |
6696 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6697 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6698 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6699 | * it as it is. | |
6700 | * | |
acc3f5d7 RR |
6701 | * The passed in 'doms_new' should be allocated using |
6702 | * alloc_sched_domains. This routine takes ownership of it and will | |
6703 | * free_sched_domains it when done with it. If the caller failed the | |
6704 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6705 | * and partition_sched_domains() will fallback to the single partition | |
6706 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6707 | * |
96f874e2 | 6708 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6709 | * ndoms_new == 0 is a special case for destroying existing domains, |
6710 | * and it will not create the default domain. | |
dfb512ec | 6711 | * |
029190c5 PJ |
6712 | * Call with hotplug lock held |
6713 | */ | |
acc3f5d7 | 6714 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6715 | struct sched_domain_attr *dattr_new) |
029190c5 | 6716 | { |
dfb512ec | 6717 | int i, j, n; |
d65bd5ec | 6718 | int new_topology; |
029190c5 | 6719 | |
712555ee | 6720 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6721 | |
7378547f MM |
6722 | /* always unregister in case we don't destroy any domains */ |
6723 | unregister_sched_domain_sysctl(); | |
6724 | ||
d65bd5ec HC |
6725 | /* Let architecture update cpu core mappings. */ |
6726 | new_topology = arch_update_cpu_topology(); | |
6727 | ||
dfb512ec | 6728 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6729 | |
6730 | /* Destroy deleted domains */ | |
6731 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6732 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6733 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6734 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6735 | goto match1; |
6736 | } | |
6737 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6738 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6739 | match1: |
6740 | ; | |
6741 | } | |
6742 | ||
c8d2d47a | 6743 | n = ndoms_cur; |
e761b772 | 6744 | if (doms_new == NULL) { |
c8d2d47a | 6745 | n = 0; |
acc3f5d7 | 6746 | doms_new = &fallback_doms; |
6ad4c188 | 6747 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6748 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6749 | } |
6750 | ||
029190c5 PJ |
6751 | /* Build new domains */ |
6752 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6753 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6754 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6755 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6756 | goto match2; |
6757 | } | |
6758 | /* no match - add a new doms_new */ | |
dce840a0 | 6759 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6760 | match2: |
6761 | ; | |
6762 | } | |
6763 | ||
6764 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6765 | if (doms_cur != &fallback_doms) |
6766 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6767 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6768 | doms_cur = doms_new; |
1d3504fc | 6769 | dattr_cur = dattr_new; |
029190c5 | 6770 | ndoms_cur = ndoms_new; |
7378547f MM |
6771 | |
6772 | register_sched_domain_sysctl(); | |
a1835615 | 6773 | |
712555ee | 6774 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6775 | } |
6776 | ||
d35be8ba SB |
6777 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6778 | ||
1da177e4 | 6779 | /* |
3a101d05 TH |
6780 | * Update cpusets according to cpu_active mask. If cpusets are |
6781 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6782 | * around partition_sched_domains(). | |
d35be8ba SB |
6783 | * |
6784 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6785 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6786 | */ |
0b2e918a TH |
6787 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6788 | void *hcpu) | |
e761b772 | 6789 | { |
d35be8ba SB |
6790 | switch (action) { |
6791 | case CPU_ONLINE_FROZEN: | |
6792 | case CPU_DOWN_FAILED_FROZEN: | |
6793 | ||
6794 | /* | |
6795 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6796 | * resume sequence. As long as this is not the last online | |
6797 | * operation in the resume sequence, just build a single sched | |
6798 | * domain, ignoring cpusets. | |
6799 | */ | |
6800 | num_cpus_frozen--; | |
6801 | if (likely(num_cpus_frozen)) { | |
6802 | partition_sched_domains(1, NULL, NULL); | |
6803 | break; | |
6804 | } | |
6805 | ||
6806 | /* | |
6807 | * This is the last CPU online operation. So fall through and | |
6808 | * restore the original sched domains by considering the | |
6809 | * cpuset configurations. | |
6810 | */ | |
6811 | ||
e761b772 | 6812 | case CPU_ONLINE: |
6ad4c188 | 6813 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6814 | cpuset_update_active_cpus(true); |
d35be8ba | 6815 | break; |
3a101d05 TH |
6816 | default: |
6817 | return NOTIFY_DONE; | |
6818 | } | |
d35be8ba | 6819 | return NOTIFY_OK; |
3a101d05 | 6820 | } |
e761b772 | 6821 | |
0b2e918a TH |
6822 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6823 | void *hcpu) | |
3a101d05 | 6824 | { |
d35be8ba | 6825 | switch (action) { |
3a101d05 | 6826 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6827 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6828 | break; |
6829 | case CPU_DOWN_PREPARE_FROZEN: | |
6830 | num_cpus_frozen++; | |
6831 | partition_sched_domains(1, NULL, NULL); | |
6832 | break; | |
e761b772 MK |
6833 | default: |
6834 | return NOTIFY_DONE; | |
6835 | } | |
d35be8ba | 6836 | return NOTIFY_OK; |
e761b772 | 6837 | } |
e761b772 | 6838 | |
1da177e4 LT |
6839 | void __init sched_init_smp(void) |
6840 | { | |
dcc30a35 RR |
6841 | cpumask_var_t non_isolated_cpus; |
6842 | ||
6843 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6844 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6845 | |
cb83b629 PZ |
6846 | sched_init_numa(); |
6847 | ||
6acce3ef PZ |
6848 | /* |
6849 | * There's no userspace yet to cause hotplug operations; hence all the | |
6850 | * cpu masks are stable and all blatant races in the below code cannot | |
6851 | * happen. | |
6852 | */ | |
712555ee | 6853 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6854 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6855 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6856 | if (cpumask_empty(non_isolated_cpus)) | |
6857 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6858 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6859 | |
301a5cba | 6860 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6861 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6862 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6863 | |
b328ca18 | 6864 | init_hrtick(); |
5c1e1767 NP |
6865 | |
6866 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6867 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6868 | BUG(); |
19978ca6 | 6869 | sched_init_granularity(); |
dcc30a35 | 6870 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6871 | |
0e3900e6 | 6872 | init_sched_rt_class(); |
1baca4ce | 6873 | init_sched_dl_class(); |
1da177e4 LT |
6874 | } |
6875 | #else | |
6876 | void __init sched_init_smp(void) | |
6877 | { | |
19978ca6 | 6878 | sched_init_granularity(); |
1da177e4 LT |
6879 | } |
6880 | #endif /* CONFIG_SMP */ | |
6881 | ||
cd1bb94b AB |
6882 | const_debug unsigned int sysctl_timer_migration = 1; |
6883 | ||
1da177e4 LT |
6884 | int in_sched_functions(unsigned long addr) |
6885 | { | |
1da177e4 LT |
6886 | return in_lock_functions(addr) || |
6887 | (addr >= (unsigned long)__sched_text_start | |
6888 | && addr < (unsigned long)__sched_text_end); | |
6889 | } | |
6890 | ||
029632fb | 6891 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6892 | /* |
6893 | * Default task group. | |
6894 | * Every task in system belongs to this group at bootup. | |
6895 | */ | |
029632fb | 6896 | struct task_group root_task_group; |
35cf4e50 | 6897 | LIST_HEAD(task_groups); |
052f1dc7 | 6898 | #endif |
6f505b16 | 6899 | |
e6252c3e | 6900 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6901 | |
1da177e4 LT |
6902 | void __init sched_init(void) |
6903 | { | |
dd41f596 | 6904 | int i, j; |
434d53b0 MT |
6905 | unsigned long alloc_size = 0, ptr; |
6906 | ||
6907 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6908 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6909 | #endif | |
6910 | #ifdef CONFIG_RT_GROUP_SCHED | |
6911 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6912 | #endif |
df7c8e84 | 6913 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6914 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6915 | #endif |
434d53b0 | 6916 | if (alloc_size) { |
36b7b6d4 | 6917 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6918 | |
6919 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6920 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6921 | ptr += nr_cpu_ids * sizeof(void **); |
6922 | ||
07e06b01 | 6923 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6924 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6925 | |
6d6bc0ad | 6926 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6927 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6928 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6929 | ptr += nr_cpu_ids * sizeof(void **); |
6930 | ||
07e06b01 | 6931 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6932 | ptr += nr_cpu_ids * sizeof(void **); |
6933 | ||
6d6bc0ad | 6934 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6935 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6936 | for_each_possible_cpu(i) { | |
e6252c3e | 6937 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6938 | ptr += cpumask_size(); |
6939 | } | |
6940 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6941 | } |
dd41f596 | 6942 | |
332ac17e DF |
6943 | init_rt_bandwidth(&def_rt_bandwidth, |
6944 | global_rt_period(), global_rt_runtime()); | |
6945 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6946 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6947 | |
57d885fe GH |
6948 | #ifdef CONFIG_SMP |
6949 | init_defrootdomain(); | |
6950 | #endif | |
6951 | ||
d0b27fa7 | 6952 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6953 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6954 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6955 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6956 | |
7c941438 | 6957 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6958 | list_add(&root_task_group.list, &task_groups); |
6959 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6960 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6961 | autogroup_init(&init_task); |
54c707e9 | 6962 | |
7c941438 | 6963 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6964 | |
0a945022 | 6965 | for_each_possible_cpu(i) { |
70b97a7f | 6966 | struct rq *rq; |
1da177e4 LT |
6967 | |
6968 | rq = cpu_rq(i); | |
05fa785c | 6969 | raw_spin_lock_init(&rq->lock); |
7897986b | 6970 | rq->nr_running = 0; |
dce48a84 TG |
6971 | rq->calc_load_active = 0; |
6972 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6973 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6974 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6975 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6976 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6977 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6978 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6979 | /* |
07e06b01 | 6980 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6981 | * |
6982 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6983 | * gets 100% of the cpu resources in the system. This overall | |
6984 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6985 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6986 | * based on each entity's (task or task-group's) weight |
6987 | * (se->load.weight). | |
6988 | * | |
07e06b01 | 6989 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6990 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6991 | * then A0's share of the cpu resource is: | |
6992 | * | |
0d905bca | 6993 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6994 | * |
07e06b01 YZ |
6995 | * We achieve this by letting root_task_group's tasks sit |
6996 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6997 | */ |
ab84d31e | 6998 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6999 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7000 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7001 | ||
7002 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7003 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7004 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7005 | #endif |
1da177e4 | 7006 | |
dd41f596 IM |
7007 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7008 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7009 | |
7010 | rq->last_load_update_tick = jiffies; | |
7011 | ||
1da177e4 | 7012 | #ifdef CONFIG_SMP |
41c7ce9a | 7013 | rq->sd = NULL; |
57d885fe | 7014 | rq->rd = NULL; |
ca8ce3d0 | 7015 | rq->cpu_capacity = SCHED_CAPACITY_SCALE; |
3f029d3c | 7016 | rq->post_schedule = 0; |
1da177e4 | 7017 | rq->active_balance = 0; |
dd41f596 | 7018 | rq->next_balance = jiffies; |
1da177e4 | 7019 | rq->push_cpu = 0; |
0a2966b4 | 7020 | rq->cpu = i; |
1f11eb6a | 7021 | rq->online = 0; |
eae0c9df MG |
7022 | rq->idle_stamp = 0; |
7023 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7024 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7025 | |
7026 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7027 | ||
dc938520 | 7028 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7029 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7030 | rq->nohz_flags = 0; |
83cd4fe2 | 7031 | #endif |
265f22a9 FW |
7032 | #ifdef CONFIG_NO_HZ_FULL |
7033 | rq->last_sched_tick = 0; | |
7034 | #endif | |
1da177e4 | 7035 | #endif |
8f4d37ec | 7036 | init_rq_hrtick(rq); |
1da177e4 | 7037 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7038 | } |
7039 | ||
2dd73a4f | 7040 | set_load_weight(&init_task); |
b50f60ce | 7041 | |
e107be36 AK |
7042 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7043 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7044 | #endif | |
7045 | ||
1da177e4 LT |
7046 | /* |
7047 | * The boot idle thread does lazy MMU switching as well: | |
7048 | */ | |
7049 | atomic_inc(&init_mm.mm_count); | |
7050 | enter_lazy_tlb(&init_mm, current); | |
7051 | ||
7052 | /* | |
7053 | * Make us the idle thread. Technically, schedule() should not be | |
7054 | * called from this thread, however somewhere below it might be, | |
7055 | * but because we are the idle thread, we just pick up running again | |
7056 | * when this runqueue becomes "idle". | |
7057 | */ | |
7058 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7059 | |
7060 | calc_load_update = jiffies + LOAD_FREQ; | |
7061 | ||
dd41f596 IM |
7062 | /* |
7063 | * During early bootup we pretend to be a normal task: | |
7064 | */ | |
7065 | current->sched_class = &fair_sched_class; | |
6892b75e | 7066 | |
bf4d83f6 | 7067 | #ifdef CONFIG_SMP |
4cb98839 | 7068 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7069 | /* May be allocated at isolcpus cmdline parse time */ |
7070 | if (cpu_isolated_map == NULL) | |
7071 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7072 | idle_thread_set_boot_cpu(); |
a803f026 | 7073 | set_cpu_rq_start_time(); |
029632fb PZ |
7074 | #endif |
7075 | init_sched_fair_class(); | |
6a7b3dc3 | 7076 | |
6892b75e | 7077 | scheduler_running = 1; |
1da177e4 LT |
7078 | } |
7079 | ||
d902db1e | 7080 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7081 | static inline int preempt_count_equals(int preempt_offset) |
7082 | { | |
234da7bc | 7083 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7084 | |
4ba8216c | 7085 | return (nested == preempt_offset); |
e4aafea2 FW |
7086 | } |
7087 | ||
d894837f | 7088 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7089 | { |
1da177e4 LT |
7090 | static unsigned long prev_jiffy; /* ratelimiting */ |
7091 | ||
b3fbab05 | 7092 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7093 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7094 | !is_idle_task(current)) || | |
e4aafea2 | 7095 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7096 | return; |
7097 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7098 | return; | |
7099 | prev_jiffy = jiffies; | |
7100 | ||
3df0fc5b PZ |
7101 | printk(KERN_ERR |
7102 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7103 | file, line); | |
7104 | printk(KERN_ERR | |
7105 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7106 | in_atomic(), irqs_disabled(), | |
7107 | current->pid, current->comm); | |
aef745fc IM |
7108 | |
7109 | debug_show_held_locks(current); | |
7110 | if (irqs_disabled()) | |
7111 | print_irqtrace_events(current); | |
8f47b187 TG |
7112 | #ifdef CONFIG_DEBUG_PREEMPT |
7113 | if (!preempt_count_equals(preempt_offset)) { | |
7114 | pr_err("Preemption disabled at:"); | |
7115 | print_ip_sym(current->preempt_disable_ip); | |
7116 | pr_cont("\n"); | |
7117 | } | |
7118 | #endif | |
aef745fc | 7119 | dump_stack(); |
1da177e4 LT |
7120 | } |
7121 | EXPORT_SYMBOL(__might_sleep); | |
7122 | #endif | |
7123 | ||
7124 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7125 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7126 | { | |
da7a735e | 7127 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7128 | struct sched_attr attr = { |
7129 | .sched_policy = SCHED_NORMAL, | |
7130 | }; | |
da7a735e | 7131 | int old_prio = p->prio; |
da0c1e65 | 7132 | int queued; |
3e51f33f | 7133 | |
da0c1e65 KT |
7134 | queued = task_on_rq_queued(p); |
7135 | if (queued) | |
4ca9b72b | 7136 | dequeue_task(rq, p, 0); |
d50dde5a | 7137 | __setscheduler(rq, p, &attr); |
da0c1e65 | 7138 | if (queued) { |
4ca9b72b | 7139 | enqueue_task(rq, p, 0); |
8875125e | 7140 | resched_curr(rq); |
3a5e4dc1 | 7141 | } |
da7a735e PZ |
7142 | |
7143 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7144 | } |
7145 | ||
1da177e4 LT |
7146 | void normalize_rt_tasks(void) |
7147 | { | |
a0f98a1c | 7148 | struct task_struct *g, *p; |
1da177e4 | 7149 | unsigned long flags; |
70b97a7f | 7150 | struct rq *rq; |
1da177e4 | 7151 | |
4cf5d77a | 7152 | read_lock_irqsave(&tasklist_lock, flags); |
5d07f420 | 7153 | for_each_process_thread(g, p) { |
178be793 IM |
7154 | /* |
7155 | * Only normalize user tasks: | |
7156 | */ | |
7157 | if (!p->mm) | |
7158 | continue; | |
7159 | ||
6cfb0d5d | 7160 | p->se.exec_start = 0; |
6cfb0d5d | 7161 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7162 | p->se.statistics.wait_start = 0; |
7163 | p->se.statistics.sleep_start = 0; | |
7164 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7165 | #endif |
dd41f596 | 7166 | |
aab03e05 | 7167 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7168 | /* |
7169 | * Renice negative nice level userspace | |
7170 | * tasks back to 0: | |
7171 | */ | |
d0ea0268 | 7172 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7173 | set_user_nice(p, 0); |
1da177e4 | 7174 | continue; |
dd41f596 | 7175 | } |
1da177e4 | 7176 | |
1d615482 | 7177 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7178 | rq = __task_rq_lock(p); |
1da177e4 | 7179 | |
178be793 | 7180 | normalize_task(rq, p); |
3a5e4dc1 | 7181 | |
b29739f9 | 7182 | __task_rq_unlock(rq); |
1d615482 | 7183 | raw_spin_unlock(&p->pi_lock); |
5d07f420 | 7184 | } |
4cf5d77a | 7185 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7186 | } |
7187 | ||
7188 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7189 | |
67fc4e0c | 7190 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7191 | /* |
67fc4e0c | 7192 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7193 | * |
7194 | * They can only be called when the whole system has been | |
7195 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7196 | * activity can take place. Using them for anything else would | |
7197 | * be a serious bug, and as a result, they aren't even visible | |
7198 | * under any other configuration. | |
7199 | */ | |
7200 | ||
7201 | /** | |
7202 | * curr_task - return the current task for a given cpu. | |
7203 | * @cpu: the processor in question. | |
7204 | * | |
7205 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7206 | * |
7207 | * Return: The current task for @cpu. | |
1df5c10a | 7208 | */ |
36c8b586 | 7209 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7210 | { |
7211 | return cpu_curr(cpu); | |
7212 | } | |
7213 | ||
67fc4e0c JW |
7214 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7215 | ||
7216 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7217 | /** |
7218 | * set_curr_task - set the current task for a given cpu. | |
7219 | * @cpu: the processor in question. | |
7220 | * @p: the task pointer to set. | |
7221 | * | |
7222 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7223 | * are serviced on a separate stack. It allows the architecture to switch the |
7224 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7225 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7226 | * and caller must save the original value of the current task (see | |
7227 | * curr_task() above) and restore that value before reenabling interrupts and | |
7228 | * re-starting the system. | |
7229 | * | |
7230 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7231 | */ | |
36c8b586 | 7232 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7233 | { |
7234 | cpu_curr(cpu) = p; | |
7235 | } | |
7236 | ||
7237 | #endif | |
29f59db3 | 7238 | |
7c941438 | 7239 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7240 | /* task_group_lock serializes the addition/removal of task groups */ |
7241 | static DEFINE_SPINLOCK(task_group_lock); | |
7242 | ||
bccbe08a PZ |
7243 | static void free_sched_group(struct task_group *tg) |
7244 | { | |
7245 | free_fair_sched_group(tg); | |
7246 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7247 | autogroup_free(tg); |
bccbe08a PZ |
7248 | kfree(tg); |
7249 | } | |
7250 | ||
7251 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7252 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7253 | { |
7254 | struct task_group *tg; | |
bccbe08a PZ |
7255 | |
7256 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7257 | if (!tg) | |
7258 | return ERR_PTR(-ENOMEM); | |
7259 | ||
ec7dc8ac | 7260 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7261 | goto err; |
7262 | ||
ec7dc8ac | 7263 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7264 | goto err; |
7265 | ||
ace783b9 LZ |
7266 | return tg; |
7267 | ||
7268 | err: | |
7269 | free_sched_group(tg); | |
7270 | return ERR_PTR(-ENOMEM); | |
7271 | } | |
7272 | ||
7273 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7274 | { | |
7275 | unsigned long flags; | |
7276 | ||
8ed36996 | 7277 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7278 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7279 | |
7280 | WARN_ON(!parent); /* root should already exist */ | |
7281 | ||
7282 | tg->parent = parent; | |
f473aa5e | 7283 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7284 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7285 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7286 | } |
7287 | ||
9b5b7751 | 7288 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7289 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7290 | { |
29f59db3 | 7291 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7292 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7293 | } |
7294 | ||
9b5b7751 | 7295 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7296 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7297 | { |
7298 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7299 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7300 | } | |
7301 | ||
7302 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7303 | { |
8ed36996 | 7304 | unsigned long flags; |
9b5b7751 | 7305 | int i; |
29f59db3 | 7306 | |
3d4b47b4 PZ |
7307 | /* end participation in shares distribution */ |
7308 | for_each_possible_cpu(i) | |
bccbe08a | 7309 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7310 | |
7311 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7312 | list_del_rcu(&tg->list); |
f473aa5e | 7313 | list_del_rcu(&tg->siblings); |
8ed36996 | 7314 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7315 | } |
7316 | ||
9b5b7751 | 7317 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7318 | * The caller of this function should have put the task in its new group |
7319 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7320 | * reflect its new group. | |
9b5b7751 SV |
7321 | */ |
7322 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7323 | { |
8323f26c | 7324 | struct task_group *tg; |
da0c1e65 | 7325 | int queued, running; |
29f59db3 SV |
7326 | unsigned long flags; |
7327 | struct rq *rq; | |
7328 | ||
7329 | rq = task_rq_lock(tsk, &flags); | |
7330 | ||
051a1d1a | 7331 | running = task_current(rq, tsk); |
da0c1e65 | 7332 | queued = task_on_rq_queued(tsk); |
29f59db3 | 7333 | |
da0c1e65 | 7334 | if (queued) |
29f59db3 | 7335 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7336 | if (unlikely(running)) |
7337 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7338 | |
073219e9 | 7339 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7340 | lockdep_is_held(&tsk->sighand->siglock)), |
7341 | struct task_group, css); | |
7342 | tg = autogroup_task_group(tsk, tg); | |
7343 | tsk->sched_task_group = tg; | |
7344 | ||
810b3817 | 7345 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 | 7346 | if (tsk->sched_class->task_move_group) |
da0c1e65 | 7347 | tsk->sched_class->task_move_group(tsk, queued); |
b2b5ce02 | 7348 | else |
810b3817 | 7349 | #endif |
b2b5ce02 | 7350 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7351 | |
0e1f3483 HS |
7352 | if (unlikely(running)) |
7353 | tsk->sched_class->set_curr_task(rq); | |
da0c1e65 | 7354 | if (queued) |
371fd7e7 | 7355 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7356 | |
0122ec5b | 7357 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7358 | } |
7c941438 | 7359 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7360 | |
a790de99 PT |
7361 | #ifdef CONFIG_RT_GROUP_SCHED |
7362 | /* | |
7363 | * Ensure that the real time constraints are schedulable. | |
7364 | */ | |
7365 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7366 | |
9a7e0b18 PZ |
7367 | /* Must be called with tasklist_lock held */ |
7368 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7369 | { |
9a7e0b18 | 7370 | struct task_struct *g, *p; |
b40b2e8e | 7371 | |
5d07f420 | 7372 | for_each_process_thread(g, p) { |
029632fb | 7373 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 | 7374 | return 1; |
5d07f420 | 7375 | } |
b40b2e8e | 7376 | |
9a7e0b18 PZ |
7377 | return 0; |
7378 | } | |
b40b2e8e | 7379 | |
9a7e0b18 PZ |
7380 | struct rt_schedulable_data { |
7381 | struct task_group *tg; | |
7382 | u64 rt_period; | |
7383 | u64 rt_runtime; | |
7384 | }; | |
b40b2e8e | 7385 | |
a790de99 | 7386 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7387 | { |
7388 | struct rt_schedulable_data *d = data; | |
7389 | struct task_group *child; | |
7390 | unsigned long total, sum = 0; | |
7391 | u64 period, runtime; | |
b40b2e8e | 7392 | |
9a7e0b18 PZ |
7393 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7394 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7395 | |
9a7e0b18 PZ |
7396 | if (tg == d->tg) { |
7397 | period = d->rt_period; | |
7398 | runtime = d->rt_runtime; | |
b40b2e8e | 7399 | } |
b40b2e8e | 7400 | |
4653f803 PZ |
7401 | /* |
7402 | * Cannot have more runtime than the period. | |
7403 | */ | |
7404 | if (runtime > period && runtime != RUNTIME_INF) | |
7405 | return -EINVAL; | |
6f505b16 | 7406 | |
4653f803 PZ |
7407 | /* |
7408 | * Ensure we don't starve existing RT tasks. | |
7409 | */ | |
9a7e0b18 PZ |
7410 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7411 | return -EBUSY; | |
6f505b16 | 7412 | |
9a7e0b18 | 7413 | total = to_ratio(period, runtime); |
6f505b16 | 7414 | |
4653f803 PZ |
7415 | /* |
7416 | * Nobody can have more than the global setting allows. | |
7417 | */ | |
7418 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7419 | return -EINVAL; | |
6f505b16 | 7420 | |
4653f803 PZ |
7421 | /* |
7422 | * The sum of our children's runtime should not exceed our own. | |
7423 | */ | |
9a7e0b18 PZ |
7424 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7425 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7426 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7427 | |
9a7e0b18 PZ |
7428 | if (child == d->tg) { |
7429 | period = d->rt_period; | |
7430 | runtime = d->rt_runtime; | |
7431 | } | |
6f505b16 | 7432 | |
9a7e0b18 | 7433 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7434 | } |
6f505b16 | 7435 | |
9a7e0b18 PZ |
7436 | if (sum > total) |
7437 | return -EINVAL; | |
7438 | ||
7439 | return 0; | |
6f505b16 PZ |
7440 | } |
7441 | ||
9a7e0b18 | 7442 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7443 | { |
8277434e PT |
7444 | int ret; |
7445 | ||
9a7e0b18 PZ |
7446 | struct rt_schedulable_data data = { |
7447 | .tg = tg, | |
7448 | .rt_period = period, | |
7449 | .rt_runtime = runtime, | |
7450 | }; | |
7451 | ||
8277434e PT |
7452 | rcu_read_lock(); |
7453 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7454 | rcu_read_unlock(); | |
7455 | ||
7456 | return ret; | |
521f1a24 DG |
7457 | } |
7458 | ||
ab84d31e | 7459 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7460 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7461 | { |
ac086bc2 | 7462 | int i, err = 0; |
9f0c1e56 | 7463 | |
9f0c1e56 | 7464 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7465 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7466 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7467 | if (err) | |
9f0c1e56 | 7468 | goto unlock; |
ac086bc2 | 7469 | |
0986b11b | 7470 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7471 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7472 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7473 | |
7474 | for_each_possible_cpu(i) { | |
7475 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7476 | ||
0986b11b | 7477 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7478 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7479 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7480 | } |
0986b11b | 7481 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7482 | unlock: |
521f1a24 | 7483 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7484 | mutex_unlock(&rt_constraints_mutex); |
7485 | ||
7486 | return err; | |
6f505b16 PZ |
7487 | } |
7488 | ||
25cc7da7 | 7489 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7490 | { |
7491 | u64 rt_runtime, rt_period; | |
7492 | ||
7493 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7494 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7495 | if (rt_runtime_us < 0) | |
7496 | rt_runtime = RUNTIME_INF; | |
7497 | ||
ab84d31e | 7498 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7499 | } |
7500 | ||
25cc7da7 | 7501 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7502 | { |
7503 | u64 rt_runtime_us; | |
7504 | ||
d0b27fa7 | 7505 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7506 | return -1; |
7507 | ||
d0b27fa7 | 7508 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7509 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7510 | return rt_runtime_us; | |
7511 | } | |
d0b27fa7 | 7512 | |
25cc7da7 | 7513 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7514 | { |
7515 | u64 rt_runtime, rt_period; | |
7516 | ||
7517 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7518 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7519 | ||
619b0488 R |
7520 | if (rt_period == 0) |
7521 | return -EINVAL; | |
7522 | ||
ab84d31e | 7523 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7524 | } |
7525 | ||
25cc7da7 | 7526 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7527 | { |
7528 | u64 rt_period_us; | |
7529 | ||
7530 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7531 | do_div(rt_period_us, NSEC_PER_USEC); | |
7532 | return rt_period_us; | |
7533 | } | |
332ac17e | 7534 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7535 | |
332ac17e | 7536 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7537 | static int sched_rt_global_constraints(void) |
7538 | { | |
7539 | int ret = 0; | |
7540 | ||
7541 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7542 | read_lock(&tasklist_lock); |
4653f803 | 7543 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7544 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7545 | mutex_unlock(&rt_constraints_mutex); |
7546 | ||
7547 | return ret; | |
7548 | } | |
54e99124 | 7549 | |
25cc7da7 | 7550 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7551 | { |
7552 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7553 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7554 | return 0; | |
7555 | ||
7556 | return 1; | |
7557 | } | |
7558 | ||
6d6bc0ad | 7559 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7560 | static int sched_rt_global_constraints(void) |
7561 | { | |
ac086bc2 | 7562 | unsigned long flags; |
332ac17e | 7563 | int i, ret = 0; |
ec5d4989 | 7564 | |
0986b11b | 7565 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7566 | for_each_possible_cpu(i) { |
7567 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7568 | ||
0986b11b | 7569 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7570 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7571 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7572 | } |
0986b11b | 7573 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7574 | |
332ac17e | 7575 | return ret; |
d0b27fa7 | 7576 | } |
6d6bc0ad | 7577 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7578 | |
332ac17e DF |
7579 | static int sched_dl_global_constraints(void) |
7580 | { | |
1724813d PZ |
7581 | u64 runtime = global_rt_runtime(); |
7582 | u64 period = global_rt_period(); | |
332ac17e | 7583 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7584 | int cpu, ret = 0; |
49516342 | 7585 | unsigned long flags; |
332ac17e DF |
7586 | |
7587 | /* | |
7588 | * Here we want to check the bandwidth not being set to some | |
7589 | * value smaller than the currently allocated bandwidth in | |
7590 | * any of the root_domains. | |
7591 | * | |
7592 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7593 | * cycling on root_domains... Discussion on different/better | |
7594 | * solutions is welcome! | |
7595 | */ | |
1724813d PZ |
7596 | for_each_possible_cpu(cpu) { |
7597 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7598 | |
49516342 | 7599 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7600 | if (new_bw < dl_b->total_bw) |
7601 | ret = -EBUSY; | |
49516342 | 7602 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7603 | |
7604 | if (ret) | |
7605 | break; | |
332ac17e DF |
7606 | } |
7607 | ||
1724813d | 7608 | return ret; |
332ac17e DF |
7609 | } |
7610 | ||
1724813d | 7611 | static void sched_dl_do_global(void) |
ce0dbbbb | 7612 | { |
1724813d PZ |
7613 | u64 new_bw = -1; |
7614 | int cpu; | |
49516342 | 7615 | unsigned long flags; |
ce0dbbbb | 7616 | |
1724813d PZ |
7617 | def_dl_bandwidth.dl_period = global_rt_period(); |
7618 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7619 | ||
7620 | if (global_rt_runtime() != RUNTIME_INF) | |
7621 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7622 | ||
7623 | /* | |
7624 | * FIXME: As above... | |
7625 | */ | |
7626 | for_each_possible_cpu(cpu) { | |
7627 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7628 | ||
49516342 | 7629 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7630 | dl_b->bw = new_bw; |
49516342 | 7631 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7632 | } |
1724813d PZ |
7633 | } |
7634 | ||
7635 | static int sched_rt_global_validate(void) | |
7636 | { | |
7637 | if (sysctl_sched_rt_period <= 0) | |
7638 | return -EINVAL; | |
7639 | ||
e9e7cb38 JL |
7640 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7641 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7642 | return -EINVAL; |
7643 | ||
7644 | return 0; | |
7645 | } | |
7646 | ||
7647 | static void sched_rt_do_global(void) | |
7648 | { | |
7649 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7650 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7651 | } |
7652 | ||
d0b27fa7 | 7653 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7654 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7655 | loff_t *ppos) |
7656 | { | |
d0b27fa7 PZ |
7657 | int old_period, old_runtime; |
7658 | static DEFINE_MUTEX(mutex); | |
1724813d | 7659 | int ret; |
d0b27fa7 PZ |
7660 | |
7661 | mutex_lock(&mutex); | |
7662 | old_period = sysctl_sched_rt_period; | |
7663 | old_runtime = sysctl_sched_rt_runtime; | |
7664 | ||
8d65af78 | 7665 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7666 | |
7667 | if (!ret && write) { | |
1724813d PZ |
7668 | ret = sched_rt_global_validate(); |
7669 | if (ret) | |
7670 | goto undo; | |
7671 | ||
d0b27fa7 | 7672 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7673 | if (ret) |
7674 | goto undo; | |
7675 | ||
7676 | ret = sched_dl_global_constraints(); | |
7677 | if (ret) | |
7678 | goto undo; | |
7679 | ||
7680 | sched_rt_do_global(); | |
7681 | sched_dl_do_global(); | |
7682 | } | |
7683 | if (0) { | |
7684 | undo: | |
7685 | sysctl_sched_rt_period = old_period; | |
7686 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7687 | } |
7688 | mutex_unlock(&mutex); | |
7689 | ||
7690 | return ret; | |
7691 | } | |
68318b8e | 7692 | |
1724813d | 7693 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7694 | void __user *buffer, size_t *lenp, |
7695 | loff_t *ppos) | |
7696 | { | |
7697 | int ret; | |
332ac17e | 7698 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7699 | |
7700 | mutex_lock(&mutex); | |
332ac17e | 7701 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7702 | /* make sure that internally we keep jiffies */ |
7703 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7704 | if (!ret && write) { |
1724813d PZ |
7705 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7706 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7707 | } |
7708 | mutex_unlock(&mutex); | |
332ac17e DF |
7709 | return ret; |
7710 | } | |
7711 | ||
052f1dc7 | 7712 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7713 | |
a7c6d554 | 7714 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7715 | { |
a7c6d554 | 7716 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7717 | } |
7718 | ||
eb95419b TH |
7719 | static struct cgroup_subsys_state * |
7720 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7721 | { |
eb95419b TH |
7722 | struct task_group *parent = css_tg(parent_css); |
7723 | struct task_group *tg; | |
68318b8e | 7724 | |
eb95419b | 7725 | if (!parent) { |
68318b8e | 7726 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7727 | return &root_task_group.css; |
68318b8e SV |
7728 | } |
7729 | ||
ec7dc8ac | 7730 | tg = sched_create_group(parent); |
68318b8e SV |
7731 | if (IS_ERR(tg)) |
7732 | return ERR_PTR(-ENOMEM); | |
7733 | ||
68318b8e SV |
7734 | return &tg->css; |
7735 | } | |
7736 | ||
eb95419b | 7737 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7738 | { |
eb95419b | 7739 | struct task_group *tg = css_tg(css); |
5c9d535b | 7740 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 7741 | |
63876986 TH |
7742 | if (parent) |
7743 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7744 | return 0; |
7745 | } | |
7746 | ||
eb95419b | 7747 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7748 | { |
eb95419b | 7749 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7750 | |
7751 | sched_destroy_group(tg); | |
7752 | } | |
7753 | ||
eb95419b | 7754 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7755 | { |
eb95419b | 7756 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7757 | |
7758 | sched_offline_group(tg); | |
7759 | } | |
7760 | ||
eb95419b | 7761 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7762 | struct cgroup_taskset *tset) |
68318b8e | 7763 | { |
bb9d97b6 TH |
7764 | struct task_struct *task; |
7765 | ||
924f0d9a | 7766 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7767 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7768 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7769 | return -EINVAL; |
b68aa230 | 7770 | #else |
bb9d97b6 TH |
7771 | /* We don't support RT-tasks being in separate groups */ |
7772 | if (task->sched_class != &fair_sched_class) | |
7773 | return -EINVAL; | |
b68aa230 | 7774 | #endif |
bb9d97b6 | 7775 | } |
be367d09 BB |
7776 | return 0; |
7777 | } | |
68318b8e | 7778 | |
eb95419b | 7779 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7780 | struct cgroup_taskset *tset) |
68318b8e | 7781 | { |
bb9d97b6 TH |
7782 | struct task_struct *task; |
7783 | ||
924f0d9a | 7784 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7785 | sched_move_task(task); |
68318b8e SV |
7786 | } |
7787 | ||
eb95419b TH |
7788 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7789 | struct cgroup_subsys_state *old_css, | |
7790 | struct task_struct *task) | |
068c5cc5 PZ |
7791 | { |
7792 | /* | |
7793 | * cgroup_exit() is called in the copy_process() failure path. | |
7794 | * Ignore this case since the task hasn't ran yet, this avoids | |
7795 | * trying to poke a half freed task state from generic code. | |
7796 | */ | |
7797 | if (!(task->flags & PF_EXITING)) | |
7798 | return; | |
7799 | ||
7800 | sched_move_task(task); | |
7801 | } | |
7802 | ||
052f1dc7 | 7803 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7804 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7805 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7806 | { |
182446d0 | 7807 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7808 | } |
7809 | ||
182446d0 TH |
7810 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7811 | struct cftype *cft) | |
68318b8e | 7812 | { |
182446d0 | 7813 | struct task_group *tg = css_tg(css); |
68318b8e | 7814 | |
c8b28116 | 7815 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7816 | } |
ab84d31e PT |
7817 | |
7818 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7819 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7820 | ||
ab84d31e PT |
7821 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7822 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7823 | ||
a790de99 PT |
7824 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7825 | ||
ab84d31e PT |
7826 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7827 | { | |
56f570e5 | 7828 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7829 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7830 | |
7831 | if (tg == &root_task_group) | |
7832 | return -EINVAL; | |
7833 | ||
7834 | /* | |
7835 | * Ensure we have at some amount of bandwidth every period. This is | |
7836 | * to prevent reaching a state of large arrears when throttled via | |
7837 | * entity_tick() resulting in prolonged exit starvation. | |
7838 | */ | |
7839 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7840 | return -EINVAL; | |
7841 | ||
7842 | /* | |
7843 | * Likewise, bound things on the otherside by preventing insane quota | |
7844 | * periods. This also allows us to normalize in computing quota | |
7845 | * feasibility. | |
7846 | */ | |
7847 | if (period > max_cfs_quota_period) | |
7848 | return -EINVAL; | |
7849 | ||
0e59bdae KT |
7850 | /* |
7851 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
7852 | * unthrottle_offline_cfs_rqs(). | |
7853 | */ | |
7854 | get_online_cpus(); | |
a790de99 PT |
7855 | mutex_lock(&cfs_constraints_mutex); |
7856 | ret = __cfs_schedulable(tg, period, quota); | |
7857 | if (ret) | |
7858 | goto out_unlock; | |
7859 | ||
58088ad0 | 7860 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7861 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7862 | /* |
7863 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7864 | * before making related changes, and on->off must occur afterwards | |
7865 | */ | |
7866 | if (runtime_enabled && !runtime_was_enabled) | |
7867 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7868 | raw_spin_lock_irq(&cfs_b->lock); |
7869 | cfs_b->period = ns_to_ktime(period); | |
7870 | cfs_b->quota = quota; | |
58088ad0 | 7871 | |
a9cf55b2 | 7872 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7873 | /* restart the period timer (if active) to handle new period expiry */ |
7874 | if (runtime_enabled && cfs_b->timer_active) { | |
7875 | /* force a reprogram */ | |
09dc4ab0 | 7876 | __start_cfs_bandwidth(cfs_b, true); |
58088ad0 | 7877 | } |
ab84d31e PT |
7878 | raw_spin_unlock_irq(&cfs_b->lock); |
7879 | ||
0e59bdae | 7880 | for_each_online_cpu(i) { |
ab84d31e | 7881 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7882 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7883 | |
7884 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7885 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7886 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7887 | |
029632fb | 7888 | if (cfs_rq->throttled) |
671fd9da | 7889 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7890 | raw_spin_unlock_irq(&rq->lock); |
7891 | } | |
1ee14e6c BS |
7892 | if (runtime_was_enabled && !runtime_enabled) |
7893 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7894 | out_unlock: |
7895 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7896 | put_online_cpus(); |
ab84d31e | 7897 | |
a790de99 | 7898 | return ret; |
ab84d31e PT |
7899 | } |
7900 | ||
7901 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7902 | { | |
7903 | u64 quota, period; | |
7904 | ||
029632fb | 7905 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7906 | if (cfs_quota_us < 0) |
7907 | quota = RUNTIME_INF; | |
7908 | else | |
7909 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7910 | ||
7911 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7912 | } | |
7913 | ||
7914 | long tg_get_cfs_quota(struct task_group *tg) | |
7915 | { | |
7916 | u64 quota_us; | |
7917 | ||
029632fb | 7918 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7919 | return -1; |
7920 | ||
029632fb | 7921 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7922 | do_div(quota_us, NSEC_PER_USEC); |
7923 | ||
7924 | return quota_us; | |
7925 | } | |
7926 | ||
7927 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7928 | { | |
7929 | u64 quota, period; | |
7930 | ||
7931 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7932 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7933 | |
ab84d31e PT |
7934 | return tg_set_cfs_bandwidth(tg, period, quota); |
7935 | } | |
7936 | ||
7937 | long tg_get_cfs_period(struct task_group *tg) | |
7938 | { | |
7939 | u64 cfs_period_us; | |
7940 | ||
029632fb | 7941 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7942 | do_div(cfs_period_us, NSEC_PER_USEC); |
7943 | ||
7944 | return cfs_period_us; | |
7945 | } | |
7946 | ||
182446d0 TH |
7947 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7948 | struct cftype *cft) | |
ab84d31e | 7949 | { |
182446d0 | 7950 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7951 | } |
7952 | ||
182446d0 TH |
7953 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7954 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7955 | { |
182446d0 | 7956 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7957 | } |
7958 | ||
182446d0 TH |
7959 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7960 | struct cftype *cft) | |
ab84d31e | 7961 | { |
182446d0 | 7962 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7963 | } |
7964 | ||
182446d0 TH |
7965 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7966 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7967 | { |
182446d0 | 7968 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7969 | } |
7970 | ||
a790de99 PT |
7971 | struct cfs_schedulable_data { |
7972 | struct task_group *tg; | |
7973 | u64 period, quota; | |
7974 | }; | |
7975 | ||
7976 | /* | |
7977 | * normalize group quota/period to be quota/max_period | |
7978 | * note: units are usecs | |
7979 | */ | |
7980 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7981 | struct cfs_schedulable_data *d) | |
7982 | { | |
7983 | u64 quota, period; | |
7984 | ||
7985 | if (tg == d->tg) { | |
7986 | period = d->period; | |
7987 | quota = d->quota; | |
7988 | } else { | |
7989 | period = tg_get_cfs_period(tg); | |
7990 | quota = tg_get_cfs_quota(tg); | |
7991 | } | |
7992 | ||
7993 | /* note: these should typically be equivalent */ | |
7994 | if (quota == RUNTIME_INF || quota == -1) | |
7995 | return RUNTIME_INF; | |
7996 | ||
7997 | return to_ratio(period, quota); | |
7998 | } | |
7999 | ||
8000 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8001 | { | |
8002 | struct cfs_schedulable_data *d = data; | |
029632fb | 8003 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8004 | s64 quota = 0, parent_quota = -1; |
8005 | ||
8006 | if (!tg->parent) { | |
8007 | quota = RUNTIME_INF; | |
8008 | } else { | |
029632fb | 8009 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8010 | |
8011 | quota = normalize_cfs_quota(tg, d); | |
8012 | parent_quota = parent_b->hierarchal_quota; | |
8013 | ||
8014 | /* | |
8015 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8016 | * limit is set | |
8017 | */ | |
8018 | if (quota == RUNTIME_INF) | |
8019 | quota = parent_quota; | |
8020 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8021 | return -EINVAL; | |
8022 | } | |
8023 | cfs_b->hierarchal_quota = quota; | |
8024 | ||
8025 | return 0; | |
8026 | } | |
8027 | ||
8028 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8029 | { | |
8277434e | 8030 | int ret; |
a790de99 PT |
8031 | struct cfs_schedulable_data data = { |
8032 | .tg = tg, | |
8033 | .period = period, | |
8034 | .quota = quota, | |
8035 | }; | |
8036 | ||
8037 | if (quota != RUNTIME_INF) { | |
8038 | do_div(data.period, NSEC_PER_USEC); | |
8039 | do_div(data.quota, NSEC_PER_USEC); | |
8040 | } | |
8041 | ||
8277434e PT |
8042 | rcu_read_lock(); |
8043 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8044 | rcu_read_unlock(); | |
8045 | ||
8046 | return ret; | |
a790de99 | 8047 | } |
e8da1b18 | 8048 | |
2da8ca82 | 8049 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8050 | { |
2da8ca82 | 8051 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8052 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8053 | |
44ffc75b TH |
8054 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8055 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8056 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8057 | |
8058 | return 0; | |
8059 | } | |
ab84d31e | 8060 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8061 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8062 | |
052f1dc7 | 8063 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8064 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8065 | struct cftype *cft, s64 val) | |
6f505b16 | 8066 | { |
182446d0 | 8067 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8068 | } |
8069 | ||
182446d0 TH |
8070 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8071 | struct cftype *cft) | |
6f505b16 | 8072 | { |
182446d0 | 8073 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8074 | } |
d0b27fa7 | 8075 | |
182446d0 TH |
8076 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8077 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8078 | { |
182446d0 | 8079 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8080 | } |
8081 | ||
182446d0 TH |
8082 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8083 | struct cftype *cft) | |
d0b27fa7 | 8084 | { |
182446d0 | 8085 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8086 | } |
6d6bc0ad | 8087 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8088 | |
fe5c7cc2 | 8089 | static struct cftype cpu_files[] = { |
052f1dc7 | 8090 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8091 | { |
8092 | .name = "shares", | |
f4c753b7 PM |
8093 | .read_u64 = cpu_shares_read_u64, |
8094 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8095 | }, |
052f1dc7 | 8096 | #endif |
ab84d31e PT |
8097 | #ifdef CONFIG_CFS_BANDWIDTH |
8098 | { | |
8099 | .name = "cfs_quota_us", | |
8100 | .read_s64 = cpu_cfs_quota_read_s64, | |
8101 | .write_s64 = cpu_cfs_quota_write_s64, | |
8102 | }, | |
8103 | { | |
8104 | .name = "cfs_period_us", | |
8105 | .read_u64 = cpu_cfs_period_read_u64, | |
8106 | .write_u64 = cpu_cfs_period_write_u64, | |
8107 | }, | |
e8da1b18 NR |
8108 | { |
8109 | .name = "stat", | |
2da8ca82 | 8110 | .seq_show = cpu_stats_show, |
e8da1b18 | 8111 | }, |
ab84d31e | 8112 | #endif |
052f1dc7 | 8113 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8114 | { |
9f0c1e56 | 8115 | .name = "rt_runtime_us", |
06ecb27c PM |
8116 | .read_s64 = cpu_rt_runtime_read, |
8117 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8118 | }, |
d0b27fa7 PZ |
8119 | { |
8120 | .name = "rt_period_us", | |
f4c753b7 PM |
8121 | .read_u64 = cpu_rt_period_read_uint, |
8122 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8123 | }, |
052f1dc7 | 8124 | #endif |
4baf6e33 | 8125 | { } /* terminate */ |
68318b8e SV |
8126 | }; |
8127 | ||
073219e9 | 8128 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8129 | .css_alloc = cpu_cgroup_css_alloc, |
8130 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8131 | .css_online = cpu_cgroup_css_online, |
8132 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8133 | .can_attach = cpu_cgroup_can_attach, |
8134 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8135 | .exit = cpu_cgroup_exit, |
5577964e | 8136 | .legacy_cftypes = cpu_files, |
68318b8e SV |
8137 | .early_init = 1, |
8138 | }; | |
8139 | ||
052f1dc7 | 8140 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8141 | |
b637a328 PM |
8142 | void dump_cpu_task(int cpu) |
8143 | { | |
8144 | pr_info("Task dump for CPU %d:\n", cpu); | |
8145 | sched_show_task(cpu_curr(cpu)); | |
8146 | } |