]>
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); |
65cc8e48 | 336 | if (likely(rq == task_rq(p))) |
3a5c359a | 337 | return rq; |
05fa785c | 338 | raw_spin_unlock(&rq->lock); |
b29739f9 | 339 | } |
b29739f9 IM |
340 | } |
341 | ||
1da177e4 | 342 | /* |
0122ec5b | 343 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 344 | */ |
70b97a7f | 345 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 346 | __acquires(p->pi_lock) |
1da177e4 LT |
347 | __acquires(rq->lock) |
348 | { | |
70b97a7f | 349 | struct rq *rq; |
1da177e4 | 350 | |
3a5c359a | 351 | for (;;) { |
0122ec5b | 352 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 353 | rq = task_rq(p); |
05fa785c | 354 | raw_spin_lock(&rq->lock); |
65cc8e48 | 355 | if (likely(rq == task_rq(p))) |
3a5c359a | 356 | return rq; |
0122ec5b PZ |
357 | raw_spin_unlock(&rq->lock); |
358 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 359 | } |
1da177e4 LT |
360 | } |
361 | ||
a9957449 | 362 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
363 | __releases(rq->lock) |
364 | { | |
05fa785c | 365 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
366 | } |
367 | ||
0122ec5b PZ |
368 | static inline void |
369 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 370 | __releases(rq->lock) |
0122ec5b | 371 | __releases(p->pi_lock) |
1da177e4 | 372 | { |
0122ec5b PZ |
373 | raw_spin_unlock(&rq->lock); |
374 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
375 | } |
376 | ||
1da177e4 | 377 | /* |
cc2a73b5 | 378 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 379 | */ |
a9957449 | 380 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
381 | __acquires(rq->lock) |
382 | { | |
70b97a7f | 383 | struct rq *rq; |
1da177e4 LT |
384 | |
385 | local_irq_disable(); | |
386 | rq = this_rq(); | |
05fa785c | 387 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
388 | |
389 | return rq; | |
390 | } | |
391 | ||
8f4d37ec PZ |
392 | #ifdef CONFIG_SCHED_HRTICK |
393 | /* | |
394 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 395 | */ |
8f4d37ec | 396 | |
8f4d37ec PZ |
397 | static void hrtick_clear(struct rq *rq) |
398 | { | |
399 | if (hrtimer_active(&rq->hrtick_timer)) | |
400 | hrtimer_cancel(&rq->hrtick_timer); | |
401 | } | |
402 | ||
8f4d37ec PZ |
403 | /* |
404 | * High-resolution timer tick. | |
405 | * Runs from hardirq context with interrupts disabled. | |
406 | */ | |
407 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
408 | { | |
409 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
410 | ||
411 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
412 | ||
05fa785c | 413 | raw_spin_lock(&rq->lock); |
3e51f33f | 414 | update_rq_clock(rq); |
8f4d37ec | 415 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 416 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
417 | |
418 | return HRTIMER_NORESTART; | |
419 | } | |
420 | ||
95e904c7 | 421 | #ifdef CONFIG_SMP |
971ee28c PZ |
422 | |
423 | static int __hrtick_restart(struct rq *rq) | |
424 | { | |
425 | struct hrtimer *timer = &rq->hrtick_timer; | |
426 | ktime_t time = hrtimer_get_softexpires(timer); | |
427 | ||
428 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
429 | } | |
430 | ||
31656519 PZ |
431 | /* |
432 | * called from hardirq (IPI) context | |
433 | */ | |
434 | static void __hrtick_start(void *arg) | |
b328ca18 | 435 | { |
31656519 | 436 | struct rq *rq = arg; |
b328ca18 | 437 | |
05fa785c | 438 | raw_spin_lock(&rq->lock); |
971ee28c | 439 | __hrtick_restart(rq); |
31656519 | 440 | rq->hrtick_csd_pending = 0; |
05fa785c | 441 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
442 | } |
443 | ||
31656519 PZ |
444 | /* |
445 | * Called to set the hrtick timer state. | |
446 | * | |
447 | * called with rq->lock held and irqs disabled | |
448 | */ | |
029632fb | 449 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 450 | { |
31656519 PZ |
451 | struct hrtimer *timer = &rq->hrtick_timer; |
452 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 453 | |
cc584b21 | 454 | hrtimer_set_expires(timer, time); |
31656519 PZ |
455 | |
456 | if (rq == this_rq()) { | |
971ee28c | 457 | __hrtick_restart(rq); |
31656519 | 458 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 459 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
460 | rq->hrtick_csd_pending = 1; |
461 | } | |
b328ca18 PZ |
462 | } |
463 | ||
464 | static int | |
465 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
466 | { | |
467 | int cpu = (int)(long)hcpu; | |
468 | ||
469 | switch (action) { | |
470 | case CPU_UP_CANCELED: | |
471 | case CPU_UP_CANCELED_FROZEN: | |
472 | case CPU_DOWN_PREPARE: | |
473 | case CPU_DOWN_PREPARE_FROZEN: | |
474 | case CPU_DEAD: | |
475 | case CPU_DEAD_FROZEN: | |
31656519 | 476 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
477 | return NOTIFY_OK; |
478 | } | |
479 | ||
480 | return NOTIFY_DONE; | |
481 | } | |
482 | ||
fa748203 | 483 | static __init void init_hrtick(void) |
b328ca18 PZ |
484 | { |
485 | hotcpu_notifier(hotplug_hrtick, 0); | |
486 | } | |
31656519 PZ |
487 | #else |
488 | /* | |
489 | * Called to set the hrtick timer state. | |
490 | * | |
491 | * called with rq->lock held and irqs disabled | |
492 | */ | |
029632fb | 493 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 494 | { |
7f1e2ca9 | 495 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 496 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 497 | } |
b328ca18 | 498 | |
006c75f1 | 499 | static inline void init_hrtick(void) |
8f4d37ec | 500 | { |
8f4d37ec | 501 | } |
31656519 | 502 | #endif /* CONFIG_SMP */ |
8f4d37ec | 503 | |
31656519 | 504 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 505 | { |
31656519 PZ |
506 | #ifdef CONFIG_SMP |
507 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 508 | |
31656519 PZ |
509 | rq->hrtick_csd.flags = 0; |
510 | rq->hrtick_csd.func = __hrtick_start; | |
511 | rq->hrtick_csd.info = rq; | |
512 | #endif | |
8f4d37ec | 513 | |
31656519 PZ |
514 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
515 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 516 | } |
006c75f1 | 517 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
518 | static inline void hrtick_clear(struct rq *rq) |
519 | { | |
520 | } | |
521 | ||
8f4d37ec PZ |
522 | static inline void init_rq_hrtick(struct rq *rq) |
523 | { | |
524 | } | |
525 | ||
b328ca18 PZ |
526 | static inline void init_hrtick(void) |
527 | { | |
528 | } | |
006c75f1 | 529 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 530 | |
fd99f91a PZ |
531 | /* |
532 | * cmpxchg based fetch_or, macro so it works for different integer types | |
533 | */ | |
534 | #define fetch_or(ptr, val) \ | |
535 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
536 | for (;;) { \ | |
537 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
538 | if (__old == __val) \ | |
539 | break; \ | |
540 | __val = __old; \ | |
541 | } \ | |
542 | __old; \ | |
543 | }) | |
544 | ||
e3baac47 | 545 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
546 | /* |
547 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
548 | * this avoids any races wrt polling state changes and thereby avoids | |
549 | * spurious IPIs. | |
550 | */ | |
551 | static bool set_nr_and_not_polling(struct task_struct *p) | |
552 | { | |
553 | struct thread_info *ti = task_thread_info(p); | |
554 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
555 | } | |
e3baac47 PZ |
556 | |
557 | /* | |
558 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
559 | * | |
560 | * If this returns true, then the idle task promises to call | |
561 | * sched_ttwu_pending() and reschedule soon. | |
562 | */ | |
563 | static bool set_nr_if_polling(struct task_struct *p) | |
564 | { | |
565 | struct thread_info *ti = task_thread_info(p); | |
566 | typeof(ti->flags) old, val = ACCESS_ONCE(ti->flags); | |
567 | ||
568 | for (;;) { | |
569 | if (!(val & _TIF_POLLING_NRFLAG)) | |
570 | return false; | |
571 | if (val & _TIF_NEED_RESCHED) | |
572 | return true; | |
573 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
574 | if (old == val) | |
575 | break; | |
576 | val = old; | |
577 | } | |
578 | return true; | |
579 | } | |
580 | ||
fd99f91a PZ |
581 | #else |
582 | static bool set_nr_and_not_polling(struct task_struct *p) | |
583 | { | |
584 | set_tsk_need_resched(p); | |
585 | return true; | |
586 | } | |
e3baac47 PZ |
587 | |
588 | #ifdef CONFIG_SMP | |
589 | static bool set_nr_if_polling(struct task_struct *p) | |
590 | { | |
591 | return false; | |
592 | } | |
593 | #endif | |
fd99f91a PZ |
594 | #endif |
595 | ||
c24d20db | 596 | /* |
8875125e | 597 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
598 | * |
599 | * On UP this means the setting of the need_resched flag, on SMP it | |
600 | * might also involve a cross-CPU call to trigger the scheduler on | |
601 | * the target CPU. | |
602 | */ | |
8875125e | 603 | void resched_curr(struct rq *rq) |
c24d20db | 604 | { |
8875125e | 605 | struct task_struct *curr = rq->curr; |
c24d20db IM |
606 | int cpu; |
607 | ||
8875125e | 608 | lockdep_assert_held(&rq->lock); |
c24d20db | 609 | |
8875125e | 610 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
611 | return; |
612 | ||
8875125e | 613 | cpu = cpu_of(rq); |
fd99f91a | 614 | |
f27dde8d | 615 | if (cpu == smp_processor_id()) { |
8875125e | 616 | set_tsk_need_resched(curr); |
f27dde8d | 617 | set_preempt_need_resched(); |
c24d20db | 618 | return; |
f27dde8d | 619 | } |
c24d20db | 620 | |
8875125e | 621 | if (set_nr_and_not_polling(curr)) |
c24d20db | 622 | smp_send_reschedule(cpu); |
dfc68f29 AL |
623 | else |
624 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
625 | } |
626 | ||
029632fb | 627 | void resched_cpu(int cpu) |
c24d20db IM |
628 | { |
629 | struct rq *rq = cpu_rq(cpu); | |
630 | unsigned long flags; | |
631 | ||
05fa785c | 632 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 633 | return; |
8875125e | 634 | resched_curr(rq); |
05fa785c | 635 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 636 | } |
06d8308c | 637 | |
b021fe3e | 638 | #ifdef CONFIG_SMP |
3451d024 | 639 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
640 | /* |
641 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
642 | * from an idle cpu. This is good for power-savings. | |
643 | * | |
644 | * We don't do similar optimization for completely idle system, as | |
645 | * selecting an idle cpu will add more delays to the timers than intended | |
646 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
647 | */ | |
6201b4d6 | 648 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
649 | { |
650 | int cpu = smp_processor_id(); | |
651 | int i; | |
652 | struct sched_domain *sd; | |
653 | ||
6201b4d6 VK |
654 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
655 | return cpu; | |
656 | ||
057f3fad | 657 | rcu_read_lock(); |
83cd4fe2 | 658 | for_each_domain(cpu, sd) { |
057f3fad PZ |
659 | for_each_cpu(i, sched_domain_span(sd)) { |
660 | if (!idle_cpu(i)) { | |
661 | cpu = i; | |
662 | goto unlock; | |
663 | } | |
664 | } | |
83cd4fe2 | 665 | } |
057f3fad PZ |
666 | unlock: |
667 | rcu_read_unlock(); | |
83cd4fe2 VP |
668 | return cpu; |
669 | } | |
06d8308c TG |
670 | /* |
671 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
672 | * idle CPU then this timer might expire before the next timer event | |
673 | * which is scheduled to wake up that CPU. In case of a completely | |
674 | * idle system the next event might even be infinite time into the | |
675 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
676 | * leaves the inner idle loop so the newly added timer is taken into | |
677 | * account when the CPU goes back to idle and evaluates the timer | |
678 | * wheel for the next timer event. | |
679 | */ | |
1c20091e | 680 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
681 | { |
682 | struct rq *rq = cpu_rq(cpu); | |
683 | ||
684 | if (cpu == smp_processor_id()) | |
685 | return; | |
686 | ||
67b9ca70 | 687 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 688 | smp_send_reschedule(cpu); |
dfc68f29 AL |
689 | else |
690 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
691 | } |
692 | ||
c5bfece2 | 693 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 694 | { |
53c5fa16 FW |
695 | /* |
696 | * We just need the target to call irq_exit() and re-evaluate | |
697 | * the next tick. The nohz full kick at least implies that. | |
698 | * If needed we can still optimize that later with an | |
699 | * empty IRQ. | |
700 | */ | |
c5bfece2 | 701 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
702 | if (cpu != smp_processor_id() || |
703 | tick_nohz_tick_stopped()) | |
53c5fa16 | 704 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
705 | return true; |
706 | } | |
707 | ||
708 | return false; | |
709 | } | |
710 | ||
711 | void wake_up_nohz_cpu(int cpu) | |
712 | { | |
c5bfece2 | 713 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
714 | wake_up_idle_cpu(cpu); |
715 | } | |
716 | ||
ca38062e | 717 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 718 | { |
1c792db7 | 719 | int cpu = smp_processor_id(); |
873b4c65 VG |
720 | |
721 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
722 | return false; | |
723 | ||
724 | if (idle_cpu(cpu) && !need_resched()) | |
725 | return true; | |
726 | ||
727 | /* | |
728 | * We can't run Idle Load Balance on this CPU for this time so we | |
729 | * cancel it and clear NOHZ_BALANCE_KICK | |
730 | */ | |
731 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
732 | return false; | |
45bf76df IM |
733 | } |
734 | ||
3451d024 | 735 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 736 | |
ca38062e | 737 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 738 | { |
ca38062e | 739 | return false; |
2069dd75 PZ |
740 | } |
741 | ||
3451d024 | 742 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 743 | |
ce831b38 FW |
744 | #ifdef CONFIG_NO_HZ_FULL |
745 | bool sched_can_stop_tick(void) | |
746 | { | |
3882ec64 FW |
747 | /* |
748 | * More than one running task need preemption. | |
749 | * nr_running update is assumed to be visible | |
750 | * after IPI is sent from wakers. | |
751 | */ | |
541b8264 VK |
752 | if (this_rq()->nr_running > 1) |
753 | return false; | |
ce831b38 | 754 | |
541b8264 | 755 | return true; |
ce831b38 FW |
756 | } |
757 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 758 | |
029632fb | 759 | void sched_avg_update(struct rq *rq) |
18d95a28 | 760 | { |
e9e9250b PZ |
761 | s64 period = sched_avg_period(); |
762 | ||
78becc27 | 763 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
764 | /* |
765 | * Inline assembly required to prevent the compiler | |
766 | * optimising this loop into a divmod call. | |
767 | * See __iter_div_u64_rem() for another example of this. | |
768 | */ | |
769 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
770 | rq->age_stamp += period; |
771 | rq->rt_avg /= 2; | |
772 | } | |
18d95a28 PZ |
773 | } |
774 | ||
6d6bc0ad | 775 | #endif /* CONFIG_SMP */ |
18d95a28 | 776 | |
a790de99 PT |
777 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
778 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 779 | /* |
8277434e PT |
780 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
781 | * node and @up when leaving it for the final time. | |
782 | * | |
783 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 784 | */ |
029632fb | 785 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 786 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
787 | { |
788 | struct task_group *parent, *child; | |
eb755805 | 789 | int ret; |
c09595f6 | 790 | |
8277434e PT |
791 | parent = from; |
792 | ||
c09595f6 | 793 | down: |
eb755805 PZ |
794 | ret = (*down)(parent, data); |
795 | if (ret) | |
8277434e | 796 | goto out; |
c09595f6 PZ |
797 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
798 | parent = child; | |
799 | goto down; | |
800 | ||
801 | up: | |
802 | continue; | |
803 | } | |
eb755805 | 804 | ret = (*up)(parent, data); |
8277434e PT |
805 | if (ret || parent == from) |
806 | goto out; | |
c09595f6 PZ |
807 | |
808 | child = parent; | |
809 | parent = parent->parent; | |
810 | if (parent) | |
811 | goto up; | |
8277434e | 812 | out: |
eb755805 | 813 | return ret; |
c09595f6 PZ |
814 | } |
815 | ||
029632fb | 816 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 817 | { |
e2b245f8 | 818 | return 0; |
eb755805 | 819 | } |
18d95a28 PZ |
820 | #endif |
821 | ||
45bf76df IM |
822 | static void set_load_weight(struct task_struct *p) |
823 | { | |
f05998d4 NR |
824 | int prio = p->static_prio - MAX_RT_PRIO; |
825 | struct load_weight *load = &p->se.load; | |
826 | ||
dd41f596 IM |
827 | /* |
828 | * SCHED_IDLE tasks get minimal weight: | |
829 | */ | |
830 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 831 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 832 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
833 | return; |
834 | } | |
71f8bd46 | 835 | |
c8b28116 | 836 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 837 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
838 | } |
839 | ||
371fd7e7 | 840 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 841 | { |
a64692a3 | 842 | update_rq_clock(rq); |
43148951 | 843 | sched_info_queued(rq, p); |
371fd7e7 | 844 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
845 | } |
846 | ||
371fd7e7 | 847 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 848 | { |
a64692a3 | 849 | update_rq_clock(rq); |
43148951 | 850 | sched_info_dequeued(rq, p); |
371fd7e7 | 851 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
852 | } |
853 | ||
029632fb | 854 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
855 | { |
856 | if (task_contributes_to_load(p)) | |
857 | rq->nr_uninterruptible--; | |
858 | ||
371fd7e7 | 859 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
860 | } |
861 | ||
029632fb | 862 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
863 | { |
864 | if (task_contributes_to_load(p)) | |
865 | rq->nr_uninterruptible++; | |
866 | ||
371fd7e7 | 867 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
868 | } |
869 | ||
fe44d621 | 870 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 871 | { |
095c0aa8 GC |
872 | /* |
873 | * In theory, the compile should just see 0 here, and optimize out the call | |
874 | * to sched_rt_avg_update. But I don't trust it... | |
875 | */ | |
876 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
877 | s64 steal = 0, irq_delta = 0; | |
878 | #endif | |
879 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 880 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
881 | |
882 | /* | |
883 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
884 | * this case when a previous update_rq_clock() happened inside a | |
885 | * {soft,}irq region. | |
886 | * | |
887 | * When this happens, we stop ->clock_task and only update the | |
888 | * prev_irq_time stamp to account for the part that fit, so that a next | |
889 | * update will consume the rest. This ensures ->clock_task is | |
890 | * monotonic. | |
891 | * | |
892 | * It does however cause some slight miss-attribution of {soft,}irq | |
893 | * time, a more accurate solution would be to update the irq_time using | |
894 | * the current rq->clock timestamp, except that would require using | |
895 | * atomic ops. | |
896 | */ | |
897 | if (irq_delta > delta) | |
898 | irq_delta = delta; | |
899 | ||
900 | rq->prev_irq_time += irq_delta; | |
901 | delta -= irq_delta; | |
095c0aa8 GC |
902 | #endif |
903 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 904 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
905 | steal = paravirt_steal_clock(cpu_of(rq)); |
906 | steal -= rq->prev_steal_time_rq; | |
907 | ||
908 | if (unlikely(steal > delta)) | |
909 | steal = delta; | |
910 | ||
095c0aa8 | 911 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
912 | delta -= steal; |
913 | } | |
914 | #endif | |
915 | ||
fe44d621 PZ |
916 | rq->clock_task += delta; |
917 | ||
095c0aa8 | 918 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 919 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
920 | sched_rt_avg_update(rq, irq_delta + steal); |
921 | #endif | |
aa483808 VP |
922 | } |
923 | ||
34f971f6 PZ |
924 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
925 | { | |
926 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
927 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
928 | ||
929 | if (stop) { | |
930 | /* | |
931 | * Make it appear like a SCHED_FIFO task, its something | |
932 | * userspace knows about and won't get confused about. | |
933 | * | |
934 | * Also, it will make PI more or less work without too | |
935 | * much confusion -- but then, stop work should not | |
936 | * rely on PI working anyway. | |
937 | */ | |
938 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
939 | ||
940 | stop->sched_class = &stop_sched_class; | |
941 | } | |
942 | ||
943 | cpu_rq(cpu)->stop = stop; | |
944 | ||
945 | if (old_stop) { | |
946 | /* | |
947 | * Reset it back to a normal scheduling class so that | |
948 | * it can die in pieces. | |
949 | */ | |
950 | old_stop->sched_class = &rt_sched_class; | |
951 | } | |
952 | } | |
953 | ||
14531189 | 954 | /* |
dd41f596 | 955 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 956 | */ |
14531189 IM |
957 | static inline int __normal_prio(struct task_struct *p) |
958 | { | |
dd41f596 | 959 | return p->static_prio; |
14531189 IM |
960 | } |
961 | ||
b29739f9 IM |
962 | /* |
963 | * Calculate the expected normal priority: i.e. priority | |
964 | * without taking RT-inheritance into account. Might be | |
965 | * boosted by interactivity modifiers. Changes upon fork, | |
966 | * setprio syscalls, and whenever the interactivity | |
967 | * estimator recalculates. | |
968 | */ | |
36c8b586 | 969 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
970 | { |
971 | int prio; | |
972 | ||
aab03e05 DF |
973 | if (task_has_dl_policy(p)) |
974 | prio = MAX_DL_PRIO-1; | |
975 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
976 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
977 | else | |
978 | prio = __normal_prio(p); | |
979 | return prio; | |
980 | } | |
981 | ||
982 | /* | |
983 | * Calculate the current priority, i.e. the priority | |
984 | * taken into account by the scheduler. This value might | |
985 | * be boosted by RT tasks, or might be boosted by | |
986 | * interactivity modifiers. Will be RT if the task got | |
987 | * RT-boosted. If not then it returns p->normal_prio. | |
988 | */ | |
36c8b586 | 989 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
990 | { |
991 | p->normal_prio = normal_prio(p); | |
992 | /* | |
993 | * If we are RT tasks or we were boosted to RT priority, | |
994 | * keep the priority unchanged. Otherwise, update priority | |
995 | * to the normal priority: | |
996 | */ | |
997 | if (!rt_prio(p->prio)) | |
998 | return p->normal_prio; | |
999 | return p->prio; | |
1000 | } | |
1001 | ||
1da177e4 LT |
1002 | /** |
1003 | * task_curr - is this task currently executing on a CPU? | |
1004 | * @p: the task in question. | |
e69f6186 YB |
1005 | * |
1006 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1007 | */ |
36c8b586 | 1008 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1009 | { |
1010 | return cpu_curr(task_cpu(p)) == p; | |
1011 | } | |
1012 | ||
cb469845 SR |
1013 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1014 | const struct sched_class *prev_class, | |
da7a735e | 1015 | int oldprio) |
cb469845 SR |
1016 | { |
1017 | if (prev_class != p->sched_class) { | |
1018 | if (prev_class->switched_from) | |
da7a735e PZ |
1019 | prev_class->switched_from(rq, p); |
1020 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 1021 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1022 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1023 | } |
1024 | ||
029632fb | 1025 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1026 | { |
1027 | const struct sched_class *class; | |
1028 | ||
1029 | if (p->sched_class == rq->curr->sched_class) { | |
1030 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1031 | } else { | |
1032 | for_each_class(class) { | |
1033 | if (class == rq->curr->sched_class) | |
1034 | break; | |
1035 | if (class == p->sched_class) { | |
8875125e | 1036 | resched_curr(rq); |
1e5a7405 PZ |
1037 | break; |
1038 | } | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * A queue event has occurred, and we're going to schedule. In | |
1044 | * this case, we can save a useless back to back clock update. | |
1045 | */ | |
fd2f4419 | 1046 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1047 | rq->skip_clock_update = 1; |
1048 | } | |
1049 | ||
1da177e4 | 1050 | #ifdef CONFIG_SMP |
dd41f596 | 1051 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1052 | { |
e2912009 PZ |
1053 | #ifdef CONFIG_SCHED_DEBUG |
1054 | /* | |
1055 | * We should never call set_task_cpu() on a blocked task, | |
1056 | * ttwu() will sort out the placement. | |
1057 | */ | |
077614ee | 1058 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1059 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1060 | |
1061 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1062 | /* |
1063 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1064 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1065 | * | |
1066 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1067 | * see task_group(). |
6c6c54e1 PZ |
1068 | * |
1069 | * Furthermore, all task_rq users should acquire both locks, see | |
1070 | * task_rq_lock(). | |
1071 | */ | |
0122ec5b PZ |
1072 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1073 | lockdep_is_held(&task_rq(p)->lock))); | |
1074 | #endif | |
e2912009 PZ |
1075 | #endif |
1076 | ||
de1d7286 | 1077 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1078 | |
0c69774e | 1079 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1080 | if (p->sched_class->migrate_task_rq) |
1081 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1082 | p->se.nr_migrations++; |
a8b0ca17 | 1083 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1084 | } |
dd41f596 IM |
1085 | |
1086 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1087 | } |
1088 | ||
ac66f547 PZ |
1089 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1090 | { | |
1091 | if (p->on_rq) { | |
1092 | struct rq *src_rq, *dst_rq; | |
1093 | ||
1094 | src_rq = task_rq(p); | |
1095 | dst_rq = cpu_rq(cpu); | |
1096 | ||
1097 | deactivate_task(src_rq, p, 0); | |
1098 | set_task_cpu(p, cpu); | |
1099 | activate_task(dst_rq, p, 0); | |
1100 | check_preempt_curr(dst_rq, p, 0); | |
1101 | } else { | |
1102 | /* | |
1103 | * Task isn't running anymore; make it appear like we migrated | |
1104 | * it before it went to sleep. This means on wakeup we make the | |
1105 | * previous cpu our targer instead of where it really is. | |
1106 | */ | |
1107 | p->wake_cpu = cpu; | |
1108 | } | |
1109 | } | |
1110 | ||
1111 | struct migration_swap_arg { | |
1112 | struct task_struct *src_task, *dst_task; | |
1113 | int src_cpu, dst_cpu; | |
1114 | }; | |
1115 | ||
1116 | static int migrate_swap_stop(void *data) | |
1117 | { | |
1118 | struct migration_swap_arg *arg = data; | |
1119 | struct rq *src_rq, *dst_rq; | |
1120 | int ret = -EAGAIN; | |
1121 | ||
1122 | src_rq = cpu_rq(arg->src_cpu); | |
1123 | dst_rq = cpu_rq(arg->dst_cpu); | |
1124 | ||
74602315 PZ |
1125 | double_raw_lock(&arg->src_task->pi_lock, |
1126 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1127 | double_rq_lock(src_rq, dst_rq); |
1128 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1129 | goto unlock; | |
1130 | ||
1131 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1132 | goto unlock; | |
1133 | ||
1134 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1135 | goto unlock; | |
1136 | ||
1137 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1138 | goto unlock; | |
1139 | ||
1140 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1141 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1142 | ||
1143 | ret = 0; | |
1144 | ||
1145 | unlock: | |
1146 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1147 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1148 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1149 | |
1150 | return ret; | |
1151 | } | |
1152 | ||
1153 | /* | |
1154 | * Cross migrate two tasks | |
1155 | */ | |
1156 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1157 | { | |
1158 | struct migration_swap_arg arg; | |
1159 | int ret = -EINVAL; | |
1160 | ||
ac66f547 PZ |
1161 | arg = (struct migration_swap_arg){ |
1162 | .src_task = cur, | |
1163 | .src_cpu = task_cpu(cur), | |
1164 | .dst_task = p, | |
1165 | .dst_cpu = task_cpu(p), | |
1166 | }; | |
1167 | ||
1168 | if (arg.src_cpu == arg.dst_cpu) | |
1169 | goto out; | |
1170 | ||
6acce3ef PZ |
1171 | /* |
1172 | * These three tests are all lockless; this is OK since all of them | |
1173 | * will be re-checked with proper locks held further down the line. | |
1174 | */ | |
ac66f547 PZ |
1175 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1176 | goto out; | |
1177 | ||
1178 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1179 | goto out; | |
1180 | ||
1181 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1182 | goto out; | |
1183 | ||
286549dc | 1184 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1185 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1186 | ||
1187 | out: | |
ac66f547 PZ |
1188 | return ret; |
1189 | } | |
1190 | ||
969c7921 | 1191 | struct migration_arg { |
36c8b586 | 1192 | struct task_struct *task; |
1da177e4 | 1193 | int dest_cpu; |
70b97a7f | 1194 | }; |
1da177e4 | 1195 | |
969c7921 TH |
1196 | static int migration_cpu_stop(void *data); |
1197 | ||
1da177e4 LT |
1198 | /* |
1199 | * wait_task_inactive - wait for a thread to unschedule. | |
1200 | * | |
85ba2d86 RM |
1201 | * If @match_state is nonzero, it's the @p->state value just checked and |
1202 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1203 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1204 | * we return a positive number (its total switch count). If a second call | |
1205 | * a short while later returns the same number, the caller can be sure that | |
1206 | * @p has remained unscheduled the whole time. | |
1207 | * | |
1da177e4 LT |
1208 | * The caller must ensure that the task *will* unschedule sometime soon, |
1209 | * else this function might spin for a *long* time. This function can't | |
1210 | * be called with interrupts off, or it may introduce deadlock with | |
1211 | * smp_call_function() if an IPI is sent by the same process we are | |
1212 | * waiting to become inactive. | |
1213 | */ | |
85ba2d86 | 1214 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1215 | { |
1216 | unsigned long flags; | |
dd41f596 | 1217 | int running, on_rq; |
85ba2d86 | 1218 | unsigned long ncsw; |
70b97a7f | 1219 | struct rq *rq; |
1da177e4 | 1220 | |
3a5c359a AK |
1221 | for (;;) { |
1222 | /* | |
1223 | * We do the initial early heuristics without holding | |
1224 | * any task-queue locks at all. We'll only try to get | |
1225 | * the runqueue lock when things look like they will | |
1226 | * work out! | |
1227 | */ | |
1228 | rq = task_rq(p); | |
fa490cfd | 1229 | |
3a5c359a AK |
1230 | /* |
1231 | * If the task is actively running on another CPU | |
1232 | * still, just relax and busy-wait without holding | |
1233 | * any locks. | |
1234 | * | |
1235 | * NOTE! Since we don't hold any locks, it's not | |
1236 | * even sure that "rq" stays as the right runqueue! | |
1237 | * But we don't care, since "task_running()" will | |
1238 | * return false if the runqueue has changed and p | |
1239 | * is actually now running somewhere else! | |
1240 | */ | |
85ba2d86 RM |
1241 | while (task_running(rq, p)) { |
1242 | if (match_state && unlikely(p->state != match_state)) | |
1243 | return 0; | |
3a5c359a | 1244 | cpu_relax(); |
85ba2d86 | 1245 | } |
fa490cfd | 1246 | |
3a5c359a AK |
1247 | /* |
1248 | * Ok, time to look more closely! We need the rq | |
1249 | * lock now, to be *sure*. If we're wrong, we'll | |
1250 | * just go back and repeat. | |
1251 | */ | |
1252 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1253 | trace_sched_wait_task(p); |
3a5c359a | 1254 | running = task_running(rq, p); |
fd2f4419 | 1255 | on_rq = p->on_rq; |
85ba2d86 | 1256 | ncsw = 0; |
f31e11d8 | 1257 | if (!match_state || p->state == match_state) |
93dcf55f | 1258 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1259 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1260 | |
85ba2d86 RM |
1261 | /* |
1262 | * If it changed from the expected state, bail out now. | |
1263 | */ | |
1264 | if (unlikely(!ncsw)) | |
1265 | break; | |
1266 | ||
3a5c359a AK |
1267 | /* |
1268 | * Was it really running after all now that we | |
1269 | * checked with the proper locks actually held? | |
1270 | * | |
1271 | * Oops. Go back and try again.. | |
1272 | */ | |
1273 | if (unlikely(running)) { | |
1274 | cpu_relax(); | |
1275 | continue; | |
1276 | } | |
fa490cfd | 1277 | |
3a5c359a AK |
1278 | /* |
1279 | * It's not enough that it's not actively running, | |
1280 | * it must be off the runqueue _entirely_, and not | |
1281 | * preempted! | |
1282 | * | |
80dd99b3 | 1283 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1284 | * running right now), it's preempted, and we should |
1285 | * yield - it could be a while. | |
1286 | */ | |
1287 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1288 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1289 | ||
1290 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1291 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1292 | continue; |
1293 | } | |
fa490cfd | 1294 | |
3a5c359a AK |
1295 | /* |
1296 | * Ahh, all good. It wasn't running, and it wasn't | |
1297 | * runnable, which means that it will never become | |
1298 | * running in the future either. We're all done! | |
1299 | */ | |
1300 | break; | |
1301 | } | |
85ba2d86 RM |
1302 | |
1303 | return ncsw; | |
1da177e4 LT |
1304 | } |
1305 | ||
1306 | /*** | |
1307 | * kick_process - kick a running thread to enter/exit the kernel | |
1308 | * @p: the to-be-kicked thread | |
1309 | * | |
1310 | * Cause a process which is running on another CPU to enter | |
1311 | * kernel-mode, without any delay. (to get signals handled.) | |
1312 | * | |
25985edc | 1313 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1314 | * because all it wants to ensure is that the remote task enters |
1315 | * the kernel. If the IPI races and the task has been migrated | |
1316 | * to another CPU then no harm is done and the purpose has been | |
1317 | * achieved as well. | |
1318 | */ | |
36c8b586 | 1319 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1320 | { |
1321 | int cpu; | |
1322 | ||
1323 | preempt_disable(); | |
1324 | cpu = task_cpu(p); | |
1325 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1326 | smp_send_reschedule(cpu); | |
1327 | preempt_enable(); | |
1328 | } | |
b43e3521 | 1329 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1330 | #endif /* CONFIG_SMP */ |
1da177e4 | 1331 | |
970b13ba | 1332 | #ifdef CONFIG_SMP |
30da688e | 1333 | /* |
013fdb80 | 1334 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1335 | */ |
5da9a0fb PZ |
1336 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1337 | { | |
aa00d89c TC |
1338 | int nid = cpu_to_node(cpu); |
1339 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1340 | enum { cpuset, possible, fail } state = cpuset; |
1341 | int dest_cpu; | |
5da9a0fb | 1342 | |
aa00d89c TC |
1343 | /* |
1344 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1345 | * will return -1. There is no cpu on the node, and we should | |
1346 | * select the cpu on the other node. | |
1347 | */ | |
1348 | if (nid != -1) { | |
1349 | nodemask = cpumask_of_node(nid); | |
1350 | ||
1351 | /* Look for allowed, online CPU in same node. */ | |
1352 | for_each_cpu(dest_cpu, nodemask) { | |
1353 | if (!cpu_online(dest_cpu)) | |
1354 | continue; | |
1355 | if (!cpu_active(dest_cpu)) | |
1356 | continue; | |
1357 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1358 | return dest_cpu; | |
1359 | } | |
2baab4e9 | 1360 | } |
5da9a0fb | 1361 | |
2baab4e9 PZ |
1362 | for (;;) { |
1363 | /* Any allowed, online CPU? */ | |
e3831edd | 1364 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1365 | if (!cpu_online(dest_cpu)) |
1366 | continue; | |
1367 | if (!cpu_active(dest_cpu)) | |
1368 | continue; | |
1369 | goto out; | |
1370 | } | |
5da9a0fb | 1371 | |
2baab4e9 PZ |
1372 | switch (state) { |
1373 | case cpuset: | |
1374 | /* No more Mr. Nice Guy. */ | |
1375 | cpuset_cpus_allowed_fallback(p); | |
1376 | state = possible; | |
1377 | break; | |
1378 | ||
1379 | case possible: | |
1380 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1381 | state = fail; | |
1382 | break; | |
1383 | ||
1384 | case fail: | |
1385 | BUG(); | |
1386 | break; | |
1387 | } | |
1388 | } | |
1389 | ||
1390 | out: | |
1391 | if (state != cpuset) { | |
1392 | /* | |
1393 | * Don't tell them about moving exiting tasks or | |
1394 | * kernel threads (both mm NULL), since they never | |
1395 | * leave kernel. | |
1396 | */ | |
1397 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1398 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1399 | task_pid_nr(p), p->comm, cpu); |
1400 | } | |
5da9a0fb PZ |
1401 | } |
1402 | ||
1403 | return dest_cpu; | |
1404 | } | |
1405 | ||
e2912009 | 1406 | /* |
013fdb80 | 1407 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1408 | */ |
970b13ba | 1409 | static inline |
ac66f547 | 1410 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1411 | { |
ac66f547 | 1412 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1413 | |
1414 | /* | |
1415 | * In order not to call set_task_cpu() on a blocking task we need | |
1416 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1417 | * cpu. | |
1418 | * | |
1419 | * Since this is common to all placement strategies, this lives here. | |
1420 | * | |
1421 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1422 | * not worry about this generic constraint ] | |
1423 | */ | |
fa17b507 | 1424 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1425 | !cpu_online(cpu))) |
5da9a0fb | 1426 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1427 | |
1428 | return cpu; | |
970b13ba | 1429 | } |
09a40af5 MG |
1430 | |
1431 | static void update_avg(u64 *avg, u64 sample) | |
1432 | { | |
1433 | s64 diff = sample - *avg; | |
1434 | *avg += diff >> 3; | |
1435 | } | |
970b13ba PZ |
1436 | #endif |
1437 | ||
d7c01d27 | 1438 | static void |
b84cb5df | 1439 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1440 | { |
d7c01d27 | 1441 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1442 | struct rq *rq = this_rq(); |
1443 | ||
d7c01d27 PZ |
1444 | #ifdef CONFIG_SMP |
1445 | int this_cpu = smp_processor_id(); | |
1446 | ||
1447 | if (cpu == this_cpu) { | |
1448 | schedstat_inc(rq, ttwu_local); | |
1449 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1450 | } else { | |
1451 | struct sched_domain *sd; | |
1452 | ||
1453 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1454 | rcu_read_lock(); |
d7c01d27 PZ |
1455 | for_each_domain(this_cpu, sd) { |
1456 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1457 | schedstat_inc(sd, ttwu_wake_remote); | |
1458 | break; | |
1459 | } | |
1460 | } | |
057f3fad | 1461 | rcu_read_unlock(); |
d7c01d27 | 1462 | } |
f339b9dc PZ |
1463 | |
1464 | if (wake_flags & WF_MIGRATED) | |
1465 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1466 | ||
d7c01d27 PZ |
1467 | #endif /* CONFIG_SMP */ |
1468 | ||
1469 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1470 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1471 | |
1472 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1473 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1474 | |
d7c01d27 PZ |
1475 | #endif /* CONFIG_SCHEDSTATS */ |
1476 | } | |
1477 | ||
1478 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1479 | { | |
9ed3811a | 1480 | activate_task(rq, p, en_flags); |
fd2f4419 | 1481 | p->on_rq = 1; |
c2f7115e PZ |
1482 | |
1483 | /* if a worker is waking up, notify workqueue */ | |
1484 | if (p->flags & PF_WQ_WORKER) | |
1485 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1486 | } |
1487 | ||
23f41eeb PZ |
1488 | /* |
1489 | * Mark the task runnable and perform wakeup-preemption. | |
1490 | */ | |
89363381 | 1491 | static void |
23f41eeb | 1492 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1493 | { |
9ed3811a | 1494 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1495 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1496 | |
1497 | p->state = TASK_RUNNING; | |
1498 | #ifdef CONFIG_SMP | |
1499 | if (p->sched_class->task_woken) | |
1500 | p->sched_class->task_woken(rq, p); | |
1501 | ||
e69c6341 | 1502 | if (rq->idle_stamp) { |
78becc27 | 1503 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1504 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1505 | |
abfafa54 JL |
1506 | update_avg(&rq->avg_idle, delta); |
1507 | ||
1508 | if (rq->avg_idle > max) | |
9ed3811a | 1509 | rq->avg_idle = max; |
abfafa54 | 1510 | |
9ed3811a TH |
1511 | rq->idle_stamp = 0; |
1512 | } | |
1513 | #endif | |
1514 | } | |
1515 | ||
c05fbafb PZ |
1516 | static void |
1517 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1518 | { | |
1519 | #ifdef CONFIG_SMP | |
1520 | if (p->sched_contributes_to_load) | |
1521 | rq->nr_uninterruptible--; | |
1522 | #endif | |
1523 | ||
1524 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1525 | ttwu_do_wakeup(rq, p, wake_flags); | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1530 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1531 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1532 | * the task is still ->on_rq. | |
1533 | */ | |
1534 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1535 | { | |
1536 | struct rq *rq; | |
1537 | int ret = 0; | |
1538 | ||
1539 | rq = __task_rq_lock(p); | |
1540 | if (p->on_rq) { | |
1ad4ec0d FW |
1541 | /* check_preempt_curr() may use rq clock */ |
1542 | update_rq_clock(rq); | |
c05fbafb PZ |
1543 | ttwu_do_wakeup(rq, p, wake_flags); |
1544 | ret = 1; | |
1545 | } | |
1546 | __task_rq_unlock(rq); | |
1547 | ||
1548 | return ret; | |
1549 | } | |
1550 | ||
317f3941 | 1551 | #ifdef CONFIG_SMP |
e3baac47 | 1552 | void sched_ttwu_pending(void) |
317f3941 PZ |
1553 | { |
1554 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1555 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1556 | struct task_struct *p; | |
e3baac47 | 1557 | unsigned long flags; |
317f3941 | 1558 | |
e3baac47 PZ |
1559 | if (!llist) |
1560 | return; | |
1561 | ||
1562 | raw_spin_lock_irqsave(&rq->lock, flags); | |
317f3941 | 1563 | |
fa14ff4a PZ |
1564 | while (llist) { |
1565 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1566 | llist = llist_next(llist); | |
317f3941 PZ |
1567 | ttwu_do_activate(rq, p, 0); |
1568 | } | |
1569 | ||
e3baac47 | 1570 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1571 | } |
1572 | ||
1573 | void scheduler_ipi(void) | |
1574 | { | |
f27dde8d PZ |
1575 | /* |
1576 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1577 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1578 | * this IPI. | |
1579 | */ | |
8cb75e0c | 1580 | preempt_fold_need_resched(); |
f27dde8d | 1581 | |
fd2ac4f4 | 1582 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1583 | return; |
1584 | ||
1585 | /* | |
1586 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1587 | * traditionally all their work was done from the interrupt return | |
1588 | * path. Now that we actually do some work, we need to make sure | |
1589 | * we do call them. | |
1590 | * | |
1591 | * Some archs already do call them, luckily irq_enter/exit nest | |
1592 | * properly. | |
1593 | * | |
1594 | * Arguably we should visit all archs and update all handlers, | |
1595 | * however a fair share of IPIs are still resched only so this would | |
1596 | * somewhat pessimize the simple resched case. | |
1597 | */ | |
1598 | irq_enter(); | |
fa14ff4a | 1599 | sched_ttwu_pending(); |
ca38062e SS |
1600 | |
1601 | /* | |
1602 | * Check if someone kicked us for doing the nohz idle load balance. | |
1603 | */ | |
873b4c65 | 1604 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1605 | this_rq()->idle_balance = 1; |
ca38062e | 1606 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1607 | } |
c5d753a5 | 1608 | irq_exit(); |
317f3941 PZ |
1609 | } |
1610 | ||
1611 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1612 | { | |
e3baac47 PZ |
1613 | struct rq *rq = cpu_rq(cpu); |
1614 | ||
1615 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { | |
1616 | if (!set_nr_if_polling(rq->idle)) | |
1617 | smp_send_reschedule(cpu); | |
1618 | else | |
1619 | trace_sched_wake_idle_without_ipi(cpu); | |
1620 | } | |
317f3941 | 1621 | } |
d6aa8f85 | 1622 | |
39be3501 | 1623 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1624 | { |
1625 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1626 | } | |
d6aa8f85 | 1627 | #endif /* CONFIG_SMP */ |
317f3941 | 1628 | |
c05fbafb PZ |
1629 | static void ttwu_queue(struct task_struct *p, int cpu) |
1630 | { | |
1631 | struct rq *rq = cpu_rq(cpu); | |
1632 | ||
17d9f311 | 1633 | #if defined(CONFIG_SMP) |
39be3501 | 1634 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1635 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1636 | ttwu_queue_remote(p, cpu); |
1637 | return; | |
1638 | } | |
1639 | #endif | |
1640 | ||
c05fbafb PZ |
1641 | raw_spin_lock(&rq->lock); |
1642 | ttwu_do_activate(rq, p, 0); | |
1643 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1644 | } |
1645 | ||
1646 | /** | |
1da177e4 | 1647 | * try_to_wake_up - wake up a thread |
9ed3811a | 1648 | * @p: the thread to be awakened |
1da177e4 | 1649 | * @state: the mask of task states that can be woken |
9ed3811a | 1650 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1651 | * |
1652 | * Put it on the run-queue if it's not already there. The "current" | |
1653 | * thread is always on the run-queue (except when the actual | |
1654 | * re-schedule is in progress), and as such you're allowed to do | |
1655 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1656 | * runnable without the overhead of this. | |
1657 | * | |
e69f6186 | 1658 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1659 | * or @state didn't match @p's state. |
1da177e4 | 1660 | */ |
e4a52bcb PZ |
1661 | static int |
1662 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1663 | { |
1da177e4 | 1664 | unsigned long flags; |
c05fbafb | 1665 | int cpu, success = 0; |
2398f2c6 | 1666 | |
e0acd0a6 ON |
1667 | /* |
1668 | * If we are going to wake up a thread waiting for CONDITION we | |
1669 | * need to ensure that CONDITION=1 done by the caller can not be | |
1670 | * reordered with p->state check below. This pairs with mb() in | |
1671 | * set_current_state() the waiting thread does. | |
1672 | */ | |
1673 | smp_mb__before_spinlock(); | |
013fdb80 | 1674 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1675 | if (!(p->state & state)) |
1da177e4 LT |
1676 | goto out; |
1677 | ||
c05fbafb | 1678 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1679 | cpu = task_cpu(p); |
1da177e4 | 1680 | |
c05fbafb PZ |
1681 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1682 | goto stat; | |
1da177e4 | 1683 | |
1da177e4 | 1684 | #ifdef CONFIG_SMP |
e9c84311 | 1685 | /* |
c05fbafb PZ |
1686 | * If the owning (remote) cpu is still in the middle of schedule() with |
1687 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1688 | */ |
f3e94786 | 1689 | while (p->on_cpu) |
e4a52bcb | 1690 | cpu_relax(); |
0970d299 | 1691 | /* |
e4a52bcb | 1692 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1693 | */ |
e4a52bcb | 1694 | smp_rmb(); |
1da177e4 | 1695 | |
a8e4f2ea | 1696 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1697 | p->state = TASK_WAKING; |
e7693a36 | 1698 | |
e4a52bcb | 1699 | if (p->sched_class->task_waking) |
74f8e4b2 | 1700 | p->sched_class->task_waking(p); |
efbbd05a | 1701 | |
ac66f547 | 1702 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1703 | if (task_cpu(p) != cpu) { |
1704 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1705 | set_task_cpu(p, cpu); |
f339b9dc | 1706 | } |
1da177e4 | 1707 | #endif /* CONFIG_SMP */ |
1da177e4 | 1708 | |
c05fbafb PZ |
1709 | ttwu_queue(p, cpu); |
1710 | stat: | |
b84cb5df | 1711 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1712 | out: |
013fdb80 | 1713 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1714 | |
1715 | return success; | |
1716 | } | |
1717 | ||
21aa9af0 TH |
1718 | /** |
1719 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1720 | * @p: the thread to be awakened | |
1721 | * | |
2acca55e | 1722 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1723 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1724 | * the current task. |
21aa9af0 TH |
1725 | */ |
1726 | static void try_to_wake_up_local(struct task_struct *p) | |
1727 | { | |
1728 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1729 | |
383efcd0 TH |
1730 | if (WARN_ON_ONCE(rq != this_rq()) || |
1731 | WARN_ON_ONCE(p == current)) | |
1732 | return; | |
1733 | ||
21aa9af0 TH |
1734 | lockdep_assert_held(&rq->lock); |
1735 | ||
2acca55e PZ |
1736 | if (!raw_spin_trylock(&p->pi_lock)) { |
1737 | raw_spin_unlock(&rq->lock); | |
1738 | raw_spin_lock(&p->pi_lock); | |
1739 | raw_spin_lock(&rq->lock); | |
1740 | } | |
1741 | ||
21aa9af0 | 1742 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1743 | goto out; |
21aa9af0 | 1744 | |
fd2f4419 | 1745 | if (!p->on_rq) |
d7c01d27 PZ |
1746 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1747 | ||
23f41eeb | 1748 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1749 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1750 | out: |
1751 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1752 | } |
1753 | ||
50fa610a DH |
1754 | /** |
1755 | * wake_up_process - Wake up a specific process | |
1756 | * @p: The process to be woken up. | |
1757 | * | |
1758 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1759 | * processes. |
1760 | * | |
1761 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1762 | * |
1763 | * It may be assumed that this function implies a write memory barrier before | |
1764 | * changing the task state if and only if any tasks are woken up. | |
1765 | */ | |
7ad5b3a5 | 1766 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1767 | { |
9067ac85 ON |
1768 | WARN_ON(task_is_stopped_or_traced(p)); |
1769 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1770 | } |
1da177e4 LT |
1771 | EXPORT_SYMBOL(wake_up_process); |
1772 | ||
7ad5b3a5 | 1773 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1774 | { |
1775 | return try_to_wake_up(p, state, 0); | |
1776 | } | |
1777 | ||
1da177e4 LT |
1778 | /* |
1779 | * Perform scheduler related setup for a newly forked process p. | |
1780 | * p is forked by current. | |
dd41f596 IM |
1781 | * |
1782 | * __sched_fork() is basic setup used by init_idle() too: | |
1783 | */ | |
5e1576ed | 1784 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1785 | { |
fd2f4419 PZ |
1786 | p->on_rq = 0; |
1787 | ||
1788 | p->se.on_rq = 0; | |
dd41f596 IM |
1789 | p->se.exec_start = 0; |
1790 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1791 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1792 | p->se.nr_migrations = 0; |
da7a735e | 1793 | p->se.vruntime = 0; |
fd2f4419 | 1794 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1795 | |
1796 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1797 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1798 | #endif |
476d139c | 1799 | |
aab03e05 DF |
1800 | RB_CLEAR_NODE(&p->dl.rb_node); |
1801 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1802 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1803 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1804 | p->dl.dl_period = 0; |
aab03e05 DF |
1805 | p->dl.flags = 0; |
1806 | ||
fa717060 | 1807 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1808 | |
e107be36 AK |
1809 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1810 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1811 | #endif | |
cbee9f88 PZ |
1812 | |
1813 | #ifdef CONFIG_NUMA_BALANCING | |
1814 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1815 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1816 | p->mm->numa_scan_seq = 0; |
1817 | } | |
1818 | ||
5e1576ed RR |
1819 | if (clone_flags & CLONE_VM) |
1820 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1821 | else | |
1822 | p->numa_preferred_nid = -1; | |
1823 | ||
cbee9f88 PZ |
1824 | p->node_stamp = 0ULL; |
1825 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1826 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1827 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1828 | p->numa_faults_memory = NULL; |
1829 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1830 | p->last_task_numa_placement = 0; |
1831 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1832 | |
1833 | INIT_LIST_HEAD(&p->numa_entry); | |
1834 | p->numa_group = NULL; | |
cbee9f88 | 1835 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1836 | } |
1837 | ||
1a687c2e | 1838 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1839 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1840 | void set_numabalancing_state(bool enabled) |
1841 | { | |
1842 | if (enabled) | |
1843 | sched_feat_set("NUMA"); | |
1844 | else | |
1845 | sched_feat_set("NO_NUMA"); | |
1846 | } | |
3105b86a MG |
1847 | #else |
1848 | __read_mostly bool numabalancing_enabled; | |
1849 | ||
1850 | void set_numabalancing_state(bool enabled) | |
1851 | { | |
1852 | numabalancing_enabled = enabled; | |
dd41f596 | 1853 | } |
3105b86a | 1854 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1855 | |
1856 | #ifdef CONFIG_PROC_SYSCTL | |
1857 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1858 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1859 | { | |
1860 | struct ctl_table t; | |
1861 | int err; | |
1862 | int state = numabalancing_enabled; | |
1863 | ||
1864 | if (write && !capable(CAP_SYS_ADMIN)) | |
1865 | return -EPERM; | |
1866 | ||
1867 | t = *table; | |
1868 | t.data = &state; | |
1869 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1870 | if (err < 0) | |
1871 | return err; | |
1872 | if (write) | |
1873 | set_numabalancing_state(state); | |
1874 | return err; | |
1875 | } | |
1876 | #endif | |
1877 | #endif | |
dd41f596 IM |
1878 | |
1879 | /* | |
1880 | * fork()/clone()-time setup: | |
1881 | */ | |
aab03e05 | 1882 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1883 | { |
0122ec5b | 1884 | unsigned long flags; |
dd41f596 IM |
1885 | int cpu = get_cpu(); |
1886 | ||
5e1576ed | 1887 | __sched_fork(clone_flags, p); |
06b83b5f | 1888 | /* |
0017d735 | 1889 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1890 | * nobody will actually run it, and a signal or other external |
1891 | * event cannot wake it up and insert it on the runqueue either. | |
1892 | */ | |
0017d735 | 1893 | p->state = TASK_RUNNING; |
dd41f596 | 1894 | |
c350a04e MG |
1895 | /* |
1896 | * Make sure we do not leak PI boosting priority to the child. | |
1897 | */ | |
1898 | p->prio = current->normal_prio; | |
1899 | ||
b9dc29e7 MG |
1900 | /* |
1901 | * Revert to default priority/policy on fork if requested. | |
1902 | */ | |
1903 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1904 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1905 | p->policy = SCHED_NORMAL; |
6c697bdf | 1906 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1907 | p->rt_priority = 0; |
1908 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1909 | p->static_prio = NICE_TO_PRIO(0); | |
1910 | ||
1911 | p->prio = p->normal_prio = __normal_prio(p); | |
1912 | set_load_weight(p); | |
6c697bdf | 1913 | |
b9dc29e7 MG |
1914 | /* |
1915 | * We don't need the reset flag anymore after the fork. It has | |
1916 | * fulfilled its duty: | |
1917 | */ | |
1918 | p->sched_reset_on_fork = 0; | |
1919 | } | |
ca94c442 | 1920 | |
aab03e05 DF |
1921 | if (dl_prio(p->prio)) { |
1922 | put_cpu(); | |
1923 | return -EAGAIN; | |
1924 | } else if (rt_prio(p->prio)) { | |
1925 | p->sched_class = &rt_sched_class; | |
1926 | } else { | |
2ddbf952 | 1927 | p->sched_class = &fair_sched_class; |
aab03e05 | 1928 | } |
b29739f9 | 1929 | |
cd29fe6f PZ |
1930 | if (p->sched_class->task_fork) |
1931 | p->sched_class->task_fork(p); | |
1932 | ||
86951599 PZ |
1933 | /* |
1934 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1935 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1936 | * is ran before sched_fork(). | |
1937 | * | |
1938 | * Silence PROVE_RCU. | |
1939 | */ | |
0122ec5b | 1940 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1941 | set_task_cpu(p, cpu); |
0122ec5b | 1942 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1943 | |
52f17b6c | 1944 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1945 | if (likely(sched_info_on())) |
52f17b6c | 1946 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1947 | #endif |
3ca7a440 PZ |
1948 | #if defined(CONFIG_SMP) |
1949 | p->on_cpu = 0; | |
4866cde0 | 1950 | #endif |
01028747 | 1951 | init_task_preempt_count(p); |
806c09a7 | 1952 | #ifdef CONFIG_SMP |
917b627d | 1953 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1954 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1955 | #endif |
917b627d | 1956 | |
476d139c | 1957 | put_cpu(); |
aab03e05 | 1958 | return 0; |
1da177e4 LT |
1959 | } |
1960 | ||
332ac17e DF |
1961 | unsigned long to_ratio(u64 period, u64 runtime) |
1962 | { | |
1963 | if (runtime == RUNTIME_INF) | |
1964 | return 1ULL << 20; | |
1965 | ||
1966 | /* | |
1967 | * Doing this here saves a lot of checks in all | |
1968 | * the calling paths, and returning zero seems | |
1969 | * safe for them anyway. | |
1970 | */ | |
1971 | if (period == 0) | |
1972 | return 0; | |
1973 | ||
1974 | return div64_u64(runtime << 20, period); | |
1975 | } | |
1976 | ||
1977 | #ifdef CONFIG_SMP | |
1978 | inline struct dl_bw *dl_bw_of(int i) | |
1979 | { | |
1980 | return &cpu_rq(i)->rd->dl_bw; | |
1981 | } | |
1982 | ||
de212f18 | 1983 | static inline int dl_bw_cpus(int i) |
332ac17e | 1984 | { |
de212f18 PZ |
1985 | struct root_domain *rd = cpu_rq(i)->rd; |
1986 | int cpus = 0; | |
1987 | ||
1988 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1989 | cpus++; | |
1990 | ||
1991 | return cpus; | |
332ac17e DF |
1992 | } |
1993 | #else | |
1994 | inline struct dl_bw *dl_bw_of(int i) | |
1995 | { | |
1996 | return &cpu_rq(i)->dl.dl_bw; | |
1997 | } | |
1998 | ||
de212f18 | 1999 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2000 | { |
2001 | return 1; | |
2002 | } | |
2003 | #endif | |
2004 | ||
2005 | static inline | |
2006 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
2007 | { | |
2008 | dl_b->total_bw -= tsk_bw; | |
2009 | } | |
2010 | ||
2011 | static inline | |
2012 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
2013 | { | |
2014 | dl_b->total_bw += tsk_bw; | |
2015 | } | |
2016 | ||
2017 | static inline | |
2018 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
2019 | { | |
2020 | return dl_b->bw != -1 && | |
2021 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * We must be sure that accepting a new task (or allowing changing the | |
2026 | * parameters of an existing one) is consistent with the bandwidth | |
2027 | * constraints. If yes, this function also accordingly updates the currently | |
2028 | * allocated bandwidth to reflect the new situation. | |
2029 | * | |
2030 | * This function is called while holding p's rq->lock. | |
2031 | */ | |
2032 | static int dl_overflow(struct task_struct *p, int policy, | |
2033 | const struct sched_attr *attr) | |
2034 | { | |
2035 | ||
2036 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2037 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2038 | u64 runtime = attr->sched_runtime; |
2039 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2040 | int cpus, err = -1; |
332ac17e DF |
2041 | |
2042 | if (new_bw == p->dl.dl_bw) | |
2043 | return 0; | |
2044 | ||
2045 | /* | |
2046 | * Either if a task, enters, leave, or stays -deadline but changes | |
2047 | * its parameters, we may need to update accordingly the total | |
2048 | * allocated bandwidth of the container. | |
2049 | */ | |
2050 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2051 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2052 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2053 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2054 | __dl_add(dl_b, new_bw); | |
2055 | err = 0; | |
2056 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2057 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2058 | __dl_clear(dl_b, p->dl.dl_bw); | |
2059 | __dl_add(dl_b, new_bw); | |
2060 | err = 0; | |
2061 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2062 | __dl_clear(dl_b, p->dl.dl_bw); | |
2063 | err = 0; | |
2064 | } | |
2065 | raw_spin_unlock(&dl_b->lock); | |
2066 | ||
2067 | return err; | |
2068 | } | |
2069 | ||
2070 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2071 | ||
1da177e4 LT |
2072 | /* |
2073 | * wake_up_new_task - wake up a newly created task for the first time. | |
2074 | * | |
2075 | * This function will do some initial scheduler statistics housekeeping | |
2076 | * that must be done for every newly created context, then puts the task | |
2077 | * on the runqueue and wakes it. | |
2078 | */ | |
3e51e3ed | 2079 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2080 | { |
2081 | unsigned long flags; | |
dd41f596 | 2082 | struct rq *rq; |
fabf318e | 2083 | |
ab2515c4 | 2084 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2085 | #ifdef CONFIG_SMP |
2086 | /* | |
2087 | * Fork balancing, do it here and not earlier because: | |
2088 | * - cpus_allowed can change in the fork path | |
2089 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2090 | */ |
ac66f547 | 2091 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2092 | #endif |
2093 | ||
a75cdaa9 AS |
2094 | /* Initialize new task's runnable average */ |
2095 | init_task_runnable_average(p); | |
ab2515c4 | 2096 | rq = __task_rq_lock(p); |
cd29fe6f | 2097 | activate_task(rq, p, 0); |
fd2f4419 | 2098 | p->on_rq = 1; |
89363381 | 2099 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2100 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2101 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2102 | if (p->sched_class->task_woken) |
2103 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2104 | #endif |
0122ec5b | 2105 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2106 | } |
2107 | ||
e107be36 AK |
2108 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2109 | ||
2110 | /** | |
80dd99b3 | 2111 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2112 | * @notifier: notifier struct to register |
e107be36 AK |
2113 | */ |
2114 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2115 | { | |
2116 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2117 | } | |
2118 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2119 | ||
2120 | /** | |
2121 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2122 | * @notifier: notifier struct to unregister |
e107be36 AK |
2123 | * |
2124 | * This is safe to call from within a preemption notifier. | |
2125 | */ | |
2126 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2127 | { | |
2128 | hlist_del(¬ifier->link); | |
2129 | } | |
2130 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2131 | ||
2132 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2133 | { | |
2134 | struct preempt_notifier *notifier; | |
e107be36 | 2135 | |
b67bfe0d | 2136 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2137 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2138 | } | |
2139 | ||
2140 | static void | |
2141 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2142 | struct task_struct *next) | |
2143 | { | |
2144 | struct preempt_notifier *notifier; | |
e107be36 | 2145 | |
b67bfe0d | 2146 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2147 | notifier->ops->sched_out(notifier, next); |
2148 | } | |
2149 | ||
6d6bc0ad | 2150 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2151 | |
2152 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2153 | { | |
2154 | } | |
2155 | ||
2156 | static void | |
2157 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2158 | struct task_struct *next) | |
2159 | { | |
2160 | } | |
2161 | ||
6d6bc0ad | 2162 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2163 | |
4866cde0 NP |
2164 | /** |
2165 | * prepare_task_switch - prepare to switch tasks | |
2166 | * @rq: the runqueue preparing to switch | |
421cee29 | 2167 | * @prev: the current task that is being switched out |
4866cde0 NP |
2168 | * @next: the task we are going to switch to. |
2169 | * | |
2170 | * This is called with the rq lock held and interrupts off. It must | |
2171 | * be paired with a subsequent finish_task_switch after the context | |
2172 | * switch. | |
2173 | * | |
2174 | * prepare_task_switch sets up locking and calls architecture specific | |
2175 | * hooks. | |
2176 | */ | |
e107be36 AK |
2177 | static inline void |
2178 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2179 | struct task_struct *next) | |
4866cde0 | 2180 | { |
895dd92c | 2181 | trace_sched_switch(prev, next); |
43148951 | 2182 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2183 | perf_event_task_sched_out(prev, next); |
e107be36 | 2184 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2185 | prepare_lock_switch(rq, next); |
2186 | prepare_arch_switch(next); | |
2187 | } | |
2188 | ||
1da177e4 LT |
2189 | /** |
2190 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2191 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2192 | * @prev: the thread we just switched away from. |
2193 | * | |
4866cde0 NP |
2194 | * finish_task_switch must be called after the context switch, paired |
2195 | * with a prepare_task_switch call before the context switch. | |
2196 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2197 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2198 | * |
2199 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2200 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2201 | * with the lock held can cause deadlocks; see schedule() for |
2202 | * details.) | |
2203 | */ | |
a9957449 | 2204 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2205 | __releases(rq->lock) |
2206 | { | |
1da177e4 | 2207 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2208 | long prev_state; |
1da177e4 LT |
2209 | |
2210 | rq->prev_mm = NULL; | |
2211 | ||
2212 | /* | |
2213 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2214 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2215 | * schedule one last time. The schedule call will never return, and |
2216 | * the scheduled task must drop that reference. | |
c394cc9f | 2217 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2218 | * still held, otherwise prev could be scheduled on another cpu, die |
2219 | * there before we look at prev->state, and then the reference would | |
2220 | * be dropped twice. | |
2221 | * Manfred Spraul <manfred@colorfullife.com> | |
2222 | */ | |
55a101f8 | 2223 | prev_state = prev->state; |
bf9fae9f | 2224 | vtime_task_switch(prev); |
4866cde0 | 2225 | finish_arch_switch(prev); |
a8d757ef | 2226 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2227 | finish_lock_switch(rq, prev); |
01f23e16 | 2228 | finish_arch_post_lock_switch(); |
e8fa1362 | 2229 | |
e107be36 | 2230 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2231 | if (mm) |
2232 | mmdrop(mm); | |
c394cc9f | 2233 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2234 | if (prev->sched_class->task_dead) |
2235 | prev->sched_class->task_dead(prev); | |
2236 | ||
c6fd91f0 | 2237 | /* |
2238 | * Remove function-return probe instances associated with this | |
2239 | * task and put them back on the free list. | |
9761eea8 | 2240 | */ |
c6fd91f0 | 2241 | kprobe_flush_task(prev); |
1da177e4 | 2242 | put_task_struct(prev); |
c6fd91f0 | 2243 | } |
99e5ada9 FW |
2244 | |
2245 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2246 | } |
2247 | ||
3f029d3c GH |
2248 | #ifdef CONFIG_SMP |
2249 | ||
3f029d3c GH |
2250 | /* rq->lock is NOT held, but preemption is disabled */ |
2251 | static inline void post_schedule(struct rq *rq) | |
2252 | { | |
2253 | if (rq->post_schedule) { | |
2254 | unsigned long flags; | |
2255 | ||
05fa785c | 2256 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2257 | if (rq->curr->sched_class->post_schedule) |
2258 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2259 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2260 | |
2261 | rq->post_schedule = 0; | |
2262 | } | |
2263 | } | |
2264 | ||
2265 | #else | |
da19ab51 | 2266 | |
3f029d3c GH |
2267 | static inline void post_schedule(struct rq *rq) |
2268 | { | |
1da177e4 LT |
2269 | } |
2270 | ||
3f029d3c GH |
2271 | #endif |
2272 | ||
1da177e4 LT |
2273 | /** |
2274 | * schedule_tail - first thing a freshly forked thread must call. | |
2275 | * @prev: the thread we just switched away from. | |
2276 | */ | |
722a9f92 | 2277 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2278 | __releases(rq->lock) |
2279 | { | |
70b97a7f IM |
2280 | struct rq *rq = this_rq(); |
2281 | ||
4866cde0 | 2282 | finish_task_switch(rq, prev); |
da19ab51 | 2283 | |
3f029d3c GH |
2284 | /* |
2285 | * FIXME: do we need to worry about rq being invalidated by the | |
2286 | * task_switch? | |
2287 | */ | |
2288 | post_schedule(rq); | |
70b97a7f | 2289 | |
4866cde0 NP |
2290 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2291 | /* In this case, finish_task_switch does not reenable preemption */ | |
2292 | preempt_enable(); | |
2293 | #endif | |
1da177e4 | 2294 | if (current->set_child_tid) |
b488893a | 2295 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2296 | } |
2297 | ||
2298 | /* | |
2299 | * context_switch - switch to the new MM and the new | |
2300 | * thread's register state. | |
2301 | */ | |
dd41f596 | 2302 | static inline void |
70b97a7f | 2303 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2304 | struct task_struct *next) |
1da177e4 | 2305 | { |
dd41f596 | 2306 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2307 | |
e107be36 | 2308 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2309 | |
dd41f596 IM |
2310 | mm = next->mm; |
2311 | oldmm = prev->active_mm; | |
9226d125 ZA |
2312 | /* |
2313 | * For paravirt, this is coupled with an exit in switch_to to | |
2314 | * combine the page table reload and the switch backend into | |
2315 | * one hypercall. | |
2316 | */ | |
224101ed | 2317 | arch_start_context_switch(prev); |
9226d125 | 2318 | |
31915ab4 | 2319 | if (!mm) { |
1da177e4 LT |
2320 | next->active_mm = oldmm; |
2321 | atomic_inc(&oldmm->mm_count); | |
2322 | enter_lazy_tlb(oldmm, next); | |
2323 | } else | |
2324 | switch_mm(oldmm, mm, next); | |
2325 | ||
31915ab4 | 2326 | if (!prev->mm) { |
1da177e4 | 2327 | prev->active_mm = NULL; |
1da177e4 LT |
2328 | rq->prev_mm = oldmm; |
2329 | } | |
3a5f5e48 IM |
2330 | /* |
2331 | * Since the runqueue lock will be released by the next | |
2332 | * task (which is an invalid locking op but in the case | |
2333 | * of the scheduler it's an obvious special-case), so we | |
2334 | * do an early lockdep release here: | |
2335 | */ | |
2336 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2337 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2338 | #endif |
1da177e4 | 2339 | |
91d1aa43 | 2340 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2341 | /* Here we just switch the register state and the stack. */ |
2342 | switch_to(prev, next, prev); | |
2343 | ||
dd41f596 IM |
2344 | barrier(); |
2345 | /* | |
2346 | * this_rq must be evaluated again because prev may have moved | |
2347 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2348 | * frame will be invalid. | |
2349 | */ | |
2350 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2351 | } |
2352 | ||
2353 | /* | |
1c3e8264 | 2354 | * nr_running and nr_context_switches: |
1da177e4 LT |
2355 | * |
2356 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2357 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2358 | */ |
2359 | unsigned long nr_running(void) | |
2360 | { | |
2361 | unsigned long i, sum = 0; | |
2362 | ||
2363 | for_each_online_cpu(i) | |
2364 | sum += cpu_rq(i)->nr_running; | |
2365 | ||
2366 | return sum; | |
f711f609 | 2367 | } |
1da177e4 | 2368 | |
1da177e4 | 2369 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2370 | { |
cc94abfc SR |
2371 | int i; |
2372 | unsigned long long sum = 0; | |
46cb4b7c | 2373 | |
0a945022 | 2374 | for_each_possible_cpu(i) |
1da177e4 | 2375 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2376 | |
1da177e4 LT |
2377 | return sum; |
2378 | } | |
483b4ee6 | 2379 | |
1da177e4 LT |
2380 | unsigned long nr_iowait(void) |
2381 | { | |
2382 | unsigned long i, sum = 0; | |
483b4ee6 | 2383 | |
0a945022 | 2384 | for_each_possible_cpu(i) |
1da177e4 | 2385 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2386 | |
1da177e4 LT |
2387 | return sum; |
2388 | } | |
483b4ee6 | 2389 | |
8c215bd3 | 2390 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2391 | { |
8c215bd3 | 2392 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2393 | return atomic_read(&this->nr_iowait); |
2394 | } | |
46cb4b7c | 2395 | |
dd41f596 | 2396 | #ifdef CONFIG_SMP |
8a0be9ef | 2397 | |
46cb4b7c | 2398 | /* |
38022906 PZ |
2399 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2400 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2401 | */ |
38022906 | 2402 | void sched_exec(void) |
46cb4b7c | 2403 | { |
38022906 | 2404 | struct task_struct *p = current; |
1da177e4 | 2405 | unsigned long flags; |
0017d735 | 2406 | int dest_cpu; |
46cb4b7c | 2407 | |
8f42ced9 | 2408 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2409 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2410 | if (dest_cpu == smp_processor_id()) |
2411 | goto unlock; | |
38022906 | 2412 | |
8f42ced9 | 2413 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2414 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2415 | |
8f42ced9 PZ |
2416 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2417 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2418 | return; |
2419 | } | |
0017d735 | 2420 | unlock: |
8f42ced9 | 2421 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2422 | } |
dd41f596 | 2423 | |
1da177e4 LT |
2424 | #endif |
2425 | ||
1da177e4 | 2426 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2427 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2428 | |
2429 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2430 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2431 | |
2432 | /* | |
c5f8d995 | 2433 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2434 | * @p in case that task is currently running. |
c5f8d995 HS |
2435 | * |
2436 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2437 | */ |
c5f8d995 HS |
2438 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2439 | { | |
2440 | u64 ns = 0; | |
2441 | ||
4036ac15 MG |
2442 | /* |
2443 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
2444 | * project cycles that may never be accounted to this | |
2445 | * thread, breaking clock_gettime(). | |
2446 | */ | |
2447 | if (task_current(rq, p) && p->on_rq) { | |
c5f8d995 | 2448 | update_rq_clock(rq); |
78becc27 | 2449 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2450 | if ((s64)ns < 0) |
2451 | ns = 0; | |
2452 | } | |
2453 | ||
2454 | return ns; | |
2455 | } | |
2456 | ||
bb34d92f | 2457 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2458 | { |
1da177e4 | 2459 | unsigned long flags; |
41b86e9c | 2460 | struct rq *rq; |
bb34d92f | 2461 | u64 ns = 0; |
48f24c4d | 2462 | |
41b86e9c | 2463 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2464 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2465 | task_rq_unlock(rq, p, &flags); |
1508487e | 2466 | |
c5f8d995 HS |
2467 | return ns; |
2468 | } | |
f06febc9 | 2469 | |
c5f8d995 HS |
2470 | /* |
2471 | * Return accounted runtime for the task. | |
2472 | * In case the task is currently running, return the runtime plus current's | |
2473 | * pending runtime that have not been accounted yet. | |
2474 | */ | |
2475 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2476 | { | |
2477 | unsigned long flags; | |
2478 | struct rq *rq; | |
2479 | u64 ns = 0; | |
2480 | ||
911b2898 PZ |
2481 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2482 | /* | |
2483 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2484 | * So we have a optimization chance when the task's delta_exec is 0. | |
2485 | * Reading ->on_cpu is racy, but this is ok. | |
2486 | * | |
2487 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2488 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2489 | * indistinguishable from the read occurring a few cycles earlier. | |
4036ac15 MG |
2490 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
2491 | * been accounted, so we're correct here as well. | |
911b2898 | 2492 | */ |
4036ac15 | 2493 | if (!p->on_cpu || !p->on_rq) |
911b2898 PZ |
2494 | return p->se.sum_exec_runtime; |
2495 | #endif | |
2496 | ||
c5f8d995 HS |
2497 | rq = task_rq_lock(p, &flags); |
2498 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2499 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2500 | |
2501 | return ns; | |
2502 | } | |
48f24c4d | 2503 | |
7835b98b CL |
2504 | /* |
2505 | * This function gets called by the timer code, with HZ frequency. | |
2506 | * We call it with interrupts disabled. | |
7835b98b CL |
2507 | */ |
2508 | void scheduler_tick(void) | |
2509 | { | |
7835b98b CL |
2510 | int cpu = smp_processor_id(); |
2511 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2512 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2513 | |
2514 | sched_clock_tick(); | |
dd41f596 | 2515 | |
05fa785c | 2516 | raw_spin_lock(&rq->lock); |
3e51f33f | 2517 | update_rq_clock(rq); |
fa85ae24 | 2518 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2519 | update_cpu_load_active(rq); |
05fa785c | 2520 | raw_spin_unlock(&rq->lock); |
7835b98b | 2521 | |
e9d2b064 | 2522 | perf_event_task_tick(); |
e220d2dc | 2523 | |
e418e1c2 | 2524 | #ifdef CONFIG_SMP |
6eb57e0d | 2525 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2526 | trigger_load_balance(rq); |
e418e1c2 | 2527 | #endif |
265f22a9 | 2528 | rq_last_tick_reset(rq); |
1da177e4 LT |
2529 | } |
2530 | ||
265f22a9 FW |
2531 | #ifdef CONFIG_NO_HZ_FULL |
2532 | /** | |
2533 | * scheduler_tick_max_deferment | |
2534 | * | |
2535 | * Keep at least one tick per second when a single | |
2536 | * active task is running because the scheduler doesn't | |
2537 | * yet completely support full dynticks environment. | |
2538 | * | |
2539 | * This makes sure that uptime, CFS vruntime, load | |
2540 | * balancing, etc... continue to move forward, even | |
2541 | * with a very low granularity. | |
e69f6186 YB |
2542 | * |
2543 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2544 | */ |
2545 | u64 scheduler_tick_max_deferment(void) | |
2546 | { | |
2547 | struct rq *rq = this_rq(); | |
2548 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2549 | ||
2550 | next = rq->last_sched_tick + HZ; | |
2551 | ||
2552 | if (time_before_eq(next, now)) | |
2553 | return 0; | |
2554 | ||
8fe8ff09 | 2555 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2556 | } |
265f22a9 | 2557 | #endif |
1da177e4 | 2558 | |
132380a0 | 2559 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2560 | { |
2561 | if (in_lock_functions(addr)) { | |
2562 | addr = CALLER_ADDR2; | |
2563 | if (in_lock_functions(addr)) | |
2564 | addr = CALLER_ADDR3; | |
2565 | } | |
2566 | return addr; | |
2567 | } | |
1da177e4 | 2568 | |
7e49fcce SR |
2569 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2570 | defined(CONFIG_PREEMPT_TRACER)) | |
2571 | ||
edafe3a5 | 2572 | void preempt_count_add(int val) |
1da177e4 | 2573 | { |
6cd8a4bb | 2574 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2575 | /* |
2576 | * Underflow? | |
2577 | */ | |
9a11b49a IM |
2578 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2579 | return; | |
6cd8a4bb | 2580 | #endif |
bdb43806 | 2581 | __preempt_count_add(val); |
6cd8a4bb | 2582 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2583 | /* |
2584 | * Spinlock count overflowing soon? | |
2585 | */ | |
33859f7f MOS |
2586 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2587 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2588 | #endif |
8f47b187 TG |
2589 | if (preempt_count() == val) { |
2590 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2591 | #ifdef CONFIG_DEBUG_PREEMPT | |
2592 | current->preempt_disable_ip = ip; | |
2593 | #endif | |
2594 | trace_preempt_off(CALLER_ADDR0, ip); | |
2595 | } | |
1da177e4 | 2596 | } |
bdb43806 | 2597 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 2598 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 2599 | |
edafe3a5 | 2600 | void preempt_count_sub(int val) |
1da177e4 | 2601 | { |
6cd8a4bb | 2602 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2603 | /* |
2604 | * Underflow? | |
2605 | */ | |
01e3eb82 | 2606 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2607 | return; |
1da177e4 LT |
2608 | /* |
2609 | * Is the spinlock portion underflowing? | |
2610 | */ | |
9a11b49a IM |
2611 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2612 | !(preempt_count() & PREEMPT_MASK))) | |
2613 | return; | |
6cd8a4bb | 2614 | #endif |
9a11b49a | 2615 | |
6cd8a4bb SR |
2616 | if (preempt_count() == val) |
2617 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2618 | __preempt_count_sub(val); |
1da177e4 | 2619 | } |
bdb43806 | 2620 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 2621 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2622 | |
2623 | #endif | |
2624 | ||
2625 | /* | |
dd41f596 | 2626 | * Print scheduling while atomic bug: |
1da177e4 | 2627 | */ |
dd41f596 | 2628 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2629 | { |
664dfa65 DJ |
2630 | if (oops_in_progress) |
2631 | return; | |
2632 | ||
3df0fc5b PZ |
2633 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2634 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2635 | |
dd41f596 | 2636 | debug_show_held_locks(prev); |
e21f5b15 | 2637 | print_modules(); |
dd41f596 IM |
2638 | if (irqs_disabled()) |
2639 | print_irqtrace_events(prev); | |
8f47b187 TG |
2640 | #ifdef CONFIG_DEBUG_PREEMPT |
2641 | if (in_atomic_preempt_off()) { | |
2642 | pr_err("Preemption disabled at:"); | |
2643 | print_ip_sym(current->preempt_disable_ip); | |
2644 | pr_cont("\n"); | |
2645 | } | |
2646 | #endif | |
6135fc1e | 2647 | dump_stack(); |
373d4d09 | 2648 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2649 | } |
1da177e4 | 2650 | |
dd41f596 IM |
2651 | /* |
2652 | * Various schedule()-time debugging checks and statistics: | |
2653 | */ | |
2654 | static inline void schedule_debug(struct task_struct *prev) | |
2655 | { | |
1da177e4 | 2656 | /* |
41a2d6cf | 2657 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2658 | * schedule() atomically, we ignore that path. Otherwise whine |
2659 | * if we are scheduling when we should not. | |
1da177e4 | 2660 | */ |
192301e7 | 2661 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2662 | __schedule_bug(prev); |
b3fbab05 | 2663 | rcu_sleep_check(); |
dd41f596 | 2664 | |
1da177e4 LT |
2665 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2666 | ||
2d72376b | 2667 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2668 | } |
2669 | ||
2670 | /* | |
2671 | * Pick up the highest-prio task: | |
2672 | */ | |
2673 | static inline struct task_struct * | |
606dba2e | 2674 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2675 | { |
37e117c0 | 2676 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2677 | struct task_struct *p; |
1da177e4 LT |
2678 | |
2679 | /* | |
dd41f596 IM |
2680 | * Optimization: we know that if all tasks are in |
2681 | * the fair class we can call that function directly: | |
1da177e4 | 2682 | */ |
37e117c0 | 2683 | if (likely(prev->sched_class == class && |
38033c37 | 2684 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2685 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2686 | if (unlikely(p == RETRY_TASK)) |
2687 | goto again; | |
2688 | ||
2689 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2690 | if (unlikely(!p)) | |
2691 | p = idle_sched_class.pick_next_task(rq, prev); | |
2692 | ||
2693 | return p; | |
1da177e4 LT |
2694 | } |
2695 | ||
37e117c0 | 2696 | again: |
34f971f6 | 2697 | for_each_class(class) { |
606dba2e | 2698 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2699 | if (p) { |
2700 | if (unlikely(p == RETRY_TASK)) | |
2701 | goto again; | |
dd41f596 | 2702 | return p; |
37e117c0 | 2703 | } |
dd41f596 | 2704 | } |
34f971f6 PZ |
2705 | |
2706 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2707 | } |
1da177e4 | 2708 | |
dd41f596 | 2709 | /* |
c259e01a | 2710 | * __schedule() is the main scheduler function. |
edde96ea PE |
2711 | * |
2712 | * The main means of driving the scheduler and thus entering this function are: | |
2713 | * | |
2714 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2715 | * | |
2716 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2717 | * paths. For example, see arch/x86/entry_64.S. | |
2718 | * | |
2719 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2720 | * interrupt handler scheduler_tick(). | |
2721 | * | |
2722 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2723 | * task to the run-queue and that's it. | |
2724 | * | |
2725 | * Now, if the new task added to the run-queue preempts the current | |
2726 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2727 | * called on the nearest possible occasion: | |
2728 | * | |
2729 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2730 | * | |
2731 | * - in syscall or exception context, at the next outmost | |
2732 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2733 | * spin_unlock()!) | |
2734 | * | |
2735 | * - in IRQ context, return from interrupt-handler to | |
2736 | * preemptible context | |
2737 | * | |
2738 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2739 | * then at the next: | |
2740 | * | |
2741 | * - cond_resched() call | |
2742 | * - explicit schedule() call | |
2743 | * - return from syscall or exception to user-space | |
2744 | * - return from interrupt-handler to user-space | |
dd41f596 | 2745 | */ |
c259e01a | 2746 | static void __sched __schedule(void) |
dd41f596 IM |
2747 | { |
2748 | struct task_struct *prev, *next; | |
67ca7bde | 2749 | unsigned long *switch_count; |
dd41f596 | 2750 | struct rq *rq; |
31656519 | 2751 | int cpu; |
dd41f596 | 2752 | |
ff743345 PZ |
2753 | need_resched: |
2754 | preempt_disable(); | |
dd41f596 IM |
2755 | cpu = smp_processor_id(); |
2756 | rq = cpu_rq(cpu); | |
25502a6c | 2757 | rcu_note_context_switch(cpu); |
dd41f596 | 2758 | prev = rq->curr; |
dd41f596 | 2759 | |
dd41f596 | 2760 | schedule_debug(prev); |
1da177e4 | 2761 | |
31656519 | 2762 | if (sched_feat(HRTICK)) |
f333fdc9 | 2763 | hrtick_clear(rq); |
8f4d37ec | 2764 | |
e0acd0a6 ON |
2765 | /* |
2766 | * Make sure that signal_pending_state()->signal_pending() below | |
2767 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2768 | * done by the caller to avoid the race with signal_wake_up(). | |
2769 | */ | |
2770 | smp_mb__before_spinlock(); | |
05fa785c | 2771 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2772 | |
246d86b5 | 2773 | switch_count = &prev->nivcsw; |
1da177e4 | 2774 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2775 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2776 | prev->state = TASK_RUNNING; |
21aa9af0 | 2777 | } else { |
2acca55e PZ |
2778 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2779 | prev->on_rq = 0; | |
2780 | ||
21aa9af0 | 2781 | /* |
2acca55e PZ |
2782 | * If a worker went to sleep, notify and ask workqueue |
2783 | * whether it wants to wake up a task to maintain | |
2784 | * concurrency. | |
21aa9af0 TH |
2785 | */ |
2786 | if (prev->flags & PF_WQ_WORKER) { | |
2787 | struct task_struct *to_wakeup; | |
2788 | ||
2789 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2790 | if (to_wakeup) | |
2791 | try_to_wake_up_local(to_wakeup); | |
2792 | } | |
21aa9af0 | 2793 | } |
dd41f596 | 2794 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2795 | } |
2796 | ||
606dba2e PZ |
2797 | if (prev->on_rq || rq->skip_clock_update < 0) |
2798 | update_rq_clock(rq); | |
2799 | ||
2800 | next = pick_next_task(rq, prev); | |
f26f9aff | 2801 | clear_tsk_need_resched(prev); |
f27dde8d | 2802 | clear_preempt_need_resched(); |
f26f9aff | 2803 | rq->skip_clock_update = 0; |
1da177e4 | 2804 | |
1da177e4 | 2805 | if (likely(prev != next)) { |
1da177e4 LT |
2806 | rq->nr_switches++; |
2807 | rq->curr = next; | |
2808 | ++*switch_count; | |
2809 | ||
dd41f596 | 2810 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2811 | /* |
246d86b5 ON |
2812 | * The context switch have flipped the stack from under us |
2813 | * and restored the local variables which were saved when | |
2814 | * this task called schedule() in the past. prev == current | |
2815 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2816 | */ |
2817 | cpu = smp_processor_id(); | |
2818 | rq = cpu_rq(cpu); | |
1da177e4 | 2819 | } else |
05fa785c | 2820 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2821 | |
3f029d3c | 2822 | post_schedule(rq); |
1da177e4 | 2823 | |
ba74c144 | 2824 | sched_preempt_enable_no_resched(); |
ff743345 | 2825 | if (need_resched()) |
1da177e4 LT |
2826 | goto need_resched; |
2827 | } | |
c259e01a | 2828 | |
9c40cef2 TG |
2829 | static inline void sched_submit_work(struct task_struct *tsk) |
2830 | { | |
3c7d5184 | 2831 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2832 | return; |
2833 | /* | |
2834 | * If we are going to sleep and we have plugged IO queued, | |
2835 | * make sure to submit it to avoid deadlocks. | |
2836 | */ | |
2837 | if (blk_needs_flush_plug(tsk)) | |
2838 | blk_schedule_flush_plug(tsk); | |
2839 | } | |
2840 | ||
722a9f92 | 2841 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2842 | { |
9c40cef2 TG |
2843 | struct task_struct *tsk = current; |
2844 | ||
2845 | sched_submit_work(tsk); | |
c259e01a TG |
2846 | __schedule(); |
2847 | } | |
1da177e4 LT |
2848 | EXPORT_SYMBOL(schedule); |
2849 | ||
91d1aa43 | 2850 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2851 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2852 | { |
2853 | /* | |
2854 | * If we come here after a random call to set_need_resched(), | |
2855 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2856 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2857 | * we find a better solution. | |
2858 | */ | |
91d1aa43 | 2859 | user_exit(); |
20ab65e3 | 2860 | schedule(); |
91d1aa43 | 2861 | user_enter(); |
20ab65e3 FW |
2862 | } |
2863 | #endif | |
2864 | ||
c5491ea7 TG |
2865 | /** |
2866 | * schedule_preempt_disabled - called with preemption disabled | |
2867 | * | |
2868 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2869 | */ | |
2870 | void __sched schedule_preempt_disabled(void) | |
2871 | { | |
ba74c144 | 2872 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2873 | schedule(); |
2874 | preempt_disable(); | |
2875 | } | |
2876 | ||
1da177e4 LT |
2877 | #ifdef CONFIG_PREEMPT |
2878 | /* | |
2ed6e34f | 2879 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2880 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2881 | * occur there and call schedule directly. |
2882 | */ | |
722a9f92 | 2883 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2884 | { |
1da177e4 LT |
2885 | /* |
2886 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2887 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2888 | */ |
fbb00b56 | 2889 | if (likely(!preemptible())) |
1da177e4 LT |
2890 | return; |
2891 | ||
3a5c359a | 2892 | do { |
bdb43806 | 2893 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2894 | __schedule(); |
bdb43806 | 2895 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2896 | |
3a5c359a AK |
2897 | /* |
2898 | * Check again in case we missed a preemption opportunity | |
2899 | * between schedule and now. | |
2900 | */ | |
2901 | barrier(); | |
5ed0cec0 | 2902 | } while (need_resched()); |
1da177e4 | 2903 | } |
376e2424 | 2904 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 2905 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2906 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2907 | |
2908 | /* | |
2ed6e34f | 2909 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2910 | * off of irq context. |
2911 | * Note, that this is called and return with irqs disabled. This will | |
2912 | * protect us against recursive calling from irq. | |
2913 | */ | |
722a9f92 | 2914 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2915 | { |
b22366cd | 2916 | enum ctx_state prev_state; |
6478d880 | 2917 | |
2ed6e34f | 2918 | /* Catch callers which need to be fixed */ |
f27dde8d | 2919 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2920 | |
b22366cd FW |
2921 | prev_state = exception_enter(); |
2922 | ||
3a5c359a | 2923 | do { |
bdb43806 | 2924 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2925 | local_irq_enable(); |
c259e01a | 2926 | __schedule(); |
3a5c359a | 2927 | local_irq_disable(); |
bdb43806 | 2928 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2929 | |
3a5c359a AK |
2930 | /* |
2931 | * Check again in case we missed a preemption opportunity | |
2932 | * between schedule and now. | |
2933 | */ | |
2934 | barrier(); | |
5ed0cec0 | 2935 | } while (need_resched()); |
b22366cd FW |
2936 | |
2937 | exception_exit(prev_state); | |
1da177e4 LT |
2938 | } |
2939 | ||
63859d4f | 2940 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2941 | void *key) |
1da177e4 | 2942 | { |
63859d4f | 2943 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2944 | } |
1da177e4 LT |
2945 | EXPORT_SYMBOL(default_wake_function); |
2946 | ||
b29739f9 IM |
2947 | #ifdef CONFIG_RT_MUTEXES |
2948 | ||
2949 | /* | |
2950 | * rt_mutex_setprio - set the current priority of a task | |
2951 | * @p: task | |
2952 | * @prio: prio value (kernel-internal form) | |
2953 | * | |
2954 | * This function changes the 'effective' priority of a task. It does | |
2955 | * not touch ->normal_prio like __setscheduler(). | |
2956 | * | |
c365c292 TG |
2957 | * Used by the rt_mutex code to implement priority inheritance |
2958 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2959 | */ |
36c8b586 | 2960 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2961 | { |
2d3d891d | 2962 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2963 | struct rq *rq; |
83ab0aa0 | 2964 | const struct sched_class *prev_class; |
b29739f9 | 2965 | |
aab03e05 | 2966 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2967 | |
0122ec5b | 2968 | rq = __task_rq_lock(p); |
b29739f9 | 2969 | |
1c4dd99b TG |
2970 | /* |
2971 | * Idle task boosting is a nono in general. There is one | |
2972 | * exception, when PREEMPT_RT and NOHZ is active: | |
2973 | * | |
2974 | * The idle task calls get_next_timer_interrupt() and holds | |
2975 | * the timer wheel base->lock on the CPU and another CPU wants | |
2976 | * to access the timer (probably to cancel it). We can safely | |
2977 | * ignore the boosting request, as the idle CPU runs this code | |
2978 | * with interrupts disabled and will complete the lock | |
2979 | * protected section without being interrupted. So there is no | |
2980 | * real need to boost. | |
2981 | */ | |
2982 | if (unlikely(p == rq->idle)) { | |
2983 | WARN_ON(p != rq->curr); | |
2984 | WARN_ON(p->pi_blocked_on); | |
2985 | goto out_unlock; | |
2986 | } | |
2987 | ||
a8027073 | 2988 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 2989 | oldprio = p->prio; |
83ab0aa0 | 2990 | prev_class = p->sched_class; |
fd2f4419 | 2991 | on_rq = p->on_rq; |
051a1d1a | 2992 | running = task_current(rq, p); |
0e1f3483 | 2993 | if (on_rq) |
69be72c1 | 2994 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2995 | if (running) |
2996 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2997 | |
2d3d891d DF |
2998 | /* |
2999 | * Boosting condition are: | |
3000 | * 1. -rt task is running and holds mutex A | |
3001 | * --> -dl task blocks on mutex A | |
3002 | * | |
3003 | * 2. -dl task is running and holds mutex A | |
3004 | * --> -dl task blocks on mutex A and could preempt the | |
3005 | * running task | |
3006 | */ | |
3007 | if (dl_prio(prio)) { | |
466af29b ON |
3008 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3009 | if (!dl_prio(p->normal_prio) || | |
3010 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d DF |
3011 | p->dl.dl_boosted = 1; |
3012 | p->dl.dl_throttled = 0; | |
3013 | enqueue_flag = ENQUEUE_REPLENISH; | |
3014 | } else | |
3015 | p->dl.dl_boosted = 0; | |
aab03e05 | 3016 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3017 | } else if (rt_prio(prio)) { |
3018 | if (dl_prio(oldprio)) | |
3019 | p->dl.dl_boosted = 0; | |
3020 | if (oldprio < prio) | |
3021 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 3022 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3023 | } else { |
3024 | if (dl_prio(oldprio)) | |
3025 | p->dl.dl_boosted = 0; | |
dd41f596 | 3026 | p->sched_class = &fair_sched_class; |
2d3d891d | 3027 | } |
dd41f596 | 3028 | |
b29739f9 IM |
3029 | p->prio = prio; |
3030 | ||
0e1f3483 HS |
3031 | if (running) |
3032 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3033 | if (on_rq) |
2d3d891d | 3034 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 3035 | |
da7a735e | 3036 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3037 | out_unlock: |
0122ec5b | 3038 | __task_rq_unlock(rq); |
b29739f9 | 3039 | } |
b29739f9 | 3040 | #endif |
d50dde5a | 3041 | |
36c8b586 | 3042 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3043 | { |
dd41f596 | 3044 | int old_prio, delta, on_rq; |
1da177e4 | 3045 | unsigned long flags; |
70b97a7f | 3046 | struct rq *rq; |
1da177e4 | 3047 | |
75e45d51 | 3048 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3049 | return; |
3050 | /* | |
3051 | * We have to be careful, if called from sys_setpriority(), | |
3052 | * the task might be in the middle of scheduling on another CPU. | |
3053 | */ | |
3054 | rq = task_rq_lock(p, &flags); | |
3055 | /* | |
3056 | * The RT priorities are set via sched_setscheduler(), but we still | |
3057 | * allow the 'normal' nice value to be set - but as expected | |
3058 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3059 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3060 | */ |
aab03e05 | 3061 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3062 | p->static_prio = NICE_TO_PRIO(nice); |
3063 | goto out_unlock; | |
3064 | } | |
fd2f4419 | 3065 | on_rq = p->on_rq; |
c09595f6 | 3066 | if (on_rq) |
69be72c1 | 3067 | dequeue_task(rq, p, 0); |
1da177e4 | 3068 | |
1da177e4 | 3069 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3070 | set_load_weight(p); |
b29739f9 IM |
3071 | old_prio = p->prio; |
3072 | p->prio = effective_prio(p); | |
3073 | delta = p->prio - old_prio; | |
1da177e4 | 3074 | |
dd41f596 | 3075 | if (on_rq) { |
371fd7e7 | 3076 | enqueue_task(rq, p, 0); |
1da177e4 | 3077 | /* |
d5f9f942 AM |
3078 | * If the task increased its priority or is running and |
3079 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3080 | */ |
d5f9f942 | 3081 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3082 | resched_curr(rq); |
1da177e4 LT |
3083 | } |
3084 | out_unlock: | |
0122ec5b | 3085 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3086 | } |
1da177e4 LT |
3087 | EXPORT_SYMBOL(set_user_nice); |
3088 | ||
e43379f1 MM |
3089 | /* |
3090 | * can_nice - check if a task can reduce its nice value | |
3091 | * @p: task | |
3092 | * @nice: nice value | |
3093 | */ | |
36c8b586 | 3094 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3095 | { |
024f4747 | 3096 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3097 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3098 | |
78d7d407 | 3099 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3100 | capable(CAP_SYS_NICE)); |
3101 | } | |
3102 | ||
1da177e4 LT |
3103 | #ifdef __ARCH_WANT_SYS_NICE |
3104 | ||
3105 | /* | |
3106 | * sys_nice - change the priority of the current process. | |
3107 | * @increment: priority increment | |
3108 | * | |
3109 | * sys_setpriority is a more generic, but much slower function that | |
3110 | * does similar things. | |
3111 | */ | |
5add95d4 | 3112 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3113 | { |
48f24c4d | 3114 | long nice, retval; |
1da177e4 LT |
3115 | |
3116 | /* | |
3117 | * Setpriority might change our priority at the same moment. | |
3118 | * We don't have to worry. Conceptually one call occurs first | |
3119 | * and we have a single winner. | |
3120 | */ | |
a9467fa3 | 3121 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3122 | nice = task_nice(current) + increment; |
1da177e4 | 3123 | |
a9467fa3 | 3124 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3125 | if (increment < 0 && !can_nice(current, nice)) |
3126 | return -EPERM; | |
3127 | ||
1da177e4 LT |
3128 | retval = security_task_setnice(current, nice); |
3129 | if (retval) | |
3130 | return retval; | |
3131 | ||
3132 | set_user_nice(current, nice); | |
3133 | return 0; | |
3134 | } | |
3135 | ||
3136 | #endif | |
3137 | ||
3138 | /** | |
3139 | * task_prio - return the priority value of a given task. | |
3140 | * @p: the task in question. | |
3141 | * | |
e69f6186 | 3142 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3143 | * RT tasks are offset by -200. Normal tasks are centered |
3144 | * around 0, value goes from -16 to +15. | |
3145 | */ | |
36c8b586 | 3146 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3147 | { |
3148 | return p->prio - MAX_RT_PRIO; | |
3149 | } | |
3150 | ||
1da177e4 LT |
3151 | /** |
3152 | * idle_cpu - is a given cpu idle currently? | |
3153 | * @cpu: the processor in question. | |
e69f6186 YB |
3154 | * |
3155 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3156 | */ |
3157 | int idle_cpu(int cpu) | |
3158 | { | |
908a3283 TG |
3159 | struct rq *rq = cpu_rq(cpu); |
3160 | ||
3161 | if (rq->curr != rq->idle) | |
3162 | return 0; | |
3163 | ||
3164 | if (rq->nr_running) | |
3165 | return 0; | |
3166 | ||
3167 | #ifdef CONFIG_SMP | |
3168 | if (!llist_empty(&rq->wake_list)) | |
3169 | return 0; | |
3170 | #endif | |
3171 | ||
3172 | return 1; | |
1da177e4 LT |
3173 | } |
3174 | ||
1da177e4 LT |
3175 | /** |
3176 | * idle_task - return the idle task for a given cpu. | |
3177 | * @cpu: the processor in question. | |
e69f6186 YB |
3178 | * |
3179 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3180 | */ |
36c8b586 | 3181 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3182 | { |
3183 | return cpu_rq(cpu)->idle; | |
3184 | } | |
3185 | ||
3186 | /** | |
3187 | * find_process_by_pid - find a process with a matching PID value. | |
3188 | * @pid: the pid in question. | |
e69f6186 YB |
3189 | * |
3190 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3191 | */ |
a9957449 | 3192 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3193 | { |
228ebcbe | 3194 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3195 | } |
3196 | ||
aab03e05 DF |
3197 | /* |
3198 | * This function initializes the sched_dl_entity of a newly becoming | |
3199 | * SCHED_DEADLINE task. | |
3200 | * | |
3201 | * Only the static values are considered here, the actual runtime and the | |
3202 | * absolute deadline will be properly calculated when the task is enqueued | |
3203 | * for the first time with its new policy. | |
3204 | */ | |
3205 | static void | |
3206 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3207 | { | |
3208 | struct sched_dl_entity *dl_se = &p->dl; | |
3209 | ||
3210 | init_dl_task_timer(dl_se); | |
3211 | dl_se->dl_runtime = attr->sched_runtime; | |
3212 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3213 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3214 | dl_se->flags = attr->sched_flags; |
332ac17e | 3215 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3216 | dl_se->dl_throttled = 0; |
3217 | dl_se->dl_new = 1; | |
5bfd126e | 3218 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3219 | } |
3220 | ||
c365c292 TG |
3221 | static void __setscheduler_params(struct task_struct *p, |
3222 | const struct sched_attr *attr) | |
1da177e4 | 3223 | { |
d50dde5a DF |
3224 | int policy = attr->sched_policy; |
3225 | ||
39fd8fd2 PZ |
3226 | if (policy == -1) /* setparam */ |
3227 | policy = p->policy; | |
3228 | ||
1da177e4 | 3229 | p->policy = policy; |
d50dde5a | 3230 | |
aab03e05 DF |
3231 | if (dl_policy(policy)) |
3232 | __setparam_dl(p, attr); | |
39fd8fd2 | 3233 | else if (fair_policy(policy)) |
d50dde5a DF |
3234 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3235 | ||
39fd8fd2 PZ |
3236 | /* |
3237 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3238 | * !rt_policy. Always setting this ensures that things like | |
3239 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3240 | */ | |
3241 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3242 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3243 | set_load_weight(p); |
3244 | } | |
39fd8fd2 | 3245 | |
c365c292 TG |
3246 | /* Actually do priority change: must hold pi & rq lock. */ |
3247 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3248 | const struct sched_attr *attr) | |
3249 | { | |
3250 | __setscheduler_params(p, attr); | |
d50dde5a | 3251 | |
383afd09 SR |
3252 | /* |
3253 | * If we get here, there was no pi waiters boosting the | |
3254 | * task. It is safe to use the normal prio. | |
3255 | */ | |
3256 | p->prio = normal_prio(p); | |
3257 | ||
aab03e05 DF |
3258 | if (dl_prio(p->prio)) |
3259 | p->sched_class = &dl_sched_class; | |
3260 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3261 | p->sched_class = &rt_sched_class; |
3262 | else | |
3263 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3264 | } |
aab03e05 DF |
3265 | |
3266 | static void | |
3267 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3268 | { | |
3269 | struct sched_dl_entity *dl_se = &p->dl; | |
3270 | ||
3271 | attr->sched_priority = p->rt_priority; | |
3272 | attr->sched_runtime = dl_se->dl_runtime; | |
3273 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3274 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3275 | attr->sched_flags = dl_se->flags; |
3276 | } | |
3277 | ||
3278 | /* | |
3279 | * This function validates the new parameters of a -deadline task. | |
3280 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3281 | * than the runtime, as well as the period of being zero or |
332ac17e | 3282 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3283 | * user parameters are above the internal resolution of 1us (we |
3284 | * check sched_runtime only since it is always the smaller one) and | |
3285 | * below 2^63 ns (we have to check both sched_deadline and | |
3286 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3287 | */ |
3288 | static bool | |
3289 | __checkparam_dl(const struct sched_attr *attr) | |
3290 | { | |
b0827819 JL |
3291 | /* deadline != 0 */ |
3292 | if (attr->sched_deadline == 0) | |
3293 | return false; | |
3294 | ||
3295 | /* | |
3296 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3297 | * that big. | |
3298 | */ | |
3299 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3300 | return false; | |
3301 | ||
3302 | /* | |
3303 | * Since we use the MSB for wrap-around and sign issues, make | |
3304 | * sure it's not set (mind that period can be equal to zero). | |
3305 | */ | |
3306 | if (attr->sched_deadline & (1ULL << 63) || | |
3307 | attr->sched_period & (1ULL << 63)) | |
3308 | return false; | |
3309 | ||
3310 | /* runtime <= deadline <= period (if period != 0) */ | |
3311 | if ((attr->sched_period != 0 && | |
3312 | attr->sched_period < attr->sched_deadline) || | |
3313 | attr->sched_deadline < attr->sched_runtime) | |
3314 | return false; | |
3315 | ||
3316 | return true; | |
aab03e05 DF |
3317 | } |
3318 | ||
c69e8d9c DH |
3319 | /* |
3320 | * check the target process has a UID that matches the current process's | |
3321 | */ | |
3322 | static bool check_same_owner(struct task_struct *p) | |
3323 | { | |
3324 | const struct cred *cred = current_cred(), *pcred; | |
3325 | bool match; | |
3326 | ||
3327 | rcu_read_lock(); | |
3328 | pcred = __task_cred(p); | |
9c806aa0 EB |
3329 | match = (uid_eq(cred->euid, pcred->euid) || |
3330 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3331 | rcu_read_unlock(); |
3332 | return match; | |
3333 | } | |
3334 | ||
d50dde5a DF |
3335 | static int __sched_setscheduler(struct task_struct *p, |
3336 | const struct sched_attr *attr, | |
3337 | bool user) | |
1da177e4 | 3338 | { |
383afd09 SR |
3339 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3340 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
83b699ed | 3341 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3342 | int policy = attr->sched_policy; |
1da177e4 | 3343 | unsigned long flags; |
83ab0aa0 | 3344 | const struct sched_class *prev_class; |
70b97a7f | 3345 | struct rq *rq; |
ca94c442 | 3346 | int reset_on_fork; |
1da177e4 | 3347 | |
66e5393a SR |
3348 | /* may grab non-irq protected spin_locks */ |
3349 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3350 | recheck: |
3351 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3352 | if (policy < 0) { |
3353 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3354 | policy = oldpolicy = p->policy; |
ca94c442 | 3355 | } else { |
7479f3c9 | 3356 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3357 | |
aab03e05 DF |
3358 | if (policy != SCHED_DEADLINE && |
3359 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3360 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3361 | policy != SCHED_IDLE) | |
3362 | return -EINVAL; | |
3363 | } | |
3364 | ||
7479f3c9 PZ |
3365 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3366 | return -EINVAL; | |
3367 | ||
1da177e4 LT |
3368 | /* |
3369 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3370 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3371 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3372 | */ |
0bb040a4 | 3373 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3374 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3375 | return -EINVAL; |
aab03e05 DF |
3376 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3377 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3378 | return -EINVAL; |
3379 | ||
37e4ab3f OC |
3380 | /* |
3381 | * Allow unprivileged RT tasks to decrease priority: | |
3382 | */ | |
961ccddd | 3383 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3384 | if (fair_policy(policy)) { |
d0ea0268 | 3385 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3386 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3387 | return -EPERM; |
3388 | } | |
3389 | ||
e05606d3 | 3390 | if (rt_policy(policy)) { |
a44702e8 ON |
3391 | unsigned long rlim_rtprio = |
3392 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3393 | |
3394 | /* can't set/change the rt policy */ | |
3395 | if (policy != p->policy && !rlim_rtprio) | |
3396 | return -EPERM; | |
3397 | ||
3398 | /* can't increase priority */ | |
d50dde5a DF |
3399 | if (attr->sched_priority > p->rt_priority && |
3400 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3401 | return -EPERM; |
3402 | } | |
c02aa73b | 3403 | |
d44753b8 JL |
3404 | /* |
3405 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3406 | * (safest behavior); in the future we would like to allow | |
3407 | * unprivileged DL tasks to increase their relative deadline | |
3408 | * or reduce their runtime (both ways reducing utilization) | |
3409 | */ | |
3410 | if (dl_policy(policy)) | |
3411 | return -EPERM; | |
3412 | ||
dd41f596 | 3413 | /* |
c02aa73b DH |
3414 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3415 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3416 | */ |
c02aa73b | 3417 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3418 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3419 | return -EPERM; |
3420 | } | |
5fe1d75f | 3421 | |
37e4ab3f | 3422 | /* can't change other user's priorities */ |
c69e8d9c | 3423 | if (!check_same_owner(p)) |
37e4ab3f | 3424 | return -EPERM; |
ca94c442 LP |
3425 | |
3426 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3427 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3428 | return -EPERM; | |
37e4ab3f | 3429 | } |
1da177e4 | 3430 | |
725aad24 | 3431 | if (user) { |
b0ae1981 | 3432 | retval = security_task_setscheduler(p); |
725aad24 JF |
3433 | if (retval) |
3434 | return retval; | |
3435 | } | |
3436 | ||
b29739f9 IM |
3437 | /* |
3438 | * make sure no PI-waiters arrive (or leave) while we are | |
3439 | * changing the priority of the task: | |
0122ec5b | 3440 | * |
25985edc | 3441 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3442 | * runqueue lock must be held. |
3443 | */ | |
0122ec5b | 3444 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3445 | |
34f971f6 PZ |
3446 | /* |
3447 | * Changing the policy of the stop threads its a very bad idea | |
3448 | */ | |
3449 | if (p == rq->stop) { | |
0122ec5b | 3450 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3451 | return -EINVAL; |
3452 | } | |
3453 | ||
a51e9198 | 3454 | /* |
d6b1e911 TG |
3455 | * If not changing anything there's no need to proceed further, |
3456 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3457 | */ |
d50dde5a | 3458 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3459 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3460 | goto change; |
3461 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3462 | goto change; | |
aab03e05 DF |
3463 | if (dl_policy(policy)) |
3464 | goto change; | |
d50dde5a | 3465 | |
d6b1e911 | 3466 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3467 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3468 | return 0; |
3469 | } | |
d50dde5a | 3470 | change: |
a51e9198 | 3471 | |
dc61b1d6 | 3472 | if (user) { |
332ac17e | 3473 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3474 | /* |
3475 | * Do not allow realtime tasks into groups that have no runtime | |
3476 | * assigned. | |
3477 | */ | |
3478 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3479 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3480 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3481 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3482 | return -EPERM; |
3483 | } | |
dc61b1d6 | 3484 | #endif |
332ac17e DF |
3485 | #ifdef CONFIG_SMP |
3486 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3487 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3488 | |
3489 | /* | |
3490 | * Don't allow tasks with an affinity mask smaller than | |
3491 | * the entire root_domain to become SCHED_DEADLINE. We | |
3492 | * will also fail if there's no bandwidth available. | |
3493 | */ | |
e4099a5e PZ |
3494 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3495 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3496 | task_rq_unlock(rq, p, &flags); |
3497 | return -EPERM; | |
3498 | } | |
3499 | } | |
3500 | #endif | |
3501 | } | |
dc61b1d6 | 3502 | |
1da177e4 LT |
3503 | /* recheck policy now with rq lock held */ |
3504 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3505 | policy = oldpolicy = -1; | |
0122ec5b | 3506 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3507 | goto recheck; |
3508 | } | |
332ac17e DF |
3509 | |
3510 | /* | |
3511 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3512 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3513 | * is available. | |
3514 | */ | |
e4099a5e | 3515 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3516 | task_rq_unlock(rq, p, &flags); |
3517 | return -EBUSY; | |
3518 | } | |
3519 | ||
c365c292 TG |
3520 | p->sched_reset_on_fork = reset_on_fork; |
3521 | oldprio = p->prio; | |
3522 | ||
3523 | /* | |
3524 | * Special case for priority boosted tasks. | |
3525 | * | |
3526 | * If the new priority is lower or equal (user space view) | |
3527 | * than the current (boosted) priority, we just store the new | |
3528 | * normal parameters and do not touch the scheduler class and | |
3529 | * the runqueue. This will be done when the task deboost | |
3530 | * itself. | |
3531 | */ | |
3532 | if (rt_mutex_check_prio(p, newprio)) { | |
3533 | __setscheduler_params(p, attr); | |
3534 | task_rq_unlock(rq, p, &flags); | |
3535 | return 0; | |
3536 | } | |
3537 | ||
fd2f4419 | 3538 | on_rq = p->on_rq; |
051a1d1a | 3539 | running = task_current(rq, p); |
0e1f3483 | 3540 | if (on_rq) |
4ca9b72b | 3541 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3542 | if (running) |
3543 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3544 | |
83ab0aa0 | 3545 | prev_class = p->sched_class; |
d50dde5a | 3546 | __setscheduler(rq, p, attr); |
f6b53205 | 3547 | |
0e1f3483 HS |
3548 | if (running) |
3549 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3550 | if (on_rq) { |
3551 | /* | |
3552 | * We enqueue to tail when the priority of a task is | |
3553 | * increased (user space view). | |
3554 | */ | |
3555 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3556 | } | |
cb469845 | 3557 | |
da7a735e | 3558 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3559 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3560 | |
95e02ca9 TG |
3561 | rt_mutex_adjust_pi(p); |
3562 | ||
1da177e4 LT |
3563 | return 0; |
3564 | } | |
961ccddd | 3565 | |
7479f3c9 PZ |
3566 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3567 | const struct sched_param *param, bool check) | |
3568 | { | |
3569 | struct sched_attr attr = { | |
3570 | .sched_policy = policy, | |
3571 | .sched_priority = param->sched_priority, | |
3572 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3573 | }; | |
3574 | ||
3575 | /* | |
3576 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3577 | */ | |
3578 | if (policy & SCHED_RESET_ON_FORK) { | |
3579 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3580 | policy &= ~SCHED_RESET_ON_FORK; | |
3581 | attr.sched_policy = policy; | |
3582 | } | |
3583 | ||
3584 | return __sched_setscheduler(p, &attr, check); | |
3585 | } | |
961ccddd RR |
3586 | /** |
3587 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3588 | * @p: the task in question. | |
3589 | * @policy: new policy. | |
3590 | * @param: structure containing the new RT priority. | |
3591 | * | |
e69f6186 YB |
3592 | * Return: 0 on success. An error code otherwise. |
3593 | * | |
961ccddd RR |
3594 | * NOTE that the task may be already dead. |
3595 | */ | |
3596 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3597 | const struct sched_param *param) |
961ccddd | 3598 | { |
7479f3c9 | 3599 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3600 | } |
1da177e4 LT |
3601 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3602 | ||
d50dde5a DF |
3603 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3604 | { | |
3605 | return __sched_setscheduler(p, attr, true); | |
3606 | } | |
3607 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3608 | ||
961ccddd RR |
3609 | /** |
3610 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3611 | * @p: the task in question. | |
3612 | * @policy: new policy. | |
3613 | * @param: structure containing the new RT priority. | |
3614 | * | |
3615 | * Just like sched_setscheduler, only don't bother checking if the | |
3616 | * current context has permission. For example, this is needed in | |
3617 | * stop_machine(): we create temporary high priority worker threads, | |
3618 | * but our caller might not have that capability. | |
e69f6186 YB |
3619 | * |
3620 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3621 | */ |
3622 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3623 | const struct sched_param *param) |
961ccddd | 3624 | { |
7479f3c9 | 3625 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3626 | } |
3627 | ||
95cdf3b7 IM |
3628 | static int |
3629 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3630 | { |
1da177e4 LT |
3631 | struct sched_param lparam; |
3632 | struct task_struct *p; | |
36c8b586 | 3633 | int retval; |
1da177e4 LT |
3634 | |
3635 | if (!param || pid < 0) | |
3636 | return -EINVAL; | |
3637 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3638 | return -EFAULT; | |
5fe1d75f ON |
3639 | |
3640 | rcu_read_lock(); | |
3641 | retval = -ESRCH; | |
1da177e4 | 3642 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3643 | if (p != NULL) |
3644 | retval = sched_setscheduler(p, policy, &lparam); | |
3645 | rcu_read_unlock(); | |
36c8b586 | 3646 | |
1da177e4 LT |
3647 | return retval; |
3648 | } | |
3649 | ||
d50dde5a DF |
3650 | /* |
3651 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3652 | */ | |
3653 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3654 | struct sched_attr *attr) | |
3655 | { | |
3656 | u32 size; | |
3657 | int ret; | |
3658 | ||
3659 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3660 | return -EFAULT; | |
3661 | ||
3662 | /* | |
3663 | * zero the full structure, so that a short copy will be nice. | |
3664 | */ | |
3665 | memset(attr, 0, sizeof(*attr)); | |
3666 | ||
3667 | ret = get_user(size, &uattr->size); | |
3668 | if (ret) | |
3669 | return ret; | |
3670 | ||
3671 | if (size > PAGE_SIZE) /* silly large */ | |
3672 | goto err_size; | |
3673 | ||
3674 | if (!size) /* abi compat */ | |
3675 | size = SCHED_ATTR_SIZE_VER0; | |
3676 | ||
3677 | if (size < SCHED_ATTR_SIZE_VER0) | |
3678 | goto err_size; | |
3679 | ||
3680 | /* | |
3681 | * If we're handed a bigger struct than we know of, | |
3682 | * ensure all the unknown bits are 0 - i.e. new | |
3683 | * user-space does not rely on any kernel feature | |
3684 | * extensions we dont know about yet. | |
3685 | */ | |
3686 | if (size > sizeof(*attr)) { | |
3687 | unsigned char __user *addr; | |
3688 | unsigned char __user *end; | |
3689 | unsigned char val; | |
3690 | ||
3691 | addr = (void __user *)uattr + sizeof(*attr); | |
3692 | end = (void __user *)uattr + size; | |
3693 | ||
3694 | for (; addr < end; addr++) { | |
3695 | ret = get_user(val, addr); | |
3696 | if (ret) | |
3697 | return ret; | |
3698 | if (val) | |
3699 | goto err_size; | |
3700 | } | |
3701 | size = sizeof(*attr); | |
3702 | } | |
3703 | ||
3704 | ret = copy_from_user(attr, uattr, size); | |
3705 | if (ret) | |
3706 | return -EFAULT; | |
3707 | ||
3708 | /* | |
3709 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3710 | * to be strict and return an error on out-of-bounds values? | |
3711 | */ | |
75e45d51 | 3712 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3713 | |
e78c7bca | 3714 | return 0; |
d50dde5a DF |
3715 | |
3716 | err_size: | |
3717 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3718 | return -E2BIG; |
d50dde5a DF |
3719 | } |
3720 | ||
1da177e4 LT |
3721 | /** |
3722 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3723 | * @pid: the pid in question. | |
3724 | * @policy: new policy. | |
3725 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3726 | * |
3727 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3728 | */ |
5add95d4 HC |
3729 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3730 | struct sched_param __user *, param) | |
1da177e4 | 3731 | { |
c21761f1 JB |
3732 | /* negative values for policy are not valid */ |
3733 | if (policy < 0) | |
3734 | return -EINVAL; | |
3735 | ||
1da177e4 LT |
3736 | return do_sched_setscheduler(pid, policy, param); |
3737 | } | |
3738 | ||
3739 | /** | |
3740 | * sys_sched_setparam - set/change the RT priority of a thread | |
3741 | * @pid: the pid in question. | |
3742 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3743 | * |
3744 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3745 | */ |
5add95d4 | 3746 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3747 | { |
3748 | return do_sched_setscheduler(pid, -1, param); | |
3749 | } | |
3750 | ||
d50dde5a DF |
3751 | /** |
3752 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3753 | * @pid: the pid in question. | |
5778fccf | 3754 | * @uattr: structure containing the extended parameters. |
db66d756 | 3755 | * @flags: for future extension. |
d50dde5a | 3756 | */ |
6d35ab48 PZ |
3757 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3758 | unsigned int, flags) | |
d50dde5a DF |
3759 | { |
3760 | struct sched_attr attr; | |
3761 | struct task_struct *p; | |
3762 | int retval; | |
3763 | ||
6d35ab48 | 3764 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3765 | return -EINVAL; |
3766 | ||
143cf23d MK |
3767 | retval = sched_copy_attr(uattr, &attr); |
3768 | if (retval) | |
3769 | return retval; | |
d50dde5a | 3770 | |
b14ed2c2 | 3771 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 3772 | return -EINVAL; |
d50dde5a DF |
3773 | |
3774 | rcu_read_lock(); | |
3775 | retval = -ESRCH; | |
3776 | p = find_process_by_pid(pid); | |
3777 | if (p != NULL) | |
3778 | retval = sched_setattr(p, &attr); | |
3779 | rcu_read_unlock(); | |
3780 | ||
3781 | return retval; | |
3782 | } | |
3783 | ||
1da177e4 LT |
3784 | /** |
3785 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3786 | * @pid: the pid in question. | |
e69f6186 YB |
3787 | * |
3788 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3789 | * code. | |
1da177e4 | 3790 | */ |
5add95d4 | 3791 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3792 | { |
36c8b586 | 3793 | struct task_struct *p; |
3a5c359a | 3794 | int retval; |
1da177e4 LT |
3795 | |
3796 | if (pid < 0) | |
3a5c359a | 3797 | return -EINVAL; |
1da177e4 LT |
3798 | |
3799 | retval = -ESRCH; | |
5fe85be0 | 3800 | rcu_read_lock(); |
1da177e4 LT |
3801 | p = find_process_by_pid(pid); |
3802 | if (p) { | |
3803 | retval = security_task_getscheduler(p); | |
3804 | if (!retval) | |
ca94c442 LP |
3805 | retval = p->policy |
3806 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3807 | } |
5fe85be0 | 3808 | rcu_read_unlock(); |
1da177e4 LT |
3809 | return retval; |
3810 | } | |
3811 | ||
3812 | /** | |
ca94c442 | 3813 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3814 | * @pid: the pid in question. |
3815 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3816 | * |
3817 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3818 | * code. | |
1da177e4 | 3819 | */ |
5add95d4 | 3820 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3821 | { |
ce5f7f82 | 3822 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3823 | struct task_struct *p; |
3a5c359a | 3824 | int retval; |
1da177e4 LT |
3825 | |
3826 | if (!param || pid < 0) | |
3a5c359a | 3827 | return -EINVAL; |
1da177e4 | 3828 | |
5fe85be0 | 3829 | rcu_read_lock(); |
1da177e4 LT |
3830 | p = find_process_by_pid(pid); |
3831 | retval = -ESRCH; | |
3832 | if (!p) | |
3833 | goto out_unlock; | |
3834 | ||
3835 | retval = security_task_getscheduler(p); | |
3836 | if (retval) | |
3837 | goto out_unlock; | |
3838 | ||
ce5f7f82 PZ |
3839 | if (task_has_rt_policy(p)) |
3840 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3841 | rcu_read_unlock(); |
1da177e4 LT |
3842 | |
3843 | /* | |
3844 | * This one might sleep, we cannot do it with a spinlock held ... | |
3845 | */ | |
3846 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3847 | ||
1da177e4 LT |
3848 | return retval; |
3849 | ||
3850 | out_unlock: | |
5fe85be0 | 3851 | rcu_read_unlock(); |
1da177e4 LT |
3852 | return retval; |
3853 | } | |
3854 | ||
d50dde5a DF |
3855 | static int sched_read_attr(struct sched_attr __user *uattr, |
3856 | struct sched_attr *attr, | |
3857 | unsigned int usize) | |
3858 | { | |
3859 | int ret; | |
3860 | ||
3861 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3862 | return -EFAULT; | |
3863 | ||
3864 | /* | |
3865 | * If we're handed a smaller struct than we know of, | |
3866 | * ensure all the unknown bits are 0 - i.e. old | |
3867 | * user-space does not get uncomplete information. | |
3868 | */ | |
3869 | if (usize < sizeof(*attr)) { | |
3870 | unsigned char *addr; | |
3871 | unsigned char *end; | |
3872 | ||
3873 | addr = (void *)attr + usize; | |
3874 | end = (void *)attr + sizeof(*attr); | |
3875 | ||
3876 | for (; addr < end; addr++) { | |
3877 | if (*addr) | |
22400674 | 3878 | return -EFBIG; |
d50dde5a DF |
3879 | } |
3880 | ||
3881 | attr->size = usize; | |
3882 | } | |
3883 | ||
4efbc454 | 3884 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3885 | if (ret) |
3886 | return -EFAULT; | |
3887 | ||
22400674 | 3888 | return 0; |
d50dde5a DF |
3889 | } |
3890 | ||
3891 | /** | |
aab03e05 | 3892 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3893 | * @pid: the pid in question. |
5778fccf | 3894 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3895 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3896 | * @flags: for future extension. |
d50dde5a | 3897 | */ |
6d35ab48 PZ |
3898 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3899 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3900 | { |
3901 | struct sched_attr attr = { | |
3902 | .size = sizeof(struct sched_attr), | |
3903 | }; | |
3904 | struct task_struct *p; | |
3905 | int retval; | |
3906 | ||
3907 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3908 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3909 | return -EINVAL; |
3910 | ||
3911 | rcu_read_lock(); | |
3912 | p = find_process_by_pid(pid); | |
3913 | retval = -ESRCH; | |
3914 | if (!p) | |
3915 | goto out_unlock; | |
3916 | ||
3917 | retval = security_task_getscheduler(p); | |
3918 | if (retval) | |
3919 | goto out_unlock; | |
3920 | ||
3921 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3922 | if (p->sched_reset_on_fork) |
3923 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3924 | if (task_has_dl_policy(p)) |
3925 | __getparam_dl(p, &attr); | |
3926 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3927 | attr.sched_priority = p->rt_priority; |
3928 | else | |
d0ea0268 | 3929 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3930 | |
3931 | rcu_read_unlock(); | |
3932 | ||
3933 | retval = sched_read_attr(uattr, &attr, size); | |
3934 | return retval; | |
3935 | ||
3936 | out_unlock: | |
3937 | rcu_read_unlock(); | |
3938 | return retval; | |
3939 | } | |
3940 | ||
96f874e2 | 3941 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3942 | { |
5a16f3d3 | 3943 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3944 | struct task_struct *p; |
3945 | int retval; | |
1da177e4 | 3946 | |
23f5d142 | 3947 | rcu_read_lock(); |
1da177e4 LT |
3948 | |
3949 | p = find_process_by_pid(pid); | |
3950 | if (!p) { | |
23f5d142 | 3951 | rcu_read_unlock(); |
1da177e4 LT |
3952 | return -ESRCH; |
3953 | } | |
3954 | ||
23f5d142 | 3955 | /* Prevent p going away */ |
1da177e4 | 3956 | get_task_struct(p); |
23f5d142 | 3957 | rcu_read_unlock(); |
1da177e4 | 3958 | |
14a40ffc TH |
3959 | if (p->flags & PF_NO_SETAFFINITY) { |
3960 | retval = -EINVAL; | |
3961 | goto out_put_task; | |
3962 | } | |
5a16f3d3 RR |
3963 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3964 | retval = -ENOMEM; | |
3965 | goto out_put_task; | |
3966 | } | |
3967 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3968 | retval = -ENOMEM; | |
3969 | goto out_free_cpus_allowed; | |
3970 | } | |
1da177e4 | 3971 | retval = -EPERM; |
4c44aaaf EB |
3972 | if (!check_same_owner(p)) { |
3973 | rcu_read_lock(); | |
3974 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3975 | rcu_read_unlock(); | |
3976 | goto out_unlock; | |
3977 | } | |
3978 | rcu_read_unlock(); | |
3979 | } | |
1da177e4 | 3980 | |
b0ae1981 | 3981 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3982 | if (retval) |
3983 | goto out_unlock; | |
3984 | ||
e4099a5e PZ |
3985 | |
3986 | cpuset_cpus_allowed(p, cpus_allowed); | |
3987 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3988 | ||
332ac17e DF |
3989 | /* |
3990 | * Since bandwidth control happens on root_domain basis, | |
3991 | * if admission test is enabled, we only admit -deadline | |
3992 | * tasks allowed to run on all the CPUs in the task's | |
3993 | * root_domain. | |
3994 | */ | |
3995 | #ifdef CONFIG_SMP | |
3996 | if (task_has_dl_policy(p)) { | |
3997 | const struct cpumask *span = task_rq(p)->rd->span; | |
3998 | ||
e4099a5e | 3999 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
4000 | retval = -EBUSY; |
4001 | goto out_unlock; | |
4002 | } | |
4003 | } | |
4004 | #endif | |
49246274 | 4005 | again: |
5a16f3d3 | 4006 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4007 | |
8707d8b8 | 4008 | if (!retval) { |
5a16f3d3 RR |
4009 | cpuset_cpus_allowed(p, cpus_allowed); |
4010 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4011 | /* |
4012 | * We must have raced with a concurrent cpuset | |
4013 | * update. Just reset the cpus_allowed to the | |
4014 | * cpuset's cpus_allowed | |
4015 | */ | |
5a16f3d3 | 4016 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4017 | goto again; |
4018 | } | |
4019 | } | |
1da177e4 | 4020 | out_unlock: |
5a16f3d3 RR |
4021 | free_cpumask_var(new_mask); |
4022 | out_free_cpus_allowed: | |
4023 | free_cpumask_var(cpus_allowed); | |
4024 | out_put_task: | |
1da177e4 | 4025 | put_task_struct(p); |
1da177e4 LT |
4026 | return retval; |
4027 | } | |
4028 | ||
4029 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4030 | struct cpumask *new_mask) |
1da177e4 | 4031 | { |
96f874e2 RR |
4032 | if (len < cpumask_size()) |
4033 | cpumask_clear(new_mask); | |
4034 | else if (len > cpumask_size()) | |
4035 | len = cpumask_size(); | |
4036 | ||
1da177e4 LT |
4037 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4038 | } | |
4039 | ||
4040 | /** | |
4041 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4042 | * @pid: pid of the process | |
4043 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4044 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
4045 | * |
4046 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4047 | */ |
5add95d4 HC |
4048 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4049 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4050 | { |
5a16f3d3 | 4051 | cpumask_var_t new_mask; |
1da177e4 LT |
4052 | int retval; |
4053 | ||
5a16f3d3 RR |
4054 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4055 | return -ENOMEM; | |
1da177e4 | 4056 | |
5a16f3d3 RR |
4057 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4058 | if (retval == 0) | |
4059 | retval = sched_setaffinity(pid, new_mask); | |
4060 | free_cpumask_var(new_mask); | |
4061 | return retval; | |
1da177e4 LT |
4062 | } |
4063 | ||
96f874e2 | 4064 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4065 | { |
36c8b586 | 4066 | struct task_struct *p; |
31605683 | 4067 | unsigned long flags; |
1da177e4 | 4068 | int retval; |
1da177e4 | 4069 | |
23f5d142 | 4070 | rcu_read_lock(); |
1da177e4 LT |
4071 | |
4072 | retval = -ESRCH; | |
4073 | p = find_process_by_pid(pid); | |
4074 | if (!p) | |
4075 | goto out_unlock; | |
4076 | ||
e7834f8f DQ |
4077 | retval = security_task_getscheduler(p); |
4078 | if (retval) | |
4079 | goto out_unlock; | |
4080 | ||
013fdb80 | 4081 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4082 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4083 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4084 | |
4085 | out_unlock: | |
23f5d142 | 4086 | rcu_read_unlock(); |
1da177e4 | 4087 | |
9531b62f | 4088 | return retval; |
1da177e4 LT |
4089 | } |
4090 | ||
4091 | /** | |
4092 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4093 | * @pid: pid of the process | |
4094 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4095 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4096 | * |
4097 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4098 | */ |
5add95d4 HC |
4099 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4100 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4101 | { |
4102 | int ret; | |
f17c8607 | 4103 | cpumask_var_t mask; |
1da177e4 | 4104 | |
84fba5ec | 4105 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4106 | return -EINVAL; |
4107 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4108 | return -EINVAL; |
4109 | ||
f17c8607 RR |
4110 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4111 | return -ENOMEM; | |
1da177e4 | 4112 | |
f17c8607 RR |
4113 | ret = sched_getaffinity(pid, mask); |
4114 | if (ret == 0) { | |
8bc037fb | 4115 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4116 | |
4117 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4118 | ret = -EFAULT; |
4119 | else | |
cd3d8031 | 4120 | ret = retlen; |
f17c8607 RR |
4121 | } |
4122 | free_cpumask_var(mask); | |
1da177e4 | 4123 | |
f17c8607 | 4124 | return ret; |
1da177e4 LT |
4125 | } |
4126 | ||
4127 | /** | |
4128 | * sys_sched_yield - yield the current processor to other threads. | |
4129 | * | |
dd41f596 IM |
4130 | * This function yields the current CPU to other tasks. If there are no |
4131 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4132 | * |
4133 | * Return: 0. | |
1da177e4 | 4134 | */ |
5add95d4 | 4135 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4136 | { |
70b97a7f | 4137 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4138 | |
2d72376b | 4139 | schedstat_inc(rq, yld_count); |
4530d7ab | 4140 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4141 | |
4142 | /* | |
4143 | * Since we are going to call schedule() anyway, there's | |
4144 | * no need to preempt or enable interrupts: | |
4145 | */ | |
4146 | __release(rq->lock); | |
8a25d5de | 4147 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4148 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4149 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4150 | |
4151 | schedule(); | |
4152 | ||
4153 | return 0; | |
4154 | } | |
4155 | ||
e7b38404 | 4156 | static void __cond_resched(void) |
1da177e4 | 4157 | { |
bdb43806 | 4158 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4159 | __schedule(); |
bdb43806 | 4160 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4161 | } |
4162 | ||
02b67cc3 | 4163 | int __sched _cond_resched(void) |
1da177e4 | 4164 | { |
ac1bea85 | 4165 | rcu_cond_resched(); |
d86ee480 | 4166 | if (should_resched()) { |
1da177e4 LT |
4167 | __cond_resched(); |
4168 | return 1; | |
4169 | } | |
4170 | return 0; | |
4171 | } | |
02b67cc3 | 4172 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4173 | |
4174 | /* | |
613afbf8 | 4175 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4176 | * call schedule, and on return reacquire the lock. |
4177 | * | |
41a2d6cf | 4178 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4179 | * operations here to prevent schedule() from being called twice (once via |
4180 | * spin_unlock(), once by hand). | |
4181 | */ | |
613afbf8 | 4182 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4183 | { |
ac1bea85 | 4184 | bool need_rcu_resched = rcu_should_resched(); |
d86ee480 | 4185 | int resched = should_resched(); |
6df3cecb JK |
4186 | int ret = 0; |
4187 | ||
f607c668 PZ |
4188 | lockdep_assert_held(lock); |
4189 | ||
ac1bea85 | 4190 | if (spin_needbreak(lock) || resched || need_rcu_resched) { |
1da177e4 | 4191 | spin_unlock(lock); |
d86ee480 | 4192 | if (resched) |
95c354fe | 4193 | __cond_resched(); |
ac1bea85 PM |
4194 | else if (unlikely(need_rcu_resched)) |
4195 | rcu_resched(); | |
95c354fe NP |
4196 | else |
4197 | cpu_relax(); | |
6df3cecb | 4198 | ret = 1; |
1da177e4 | 4199 | spin_lock(lock); |
1da177e4 | 4200 | } |
6df3cecb | 4201 | return ret; |
1da177e4 | 4202 | } |
613afbf8 | 4203 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4204 | |
613afbf8 | 4205 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4206 | { |
4207 | BUG_ON(!in_softirq()); | |
4208 | ||
ac1bea85 | 4209 | rcu_cond_resched(); /* BH disabled OK, just recording QSes. */ |
d86ee480 | 4210 | if (should_resched()) { |
98d82567 | 4211 | local_bh_enable(); |
1da177e4 LT |
4212 | __cond_resched(); |
4213 | local_bh_disable(); | |
4214 | return 1; | |
4215 | } | |
4216 | return 0; | |
4217 | } | |
613afbf8 | 4218 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4219 | |
1da177e4 LT |
4220 | /** |
4221 | * yield - yield the current processor to other threads. | |
4222 | * | |
8e3fabfd PZ |
4223 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4224 | * | |
4225 | * The scheduler is at all times free to pick the calling task as the most | |
4226 | * eligible task to run, if removing the yield() call from your code breaks | |
4227 | * it, its already broken. | |
4228 | * | |
4229 | * Typical broken usage is: | |
4230 | * | |
4231 | * while (!event) | |
4232 | * yield(); | |
4233 | * | |
4234 | * where one assumes that yield() will let 'the other' process run that will | |
4235 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4236 | * happen. Never use yield() as a progress guarantee!! | |
4237 | * | |
4238 | * If you want to use yield() to wait for something, use wait_event(). | |
4239 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4240 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4241 | */ |
4242 | void __sched yield(void) | |
4243 | { | |
4244 | set_current_state(TASK_RUNNING); | |
4245 | sys_sched_yield(); | |
4246 | } | |
1da177e4 LT |
4247 | EXPORT_SYMBOL(yield); |
4248 | ||
d95f4122 MG |
4249 | /** |
4250 | * yield_to - yield the current processor to another thread in | |
4251 | * your thread group, or accelerate that thread toward the | |
4252 | * processor it's on. | |
16addf95 RD |
4253 | * @p: target task |
4254 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4255 | * |
4256 | * It's the caller's job to ensure that the target task struct | |
4257 | * can't go away on us before we can do any checks. | |
4258 | * | |
e69f6186 | 4259 | * Return: |
7b270f60 PZ |
4260 | * true (>0) if we indeed boosted the target task. |
4261 | * false (0) if we failed to boost the target. | |
4262 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4263 | */ |
fa93384f | 4264 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4265 | { |
4266 | struct task_struct *curr = current; | |
4267 | struct rq *rq, *p_rq; | |
4268 | unsigned long flags; | |
c3c18640 | 4269 | int yielded = 0; |
d95f4122 MG |
4270 | |
4271 | local_irq_save(flags); | |
4272 | rq = this_rq(); | |
4273 | ||
4274 | again: | |
4275 | p_rq = task_rq(p); | |
7b270f60 PZ |
4276 | /* |
4277 | * If we're the only runnable task on the rq and target rq also | |
4278 | * has only one task, there's absolutely no point in yielding. | |
4279 | */ | |
4280 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4281 | yielded = -ESRCH; | |
4282 | goto out_irq; | |
4283 | } | |
4284 | ||
d95f4122 | 4285 | double_rq_lock(rq, p_rq); |
39e24d8f | 4286 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4287 | double_rq_unlock(rq, p_rq); |
4288 | goto again; | |
4289 | } | |
4290 | ||
4291 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4292 | goto out_unlock; |
d95f4122 MG |
4293 | |
4294 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4295 | goto out_unlock; |
d95f4122 MG |
4296 | |
4297 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4298 | goto out_unlock; |
d95f4122 MG |
4299 | |
4300 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4301 | if (yielded) { |
d95f4122 | 4302 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4303 | /* |
4304 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4305 | * fairness. | |
4306 | */ | |
4307 | if (preempt && rq != p_rq) | |
8875125e | 4308 | resched_curr(p_rq); |
6d1cafd8 | 4309 | } |
d95f4122 | 4310 | |
7b270f60 | 4311 | out_unlock: |
d95f4122 | 4312 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4313 | out_irq: |
d95f4122 MG |
4314 | local_irq_restore(flags); |
4315 | ||
7b270f60 | 4316 | if (yielded > 0) |
d95f4122 MG |
4317 | schedule(); |
4318 | ||
4319 | return yielded; | |
4320 | } | |
4321 | EXPORT_SYMBOL_GPL(yield_to); | |
4322 | ||
1da177e4 | 4323 | /* |
41a2d6cf | 4324 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4325 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4326 | */ |
4327 | void __sched io_schedule(void) | |
4328 | { | |
54d35f29 | 4329 | struct rq *rq = raw_rq(); |
1da177e4 | 4330 | |
0ff92245 | 4331 | delayacct_blkio_start(); |
1da177e4 | 4332 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4333 | blk_flush_plug(current); |
8f0dfc34 | 4334 | current->in_iowait = 1; |
1da177e4 | 4335 | schedule(); |
8f0dfc34 | 4336 | current->in_iowait = 0; |
1da177e4 | 4337 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4338 | delayacct_blkio_end(); |
1da177e4 | 4339 | } |
1da177e4 LT |
4340 | EXPORT_SYMBOL(io_schedule); |
4341 | ||
4342 | long __sched io_schedule_timeout(long timeout) | |
4343 | { | |
54d35f29 | 4344 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4345 | long ret; |
4346 | ||
0ff92245 | 4347 | delayacct_blkio_start(); |
1da177e4 | 4348 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4349 | blk_flush_plug(current); |
8f0dfc34 | 4350 | current->in_iowait = 1; |
1da177e4 | 4351 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4352 | current->in_iowait = 0; |
1da177e4 | 4353 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4354 | delayacct_blkio_end(); |
1da177e4 LT |
4355 | return ret; |
4356 | } | |
4357 | ||
4358 | /** | |
4359 | * sys_sched_get_priority_max - return maximum RT priority. | |
4360 | * @policy: scheduling class. | |
4361 | * | |
e69f6186 YB |
4362 | * Return: On success, this syscall returns the maximum |
4363 | * rt_priority that can be used by a given scheduling class. | |
4364 | * On failure, a negative error code is returned. | |
1da177e4 | 4365 | */ |
5add95d4 | 4366 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4367 | { |
4368 | int ret = -EINVAL; | |
4369 | ||
4370 | switch (policy) { | |
4371 | case SCHED_FIFO: | |
4372 | case SCHED_RR: | |
4373 | ret = MAX_USER_RT_PRIO-1; | |
4374 | break; | |
aab03e05 | 4375 | case SCHED_DEADLINE: |
1da177e4 | 4376 | case SCHED_NORMAL: |
b0a9499c | 4377 | case SCHED_BATCH: |
dd41f596 | 4378 | case SCHED_IDLE: |
1da177e4 LT |
4379 | ret = 0; |
4380 | break; | |
4381 | } | |
4382 | return ret; | |
4383 | } | |
4384 | ||
4385 | /** | |
4386 | * sys_sched_get_priority_min - return minimum RT priority. | |
4387 | * @policy: scheduling class. | |
4388 | * | |
e69f6186 YB |
4389 | * Return: On success, this syscall returns the minimum |
4390 | * rt_priority that can be used by a given scheduling class. | |
4391 | * On failure, a negative error code is returned. | |
1da177e4 | 4392 | */ |
5add95d4 | 4393 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4394 | { |
4395 | int ret = -EINVAL; | |
4396 | ||
4397 | switch (policy) { | |
4398 | case SCHED_FIFO: | |
4399 | case SCHED_RR: | |
4400 | ret = 1; | |
4401 | break; | |
aab03e05 | 4402 | case SCHED_DEADLINE: |
1da177e4 | 4403 | case SCHED_NORMAL: |
b0a9499c | 4404 | case SCHED_BATCH: |
dd41f596 | 4405 | case SCHED_IDLE: |
1da177e4 LT |
4406 | ret = 0; |
4407 | } | |
4408 | return ret; | |
4409 | } | |
4410 | ||
4411 | /** | |
4412 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4413 | * @pid: pid of the process. | |
4414 | * @interval: userspace pointer to the timeslice value. | |
4415 | * | |
4416 | * this syscall writes the default timeslice value of a given process | |
4417 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4418 | * |
4419 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4420 | * an error code. | |
1da177e4 | 4421 | */ |
17da2bd9 | 4422 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4423 | struct timespec __user *, interval) |
1da177e4 | 4424 | { |
36c8b586 | 4425 | struct task_struct *p; |
a4ec24b4 | 4426 | unsigned int time_slice; |
dba091b9 TG |
4427 | unsigned long flags; |
4428 | struct rq *rq; | |
3a5c359a | 4429 | int retval; |
1da177e4 | 4430 | struct timespec t; |
1da177e4 LT |
4431 | |
4432 | if (pid < 0) | |
3a5c359a | 4433 | return -EINVAL; |
1da177e4 LT |
4434 | |
4435 | retval = -ESRCH; | |
1a551ae7 | 4436 | rcu_read_lock(); |
1da177e4 LT |
4437 | p = find_process_by_pid(pid); |
4438 | if (!p) | |
4439 | goto out_unlock; | |
4440 | ||
4441 | retval = security_task_getscheduler(p); | |
4442 | if (retval) | |
4443 | goto out_unlock; | |
4444 | ||
dba091b9 | 4445 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4446 | time_slice = 0; |
4447 | if (p->sched_class->get_rr_interval) | |
4448 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4449 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4450 | |
1a551ae7 | 4451 | rcu_read_unlock(); |
a4ec24b4 | 4452 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4453 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4454 | return retval; |
3a5c359a | 4455 | |
1da177e4 | 4456 | out_unlock: |
1a551ae7 | 4457 | rcu_read_unlock(); |
1da177e4 LT |
4458 | return retval; |
4459 | } | |
4460 | ||
7c731e0a | 4461 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4462 | |
82a1fcb9 | 4463 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4464 | { |
1da177e4 | 4465 | unsigned long free = 0; |
4e79752c | 4466 | int ppid; |
36c8b586 | 4467 | unsigned state; |
1da177e4 | 4468 | |
1da177e4 | 4469 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4470 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4471 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4472 | #if BITS_PER_LONG == 32 |
1da177e4 | 4473 | if (state == TASK_RUNNING) |
3df0fc5b | 4474 | printk(KERN_CONT " running "); |
1da177e4 | 4475 | else |
3df0fc5b | 4476 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4477 | #else |
4478 | if (state == TASK_RUNNING) | |
3df0fc5b | 4479 | printk(KERN_CONT " running task "); |
1da177e4 | 4480 | else |
3df0fc5b | 4481 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4482 | #endif |
4483 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4484 | free = stack_not_used(p); |
1da177e4 | 4485 | #endif |
4e79752c PM |
4486 | rcu_read_lock(); |
4487 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4488 | rcu_read_unlock(); | |
3df0fc5b | 4489 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4490 | task_pid_nr(p), ppid, |
aa47b7e0 | 4491 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4492 | |
3d1cb205 | 4493 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4494 | show_stack(p, NULL); |
1da177e4 LT |
4495 | } |
4496 | ||
e59e2ae2 | 4497 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4498 | { |
36c8b586 | 4499 | struct task_struct *g, *p; |
1da177e4 | 4500 | |
4bd77321 | 4501 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4502 | printk(KERN_INFO |
4503 | " task PC stack pid father\n"); | |
1da177e4 | 4504 | #else |
3df0fc5b PZ |
4505 | printk(KERN_INFO |
4506 | " task PC stack pid father\n"); | |
1da177e4 | 4507 | #endif |
510f5acc | 4508 | rcu_read_lock(); |
1da177e4 LT |
4509 | do_each_thread(g, p) { |
4510 | /* | |
4511 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4512 | * console might take a lot of time: |
1da177e4 LT |
4513 | */ |
4514 | touch_nmi_watchdog(); | |
39bc89fd | 4515 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4516 | sched_show_task(p); |
1da177e4 LT |
4517 | } while_each_thread(g, p); |
4518 | ||
04c9167f JF |
4519 | touch_all_softlockup_watchdogs(); |
4520 | ||
dd41f596 IM |
4521 | #ifdef CONFIG_SCHED_DEBUG |
4522 | sysrq_sched_debug_show(); | |
4523 | #endif | |
510f5acc | 4524 | rcu_read_unlock(); |
e59e2ae2 IM |
4525 | /* |
4526 | * Only show locks if all tasks are dumped: | |
4527 | */ | |
93335a21 | 4528 | if (!state_filter) |
e59e2ae2 | 4529 | debug_show_all_locks(); |
1da177e4 LT |
4530 | } |
4531 | ||
0db0628d | 4532 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4533 | { |
dd41f596 | 4534 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4535 | } |
4536 | ||
f340c0d1 IM |
4537 | /** |
4538 | * init_idle - set up an idle thread for a given CPU | |
4539 | * @idle: task in question | |
4540 | * @cpu: cpu the idle task belongs to | |
4541 | * | |
4542 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4543 | * flag, to make booting more robust. | |
4544 | */ | |
0db0628d | 4545 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4546 | { |
70b97a7f | 4547 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4548 | unsigned long flags; |
4549 | ||
05fa785c | 4550 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4551 | |
5e1576ed | 4552 | __sched_fork(0, idle); |
06b83b5f | 4553 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4554 | idle->se.exec_start = sched_clock(); |
4555 | ||
1e1b6c51 | 4556 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4557 | /* |
4558 | * We're having a chicken and egg problem, even though we are | |
4559 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4560 | * lockdep check in task_group() will fail. | |
4561 | * | |
4562 | * Similar case to sched_fork(). / Alternatively we could | |
4563 | * use task_rq_lock() here and obtain the other rq->lock. | |
4564 | * | |
4565 | * Silence PROVE_RCU | |
4566 | */ | |
4567 | rcu_read_lock(); | |
dd41f596 | 4568 | __set_task_cpu(idle, cpu); |
6506cf6c | 4569 | rcu_read_unlock(); |
1da177e4 | 4570 | |
1da177e4 | 4571 | rq->curr = rq->idle = idle; |
77177856 | 4572 | idle->on_rq = 1; |
3ca7a440 PZ |
4573 | #if defined(CONFIG_SMP) |
4574 | idle->on_cpu = 1; | |
4866cde0 | 4575 | #endif |
05fa785c | 4576 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4577 | |
4578 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4579 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4580 | |
dd41f596 IM |
4581 | /* |
4582 | * The idle tasks have their own, simple scheduling class: | |
4583 | */ | |
4584 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4585 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4586 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4587 | #if defined(CONFIG_SMP) |
4588 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4589 | #endif | |
19978ca6 IM |
4590 | } |
4591 | ||
1da177e4 | 4592 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4593 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4594 | { | |
4595 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4596 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4597 | |
4598 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4599 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4600 | } |
4601 | ||
1da177e4 LT |
4602 | /* |
4603 | * This is how migration works: | |
4604 | * | |
969c7921 TH |
4605 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4606 | * stop_one_cpu(). | |
4607 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4608 | * off the CPU) | |
4609 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4610 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4611 | * it and puts it into the right queue. |
969c7921 TH |
4612 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4613 | * is done. | |
1da177e4 LT |
4614 | */ |
4615 | ||
4616 | /* | |
4617 | * Change a given task's CPU affinity. Migrate the thread to a | |
4618 | * proper CPU and schedule it away if the CPU it's executing on | |
4619 | * is removed from the allowed bitmask. | |
4620 | * | |
4621 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4622 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4623 | * call is not atomic; no spinlocks may be held. |
4624 | */ | |
96f874e2 | 4625 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4626 | { |
4627 | unsigned long flags; | |
70b97a7f | 4628 | struct rq *rq; |
969c7921 | 4629 | unsigned int dest_cpu; |
48f24c4d | 4630 | int ret = 0; |
1da177e4 LT |
4631 | |
4632 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4633 | |
db44fc01 YZ |
4634 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4635 | goto out; | |
4636 | ||
6ad4c188 | 4637 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4638 | ret = -EINVAL; |
4639 | goto out; | |
4640 | } | |
4641 | ||
1e1b6c51 | 4642 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4643 | |
1da177e4 | 4644 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4645 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4646 | goto out; |
4647 | ||
969c7921 | 4648 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4649 | if (p->on_rq) { |
969c7921 | 4650 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4651 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4652 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4653 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4654 | tlb_migrate_finish(p->mm); |
4655 | return 0; | |
4656 | } | |
4657 | out: | |
0122ec5b | 4658 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4659 | |
1da177e4 LT |
4660 | return ret; |
4661 | } | |
cd8ba7cd | 4662 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4663 | |
4664 | /* | |
41a2d6cf | 4665 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4666 | * this because either it can't run here any more (set_cpus_allowed() |
4667 | * away from this CPU, or CPU going down), or because we're | |
4668 | * attempting to rebalance this task on exec (sched_exec). | |
4669 | * | |
4670 | * So we race with normal scheduler movements, but that's OK, as long | |
4671 | * as the task is no longer on this CPU. | |
efc30814 KK |
4672 | * |
4673 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4674 | */ |
efc30814 | 4675 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4676 | { |
70b97a7f | 4677 | struct rq *rq_dest, *rq_src; |
e2912009 | 4678 | int ret = 0; |
1da177e4 | 4679 | |
e761b772 | 4680 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4681 | return ret; |
1da177e4 LT |
4682 | |
4683 | rq_src = cpu_rq(src_cpu); | |
4684 | rq_dest = cpu_rq(dest_cpu); | |
4685 | ||
0122ec5b | 4686 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4687 | double_rq_lock(rq_src, rq_dest); |
4688 | /* Already moved. */ | |
4689 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4690 | goto done; |
1da177e4 | 4691 | /* Affinity changed (again). */ |
fa17b507 | 4692 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4693 | goto fail; |
1da177e4 | 4694 | |
e2912009 PZ |
4695 | /* |
4696 | * If we're not on a rq, the next wake-up will ensure we're | |
4697 | * placed properly. | |
4698 | */ | |
fd2f4419 | 4699 | if (p->on_rq) { |
4ca9b72b | 4700 | dequeue_task(rq_src, p, 0); |
e2912009 | 4701 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4702 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4703 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4704 | } |
b1e38734 | 4705 | done: |
efc30814 | 4706 | ret = 1; |
b1e38734 | 4707 | fail: |
1da177e4 | 4708 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4709 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4710 | return ret; |
1da177e4 LT |
4711 | } |
4712 | ||
e6628d5b MG |
4713 | #ifdef CONFIG_NUMA_BALANCING |
4714 | /* Migrate current task p to target_cpu */ | |
4715 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4716 | { | |
4717 | struct migration_arg arg = { p, target_cpu }; | |
4718 | int curr_cpu = task_cpu(p); | |
4719 | ||
4720 | if (curr_cpu == target_cpu) | |
4721 | return 0; | |
4722 | ||
4723 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4724 | return -EINVAL; | |
4725 | ||
4726 | /* TODO: This is not properly updating schedstats */ | |
4727 | ||
286549dc | 4728 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4729 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4730 | } | |
0ec8aa00 PZ |
4731 | |
4732 | /* | |
4733 | * Requeue a task on a given node and accurately track the number of NUMA | |
4734 | * tasks on the runqueues | |
4735 | */ | |
4736 | void sched_setnuma(struct task_struct *p, int nid) | |
4737 | { | |
4738 | struct rq *rq; | |
4739 | unsigned long flags; | |
4740 | bool on_rq, running; | |
4741 | ||
4742 | rq = task_rq_lock(p, &flags); | |
4743 | on_rq = p->on_rq; | |
4744 | running = task_current(rq, p); | |
4745 | ||
4746 | if (on_rq) | |
4747 | dequeue_task(rq, p, 0); | |
4748 | if (running) | |
4749 | p->sched_class->put_prev_task(rq, p); | |
4750 | ||
4751 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4752 | |
4753 | if (running) | |
4754 | p->sched_class->set_curr_task(rq); | |
4755 | if (on_rq) | |
4756 | enqueue_task(rq, p, 0); | |
4757 | task_rq_unlock(rq, p, &flags); | |
4758 | } | |
e6628d5b MG |
4759 | #endif |
4760 | ||
1da177e4 | 4761 | /* |
969c7921 TH |
4762 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4763 | * and performs thread migration by bumping thread off CPU then | |
4764 | * 'pushing' onto another runqueue. | |
1da177e4 | 4765 | */ |
969c7921 | 4766 | static int migration_cpu_stop(void *data) |
1da177e4 | 4767 | { |
969c7921 | 4768 | struct migration_arg *arg = data; |
f7b4cddc | 4769 | |
969c7921 TH |
4770 | /* |
4771 | * The original target cpu might have gone down and we might | |
4772 | * be on another cpu but it doesn't matter. | |
4773 | */ | |
f7b4cddc | 4774 | local_irq_disable(); |
969c7921 | 4775 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4776 | local_irq_enable(); |
1da177e4 | 4777 | return 0; |
f7b4cddc ON |
4778 | } |
4779 | ||
1da177e4 | 4780 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4781 | |
054b9108 | 4782 | /* |
48c5ccae PZ |
4783 | * Ensures that the idle task is using init_mm right before its cpu goes |
4784 | * offline. | |
054b9108 | 4785 | */ |
48c5ccae | 4786 | void idle_task_exit(void) |
1da177e4 | 4787 | { |
48c5ccae | 4788 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4789 | |
48c5ccae | 4790 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4791 | |
a53efe5f | 4792 | if (mm != &init_mm) { |
48c5ccae | 4793 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4794 | finish_arch_post_lock_switch(); |
4795 | } | |
48c5ccae | 4796 | mmdrop(mm); |
1da177e4 LT |
4797 | } |
4798 | ||
4799 | /* | |
5d180232 PZ |
4800 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4801 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4802 | * nr_active count is stable. | |
4803 | * | |
4804 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4805 | */ |
5d180232 | 4806 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4807 | { |
5d180232 PZ |
4808 | long delta = calc_load_fold_active(rq); |
4809 | if (delta) | |
4810 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4811 | } |
4812 | ||
3f1d2a31 PZ |
4813 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4814 | { | |
4815 | } | |
4816 | ||
4817 | static const struct sched_class fake_sched_class = { | |
4818 | .put_prev_task = put_prev_task_fake, | |
4819 | }; | |
4820 | ||
4821 | static struct task_struct fake_task = { | |
4822 | /* | |
4823 | * Avoid pull_{rt,dl}_task() | |
4824 | */ | |
4825 | .prio = MAX_PRIO + 1, | |
4826 | .sched_class = &fake_sched_class, | |
4827 | }; | |
4828 | ||
48f24c4d | 4829 | /* |
48c5ccae PZ |
4830 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4831 | * try_to_wake_up()->select_task_rq(). | |
4832 | * | |
4833 | * Called with rq->lock held even though we'er in stop_machine() and | |
4834 | * there's no concurrency possible, we hold the required locks anyway | |
4835 | * because of lock validation efforts. | |
1da177e4 | 4836 | */ |
48c5ccae | 4837 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4838 | { |
70b97a7f | 4839 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4840 | struct task_struct *next, *stop = rq->stop; |
4841 | int dest_cpu; | |
1da177e4 LT |
4842 | |
4843 | /* | |
48c5ccae PZ |
4844 | * Fudge the rq selection such that the below task selection loop |
4845 | * doesn't get stuck on the currently eligible stop task. | |
4846 | * | |
4847 | * We're currently inside stop_machine() and the rq is either stuck | |
4848 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4849 | * either way we should never end up calling schedule() until we're | |
4850 | * done here. | |
1da177e4 | 4851 | */ |
48c5ccae | 4852 | rq->stop = NULL; |
48f24c4d | 4853 | |
77bd3970 FW |
4854 | /* |
4855 | * put_prev_task() and pick_next_task() sched | |
4856 | * class method both need to have an up-to-date | |
4857 | * value of rq->clock[_task] | |
4858 | */ | |
4859 | update_rq_clock(rq); | |
4860 | ||
dd41f596 | 4861 | for ( ; ; ) { |
48c5ccae PZ |
4862 | /* |
4863 | * There's this thread running, bail when that's the only | |
4864 | * remaining thread. | |
4865 | */ | |
4866 | if (rq->nr_running == 1) | |
dd41f596 | 4867 | break; |
48c5ccae | 4868 | |
3f1d2a31 | 4869 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4870 | BUG_ON(!next); |
79c53799 | 4871 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4872 | |
48c5ccae PZ |
4873 | /* Find suitable destination for @next, with force if needed. */ |
4874 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4875 | raw_spin_unlock(&rq->lock); | |
4876 | ||
4877 | __migrate_task(next, dead_cpu, dest_cpu); | |
4878 | ||
4879 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4880 | } |
dce48a84 | 4881 | |
48c5ccae | 4882 | rq->stop = stop; |
dce48a84 | 4883 | } |
48c5ccae | 4884 | |
1da177e4 LT |
4885 | #endif /* CONFIG_HOTPLUG_CPU */ |
4886 | ||
e692ab53 NP |
4887 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4888 | ||
4889 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4890 | { |
4891 | .procname = "sched_domain", | |
c57baf1e | 4892 | .mode = 0555, |
e0361851 | 4893 | }, |
56992309 | 4894 | {} |
e692ab53 NP |
4895 | }; |
4896 | ||
4897 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4898 | { |
4899 | .procname = "kernel", | |
c57baf1e | 4900 | .mode = 0555, |
e0361851 AD |
4901 | .child = sd_ctl_dir, |
4902 | }, | |
56992309 | 4903 | {} |
e692ab53 NP |
4904 | }; |
4905 | ||
4906 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4907 | { | |
4908 | struct ctl_table *entry = | |
5cf9f062 | 4909 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4910 | |
e692ab53 NP |
4911 | return entry; |
4912 | } | |
4913 | ||
6382bc90 MM |
4914 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4915 | { | |
cd790076 | 4916 | struct ctl_table *entry; |
6382bc90 | 4917 | |
cd790076 MM |
4918 | /* |
4919 | * In the intermediate directories, both the child directory and | |
4920 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4921 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4922 | * static strings and all have proc handlers. |
4923 | */ | |
4924 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4925 | if (entry->child) |
4926 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4927 | if (entry->proc_handler == NULL) |
4928 | kfree(entry->procname); | |
4929 | } | |
6382bc90 MM |
4930 | |
4931 | kfree(*tablep); | |
4932 | *tablep = NULL; | |
4933 | } | |
4934 | ||
201c373e | 4935 | static int min_load_idx = 0; |
fd9b86d3 | 4936 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4937 | |
e692ab53 | 4938 | static void |
e0361851 | 4939 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4940 | const char *procname, void *data, int maxlen, |
201c373e NK |
4941 | umode_t mode, proc_handler *proc_handler, |
4942 | bool load_idx) | |
e692ab53 | 4943 | { |
e692ab53 NP |
4944 | entry->procname = procname; |
4945 | entry->data = data; | |
4946 | entry->maxlen = maxlen; | |
4947 | entry->mode = mode; | |
4948 | entry->proc_handler = proc_handler; | |
201c373e NK |
4949 | |
4950 | if (load_idx) { | |
4951 | entry->extra1 = &min_load_idx; | |
4952 | entry->extra2 = &max_load_idx; | |
4953 | } | |
e692ab53 NP |
4954 | } |
4955 | ||
4956 | static struct ctl_table * | |
4957 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4958 | { | |
37e6bae8 | 4959 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4960 | |
ad1cdc1d MM |
4961 | if (table == NULL) |
4962 | return NULL; | |
4963 | ||
e0361851 | 4964 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4965 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4966 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4967 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4968 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4969 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4970 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4971 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4972 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4973 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4974 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4975 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4976 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4977 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4978 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4979 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4980 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4981 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4982 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4983 | &sd->cache_nice_tries, |
201c373e | 4984 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4985 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4986 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4987 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4988 | &sd->max_newidle_lb_cost, | |
4989 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4990 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4991 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4992 | /* &table[13] is terminator */ |
e692ab53 NP |
4993 | |
4994 | return table; | |
4995 | } | |
4996 | ||
be7002e6 | 4997 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4998 | { |
4999 | struct ctl_table *entry, *table; | |
5000 | struct sched_domain *sd; | |
5001 | int domain_num = 0, i; | |
5002 | char buf[32]; | |
5003 | ||
5004 | for_each_domain(cpu, sd) | |
5005 | domain_num++; | |
5006 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5007 | if (table == NULL) |
5008 | return NULL; | |
e692ab53 NP |
5009 | |
5010 | i = 0; | |
5011 | for_each_domain(cpu, sd) { | |
5012 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5013 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5014 | entry->mode = 0555; |
e692ab53 NP |
5015 | entry->child = sd_alloc_ctl_domain_table(sd); |
5016 | entry++; | |
5017 | i++; | |
5018 | } | |
5019 | return table; | |
5020 | } | |
5021 | ||
5022 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5023 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5024 | { |
6ad4c188 | 5025 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5026 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5027 | char buf[32]; | |
5028 | ||
7378547f MM |
5029 | WARN_ON(sd_ctl_dir[0].child); |
5030 | sd_ctl_dir[0].child = entry; | |
5031 | ||
ad1cdc1d MM |
5032 | if (entry == NULL) |
5033 | return; | |
5034 | ||
6ad4c188 | 5035 | for_each_possible_cpu(i) { |
e692ab53 | 5036 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5037 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5038 | entry->mode = 0555; |
e692ab53 | 5039 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5040 | entry++; |
e692ab53 | 5041 | } |
7378547f MM |
5042 | |
5043 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5044 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5045 | } | |
6382bc90 | 5046 | |
7378547f | 5047 | /* may be called multiple times per register */ |
6382bc90 MM |
5048 | static void unregister_sched_domain_sysctl(void) |
5049 | { | |
7378547f MM |
5050 | if (sd_sysctl_header) |
5051 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5052 | sd_sysctl_header = NULL; |
7378547f MM |
5053 | if (sd_ctl_dir[0].child) |
5054 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5055 | } |
e692ab53 | 5056 | #else |
6382bc90 MM |
5057 | static void register_sched_domain_sysctl(void) |
5058 | { | |
5059 | } | |
5060 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5061 | { |
5062 | } | |
5063 | #endif | |
5064 | ||
1f11eb6a GH |
5065 | static void set_rq_online(struct rq *rq) |
5066 | { | |
5067 | if (!rq->online) { | |
5068 | const struct sched_class *class; | |
5069 | ||
c6c4927b | 5070 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5071 | rq->online = 1; |
5072 | ||
5073 | for_each_class(class) { | |
5074 | if (class->rq_online) | |
5075 | class->rq_online(rq); | |
5076 | } | |
5077 | } | |
5078 | } | |
5079 | ||
5080 | static void set_rq_offline(struct rq *rq) | |
5081 | { | |
5082 | if (rq->online) { | |
5083 | const struct sched_class *class; | |
5084 | ||
5085 | for_each_class(class) { | |
5086 | if (class->rq_offline) | |
5087 | class->rq_offline(rq); | |
5088 | } | |
5089 | ||
c6c4927b | 5090 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5091 | rq->online = 0; |
5092 | } | |
5093 | } | |
5094 | ||
1da177e4 LT |
5095 | /* |
5096 | * migration_call - callback that gets triggered when a CPU is added. | |
5097 | * Here we can start up the necessary migration thread for the new CPU. | |
5098 | */ | |
0db0628d | 5099 | static int |
48f24c4d | 5100 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5101 | { |
48f24c4d | 5102 | int cpu = (long)hcpu; |
1da177e4 | 5103 | unsigned long flags; |
969c7921 | 5104 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5105 | |
48c5ccae | 5106 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5107 | |
1da177e4 | 5108 | case CPU_UP_PREPARE: |
a468d389 | 5109 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5110 | break; |
48f24c4d | 5111 | |
1da177e4 | 5112 | case CPU_ONLINE: |
1f94ef59 | 5113 | /* Update our root-domain */ |
05fa785c | 5114 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5115 | if (rq->rd) { |
c6c4927b | 5116 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5117 | |
5118 | set_rq_online(rq); | |
1f94ef59 | 5119 | } |
05fa785c | 5120 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5121 | break; |
48f24c4d | 5122 | |
1da177e4 | 5123 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5124 | case CPU_DYING: |
317f3941 | 5125 | sched_ttwu_pending(); |
57d885fe | 5126 | /* Update our root-domain */ |
05fa785c | 5127 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5128 | if (rq->rd) { |
c6c4927b | 5129 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5130 | set_rq_offline(rq); |
57d885fe | 5131 | } |
48c5ccae PZ |
5132 | migrate_tasks(cpu); |
5133 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5134 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5135 | break; |
48c5ccae | 5136 | |
5d180232 | 5137 | case CPU_DEAD: |
f319da0c | 5138 | calc_load_migrate(rq); |
57d885fe | 5139 | break; |
1da177e4 LT |
5140 | #endif |
5141 | } | |
49c022e6 PZ |
5142 | |
5143 | update_max_interval(); | |
5144 | ||
1da177e4 LT |
5145 | return NOTIFY_OK; |
5146 | } | |
5147 | ||
f38b0820 PM |
5148 | /* |
5149 | * Register at high priority so that task migration (migrate_all_tasks) | |
5150 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5151 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5152 | */ |
0db0628d | 5153 | static struct notifier_block migration_notifier = { |
1da177e4 | 5154 | .notifier_call = migration_call, |
50a323b7 | 5155 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5156 | }; |
5157 | ||
a803f026 CM |
5158 | static void __cpuinit set_cpu_rq_start_time(void) |
5159 | { | |
5160 | int cpu = smp_processor_id(); | |
5161 | struct rq *rq = cpu_rq(cpu); | |
5162 | rq->age_stamp = sched_clock_cpu(cpu); | |
5163 | } | |
5164 | ||
0db0628d | 5165 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5166 | unsigned long action, void *hcpu) |
5167 | { | |
5168 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5169 | case CPU_STARTING: |
5170 | set_cpu_rq_start_time(); | |
5171 | return NOTIFY_OK; | |
3a101d05 TH |
5172 | case CPU_DOWN_FAILED: |
5173 | set_cpu_active((long)hcpu, true); | |
5174 | return NOTIFY_OK; | |
5175 | default: | |
5176 | return NOTIFY_DONE; | |
5177 | } | |
5178 | } | |
5179 | ||
0db0628d | 5180 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5181 | unsigned long action, void *hcpu) |
5182 | { | |
de212f18 PZ |
5183 | unsigned long flags; |
5184 | long cpu = (long)hcpu; | |
5185 | ||
3a101d05 TH |
5186 | switch (action & ~CPU_TASKS_FROZEN) { |
5187 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5188 | set_cpu_active(cpu, false); |
5189 | ||
5190 | /* explicitly allow suspend */ | |
5191 | if (!(action & CPU_TASKS_FROZEN)) { | |
5192 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5193 | bool overflow; | |
5194 | int cpus; | |
5195 | ||
5196 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5197 | cpus = dl_bw_cpus(cpu); | |
5198 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5199 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5200 | ||
5201 | if (overflow) | |
5202 | return notifier_from_errno(-EBUSY); | |
5203 | } | |
3a101d05 | 5204 | return NOTIFY_OK; |
3a101d05 | 5205 | } |
de212f18 PZ |
5206 | |
5207 | return NOTIFY_DONE; | |
3a101d05 TH |
5208 | } |
5209 | ||
7babe8db | 5210 | static int __init migration_init(void) |
1da177e4 LT |
5211 | { |
5212 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5213 | int err; |
48f24c4d | 5214 | |
3a101d05 | 5215 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5216 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5217 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5218 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5219 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5220 | |
3a101d05 TH |
5221 | /* Register cpu active notifiers */ |
5222 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5223 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5224 | ||
a004cd42 | 5225 | return 0; |
1da177e4 | 5226 | } |
7babe8db | 5227 | early_initcall(migration_init); |
1da177e4 LT |
5228 | #endif |
5229 | ||
5230 | #ifdef CONFIG_SMP | |
476f3534 | 5231 | |
4cb98839 PZ |
5232 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5233 | ||
3e9830dc | 5234 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5235 | |
d039ac60 | 5236 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5237 | |
d039ac60 | 5238 | static int __init sched_debug_setup(char *str) |
f6630114 | 5239 | { |
d039ac60 | 5240 | sched_debug_enabled = 1; |
f6630114 MT |
5241 | |
5242 | return 0; | |
5243 | } | |
d039ac60 PZ |
5244 | early_param("sched_debug", sched_debug_setup); |
5245 | ||
5246 | static inline bool sched_debug(void) | |
5247 | { | |
5248 | return sched_debug_enabled; | |
5249 | } | |
f6630114 | 5250 | |
7c16ec58 | 5251 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5252 | struct cpumask *groupmask) |
1da177e4 | 5253 | { |
4dcf6aff | 5254 | struct sched_group *group = sd->groups; |
434d53b0 | 5255 | char str[256]; |
1da177e4 | 5256 | |
968ea6d8 | 5257 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5258 | cpumask_clear(groupmask); |
4dcf6aff IM |
5259 | |
5260 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5261 | ||
5262 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5263 | printk("does not load-balance\n"); |
4dcf6aff | 5264 | if (sd->parent) |
3df0fc5b PZ |
5265 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5266 | " has parent"); | |
4dcf6aff | 5267 | return -1; |
41c7ce9a NP |
5268 | } |
5269 | ||
3df0fc5b | 5270 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5271 | |
758b2cdc | 5272 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5273 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5274 | "CPU%d\n", cpu); | |
4dcf6aff | 5275 | } |
758b2cdc | 5276 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5277 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5278 | " CPU%d\n", cpu); | |
4dcf6aff | 5279 | } |
1da177e4 | 5280 | |
4dcf6aff | 5281 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5282 | do { |
4dcf6aff | 5283 | if (!group) { |
3df0fc5b PZ |
5284 | printk("\n"); |
5285 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5286 | break; |
5287 | } | |
5288 | ||
c3decf0d | 5289 | /* |
63b2ca30 NP |
5290 | * Even though we initialize ->capacity to something semi-sane, |
5291 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5292 | * domain iteration is still funny without causing /0 traps. |
5293 | */ | |
63b2ca30 | 5294 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5295 | printk(KERN_CONT "\n"); |
63b2ca30 | 5296 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5297 | break; |
5298 | } | |
1da177e4 | 5299 | |
758b2cdc | 5300 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5301 | printk(KERN_CONT "\n"); |
5302 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5303 | break; |
5304 | } | |
1da177e4 | 5305 | |
cb83b629 PZ |
5306 | if (!(sd->flags & SD_OVERLAP) && |
5307 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5308 | printk(KERN_CONT "\n"); |
5309 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5310 | break; |
5311 | } | |
1da177e4 | 5312 | |
758b2cdc | 5313 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5314 | |
968ea6d8 | 5315 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5316 | |
3df0fc5b | 5317 | printk(KERN_CONT " %s", str); |
ca8ce3d0 | 5318 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5319 | printk(KERN_CONT " (cpu_capacity = %d)", |
5320 | group->sgc->capacity); | |
381512cf | 5321 | } |
1da177e4 | 5322 | |
4dcf6aff IM |
5323 | group = group->next; |
5324 | } while (group != sd->groups); | |
3df0fc5b | 5325 | printk(KERN_CONT "\n"); |
1da177e4 | 5326 | |
758b2cdc | 5327 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5328 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5329 | |
758b2cdc RR |
5330 | if (sd->parent && |
5331 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5332 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5333 | "of domain->span\n"); | |
4dcf6aff IM |
5334 | return 0; |
5335 | } | |
1da177e4 | 5336 | |
4dcf6aff IM |
5337 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5338 | { | |
5339 | int level = 0; | |
1da177e4 | 5340 | |
d039ac60 | 5341 | if (!sched_debug_enabled) |
f6630114 MT |
5342 | return; |
5343 | ||
4dcf6aff IM |
5344 | if (!sd) { |
5345 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5346 | return; | |
5347 | } | |
1da177e4 | 5348 | |
4dcf6aff IM |
5349 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5350 | ||
5351 | for (;;) { | |
4cb98839 | 5352 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5353 | break; |
1da177e4 LT |
5354 | level++; |
5355 | sd = sd->parent; | |
33859f7f | 5356 | if (!sd) |
4dcf6aff IM |
5357 | break; |
5358 | } | |
1da177e4 | 5359 | } |
6d6bc0ad | 5360 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5361 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5362 | static inline bool sched_debug(void) |
5363 | { | |
5364 | return false; | |
5365 | } | |
6d6bc0ad | 5366 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5367 | |
1a20ff27 | 5368 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5369 | { |
758b2cdc | 5370 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5371 | return 1; |
5372 | ||
5373 | /* Following flags need at least 2 groups */ | |
5374 | if (sd->flags & (SD_LOAD_BALANCE | | |
5375 | SD_BALANCE_NEWIDLE | | |
5376 | SD_BALANCE_FORK | | |
89c4710e | 5377 | SD_BALANCE_EXEC | |
5d4dfddd | 5378 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5379 | SD_SHARE_PKG_RESOURCES | |
5380 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5381 | if (sd->groups != sd->groups->next) |
5382 | return 0; | |
5383 | } | |
5384 | ||
5385 | /* Following flags don't use groups */ | |
c88d5910 | 5386 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5387 | return 0; |
5388 | ||
5389 | return 1; | |
5390 | } | |
5391 | ||
48f24c4d IM |
5392 | static int |
5393 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5394 | { |
5395 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5396 | ||
5397 | if (sd_degenerate(parent)) | |
5398 | return 1; | |
5399 | ||
758b2cdc | 5400 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5401 | return 0; |
5402 | ||
245af2c7 SS |
5403 | /* Flags needing groups don't count if only 1 group in parent */ |
5404 | if (parent->groups == parent->groups->next) { | |
5405 | pflags &= ~(SD_LOAD_BALANCE | | |
5406 | SD_BALANCE_NEWIDLE | | |
5407 | SD_BALANCE_FORK | | |
89c4710e | 5408 | SD_BALANCE_EXEC | |
5d4dfddd | 5409 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5410 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5411 | SD_PREFER_SIBLING | |
5412 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5413 | if (nr_node_ids == 1) |
5414 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5415 | } |
5416 | if (~cflags & pflags) | |
5417 | return 0; | |
5418 | ||
5419 | return 1; | |
5420 | } | |
5421 | ||
dce840a0 | 5422 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5423 | { |
dce840a0 | 5424 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5425 | |
68e74568 | 5426 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5427 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5428 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5429 | free_cpumask_var(rd->rto_mask); |
5430 | free_cpumask_var(rd->online); | |
5431 | free_cpumask_var(rd->span); | |
5432 | kfree(rd); | |
5433 | } | |
5434 | ||
57d885fe GH |
5435 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5436 | { | |
a0490fa3 | 5437 | struct root_domain *old_rd = NULL; |
57d885fe | 5438 | unsigned long flags; |
57d885fe | 5439 | |
05fa785c | 5440 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5441 | |
5442 | if (rq->rd) { | |
a0490fa3 | 5443 | old_rd = rq->rd; |
57d885fe | 5444 | |
c6c4927b | 5445 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5446 | set_rq_offline(rq); |
57d885fe | 5447 | |
c6c4927b | 5448 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5449 | |
a0490fa3 | 5450 | /* |
0515973f | 5451 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5452 | * set old_rd to NULL to skip the freeing later |
5453 | * in this function: | |
5454 | */ | |
5455 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5456 | old_rd = NULL; | |
57d885fe GH |
5457 | } |
5458 | ||
5459 | atomic_inc(&rd->refcount); | |
5460 | rq->rd = rd; | |
5461 | ||
c6c4927b | 5462 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5463 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5464 | set_rq_online(rq); |
57d885fe | 5465 | |
05fa785c | 5466 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5467 | |
5468 | if (old_rd) | |
dce840a0 | 5469 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5470 | } |
5471 | ||
68c38fc3 | 5472 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5473 | { |
5474 | memset(rd, 0, sizeof(*rd)); | |
5475 | ||
68c38fc3 | 5476 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5477 | goto out; |
68c38fc3 | 5478 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5479 | goto free_span; |
1baca4ce | 5480 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5481 | goto free_online; |
1baca4ce JL |
5482 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5483 | goto free_dlo_mask; | |
6e0534f2 | 5484 | |
332ac17e | 5485 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5486 | if (cpudl_init(&rd->cpudl) != 0) |
5487 | goto free_dlo_mask; | |
332ac17e | 5488 | |
68c38fc3 | 5489 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5490 | goto free_rto_mask; |
c6c4927b | 5491 | return 0; |
6e0534f2 | 5492 | |
68e74568 RR |
5493 | free_rto_mask: |
5494 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5495 | free_dlo_mask: |
5496 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5497 | free_online: |
5498 | free_cpumask_var(rd->online); | |
5499 | free_span: | |
5500 | free_cpumask_var(rd->span); | |
0c910d28 | 5501 | out: |
c6c4927b | 5502 | return -ENOMEM; |
57d885fe GH |
5503 | } |
5504 | ||
029632fb PZ |
5505 | /* |
5506 | * By default the system creates a single root-domain with all cpus as | |
5507 | * members (mimicking the global state we have today). | |
5508 | */ | |
5509 | struct root_domain def_root_domain; | |
5510 | ||
57d885fe GH |
5511 | static void init_defrootdomain(void) |
5512 | { | |
68c38fc3 | 5513 | init_rootdomain(&def_root_domain); |
c6c4927b | 5514 | |
57d885fe GH |
5515 | atomic_set(&def_root_domain.refcount, 1); |
5516 | } | |
5517 | ||
dc938520 | 5518 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5519 | { |
5520 | struct root_domain *rd; | |
5521 | ||
5522 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5523 | if (!rd) | |
5524 | return NULL; | |
5525 | ||
68c38fc3 | 5526 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5527 | kfree(rd); |
5528 | return NULL; | |
5529 | } | |
57d885fe GH |
5530 | |
5531 | return rd; | |
5532 | } | |
5533 | ||
63b2ca30 | 5534 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5535 | { |
5536 | struct sched_group *tmp, *first; | |
5537 | ||
5538 | if (!sg) | |
5539 | return; | |
5540 | ||
5541 | first = sg; | |
5542 | do { | |
5543 | tmp = sg->next; | |
5544 | ||
63b2ca30 NP |
5545 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5546 | kfree(sg->sgc); | |
e3589f6c PZ |
5547 | |
5548 | kfree(sg); | |
5549 | sg = tmp; | |
5550 | } while (sg != first); | |
5551 | } | |
5552 | ||
dce840a0 PZ |
5553 | static void free_sched_domain(struct rcu_head *rcu) |
5554 | { | |
5555 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5556 | |
5557 | /* | |
5558 | * If its an overlapping domain it has private groups, iterate and | |
5559 | * nuke them all. | |
5560 | */ | |
5561 | if (sd->flags & SD_OVERLAP) { | |
5562 | free_sched_groups(sd->groups, 1); | |
5563 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5564 | kfree(sd->groups->sgc); |
dce840a0 | 5565 | kfree(sd->groups); |
9c3f75cb | 5566 | } |
dce840a0 PZ |
5567 | kfree(sd); |
5568 | } | |
5569 | ||
5570 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5571 | { | |
5572 | call_rcu(&sd->rcu, free_sched_domain); | |
5573 | } | |
5574 | ||
5575 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5576 | { | |
5577 | for (; sd; sd = sd->parent) | |
5578 | destroy_sched_domain(sd, cpu); | |
5579 | } | |
5580 | ||
518cd623 PZ |
5581 | /* |
5582 | * Keep a special pointer to the highest sched_domain that has | |
5583 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5584 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5585 | * | |
5586 | * Also keep a unique ID per domain (we use the first cpu number in | |
5587 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5588 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5589 | */ |
5590 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5591 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5592 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5593 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5594 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5595 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5596 | |
5597 | static void update_top_cache_domain(int cpu) | |
5598 | { | |
5599 | struct sched_domain *sd; | |
5d4cf996 | 5600 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5601 | int id = cpu; |
7d9ffa89 | 5602 | int size = 1; |
518cd623 PZ |
5603 | |
5604 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5605 | if (sd) { |
518cd623 | 5606 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5607 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5608 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5609 | } |
5d4cf996 | 5610 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5611 | |
5612 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5613 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5614 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5615 | |
5616 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5617 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5618 | |
5619 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5620 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5621 | } |
5622 | ||
1da177e4 | 5623 | /* |
0eab9146 | 5624 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5625 | * hold the hotplug lock. |
5626 | */ | |
0eab9146 IM |
5627 | static void |
5628 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5629 | { |
70b97a7f | 5630 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5631 | struct sched_domain *tmp; |
5632 | ||
5633 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5634 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5635 | struct sched_domain *parent = tmp->parent; |
5636 | if (!parent) | |
5637 | break; | |
f29c9b1c | 5638 | |
1a848870 | 5639 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5640 | tmp->parent = parent->parent; |
1a848870 SS |
5641 | if (parent->parent) |
5642 | parent->parent->child = tmp; | |
10866e62 PZ |
5643 | /* |
5644 | * Transfer SD_PREFER_SIBLING down in case of a | |
5645 | * degenerate parent; the spans match for this | |
5646 | * so the property transfers. | |
5647 | */ | |
5648 | if (parent->flags & SD_PREFER_SIBLING) | |
5649 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5650 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5651 | } else |
5652 | tmp = tmp->parent; | |
245af2c7 SS |
5653 | } |
5654 | ||
1a848870 | 5655 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5656 | tmp = sd; |
245af2c7 | 5657 | sd = sd->parent; |
dce840a0 | 5658 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5659 | if (sd) |
5660 | sd->child = NULL; | |
5661 | } | |
1da177e4 | 5662 | |
4cb98839 | 5663 | sched_domain_debug(sd, cpu); |
1da177e4 | 5664 | |
57d885fe | 5665 | rq_attach_root(rq, rd); |
dce840a0 | 5666 | tmp = rq->sd; |
674311d5 | 5667 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5668 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5669 | |
5670 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5671 | } |
5672 | ||
5673 | /* cpus with isolated domains */ | |
dcc30a35 | 5674 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5675 | |
5676 | /* Setup the mask of cpus configured for isolated domains */ | |
5677 | static int __init isolated_cpu_setup(char *str) | |
5678 | { | |
bdddd296 | 5679 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5680 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5681 | return 1; |
5682 | } | |
5683 | ||
8927f494 | 5684 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5685 | |
49a02c51 | 5686 | struct s_data { |
21d42ccf | 5687 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5688 | struct root_domain *rd; |
5689 | }; | |
5690 | ||
2109b99e | 5691 | enum s_alloc { |
2109b99e | 5692 | sa_rootdomain, |
21d42ccf | 5693 | sa_sd, |
dce840a0 | 5694 | sa_sd_storage, |
2109b99e AH |
5695 | sa_none, |
5696 | }; | |
5697 | ||
c1174876 PZ |
5698 | /* |
5699 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5700 | * domain traversal. | |
5701 | * | |
5702 | * Asymmetric node setups can result in situations where the domain tree is of | |
5703 | * unequal depth, make sure to skip domains that already cover the entire | |
5704 | * range. | |
5705 | * | |
5706 | * In that case build_sched_domains() will have terminated the iteration early | |
5707 | * and our sibling sd spans will be empty. Domains should always include the | |
5708 | * cpu they're built on, so check that. | |
5709 | * | |
5710 | */ | |
5711 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5712 | { | |
5713 | const struct cpumask *span = sched_domain_span(sd); | |
5714 | struct sd_data *sdd = sd->private; | |
5715 | struct sched_domain *sibling; | |
5716 | int i; | |
5717 | ||
5718 | for_each_cpu(i, span) { | |
5719 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5720 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5721 | continue; | |
5722 | ||
5723 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5724 | } | |
5725 | } | |
5726 | ||
5727 | /* | |
5728 | * Return the canonical balance cpu for this group, this is the first cpu | |
5729 | * of this group that's also in the iteration mask. | |
5730 | */ | |
5731 | int group_balance_cpu(struct sched_group *sg) | |
5732 | { | |
5733 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5734 | } | |
5735 | ||
e3589f6c PZ |
5736 | static int |
5737 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5738 | { | |
5739 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5740 | const struct cpumask *span = sched_domain_span(sd); | |
5741 | struct cpumask *covered = sched_domains_tmpmask; | |
5742 | struct sd_data *sdd = sd->private; | |
5743 | struct sched_domain *child; | |
5744 | int i; | |
5745 | ||
5746 | cpumask_clear(covered); | |
5747 | ||
5748 | for_each_cpu(i, span) { | |
5749 | struct cpumask *sg_span; | |
5750 | ||
5751 | if (cpumask_test_cpu(i, covered)) | |
5752 | continue; | |
5753 | ||
c1174876 PZ |
5754 | child = *per_cpu_ptr(sdd->sd, i); |
5755 | ||
5756 | /* See the comment near build_group_mask(). */ | |
5757 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5758 | continue; | |
5759 | ||
e3589f6c | 5760 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5761 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5762 | |
5763 | if (!sg) | |
5764 | goto fail; | |
5765 | ||
5766 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5767 | if (child->child) { |
5768 | child = child->child; | |
5769 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5770 | } else | |
5771 | cpumask_set_cpu(i, sg_span); | |
5772 | ||
5773 | cpumask_or(covered, covered, sg_span); | |
5774 | ||
63b2ca30 NP |
5775 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5776 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5777 | build_group_mask(sd, sg); |
5778 | ||
c3decf0d | 5779 | /* |
63b2ca30 | 5780 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5781 | * domains and no possible iteration will get us here, we won't |
5782 | * die on a /0 trap. | |
5783 | */ | |
ca8ce3d0 | 5784 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
63b2ca30 | 5785 | sg->sgc->capacity_orig = sg->sgc->capacity; |
e3589f6c | 5786 | |
c1174876 PZ |
5787 | /* |
5788 | * Make sure the first group of this domain contains the | |
5789 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5790 | * breaks. See update_sg_lb_stats(). | |
5791 | */ | |
74a5ce20 | 5792 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5793 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5794 | groups = sg; |
5795 | ||
5796 | if (!first) | |
5797 | first = sg; | |
5798 | if (last) | |
5799 | last->next = sg; | |
5800 | last = sg; | |
5801 | last->next = first; | |
5802 | } | |
5803 | sd->groups = groups; | |
5804 | ||
5805 | return 0; | |
5806 | ||
5807 | fail: | |
5808 | free_sched_groups(first, 0); | |
5809 | ||
5810 | return -ENOMEM; | |
5811 | } | |
5812 | ||
dce840a0 | 5813 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5814 | { |
dce840a0 PZ |
5815 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5816 | struct sched_domain *child = sd->child; | |
1da177e4 | 5817 | |
dce840a0 PZ |
5818 | if (child) |
5819 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5820 | |
9c3f75cb | 5821 | if (sg) { |
dce840a0 | 5822 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5823 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5824 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5825 | } |
dce840a0 PZ |
5826 | |
5827 | return cpu; | |
1e9f28fa | 5828 | } |
1e9f28fa | 5829 | |
01a08546 | 5830 | /* |
dce840a0 PZ |
5831 | * build_sched_groups will build a circular linked list of the groups |
5832 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5833 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5834 | * |
5835 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5836 | */ |
e3589f6c PZ |
5837 | static int |
5838 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5839 | { |
dce840a0 PZ |
5840 | struct sched_group *first = NULL, *last = NULL; |
5841 | struct sd_data *sdd = sd->private; | |
5842 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5843 | struct cpumask *covered; |
dce840a0 | 5844 | int i; |
9c1cfda2 | 5845 | |
e3589f6c PZ |
5846 | get_group(cpu, sdd, &sd->groups); |
5847 | atomic_inc(&sd->groups->ref); | |
5848 | ||
0936629f | 5849 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5850 | return 0; |
5851 | ||
f96225fd PZ |
5852 | lockdep_assert_held(&sched_domains_mutex); |
5853 | covered = sched_domains_tmpmask; | |
5854 | ||
dce840a0 | 5855 | cpumask_clear(covered); |
6711cab4 | 5856 | |
dce840a0 PZ |
5857 | for_each_cpu(i, span) { |
5858 | struct sched_group *sg; | |
cd08e923 | 5859 | int group, j; |
6711cab4 | 5860 | |
dce840a0 PZ |
5861 | if (cpumask_test_cpu(i, covered)) |
5862 | continue; | |
6711cab4 | 5863 | |
cd08e923 | 5864 | group = get_group(i, sdd, &sg); |
c1174876 | 5865 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5866 | |
dce840a0 PZ |
5867 | for_each_cpu(j, span) { |
5868 | if (get_group(j, sdd, NULL) != group) | |
5869 | continue; | |
0601a88d | 5870 | |
dce840a0 PZ |
5871 | cpumask_set_cpu(j, covered); |
5872 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5873 | } | |
0601a88d | 5874 | |
dce840a0 PZ |
5875 | if (!first) |
5876 | first = sg; | |
5877 | if (last) | |
5878 | last->next = sg; | |
5879 | last = sg; | |
5880 | } | |
5881 | last->next = first; | |
e3589f6c PZ |
5882 | |
5883 | return 0; | |
0601a88d | 5884 | } |
51888ca2 | 5885 | |
89c4710e | 5886 | /* |
63b2ca30 | 5887 | * Initialize sched groups cpu_capacity. |
89c4710e | 5888 | * |
63b2ca30 | 5889 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 5890 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
5891 | * Typically cpu_capacity for all the groups in a sched domain will be same |
5892 | * unless there are asymmetries in the topology. If there are asymmetries, | |
5893 | * group having more cpu_capacity will pickup more load compared to the | |
5894 | * group having less cpu_capacity. | |
89c4710e | 5895 | */ |
63b2ca30 | 5896 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 5897 | { |
e3589f6c | 5898 | struct sched_group *sg = sd->groups; |
89c4710e | 5899 | |
94c95ba6 | 5900 | WARN_ON(!sg); |
e3589f6c PZ |
5901 | |
5902 | do { | |
5903 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5904 | sg = sg->next; | |
5905 | } while (sg != sd->groups); | |
89c4710e | 5906 | |
c1174876 | 5907 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5908 | return; |
aae6d3dd | 5909 | |
63b2ca30 NP |
5910 | update_group_capacity(sd, cpu); |
5911 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
5912 | } |
5913 | ||
7c16ec58 MT |
5914 | /* |
5915 | * Initializers for schedule domains | |
5916 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5917 | */ | |
5918 | ||
1d3504fc | 5919 | static int default_relax_domain_level = -1; |
60495e77 | 5920 | int sched_domain_level_max; |
1d3504fc HS |
5921 | |
5922 | static int __init setup_relax_domain_level(char *str) | |
5923 | { | |
a841f8ce DS |
5924 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5925 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5926 | |
1d3504fc HS |
5927 | return 1; |
5928 | } | |
5929 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5930 | ||
5931 | static void set_domain_attribute(struct sched_domain *sd, | |
5932 | struct sched_domain_attr *attr) | |
5933 | { | |
5934 | int request; | |
5935 | ||
5936 | if (!attr || attr->relax_domain_level < 0) { | |
5937 | if (default_relax_domain_level < 0) | |
5938 | return; | |
5939 | else | |
5940 | request = default_relax_domain_level; | |
5941 | } else | |
5942 | request = attr->relax_domain_level; | |
5943 | if (request < sd->level) { | |
5944 | /* turn off idle balance on this domain */ | |
c88d5910 | 5945 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5946 | } else { |
5947 | /* turn on idle balance on this domain */ | |
c88d5910 | 5948 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5949 | } |
5950 | } | |
5951 | ||
54ab4ff4 PZ |
5952 | static void __sdt_free(const struct cpumask *cpu_map); |
5953 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5954 | ||
2109b99e AH |
5955 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5956 | const struct cpumask *cpu_map) | |
5957 | { | |
5958 | switch (what) { | |
2109b99e | 5959 | case sa_rootdomain: |
822ff793 PZ |
5960 | if (!atomic_read(&d->rd->refcount)) |
5961 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5962 | case sa_sd: |
5963 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5964 | case sa_sd_storage: |
54ab4ff4 | 5965 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5966 | case sa_none: |
5967 | break; | |
5968 | } | |
5969 | } | |
3404c8d9 | 5970 | |
2109b99e AH |
5971 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5972 | const struct cpumask *cpu_map) | |
5973 | { | |
dce840a0 PZ |
5974 | memset(d, 0, sizeof(*d)); |
5975 | ||
54ab4ff4 PZ |
5976 | if (__sdt_alloc(cpu_map)) |
5977 | return sa_sd_storage; | |
dce840a0 PZ |
5978 | d->sd = alloc_percpu(struct sched_domain *); |
5979 | if (!d->sd) | |
5980 | return sa_sd_storage; | |
2109b99e | 5981 | d->rd = alloc_rootdomain(); |
dce840a0 | 5982 | if (!d->rd) |
21d42ccf | 5983 | return sa_sd; |
2109b99e AH |
5984 | return sa_rootdomain; |
5985 | } | |
57d885fe | 5986 | |
dce840a0 PZ |
5987 | /* |
5988 | * NULL the sd_data elements we've used to build the sched_domain and | |
5989 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5990 | * will not free the data we're using. | |
5991 | */ | |
5992 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5993 | { | |
5994 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5995 | |
5996 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5997 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5998 | ||
e3589f6c | 5999 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6000 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 6001 | |
63b2ca30 NP |
6002 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
6003 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
6004 | } |
6005 | ||
cb83b629 | 6006 | #ifdef CONFIG_NUMA |
cb83b629 | 6007 | static int sched_domains_numa_levels; |
cb83b629 PZ |
6008 | static int *sched_domains_numa_distance; |
6009 | static struct cpumask ***sched_domains_numa_masks; | |
6010 | static int sched_domains_curr_level; | |
143e1e28 | 6011 | #endif |
cb83b629 | 6012 | |
143e1e28 VG |
6013 | /* |
6014 | * SD_flags allowed in topology descriptions. | |
6015 | * | |
5d4dfddd | 6016 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
6017 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
6018 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 6019 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
6020 | * |
6021 | * Odd one out: | |
6022 | * SD_ASYM_PACKING - describes SMT quirks | |
6023 | */ | |
6024 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 6025 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
6026 | SD_SHARE_PKG_RESOURCES | \ |
6027 | SD_NUMA | \ | |
d77b3ed5 VG |
6028 | SD_ASYM_PACKING | \ |
6029 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
6030 | |
6031 | static struct sched_domain * | |
143e1e28 | 6032 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
6033 | { |
6034 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
6035 | int sd_weight, sd_flags = 0; |
6036 | ||
6037 | #ifdef CONFIG_NUMA | |
6038 | /* | |
6039 | * Ugly hack to pass state to sd_numa_mask()... | |
6040 | */ | |
6041 | sched_domains_curr_level = tl->numa_level; | |
6042 | #endif | |
6043 | ||
6044 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
6045 | ||
6046 | if (tl->sd_flags) | |
6047 | sd_flags = (*tl->sd_flags)(); | |
6048 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
6049 | "wrong sd_flags in topology description\n")) | |
6050 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
6051 | |
6052 | *sd = (struct sched_domain){ | |
6053 | .min_interval = sd_weight, | |
6054 | .max_interval = 2*sd_weight, | |
6055 | .busy_factor = 32, | |
870a0bb5 | 6056 | .imbalance_pct = 125, |
143e1e28 VG |
6057 | |
6058 | .cache_nice_tries = 0, | |
6059 | .busy_idx = 0, | |
6060 | .idle_idx = 0, | |
cb83b629 PZ |
6061 | .newidle_idx = 0, |
6062 | .wake_idx = 0, | |
6063 | .forkexec_idx = 0, | |
6064 | ||
6065 | .flags = 1*SD_LOAD_BALANCE | |
6066 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6067 | | 1*SD_BALANCE_EXEC |
6068 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6069 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6070 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 6071 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 6072 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6073 | | 0*SD_SERIALIZE |
cb83b629 | 6074 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6075 | | 0*SD_NUMA |
6076 | | sd_flags | |
cb83b629 | 6077 | , |
143e1e28 | 6078 | |
cb83b629 PZ |
6079 | .last_balance = jiffies, |
6080 | .balance_interval = sd_weight, | |
143e1e28 | 6081 | .smt_gain = 0, |
2b4cfe64 JL |
6082 | .max_newidle_lb_cost = 0, |
6083 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6084 | #ifdef CONFIG_SCHED_DEBUG |
6085 | .name = tl->name, | |
6086 | #endif | |
cb83b629 | 6087 | }; |
cb83b629 PZ |
6088 | |
6089 | /* | |
143e1e28 | 6090 | * Convert topological properties into behaviour. |
cb83b629 | 6091 | */ |
143e1e28 | 6092 | |
5d4dfddd | 6093 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
143e1e28 VG |
6094 | sd->imbalance_pct = 110; |
6095 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6096 | |
6097 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6098 | sd->imbalance_pct = 117; | |
6099 | sd->cache_nice_tries = 1; | |
6100 | sd->busy_idx = 2; | |
6101 | ||
6102 | #ifdef CONFIG_NUMA | |
6103 | } else if (sd->flags & SD_NUMA) { | |
6104 | sd->cache_nice_tries = 2; | |
6105 | sd->busy_idx = 3; | |
6106 | sd->idle_idx = 2; | |
6107 | ||
6108 | sd->flags |= SD_SERIALIZE; | |
6109 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6110 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6111 | SD_BALANCE_FORK | | |
6112 | SD_WAKE_AFFINE); | |
6113 | } | |
6114 | ||
6115 | #endif | |
6116 | } else { | |
6117 | sd->flags |= SD_PREFER_SIBLING; | |
6118 | sd->cache_nice_tries = 1; | |
6119 | sd->busy_idx = 2; | |
6120 | sd->idle_idx = 1; | |
6121 | } | |
6122 | ||
6123 | sd->private = &tl->data; | |
cb83b629 PZ |
6124 | |
6125 | return sd; | |
6126 | } | |
6127 | ||
143e1e28 VG |
6128 | /* |
6129 | * Topology list, bottom-up. | |
6130 | */ | |
6131 | static struct sched_domain_topology_level default_topology[] = { | |
6132 | #ifdef CONFIG_SCHED_SMT | |
6133 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6134 | #endif | |
6135 | #ifdef CONFIG_SCHED_MC | |
6136 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6137 | #endif |
6138 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6139 | { NULL, }, | |
6140 | }; | |
6141 | ||
6142 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6143 | ||
6144 | #define for_each_sd_topology(tl) \ | |
6145 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6146 | ||
6147 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6148 | { | |
6149 | sched_domain_topology = tl; | |
6150 | } | |
6151 | ||
6152 | #ifdef CONFIG_NUMA | |
6153 | ||
cb83b629 PZ |
6154 | static const struct cpumask *sd_numa_mask(int cpu) |
6155 | { | |
6156 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6157 | } | |
6158 | ||
d039ac60 PZ |
6159 | static void sched_numa_warn(const char *str) |
6160 | { | |
6161 | static int done = false; | |
6162 | int i,j; | |
6163 | ||
6164 | if (done) | |
6165 | return; | |
6166 | ||
6167 | done = true; | |
6168 | ||
6169 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6170 | ||
6171 | for (i = 0; i < nr_node_ids; i++) { | |
6172 | printk(KERN_WARNING " "); | |
6173 | for (j = 0; j < nr_node_ids; j++) | |
6174 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6175 | printk(KERN_CONT "\n"); | |
6176 | } | |
6177 | printk(KERN_WARNING "\n"); | |
6178 | } | |
6179 | ||
6180 | static bool find_numa_distance(int distance) | |
6181 | { | |
6182 | int i; | |
6183 | ||
6184 | if (distance == node_distance(0, 0)) | |
6185 | return true; | |
6186 | ||
6187 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6188 | if (sched_domains_numa_distance[i] == distance) | |
6189 | return true; | |
6190 | } | |
6191 | ||
6192 | return false; | |
6193 | } | |
6194 | ||
cb83b629 PZ |
6195 | static void sched_init_numa(void) |
6196 | { | |
6197 | int next_distance, curr_distance = node_distance(0, 0); | |
6198 | struct sched_domain_topology_level *tl; | |
6199 | int level = 0; | |
6200 | int i, j, k; | |
6201 | ||
cb83b629 PZ |
6202 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6203 | if (!sched_domains_numa_distance) | |
6204 | return; | |
6205 | ||
6206 | /* | |
6207 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6208 | * unique distances in the node_distance() table. | |
6209 | * | |
6210 | * Assumes node_distance(0,j) includes all distances in | |
6211 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6212 | */ |
6213 | next_distance = curr_distance; | |
6214 | for (i = 0; i < nr_node_ids; i++) { | |
6215 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6216 | for (k = 0; k < nr_node_ids; k++) { |
6217 | int distance = node_distance(i, k); | |
6218 | ||
6219 | if (distance > curr_distance && | |
6220 | (distance < next_distance || | |
6221 | next_distance == curr_distance)) | |
6222 | next_distance = distance; | |
6223 | ||
6224 | /* | |
6225 | * While not a strong assumption it would be nice to know | |
6226 | * about cases where if node A is connected to B, B is not | |
6227 | * equally connected to A. | |
6228 | */ | |
6229 | if (sched_debug() && node_distance(k, i) != distance) | |
6230 | sched_numa_warn("Node-distance not symmetric"); | |
6231 | ||
6232 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6233 | sched_numa_warn("Node-0 not representative"); | |
6234 | } | |
6235 | if (next_distance != curr_distance) { | |
6236 | sched_domains_numa_distance[level++] = next_distance; | |
6237 | sched_domains_numa_levels = level; | |
6238 | curr_distance = next_distance; | |
6239 | } else break; | |
cb83b629 | 6240 | } |
d039ac60 PZ |
6241 | |
6242 | /* | |
6243 | * In case of sched_debug() we verify the above assumption. | |
6244 | */ | |
6245 | if (!sched_debug()) | |
6246 | break; | |
cb83b629 PZ |
6247 | } |
6248 | /* | |
6249 | * 'level' contains the number of unique distances, excluding the | |
6250 | * identity distance node_distance(i,i). | |
6251 | * | |
28b4a521 | 6252 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6253 | * numbers. |
6254 | */ | |
6255 | ||
5f7865f3 TC |
6256 | /* |
6257 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6258 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6259 | * the array will contain less then 'level' members. This could be | |
6260 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6261 | * in other functions. | |
6262 | * | |
6263 | * We reset it to 'level' at the end of this function. | |
6264 | */ | |
6265 | sched_domains_numa_levels = 0; | |
6266 | ||
cb83b629 PZ |
6267 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6268 | if (!sched_domains_numa_masks) | |
6269 | return; | |
6270 | ||
6271 | /* | |
6272 | * Now for each level, construct a mask per node which contains all | |
6273 | * cpus of nodes that are that many hops away from us. | |
6274 | */ | |
6275 | for (i = 0; i < level; i++) { | |
6276 | sched_domains_numa_masks[i] = | |
6277 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6278 | if (!sched_domains_numa_masks[i]) | |
6279 | return; | |
6280 | ||
6281 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6282 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6283 | if (!mask) |
6284 | return; | |
6285 | ||
6286 | sched_domains_numa_masks[i][j] = mask; | |
6287 | ||
6288 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6289 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6290 | continue; |
6291 | ||
6292 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6293 | } | |
6294 | } | |
6295 | } | |
6296 | ||
143e1e28 VG |
6297 | /* Compute default topology size */ |
6298 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6299 | ||
c515db8c | 6300 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6301 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6302 | if (!tl) | |
6303 | return; | |
6304 | ||
6305 | /* | |
6306 | * Copy the default topology bits.. | |
6307 | */ | |
143e1e28 VG |
6308 | for (i = 0; sched_domain_topology[i].mask; i++) |
6309 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6310 | |
6311 | /* | |
6312 | * .. and append 'j' levels of NUMA goodness. | |
6313 | */ | |
6314 | for (j = 0; j < level; i++, j++) { | |
6315 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6316 | .mask = sd_numa_mask, |
143e1e28 | 6317 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6318 | .flags = SDTL_OVERLAP, |
6319 | .numa_level = j, | |
143e1e28 | 6320 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6321 | }; |
6322 | } | |
6323 | ||
6324 | sched_domain_topology = tl; | |
5f7865f3 TC |
6325 | |
6326 | sched_domains_numa_levels = level; | |
cb83b629 | 6327 | } |
301a5cba TC |
6328 | |
6329 | static void sched_domains_numa_masks_set(int cpu) | |
6330 | { | |
6331 | int i, j; | |
6332 | int node = cpu_to_node(cpu); | |
6333 | ||
6334 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6335 | for (j = 0; j < nr_node_ids; j++) { | |
6336 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6337 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6338 | } | |
6339 | } | |
6340 | } | |
6341 | ||
6342 | static void sched_domains_numa_masks_clear(int cpu) | |
6343 | { | |
6344 | int i, j; | |
6345 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6346 | for (j = 0; j < nr_node_ids; j++) | |
6347 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6348 | } | |
6349 | } | |
6350 | ||
6351 | /* | |
6352 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6353 | * are onlined. | |
6354 | */ | |
6355 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6356 | unsigned long action, | |
6357 | void *hcpu) | |
6358 | { | |
6359 | int cpu = (long)hcpu; | |
6360 | ||
6361 | switch (action & ~CPU_TASKS_FROZEN) { | |
6362 | case CPU_ONLINE: | |
6363 | sched_domains_numa_masks_set(cpu); | |
6364 | break; | |
6365 | ||
6366 | case CPU_DEAD: | |
6367 | sched_domains_numa_masks_clear(cpu); | |
6368 | break; | |
6369 | ||
6370 | default: | |
6371 | return NOTIFY_DONE; | |
6372 | } | |
6373 | ||
6374 | return NOTIFY_OK; | |
cb83b629 PZ |
6375 | } |
6376 | #else | |
6377 | static inline void sched_init_numa(void) | |
6378 | { | |
6379 | } | |
301a5cba TC |
6380 | |
6381 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6382 | unsigned long action, | |
6383 | void *hcpu) | |
6384 | { | |
6385 | return 0; | |
6386 | } | |
cb83b629 PZ |
6387 | #endif /* CONFIG_NUMA */ |
6388 | ||
54ab4ff4 PZ |
6389 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6390 | { | |
6391 | struct sched_domain_topology_level *tl; | |
6392 | int j; | |
6393 | ||
27723a68 | 6394 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6395 | struct sd_data *sdd = &tl->data; |
6396 | ||
6397 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6398 | if (!sdd->sd) | |
6399 | return -ENOMEM; | |
6400 | ||
6401 | sdd->sg = alloc_percpu(struct sched_group *); | |
6402 | if (!sdd->sg) | |
6403 | return -ENOMEM; | |
6404 | ||
63b2ca30 NP |
6405 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6406 | if (!sdd->sgc) | |
9c3f75cb PZ |
6407 | return -ENOMEM; |
6408 | ||
54ab4ff4 PZ |
6409 | for_each_cpu(j, cpu_map) { |
6410 | struct sched_domain *sd; | |
6411 | struct sched_group *sg; | |
63b2ca30 | 6412 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6413 | |
6414 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6415 | GFP_KERNEL, cpu_to_node(j)); | |
6416 | if (!sd) | |
6417 | return -ENOMEM; | |
6418 | ||
6419 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6420 | ||
6421 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6422 | GFP_KERNEL, cpu_to_node(j)); | |
6423 | if (!sg) | |
6424 | return -ENOMEM; | |
6425 | ||
30b4e9eb IM |
6426 | sg->next = sg; |
6427 | ||
54ab4ff4 | 6428 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6429 | |
63b2ca30 | 6430 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6431 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6432 | if (!sgc) |
9c3f75cb PZ |
6433 | return -ENOMEM; |
6434 | ||
63b2ca30 | 6435 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6436 | } |
6437 | } | |
6438 | ||
6439 | return 0; | |
6440 | } | |
6441 | ||
6442 | static void __sdt_free(const struct cpumask *cpu_map) | |
6443 | { | |
6444 | struct sched_domain_topology_level *tl; | |
6445 | int j; | |
6446 | ||
27723a68 | 6447 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6448 | struct sd_data *sdd = &tl->data; |
6449 | ||
6450 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6451 | struct sched_domain *sd; |
6452 | ||
6453 | if (sdd->sd) { | |
6454 | sd = *per_cpu_ptr(sdd->sd, j); | |
6455 | if (sd && (sd->flags & SD_OVERLAP)) | |
6456 | free_sched_groups(sd->groups, 0); | |
6457 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6458 | } | |
6459 | ||
6460 | if (sdd->sg) | |
6461 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6462 | if (sdd->sgc) |
6463 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6464 | } |
6465 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6466 | sdd->sd = NULL; |
54ab4ff4 | 6467 | free_percpu(sdd->sg); |
fb2cf2c6 | 6468 | sdd->sg = NULL; |
63b2ca30 NP |
6469 | free_percpu(sdd->sgc); |
6470 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6471 | } |
6472 | } | |
6473 | ||
2c402dc3 | 6474 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6475 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6476 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6477 | { |
143e1e28 | 6478 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6479 | if (!sd) |
d069b916 | 6480 | return child; |
2c402dc3 | 6481 | |
2c402dc3 | 6482 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6483 | if (child) { |
6484 | sd->level = child->level + 1; | |
6485 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6486 | child->parent = sd; |
c75e0128 | 6487 | sd->child = child; |
60495e77 | 6488 | } |
a841f8ce | 6489 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6490 | |
6491 | return sd; | |
6492 | } | |
6493 | ||
2109b99e AH |
6494 | /* |
6495 | * Build sched domains for a given set of cpus and attach the sched domains | |
6496 | * to the individual cpus | |
6497 | */ | |
dce840a0 PZ |
6498 | static int build_sched_domains(const struct cpumask *cpu_map, |
6499 | struct sched_domain_attr *attr) | |
2109b99e | 6500 | { |
1c632169 | 6501 | enum s_alloc alloc_state; |
dce840a0 | 6502 | struct sched_domain *sd; |
2109b99e | 6503 | struct s_data d; |
822ff793 | 6504 | int i, ret = -ENOMEM; |
9c1cfda2 | 6505 | |
2109b99e AH |
6506 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6507 | if (alloc_state != sa_rootdomain) | |
6508 | goto error; | |
9c1cfda2 | 6509 | |
dce840a0 | 6510 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6511 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6512 | struct sched_domain_topology_level *tl; |
6513 | ||
3bd65a80 | 6514 | sd = NULL; |
27723a68 | 6515 | for_each_sd_topology(tl) { |
4a850cbe | 6516 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6517 | if (tl == sched_domain_topology) |
6518 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6519 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6520 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6521 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6522 | break; | |
e3589f6c | 6523 | } |
dce840a0 PZ |
6524 | } |
6525 | ||
6526 | /* Build the groups for the domains */ | |
6527 | for_each_cpu(i, cpu_map) { | |
6528 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6529 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6530 | if (sd->flags & SD_OVERLAP) { |
6531 | if (build_overlap_sched_groups(sd, i)) | |
6532 | goto error; | |
6533 | } else { | |
6534 | if (build_sched_groups(sd, i)) | |
6535 | goto error; | |
6536 | } | |
1cf51902 | 6537 | } |
a06dadbe | 6538 | } |
9c1cfda2 | 6539 | |
ced549fa | 6540 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6541 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6542 | if (!cpumask_test_cpu(i, cpu_map)) | |
6543 | continue; | |
9c1cfda2 | 6544 | |
dce840a0 PZ |
6545 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6546 | claim_allocations(i, sd); | |
63b2ca30 | 6547 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6548 | } |
f712c0c7 | 6549 | } |
9c1cfda2 | 6550 | |
1da177e4 | 6551 | /* Attach the domains */ |
dce840a0 | 6552 | rcu_read_lock(); |
abcd083a | 6553 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6554 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6555 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6556 | } |
dce840a0 | 6557 | rcu_read_unlock(); |
51888ca2 | 6558 | |
822ff793 | 6559 | ret = 0; |
51888ca2 | 6560 | error: |
2109b99e | 6561 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6562 | return ret; |
1da177e4 | 6563 | } |
029190c5 | 6564 | |
acc3f5d7 | 6565 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6566 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6567 | static struct sched_domain_attr *dattr_cur; |
6568 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6569 | |
6570 | /* | |
6571 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6572 | * cpumask) fails, then fallback to a single sched domain, |
6573 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6574 | */ |
4212823f | 6575 | static cpumask_var_t fallback_doms; |
029190c5 | 6576 | |
ee79d1bd HC |
6577 | /* |
6578 | * arch_update_cpu_topology lets virtualized architectures update the | |
6579 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6580 | * or 0 if it stayed the same. | |
6581 | */ | |
52f5684c | 6582 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6583 | { |
ee79d1bd | 6584 | return 0; |
22e52b07 HC |
6585 | } |
6586 | ||
acc3f5d7 RR |
6587 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6588 | { | |
6589 | int i; | |
6590 | cpumask_var_t *doms; | |
6591 | ||
6592 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6593 | if (!doms) | |
6594 | return NULL; | |
6595 | for (i = 0; i < ndoms; i++) { | |
6596 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6597 | free_sched_domains(doms, i); | |
6598 | return NULL; | |
6599 | } | |
6600 | } | |
6601 | return doms; | |
6602 | } | |
6603 | ||
6604 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6605 | { | |
6606 | unsigned int i; | |
6607 | for (i = 0; i < ndoms; i++) | |
6608 | free_cpumask_var(doms[i]); | |
6609 | kfree(doms); | |
6610 | } | |
6611 | ||
1a20ff27 | 6612 | /* |
41a2d6cf | 6613 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6614 | * For now this just excludes isolated cpus, but could be used to |
6615 | * exclude other special cases in the future. | |
1a20ff27 | 6616 | */ |
c4a8849a | 6617 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6618 | { |
7378547f MM |
6619 | int err; |
6620 | ||
22e52b07 | 6621 | arch_update_cpu_topology(); |
029190c5 | 6622 | ndoms_cur = 1; |
acc3f5d7 | 6623 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6624 | if (!doms_cur) |
acc3f5d7 RR |
6625 | doms_cur = &fallback_doms; |
6626 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6627 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6628 | register_sched_domain_sysctl(); |
7378547f MM |
6629 | |
6630 | return err; | |
1a20ff27 DG |
6631 | } |
6632 | ||
1a20ff27 DG |
6633 | /* |
6634 | * Detach sched domains from a group of cpus specified in cpu_map | |
6635 | * These cpus will now be attached to the NULL domain | |
6636 | */ | |
96f874e2 | 6637 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6638 | { |
6639 | int i; | |
6640 | ||
dce840a0 | 6641 | rcu_read_lock(); |
abcd083a | 6642 | for_each_cpu(i, cpu_map) |
57d885fe | 6643 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6644 | rcu_read_unlock(); |
1a20ff27 DG |
6645 | } |
6646 | ||
1d3504fc HS |
6647 | /* handle null as "default" */ |
6648 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6649 | struct sched_domain_attr *new, int idx_new) | |
6650 | { | |
6651 | struct sched_domain_attr tmp; | |
6652 | ||
6653 | /* fast path */ | |
6654 | if (!new && !cur) | |
6655 | return 1; | |
6656 | ||
6657 | tmp = SD_ATTR_INIT; | |
6658 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6659 | new ? (new + idx_new) : &tmp, | |
6660 | sizeof(struct sched_domain_attr)); | |
6661 | } | |
6662 | ||
029190c5 PJ |
6663 | /* |
6664 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6665 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6666 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6667 | * It destroys each deleted domain and builds each new domain. | |
6668 | * | |
acc3f5d7 | 6669 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6670 | * The masks don't intersect (don't overlap.) We should setup one |
6671 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6672 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6673 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6674 | * it as it is. | |
6675 | * | |
acc3f5d7 RR |
6676 | * The passed in 'doms_new' should be allocated using |
6677 | * alloc_sched_domains. This routine takes ownership of it and will | |
6678 | * free_sched_domains it when done with it. If the caller failed the | |
6679 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6680 | * and partition_sched_domains() will fallback to the single partition | |
6681 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6682 | * |
96f874e2 | 6683 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6684 | * ndoms_new == 0 is a special case for destroying existing domains, |
6685 | * and it will not create the default domain. | |
dfb512ec | 6686 | * |
029190c5 PJ |
6687 | * Call with hotplug lock held |
6688 | */ | |
acc3f5d7 | 6689 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6690 | struct sched_domain_attr *dattr_new) |
029190c5 | 6691 | { |
dfb512ec | 6692 | int i, j, n; |
d65bd5ec | 6693 | int new_topology; |
029190c5 | 6694 | |
712555ee | 6695 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6696 | |
7378547f MM |
6697 | /* always unregister in case we don't destroy any domains */ |
6698 | unregister_sched_domain_sysctl(); | |
6699 | ||
d65bd5ec HC |
6700 | /* Let architecture update cpu core mappings. */ |
6701 | new_topology = arch_update_cpu_topology(); | |
6702 | ||
dfb512ec | 6703 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6704 | |
6705 | /* Destroy deleted domains */ | |
6706 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6707 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6708 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6709 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6710 | goto match1; |
6711 | } | |
6712 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6713 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6714 | match1: |
6715 | ; | |
6716 | } | |
6717 | ||
c8d2d47a | 6718 | n = ndoms_cur; |
e761b772 | 6719 | if (doms_new == NULL) { |
c8d2d47a | 6720 | n = 0; |
acc3f5d7 | 6721 | doms_new = &fallback_doms; |
6ad4c188 | 6722 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6723 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6724 | } |
6725 | ||
029190c5 PJ |
6726 | /* Build new domains */ |
6727 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6728 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6729 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6730 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6731 | goto match2; |
6732 | } | |
6733 | /* no match - add a new doms_new */ | |
dce840a0 | 6734 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6735 | match2: |
6736 | ; | |
6737 | } | |
6738 | ||
6739 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6740 | if (doms_cur != &fallback_doms) |
6741 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6742 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6743 | doms_cur = doms_new; |
1d3504fc | 6744 | dattr_cur = dattr_new; |
029190c5 | 6745 | ndoms_cur = ndoms_new; |
7378547f MM |
6746 | |
6747 | register_sched_domain_sysctl(); | |
a1835615 | 6748 | |
712555ee | 6749 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6750 | } |
6751 | ||
d35be8ba SB |
6752 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6753 | ||
1da177e4 | 6754 | /* |
3a101d05 TH |
6755 | * Update cpusets according to cpu_active mask. If cpusets are |
6756 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6757 | * around partition_sched_domains(). | |
d35be8ba SB |
6758 | * |
6759 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6760 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6761 | */ |
0b2e918a TH |
6762 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6763 | void *hcpu) | |
e761b772 | 6764 | { |
d35be8ba SB |
6765 | switch (action) { |
6766 | case CPU_ONLINE_FROZEN: | |
6767 | case CPU_DOWN_FAILED_FROZEN: | |
6768 | ||
6769 | /* | |
6770 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6771 | * resume sequence. As long as this is not the last online | |
6772 | * operation in the resume sequence, just build a single sched | |
6773 | * domain, ignoring cpusets. | |
6774 | */ | |
6775 | num_cpus_frozen--; | |
6776 | if (likely(num_cpus_frozen)) { | |
6777 | partition_sched_domains(1, NULL, NULL); | |
6778 | break; | |
6779 | } | |
6780 | ||
6781 | /* | |
6782 | * This is the last CPU online operation. So fall through and | |
6783 | * restore the original sched domains by considering the | |
6784 | * cpuset configurations. | |
6785 | */ | |
6786 | ||
e761b772 | 6787 | case CPU_ONLINE: |
6ad4c188 | 6788 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6789 | cpuset_update_active_cpus(true); |
d35be8ba | 6790 | break; |
3a101d05 TH |
6791 | default: |
6792 | return NOTIFY_DONE; | |
6793 | } | |
d35be8ba | 6794 | return NOTIFY_OK; |
3a101d05 | 6795 | } |
e761b772 | 6796 | |
0b2e918a TH |
6797 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6798 | void *hcpu) | |
3a101d05 | 6799 | { |
d35be8ba | 6800 | switch (action) { |
3a101d05 | 6801 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6802 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6803 | break; |
6804 | case CPU_DOWN_PREPARE_FROZEN: | |
6805 | num_cpus_frozen++; | |
6806 | partition_sched_domains(1, NULL, NULL); | |
6807 | break; | |
e761b772 MK |
6808 | default: |
6809 | return NOTIFY_DONE; | |
6810 | } | |
d35be8ba | 6811 | return NOTIFY_OK; |
e761b772 | 6812 | } |
e761b772 | 6813 | |
1da177e4 LT |
6814 | void __init sched_init_smp(void) |
6815 | { | |
dcc30a35 RR |
6816 | cpumask_var_t non_isolated_cpus; |
6817 | ||
6818 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6819 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6820 | |
cb83b629 PZ |
6821 | sched_init_numa(); |
6822 | ||
6acce3ef PZ |
6823 | /* |
6824 | * There's no userspace yet to cause hotplug operations; hence all the | |
6825 | * cpu masks are stable and all blatant races in the below code cannot | |
6826 | * happen. | |
6827 | */ | |
712555ee | 6828 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6829 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6830 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6831 | if (cpumask_empty(non_isolated_cpus)) | |
6832 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6833 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6834 | |
301a5cba | 6835 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6836 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6837 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6838 | |
b328ca18 | 6839 | init_hrtick(); |
5c1e1767 NP |
6840 | |
6841 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6842 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6843 | BUG(); |
19978ca6 | 6844 | sched_init_granularity(); |
dcc30a35 | 6845 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6846 | |
0e3900e6 | 6847 | init_sched_rt_class(); |
1baca4ce | 6848 | init_sched_dl_class(); |
1da177e4 LT |
6849 | } |
6850 | #else | |
6851 | void __init sched_init_smp(void) | |
6852 | { | |
19978ca6 | 6853 | sched_init_granularity(); |
1da177e4 LT |
6854 | } |
6855 | #endif /* CONFIG_SMP */ | |
6856 | ||
cd1bb94b AB |
6857 | const_debug unsigned int sysctl_timer_migration = 1; |
6858 | ||
1da177e4 LT |
6859 | int in_sched_functions(unsigned long addr) |
6860 | { | |
1da177e4 LT |
6861 | return in_lock_functions(addr) || |
6862 | (addr >= (unsigned long)__sched_text_start | |
6863 | && addr < (unsigned long)__sched_text_end); | |
6864 | } | |
6865 | ||
029632fb | 6866 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6867 | /* |
6868 | * Default task group. | |
6869 | * Every task in system belongs to this group at bootup. | |
6870 | */ | |
029632fb | 6871 | struct task_group root_task_group; |
35cf4e50 | 6872 | LIST_HEAD(task_groups); |
052f1dc7 | 6873 | #endif |
6f505b16 | 6874 | |
e6252c3e | 6875 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6876 | |
1da177e4 LT |
6877 | void __init sched_init(void) |
6878 | { | |
dd41f596 | 6879 | int i, j; |
434d53b0 MT |
6880 | unsigned long alloc_size = 0, ptr; |
6881 | ||
6882 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6883 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6884 | #endif | |
6885 | #ifdef CONFIG_RT_GROUP_SCHED | |
6886 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6887 | #endif |
df7c8e84 | 6888 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6889 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6890 | #endif |
434d53b0 | 6891 | if (alloc_size) { |
36b7b6d4 | 6892 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6893 | |
6894 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6895 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6896 | ptr += nr_cpu_ids * sizeof(void **); |
6897 | ||
07e06b01 | 6898 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6899 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6900 | |
6d6bc0ad | 6901 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6902 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6903 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6904 | ptr += nr_cpu_ids * sizeof(void **); |
6905 | ||
07e06b01 | 6906 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6907 | ptr += nr_cpu_ids * sizeof(void **); |
6908 | ||
6d6bc0ad | 6909 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6910 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6911 | for_each_possible_cpu(i) { | |
e6252c3e | 6912 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6913 | ptr += cpumask_size(); |
6914 | } | |
6915 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6916 | } |
dd41f596 | 6917 | |
332ac17e DF |
6918 | init_rt_bandwidth(&def_rt_bandwidth, |
6919 | global_rt_period(), global_rt_runtime()); | |
6920 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6921 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6922 | |
57d885fe GH |
6923 | #ifdef CONFIG_SMP |
6924 | init_defrootdomain(); | |
6925 | #endif | |
6926 | ||
d0b27fa7 | 6927 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6928 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6929 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6930 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6931 | |
7c941438 | 6932 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6933 | list_add(&root_task_group.list, &task_groups); |
6934 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6935 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6936 | autogroup_init(&init_task); |
54c707e9 | 6937 | |
7c941438 | 6938 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6939 | |
0a945022 | 6940 | for_each_possible_cpu(i) { |
70b97a7f | 6941 | struct rq *rq; |
1da177e4 LT |
6942 | |
6943 | rq = cpu_rq(i); | |
05fa785c | 6944 | raw_spin_lock_init(&rq->lock); |
7897986b | 6945 | rq->nr_running = 0; |
dce48a84 TG |
6946 | rq->calc_load_active = 0; |
6947 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6948 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6949 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6950 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6951 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6952 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6953 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6954 | /* |
07e06b01 | 6955 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6956 | * |
6957 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6958 | * gets 100% of the cpu resources in the system. This overall | |
6959 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6960 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6961 | * based on each entity's (task or task-group's) weight |
6962 | * (se->load.weight). | |
6963 | * | |
07e06b01 | 6964 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6965 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6966 | * then A0's share of the cpu resource is: | |
6967 | * | |
0d905bca | 6968 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6969 | * |
07e06b01 YZ |
6970 | * We achieve this by letting root_task_group's tasks sit |
6971 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6972 | */ |
ab84d31e | 6973 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6974 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6975 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6976 | ||
6977 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6978 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6979 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6980 | #endif |
1da177e4 | 6981 | |
dd41f596 IM |
6982 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6983 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6984 | |
6985 | rq->last_load_update_tick = jiffies; | |
6986 | ||
1da177e4 | 6987 | #ifdef CONFIG_SMP |
41c7ce9a | 6988 | rq->sd = NULL; |
57d885fe | 6989 | rq->rd = NULL; |
ca8ce3d0 | 6990 | rq->cpu_capacity = SCHED_CAPACITY_SCALE; |
3f029d3c | 6991 | rq->post_schedule = 0; |
1da177e4 | 6992 | rq->active_balance = 0; |
dd41f596 | 6993 | rq->next_balance = jiffies; |
1da177e4 | 6994 | rq->push_cpu = 0; |
0a2966b4 | 6995 | rq->cpu = i; |
1f11eb6a | 6996 | rq->online = 0; |
eae0c9df MG |
6997 | rq->idle_stamp = 0; |
6998 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6999 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7000 | |
7001 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7002 | ||
dc938520 | 7003 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7004 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 7005 | rq->nohz_flags = 0; |
83cd4fe2 | 7006 | #endif |
265f22a9 FW |
7007 | #ifdef CONFIG_NO_HZ_FULL |
7008 | rq->last_sched_tick = 0; | |
7009 | #endif | |
1da177e4 | 7010 | #endif |
8f4d37ec | 7011 | init_rq_hrtick(rq); |
1da177e4 | 7012 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7013 | } |
7014 | ||
2dd73a4f | 7015 | set_load_weight(&init_task); |
b50f60ce | 7016 | |
e107be36 AK |
7017 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7018 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7019 | #endif | |
7020 | ||
1da177e4 LT |
7021 | /* |
7022 | * The boot idle thread does lazy MMU switching as well: | |
7023 | */ | |
7024 | atomic_inc(&init_mm.mm_count); | |
7025 | enter_lazy_tlb(&init_mm, current); | |
7026 | ||
7027 | /* | |
7028 | * Make us the idle thread. Technically, schedule() should not be | |
7029 | * called from this thread, however somewhere below it might be, | |
7030 | * but because we are the idle thread, we just pick up running again | |
7031 | * when this runqueue becomes "idle". | |
7032 | */ | |
7033 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7034 | |
7035 | calc_load_update = jiffies + LOAD_FREQ; | |
7036 | ||
dd41f596 IM |
7037 | /* |
7038 | * During early bootup we pretend to be a normal task: | |
7039 | */ | |
7040 | current->sched_class = &fair_sched_class; | |
6892b75e | 7041 | |
bf4d83f6 | 7042 | #ifdef CONFIG_SMP |
4cb98839 | 7043 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7044 | /* May be allocated at isolcpus cmdline parse time */ |
7045 | if (cpu_isolated_map == NULL) | |
7046 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7047 | idle_thread_set_boot_cpu(); |
a803f026 | 7048 | set_cpu_rq_start_time(); |
029632fb PZ |
7049 | #endif |
7050 | init_sched_fair_class(); | |
6a7b3dc3 | 7051 | |
6892b75e | 7052 | scheduler_running = 1; |
1da177e4 LT |
7053 | } |
7054 | ||
d902db1e | 7055 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7056 | static inline int preempt_count_equals(int preempt_offset) |
7057 | { | |
234da7bc | 7058 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7059 | |
4ba8216c | 7060 | return (nested == preempt_offset); |
e4aafea2 FW |
7061 | } |
7062 | ||
d894837f | 7063 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7064 | { |
1da177e4 LT |
7065 | static unsigned long prev_jiffy; /* ratelimiting */ |
7066 | ||
b3fbab05 | 7067 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7068 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7069 | !is_idle_task(current)) || | |
e4aafea2 | 7070 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7071 | return; |
7072 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7073 | return; | |
7074 | prev_jiffy = jiffies; | |
7075 | ||
3df0fc5b PZ |
7076 | printk(KERN_ERR |
7077 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7078 | file, line); | |
7079 | printk(KERN_ERR | |
7080 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7081 | in_atomic(), irqs_disabled(), | |
7082 | current->pid, current->comm); | |
aef745fc IM |
7083 | |
7084 | debug_show_held_locks(current); | |
7085 | if (irqs_disabled()) | |
7086 | print_irqtrace_events(current); | |
8f47b187 TG |
7087 | #ifdef CONFIG_DEBUG_PREEMPT |
7088 | if (!preempt_count_equals(preempt_offset)) { | |
7089 | pr_err("Preemption disabled at:"); | |
7090 | print_ip_sym(current->preempt_disable_ip); | |
7091 | pr_cont("\n"); | |
7092 | } | |
7093 | #endif | |
aef745fc | 7094 | dump_stack(); |
1da177e4 LT |
7095 | } |
7096 | EXPORT_SYMBOL(__might_sleep); | |
7097 | #endif | |
7098 | ||
7099 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7100 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7101 | { | |
da7a735e | 7102 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7103 | struct sched_attr attr = { |
7104 | .sched_policy = SCHED_NORMAL, | |
7105 | }; | |
da7a735e | 7106 | int old_prio = p->prio; |
3a5e4dc1 | 7107 | int on_rq; |
3e51f33f | 7108 | |
fd2f4419 | 7109 | on_rq = p->on_rq; |
3a5e4dc1 | 7110 | if (on_rq) |
4ca9b72b | 7111 | dequeue_task(rq, p, 0); |
d50dde5a | 7112 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7113 | if (on_rq) { |
4ca9b72b | 7114 | enqueue_task(rq, p, 0); |
8875125e | 7115 | resched_curr(rq); |
3a5e4dc1 | 7116 | } |
da7a735e PZ |
7117 | |
7118 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7119 | } |
7120 | ||
1da177e4 LT |
7121 | void normalize_rt_tasks(void) |
7122 | { | |
a0f98a1c | 7123 | struct task_struct *g, *p; |
1da177e4 | 7124 | unsigned long flags; |
70b97a7f | 7125 | struct rq *rq; |
1da177e4 | 7126 | |
4cf5d77a | 7127 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7128 | do_each_thread(g, p) { |
178be793 IM |
7129 | /* |
7130 | * Only normalize user tasks: | |
7131 | */ | |
7132 | if (!p->mm) | |
7133 | continue; | |
7134 | ||
6cfb0d5d | 7135 | p->se.exec_start = 0; |
6cfb0d5d | 7136 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7137 | p->se.statistics.wait_start = 0; |
7138 | p->se.statistics.sleep_start = 0; | |
7139 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7140 | #endif |
dd41f596 | 7141 | |
aab03e05 | 7142 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7143 | /* |
7144 | * Renice negative nice level userspace | |
7145 | * tasks back to 0: | |
7146 | */ | |
d0ea0268 | 7147 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7148 | set_user_nice(p, 0); |
1da177e4 | 7149 | continue; |
dd41f596 | 7150 | } |
1da177e4 | 7151 | |
1d615482 | 7152 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7153 | rq = __task_rq_lock(p); |
1da177e4 | 7154 | |
178be793 | 7155 | normalize_task(rq, p); |
3a5e4dc1 | 7156 | |
b29739f9 | 7157 | __task_rq_unlock(rq); |
1d615482 | 7158 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7159 | } while_each_thread(g, p); |
7160 | ||
4cf5d77a | 7161 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7162 | } |
7163 | ||
7164 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7165 | |
67fc4e0c | 7166 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7167 | /* |
67fc4e0c | 7168 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7169 | * |
7170 | * They can only be called when the whole system has been | |
7171 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7172 | * activity can take place. Using them for anything else would | |
7173 | * be a serious bug, and as a result, they aren't even visible | |
7174 | * under any other configuration. | |
7175 | */ | |
7176 | ||
7177 | /** | |
7178 | * curr_task - return the current task for a given cpu. | |
7179 | * @cpu: the processor in question. | |
7180 | * | |
7181 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7182 | * |
7183 | * Return: The current task for @cpu. | |
1df5c10a | 7184 | */ |
36c8b586 | 7185 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7186 | { |
7187 | return cpu_curr(cpu); | |
7188 | } | |
7189 | ||
67fc4e0c JW |
7190 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7191 | ||
7192 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7193 | /** |
7194 | * set_curr_task - set the current task for a given cpu. | |
7195 | * @cpu: the processor in question. | |
7196 | * @p: the task pointer to set. | |
7197 | * | |
7198 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7199 | * are serviced on a separate stack. It allows the architecture to switch the |
7200 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7201 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7202 | * and caller must save the original value of the current task (see | |
7203 | * curr_task() above) and restore that value before reenabling interrupts and | |
7204 | * re-starting the system. | |
7205 | * | |
7206 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7207 | */ | |
36c8b586 | 7208 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7209 | { |
7210 | cpu_curr(cpu) = p; | |
7211 | } | |
7212 | ||
7213 | #endif | |
29f59db3 | 7214 | |
7c941438 | 7215 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7216 | /* task_group_lock serializes the addition/removal of task groups */ |
7217 | static DEFINE_SPINLOCK(task_group_lock); | |
7218 | ||
bccbe08a PZ |
7219 | static void free_sched_group(struct task_group *tg) |
7220 | { | |
7221 | free_fair_sched_group(tg); | |
7222 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7223 | autogroup_free(tg); |
bccbe08a PZ |
7224 | kfree(tg); |
7225 | } | |
7226 | ||
7227 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7228 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7229 | { |
7230 | struct task_group *tg; | |
bccbe08a PZ |
7231 | |
7232 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7233 | if (!tg) | |
7234 | return ERR_PTR(-ENOMEM); | |
7235 | ||
ec7dc8ac | 7236 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7237 | goto err; |
7238 | ||
ec7dc8ac | 7239 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7240 | goto err; |
7241 | ||
ace783b9 LZ |
7242 | return tg; |
7243 | ||
7244 | err: | |
7245 | free_sched_group(tg); | |
7246 | return ERR_PTR(-ENOMEM); | |
7247 | } | |
7248 | ||
7249 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7250 | { | |
7251 | unsigned long flags; | |
7252 | ||
8ed36996 | 7253 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7254 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7255 | |
7256 | WARN_ON(!parent); /* root should already exist */ | |
7257 | ||
7258 | tg->parent = parent; | |
f473aa5e | 7259 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7260 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7261 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7262 | } |
7263 | ||
9b5b7751 | 7264 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7265 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7266 | { |
29f59db3 | 7267 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7268 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7269 | } |
7270 | ||
9b5b7751 | 7271 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7272 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7273 | { |
7274 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7275 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7276 | } | |
7277 | ||
7278 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7279 | { |
8ed36996 | 7280 | unsigned long flags; |
9b5b7751 | 7281 | int i; |
29f59db3 | 7282 | |
3d4b47b4 PZ |
7283 | /* end participation in shares distribution */ |
7284 | for_each_possible_cpu(i) | |
bccbe08a | 7285 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7286 | |
7287 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7288 | list_del_rcu(&tg->list); |
f473aa5e | 7289 | list_del_rcu(&tg->siblings); |
8ed36996 | 7290 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7291 | } |
7292 | ||
9b5b7751 | 7293 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7294 | * The caller of this function should have put the task in its new group |
7295 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7296 | * reflect its new group. | |
9b5b7751 SV |
7297 | */ |
7298 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7299 | { |
8323f26c | 7300 | struct task_group *tg; |
29f59db3 SV |
7301 | int on_rq, running; |
7302 | unsigned long flags; | |
7303 | struct rq *rq; | |
7304 | ||
7305 | rq = task_rq_lock(tsk, &flags); | |
7306 | ||
051a1d1a | 7307 | running = task_current(rq, tsk); |
fd2f4419 | 7308 | on_rq = tsk->on_rq; |
29f59db3 | 7309 | |
0e1f3483 | 7310 | if (on_rq) |
29f59db3 | 7311 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7312 | if (unlikely(running)) |
7313 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7314 | |
073219e9 | 7315 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7316 | lockdep_is_held(&tsk->sighand->siglock)), |
7317 | struct task_group, css); | |
7318 | tg = autogroup_task_group(tsk, tg); | |
7319 | tsk->sched_task_group = tg; | |
7320 | ||
810b3817 | 7321 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7322 | if (tsk->sched_class->task_move_group) |
7323 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7324 | else | |
810b3817 | 7325 | #endif |
b2b5ce02 | 7326 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7327 | |
0e1f3483 HS |
7328 | if (unlikely(running)) |
7329 | tsk->sched_class->set_curr_task(rq); | |
7330 | if (on_rq) | |
371fd7e7 | 7331 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7332 | |
0122ec5b | 7333 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7334 | } |
7c941438 | 7335 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7336 | |
a790de99 PT |
7337 | #ifdef CONFIG_RT_GROUP_SCHED |
7338 | /* | |
7339 | * Ensure that the real time constraints are schedulable. | |
7340 | */ | |
7341 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7342 | |
9a7e0b18 PZ |
7343 | /* Must be called with tasklist_lock held */ |
7344 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7345 | { |
9a7e0b18 | 7346 | struct task_struct *g, *p; |
b40b2e8e | 7347 | |
9a7e0b18 | 7348 | do_each_thread(g, p) { |
029632fb | 7349 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7350 | return 1; |
7351 | } while_each_thread(g, p); | |
b40b2e8e | 7352 | |
9a7e0b18 PZ |
7353 | return 0; |
7354 | } | |
b40b2e8e | 7355 | |
9a7e0b18 PZ |
7356 | struct rt_schedulable_data { |
7357 | struct task_group *tg; | |
7358 | u64 rt_period; | |
7359 | u64 rt_runtime; | |
7360 | }; | |
b40b2e8e | 7361 | |
a790de99 | 7362 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7363 | { |
7364 | struct rt_schedulable_data *d = data; | |
7365 | struct task_group *child; | |
7366 | unsigned long total, sum = 0; | |
7367 | u64 period, runtime; | |
b40b2e8e | 7368 | |
9a7e0b18 PZ |
7369 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7370 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7371 | |
9a7e0b18 PZ |
7372 | if (tg == d->tg) { |
7373 | period = d->rt_period; | |
7374 | runtime = d->rt_runtime; | |
b40b2e8e | 7375 | } |
b40b2e8e | 7376 | |
4653f803 PZ |
7377 | /* |
7378 | * Cannot have more runtime than the period. | |
7379 | */ | |
7380 | if (runtime > period && runtime != RUNTIME_INF) | |
7381 | return -EINVAL; | |
6f505b16 | 7382 | |
4653f803 PZ |
7383 | /* |
7384 | * Ensure we don't starve existing RT tasks. | |
7385 | */ | |
9a7e0b18 PZ |
7386 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7387 | return -EBUSY; | |
6f505b16 | 7388 | |
9a7e0b18 | 7389 | total = to_ratio(period, runtime); |
6f505b16 | 7390 | |
4653f803 PZ |
7391 | /* |
7392 | * Nobody can have more than the global setting allows. | |
7393 | */ | |
7394 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7395 | return -EINVAL; | |
6f505b16 | 7396 | |
4653f803 PZ |
7397 | /* |
7398 | * The sum of our children's runtime should not exceed our own. | |
7399 | */ | |
9a7e0b18 PZ |
7400 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7401 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7402 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7403 | |
9a7e0b18 PZ |
7404 | if (child == d->tg) { |
7405 | period = d->rt_period; | |
7406 | runtime = d->rt_runtime; | |
7407 | } | |
6f505b16 | 7408 | |
9a7e0b18 | 7409 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7410 | } |
6f505b16 | 7411 | |
9a7e0b18 PZ |
7412 | if (sum > total) |
7413 | return -EINVAL; | |
7414 | ||
7415 | return 0; | |
6f505b16 PZ |
7416 | } |
7417 | ||
9a7e0b18 | 7418 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7419 | { |
8277434e PT |
7420 | int ret; |
7421 | ||
9a7e0b18 PZ |
7422 | struct rt_schedulable_data data = { |
7423 | .tg = tg, | |
7424 | .rt_period = period, | |
7425 | .rt_runtime = runtime, | |
7426 | }; | |
7427 | ||
8277434e PT |
7428 | rcu_read_lock(); |
7429 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7430 | rcu_read_unlock(); | |
7431 | ||
7432 | return ret; | |
521f1a24 DG |
7433 | } |
7434 | ||
ab84d31e | 7435 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7436 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7437 | { |
ac086bc2 | 7438 | int i, err = 0; |
9f0c1e56 | 7439 | |
9f0c1e56 | 7440 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7441 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7442 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7443 | if (err) | |
9f0c1e56 | 7444 | goto unlock; |
ac086bc2 | 7445 | |
0986b11b | 7446 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7447 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7448 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7449 | |
7450 | for_each_possible_cpu(i) { | |
7451 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7452 | ||
0986b11b | 7453 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7454 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7455 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7456 | } |
0986b11b | 7457 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7458 | unlock: |
521f1a24 | 7459 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7460 | mutex_unlock(&rt_constraints_mutex); |
7461 | ||
7462 | return err; | |
6f505b16 PZ |
7463 | } |
7464 | ||
25cc7da7 | 7465 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7466 | { |
7467 | u64 rt_runtime, rt_period; | |
7468 | ||
7469 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7470 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7471 | if (rt_runtime_us < 0) | |
7472 | rt_runtime = RUNTIME_INF; | |
7473 | ||
ab84d31e | 7474 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7475 | } |
7476 | ||
25cc7da7 | 7477 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7478 | { |
7479 | u64 rt_runtime_us; | |
7480 | ||
d0b27fa7 | 7481 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7482 | return -1; |
7483 | ||
d0b27fa7 | 7484 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7485 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7486 | return rt_runtime_us; | |
7487 | } | |
d0b27fa7 | 7488 | |
25cc7da7 | 7489 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7490 | { |
7491 | u64 rt_runtime, rt_period; | |
7492 | ||
7493 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7494 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7495 | ||
619b0488 R |
7496 | if (rt_period == 0) |
7497 | return -EINVAL; | |
7498 | ||
ab84d31e | 7499 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7500 | } |
7501 | ||
25cc7da7 | 7502 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7503 | { |
7504 | u64 rt_period_us; | |
7505 | ||
7506 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7507 | do_div(rt_period_us, NSEC_PER_USEC); | |
7508 | return rt_period_us; | |
7509 | } | |
332ac17e | 7510 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7511 | |
332ac17e | 7512 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7513 | static int sched_rt_global_constraints(void) |
7514 | { | |
7515 | int ret = 0; | |
7516 | ||
7517 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7518 | read_lock(&tasklist_lock); |
4653f803 | 7519 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7520 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7521 | mutex_unlock(&rt_constraints_mutex); |
7522 | ||
7523 | return ret; | |
7524 | } | |
54e99124 | 7525 | |
25cc7da7 | 7526 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7527 | { |
7528 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7529 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7530 | return 0; | |
7531 | ||
7532 | return 1; | |
7533 | } | |
7534 | ||
6d6bc0ad | 7535 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7536 | static int sched_rt_global_constraints(void) |
7537 | { | |
ac086bc2 | 7538 | unsigned long flags; |
332ac17e | 7539 | int i, ret = 0; |
ec5d4989 | 7540 | |
0986b11b | 7541 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7542 | for_each_possible_cpu(i) { |
7543 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7544 | ||
0986b11b | 7545 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7546 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7547 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7548 | } |
0986b11b | 7549 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7550 | |
332ac17e | 7551 | return ret; |
d0b27fa7 | 7552 | } |
6d6bc0ad | 7553 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7554 | |
332ac17e DF |
7555 | static int sched_dl_global_constraints(void) |
7556 | { | |
1724813d PZ |
7557 | u64 runtime = global_rt_runtime(); |
7558 | u64 period = global_rt_period(); | |
332ac17e | 7559 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7560 | int cpu, ret = 0; |
49516342 | 7561 | unsigned long flags; |
332ac17e DF |
7562 | |
7563 | /* | |
7564 | * Here we want to check the bandwidth not being set to some | |
7565 | * value smaller than the currently allocated bandwidth in | |
7566 | * any of the root_domains. | |
7567 | * | |
7568 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7569 | * cycling on root_domains... Discussion on different/better | |
7570 | * solutions is welcome! | |
7571 | */ | |
1724813d PZ |
7572 | for_each_possible_cpu(cpu) { |
7573 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7574 | |
49516342 | 7575 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7576 | if (new_bw < dl_b->total_bw) |
7577 | ret = -EBUSY; | |
49516342 | 7578 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7579 | |
7580 | if (ret) | |
7581 | break; | |
332ac17e DF |
7582 | } |
7583 | ||
1724813d | 7584 | return ret; |
332ac17e DF |
7585 | } |
7586 | ||
1724813d | 7587 | static void sched_dl_do_global(void) |
ce0dbbbb | 7588 | { |
1724813d PZ |
7589 | u64 new_bw = -1; |
7590 | int cpu; | |
49516342 | 7591 | unsigned long flags; |
ce0dbbbb | 7592 | |
1724813d PZ |
7593 | def_dl_bandwidth.dl_period = global_rt_period(); |
7594 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7595 | ||
7596 | if (global_rt_runtime() != RUNTIME_INF) | |
7597 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7598 | ||
7599 | /* | |
7600 | * FIXME: As above... | |
7601 | */ | |
7602 | for_each_possible_cpu(cpu) { | |
7603 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7604 | ||
49516342 | 7605 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7606 | dl_b->bw = new_bw; |
49516342 | 7607 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7608 | } |
1724813d PZ |
7609 | } |
7610 | ||
7611 | static int sched_rt_global_validate(void) | |
7612 | { | |
7613 | if (sysctl_sched_rt_period <= 0) | |
7614 | return -EINVAL; | |
7615 | ||
e9e7cb38 JL |
7616 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7617 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7618 | return -EINVAL; |
7619 | ||
7620 | return 0; | |
7621 | } | |
7622 | ||
7623 | static void sched_rt_do_global(void) | |
7624 | { | |
7625 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7626 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7627 | } |
7628 | ||
d0b27fa7 | 7629 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7630 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7631 | loff_t *ppos) |
7632 | { | |
d0b27fa7 PZ |
7633 | int old_period, old_runtime; |
7634 | static DEFINE_MUTEX(mutex); | |
1724813d | 7635 | int ret; |
d0b27fa7 PZ |
7636 | |
7637 | mutex_lock(&mutex); | |
7638 | old_period = sysctl_sched_rt_period; | |
7639 | old_runtime = sysctl_sched_rt_runtime; | |
7640 | ||
8d65af78 | 7641 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7642 | |
7643 | if (!ret && write) { | |
1724813d PZ |
7644 | ret = sched_rt_global_validate(); |
7645 | if (ret) | |
7646 | goto undo; | |
7647 | ||
d0b27fa7 | 7648 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7649 | if (ret) |
7650 | goto undo; | |
7651 | ||
7652 | ret = sched_dl_global_constraints(); | |
7653 | if (ret) | |
7654 | goto undo; | |
7655 | ||
7656 | sched_rt_do_global(); | |
7657 | sched_dl_do_global(); | |
7658 | } | |
7659 | if (0) { | |
7660 | undo: | |
7661 | sysctl_sched_rt_period = old_period; | |
7662 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7663 | } |
7664 | mutex_unlock(&mutex); | |
7665 | ||
7666 | return ret; | |
7667 | } | |
68318b8e | 7668 | |
1724813d | 7669 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7670 | void __user *buffer, size_t *lenp, |
7671 | loff_t *ppos) | |
7672 | { | |
7673 | int ret; | |
332ac17e | 7674 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7675 | |
7676 | mutex_lock(&mutex); | |
332ac17e | 7677 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7678 | /* make sure that internally we keep jiffies */ |
7679 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7680 | if (!ret && write) { |
1724813d PZ |
7681 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7682 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7683 | } |
7684 | mutex_unlock(&mutex); | |
332ac17e DF |
7685 | return ret; |
7686 | } | |
7687 | ||
052f1dc7 | 7688 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7689 | |
a7c6d554 | 7690 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7691 | { |
a7c6d554 | 7692 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7693 | } |
7694 | ||
eb95419b TH |
7695 | static struct cgroup_subsys_state * |
7696 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7697 | { |
eb95419b TH |
7698 | struct task_group *parent = css_tg(parent_css); |
7699 | struct task_group *tg; | |
68318b8e | 7700 | |
eb95419b | 7701 | if (!parent) { |
68318b8e | 7702 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7703 | return &root_task_group.css; |
68318b8e SV |
7704 | } |
7705 | ||
ec7dc8ac | 7706 | tg = sched_create_group(parent); |
68318b8e SV |
7707 | if (IS_ERR(tg)) |
7708 | return ERR_PTR(-ENOMEM); | |
7709 | ||
68318b8e SV |
7710 | return &tg->css; |
7711 | } | |
7712 | ||
eb95419b | 7713 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7714 | { |
eb95419b | 7715 | struct task_group *tg = css_tg(css); |
5c9d535b | 7716 | struct task_group *parent = css_tg(css->parent); |
ace783b9 | 7717 | |
63876986 TH |
7718 | if (parent) |
7719 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7720 | return 0; |
7721 | } | |
7722 | ||
eb95419b | 7723 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7724 | { |
eb95419b | 7725 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7726 | |
7727 | sched_destroy_group(tg); | |
7728 | } | |
7729 | ||
eb95419b | 7730 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7731 | { |
eb95419b | 7732 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7733 | |
7734 | sched_offline_group(tg); | |
7735 | } | |
7736 | ||
eb95419b | 7737 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7738 | struct cgroup_taskset *tset) |
68318b8e | 7739 | { |
bb9d97b6 TH |
7740 | struct task_struct *task; |
7741 | ||
924f0d9a | 7742 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7743 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7744 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7745 | return -EINVAL; |
b68aa230 | 7746 | #else |
bb9d97b6 TH |
7747 | /* We don't support RT-tasks being in separate groups */ |
7748 | if (task->sched_class != &fair_sched_class) | |
7749 | return -EINVAL; | |
b68aa230 | 7750 | #endif |
bb9d97b6 | 7751 | } |
be367d09 BB |
7752 | return 0; |
7753 | } | |
68318b8e | 7754 | |
eb95419b | 7755 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7756 | struct cgroup_taskset *tset) |
68318b8e | 7757 | { |
bb9d97b6 TH |
7758 | struct task_struct *task; |
7759 | ||
924f0d9a | 7760 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7761 | sched_move_task(task); |
68318b8e SV |
7762 | } |
7763 | ||
eb95419b TH |
7764 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7765 | struct cgroup_subsys_state *old_css, | |
7766 | struct task_struct *task) | |
068c5cc5 PZ |
7767 | { |
7768 | /* | |
7769 | * cgroup_exit() is called in the copy_process() failure path. | |
7770 | * Ignore this case since the task hasn't ran yet, this avoids | |
7771 | * trying to poke a half freed task state from generic code. | |
7772 | */ | |
7773 | if (!(task->flags & PF_EXITING)) | |
7774 | return; | |
7775 | ||
7776 | sched_move_task(task); | |
7777 | } | |
7778 | ||
052f1dc7 | 7779 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7780 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7781 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7782 | { |
182446d0 | 7783 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7784 | } |
7785 | ||
182446d0 TH |
7786 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7787 | struct cftype *cft) | |
68318b8e | 7788 | { |
182446d0 | 7789 | struct task_group *tg = css_tg(css); |
68318b8e | 7790 | |
c8b28116 | 7791 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7792 | } |
ab84d31e PT |
7793 | |
7794 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7795 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7796 | ||
ab84d31e PT |
7797 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7798 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7799 | ||
a790de99 PT |
7800 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7801 | ||
ab84d31e PT |
7802 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7803 | { | |
56f570e5 | 7804 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7805 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7806 | |
7807 | if (tg == &root_task_group) | |
7808 | return -EINVAL; | |
7809 | ||
7810 | /* | |
7811 | * Ensure we have at some amount of bandwidth every period. This is | |
7812 | * to prevent reaching a state of large arrears when throttled via | |
7813 | * entity_tick() resulting in prolonged exit starvation. | |
7814 | */ | |
7815 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7816 | return -EINVAL; | |
7817 | ||
7818 | /* | |
7819 | * Likewise, bound things on the otherside by preventing insane quota | |
7820 | * periods. This also allows us to normalize in computing quota | |
7821 | * feasibility. | |
7822 | */ | |
7823 | if (period > max_cfs_quota_period) | |
7824 | return -EINVAL; | |
7825 | ||
0e59bdae KT |
7826 | /* |
7827 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
7828 | * unthrottle_offline_cfs_rqs(). | |
7829 | */ | |
7830 | get_online_cpus(); | |
a790de99 PT |
7831 | mutex_lock(&cfs_constraints_mutex); |
7832 | ret = __cfs_schedulable(tg, period, quota); | |
7833 | if (ret) | |
7834 | goto out_unlock; | |
7835 | ||
58088ad0 | 7836 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7837 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7838 | /* |
7839 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7840 | * before making related changes, and on->off must occur afterwards | |
7841 | */ | |
7842 | if (runtime_enabled && !runtime_was_enabled) | |
7843 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7844 | raw_spin_lock_irq(&cfs_b->lock); |
7845 | cfs_b->period = ns_to_ktime(period); | |
7846 | cfs_b->quota = quota; | |
58088ad0 | 7847 | |
a9cf55b2 | 7848 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7849 | /* restart the period timer (if active) to handle new period expiry */ |
7850 | if (runtime_enabled && cfs_b->timer_active) { | |
7851 | /* force a reprogram */ | |
09dc4ab0 | 7852 | __start_cfs_bandwidth(cfs_b, true); |
58088ad0 | 7853 | } |
ab84d31e PT |
7854 | raw_spin_unlock_irq(&cfs_b->lock); |
7855 | ||
0e59bdae | 7856 | for_each_online_cpu(i) { |
ab84d31e | 7857 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7858 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7859 | |
7860 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7861 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7862 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7863 | |
029632fb | 7864 | if (cfs_rq->throttled) |
671fd9da | 7865 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7866 | raw_spin_unlock_irq(&rq->lock); |
7867 | } | |
1ee14e6c BS |
7868 | if (runtime_was_enabled && !runtime_enabled) |
7869 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7870 | out_unlock: |
7871 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7872 | put_online_cpus(); |
ab84d31e | 7873 | |
a790de99 | 7874 | return ret; |
ab84d31e PT |
7875 | } |
7876 | ||
7877 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7878 | { | |
7879 | u64 quota, period; | |
7880 | ||
029632fb | 7881 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7882 | if (cfs_quota_us < 0) |
7883 | quota = RUNTIME_INF; | |
7884 | else | |
7885 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7886 | ||
7887 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7888 | } | |
7889 | ||
7890 | long tg_get_cfs_quota(struct task_group *tg) | |
7891 | { | |
7892 | u64 quota_us; | |
7893 | ||
029632fb | 7894 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7895 | return -1; |
7896 | ||
029632fb | 7897 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7898 | do_div(quota_us, NSEC_PER_USEC); |
7899 | ||
7900 | return quota_us; | |
7901 | } | |
7902 | ||
7903 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7904 | { | |
7905 | u64 quota, period; | |
7906 | ||
7907 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7908 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7909 | |
ab84d31e PT |
7910 | return tg_set_cfs_bandwidth(tg, period, quota); |
7911 | } | |
7912 | ||
7913 | long tg_get_cfs_period(struct task_group *tg) | |
7914 | { | |
7915 | u64 cfs_period_us; | |
7916 | ||
029632fb | 7917 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7918 | do_div(cfs_period_us, NSEC_PER_USEC); |
7919 | ||
7920 | return cfs_period_us; | |
7921 | } | |
7922 | ||
182446d0 TH |
7923 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7924 | struct cftype *cft) | |
ab84d31e | 7925 | { |
182446d0 | 7926 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7927 | } |
7928 | ||
182446d0 TH |
7929 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7930 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7931 | { |
182446d0 | 7932 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7933 | } |
7934 | ||
182446d0 TH |
7935 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7936 | struct cftype *cft) | |
ab84d31e | 7937 | { |
182446d0 | 7938 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7939 | } |
7940 | ||
182446d0 TH |
7941 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7942 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7943 | { |
182446d0 | 7944 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7945 | } |
7946 | ||
a790de99 PT |
7947 | struct cfs_schedulable_data { |
7948 | struct task_group *tg; | |
7949 | u64 period, quota; | |
7950 | }; | |
7951 | ||
7952 | /* | |
7953 | * normalize group quota/period to be quota/max_period | |
7954 | * note: units are usecs | |
7955 | */ | |
7956 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7957 | struct cfs_schedulable_data *d) | |
7958 | { | |
7959 | u64 quota, period; | |
7960 | ||
7961 | if (tg == d->tg) { | |
7962 | period = d->period; | |
7963 | quota = d->quota; | |
7964 | } else { | |
7965 | period = tg_get_cfs_period(tg); | |
7966 | quota = tg_get_cfs_quota(tg); | |
7967 | } | |
7968 | ||
7969 | /* note: these should typically be equivalent */ | |
7970 | if (quota == RUNTIME_INF || quota == -1) | |
7971 | return RUNTIME_INF; | |
7972 | ||
7973 | return to_ratio(period, quota); | |
7974 | } | |
7975 | ||
7976 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7977 | { | |
7978 | struct cfs_schedulable_data *d = data; | |
029632fb | 7979 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7980 | s64 quota = 0, parent_quota = -1; |
7981 | ||
7982 | if (!tg->parent) { | |
7983 | quota = RUNTIME_INF; | |
7984 | } else { | |
029632fb | 7985 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7986 | |
7987 | quota = normalize_cfs_quota(tg, d); | |
7988 | parent_quota = parent_b->hierarchal_quota; | |
7989 | ||
7990 | /* | |
7991 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7992 | * limit is set | |
7993 | */ | |
7994 | if (quota == RUNTIME_INF) | |
7995 | quota = parent_quota; | |
7996 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7997 | return -EINVAL; | |
7998 | } | |
7999 | cfs_b->hierarchal_quota = quota; | |
8000 | ||
8001 | return 0; | |
8002 | } | |
8003 | ||
8004 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8005 | { | |
8277434e | 8006 | int ret; |
a790de99 PT |
8007 | struct cfs_schedulable_data data = { |
8008 | .tg = tg, | |
8009 | .period = period, | |
8010 | .quota = quota, | |
8011 | }; | |
8012 | ||
8013 | if (quota != RUNTIME_INF) { | |
8014 | do_div(data.period, NSEC_PER_USEC); | |
8015 | do_div(data.quota, NSEC_PER_USEC); | |
8016 | } | |
8017 | ||
8277434e PT |
8018 | rcu_read_lock(); |
8019 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8020 | rcu_read_unlock(); | |
8021 | ||
8022 | return ret; | |
a790de99 | 8023 | } |
e8da1b18 | 8024 | |
2da8ca82 | 8025 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 8026 | { |
2da8ca82 | 8027 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8028 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8029 | |
44ffc75b TH |
8030 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8031 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8032 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
8033 | |
8034 | return 0; | |
8035 | } | |
ab84d31e | 8036 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8037 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8038 | |
052f1dc7 | 8039 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8040 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8041 | struct cftype *cft, s64 val) | |
6f505b16 | 8042 | { |
182446d0 | 8043 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8044 | } |
8045 | ||
182446d0 TH |
8046 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8047 | struct cftype *cft) | |
6f505b16 | 8048 | { |
182446d0 | 8049 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8050 | } |
d0b27fa7 | 8051 | |
182446d0 TH |
8052 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8053 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8054 | { |
182446d0 | 8055 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8056 | } |
8057 | ||
182446d0 TH |
8058 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8059 | struct cftype *cft) | |
d0b27fa7 | 8060 | { |
182446d0 | 8061 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8062 | } |
6d6bc0ad | 8063 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8064 | |
fe5c7cc2 | 8065 | static struct cftype cpu_files[] = { |
052f1dc7 | 8066 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8067 | { |
8068 | .name = "shares", | |
f4c753b7 PM |
8069 | .read_u64 = cpu_shares_read_u64, |
8070 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8071 | }, |
052f1dc7 | 8072 | #endif |
ab84d31e PT |
8073 | #ifdef CONFIG_CFS_BANDWIDTH |
8074 | { | |
8075 | .name = "cfs_quota_us", | |
8076 | .read_s64 = cpu_cfs_quota_read_s64, | |
8077 | .write_s64 = cpu_cfs_quota_write_s64, | |
8078 | }, | |
8079 | { | |
8080 | .name = "cfs_period_us", | |
8081 | .read_u64 = cpu_cfs_period_read_u64, | |
8082 | .write_u64 = cpu_cfs_period_write_u64, | |
8083 | }, | |
e8da1b18 NR |
8084 | { |
8085 | .name = "stat", | |
2da8ca82 | 8086 | .seq_show = cpu_stats_show, |
e8da1b18 | 8087 | }, |
ab84d31e | 8088 | #endif |
052f1dc7 | 8089 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8090 | { |
9f0c1e56 | 8091 | .name = "rt_runtime_us", |
06ecb27c PM |
8092 | .read_s64 = cpu_rt_runtime_read, |
8093 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8094 | }, |
d0b27fa7 PZ |
8095 | { |
8096 | .name = "rt_period_us", | |
f4c753b7 PM |
8097 | .read_u64 = cpu_rt_period_read_uint, |
8098 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8099 | }, |
052f1dc7 | 8100 | #endif |
4baf6e33 | 8101 | { } /* terminate */ |
68318b8e SV |
8102 | }; |
8103 | ||
073219e9 | 8104 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8105 | .css_alloc = cpu_cgroup_css_alloc, |
8106 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8107 | .css_online = cpu_cgroup_css_online, |
8108 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8109 | .can_attach = cpu_cgroup_can_attach, |
8110 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8111 | .exit = cpu_cgroup_exit, |
4baf6e33 | 8112 | .base_cftypes = cpu_files, |
68318b8e SV |
8113 | .early_init = 1, |
8114 | }; | |
8115 | ||
052f1dc7 | 8116 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8117 | |
b637a328 PM |
8118 | void dump_cpu_task(int cpu) |
8119 | { | |
8120 | pr_info("Task dump for CPU %d:\n", cpu); | |
8121 | sched_show_task(cpu_curr(cpu)); | |
8122 | } |