]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
1da177e4 | 76 | |
96f951ed | 77 | #include <asm/switch_to.h> |
5517d86b | 78 | #include <asm/tlb.h> |
838225b4 | 79 | #include <asm/irq_regs.h> |
db7e527d | 80 | #include <asm/mutex.h> |
e6e6685a GC |
81 | #ifdef CONFIG_PARAVIRT |
82 | #include <asm/paravirt.h> | |
83 | #endif | |
1da177e4 | 84 | |
029632fb | 85 | #include "sched.h" |
ea138446 | 86 | #include "../workqueue_internal.h" |
29d5e047 | 87 | #include "../smpboot.h" |
6e0534f2 | 88 | |
a8d154b0 | 89 | #define CREATE_TRACE_POINTS |
ad8d75ff | 90 | #include <trace/events/sched.h> |
a8d154b0 | 91 | |
029632fb | 92 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 93 | { |
58088ad0 PT |
94 | unsigned long delta; |
95 | ktime_t soft, hard, now; | |
d0b27fa7 | 96 | |
58088ad0 PT |
97 | for (;;) { |
98 | if (hrtimer_active(period_timer)) | |
99 | break; | |
100 | ||
101 | now = hrtimer_cb_get_time(period_timer); | |
102 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 103 | |
58088ad0 PT |
104 | soft = hrtimer_get_softexpires(period_timer); |
105 | hard = hrtimer_get_expires(period_timer); | |
106 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
107 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
108 | HRTIMER_MODE_ABS_PINNED, 0); | |
109 | } | |
110 | } | |
111 | ||
029632fb PZ |
112 | DEFINE_MUTEX(sched_domains_mutex); |
113 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 114 | |
fe44d621 | 115 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 116 | |
029632fb | 117 | void update_rq_clock(struct rq *rq) |
3e51f33f | 118 | { |
fe44d621 | 119 | s64 delta; |
305e6835 | 120 | |
61eadef6 | 121 | if (rq->skip_clock_update > 0) |
f26f9aff | 122 | return; |
aa483808 | 123 | |
fe44d621 PZ |
124 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
125 | rq->clock += delta; | |
126 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
127 | } |
128 | ||
bf5c91ba IM |
129 | /* |
130 | * Debugging: various feature bits | |
131 | */ | |
f00b45c1 | 132 | |
f00b45c1 PZ |
133 | #define SCHED_FEAT(name, enabled) \ |
134 | (1UL << __SCHED_FEAT_##name) * enabled | | |
135 | ||
bf5c91ba | 136 | const_debug unsigned int sysctl_sched_features = |
391e43da | 137 | #include "features.h" |
f00b45c1 PZ |
138 | 0; |
139 | ||
140 | #undef SCHED_FEAT | |
141 | ||
142 | #ifdef CONFIG_SCHED_DEBUG | |
143 | #define SCHED_FEAT(name, enabled) \ | |
144 | #name , | |
145 | ||
1292531f | 146 | static const char * const sched_feat_names[] = { |
391e43da | 147 | #include "features.h" |
f00b45c1 PZ |
148 | }; |
149 | ||
150 | #undef SCHED_FEAT | |
151 | ||
34f3a814 | 152 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 153 | { |
f00b45c1 PZ |
154 | int i; |
155 | ||
f8b6d1cc | 156 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
157 | if (!(sysctl_sched_features & (1UL << i))) |
158 | seq_puts(m, "NO_"); | |
159 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 160 | } |
34f3a814 | 161 | seq_puts(m, "\n"); |
f00b45c1 | 162 | |
34f3a814 | 163 | return 0; |
f00b45c1 PZ |
164 | } |
165 | ||
f8b6d1cc PZ |
166 | #ifdef HAVE_JUMP_LABEL |
167 | ||
c5905afb IM |
168 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
169 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
170 | |
171 | #define SCHED_FEAT(name, enabled) \ | |
172 | jump_label_key__##enabled , | |
173 | ||
c5905afb | 174 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
175 | #include "features.h" |
176 | }; | |
177 | ||
178 | #undef SCHED_FEAT | |
179 | ||
180 | static void sched_feat_disable(int i) | |
181 | { | |
c5905afb IM |
182 | if (static_key_enabled(&sched_feat_keys[i])) |
183 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
184 | } |
185 | ||
186 | static void sched_feat_enable(int i) | |
187 | { | |
c5905afb IM |
188 | if (!static_key_enabled(&sched_feat_keys[i])) |
189 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
190 | } |
191 | #else | |
192 | static void sched_feat_disable(int i) { }; | |
193 | static void sched_feat_enable(int i) { }; | |
194 | #endif /* HAVE_JUMP_LABEL */ | |
195 | ||
1a687c2e | 196 | static int sched_feat_set(char *cmp) |
f00b45c1 | 197 | { |
f00b45c1 | 198 | int i; |
1a687c2e | 199 | int neg = 0; |
f00b45c1 | 200 | |
524429c3 | 201 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
202 | neg = 1; |
203 | cmp += 3; | |
204 | } | |
205 | ||
f8b6d1cc | 206 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 207 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 208 | if (neg) { |
f00b45c1 | 209 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
210 | sched_feat_disable(i); |
211 | } else { | |
f00b45c1 | 212 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
213 | sched_feat_enable(i); |
214 | } | |
f00b45c1 PZ |
215 | break; |
216 | } | |
217 | } | |
218 | ||
1a687c2e MG |
219 | return i; |
220 | } | |
221 | ||
222 | static ssize_t | |
223 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
224 | size_t cnt, loff_t *ppos) | |
225 | { | |
226 | char buf[64]; | |
227 | char *cmp; | |
228 | int i; | |
229 | ||
230 | if (cnt > 63) | |
231 | cnt = 63; | |
232 | ||
233 | if (copy_from_user(&buf, ubuf, cnt)) | |
234 | return -EFAULT; | |
235 | ||
236 | buf[cnt] = 0; | |
237 | cmp = strstrip(buf); | |
238 | ||
239 | i = sched_feat_set(cmp); | |
f8b6d1cc | 240 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
241 | return -EINVAL; |
242 | ||
42994724 | 243 | *ppos += cnt; |
f00b45c1 PZ |
244 | |
245 | return cnt; | |
246 | } | |
247 | ||
34f3a814 LZ |
248 | static int sched_feat_open(struct inode *inode, struct file *filp) |
249 | { | |
250 | return single_open(filp, sched_feat_show, NULL); | |
251 | } | |
252 | ||
828c0950 | 253 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
254 | .open = sched_feat_open, |
255 | .write = sched_feat_write, | |
256 | .read = seq_read, | |
257 | .llseek = seq_lseek, | |
258 | .release = single_release, | |
f00b45c1 PZ |
259 | }; |
260 | ||
261 | static __init int sched_init_debug(void) | |
262 | { | |
f00b45c1 PZ |
263 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
264 | &sched_feat_fops); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | late_initcall(sched_init_debug); | |
f8b6d1cc | 269 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 270 | |
b82d9fdd PZ |
271 | /* |
272 | * Number of tasks to iterate in a single balance run. | |
273 | * Limited because this is done with IRQs disabled. | |
274 | */ | |
275 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
276 | ||
e9e9250b PZ |
277 | /* |
278 | * period over which we average the RT time consumption, measured | |
279 | * in ms. | |
280 | * | |
281 | * default: 1s | |
282 | */ | |
283 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
284 | ||
fa85ae24 | 285 | /* |
9f0c1e56 | 286 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
287 | * default: 1s |
288 | */ | |
9f0c1e56 | 289 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 290 | |
029632fb | 291 | __read_mostly int scheduler_running; |
6892b75e | 292 | |
9f0c1e56 PZ |
293 | /* |
294 | * part of the period that we allow rt tasks to run in us. | |
295 | * default: 0.95s | |
296 | */ | |
297 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 298 | |
0970d299 | 299 | /* |
0122ec5b | 300 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 301 | */ |
70b97a7f | 302 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
303 | __acquires(rq->lock) |
304 | { | |
0970d299 PZ |
305 | struct rq *rq; |
306 | ||
0122ec5b PZ |
307 | lockdep_assert_held(&p->pi_lock); |
308 | ||
3a5c359a | 309 | for (;;) { |
0970d299 | 310 | rq = task_rq(p); |
05fa785c | 311 | raw_spin_lock(&rq->lock); |
65cc8e48 | 312 | if (likely(rq == task_rq(p))) |
3a5c359a | 313 | return rq; |
05fa785c | 314 | raw_spin_unlock(&rq->lock); |
b29739f9 | 315 | } |
b29739f9 IM |
316 | } |
317 | ||
1da177e4 | 318 | /* |
0122ec5b | 319 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 320 | */ |
70b97a7f | 321 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 322 | __acquires(p->pi_lock) |
1da177e4 LT |
323 | __acquires(rq->lock) |
324 | { | |
70b97a7f | 325 | struct rq *rq; |
1da177e4 | 326 | |
3a5c359a | 327 | for (;;) { |
0122ec5b | 328 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 329 | rq = task_rq(p); |
05fa785c | 330 | raw_spin_lock(&rq->lock); |
65cc8e48 | 331 | if (likely(rq == task_rq(p))) |
3a5c359a | 332 | return rq; |
0122ec5b PZ |
333 | raw_spin_unlock(&rq->lock); |
334 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 335 | } |
1da177e4 LT |
336 | } |
337 | ||
a9957449 | 338 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
339 | __releases(rq->lock) |
340 | { | |
05fa785c | 341 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
342 | } |
343 | ||
0122ec5b PZ |
344 | static inline void |
345 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 346 | __releases(rq->lock) |
0122ec5b | 347 | __releases(p->pi_lock) |
1da177e4 | 348 | { |
0122ec5b PZ |
349 | raw_spin_unlock(&rq->lock); |
350 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
351 | } |
352 | ||
1da177e4 | 353 | /* |
cc2a73b5 | 354 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 355 | */ |
a9957449 | 356 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
357 | __acquires(rq->lock) |
358 | { | |
70b97a7f | 359 | struct rq *rq; |
1da177e4 LT |
360 | |
361 | local_irq_disable(); | |
362 | rq = this_rq(); | |
05fa785c | 363 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
364 | |
365 | return rq; | |
366 | } | |
367 | ||
8f4d37ec PZ |
368 | #ifdef CONFIG_SCHED_HRTICK |
369 | /* | |
370 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 371 | */ |
8f4d37ec | 372 | |
8f4d37ec PZ |
373 | static void hrtick_clear(struct rq *rq) |
374 | { | |
375 | if (hrtimer_active(&rq->hrtick_timer)) | |
376 | hrtimer_cancel(&rq->hrtick_timer); | |
377 | } | |
378 | ||
8f4d37ec PZ |
379 | /* |
380 | * High-resolution timer tick. | |
381 | * Runs from hardirq context with interrupts disabled. | |
382 | */ | |
383 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
384 | { | |
385 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
386 | ||
387 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
388 | ||
05fa785c | 389 | raw_spin_lock(&rq->lock); |
3e51f33f | 390 | update_rq_clock(rq); |
8f4d37ec | 391 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 392 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
393 | |
394 | return HRTIMER_NORESTART; | |
395 | } | |
396 | ||
95e904c7 | 397 | #ifdef CONFIG_SMP |
971ee28c PZ |
398 | |
399 | static int __hrtick_restart(struct rq *rq) | |
400 | { | |
401 | struct hrtimer *timer = &rq->hrtick_timer; | |
402 | ktime_t time = hrtimer_get_softexpires(timer); | |
403 | ||
404 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
405 | } | |
406 | ||
31656519 PZ |
407 | /* |
408 | * called from hardirq (IPI) context | |
409 | */ | |
410 | static void __hrtick_start(void *arg) | |
b328ca18 | 411 | { |
31656519 | 412 | struct rq *rq = arg; |
b328ca18 | 413 | |
05fa785c | 414 | raw_spin_lock(&rq->lock); |
971ee28c | 415 | __hrtick_restart(rq); |
31656519 | 416 | rq->hrtick_csd_pending = 0; |
05fa785c | 417 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
418 | } |
419 | ||
31656519 PZ |
420 | /* |
421 | * Called to set the hrtick timer state. | |
422 | * | |
423 | * called with rq->lock held and irqs disabled | |
424 | */ | |
029632fb | 425 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 426 | { |
31656519 PZ |
427 | struct hrtimer *timer = &rq->hrtick_timer; |
428 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 429 | |
cc584b21 | 430 | hrtimer_set_expires(timer, time); |
31656519 PZ |
431 | |
432 | if (rq == this_rq()) { | |
971ee28c | 433 | __hrtick_restart(rq); |
31656519 | 434 | } else if (!rq->hrtick_csd_pending) { |
6e275637 | 435 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
436 | rq->hrtick_csd_pending = 1; |
437 | } | |
b328ca18 PZ |
438 | } |
439 | ||
440 | static int | |
441 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
442 | { | |
443 | int cpu = (int)(long)hcpu; | |
444 | ||
445 | switch (action) { | |
446 | case CPU_UP_CANCELED: | |
447 | case CPU_UP_CANCELED_FROZEN: | |
448 | case CPU_DOWN_PREPARE: | |
449 | case CPU_DOWN_PREPARE_FROZEN: | |
450 | case CPU_DEAD: | |
451 | case CPU_DEAD_FROZEN: | |
31656519 | 452 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
453 | return NOTIFY_OK; |
454 | } | |
455 | ||
456 | return NOTIFY_DONE; | |
457 | } | |
458 | ||
fa748203 | 459 | static __init void init_hrtick(void) |
b328ca18 PZ |
460 | { |
461 | hotcpu_notifier(hotplug_hrtick, 0); | |
462 | } | |
31656519 PZ |
463 | #else |
464 | /* | |
465 | * Called to set the hrtick timer state. | |
466 | * | |
467 | * called with rq->lock held and irqs disabled | |
468 | */ | |
029632fb | 469 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 470 | { |
7f1e2ca9 | 471 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 472 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 473 | } |
b328ca18 | 474 | |
006c75f1 | 475 | static inline void init_hrtick(void) |
8f4d37ec | 476 | { |
8f4d37ec | 477 | } |
31656519 | 478 | #endif /* CONFIG_SMP */ |
8f4d37ec | 479 | |
31656519 | 480 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 481 | { |
31656519 PZ |
482 | #ifdef CONFIG_SMP |
483 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 484 | |
31656519 PZ |
485 | rq->hrtick_csd.flags = 0; |
486 | rq->hrtick_csd.func = __hrtick_start; | |
487 | rq->hrtick_csd.info = rq; | |
488 | #endif | |
8f4d37ec | 489 | |
31656519 PZ |
490 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
491 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 492 | } |
006c75f1 | 493 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
494 | static inline void hrtick_clear(struct rq *rq) |
495 | { | |
496 | } | |
497 | ||
8f4d37ec PZ |
498 | static inline void init_rq_hrtick(struct rq *rq) |
499 | { | |
500 | } | |
501 | ||
b328ca18 PZ |
502 | static inline void init_hrtick(void) |
503 | { | |
504 | } | |
006c75f1 | 505 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 506 | |
c24d20db IM |
507 | /* |
508 | * resched_task - mark a task 'to be rescheduled now'. | |
509 | * | |
510 | * On UP this means the setting of the need_resched flag, on SMP it | |
511 | * might also involve a cross-CPU call to trigger the scheduler on | |
512 | * the target CPU. | |
513 | */ | |
029632fb | 514 | void resched_task(struct task_struct *p) |
c24d20db IM |
515 | { |
516 | int cpu; | |
517 | ||
b021fe3e | 518 | lockdep_assert_held(&task_rq(p)->lock); |
c24d20db | 519 | |
5ed0cec0 | 520 | if (test_tsk_need_resched(p)) |
c24d20db IM |
521 | return; |
522 | ||
5ed0cec0 | 523 | set_tsk_need_resched(p); |
c24d20db IM |
524 | |
525 | cpu = task_cpu(p); | |
f27dde8d PZ |
526 | if (cpu == smp_processor_id()) { |
527 | set_preempt_need_resched(); | |
c24d20db | 528 | return; |
f27dde8d | 529 | } |
c24d20db IM |
530 | |
531 | /* NEED_RESCHED must be visible before we test polling */ | |
532 | smp_mb(); | |
533 | if (!tsk_is_polling(p)) | |
534 | smp_send_reschedule(cpu); | |
535 | } | |
536 | ||
029632fb | 537 | void resched_cpu(int cpu) |
c24d20db IM |
538 | { |
539 | struct rq *rq = cpu_rq(cpu); | |
540 | unsigned long flags; | |
541 | ||
05fa785c | 542 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
543 | return; |
544 | resched_task(cpu_curr(cpu)); | |
05fa785c | 545 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 546 | } |
06d8308c | 547 | |
b021fe3e | 548 | #ifdef CONFIG_SMP |
3451d024 | 549 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
550 | /* |
551 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
552 | * from an idle cpu. This is good for power-savings. | |
553 | * | |
554 | * We don't do similar optimization for completely idle system, as | |
555 | * selecting an idle cpu will add more delays to the timers than intended | |
556 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
557 | */ | |
558 | int get_nohz_timer_target(void) | |
559 | { | |
560 | int cpu = smp_processor_id(); | |
561 | int i; | |
562 | struct sched_domain *sd; | |
563 | ||
057f3fad | 564 | rcu_read_lock(); |
83cd4fe2 | 565 | for_each_domain(cpu, sd) { |
057f3fad PZ |
566 | for_each_cpu(i, sched_domain_span(sd)) { |
567 | if (!idle_cpu(i)) { | |
568 | cpu = i; | |
569 | goto unlock; | |
570 | } | |
571 | } | |
83cd4fe2 | 572 | } |
057f3fad PZ |
573 | unlock: |
574 | rcu_read_unlock(); | |
83cd4fe2 VP |
575 | return cpu; |
576 | } | |
06d8308c TG |
577 | /* |
578 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
579 | * idle CPU then this timer might expire before the next timer event | |
580 | * which is scheduled to wake up that CPU. In case of a completely | |
581 | * idle system the next event might even be infinite time into the | |
582 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
583 | * leaves the inner idle loop so the newly added timer is taken into | |
584 | * account when the CPU goes back to idle and evaluates the timer | |
585 | * wheel for the next timer event. | |
586 | */ | |
1c20091e | 587 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
588 | { |
589 | struct rq *rq = cpu_rq(cpu); | |
590 | ||
591 | if (cpu == smp_processor_id()) | |
592 | return; | |
593 | ||
594 | /* | |
595 | * This is safe, as this function is called with the timer | |
596 | * wheel base lock of (cpu) held. When the CPU is on the way | |
597 | * to idle and has not yet set rq->curr to idle then it will | |
598 | * be serialized on the timer wheel base lock and take the new | |
599 | * timer into account automatically. | |
600 | */ | |
601 | if (rq->curr != rq->idle) | |
602 | return; | |
45bf76df | 603 | |
45bf76df | 604 | /* |
06d8308c TG |
605 | * We can set TIF_RESCHED on the idle task of the other CPU |
606 | * lockless. The worst case is that the other CPU runs the | |
607 | * idle task through an additional NOOP schedule() | |
45bf76df | 608 | */ |
5ed0cec0 | 609 | set_tsk_need_resched(rq->idle); |
45bf76df | 610 | |
06d8308c TG |
611 | /* NEED_RESCHED must be visible before we test polling */ |
612 | smp_mb(); | |
613 | if (!tsk_is_polling(rq->idle)) | |
614 | smp_send_reschedule(cpu); | |
45bf76df IM |
615 | } |
616 | ||
c5bfece2 | 617 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 618 | { |
c5bfece2 | 619 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
620 | if (cpu != smp_processor_id() || |
621 | tick_nohz_tick_stopped()) | |
622 | smp_send_reschedule(cpu); | |
623 | return true; | |
624 | } | |
625 | ||
626 | return false; | |
627 | } | |
628 | ||
629 | void wake_up_nohz_cpu(int cpu) | |
630 | { | |
c5bfece2 | 631 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
632 | wake_up_idle_cpu(cpu); |
633 | } | |
634 | ||
ca38062e | 635 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 636 | { |
1c792db7 | 637 | int cpu = smp_processor_id(); |
873b4c65 VG |
638 | |
639 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
640 | return false; | |
641 | ||
642 | if (idle_cpu(cpu) && !need_resched()) | |
643 | return true; | |
644 | ||
645 | /* | |
646 | * We can't run Idle Load Balance on this CPU for this time so we | |
647 | * cancel it and clear NOHZ_BALANCE_KICK | |
648 | */ | |
649 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
650 | return false; | |
45bf76df IM |
651 | } |
652 | ||
3451d024 | 653 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 654 | |
ca38062e | 655 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 656 | { |
ca38062e | 657 | return false; |
2069dd75 PZ |
658 | } |
659 | ||
3451d024 | 660 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 661 | |
ce831b38 FW |
662 | #ifdef CONFIG_NO_HZ_FULL |
663 | bool sched_can_stop_tick(void) | |
664 | { | |
665 | struct rq *rq; | |
666 | ||
667 | rq = this_rq(); | |
668 | ||
669 | /* Make sure rq->nr_running update is visible after the IPI */ | |
670 | smp_rmb(); | |
671 | ||
672 | /* More than one running task need preemption */ | |
673 | if (rq->nr_running > 1) | |
674 | return false; | |
675 | ||
676 | return true; | |
677 | } | |
678 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 679 | |
029632fb | 680 | void sched_avg_update(struct rq *rq) |
18d95a28 | 681 | { |
e9e9250b PZ |
682 | s64 period = sched_avg_period(); |
683 | ||
78becc27 | 684 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
685 | /* |
686 | * Inline assembly required to prevent the compiler | |
687 | * optimising this loop into a divmod call. | |
688 | * See __iter_div_u64_rem() for another example of this. | |
689 | */ | |
690 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
691 | rq->age_stamp += period; |
692 | rq->rt_avg /= 2; | |
693 | } | |
18d95a28 PZ |
694 | } |
695 | ||
6d6bc0ad | 696 | #endif /* CONFIG_SMP */ |
18d95a28 | 697 | |
a790de99 PT |
698 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
699 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 700 | /* |
8277434e PT |
701 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
702 | * node and @up when leaving it for the final time. | |
703 | * | |
704 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 705 | */ |
029632fb | 706 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 707 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
708 | { |
709 | struct task_group *parent, *child; | |
eb755805 | 710 | int ret; |
c09595f6 | 711 | |
8277434e PT |
712 | parent = from; |
713 | ||
c09595f6 | 714 | down: |
eb755805 PZ |
715 | ret = (*down)(parent, data); |
716 | if (ret) | |
8277434e | 717 | goto out; |
c09595f6 PZ |
718 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
719 | parent = child; | |
720 | goto down; | |
721 | ||
722 | up: | |
723 | continue; | |
724 | } | |
eb755805 | 725 | ret = (*up)(parent, data); |
8277434e PT |
726 | if (ret || parent == from) |
727 | goto out; | |
c09595f6 PZ |
728 | |
729 | child = parent; | |
730 | parent = parent->parent; | |
731 | if (parent) | |
732 | goto up; | |
8277434e | 733 | out: |
eb755805 | 734 | return ret; |
c09595f6 PZ |
735 | } |
736 | ||
029632fb | 737 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 738 | { |
e2b245f8 | 739 | return 0; |
eb755805 | 740 | } |
18d95a28 PZ |
741 | #endif |
742 | ||
45bf76df IM |
743 | static void set_load_weight(struct task_struct *p) |
744 | { | |
f05998d4 NR |
745 | int prio = p->static_prio - MAX_RT_PRIO; |
746 | struct load_weight *load = &p->se.load; | |
747 | ||
dd41f596 IM |
748 | /* |
749 | * SCHED_IDLE tasks get minimal weight: | |
750 | */ | |
751 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 752 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 753 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
754 | return; |
755 | } | |
71f8bd46 | 756 | |
c8b28116 | 757 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 758 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
759 | } |
760 | ||
371fd7e7 | 761 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 762 | { |
a64692a3 | 763 | update_rq_clock(rq); |
43148951 | 764 | sched_info_queued(rq, p); |
371fd7e7 | 765 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
766 | } |
767 | ||
371fd7e7 | 768 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 769 | { |
a64692a3 | 770 | update_rq_clock(rq); |
43148951 | 771 | sched_info_dequeued(rq, p); |
371fd7e7 | 772 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
773 | } |
774 | ||
029632fb | 775 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
776 | { |
777 | if (task_contributes_to_load(p)) | |
778 | rq->nr_uninterruptible--; | |
779 | ||
371fd7e7 | 780 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
781 | } |
782 | ||
029632fb | 783 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
784 | { |
785 | if (task_contributes_to_load(p)) | |
786 | rq->nr_uninterruptible++; | |
787 | ||
371fd7e7 | 788 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
789 | } |
790 | ||
fe44d621 | 791 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 792 | { |
095c0aa8 GC |
793 | /* |
794 | * In theory, the compile should just see 0 here, and optimize out the call | |
795 | * to sched_rt_avg_update. But I don't trust it... | |
796 | */ | |
797 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
798 | s64 steal = 0, irq_delta = 0; | |
799 | #endif | |
800 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 801 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
802 | |
803 | /* | |
804 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
805 | * this case when a previous update_rq_clock() happened inside a | |
806 | * {soft,}irq region. | |
807 | * | |
808 | * When this happens, we stop ->clock_task and only update the | |
809 | * prev_irq_time stamp to account for the part that fit, so that a next | |
810 | * update will consume the rest. This ensures ->clock_task is | |
811 | * monotonic. | |
812 | * | |
813 | * It does however cause some slight miss-attribution of {soft,}irq | |
814 | * time, a more accurate solution would be to update the irq_time using | |
815 | * the current rq->clock timestamp, except that would require using | |
816 | * atomic ops. | |
817 | */ | |
818 | if (irq_delta > delta) | |
819 | irq_delta = delta; | |
820 | ||
821 | rq->prev_irq_time += irq_delta; | |
822 | delta -= irq_delta; | |
095c0aa8 GC |
823 | #endif |
824 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 825 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
826 | u64 st; |
827 | ||
828 | steal = paravirt_steal_clock(cpu_of(rq)); | |
829 | steal -= rq->prev_steal_time_rq; | |
830 | ||
831 | if (unlikely(steal > delta)) | |
832 | steal = delta; | |
833 | ||
834 | st = steal_ticks(steal); | |
835 | steal = st * TICK_NSEC; | |
836 | ||
837 | rq->prev_steal_time_rq += steal; | |
838 | ||
839 | delta -= steal; | |
840 | } | |
841 | #endif | |
842 | ||
fe44d621 PZ |
843 | rq->clock_task += delta; |
844 | ||
095c0aa8 GC |
845 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
846 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
847 | sched_rt_avg_update(rq, irq_delta + steal); | |
848 | #endif | |
aa483808 VP |
849 | } |
850 | ||
34f971f6 PZ |
851 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
852 | { | |
853 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
854 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
855 | ||
856 | if (stop) { | |
857 | /* | |
858 | * Make it appear like a SCHED_FIFO task, its something | |
859 | * userspace knows about and won't get confused about. | |
860 | * | |
861 | * Also, it will make PI more or less work without too | |
862 | * much confusion -- but then, stop work should not | |
863 | * rely on PI working anyway. | |
864 | */ | |
865 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
866 | ||
867 | stop->sched_class = &stop_sched_class; | |
868 | } | |
869 | ||
870 | cpu_rq(cpu)->stop = stop; | |
871 | ||
872 | if (old_stop) { | |
873 | /* | |
874 | * Reset it back to a normal scheduling class so that | |
875 | * it can die in pieces. | |
876 | */ | |
877 | old_stop->sched_class = &rt_sched_class; | |
878 | } | |
879 | } | |
880 | ||
14531189 | 881 | /* |
dd41f596 | 882 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 883 | */ |
14531189 IM |
884 | static inline int __normal_prio(struct task_struct *p) |
885 | { | |
dd41f596 | 886 | return p->static_prio; |
14531189 IM |
887 | } |
888 | ||
b29739f9 IM |
889 | /* |
890 | * Calculate the expected normal priority: i.e. priority | |
891 | * without taking RT-inheritance into account. Might be | |
892 | * boosted by interactivity modifiers. Changes upon fork, | |
893 | * setprio syscalls, and whenever the interactivity | |
894 | * estimator recalculates. | |
895 | */ | |
36c8b586 | 896 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
897 | { |
898 | int prio; | |
899 | ||
aab03e05 DF |
900 | if (task_has_dl_policy(p)) |
901 | prio = MAX_DL_PRIO-1; | |
902 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
903 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
904 | else | |
905 | prio = __normal_prio(p); | |
906 | return prio; | |
907 | } | |
908 | ||
909 | /* | |
910 | * Calculate the current priority, i.e. the priority | |
911 | * taken into account by the scheduler. This value might | |
912 | * be boosted by RT tasks, or might be boosted by | |
913 | * interactivity modifiers. Will be RT if the task got | |
914 | * RT-boosted. If not then it returns p->normal_prio. | |
915 | */ | |
36c8b586 | 916 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
917 | { |
918 | p->normal_prio = normal_prio(p); | |
919 | /* | |
920 | * If we are RT tasks or we were boosted to RT priority, | |
921 | * keep the priority unchanged. Otherwise, update priority | |
922 | * to the normal priority: | |
923 | */ | |
924 | if (!rt_prio(p->prio)) | |
925 | return p->normal_prio; | |
926 | return p->prio; | |
927 | } | |
928 | ||
1da177e4 LT |
929 | /** |
930 | * task_curr - is this task currently executing on a CPU? | |
931 | * @p: the task in question. | |
e69f6186 YB |
932 | * |
933 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 934 | */ |
36c8b586 | 935 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
936 | { |
937 | return cpu_curr(task_cpu(p)) == p; | |
938 | } | |
939 | ||
cb469845 SR |
940 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
941 | const struct sched_class *prev_class, | |
da7a735e | 942 | int oldprio) |
cb469845 SR |
943 | { |
944 | if (prev_class != p->sched_class) { | |
945 | if (prev_class->switched_from) | |
da7a735e PZ |
946 | prev_class->switched_from(rq, p); |
947 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 948 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 949 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
950 | } |
951 | ||
029632fb | 952 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
953 | { |
954 | const struct sched_class *class; | |
955 | ||
956 | if (p->sched_class == rq->curr->sched_class) { | |
957 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
958 | } else { | |
959 | for_each_class(class) { | |
960 | if (class == rq->curr->sched_class) | |
961 | break; | |
962 | if (class == p->sched_class) { | |
963 | resched_task(rq->curr); | |
964 | break; | |
965 | } | |
966 | } | |
967 | } | |
968 | ||
969 | /* | |
970 | * A queue event has occurred, and we're going to schedule. In | |
971 | * this case, we can save a useless back to back clock update. | |
972 | */ | |
fd2f4419 | 973 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
974 | rq->skip_clock_update = 1; |
975 | } | |
976 | ||
1da177e4 | 977 | #ifdef CONFIG_SMP |
dd41f596 | 978 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 979 | { |
e2912009 PZ |
980 | #ifdef CONFIG_SCHED_DEBUG |
981 | /* | |
982 | * We should never call set_task_cpu() on a blocked task, | |
983 | * ttwu() will sort out the placement. | |
984 | */ | |
077614ee | 985 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 986 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
987 | |
988 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
989 | /* |
990 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
991 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
992 | * | |
993 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 994 | * see task_group(). |
6c6c54e1 PZ |
995 | * |
996 | * Furthermore, all task_rq users should acquire both locks, see | |
997 | * task_rq_lock(). | |
998 | */ | |
0122ec5b PZ |
999 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1000 | lockdep_is_held(&task_rq(p)->lock))); | |
1001 | #endif | |
e2912009 PZ |
1002 | #endif |
1003 | ||
de1d7286 | 1004 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1005 | |
0c69774e | 1006 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1007 | if (p->sched_class->migrate_task_rq) |
1008 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1009 | p->se.nr_migrations++; |
a8b0ca17 | 1010 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1011 | } |
dd41f596 IM |
1012 | |
1013 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1014 | } |
1015 | ||
ac66f547 PZ |
1016 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1017 | { | |
1018 | if (p->on_rq) { | |
1019 | struct rq *src_rq, *dst_rq; | |
1020 | ||
1021 | src_rq = task_rq(p); | |
1022 | dst_rq = cpu_rq(cpu); | |
1023 | ||
1024 | deactivate_task(src_rq, p, 0); | |
1025 | set_task_cpu(p, cpu); | |
1026 | activate_task(dst_rq, p, 0); | |
1027 | check_preempt_curr(dst_rq, p, 0); | |
1028 | } else { | |
1029 | /* | |
1030 | * Task isn't running anymore; make it appear like we migrated | |
1031 | * it before it went to sleep. This means on wakeup we make the | |
1032 | * previous cpu our targer instead of where it really is. | |
1033 | */ | |
1034 | p->wake_cpu = cpu; | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | struct migration_swap_arg { | |
1039 | struct task_struct *src_task, *dst_task; | |
1040 | int src_cpu, dst_cpu; | |
1041 | }; | |
1042 | ||
1043 | static int migrate_swap_stop(void *data) | |
1044 | { | |
1045 | struct migration_swap_arg *arg = data; | |
1046 | struct rq *src_rq, *dst_rq; | |
1047 | int ret = -EAGAIN; | |
1048 | ||
1049 | src_rq = cpu_rq(arg->src_cpu); | |
1050 | dst_rq = cpu_rq(arg->dst_cpu); | |
1051 | ||
74602315 PZ |
1052 | double_raw_lock(&arg->src_task->pi_lock, |
1053 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1054 | double_rq_lock(src_rq, dst_rq); |
1055 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1056 | goto unlock; | |
1057 | ||
1058 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1059 | goto unlock; | |
1060 | ||
1061 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1062 | goto unlock; | |
1063 | ||
1064 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1065 | goto unlock; | |
1066 | ||
1067 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1068 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1069 | ||
1070 | ret = 0; | |
1071 | ||
1072 | unlock: | |
1073 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1074 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1075 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1076 | |
1077 | return ret; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Cross migrate two tasks | |
1082 | */ | |
1083 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1084 | { | |
1085 | struct migration_swap_arg arg; | |
1086 | int ret = -EINVAL; | |
1087 | ||
ac66f547 PZ |
1088 | arg = (struct migration_swap_arg){ |
1089 | .src_task = cur, | |
1090 | .src_cpu = task_cpu(cur), | |
1091 | .dst_task = p, | |
1092 | .dst_cpu = task_cpu(p), | |
1093 | }; | |
1094 | ||
1095 | if (arg.src_cpu == arg.dst_cpu) | |
1096 | goto out; | |
1097 | ||
6acce3ef PZ |
1098 | /* |
1099 | * These three tests are all lockless; this is OK since all of them | |
1100 | * will be re-checked with proper locks held further down the line. | |
1101 | */ | |
ac66f547 PZ |
1102 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1103 | goto out; | |
1104 | ||
1105 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1106 | goto out; | |
1107 | ||
1108 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1109 | goto out; | |
1110 | ||
286549dc | 1111 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1112 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1113 | ||
1114 | out: | |
ac66f547 PZ |
1115 | return ret; |
1116 | } | |
1117 | ||
969c7921 | 1118 | struct migration_arg { |
36c8b586 | 1119 | struct task_struct *task; |
1da177e4 | 1120 | int dest_cpu; |
70b97a7f | 1121 | }; |
1da177e4 | 1122 | |
969c7921 TH |
1123 | static int migration_cpu_stop(void *data); |
1124 | ||
1da177e4 LT |
1125 | /* |
1126 | * wait_task_inactive - wait for a thread to unschedule. | |
1127 | * | |
85ba2d86 RM |
1128 | * If @match_state is nonzero, it's the @p->state value just checked and |
1129 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1130 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1131 | * we return a positive number (its total switch count). If a second call | |
1132 | * a short while later returns the same number, the caller can be sure that | |
1133 | * @p has remained unscheduled the whole time. | |
1134 | * | |
1da177e4 LT |
1135 | * The caller must ensure that the task *will* unschedule sometime soon, |
1136 | * else this function might spin for a *long* time. This function can't | |
1137 | * be called with interrupts off, or it may introduce deadlock with | |
1138 | * smp_call_function() if an IPI is sent by the same process we are | |
1139 | * waiting to become inactive. | |
1140 | */ | |
85ba2d86 | 1141 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1142 | { |
1143 | unsigned long flags; | |
dd41f596 | 1144 | int running, on_rq; |
85ba2d86 | 1145 | unsigned long ncsw; |
70b97a7f | 1146 | struct rq *rq; |
1da177e4 | 1147 | |
3a5c359a AK |
1148 | for (;;) { |
1149 | /* | |
1150 | * We do the initial early heuristics without holding | |
1151 | * any task-queue locks at all. We'll only try to get | |
1152 | * the runqueue lock when things look like they will | |
1153 | * work out! | |
1154 | */ | |
1155 | rq = task_rq(p); | |
fa490cfd | 1156 | |
3a5c359a AK |
1157 | /* |
1158 | * If the task is actively running on another CPU | |
1159 | * still, just relax and busy-wait without holding | |
1160 | * any locks. | |
1161 | * | |
1162 | * NOTE! Since we don't hold any locks, it's not | |
1163 | * even sure that "rq" stays as the right runqueue! | |
1164 | * But we don't care, since "task_running()" will | |
1165 | * return false if the runqueue has changed and p | |
1166 | * is actually now running somewhere else! | |
1167 | */ | |
85ba2d86 RM |
1168 | while (task_running(rq, p)) { |
1169 | if (match_state && unlikely(p->state != match_state)) | |
1170 | return 0; | |
3a5c359a | 1171 | cpu_relax(); |
85ba2d86 | 1172 | } |
fa490cfd | 1173 | |
3a5c359a AK |
1174 | /* |
1175 | * Ok, time to look more closely! We need the rq | |
1176 | * lock now, to be *sure*. If we're wrong, we'll | |
1177 | * just go back and repeat. | |
1178 | */ | |
1179 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1180 | trace_sched_wait_task(p); |
3a5c359a | 1181 | running = task_running(rq, p); |
fd2f4419 | 1182 | on_rq = p->on_rq; |
85ba2d86 | 1183 | ncsw = 0; |
f31e11d8 | 1184 | if (!match_state || p->state == match_state) |
93dcf55f | 1185 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1186 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1187 | |
85ba2d86 RM |
1188 | /* |
1189 | * If it changed from the expected state, bail out now. | |
1190 | */ | |
1191 | if (unlikely(!ncsw)) | |
1192 | break; | |
1193 | ||
3a5c359a AK |
1194 | /* |
1195 | * Was it really running after all now that we | |
1196 | * checked with the proper locks actually held? | |
1197 | * | |
1198 | * Oops. Go back and try again.. | |
1199 | */ | |
1200 | if (unlikely(running)) { | |
1201 | cpu_relax(); | |
1202 | continue; | |
1203 | } | |
fa490cfd | 1204 | |
3a5c359a AK |
1205 | /* |
1206 | * It's not enough that it's not actively running, | |
1207 | * it must be off the runqueue _entirely_, and not | |
1208 | * preempted! | |
1209 | * | |
80dd99b3 | 1210 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1211 | * running right now), it's preempted, and we should |
1212 | * yield - it could be a while. | |
1213 | */ | |
1214 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1215 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1216 | ||
1217 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1218 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1219 | continue; |
1220 | } | |
fa490cfd | 1221 | |
3a5c359a AK |
1222 | /* |
1223 | * Ahh, all good. It wasn't running, and it wasn't | |
1224 | * runnable, which means that it will never become | |
1225 | * running in the future either. We're all done! | |
1226 | */ | |
1227 | break; | |
1228 | } | |
85ba2d86 RM |
1229 | |
1230 | return ncsw; | |
1da177e4 LT |
1231 | } |
1232 | ||
1233 | /*** | |
1234 | * kick_process - kick a running thread to enter/exit the kernel | |
1235 | * @p: the to-be-kicked thread | |
1236 | * | |
1237 | * Cause a process which is running on another CPU to enter | |
1238 | * kernel-mode, without any delay. (to get signals handled.) | |
1239 | * | |
25985edc | 1240 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1241 | * because all it wants to ensure is that the remote task enters |
1242 | * the kernel. If the IPI races and the task has been migrated | |
1243 | * to another CPU then no harm is done and the purpose has been | |
1244 | * achieved as well. | |
1245 | */ | |
36c8b586 | 1246 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1247 | { |
1248 | int cpu; | |
1249 | ||
1250 | preempt_disable(); | |
1251 | cpu = task_cpu(p); | |
1252 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1253 | smp_send_reschedule(cpu); | |
1254 | preempt_enable(); | |
1255 | } | |
b43e3521 | 1256 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1257 | #endif /* CONFIG_SMP */ |
1da177e4 | 1258 | |
970b13ba | 1259 | #ifdef CONFIG_SMP |
30da688e | 1260 | /* |
013fdb80 | 1261 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1262 | */ |
5da9a0fb PZ |
1263 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1264 | { | |
aa00d89c TC |
1265 | int nid = cpu_to_node(cpu); |
1266 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1267 | enum { cpuset, possible, fail } state = cpuset; |
1268 | int dest_cpu; | |
5da9a0fb | 1269 | |
aa00d89c TC |
1270 | /* |
1271 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1272 | * will return -1. There is no cpu on the node, and we should | |
1273 | * select the cpu on the other node. | |
1274 | */ | |
1275 | if (nid != -1) { | |
1276 | nodemask = cpumask_of_node(nid); | |
1277 | ||
1278 | /* Look for allowed, online CPU in same node. */ | |
1279 | for_each_cpu(dest_cpu, nodemask) { | |
1280 | if (!cpu_online(dest_cpu)) | |
1281 | continue; | |
1282 | if (!cpu_active(dest_cpu)) | |
1283 | continue; | |
1284 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1285 | return dest_cpu; | |
1286 | } | |
2baab4e9 | 1287 | } |
5da9a0fb | 1288 | |
2baab4e9 PZ |
1289 | for (;;) { |
1290 | /* Any allowed, online CPU? */ | |
e3831edd | 1291 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1292 | if (!cpu_online(dest_cpu)) |
1293 | continue; | |
1294 | if (!cpu_active(dest_cpu)) | |
1295 | continue; | |
1296 | goto out; | |
1297 | } | |
5da9a0fb | 1298 | |
2baab4e9 PZ |
1299 | switch (state) { |
1300 | case cpuset: | |
1301 | /* No more Mr. Nice Guy. */ | |
1302 | cpuset_cpus_allowed_fallback(p); | |
1303 | state = possible; | |
1304 | break; | |
1305 | ||
1306 | case possible: | |
1307 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1308 | state = fail; | |
1309 | break; | |
1310 | ||
1311 | case fail: | |
1312 | BUG(); | |
1313 | break; | |
1314 | } | |
1315 | } | |
1316 | ||
1317 | out: | |
1318 | if (state != cpuset) { | |
1319 | /* | |
1320 | * Don't tell them about moving exiting tasks or | |
1321 | * kernel threads (both mm NULL), since they never | |
1322 | * leave kernel. | |
1323 | */ | |
1324 | if (p->mm && printk_ratelimit()) { | |
1325 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1326 | task_pid_nr(p), p->comm, cpu); | |
1327 | } | |
5da9a0fb PZ |
1328 | } |
1329 | ||
1330 | return dest_cpu; | |
1331 | } | |
1332 | ||
e2912009 | 1333 | /* |
013fdb80 | 1334 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1335 | */ |
970b13ba | 1336 | static inline |
ac66f547 | 1337 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1338 | { |
ac66f547 | 1339 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1340 | |
1341 | /* | |
1342 | * In order not to call set_task_cpu() on a blocking task we need | |
1343 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1344 | * cpu. | |
1345 | * | |
1346 | * Since this is common to all placement strategies, this lives here. | |
1347 | * | |
1348 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1349 | * not worry about this generic constraint ] | |
1350 | */ | |
fa17b507 | 1351 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1352 | !cpu_online(cpu))) |
5da9a0fb | 1353 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1354 | |
1355 | return cpu; | |
970b13ba | 1356 | } |
09a40af5 MG |
1357 | |
1358 | static void update_avg(u64 *avg, u64 sample) | |
1359 | { | |
1360 | s64 diff = sample - *avg; | |
1361 | *avg += diff >> 3; | |
1362 | } | |
970b13ba PZ |
1363 | #endif |
1364 | ||
d7c01d27 | 1365 | static void |
b84cb5df | 1366 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1367 | { |
d7c01d27 | 1368 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1369 | struct rq *rq = this_rq(); |
1370 | ||
d7c01d27 PZ |
1371 | #ifdef CONFIG_SMP |
1372 | int this_cpu = smp_processor_id(); | |
1373 | ||
1374 | if (cpu == this_cpu) { | |
1375 | schedstat_inc(rq, ttwu_local); | |
1376 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1377 | } else { | |
1378 | struct sched_domain *sd; | |
1379 | ||
1380 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1381 | rcu_read_lock(); |
d7c01d27 PZ |
1382 | for_each_domain(this_cpu, sd) { |
1383 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1384 | schedstat_inc(sd, ttwu_wake_remote); | |
1385 | break; | |
1386 | } | |
1387 | } | |
057f3fad | 1388 | rcu_read_unlock(); |
d7c01d27 | 1389 | } |
f339b9dc PZ |
1390 | |
1391 | if (wake_flags & WF_MIGRATED) | |
1392 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1393 | ||
d7c01d27 PZ |
1394 | #endif /* CONFIG_SMP */ |
1395 | ||
1396 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1397 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1398 | |
1399 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1400 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1401 | |
d7c01d27 PZ |
1402 | #endif /* CONFIG_SCHEDSTATS */ |
1403 | } | |
1404 | ||
1405 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1406 | { | |
9ed3811a | 1407 | activate_task(rq, p, en_flags); |
fd2f4419 | 1408 | p->on_rq = 1; |
c2f7115e PZ |
1409 | |
1410 | /* if a worker is waking up, notify workqueue */ | |
1411 | if (p->flags & PF_WQ_WORKER) | |
1412 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1413 | } |
1414 | ||
23f41eeb PZ |
1415 | /* |
1416 | * Mark the task runnable and perform wakeup-preemption. | |
1417 | */ | |
89363381 | 1418 | static void |
23f41eeb | 1419 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1420 | { |
9ed3811a | 1421 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1422 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1423 | |
1424 | p->state = TASK_RUNNING; | |
1425 | #ifdef CONFIG_SMP | |
1426 | if (p->sched_class->task_woken) | |
1427 | p->sched_class->task_woken(rq, p); | |
1428 | ||
e69c6341 | 1429 | if (rq->idle_stamp) { |
78becc27 | 1430 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1431 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1432 | |
abfafa54 JL |
1433 | update_avg(&rq->avg_idle, delta); |
1434 | ||
1435 | if (rq->avg_idle > max) | |
9ed3811a | 1436 | rq->avg_idle = max; |
abfafa54 | 1437 | |
9ed3811a TH |
1438 | rq->idle_stamp = 0; |
1439 | } | |
1440 | #endif | |
1441 | } | |
1442 | ||
c05fbafb PZ |
1443 | static void |
1444 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1445 | { | |
1446 | #ifdef CONFIG_SMP | |
1447 | if (p->sched_contributes_to_load) | |
1448 | rq->nr_uninterruptible--; | |
1449 | #endif | |
1450 | ||
1451 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1452 | ttwu_do_wakeup(rq, p, wake_flags); | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1457 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1458 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1459 | * the task is still ->on_rq. | |
1460 | */ | |
1461 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1462 | { | |
1463 | struct rq *rq; | |
1464 | int ret = 0; | |
1465 | ||
1466 | rq = __task_rq_lock(p); | |
1467 | if (p->on_rq) { | |
1ad4ec0d FW |
1468 | /* check_preempt_curr() may use rq clock */ |
1469 | update_rq_clock(rq); | |
c05fbafb PZ |
1470 | ttwu_do_wakeup(rq, p, wake_flags); |
1471 | ret = 1; | |
1472 | } | |
1473 | __task_rq_unlock(rq); | |
1474 | ||
1475 | return ret; | |
1476 | } | |
1477 | ||
317f3941 | 1478 | #ifdef CONFIG_SMP |
fa14ff4a | 1479 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1480 | { |
1481 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1482 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1483 | struct task_struct *p; | |
317f3941 PZ |
1484 | |
1485 | raw_spin_lock(&rq->lock); | |
1486 | ||
fa14ff4a PZ |
1487 | while (llist) { |
1488 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1489 | llist = llist_next(llist); | |
317f3941 PZ |
1490 | ttwu_do_activate(rq, p, 0); |
1491 | } | |
1492 | ||
1493 | raw_spin_unlock(&rq->lock); | |
1494 | } | |
1495 | ||
1496 | void scheduler_ipi(void) | |
1497 | { | |
f27dde8d PZ |
1498 | /* |
1499 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1500 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1501 | * this IPI. | |
1502 | */ | |
8cb75e0c | 1503 | preempt_fold_need_resched(); |
f27dde8d | 1504 | |
873b4c65 VG |
1505 | if (llist_empty(&this_rq()->wake_list) |
1506 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1507 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1508 | return; |
1509 | ||
1510 | /* | |
1511 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1512 | * traditionally all their work was done from the interrupt return | |
1513 | * path. Now that we actually do some work, we need to make sure | |
1514 | * we do call them. | |
1515 | * | |
1516 | * Some archs already do call them, luckily irq_enter/exit nest | |
1517 | * properly. | |
1518 | * | |
1519 | * Arguably we should visit all archs and update all handlers, | |
1520 | * however a fair share of IPIs are still resched only so this would | |
1521 | * somewhat pessimize the simple resched case. | |
1522 | */ | |
1523 | irq_enter(); | |
ff442c51 | 1524 | tick_nohz_full_check(); |
fa14ff4a | 1525 | sched_ttwu_pending(); |
ca38062e SS |
1526 | |
1527 | /* | |
1528 | * Check if someone kicked us for doing the nohz idle load balance. | |
1529 | */ | |
873b4c65 | 1530 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1531 | this_rq()->idle_balance = 1; |
ca38062e | 1532 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1533 | } |
c5d753a5 | 1534 | irq_exit(); |
317f3941 PZ |
1535 | } |
1536 | ||
1537 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1538 | { | |
fa14ff4a | 1539 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1540 | smp_send_reschedule(cpu); |
1541 | } | |
d6aa8f85 | 1542 | |
39be3501 | 1543 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1544 | { |
1545 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1546 | } | |
d6aa8f85 | 1547 | #endif /* CONFIG_SMP */ |
317f3941 | 1548 | |
c05fbafb PZ |
1549 | static void ttwu_queue(struct task_struct *p, int cpu) |
1550 | { | |
1551 | struct rq *rq = cpu_rq(cpu); | |
1552 | ||
17d9f311 | 1553 | #if defined(CONFIG_SMP) |
39be3501 | 1554 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1555 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1556 | ttwu_queue_remote(p, cpu); |
1557 | return; | |
1558 | } | |
1559 | #endif | |
1560 | ||
c05fbafb PZ |
1561 | raw_spin_lock(&rq->lock); |
1562 | ttwu_do_activate(rq, p, 0); | |
1563 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1564 | } |
1565 | ||
1566 | /** | |
1da177e4 | 1567 | * try_to_wake_up - wake up a thread |
9ed3811a | 1568 | * @p: the thread to be awakened |
1da177e4 | 1569 | * @state: the mask of task states that can be woken |
9ed3811a | 1570 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1571 | * |
1572 | * Put it on the run-queue if it's not already there. The "current" | |
1573 | * thread is always on the run-queue (except when the actual | |
1574 | * re-schedule is in progress), and as such you're allowed to do | |
1575 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1576 | * runnable without the overhead of this. | |
1577 | * | |
e69f6186 | 1578 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1579 | * or @state didn't match @p's state. |
1da177e4 | 1580 | */ |
e4a52bcb PZ |
1581 | static int |
1582 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1583 | { |
1da177e4 | 1584 | unsigned long flags; |
c05fbafb | 1585 | int cpu, success = 0; |
2398f2c6 | 1586 | |
e0acd0a6 ON |
1587 | /* |
1588 | * If we are going to wake up a thread waiting for CONDITION we | |
1589 | * need to ensure that CONDITION=1 done by the caller can not be | |
1590 | * reordered with p->state check below. This pairs with mb() in | |
1591 | * set_current_state() the waiting thread does. | |
1592 | */ | |
1593 | smp_mb__before_spinlock(); | |
013fdb80 | 1594 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1595 | if (!(p->state & state)) |
1da177e4 LT |
1596 | goto out; |
1597 | ||
c05fbafb | 1598 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1599 | cpu = task_cpu(p); |
1da177e4 | 1600 | |
c05fbafb PZ |
1601 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1602 | goto stat; | |
1da177e4 | 1603 | |
1da177e4 | 1604 | #ifdef CONFIG_SMP |
e9c84311 | 1605 | /* |
c05fbafb PZ |
1606 | * If the owning (remote) cpu is still in the middle of schedule() with |
1607 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1608 | */ |
f3e94786 | 1609 | while (p->on_cpu) |
e4a52bcb | 1610 | cpu_relax(); |
0970d299 | 1611 | /* |
e4a52bcb | 1612 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1613 | */ |
e4a52bcb | 1614 | smp_rmb(); |
1da177e4 | 1615 | |
a8e4f2ea | 1616 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1617 | p->state = TASK_WAKING; |
e7693a36 | 1618 | |
e4a52bcb | 1619 | if (p->sched_class->task_waking) |
74f8e4b2 | 1620 | p->sched_class->task_waking(p); |
efbbd05a | 1621 | |
ac66f547 | 1622 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1623 | if (task_cpu(p) != cpu) { |
1624 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1625 | set_task_cpu(p, cpu); |
f339b9dc | 1626 | } |
1da177e4 | 1627 | #endif /* CONFIG_SMP */ |
1da177e4 | 1628 | |
c05fbafb PZ |
1629 | ttwu_queue(p, cpu); |
1630 | stat: | |
b84cb5df | 1631 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1632 | out: |
013fdb80 | 1633 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1634 | |
1635 | return success; | |
1636 | } | |
1637 | ||
21aa9af0 TH |
1638 | /** |
1639 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1640 | * @p: the thread to be awakened | |
1641 | * | |
2acca55e | 1642 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1643 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1644 | * the current task. |
21aa9af0 TH |
1645 | */ |
1646 | static void try_to_wake_up_local(struct task_struct *p) | |
1647 | { | |
1648 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1649 | |
383efcd0 TH |
1650 | if (WARN_ON_ONCE(rq != this_rq()) || |
1651 | WARN_ON_ONCE(p == current)) | |
1652 | return; | |
1653 | ||
21aa9af0 TH |
1654 | lockdep_assert_held(&rq->lock); |
1655 | ||
2acca55e PZ |
1656 | if (!raw_spin_trylock(&p->pi_lock)) { |
1657 | raw_spin_unlock(&rq->lock); | |
1658 | raw_spin_lock(&p->pi_lock); | |
1659 | raw_spin_lock(&rq->lock); | |
1660 | } | |
1661 | ||
21aa9af0 | 1662 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1663 | goto out; |
21aa9af0 | 1664 | |
fd2f4419 | 1665 | if (!p->on_rq) |
d7c01d27 PZ |
1666 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1667 | ||
23f41eeb | 1668 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1669 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1670 | out: |
1671 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1672 | } |
1673 | ||
50fa610a DH |
1674 | /** |
1675 | * wake_up_process - Wake up a specific process | |
1676 | * @p: The process to be woken up. | |
1677 | * | |
1678 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1679 | * processes. |
1680 | * | |
1681 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1682 | * |
1683 | * It may be assumed that this function implies a write memory barrier before | |
1684 | * changing the task state if and only if any tasks are woken up. | |
1685 | */ | |
7ad5b3a5 | 1686 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1687 | { |
9067ac85 ON |
1688 | WARN_ON(task_is_stopped_or_traced(p)); |
1689 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1690 | } |
1da177e4 LT |
1691 | EXPORT_SYMBOL(wake_up_process); |
1692 | ||
7ad5b3a5 | 1693 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1694 | { |
1695 | return try_to_wake_up(p, state, 0); | |
1696 | } | |
1697 | ||
1da177e4 LT |
1698 | /* |
1699 | * Perform scheduler related setup for a newly forked process p. | |
1700 | * p is forked by current. | |
dd41f596 IM |
1701 | * |
1702 | * __sched_fork() is basic setup used by init_idle() too: | |
1703 | */ | |
5e1576ed | 1704 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1705 | { |
fd2f4419 PZ |
1706 | p->on_rq = 0; |
1707 | ||
1708 | p->se.on_rq = 0; | |
dd41f596 IM |
1709 | p->se.exec_start = 0; |
1710 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1711 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1712 | p->se.nr_migrations = 0; |
da7a735e | 1713 | p->se.vruntime = 0; |
fd2f4419 | 1714 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1715 | |
1716 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1717 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1718 | #endif |
476d139c | 1719 | |
aab03e05 DF |
1720 | RB_CLEAR_NODE(&p->dl.rb_node); |
1721 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1722 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1723 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1724 | p->dl.dl_period = 0; |
aab03e05 DF |
1725 | p->dl.flags = 0; |
1726 | ||
fa717060 | 1727 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1728 | |
e107be36 AK |
1729 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1730 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1731 | #endif | |
cbee9f88 PZ |
1732 | |
1733 | #ifdef CONFIG_NUMA_BALANCING | |
1734 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1735 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1736 | p->mm->numa_scan_seq = 0; |
1737 | } | |
1738 | ||
5e1576ed RR |
1739 | if (clone_flags & CLONE_VM) |
1740 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1741 | else | |
1742 | p->numa_preferred_nid = -1; | |
1743 | ||
cbee9f88 PZ |
1744 | p->node_stamp = 0ULL; |
1745 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1746 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1747 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1748 | p->numa_faults_memory = NULL; |
1749 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1750 | p->last_task_numa_placement = 0; |
1751 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1752 | |
1753 | INIT_LIST_HEAD(&p->numa_entry); | |
1754 | p->numa_group = NULL; | |
cbee9f88 | 1755 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1756 | } |
1757 | ||
1a687c2e | 1758 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1759 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1760 | void set_numabalancing_state(bool enabled) |
1761 | { | |
1762 | if (enabled) | |
1763 | sched_feat_set("NUMA"); | |
1764 | else | |
1765 | sched_feat_set("NO_NUMA"); | |
1766 | } | |
3105b86a MG |
1767 | #else |
1768 | __read_mostly bool numabalancing_enabled; | |
1769 | ||
1770 | void set_numabalancing_state(bool enabled) | |
1771 | { | |
1772 | numabalancing_enabled = enabled; | |
dd41f596 | 1773 | } |
3105b86a | 1774 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1775 | |
1776 | #ifdef CONFIG_PROC_SYSCTL | |
1777 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1778 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1779 | { | |
1780 | struct ctl_table t; | |
1781 | int err; | |
1782 | int state = numabalancing_enabled; | |
1783 | ||
1784 | if (write && !capable(CAP_SYS_ADMIN)) | |
1785 | return -EPERM; | |
1786 | ||
1787 | t = *table; | |
1788 | t.data = &state; | |
1789 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1790 | if (err < 0) | |
1791 | return err; | |
1792 | if (write) | |
1793 | set_numabalancing_state(state); | |
1794 | return err; | |
1795 | } | |
1796 | #endif | |
1797 | #endif | |
dd41f596 IM |
1798 | |
1799 | /* | |
1800 | * fork()/clone()-time setup: | |
1801 | */ | |
aab03e05 | 1802 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1803 | { |
0122ec5b | 1804 | unsigned long flags; |
dd41f596 IM |
1805 | int cpu = get_cpu(); |
1806 | ||
5e1576ed | 1807 | __sched_fork(clone_flags, p); |
06b83b5f | 1808 | /* |
0017d735 | 1809 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1810 | * nobody will actually run it, and a signal or other external |
1811 | * event cannot wake it up and insert it on the runqueue either. | |
1812 | */ | |
0017d735 | 1813 | p->state = TASK_RUNNING; |
dd41f596 | 1814 | |
c350a04e MG |
1815 | /* |
1816 | * Make sure we do not leak PI boosting priority to the child. | |
1817 | */ | |
1818 | p->prio = current->normal_prio; | |
1819 | ||
b9dc29e7 MG |
1820 | /* |
1821 | * Revert to default priority/policy on fork if requested. | |
1822 | */ | |
1823 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1824 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1825 | p->policy = SCHED_NORMAL; |
6c697bdf | 1826 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1827 | p->rt_priority = 0; |
1828 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1829 | p->static_prio = NICE_TO_PRIO(0); | |
1830 | ||
1831 | p->prio = p->normal_prio = __normal_prio(p); | |
1832 | set_load_weight(p); | |
6c697bdf | 1833 | |
b9dc29e7 MG |
1834 | /* |
1835 | * We don't need the reset flag anymore after the fork. It has | |
1836 | * fulfilled its duty: | |
1837 | */ | |
1838 | p->sched_reset_on_fork = 0; | |
1839 | } | |
ca94c442 | 1840 | |
aab03e05 DF |
1841 | if (dl_prio(p->prio)) { |
1842 | put_cpu(); | |
1843 | return -EAGAIN; | |
1844 | } else if (rt_prio(p->prio)) { | |
1845 | p->sched_class = &rt_sched_class; | |
1846 | } else { | |
2ddbf952 | 1847 | p->sched_class = &fair_sched_class; |
aab03e05 | 1848 | } |
b29739f9 | 1849 | |
cd29fe6f PZ |
1850 | if (p->sched_class->task_fork) |
1851 | p->sched_class->task_fork(p); | |
1852 | ||
86951599 PZ |
1853 | /* |
1854 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1855 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1856 | * is ran before sched_fork(). | |
1857 | * | |
1858 | * Silence PROVE_RCU. | |
1859 | */ | |
0122ec5b | 1860 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1861 | set_task_cpu(p, cpu); |
0122ec5b | 1862 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1863 | |
52f17b6c | 1864 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1865 | if (likely(sched_info_on())) |
52f17b6c | 1866 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1867 | #endif |
3ca7a440 PZ |
1868 | #if defined(CONFIG_SMP) |
1869 | p->on_cpu = 0; | |
4866cde0 | 1870 | #endif |
01028747 | 1871 | init_task_preempt_count(p); |
806c09a7 | 1872 | #ifdef CONFIG_SMP |
917b627d | 1873 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1874 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1875 | #endif |
917b627d | 1876 | |
476d139c | 1877 | put_cpu(); |
aab03e05 | 1878 | return 0; |
1da177e4 LT |
1879 | } |
1880 | ||
332ac17e DF |
1881 | unsigned long to_ratio(u64 period, u64 runtime) |
1882 | { | |
1883 | if (runtime == RUNTIME_INF) | |
1884 | return 1ULL << 20; | |
1885 | ||
1886 | /* | |
1887 | * Doing this here saves a lot of checks in all | |
1888 | * the calling paths, and returning zero seems | |
1889 | * safe for them anyway. | |
1890 | */ | |
1891 | if (period == 0) | |
1892 | return 0; | |
1893 | ||
1894 | return div64_u64(runtime << 20, period); | |
1895 | } | |
1896 | ||
1897 | #ifdef CONFIG_SMP | |
1898 | inline struct dl_bw *dl_bw_of(int i) | |
1899 | { | |
1900 | return &cpu_rq(i)->rd->dl_bw; | |
1901 | } | |
1902 | ||
de212f18 | 1903 | static inline int dl_bw_cpus(int i) |
332ac17e | 1904 | { |
de212f18 PZ |
1905 | struct root_domain *rd = cpu_rq(i)->rd; |
1906 | int cpus = 0; | |
1907 | ||
1908 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1909 | cpus++; | |
1910 | ||
1911 | return cpus; | |
332ac17e DF |
1912 | } |
1913 | #else | |
1914 | inline struct dl_bw *dl_bw_of(int i) | |
1915 | { | |
1916 | return &cpu_rq(i)->dl.dl_bw; | |
1917 | } | |
1918 | ||
de212f18 | 1919 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
1920 | { |
1921 | return 1; | |
1922 | } | |
1923 | #endif | |
1924 | ||
1925 | static inline | |
1926 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
1927 | { | |
1928 | dl_b->total_bw -= tsk_bw; | |
1929 | } | |
1930 | ||
1931 | static inline | |
1932 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
1933 | { | |
1934 | dl_b->total_bw += tsk_bw; | |
1935 | } | |
1936 | ||
1937 | static inline | |
1938 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
1939 | { | |
1940 | return dl_b->bw != -1 && | |
1941 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * We must be sure that accepting a new task (or allowing changing the | |
1946 | * parameters of an existing one) is consistent with the bandwidth | |
1947 | * constraints. If yes, this function also accordingly updates the currently | |
1948 | * allocated bandwidth to reflect the new situation. | |
1949 | * | |
1950 | * This function is called while holding p's rq->lock. | |
1951 | */ | |
1952 | static int dl_overflow(struct task_struct *p, int policy, | |
1953 | const struct sched_attr *attr) | |
1954 | { | |
1955 | ||
1956 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
1957 | u64 period = attr->sched_period; | |
1958 | u64 runtime = attr->sched_runtime; | |
1959 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 1960 | int cpus, err = -1; |
332ac17e DF |
1961 | |
1962 | if (new_bw == p->dl.dl_bw) | |
1963 | return 0; | |
1964 | ||
1965 | /* | |
1966 | * Either if a task, enters, leave, or stays -deadline but changes | |
1967 | * its parameters, we may need to update accordingly the total | |
1968 | * allocated bandwidth of the container. | |
1969 | */ | |
1970 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 1971 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
1972 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
1973 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
1974 | __dl_add(dl_b, new_bw); | |
1975 | err = 0; | |
1976 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
1977 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
1978 | __dl_clear(dl_b, p->dl.dl_bw); | |
1979 | __dl_add(dl_b, new_bw); | |
1980 | err = 0; | |
1981 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
1982 | __dl_clear(dl_b, p->dl.dl_bw); | |
1983 | err = 0; | |
1984 | } | |
1985 | raw_spin_unlock(&dl_b->lock); | |
1986 | ||
1987 | return err; | |
1988 | } | |
1989 | ||
1990 | extern void init_dl_bw(struct dl_bw *dl_b); | |
1991 | ||
1da177e4 LT |
1992 | /* |
1993 | * wake_up_new_task - wake up a newly created task for the first time. | |
1994 | * | |
1995 | * This function will do some initial scheduler statistics housekeeping | |
1996 | * that must be done for every newly created context, then puts the task | |
1997 | * on the runqueue and wakes it. | |
1998 | */ | |
3e51e3ed | 1999 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2000 | { |
2001 | unsigned long flags; | |
dd41f596 | 2002 | struct rq *rq; |
fabf318e | 2003 | |
ab2515c4 | 2004 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2005 | #ifdef CONFIG_SMP |
2006 | /* | |
2007 | * Fork balancing, do it here and not earlier because: | |
2008 | * - cpus_allowed can change in the fork path | |
2009 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2010 | */ |
ac66f547 | 2011 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2012 | #endif |
2013 | ||
a75cdaa9 AS |
2014 | /* Initialize new task's runnable average */ |
2015 | init_task_runnable_average(p); | |
ab2515c4 | 2016 | rq = __task_rq_lock(p); |
cd29fe6f | 2017 | activate_task(rq, p, 0); |
fd2f4419 | 2018 | p->on_rq = 1; |
89363381 | 2019 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2020 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2021 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2022 | if (p->sched_class->task_woken) |
2023 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2024 | #endif |
0122ec5b | 2025 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2026 | } |
2027 | ||
e107be36 AK |
2028 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2029 | ||
2030 | /** | |
80dd99b3 | 2031 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2032 | * @notifier: notifier struct to register |
e107be36 AK |
2033 | */ |
2034 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2035 | { | |
2036 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2037 | } | |
2038 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2039 | ||
2040 | /** | |
2041 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2042 | * @notifier: notifier struct to unregister |
e107be36 AK |
2043 | * |
2044 | * This is safe to call from within a preemption notifier. | |
2045 | */ | |
2046 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2047 | { | |
2048 | hlist_del(¬ifier->link); | |
2049 | } | |
2050 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2051 | ||
2052 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2053 | { | |
2054 | struct preempt_notifier *notifier; | |
e107be36 | 2055 | |
b67bfe0d | 2056 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2057 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2058 | } | |
2059 | ||
2060 | static void | |
2061 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2062 | struct task_struct *next) | |
2063 | { | |
2064 | struct preempt_notifier *notifier; | |
e107be36 | 2065 | |
b67bfe0d | 2066 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2067 | notifier->ops->sched_out(notifier, next); |
2068 | } | |
2069 | ||
6d6bc0ad | 2070 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2071 | |
2072 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2073 | { | |
2074 | } | |
2075 | ||
2076 | static void | |
2077 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2078 | struct task_struct *next) | |
2079 | { | |
2080 | } | |
2081 | ||
6d6bc0ad | 2082 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2083 | |
4866cde0 NP |
2084 | /** |
2085 | * prepare_task_switch - prepare to switch tasks | |
2086 | * @rq: the runqueue preparing to switch | |
421cee29 | 2087 | * @prev: the current task that is being switched out |
4866cde0 NP |
2088 | * @next: the task we are going to switch to. |
2089 | * | |
2090 | * This is called with the rq lock held and interrupts off. It must | |
2091 | * be paired with a subsequent finish_task_switch after the context | |
2092 | * switch. | |
2093 | * | |
2094 | * prepare_task_switch sets up locking and calls architecture specific | |
2095 | * hooks. | |
2096 | */ | |
e107be36 AK |
2097 | static inline void |
2098 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2099 | struct task_struct *next) | |
4866cde0 | 2100 | { |
895dd92c | 2101 | trace_sched_switch(prev, next); |
43148951 | 2102 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2103 | perf_event_task_sched_out(prev, next); |
e107be36 | 2104 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2105 | prepare_lock_switch(rq, next); |
2106 | prepare_arch_switch(next); | |
2107 | } | |
2108 | ||
1da177e4 LT |
2109 | /** |
2110 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2111 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2112 | * @prev: the thread we just switched away from. |
2113 | * | |
4866cde0 NP |
2114 | * finish_task_switch must be called after the context switch, paired |
2115 | * with a prepare_task_switch call before the context switch. | |
2116 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2117 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2118 | * |
2119 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2120 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2121 | * with the lock held can cause deadlocks; see schedule() for |
2122 | * details.) | |
2123 | */ | |
a9957449 | 2124 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2125 | __releases(rq->lock) |
2126 | { | |
1da177e4 | 2127 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2128 | long prev_state; |
1da177e4 LT |
2129 | |
2130 | rq->prev_mm = NULL; | |
2131 | ||
2132 | /* | |
2133 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2134 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2135 | * schedule one last time. The schedule call will never return, and |
2136 | * the scheduled task must drop that reference. | |
c394cc9f | 2137 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2138 | * still held, otherwise prev could be scheduled on another cpu, die |
2139 | * there before we look at prev->state, and then the reference would | |
2140 | * be dropped twice. | |
2141 | * Manfred Spraul <manfred@colorfullife.com> | |
2142 | */ | |
55a101f8 | 2143 | prev_state = prev->state; |
bf9fae9f | 2144 | vtime_task_switch(prev); |
4866cde0 | 2145 | finish_arch_switch(prev); |
a8d757ef | 2146 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2147 | finish_lock_switch(rq, prev); |
01f23e16 | 2148 | finish_arch_post_lock_switch(); |
e8fa1362 | 2149 | |
e107be36 | 2150 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2151 | if (mm) |
2152 | mmdrop(mm); | |
c394cc9f | 2153 | if (unlikely(prev_state == TASK_DEAD)) { |
f809ca9a MG |
2154 | task_numa_free(prev); |
2155 | ||
e6c390f2 DF |
2156 | if (prev->sched_class->task_dead) |
2157 | prev->sched_class->task_dead(prev); | |
2158 | ||
c6fd91f0 | 2159 | /* |
2160 | * Remove function-return probe instances associated with this | |
2161 | * task and put them back on the free list. | |
9761eea8 | 2162 | */ |
c6fd91f0 | 2163 | kprobe_flush_task(prev); |
1da177e4 | 2164 | put_task_struct(prev); |
c6fd91f0 | 2165 | } |
99e5ada9 FW |
2166 | |
2167 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2168 | } |
2169 | ||
3f029d3c GH |
2170 | #ifdef CONFIG_SMP |
2171 | ||
3f029d3c GH |
2172 | /* rq->lock is NOT held, but preemption is disabled */ |
2173 | static inline void post_schedule(struct rq *rq) | |
2174 | { | |
2175 | if (rq->post_schedule) { | |
2176 | unsigned long flags; | |
2177 | ||
05fa785c | 2178 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2179 | if (rq->curr->sched_class->post_schedule) |
2180 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2181 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2182 | |
2183 | rq->post_schedule = 0; | |
2184 | } | |
2185 | } | |
2186 | ||
2187 | #else | |
da19ab51 | 2188 | |
3f029d3c GH |
2189 | static inline void post_schedule(struct rq *rq) |
2190 | { | |
1da177e4 LT |
2191 | } |
2192 | ||
3f029d3c GH |
2193 | #endif |
2194 | ||
1da177e4 LT |
2195 | /** |
2196 | * schedule_tail - first thing a freshly forked thread must call. | |
2197 | * @prev: the thread we just switched away from. | |
2198 | */ | |
36c8b586 | 2199 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2200 | __releases(rq->lock) |
2201 | { | |
70b97a7f IM |
2202 | struct rq *rq = this_rq(); |
2203 | ||
4866cde0 | 2204 | finish_task_switch(rq, prev); |
da19ab51 | 2205 | |
3f029d3c GH |
2206 | /* |
2207 | * FIXME: do we need to worry about rq being invalidated by the | |
2208 | * task_switch? | |
2209 | */ | |
2210 | post_schedule(rq); | |
70b97a7f | 2211 | |
4866cde0 NP |
2212 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2213 | /* In this case, finish_task_switch does not reenable preemption */ | |
2214 | preempt_enable(); | |
2215 | #endif | |
1da177e4 | 2216 | if (current->set_child_tid) |
b488893a | 2217 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2218 | } |
2219 | ||
2220 | /* | |
2221 | * context_switch - switch to the new MM and the new | |
2222 | * thread's register state. | |
2223 | */ | |
dd41f596 | 2224 | static inline void |
70b97a7f | 2225 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2226 | struct task_struct *next) |
1da177e4 | 2227 | { |
dd41f596 | 2228 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2229 | |
e107be36 | 2230 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2231 | |
dd41f596 IM |
2232 | mm = next->mm; |
2233 | oldmm = prev->active_mm; | |
9226d125 ZA |
2234 | /* |
2235 | * For paravirt, this is coupled with an exit in switch_to to | |
2236 | * combine the page table reload and the switch backend into | |
2237 | * one hypercall. | |
2238 | */ | |
224101ed | 2239 | arch_start_context_switch(prev); |
9226d125 | 2240 | |
31915ab4 | 2241 | if (!mm) { |
1da177e4 LT |
2242 | next->active_mm = oldmm; |
2243 | atomic_inc(&oldmm->mm_count); | |
2244 | enter_lazy_tlb(oldmm, next); | |
2245 | } else | |
2246 | switch_mm(oldmm, mm, next); | |
2247 | ||
31915ab4 | 2248 | if (!prev->mm) { |
1da177e4 | 2249 | prev->active_mm = NULL; |
1da177e4 LT |
2250 | rq->prev_mm = oldmm; |
2251 | } | |
3a5f5e48 IM |
2252 | /* |
2253 | * Since the runqueue lock will be released by the next | |
2254 | * task (which is an invalid locking op but in the case | |
2255 | * of the scheduler it's an obvious special-case), so we | |
2256 | * do an early lockdep release here: | |
2257 | */ | |
2258 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2259 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2260 | #endif |
1da177e4 | 2261 | |
91d1aa43 | 2262 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2263 | /* Here we just switch the register state and the stack. */ |
2264 | switch_to(prev, next, prev); | |
2265 | ||
dd41f596 IM |
2266 | barrier(); |
2267 | /* | |
2268 | * this_rq must be evaluated again because prev may have moved | |
2269 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2270 | * frame will be invalid. | |
2271 | */ | |
2272 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2273 | } |
2274 | ||
2275 | /* | |
1c3e8264 | 2276 | * nr_running and nr_context_switches: |
1da177e4 LT |
2277 | * |
2278 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2279 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2280 | */ |
2281 | unsigned long nr_running(void) | |
2282 | { | |
2283 | unsigned long i, sum = 0; | |
2284 | ||
2285 | for_each_online_cpu(i) | |
2286 | sum += cpu_rq(i)->nr_running; | |
2287 | ||
2288 | return sum; | |
f711f609 | 2289 | } |
1da177e4 | 2290 | |
1da177e4 | 2291 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2292 | { |
cc94abfc SR |
2293 | int i; |
2294 | unsigned long long sum = 0; | |
46cb4b7c | 2295 | |
0a945022 | 2296 | for_each_possible_cpu(i) |
1da177e4 | 2297 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2298 | |
1da177e4 LT |
2299 | return sum; |
2300 | } | |
483b4ee6 | 2301 | |
1da177e4 LT |
2302 | unsigned long nr_iowait(void) |
2303 | { | |
2304 | unsigned long i, sum = 0; | |
483b4ee6 | 2305 | |
0a945022 | 2306 | for_each_possible_cpu(i) |
1da177e4 | 2307 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2308 | |
1da177e4 LT |
2309 | return sum; |
2310 | } | |
483b4ee6 | 2311 | |
8c215bd3 | 2312 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2313 | { |
8c215bd3 | 2314 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2315 | return atomic_read(&this->nr_iowait); |
2316 | } | |
46cb4b7c | 2317 | |
dd41f596 | 2318 | #ifdef CONFIG_SMP |
8a0be9ef | 2319 | |
46cb4b7c | 2320 | /* |
38022906 PZ |
2321 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2322 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2323 | */ |
38022906 | 2324 | void sched_exec(void) |
46cb4b7c | 2325 | { |
38022906 | 2326 | struct task_struct *p = current; |
1da177e4 | 2327 | unsigned long flags; |
0017d735 | 2328 | int dest_cpu; |
46cb4b7c | 2329 | |
8f42ced9 | 2330 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2331 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2332 | if (dest_cpu == smp_processor_id()) |
2333 | goto unlock; | |
38022906 | 2334 | |
8f42ced9 | 2335 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2336 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2337 | |
8f42ced9 PZ |
2338 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2339 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2340 | return; |
2341 | } | |
0017d735 | 2342 | unlock: |
8f42ced9 | 2343 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2344 | } |
dd41f596 | 2345 | |
1da177e4 LT |
2346 | #endif |
2347 | ||
1da177e4 | 2348 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2349 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2350 | |
2351 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2352 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2353 | |
2354 | /* | |
c5f8d995 | 2355 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2356 | * @p in case that task is currently running. |
c5f8d995 HS |
2357 | * |
2358 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2359 | */ |
c5f8d995 HS |
2360 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2361 | { | |
2362 | u64 ns = 0; | |
2363 | ||
2364 | if (task_current(rq, p)) { | |
2365 | update_rq_clock(rq); | |
78becc27 | 2366 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2367 | if ((s64)ns < 0) |
2368 | ns = 0; | |
2369 | } | |
2370 | ||
2371 | return ns; | |
2372 | } | |
2373 | ||
bb34d92f | 2374 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2375 | { |
1da177e4 | 2376 | unsigned long flags; |
41b86e9c | 2377 | struct rq *rq; |
bb34d92f | 2378 | u64 ns = 0; |
48f24c4d | 2379 | |
41b86e9c | 2380 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2381 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2382 | task_rq_unlock(rq, p, &flags); |
1508487e | 2383 | |
c5f8d995 HS |
2384 | return ns; |
2385 | } | |
f06febc9 | 2386 | |
c5f8d995 HS |
2387 | /* |
2388 | * Return accounted runtime for the task. | |
2389 | * In case the task is currently running, return the runtime plus current's | |
2390 | * pending runtime that have not been accounted yet. | |
2391 | */ | |
2392 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2393 | { | |
2394 | unsigned long flags; | |
2395 | struct rq *rq; | |
2396 | u64 ns = 0; | |
2397 | ||
911b2898 PZ |
2398 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2399 | /* | |
2400 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2401 | * So we have a optimization chance when the task's delta_exec is 0. | |
2402 | * Reading ->on_cpu is racy, but this is ok. | |
2403 | * | |
2404 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2405 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2406 | * indistinguishable from the read occurring a few cycles earlier. | |
2407 | */ | |
2408 | if (!p->on_cpu) | |
2409 | return p->se.sum_exec_runtime; | |
2410 | #endif | |
2411 | ||
c5f8d995 HS |
2412 | rq = task_rq_lock(p, &flags); |
2413 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2414 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2415 | |
2416 | return ns; | |
2417 | } | |
48f24c4d | 2418 | |
7835b98b CL |
2419 | /* |
2420 | * This function gets called by the timer code, with HZ frequency. | |
2421 | * We call it with interrupts disabled. | |
7835b98b CL |
2422 | */ |
2423 | void scheduler_tick(void) | |
2424 | { | |
7835b98b CL |
2425 | int cpu = smp_processor_id(); |
2426 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2427 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2428 | |
2429 | sched_clock_tick(); | |
dd41f596 | 2430 | |
05fa785c | 2431 | raw_spin_lock(&rq->lock); |
3e51f33f | 2432 | update_rq_clock(rq); |
fa85ae24 | 2433 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2434 | update_cpu_load_active(rq); |
05fa785c | 2435 | raw_spin_unlock(&rq->lock); |
7835b98b | 2436 | |
e9d2b064 | 2437 | perf_event_task_tick(); |
e220d2dc | 2438 | |
e418e1c2 | 2439 | #ifdef CONFIG_SMP |
6eb57e0d | 2440 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2441 | trigger_load_balance(rq); |
e418e1c2 | 2442 | #endif |
265f22a9 | 2443 | rq_last_tick_reset(rq); |
1da177e4 LT |
2444 | } |
2445 | ||
265f22a9 FW |
2446 | #ifdef CONFIG_NO_HZ_FULL |
2447 | /** | |
2448 | * scheduler_tick_max_deferment | |
2449 | * | |
2450 | * Keep at least one tick per second when a single | |
2451 | * active task is running because the scheduler doesn't | |
2452 | * yet completely support full dynticks environment. | |
2453 | * | |
2454 | * This makes sure that uptime, CFS vruntime, load | |
2455 | * balancing, etc... continue to move forward, even | |
2456 | * with a very low granularity. | |
e69f6186 YB |
2457 | * |
2458 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2459 | */ |
2460 | u64 scheduler_tick_max_deferment(void) | |
2461 | { | |
2462 | struct rq *rq = this_rq(); | |
2463 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2464 | ||
2465 | next = rq->last_sched_tick + HZ; | |
2466 | ||
2467 | if (time_before_eq(next, now)) | |
2468 | return 0; | |
2469 | ||
8fe8ff09 | 2470 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2471 | } |
265f22a9 | 2472 | #endif |
1da177e4 | 2473 | |
132380a0 | 2474 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2475 | { |
2476 | if (in_lock_functions(addr)) { | |
2477 | addr = CALLER_ADDR2; | |
2478 | if (in_lock_functions(addr)) | |
2479 | addr = CALLER_ADDR3; | |
2480 | } | |
2481 | return addr; | |
2482 | } | |
1da177e4 | 2483 | |
7e49fcce SR |
2484 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2485 | defined(CONFIG_PREEMPT_TRACER)) | |
2486 | ||
bdb43806 | 2487 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2488 | { |
6cd8a4bb | 2489 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2490 | /* |
2491 | * Underflow? | |
2492 | */ | |
9a11b49a IM |
2493 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2494 | return; | |
6cd8a4bb | 2495 | #endif |
bdb43806 | 2496 | __preempt_count_add(val); |
6cd8a4bb | 2497 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2498 | /* |
2499 | * Spinlock count overflowing soon? | |
2500 | */ | |
33859f7f MOS |
2501 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2502 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2503 | #endif |
8f47b187 TG |
2504 | if (preempt_count() == val) { |
2505 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2506 | #ifdef CONFIG_DEBUG_PREEMPT | |
2507 | current->preempt_disable_ip = ip; | |
2508 | #endif | |
2509 | trace_preempt_off(CALLER_ADDR0, ip); | |
2510 | } | |
1da177e4 | 2511 | } |
bdb43806 | 2512 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2513 | |
bdb43806 | 2514 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2515 | { |
6cd8a4bb | 2516 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2517 | /* |
2518 | * Underflow? | |
2519 | */ | |
01e3eb82 | 2520 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2521 | return; |
1da177e4 LT |
2522 | /* |
2523 | * Is the spinlock portion underflowing? | |
2524 | */ | |
9a11b49a IM |
2525 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2526 | !(preempt_count() & PREEMPT_MASK))) | |
2527 | return; | |
6cd8a4bb | 2528 | #endif |
9a11b49a | 2529 | |
6cd8a4bb SR |
2530 | if (preempt_count() == val) |
2531 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2532 | __preempt_count_sub(val); |
1da177e4 | 2533 | } |
bdb43806 | 2534 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2535 | |
2536 | #endif | |
2537 | ||
2538 | /* | |
dd41f596 | 2539 | * Print scheduling while atomic bug: |
1da177e4 | 2540 | */ |
dd41f596 | 2541 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2542 | { |
664dfa65 DJ |
2543 | if (oops_in_progress) |
2544 | return; | |
2545 | ||
3df0fc5b PZ |
2546 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2547 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2548 | |
dd41f596 | 2549 | debug_show_held_locks(prev); |
e21f5b15 | 2550 | print_modules(); |
dd41f596 IM |
2551 | if (irqs_disabled()) |
2552 | print_irqtrace_events(prev); | |
8f47b187 TG |
2553 | #ifdef CONFIG_DEBUG_PREEMPT |
2554 | if (in_atomic_preempt_off()) { | |
2555 | pr_err("Preemption disabled at:"); | |
2556 | print_ip_sym(current->preempt_disable_ip); | |
2557 | pr_cont("\n"); | |
2558 | } | |
2559 | #endif | |
6135fc1e | 2560 | dump_stack(); |
373d4d09 | 2561 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2562 | } |
1da177e4 | 2563 | |
dd41f596 IM |
2564 | /* |
2565 | * Various schedule()-time debugging checks and statistics: | |
2566 | */ | |
2567 | static inline void schedule_debug(struct task_struct *prev) | |
2568 | { | |
1da177e4 | 2569 | /* |
41a2d6cf | 2570 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2571 | * schedule() atomically, we ignore that path. Otherwise whine |
2572 | * if we are scheduling when we should not. | |
1da177e4 | 2573 | */ |
192301e7 | 2574 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2575 | __schedule_bug(prev); |
b3fbab05 | 2576 | rcu_sleep_check(); |
dd41f596 | 2577 | |
1da177e4 LT |
2578 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2579 | ||
2d72376b | 2580 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2581 | } |
2582 | ||
2583 | /* | |
2584 | * Pick up the highest-prio task: | |
2585 | */ | |
2586 | static inline struct task_struct * | |
606dba2e | 2587 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2588 | { |
5522d5d5 | 2589 | const struct sched_class *class; |
dd41f596 | 2590 | struct task_struct *p; |
1da177e4 LT |
2591 | |
2592 | /* | |
dd41f596 IM |
2593 | * Optimization: we know that if all tasks are in |
2594 | * the fair class we can call that function directly: | |
1da177e4 | 2595 | */ |
38033c37 PZ |
2596 | if (likely(prev->sched_class == &fair_sched_class && |
2597 | rq->nr_running == rq->cfs.h_nr_running)) { | |
606dba2e | 2598 | p = fair_sched_class.pick_next_task(rq, prev); |
dd41f596 IM |
2599 | if (likely(p)) |
2600 | return p; | |
1da177e4 LT |
2601 | } |
2602 | ||
34f971f6 | 2603 | for_each_class(class) { |
606dba2e | 2604 | p = class->pick_next_task(rq, prev); |
dd41f596 IM |
2605 | if (p) |
2606 | return p; | |
dd41f596 | 2607 | } |
34f971f6 PZ |
2608 | |
2609 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2610 | } |
1da177e4 | 2611 | |
dd41f596 | 2612 | /* |
c259e01a | 2613 | * __schedule() is the main scheduler function. |
edde96ea PE |
2614 | * |
2615 | * The main means of driving the scheduler and thus entering this function are: | |
2616 | * | |
2617 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2618 | * | |
2619 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2620 | * paths. For example, see arch/x86/entry_64.S. | |
2621 | * | |
2622 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2623 | * interrupt handler scheduler_tick(). | |
2624 | * | |
2625 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2626 | * task to the run-queue and that's it. | |
2627 | * | |
2628 | * Now, if the new task added to the run-queue preempts the current | |
2629 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2630 | * called on the nearest possible occasion: | |
2631 | * | |
2632 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2633 | * | |
2634 | * - in syscall or exception context, at the next outmost | |
2635 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2636 | * spin_unlock()!) | |
2637 | * | |
2638 | * - in IRQ context, return from interrupt-handler to | |
2639 | * preemptible context | |
2640 | * | |
2641 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2642 | * then at the next: | |
2643 | * | |
2644 | * - cond_resched() call | |
2645 | * - explicit schedule() call | |
2646 | * - return from syscall or exception to user-space | |
2647 | * - return from interrupt-handler to user-space | |
dd41f596 | 2648 | */ |
c259e01a | 2649 | static void __sched __schedule(void) |
dd41f596 IM |
2650 | { |
2651 | struct task_struct *prev, *next; | |
67ca7bde | 2652 | unsigned long *switch_count; |
dd41f596 | 2653 | struct rq *rq; |
31656519 | 2654 | int cpu; |
dd41f596 | 2655 | |
ff743345 PZ |
2656 | need_resched: |
2657 | preempt_disable(); | |
dd41f596 IM |
2658 | cpu = smp_processor_id(); |
2659 | rq = cpu_rq(cpu); | |
25502a6c | 2660 | rcu_note_context_switch(cpu); |
dd41f596 | 2661 | prev = rq->curr; |
dd41f596 | 2662 | |
dd41f596 | 2663 | schedule_debug(prev); |
1da177e4 | 2664 | |
31656519 | 2665 | if (sched_feat(HRTICK)) |
f333fdc9 | 2666 | hrtick_clear(rq); |
8f4d37ec | 2667 | |
e0acd0a6 ON |
2668 | /* |
2669 | * Make sure that signal_pending_state()->signal_pending() below | |
2670 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2671 | * done by the caller to avoid the race with signal_wake_up(). | |
2672 | */ | |
2673 | smp_mb__before_spinlock(); | |
05fa785c | 2674 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2675 | |
246d86b5 | 2676 | switch_count = &prev->nivcsw; |
1da177e4 | 2677 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2678 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2679 | prev->state = TASK_RUNNING; |
21aa9af0 | 2680 | } else { |
2acca55e PZ |
2681 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2682 | prev->on_rq = 0; | |
2683 | ||
21aa9af0 | 2684 | /* |
2acca55e PZ |
2685 | * If a worker went to sleep, notify and ask workqueue |
2686 | * whether it wants to wake up a task to maintain | |
2687 | * concurrency. | |
21aa9af0 TH |
2688 | */ |
2689 | if (prev->flags & PF_WQ_WORKER) { | |
2690 | struct task_struct *to_wakeup; | |
2691 | ||
2692 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2693 | if (to_wakeup) | |
2694 | try_to_wake_up_local(to_wakeup); | |
2695 | } | |
21aa9af0 | 2696 | } |
dd41f596 | 2697 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2698 | } |
2699 | ||
606dba2e PZ |
2700 | if (prev->on_rq || rq->skip_clock_update < 0) |
2701 | update_rq_clock(rq); | |
2702 | ||
2703 | next = pick_next_task(rq, prev); | |
f26f9aff | 2704 | clear_tsk_need_resched(prev); |
f27dde8d | 2705 | clear_preempt_need_resched(); |
f26f9aff | 2706 | rq->skip_clock_update = 0; |
1da177e4 | 2707 | |
1da177e4 | 2708 | if (likely(prev != next)) { |
1da177e4 LT |
2709 | rq->nr_switches++; |
2710 | rq->curr = next; | |
2711 | ++*switch_count; | |
2712 | ||
dd41f596 | 2713 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2714 | /* |
246d86b5 ON |
2715 | * The context switch have flipped the stack from under us |
2716 | * and restored the local variables which were saved when | |
2717 | * this task called schedule() in the past. prev == current | |
2718 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2719 | */ |
2720 | cpu = smp_processor_id(); | |
2721 | rq = cpu_rq(cpu); | |
1da177e4 | 2722 | } else |
05fa785c | 2723 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2724 | |
3f029d3c | 2725 | post_schedule(rq); |
1da177e4 | 2726 | |
ba74c144 | 2727 | sched_preempt_enable_no_resched(); |
ff743345 | 2728 | if (need_resched()) |
1da177e4 LT |
2729 | goto need_resched; |
2730 | } | |
c259e01a | 2731 | |
9c40cef2 TG |
2732 | static inline void sched_submit_work(struct task_struct *tsk) |
2733 | { | |
3c7d5184 | 2734 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2735 | return; |
2736 | /* | |
2737 | * If we are going to sleep and we have plugged IO queued, | |
2738 | * make sure to submit it to avoid deadlocks. | |
2739 | */ | |
2740 | if (blk_needs_flush_plug(tsk)) | |
2741 | blk_schedule_flush_plug(tsk); | |
2742 | } | |
2743 | ||
6ebbe7a0 | 2744 | asmlinkage void __sched schedule(void) |
c259e01a | 2745 | { |
9c40cef2 TG |
2746 | struct task_struct *tsk = current; |
2747 | ||
2748 | sched_submit_work(tsk); | |
c259e01a TG |
2749 | __schedule(); |
2750 | } | |
1da177e4 LT |
2751 | EXPORT_SYMBOL(schedule); |
2752 | ||
91d1aa43 | 2753 | #ifdef CONFIG_CONTEXT_TRACKING |
20ab65e3 FW |
2754 | asmlinkage void __sched schedule_user(void) |
2755 | { | |
2756 | /* | |
2757 | * If we come here after a random call to set_need_resched(), | |
2758 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2759 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2760 | * we find a better solution. | |
2761 | */ | |
91d1aa43 | 2762 | user_exit(); |
20ab65e3 | 2763 | schedule(); |
91d1aa43 | 2764 | user_enter(); |
20ab65e3 FW |
2765 | } |
2766 | #endif | |
2767 | ||
c5491ea7 TG |
2768 | /** |
2769 | * schedule_preempt_disabled - called with preemption disabled | |
2770 | * | |
2771 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2772 | */ | |
2773 | void __sched schedule_preempt_disabled(void) | |
2774 | { | |
ba74c144 | 2775 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2776 | schedule(); |
2777 | preempt_disable(); | |
2778 | } | |
2779 | ||
1da177e4 LT |
2780 | #ifdef CONFIG_PREEMPT |
2781 | /* | |
2ed6e34f | 2782 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2783 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2784 | * occur there and call schedule directly. |
2785 | */ | |
d1f74e20 | 2786 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 | 2787 | { |
1da177e4 LT |
2788 | /* |
2789 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2790 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2791 | */ |
fbb00b56 | 2792 | if (likely(!preemptible())) |
1da177e4 LT |
2793 | return; |
2794 | ||
3a5c359a | 2795 | do { |
bdb43806 | 2796 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2797 | __schedule(); |
bdb43806 | 2798 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2799 | |
3a5c359a AK |
2800 | /* |
2801 | * Check again in case we missed a preemption opportunity | |
2802 | * between schedule and now. | |
2803 | */ | |
2804 | barrier(); | |
5ed0cec0 | 2805 | } while (need_resched()); |
1da177e4 | 2806 | } |
1da177e4 | 2807 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2808 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2809 | |
2810 | /* | |
2ed6e34f | 2811 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2812 | * off of irq context. |
2813 | * Note, that this is called and return with irqs disabled. This will | |
2814 | * protect us against recursive calling from irq. | |
2815 | */ | |
2816 | asmlinkage void __sched preempt_schedule_irq(void) | |
2817 | { | |
b22366cd | 2818 | enum ctx_state prev_state; |
6478d880 | 2819 | |
2ed6e34f | 2820 | /* Catch callers which need to be fixed */ |
f27dde8d | 2821 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2822 | |
b22366cd FW |
2823 | prev_state = exception_enter(); |
2824 | ||
3a5c359a | 2825 | do { |
bdb43806 | 2826 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2827 | local_irq_enable(); |
c259e01a | 2828 | __schedule(); |
3a5c359a | 2829 | local_irq_disable(); |
bdb43806 | 2830 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2831 | |
3a5c359a AK |
2832 | /* |
2833 | * Check again in case we missed a preemption opportunity | |
2834 | * between schedule and now. | |
2835 | */ | |
2836 | barrier(); | |
5ed0cec0 | 2837 | } while (need_resched()); |
b22366cd FW |
2838 | |
2839 | exception_exit(prev_state); | |
1da177e4 LT |
2840 | } |
2841 | ||
63859d4f | 2842 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2843 | void *key) |
1da177e4 | 2844 | { |
63859d4f | 2845 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2846 | } |
1da177e4 LT |
2847 | EXPORT_SYMBOL(default_wake_function); |
2848 | ||
8cbbe86d AK |
2849 | static long __sched |
2850 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 2851 | { |
0fec171c IM |
2852 | unsigned long flags; |
2853 | wait_queue_t wait; | |
2854 | ||
2855 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 2856 | |
8cbbe86d | 2857 | __set_current_state(state); |
1da177e4 | 2858 | |
8cbbe86d AK |
2859 | spin_lock_irqsave(&q->lock, flags); |
2860 | __add_wait_queue(q, &wait); | |
2861 | spin_unlock(&q->lock); | |
2862 | timeout = schedule_timeout(timeout); | |
2863 | spin_lock_irq(&q->lock); | |
2864 | __remove_wait_queue(q, &wait); | |
2865 | spin_unlock_irqrestore(&q->lock, flags); | |
2866 | ||
2867 | return timeout; | |
2868 | } | |
2869 | ||
2870 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
2871 | { | |
2872 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 2873 | } |
1da177e4 LT |
2874 | EXPORT_SYMBOL(interruptible_sleep_on); |
2875 | ||
0fec171c | 2876 | long __sched |
95cdf3b7 | 2877 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2878 | { |
8cbbe86d | 2879 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 2880 | } |
1da177e4 LT |
2881 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
2882 | ||
0fec171c | 2883 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 2884 | { |
8cbbe86d | 2885 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 2886 | } |
1da177e4 LT |
2887 | EXPORT_SYMBOL(sleep_on); |
2888 | ||
0fec171c | 2889 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 2890 | { |
8cbbe86d | 2891 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 2892 | } |
1da177e4 LT |
2893 | EXPORT_SYMBOL(sleep_on_timeout); |
2894 | ||
b29739f9 IM |
2895 | #ifdef CONFIG_RT_MUTEXES |
2896 | ||
2897 | /* | |
2898 | * rt_mutex_setprio - set the current priority of a task | |
2899 | * @p: task | |
2900 | * @prio: prio value (kernel-internal form) | |
2901 | * | |
2902 | * This function changes the 'effective' priority of a task. It does | |
2903 | * not touch ->normal_prio like __setscheduler(). | |
2904 | * | |
c365c292 TG |
2905 | * Used by the rt_mutex code to implement priority inheritance |
2906 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2907 | */ |
36c8b586 | 2908 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2909 | { |
2d3d891d | 2910 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2911 | struct rq *rq; |
83ab0aa0 | 2912 | const struct sched_class *prev_class; |
b29739f9 | 2913 | |
aab03e05 | 2914 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2915 | |
0122ec5b | 2916 | rq = __task_rq_lock(p); |
b29739f9 | 2917 | |
1c4dd99b TG |
2918 | /* |
2919 | * Idle task boosting is a nono in general. There is one | |
2920 | * exception, when PREEMPT_RT and NOHZ is active: | |
2921 | * | |
2922 | * The idle task calls get_next_timer_interrupt() and holds | |
2923 | * the timer wheel base->lock on the CPU and another CPU wants | |
2924 | * to access the timer (probably to cancel it). We can safely | |
2925 | * ignore the boosting request, as the idle CPU runs this code | |
2926 | * with interrupts disabled and will complete the lock | |
2927 | * protected section without being interrupted. So there is no | |
2928 | * real need to boost. | |
2929 | */ | |
2930 | if (unlikely(p == rq->idle)) { | |
2931 | WARN_ON(p != rq->curr); | |
2932 | WARN_ON(p->pi_blocked_on); | |
2933 | goto out_unlock; | |
2934 | } | |
2935 | ||
a8027073 | 2936 | trace_sched_pi_setprio(p, prio); |
2d3d891d | 2937 | p->pi_top_task = rt_mutex_get_top_task(p); |
d5f9f942 | 2938 | oldprio = p->prio; |
83ab0aa0 | 2939 | prev_class = p->sched_class; |
fd2f4419 | 2940 | on_rq = p->on_rq; |
051a1d1a | 2941 | running = task_current(rq, p); |
0e1f3483 | 2942 | if (on_rq) |
69be72c1 | 2943 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2944 | if (running) |
2945 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2946 | |
2d3d891d DF |
2947 | /* |
2948 | * Boosting condition are: | |
2949 | * 1. -rt task is running and holds mutex A | |
2950 | * --> -dl task blocks on mutex A | |
2951 | * | |
2952 | * 2. -dl task is running and holds mutex A | |
2953 | * --> -dl task blocks on mutex A and could preempt the | |
2954 | * running task | |
2955 | */ | |
2956 | if (dl_prio(prio)) { | |
2957 | if (!dl_prio(p->normal_prio) || (p->pi_top_task && | |
2958 | dl_entity_preempt(&p->pi_top_task->dl, &p->dl))) { | |
2959 | p->dl.dl_boosted = 1; | |
2960 | p->dl.dl_throttled = 0; | |
2961 | enqueue_flag = ENQUEUE_REPLENISH; | |
2962 | } else | |
2963 | p->dl.dl_boosted = 0; | |
aab03e05 | 2964 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
2965 | } else if (rt_prio(prio)) { |
2966 | if (dl_prio(oldprio)) | |
2967 | p->dl.dl_boosted = 0; | |
2968 | if (oldprio < prio) | |
2969 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 2970 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
2971 | } else { |
2972 | if (dl_prio(oldprio)) | |
2973 | p->dl.dl_boosted = 0; | |
dd41f596 | 2974 | p->sched_class = &fair_sched_class; |
2d3d891d | 2975 | } |
dd41f596 | 2976 | |
b29739f9 IM |
2977 | p->prio = prio; |
2978 | ||
0e1f3483 HS |
2979 | if (running) |
2980 | p->sched_class->set_curr_task(rq); | |
da7a735e | 2981 | if (on_rq) |
2d3d891d | 2982 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 2983 | |
da7a735e | 2984 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 2985 | out_unlock: |
0122ec5b | 2986 | __task_rq_unlock(rq); |
b29739f9 | 2987 | } |
b29739f9 | 2988 | #endif |
d50dde5a | 2989 | |
36c8b586 | 2990 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 2991 | { |
dd41f596 | 2992 | int old_prio, delta, on_rq; |
1da177e4 | 2993 | unsigned long flags; |
70b97a7f | 2994 | struct rq *rq; |
1da177e4 | 2995 | |
d0ea0268 | 2996 | if (task_nice(p) == nice || nice < -20 || nice > 19) |
1da177e4 LT |
2997 | return; |
2998 | /* | |
2999 | * We have to be careful, if called from sys_setpriority(), | |
3000 | * the task might be in the middle of scheduling on another CPU. | |
3001 | */ | |
3002 | rq = task_rq_lock(p, &flags); | |
3003 | /* | |
3004 | * The RT priorities are set via sched_setscheduler(), but we still | |
3005 | * allow the 'normal' nice value to be set - but as expected | |
3006 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3007 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3008 | */ |
aab03e05 | 3009 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3010 | p->static_prio = NICE_TO_PRIO(nice); |
3011 | goto out_unlock; | |
3012 | } | |
fd2f4419 | 3013 | on_rq = p->on_rq; |
c09595f6 | 3014 | if (on_rq) |
69be72c1 | 3015 | dequeue_task(rq, p, 0); |
1da177e4 | 3016 | |
1da177e4 | 3017 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3018 | set_load_weight(p); |
b29739f9 IM |
3019 | old_prio = p->prio; |
3020 | p->prio = effective_prio(p); | |
3021 | delta = p->prio - old_prio; | |
1da177e4 | 3022 | |
dd41f596 | 3023 | if (on_rq) { |
371fd7e7 | 3024 | enqueue_task(rq, p, 0); |
1da177e4 | 3025 | /* |
d5f9f942 AM |
3026 | * If the task increased its priority or is running and |
3027 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3028 | */ |
d5f9f942 | 3029 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3030 | resched_task(rq->curr); |
3031 | } | |
3032 | out_unlock: | |
0122ec5b | 3033 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3034 | } |
1da177e4 LT |
3035 | EXPORT_SYMBOL(set_user_nice); |
3036 | ||
e43379f1 MM |
3037 | /* |
3038 | * can_nice - check if a task can reduce its nice value | |
3039 | * @p: task | |
3040 | * @nice: nice value | |
3041 | */ | |
36c8b586 | 3042 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3043 | { |
024f4747 MM |
3044 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3045 | int nice_rlim = 20 - nice; | |
48f24c4d | 3046 | |
78d7d407 | 3047 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3048 | capable(CAP_SYS_NICE)); |
3049 | } | |
3050 | ||
1da177e4 LT |
3051 | #ifdef __ARCH_WANT_SYS_NICE |
3052 | ||
3053 | /* | |
3054 | * sys_nice - change the priority of the current process. | |
3055 | * @increment: priority increment | |
3056 | * | |
3057 | * sys_setpriority is a more generic, but much slower function that | |
3058 | * does similar things. | |
3059 | */ | |
5add95d4 | 3060 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3061 | { |
48f24c4d | 3062 | long nice, retval; |
1da177e4 LT |
3063 | |
3064 | /* | |
3065 | * Setpriority might change our priority at the same moment. | |
3066 | * We don't have to worry. Conceptually one call occurs first | |
3067 | * and we have a single winner. | |
3068 | */ | |
e43379f1 MM |
3069 | if (increment < -40) |
3070 | increment = -40; | |
1da177e4 LT |
3071 | if (increment > 40) |
3072 | increment = 40; | |
3073 | ||
d0ea0268 | 3074 | nice = task_nice(current) + increment; |
1da177e4 LT |
3075 | if (nice < -20) |
3076 | nice = -20; | |
3077 | if (nice > 19) | |
3078 | nice = 19; | |
3079 | ||
e43379f1 MM |
3080 | if (increment < 0 && !can_nice(current, nice)) |
3081 | return -EPERM; | |
3082 | ||
1da177e4 LT |
3083 | retval = security_task_setnice(current, nice); |
3084 | if (retval) | |
3085 | return retval; | |
3086 | ||
3087 | set_user_nice(current, nice); | |
3088 | return 0; | |
3089 | } | |
3090 | ||
3091 | #endif | |
3092 | ||
3093 | /** | |
3094 | * task_prio - return the priority value of a given task. | |
3095 | * @p: the task in question. | |
3096 | * | |
e69f6186 | 3097 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3098 | * RT tasks are offset by -200. Normal tasks are centered |
3099 | * around 0, value goes from -16 to +15. | |
3100 | */ | |
36c8b586 | 3101 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3102 | { |
3103 | return p->prio - MAX_RT_PRIO; | |
3104 | } | |
3105 | ||
1da177e4 LT |
3106 | /** |
3107 | * idle_cpu - is a given cpu idle currently? | |
3108 | * @cpu: the processor in question. | |
e69f6186 YB |
3109 | * |
3110 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3111 | */ |
3112 | int idle_cpu(int cpu) | |
3113 | { | |
908a3283 TG |
3114 | struct rq *rq = cpu_rq(cpu); |
3115 | ||
3116 | if (rq->curr != rq->idle) | |
3117 | return 0; | |
3118 | ||
3119 | if (rq->nr_running) | |
3120 | return 0; | |
3121 | ||
3122 | #ifdef CONFIG_SMP | |
3123 | if (!llist_empty(&rq->wake_list)) | |
3124 | return 0; | |
3125 | #endif | |
3126 | ||
3127 | return 1; | |
1da177e4 LT |
3128 | } |
3129 | ||
1da177e4 LT |
3130 | /** |
3131 | * idle_task - return the idle task for a given cpu. | |
3132 | * @cpu: the processor in question. | |
e69f6186 YB |
3133 | * |
3134 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3135 | */ |
36c8b586 | 3136 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3137 | { |
3138 | return cpu_rq(cpu)->idle; | |
3139 | } | |
3140 | ||
3141 | /** | |
3142 | * find_process_by_pid - find a process with a matching PID value. | |
3143 | * @pid: the pid in question. | |
e69f6186 YB |
3144 | * |
3145 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3146 | */ |
a9957449 | 3147 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3148 | { |
228ebcbe | 3149 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3150 | } |
3151 | ||
aab03e05 DF |
3152 | /* |
3153 | * This function initializes the sched_dl_entity of a newly becoming | |
3154 | * SCHED_DEADLINE task. | |
3155 | * | |
3156 | * Only the static values are considered here, the actual runtime and the | |
3157 | * absolute deadline will be properly calculated when the task is enqueued | |
3158 | * for the first time with its new policy. | |
3159 | */ | |
3160 | static void | |
3161 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3162 | { | |
3163 | struct sched_dl_entity *dl_se = &p->dl; | |
3164 | ||
3165 | init_dl_task_timer(dl_se); | |
3166 | dl_se->dl_runtime = attr->sched_runtime; | |
3167 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3168 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3169 | dl_se->flags = attr->sched_flags; |
332ac17e | 3170 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3171 | dl_se->dl_throttled = 0; |
3172 | dl_se->dl_new = 1; | |
3173 | } | |
3174 | ||
c365c292 TG |
3175 | static void __setscheduler_params(struct task_struct *p, |
3176 | const struct sched_attr *attr) | |
1da177e4 | 3177 | { |
d50dde5a DF |
3178 | int policy = attr->sched_policy; |
3179 | ||
39fd8fd2 PZ |
3180 | if (policy == -1) /* setparam */ |
3181 | policy = p->policy; | |
3182 | ||
1da177e4 | 3183 | p->policy = policy; |
d50dde5a | 3184 | |
aab03e05 DF |
3185 | if (dl_policy(policy)) |
3186 | __setparam_dl(p, attr); | |
39fd8fd2 | 3187 | else if (fair_policy(policy)) |
d50dde5a DF |
3188 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3189 | ||
39fd8fd2 PZ |
3190 | /* |
3191 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3192 | * !rt_policy. Always setting this ensures that things like | |
3193 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3194 | */ | |
3195 | p->rt_priority = attr->sched_priority; | |
c365c292 TG |
3196 | set_load_weight(p); |
3197 | } | |
39fd8fd2 | 3198 | |
c365c292 TG |
3199 | /* Actually do priority change: must hold pi & rq lock. */ |
3200 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3201 | const struct sched_attr *attr) | |
3202 | { | |
3203 | __setscheduler_params(p, attr); | |
d50dde5a | 3204 | |
aab03e05 DF |
3205 | if (dl_prio(p->prio)) |
3206 | p->sched_class = &dl_sched_class; | |
3207 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3208 | p->sched_class = &rt_sched_class; |
3209 | else | |
3210 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3211 | } |
aab03e05 DF |
3212 | |
3213 | static void | |
3214 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3215 | { | |
3216 | struct sched_dl_entity *dl_se = &p->dl; | |
3217 | ||
3218 | attr->sched_priority = p->rt_priority; | |
3219 | attr->sched_runtime = dl_se->dl_runtime; | |
3220 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3221 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3222 | attr->sched_flags = dl_se->flags; |
3223 | } | |
3224 | ||
3225 | /* | |
3226 | * This function validates the new parameters of a -deadline task. | |
3227 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3228 | * than the runtime, as well as the period of being zero or |
332ac17e DF |
3229 | * greater than deadline. Furthermore, we have to be sure that |
3230 | * user parameters are above the internal resolution (1us); we | |
3231 | * check sched_runtime only since it is always the smaller one. | |
aab03e05 DF |
3232 | */ |
3233 | static bool | |
3234 | __checkparam_dl(const struct sched_attr *attr) | |
3235 | { | |
3236 | return attr && attr->sched_deadline != 0 && | |
755378a4 HG |
3237 | (attr->sched_period == 0 || |
3238 | (s64)(attr->sched_period - attr->sched_deadline) >= 0) && | |
332ac17e DF |
3239 | (s64)(attr->sched_deadline - attr->sched_runtime ) >= 0 && |
3240 | attr->sched_runtime >= (2 << (DL_SCALE - 1)); | |
aab03e05 DF |
3241 | } |
3242 | ||
c69e8d9c DH |
3243 | /* |
3244 | * check the target process has a UID that matches the current process's | |
3245 | */ | |
3246 | static bool check_same_owner(struct task_struct *p) | |
3247 | { | |
3248 | const struct cred *cred = current_cred(), *pcred; | |
3249 | bool match; | |
3250 | ||
3251 | rcu_read_lock(); | |
3252 | pcred = __task_cred(p); | |
9c806aa0 EB |
3253 | match = (uid_eq(cred->euid, pcred->euid) || |
3254 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3255 | rcu_read_unlock(); |
3256 | return match; | |
3257 | } | |
3258 | ||
d50dde5a DF |
3259 | static int __sched_setscheduler(struct task_struct *p, |
3260 | const struct sched_attr *attr, | |
3261 | bool user) | |
1da177e4 | 3262 | { |
c365c292 | 3263 | int newprio = MAX_RT_PRIO - 1 - attr->sched_priority; |
83b699ed | 3264 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3265 | int policy = attr->sched_policy; |
1da177e4 | 3266 | unsigned long flags; |
83ab0aa0 | 3267 | const struct sched_class *prev_class; |
70b97a7f | 3268 | struct rq *rq; |
ca94c442 | 3269 | int reset_on_fork; |
1da177e4 | 3270 | |
66e5393a SR |
3271 | /* may grab non-irq protected spin_locks */ |
3272 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3273 | recheck: |
3274 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3275 | if (policy < 0) { |
3276 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3277 | policy = oldpolicy = p->policy; |
ca94c442 | 3278 | } else { |
7479f3c9 | 3279 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3280 | |
aab03e05 DF |
3281 | if (policy != SCHED_DEADLINE && |
3282 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3283 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3284 | policy != SCHED_IDLE) | |
3285 | return -EINVAL; | |
3286 | } | |
3287 | ||
7479f3c9 PZ |
3288 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3289 | return -EINVAL; | |
3290 | ||
1da177e4 LT |
3291 | /* |
3292 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3293 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3294 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3295 | */ |
0bb040a4 | 3296 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3297 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3298 | return -EINVAL; |
aab03e05 DF |
3299 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3300 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3301 | return -EINVAL; |
3302 | ||
37e4ab3f OC |
3303 | /* |
3304 | * Allow unprivileged RT tasks to decrease priority: | |
3305 | */ | |
961ccddd | 3306 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3307 | if (fair_policy(policy)) { |
d0ea0268 | 3308 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3309 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3310 | return -EPERM; |
3311 | } | |
3312 | ||
e05606d3 | 3313 | if (rt_policy(policy)) { |
a44702e8 ON |
3314 | unsigned long rlim_rtprio = |
3315 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3316 | |
3317 | /* can't set/change the rt policy */ | |
3318 | if (policy != p->policy && !rlim_rtprio) | |
3319 | return -EPERM; | |
3320 | ||
3321 | /* can't increase priority */ | |
d50dde5a DF |
3322 | if (attr->sched_priority > p->rt_priority && |
3323 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3324 | return -EPERM; |
3325 | } | |
c02aa73b | 3326 | |
dd41f596 | 3327 | /* |
c02aa73b DH |
3328 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3329 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3330 | */ |
c02aa73b | 3331 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3332 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3333 | return -EPERM; |
3334 | } | |
5fe1d75f | 3335 | |
37e4ab3f | 3336 | /* can't change other user's priorities */ |
c69e8d9c | 3337 | if (!check_same_owner(p)) |
37e4ab3f | 3338 | return -EPERM; |
ca94c442 LP |
3339 | |
3340 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3341 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3342 | return -EPERM; | |
37e4ab3f | 3343 | } |
1da177e4 | 3344 | |
725aad24 | 3345 | if (user) { |
b0ae1981 | 3346 | retval = security_task_setscheduler(p); |
725aad24 JF |
3347 | if (retval) |
3348 | return retval; | |
3349 | } | |
3350 | ||
b29739f9 IM |
3351 | /* |
3352 | * make sure no PI-waiters arrive (or leave) while we are | |
3353 | * changing the priority of the task: | |
0122ec5b | 3354 | * |
25985edc | 3355 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3356 | * runqueue lock must be held. |
3357 | */ | |
0122ec5b | 3358 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3359 | |
34f971f6 PZ |
3360 | /* |
3361 | * Changing the policy of the stop threads its a very bad idea | |
3362 | */ | |
3363 | if (p == rq->stop) { | |
0122ec5b | 3364 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3365 | return -EINVAL; |
3366 | } | |
3367 | ||
a51e9198 | 3368 | /* |
d6b1e911 TG |
3369 | * If not changing anything there's no need to proceed further, |
3370 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3371 | */ |
d50dde5a | 3372 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3373 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3374 | goto change; |
3375 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3376 | goto change; | |
aab03e05 DF |
3377 | if (dl_policy(policy)) |
3378 | goto change; | |
d50dde5a | 3379 | |
d6b1e911 | 3380 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3381 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3382 | return 0; |
3383 | } | |
d50dde5a | 3384 | change: |
a51e9198 | 3385 | |
dc61b1d6 | 3386 | if (user) { |
332ac17e | 3387 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3388 | /* |
3389 | * Do not allow realtime tasks into groups that have no runtime | |
3390 | * assigned. | |
3391 | */ | |
3392 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3393 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3394 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3395 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3396 | return -EPERM; |
3397 | } | |
dc61b1d6 | 3398 | #endif |
332ac17e DF |
3399 | #ifdef CONFIG_SMP |
3400 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3401 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3402 | |
3403 | /* | |
3404 | * Don't allow tasks with an affinity mask smaller than | |
3405 | * the entire root_domain to become SCHED_DEADLINE. We | |
3406 | * will also fail if there's no bandwidth available. | |
3407 | */ | |
e4099a5e PZ |
3408 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3409 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3410 | task_rq_unlock(rq, p, &flags); |
3411 | return -EPERM; | |
3412 | } | |
3413 | } | |
3414 | #endif | |
3415 | } | |
dc61b1d6 | 3416 | |
1da177e4 LT |
3417 | /* recheck policy now with rq lock held */ |
3418 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3419 | policy = oldpolicy = -1; | |
0122ec5b | 3420 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3421 | goto recheck; |
3422 | } | |
332ac17e DF |
3423 | |
3424 | /* | |
3425 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3426 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3427 | * is available. | |
3428 | */ | |
e4099a5e | 3429 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3430 | task_rq_unlock(rq, p, &flags); |
3431 | return -EBUSY; | |
3432 | } | |
3433 | ||
c365c292 TG |
3434 | p->sched_reset_on_fork = reset_on_fork; |
3435 | oldprio = p->prio; | |
3436 | ||
3437 | /* | |
3438 | * Special case for priority boosted tasks. | |
3439 | * | |
3440 | * If the new priority is lower or equal (user space view) | |
3441 | * than the current (boosted) priority, we just store the new | |
3442 | * normal parameters and do not touch the scheduler class and | |
3443 | * the runqueue. This will be done when the task deboost | |
3444 | * itself. | |
3445 | */ | |
3446 | if (rt_mutex_check_prio(p, newprio)) { | |
3447 | __setscheduler_params(p, attr); | |
3448 | task_rq_unlock(rq, p, &flags); | |
3449 | return 0; | |
3450 | } | |
3451 | ||
fd2f4419 | 3452 | on_rq = p->on_rq; |
051a1d1a | 3453 | running = task_current(rq, p); |
0e1f3483 | 3454 | if (on_rq) |
4ca9b72b | 3455 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3456 | if (running) |
3457 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3458 | |
83ab0aa0 | 3459 | prev_class = p->sched_class; |
d50dde5a | 3460 | __setscheduler(rq, p, attr); |
f6b53205 | 3461 | |
0e1f3483 HS |
3462 | if (running) |
3463 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3464 | if (on_rq) { |
3465 | /* | |
3466 | * We enqueue to tail when the priority of a task is | |
3467 | * increased (user space view). | |
3468 | */ | |
3469 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3470 | } | |
cb469845 | 3471 | |
da7a735e | 3472 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3473 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3474 | |
95e02ca9 TG |
3475 | rt_mutex_adjust_pi(p); |
3476 | ||
1da177e4 LT |
3477 | return 0; |
3478 | } | |
961ccddd | 3479 | |
7479f3c9 PZ |
3480 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3481 | const struct sched_param *param, bool check) | |
3482 | { | |
3483 | struct sched_attr attr = { | |
3484 | .sched_policy = policy, | |
3485 | .sched_priority = param->sched_priority, | |
3486 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3487 | }; | |
3488 | ||
3489 | /* | |
3490 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3491 | */ | |
3492 | if (policy & SCHED_RESET_ON_FORK) { | |
3493 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3494 | policy &= ~SCHED_RESET_ON_FORK; | |
3495 | attr.sched_policy = policy; | |
3496 | } | |
3497 | ||
3498 | return __sched_setscheduler(p, &attr, check); | |
3499 | } | |
961ccddd RR |
3500 | /** |
3501 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3502 | * @p: the task in question. | |
3503 | * @policy: new policy. | |
3504 | * @param: structure containing the new RT priority. | |
3505 | * | |
e69f6186 YB |
3506 | * Return: 0 on success. An error code otherwise. |
3507 | * | |
961ccddd RR |
3508 | * NOTE that the task may be already dead. |
3509 | */ | |
3510 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3511 | const struct sched_param *param) |
961ccddd | 3512 | { |
7479f3c9 | 3513 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3514 | } |
1da177e4 LT |
3515 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3516 | ||
d50dde5a DF |
3517 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3518 | { | |
3519 | return __sched_setscheduler(p, attr, true); | |
3520 | } | |
3521 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3522 | ||
961ccddd RR |
3523 | /** |
3524 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3525 | * @p: the task in question. | |
3526 | * @policy: new policy. | |
3527 | * @param: structure containing the new RT priority. | |
3528 | * | |
3529 | * Just like sched_setscheduler, only don't bother checking if the | |
3530 | * current context has permission. For example, this is needed in | |
3531 | * stop_machine(): we create temporary high priority worker threads, | |
3532 | * but our caller might not have that capability. | |
e69f6186 YB |
3533 | * |
3534 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3535 | */ |
3536 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3537 | const struct sched_param *param) |
961ccddd | 3538 | { |
7479f3c9 | 3539 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3540 | } |
3541 | ||
95cdf3b7 IM |
3542 | static int |
3543 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3544 | { |
1da177e4 LT |
3545 | struct sched_param lparam; |
3546 | struct task_struct *p; | |
36c8b586 | 3547 | int retval; |
1da177e4 LT |
3548 | |
3549 | if (!param || pid < 0) | |
3550 | return -EINVAL; | |
3551 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3552 | return -EFAULT; | |
5fe1d75f ON |
3553 | |
3554 | rcu_read_lock(); | |
3555 | retval = -ESRCH; | |
1da177e4 | 3556 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3557 | if (p != NULL) |
3558 | retval = sched_setscheduler(p, policy, &lparam); | |
3559 | rcu_read_unlock(); | |
36c8b586 | 3560 | |
1da177e4 LT |
3561 | return retval; |
3562 | } | |
3563 | ||
d50dde5a DF |
3564 | /* |
3565 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3566 | */ | |
3567 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3568 | struct sched_attr *attr) | |
3569 | { | |
3570 | u32 size; | |
3571 | int ret; | |
3572 | ||
3573 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3574 | return -EFAULT; | |
3575 | ||
3576 | /* | |
3577 | * zero the full structure, so that a short copy will be nice. | |
3578 | */ | |
3579 | memset(attr, 0, sizeof(*attr)); | |
3580 | ||
3581 | ret = get_user(size, &uattr->size); | |
3582 | if (ret) | |
3583 | return ret; | |
3584 | ||
3585 | if (size > PAGE_SIZE) /* silly large */ | |
3586 | goto err_size; | |
3587 | ||
3588 | if (!size) /* abi compat */ | |
3589 | size = SCHED_ATTR_SIZE_VER0; | |
3590 | ||
3591 | if (size < SCHED_ATTR_SIZE_VER0) | |
3592 | goto err_size; | |
3593 | ||
3594 | /* | |
3595 | * If we're handed a bigger struct than we know of, | |
3596 | * ensure all the unknown bits are 0 - i.e. new | |
3597 | * user-space does not rely on any kernel feature | |
3598 | * extensions we dont know about yet. | |
3599 | */ | |
3600 | if (size > sizeof(*attr)) { | |
3601 | unsigned char __user *addr; | |
3602 | unsigned char __user *end; | |
3603 | unsigned char val; | |
3604 | ||
3605 | addr = (void __user *)uattr + sizeof(*attr); | |
3606 | end = (void __user *)uattr + size; | |
3607 | ||
3608 | for (; addr < end; addr++) { | |
3609 | ret = get_user(val, addr); | |
3610 | if (ret) | |
3611 | return ret; | |
3612 | if (val) | |
3613 | goto err_size; | |
3614 | } | |
3615 | size = sizeof(*attr); | |
3616 | } | |
3617 | ||
3618 | ret = copy_from_user(attr, uattr, size); | |
3619 | if (ret) | |
3620 | return -EFAULT; | |
3621 | ||
3622 | /* | |
3623 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3624 | * to be strict and return an error on out-of-bounds values? | |
3625 | */ | |
3626 | attr->sched_nice = clamp(attr->sched_nice, -20, 19); | |
3627 | ||
3628 | out: | |
3629 | return ret; | |
3630 | ||
3631 | err_size: | |
3632 | put_user(sizeof(*attr), &uattr->size); | |
3633 | ret = -E2BIG; | |
3634 | goto out; | |
3635 | } | |
3636 | ||
1da177e4 LT |
3637 | /** |
3638 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3639 | * @pid: the pid in question. | |
3640 | * @policy: new policy. | |
3641 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3642 | * |
3643 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3644 | */ |
5add95d4 HC |
3645 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3646 | struct sched_param __user *, param) | |
1da177e4 | 3647 | { |
c21761f1 JB |
3648 | /* negative values for policy are not valid */ |
3649 | if (policy < 0) | |
3650 | return -EINVAL; | |
3651 | ||
1da177e4 LT |
3652 | return do_sched_setscheduler(pid, policy, param); |
3653 | } | |
3654 | ||
3655 | /** | |
3656 | * sys_sched_setparam - set/change the RT priority of a thread | |
3657 | * @pid: the pid in question. | |
3658 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3659 | * |
3660 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3661 | */ |
5add95d4 | 3662 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3663 | { |
3664 | return do_sched_setscheduler(pid, -1, param); | |
3665 | } | |
3666 | ||
d50dde5a DF |
3667 | /** |
3668 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3669 | * @pid: the pid in question. | |
5778fccf | 3670 | * @uattr: structure containing the extended parameters. |
d50dde5a DF |
3671 | */ |
3672 | SYSCALL_DEFINE2(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr) | |
3673 | { | |
3674 | struct sched_attr attr; | |
3675 | struct task_struct *p; | |
3676 | int retval; | |
3677 | ||
3678 | if (!uattr || pid < 0) | |
3679 | return -EINVAL; | |
3680 | ||
3681 | if (sched_copy_attr(uattr, &attr)) | |
3682 | return -EFAULT; | |
3683 | ||
3684 | rcu_read_lock(); | |
3685 | retval = -ESRCH; | |
3686 | p = find_process_by_pid(pid); | |
3687 | if (p != NULL) | |
3688 | retval = sched_setattr(p, &attr); | |
3689 | rcu_read_unlock(); | |
3690 | ||
3691 | return retval; | |
3692 | } | |
3693 | ||
1da177e4 LT |
3694 | /** |
3695 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3696 | * @pid: the pid in question. | |
e69f6186 YB |
3697 | * |
3698 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3699 | * code. | |
1da177e4 | 3700 | */ |
5add95d4 | 3701 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3702 | { |
36c8b586 | 3703 | struct task_struct *p; |
3a5c359a | 3704 | int retval; |
1da177e4 LT |
3705 | |
3706 | if (pid < 0) | |
3a5c359a | 3707 | return -EINVAL; |
1da177e4 LT |
3708 | |
3709 | retval = -ESRCH; | |
5fe85be0 | 3710 | rcu_read_lock(); |
1da177e4 LT |
3711 | p = find_process_by_pid(pid); |
3712 | if (p) { | |
3713 | retval = security_task_getscheduler(p); | |
3714 | if (!retval) | |
ca94c442 LP |
3715 | retval = p->policy |
3716 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3717 | } |
5fe85be0 | 3718 | rcu_read_unlock(); |
1da177e4 LT |
3719 | return retval; |
3720 | } | |
3721 | ||
3722 | /** | |
ca94c442 | 3723 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3724 | * @pid: the pid in question. |
3725 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3726 | * |
3727 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3728 | * code. | |
1da177e4 | 3729 | */ |
5add95d4 | 3730 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3731 | { |
3732 | struct sched_param lp; | |
36c8b586 | 3733 | struct task_struct *p; |
3a5c359a | 3734 | int retval; |
1da177e4 LT |
3735 | |
3736 | if (!param || pid < 0) | |
3a5c359a | 3737 | return -EINVAL; |
1da177e4 | 3738 | |
5fe85be0 | 3739 | rcu_read_lock(); |
1da177e4 LT |
3740 | p = find_process_by_pid(pid); |
3741 | retval = -ESRCH; | |
3742 | if (!p) | |
3743 | goto out_unlock; | |
3744 | ||
3745 | retval = security_task_getscheduler(p); | |
3746 | if (retval) | |
3747 | goto out_unlock; | |
3748 | ||
aab03e05 DF |
3749 | if (task_has_dl_policy(p)) { |
3750 | retval = -EINVAL; | |
3751 | goto out_unlock; | |
3752 | } | |
1da177e4 | 3753 | lp.sched_priority = p->rt_priority; |
5fe85be0 | 3754 | rcu_read_unlock(); |
1da177e4 LT |
3755 | |
3756 | /* | |
3757 | * This one might sleep, we cannot do it with a spinlock held ... | |
3758 | */ | |
3759 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3760 | ||
1da177e4 LT |
3761 | return retval; |
3762 | ||
3763 | out_unlock: | |
5fe85be0 | 3764 | rcu_read_unlock(); |
1da177e4 LT |
3765 | return retval; |
3766 | } | |
3767 | ||
d50dde5a DF |
3768 | static int sched_read_attr(struct sched_attr __user *uattr, |
3769 | struct sched_attr *attr, | |
3770 | unsigned int usize) | |
3771 | { | |
3772 | int ret; | |
3773 | ||
3774 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3775 | return -EFAULT; | |
3776 | ||
3777 | /* | |
3778 | * If we're handed a smaller struct than we know of, | |
3779 | * ensure all the unknown bits are 0 - i.e. old | |
3780 | * user-space does not get uncomplete information. | |
3781 | */ | |
3782 | if (usize < sizeof(*attr)) { | |
3783 | unsigned char *addr; | |
3784 | unsigned char *end; | |
3785 | ||
3786 | addr = (void *)attr + usize; | |
3787 | end = (void *)attr + sizeof(*attr); | |
3788 | ||
3789 | for (; addr < end; addr++) { | |
3790 | if (*addr) | |
3791 | goto err_size; | |
3792 | } | |
3793 | ||
3794 | attr->size = usize; | |
3795 | } | |
3796 | ||
3797 | ret = copy_to_user(uattr, attr, usize); | |
3798 | if (ret) | |
3799 | return -EFAULT; | |
3800 | ||
3801 | out: | |
3802 | return ret; | |
3803 | ||
3804 | err_size: | |
3805 | ret = -E2BIG; | |
3806 | goto out; | |
3807 | } | |
3808 | ||
3809 | /** | |
aab03e05 | 3810 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3811 | * @pid: the pid in question. |
5778fccf | 3812 | * @uattr: structure containing the extended parameters. |
d50dde5a DF |
3813 | * @size: sizeof(attr) for fwd/bwd comp. |
3814 | */ | |
3815 | SYSCALL_DEFINE3(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, | |
3816 | unsigned int, size) | |
3817 | { | |
3818 | struct sched_attr attr = { | |
3819 | .size = sizeof(struct sched_attr), | |
3820 | }; | |
3821 | struct task_struct *p; | |
3822 | int retval; | |
3823 | ||
3824 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
3825 | size < SCHED_ATTR_SIZE_VER0) | |
3826 | return -EINVAL; | |
3827 | ||
3828 | rcu_read_lock(); | |
3829 | p = find_process_by_pid(pid); | |
3830 | retval = -ESRCH; | |
3831 | if (!p) | |
3832 | goto out_unlock; | |
3833 | ||
3834 | retval = security_task_getscheduler(p); | |
3835 | if (retval) | |
3836 | goto out_unlock; | |
3837 | ||
3838 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3839 | if (p->sched_reset_on_fork) |
3840 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3841 | if (task_has_dl_policy(p)) |
3842 | __getparam_dl(p, &attr); | |
3843 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3844 | attr.sched_priority = p->rt_priority; |
3845 | else | |
d0ea0268 | 3846 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3847 | |
3848 | rcu_read_unlock(); | |
3849 | ||
3850 | retval = sched_read_attr(uattr, &attr, size); | |
3851 | return retval; | |
3852 | ||
3853 | out_unlock: | |
3854 | rcu_read_unlock(); | |
3855 | return retval; | |
3856 | } | |
3857 | ||
96f874e2 | 3858 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3859 | { |
5a16f3d3 | 3860 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3861 | struct task_struct *p; |
3862 | int retval; | |
1da177e4 | 3863 | |
23f5d142 | 3864 | rcu_read_lock(); |
1da177e4 LT |
3865 | |
3866 | p = find_process_by_pid(pid); | |
3867 | if (!p) { | |
23f5d142 | 3868 | rcu_read_unlock(); |
1da177e4 LT |
3869 | return -ESRCH; |
3870 | } | |
3871 | ||
23f5d142 | 3872 | /* Prevent p going away */ |
1da177e4 | 3873 | get_task_struct(p); |
23f5d142 | 3874 | rcu_read_unlock(); |
1da177e4 | 3875 | |
14a40ffc TH |
3876 | if (p->flags & PF_NO_SETAFFINITY) { |
3877 | retval = -EINVAL; | |
3878 | goto out_put_task; | |
3879 | } | |
5a16f3d3 RR |
3880 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3881 | retval = -ENOMEM; | |
3882 | goto out_put_task; | |
3883 | } | |
3884 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3885 | retval = -ENOMEM; | |
3886 | goto out_free_cpus_allowed; | |
3887 | } | |
1da177e4 | 3888 | retval = -EPERM; |
4c44aaaf EB |
3889 | if (!check_same_owner(p)) { |
3890 | rcu_read_lock(); | |
3891 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3892 | rcu_read_unlock(); | |
3893 | goto out_unlock; | |
3894 | } | |
3895 | rcu_read_unlock(); | |
3896 | } | |
1da177e4 | 3897 | |
b0ae1981 | 3898 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3899 | if (retval) |
3900 | goto out_unlock; | |
3901 | ||
e4099a5e PZ |
3902 | |
3903 | cpuset_cpus_allowed(p, cpus_allowed); | |
3904 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3905 | ||
332ac17e DF |
3906 | /* |
3907 | * Since bandwidth control happens on root_domain basis, | |
3908 | * if admission test is enabled, we only admit -deadline | |
3909 | * tasks allowed to run on all the CPUs in the task's | |
3910 | * root_domain. | |
3911 | */ | |
3912 | #ifdef CONFIG_SMP | |
3913 | if (task_has_dl_policy(p)) { | |
3914 | const struct cpumask *span = task_rq(p)->rd->span; | |
3915 | ||
e4099a5e | 3916 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
3917 | retval = -EBUSY; |
3918 | goto out_unlock; | |
3919 | } | |
3920 | } | |
3921 | #endif | |
49246274 | 3922 | again: |
5a16f3d3 | 3923 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3924 | |
8707d8b8 | 3925 | if (!retval) { |
5a16f3d3 RR |
3926 | cpuset_cpus_allowed(p, cpus_allowed); |
3927 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3928 | /* |
3929 | * We must have raced with a concurrent cpuset | |
3930 | * update. Just reset the cpus_allowed to the | |
3931 | * cpuset's cpus_allowed | |
3932 | */ | |
5a16f3d3 | 3933 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3934 | goto again; |
3935 | } | |
3936 | } | |
1da177e4 | 3937 | out_unlock: |
5a16f3d3 RR |
3938 | free_cpumask_var(new_mask); |
3939 | out_free_cpus_allowed: | |
3940 | free_cpumask_var(cpus_allowed); | |
3941 | out_put_task: | |
1da177e4 | 3942 | put_task_struct(p); |
1da177e4 LT |
3943 | return retval; |
3944 | } | |
3945 | ||
3946 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3947 | struct cpumask *new_mask) |
1da177e4 | 3948 | { |
96f874e2 RR |
3949 | if (len < cpumask_size()) |
3950 | cpumask_clear(new_mask); | |
3951 | else if (len > cpumask_size()) | |
3952 | len = cpumask_size(); | |
3953 | ||
1da177e4 LT |
3954 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3955 | } | |
3956 | ||
3957 | /** | |
3958 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3959 | * @pid: pid of the process | |
3960 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3961 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3962 | * |
3963 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3964 | */ |
5add95d4 HC |
3965 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3966 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3967 | { |
5a16f3d3 | 3968 | cpumask_var_t new_mask; |
1da177e4 LT |
3969 | int retval; |
3970 | ||
5a16f3d3 RR |
3971 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3972 | return -ENOMEM; | |
1da177e4 | 3973 | |
5a16f3d3 RR |
3974 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3975 | if (retval == 0) | |
3976 | retval = sched_setaffinity(pid, new_mask); | |
3977 | free_cpumask_var(new_mask); | |
3978 | return retval; | |
1da177e4 LT |
3979 | } |
3980 | ||
96f874e2 | 3981 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 3982 | { |
36c8b586 | 3983 | struct task_struct *p; |
31605683 | 3984 | unsigned long flags; |
1da177e4 | 3985 | int retval; |
1da177e4 | 3986 | |
23f5d142 | 3987 | rcu_read_lock(); |
1da177e4 LT |
3988 | |
3989 | retval = -ESRCH; | |
3990 | p = find_process_by_pid(pid); | |
3991 | if (!p) | |
3992 | goto out_unlock; | |
3993 | ||
e7834f8f DQ |
3994 | retval = security_task_getscheduler(p); |
3995 | if (retval) | |
3996 | goto out_unlock; | |
3997 | ||
013fdb80 | 3998 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 3999 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4000 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4001 | |
4002 | out_unlock: | |
23f5d142 | 4003 | rcu_read_unlock(); |
1da177e4 | 4004 | |
9531b62f | 4005 | return retval; |
1da177e4 LT |
4006 | } |
4007 | ||
4008 | /** | |
4009 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4010 | * @pid: pid of the process | |
4011 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4012 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4013 | * |
4014 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4015 | */ |
5add95d4 HC |
4016 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4017 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4018 | { |
4019 | int ret; | |
f17c8607 | 4020 | cpumask_var_t mask; |
1da177e4 | 4021 | |
84fba5ec | 4022 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4023 | return -EINVAL; |
4024 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4025 | return -EINVAL; |
4026 | ||
f17c8607 RR |
4027 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4028 | return -ENOMEM; | |
1da177e4 | 4029 | |
f17c8607 RR |
4030 | ret = sched_getaffinity(pid, mask); |
4031 | if (ret == 0) { | |
8bc037fb | 4032 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4033 | |
4034 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4035 | ret = -EFAULT; |
4036 | else | |
cd3d8031 | 4037 | ret = retlen; |
f17c8607 RR |
4038 | } |
4039 | free_cpumask_var(mask); | |
1da177e4 | 4040 | |
f17c8607 | 4041 | return ret; |
1da177e4 LT |
4042 | } |
4043 | ||
4044 | /** | |
4045 | * sys_sched_yield - yield the current processor to other threads. | |
4046 | * | |
dd41f596 IM |
4047 | * This function yields the current CPU to other tasks. If there are no |
4048 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4049 | * |
4050 | * Return: 0. | |
1da177e4 | 4051 | */ |
5add95d4 | 4052 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4053 | { |
70b97a7f | 4054 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4055 | |
2d72376b | 4056 | schedstat_inc(rq, yld_count); |
4530d7ab | 4057 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4058 | |
4059 | /* | |
4060 | * Since we are going to call schedule() anyway, there's | |
4061 | * no need to preempt or enable interrupts: | |
4062 | */ | |
4063 | __release(rq->lock); | |
8a25d5de | 4064 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4065 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4066 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4067 | |
4068 | schedule(); | |
4069 | ||
4070 | return 0; | |
4071 | } | |
4072 | ||
e7b38404 | 4073 | static void __cond_resched(void) |
1da177e4 | 4074 | { |
bdb43806 | 4075 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4076 | __schedule(); |
bdb43806 | 4077 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4078 | } |
4079 | ||
02b67cc3 | 4080 | int __sched _cond_resched(void) |
1da177e4 | 4081 | { |
d86ee480 | 4082 | if (should_resched()) { |
1da177e4 LT |
4083 | __cond_resched(); |
4084 | return 1; | |
4085 | } | |
4086 | return 0; | |
4087 | } | |
02b67cc3 | 4088 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4089 | |
4090 | /* | |
613afbf8 | 4091 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4092 | * call schedule, and on return reacquire the lock. |
4093 | * | |
41a2d6cf | 4094 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4095 | * operations here to prevent schedule() from being called twice (once via |
4096 | * spin_unlock(), once by hand). | |
4097 | */ | |
613afbf8 | 4098 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4099 | { |
d86ee480 | 4100 | int resched = should_resched(); |
6df3cecb JK |
4101 | int ret = 0; |
4102 | ||
f607c668 PZ |
4103 | lockdep_assert_held(lock); |
4104 | ||
95c354fe | 4105 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4106 | spin_unlock(lock); |
d86ee480 | 4107 | if (resched) |
95c354fe NP |
4108 | __cond_resched(); |
4109 | else | |
4110 | cpu_relax(); | |
6df3cecb | 4111 | ret = 1; |
1da177e4 | 4112 | spin_lock(lock); |
1da177e4 | 4113 | } |
6df3cecb | 4114 | return ret; |
1da177e4 | 4115 | } |
613afbf8 | 4116 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4117 | |
613afbf8 | 4118 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4119 | { |
4120 | BUG_ON(!in_softirq()); | |
4121 | ||
d86ee480 | 4122 | if (should_resched()) { |
98d82567 | 4123 | local_bh_enable(); |
1da177e4 LT |
4124 | __cond_resched(); |
4125 | local_bh_disable(); | |
4126 | return 1; | |
4127 | } | |
4128 | return 0; | |
4129 | } | |
613afbf8 | 4130 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4131 | |
1da177e4 LT |
4132 | /** |
4133 | * yield - yield the current processor to other threads. | |
4134 | * | |
8e3fabfd PZ |
4135 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4136 | * | |
4137 | * The scheduler is at all times free to pick the calling task as the most | |
4138 | * eligible task to run, if removing the yield() call from your code breaks | |
4139 | * it, its already broken. | |
4140 | * | |
4141 | * Typical broken usage is: | |
4142 | * | |
4143 | * while (!event) | |
4144 | * yield(); | |
4145 | * | |
4146 | * where one assumes that yield() will let 'the other' process run that will | |
4147 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4148 | * happen. Never use yield() as a progress guarantee!! | |
4149 | * | |
4150 | * If you want to use yield() to wait for something, use wait_event(). | |
4151 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4152 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4153 | */ |
4154 | void __sched yield(void) | |
4155 | { | |
4156 | set_current_state(TASK_RUNNING); | |
4157 | sys_sched_yield(); | |
4158 | } | |
1da177e4 LT |
4159 | EXPORT_SYMBOL(yield); |
4160 | ||
d95f4122 MG |
4161 | /** |
4162 | * yield_to - yield the current processor to another thread in | |
4163 | * your thread group, or accelerate that thread toward the | |
4164 | * processor it's on. | |
16addf95 RD |
4165 | * @p: target task |
4166 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4167 | * |
4168 | * It's the caller's job to ensure that the target task struct | |
4169 | * can't go away on us before we can do any checks. | |
4170 | * | |
e69f6186 | 4171 | * Return: |
7b270f60 PZ |
4172 | * true (>0) if we indeed boosted the target task. |
4173 | * false (0) if we failed to boost the target. | |
4174 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4175 | */ |
4176 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4177 | { | |
4178 | struct task_struct *curr = current; | |
4179 | struct rq *rq, *p_rq; | |
4180 | unsigned long flags; | |
c3c18640 | 4181 | int yielded = 0; |
d95f4122 MG |
4182 | |
4183 | local_irq_save(flags); | |
4184 | rq = this_rq(); | |
4185 | ||
4186 | again: | |
4187 | p_rq = task_rq(p); | |
7b270f60 PZ |
4188 | /* |
4189 | * If we're the only runnable task on the rq and target rq also | |
4190 | * has only one task, there's absolutely no point in yielding. | |
4191 | */ | |
4192 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4193 | yielded = -ESRCH; | |
4194 | goto out_irq; | |
4195 | } | |
4196 | ||
d95f4122 | 4197 | double_rq_lock(rq, p_rq); |
39e24d8f | 4198 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4199 | double_rq_unlock(rq, p_rq); |
4200 | goto again; | |
4201 | } | |
4202 | ||
4203 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4204 | goto out_unlock; |
d95f4122 MG |
4205 | |
4206 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4207 | goto out_unlock; |
d95f4122 MG |
4208 | |
4209 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4210 | goto out_unlock; |
d95f4122 MG |
4211 | |
4212 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4213 | if (yielded) { |
d95f4122 | 4214 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4215 | /* |
4216 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4217 | * fairness. | |
4218 | */ | |
4219 | if (preempt && rq != p_rq) | |
4220 | resched_task(p_rq->curr); | |
4221 | } | |
d95f4122 | 4222 | |
7b270f60 | 4223 | out_unlock: |
d95f4122 | 4224 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4225 | out_irq: |
d95f4122 MG |
4226 | local_irq_restore(flags); |
4227 | ||
7b270f60 | 4228 | if (yielded > 0) |
d95f4122 MG |
4229 | schedule(); |
4230 | ||
4231 | return yielded; | |
4232 | } | |
4233 | EXPORT_SYMBOL_GPL(yield_to); | |
4234 | ||
1da177e4 | 4235 | /* |
41a2d6cf | 4236 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4237 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4238 | */ |
4239 | void __sched io_schedule(void) | |
4240 | { | |
54d35f29 | 4241 | struct rq *rq = raw_rq(); |
1da177e4 | 4242 | |
0ff92245 | 4243 | delayacct_blkio_start(); |
1da177e4 | 4244 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4245 | blk_flush_plug(current); |
8f0dfc34 | 4246 | current->in_iowait = 1; |
1da177e4 | 4247 | schedule(); |
8f0dfc34 | 4248 | current->in_iowait = 0; |
1da177e4 | 4249 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4250 | delayacct_blkio_end(); |
1da177e4 | 4251 | } |
1da177e4 LT |
4252 | EXPORT_SYMBOL(io_schedule); |
4253 | ||
4254 | long __sched io_schedule_timeout(long timeout) | |
4255 | { | |
54d35f29 | 4256 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4257 | long ret; |
4258 | ||
0ff92245 | 4259 | delayacct_blkio_start(); |
1da177e4 | 4260 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4261 | blk_flush_plug(current); |
8f0dfc34 | 4262 | current->in_iowait = 1; |
1da177e4 | 4263 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4264 | current->in_iowait = 0; |
1da177e4 | 4265 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4266 | delayacct_blkio_end(); |
1da177e4 LT |
4267 | return ret; |
4268 | } | |
4269 | ||
4270 | /** | |
4271 | * sys_sched_get_priority_max - return maximum RT priority. | |
4272 | * @policy: scheduling class. | |
4273 | * | |
e69f6186 YB |
4274 | * Return: On success, this syscall returns the maximum |
4275 | * rt_priority that can be used by a given scheduling class. | |
4276 | * On failure, a negative error code is returned. | |
1da177e4 | 4277 | */ |
5add95d4 | 4278 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4279 | { |
4280 | int ret = -EINVAL; | |
4281 | ||
4282 | switch (policy) { | |
4283 | case SCHED_FIFO: | |
4284 | case SCHED_RR: | |
4285 | ret = MAX_USER_RT_PRIO-1; | |
4286 | break; | |
aab03e05 | 4287 | case SCHED_DEADLINE: |
1da177e4 | 4288 | case SCHED_NORMAL: |
b0a9499c | 4289 | case SCHED_BATCH: |
dd41f596 | 4290 | case SCHED_IDLE: |
1da177e4 LT |
4291 | ret = 0; |
4292 | break; | |
4293 | } | |
4294 | return ret; | |
4295 | } | |
4296 | ||
4297 | /** | |
4298 | * sys_sched_get_priority_min - return minimum RT priority. | |
4299 | * @policy: scheduling class. | |
4300 | * | |
e69f6186 YB |
4301 | * Return: On success, this syscall returns the minimum |
4302 | * rt_priority that can be used by a given scheduling class. | |
4303 | * On failure, a negative error code is returned. | |
1da177e4 | 4304 | */ |
5add95d4 | 4305 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4306 | { |
4307 | int ret = -EINVAL; | |
4308 | ||
4309 | switch (policy) { | |
4310 | case SCHED_FIFO: | |
4311 | case SCHED_RR: | |
4312 | ret = 1; | |
4313 | break; | |
aab03e05 | 4314 | case SCHED_DEADLINE: |
1da177e4 | 4315 | case SCHED_NORMAL: |
b0a9499c | 4316 | case SCHED_BATCH: |
dd41f596 | 4317 | case SCHED_IDLE: |
1da177e4 LT |
4318 | ret = 0; |
4319 | } | |
4320 | return ret; | |
4321 | } | |
4322 | ||
4323 | /** | |
4324 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4325 | * @pid: pid of the process. | |
4326 | * @interval: userspace pointer to the timeslice value. | |
4327 | * | |
4328 | * this syscall writes the default timeslice value of a given process | |
4329 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4330 | * |
4331 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4332 | * an error code. | |
1da177e4 | 4333 | */ |
17da2bd9 | 4334 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4335 | struct timespec __user *, interval) |
1da177e4 | 4336 | { |
36c8b586 | 4337 | struct task_struct *p; |
a4ec24b4 | 4338 | unsigned int time_slice; |
dba091b9 TG |
4339 | unsigned long flags; |
4340 | struct rq *rq; | |
3a5c359a | 4341 | int retval; |
1da177e4 | 4342 | struct timespec t; |
1da177e4 LT |
4343 | |
4344 | if (pid < 0) | |
3a5c359a | 4345 | return -EINVAL; |
1da177e4 LT |
4346 | |
4347 | retval = -ESRCH; | |
1a551ae7 | 4348 | rcu_read_lock(); |
1da177e4 LT |
4349 | p = find_process_by_pid(pid); |
4350 | if (!p) | |
4351 | goto out_unlock; | |
4352 | ||
4353 | retval = security_task_getscheduler(p); | |
4354 | if (retval) | |
4355 | goto out_unlock; | |
4356 | ||
dba091b9 | 4357 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4358 | time_slice = 0; |
4359 | if (p->sched_class->get_rr_interval) | |
4360 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4361 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4362 | |
1a551ae7 | 4363 | rcu_read_unlock(); |
a4ec24b4 | 4364 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4365 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4366 | return retval; |
3a5c359a | 4367 | |
1da177e4 | 4368 | out_unlock: |
1a551ae7 | 4369 | rcu_read_unlock(); |
1da177e4 LT |
4370 | return retval; |
4371 | } | |
4372 | ||
7c731e0a | 4373 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4374 | |
82a1fcb9 | 4375 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4376 | { |
1da177e4 | 4377 | unsigned long free = 0; |
4e79752c | 4378 | int ppid; |
36c8b586 | 4379 | unsigned state; |
1da177e4 | 4380 | |
1da177e4 | 4381 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4382 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4383 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4384 | #if BITS_PER_LONG == 32 |
1da177e4 | 4385 | if (state == TASK_RUNNING) |
3df0fc5b | 4386 | printk(KERN_CONT " running "); |
1da177e4 | 4387 | else |
3df0fc5b | 4388 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4389 | #else |
4390 | if (state == TASK_RUNNING) | |
3df0fc5b | 4391 | printk(KERN_CONT " running task "); |
1da177e4 | 4392 | else |
3df0fc5b | 4393 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4394 | #endif |
4395 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4396 | free = stack_not_used(p); |
1da177e4 | 4397 | #endif |
4e79752c PM |
4398 | rcu_read_lock(); |
4399 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4400 | rcu_read_unlock(); | |
3df0fc5b | 4401 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4402 | task_pid_nr(p), ppid, |
aa47b7e0 | 4403 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4404 | |
3d1cb205 | 4405 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4406 | show_stack(p, NULL); |
1da177e4 LT |
4407 | } |
4408 | ||
e59e2ae2 | 4409 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4410 | { |
36c8b586 | 4411 | struct task_struct *g, *p; |
1da177e4 | 4412 | |
4bd77321 | 4413 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4414 | printk(KERN_INFO |
4415 | " task PC stack pid father\n"); | |
1da177e4 | 4416 | #else |
3df0fc5b PZ |
4417 | printk(KERN_INFO |
4418 | " task PC stack pid father\n"); | |
1da177e4 | 4419 | #endif |
510f5acc | 4420 | rcu_read_lock(); |
1da177e4 LT |
4421 | do_each_thread(g, p) { |
4422 | /* | |
4423 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4424 | * console might take a lot of time: |
1da177e4 LT |
4425 | */ |
4426 | touch_nmi_watchdog(); | |
39bc89fd | 4427 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4428 | sched_show_task(p); |
1da177e4 LT |
4429 | } while_each_thread(g, p); |
4430 | ||
04c9167f JF |
4431 | touch_all_softlockup_watchdogs(); |
4432 | ||
dd41f596 IM |
4433 | #ifdef CONFIG_SCHED_DEBUG |
4434 | sysrq_sched_debug_show(); | |
4435 | #endif | |
510f5acc | 4436 | rcu_read_unlock(); |
e59e2ae2 IM |
4437 | /* |
4438 | * Only show locks if all tasks are dumped: | |
4439 | */ | |
93335a21 | 4440 | if (!state_filter) |
e59e2ae2 | 4441 | debug_show_all_locks(); |
1da177e4 LT |
4442 | } |
4443 | ||
0db0628d | 4444 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4445 | { |
dd41f596 | 4446 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4447 | } |
4448 | ||
f340c0d1 IM |
4449 | /** |
4450 | * init_idle - set up an idle thread for a given CPU | |
4451 | * @idle: task in question | |
4452 | * @cpu: cpu the idle task belongs to | |
4453 | * | |
4454 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4455 | * flag, to make booting more robust. | |
4456 | */ | |
0db0628d | 4457 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4458 | { |
70b97a7f | 4459 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4460 | unsigned long flags; |
4461 | ||
05fa785c | 4462 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4463 | |
5e1576ed | 4464 | __sched_fork(0, idle); |
06b83b5f | 4465 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4466 | idle->se.exec_start = sched_clock(); |
4467 | ||
1e1b6c51 | 4468 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4469 | /* |
4470 | * We're having a chicken and egg problem, even though we are | |
4471 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4472 | * lockdep check in task_group() will fail. | |
4473 | * | |
4474 | * Similar case to sched_fork(). / Alternatively we could | |
4475 | * use task_rq_lock() here and obtain the other rq->lock. | |
4476 | * | |
4477 | * Silence PROVE_RCU | |
4478 | */ | |
4479 | rcu_read_lock(); | |
dd41f596 | 4480 | __set_task_cpu(idle, cpu); |
6506cf6c | 4481 | rcu_read_unlock(); |
1da177e4 | 4482 | |
1da177e4 | 4483 | rq->curr = rq->idle = idle; |
77177856 | 4484 | idle->on_rq = 1; |
3ca7a440 PZ |
4485 | #if defined(CONFIG_SMP) |
4486 | idle->on_cpu = 1; | |
4866cde0 | 4487 | #endif |
05fa785c | 4488 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4489 | |
4490 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4491 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4492 | |
dd41f596 IM |
4493 | /* |
4494 | * The idle tasks have their own, simple scheduling class: | |
4495 | */ | |
4496 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4497 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4498 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4499 | #if defined(CONFIG_SMP) |
4500 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4501 | #endif | |
19978ca6 IM |
4502 | } |
4503 | ||
1da177e4 | 4504 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4505 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4506 | { | |
4507 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4508 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4509 | |
4510 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4511 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4512 | } |
4513 | ||
1da177e4 LT |
4514 | /* |
4515 | * This is how migration works: | |
4516 | * | |
969c7921 TH |
4517 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4518 | * stop_one_cpu(). | |
4519 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4520 | * off the CPU) | |
4521 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4522 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4523 | * it and puts it into the right queue. |
969c7921 TH |
4524 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4525 | * is done. | |
1da177e4 LT |
4526 | */ |
4527 | ||
4528 | /* | |
4529 | * Change a given task's CPU affinity. Migrate the thread to a | |
4530 | * proper CPU and schedule it away if the CPU it's executing on | |
4531 | * is removed from the allowed bitmask. | |
4532 | * | |
4533 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4534 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4535 | * call is not atomic; no spinlocks may be held. |
4536 | */ | |
96f874e2 | 4537 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4538 | { |
4539 | unsigned long flags; | |
70b97a7f | 4540 | struct rq *rq; |
969c7921 | 4541 | unsigned int dest_cpu; |
48f24c4d | 4542 | int ret = 0; |
1da177e4 LT |
4543 | |
4544 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4545 | |
db44fc01 YZ |
4546 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4547 | goto out; | |
4548 | ||
6ad4c188 | 4549 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4550 | ret = -EINVAL; |
4551 | goto out; | |
4552 | } | |
4553 | ||
1e1b6c51 | 4554 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4555 | |
1da177e4 | 4556 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4557 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4558 | goto out; |
4559 | ||
969c7921 | 4560 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4561 | if (p->on_rq) { |
969c7921 | 4562 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4563 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4564 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4565 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4566 | tlb_migrate_finish(p->mm); |
4567 | return 0; | |
4568 | } | |
4569 | out: | |
0122ec5b | 4570 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4571 | |
1da177e4 LT |
4572 | return ret; |
4573 | } | |
cd8ba7cd | 4574 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4575 | |
4576 | /* | |
41a2d6cf | 4577 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4578 | * this because either it can't run here any more (set_cpus_allowed() |
4579 | * away from this CPU, or CPU going down), or because we're | |
4580 | * attempting to rebalance this task on exec (sched_exec). | |
4581 | * | |
4582 | * So we race with normal scheduler movements, but that's OK, as long | |
4583 | * as the task is no longer on this CPU. | |
efc30814 KK |
4584 | * |
4585 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4586 | */ |
efc30814 | 4587 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4588 | { |
70b97a7f | 4589 | struct rq *rq_dest, *rq_src; |
e2912009 | 4590 | int ret = 0; |
1da177e4 | 4591 | |
e761b772 | 4592 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4593 | return ret; |
1da177e4 LT |
4594 | |
4595 | rq_src = cpu_rq(src_cpu); | |
4596 | rq_dest = cpu_rq(dest_cpu); | |
4597 | ||
0122ec5b | 4598 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4599 | double_rq_lock(rq_src, rq_dest); |
4600 | /* Already moved. */ | |
4601 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4602 | goto done; |
1da177e4 | 4603 | /* Affinity changed (again). */ |
fa17b507 | 4604 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4605 | goto fail; |
1da177e4 | 4606 | |
e2912009 PZ |
4607 | /* |
4608 | * If we're not on a rq, the next wake-up will ensure we're | |
4609 | * placed properly. | |
4610 | */ | |
fd2f4419 | 4611 | if (p->on_rq) { |
4ca9b72b | 4612 | dequeue_task(rq_src, p, 0); |
e2912009 | 4613 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4614 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4615 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4616 | } |
b1e38734 | 4617 | done: |
efc30814 | 4618 | ret = 1; |
b1e38734 | 4619 | fail: |
1da177e4 | 4620 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4621 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4622 | return ret; |
1da177e4 LT |
4623 | } |
4624 | ||
e6628d5b MG |
4625 | #ifdef CONFIG_NUMA_BALANCING |
4626 | /* Migrate current task p to target_cpu */ | |
4627 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4628 | { | |
4629 | struct migration_arg arg = { p, target_cpu }; | |
4630 | int curr_cpu = task_cpu(p); | |
4631 | ||
4632 | if (curr_cpu == target_cpu) | |
4633 | return 0; | |
4634 | ||
4635 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4636 | return -EINVAL; | |
4637 | ||
4638 | /* TODO: This is not properly updating schedstats */ | |
4639 | ||
286549dc | 4640 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4641 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4642 | } | |
0ec8aa00 PZ |
4643 | |
4644 | /* | |
4645 | * Requeue a task on a given node and accurately track the number of NUMA | |
4646 | * tasks on the runqueues | |
4647 | */ | |
4648 | void sched_setnuma(struct task_struct *p, int nid) | |
4649 | { | |
4650 | struct rq *rq; | |
4651 | unsigned long flags; | |
4652 | bool on_rq, running; | |
4653 | ||
4654 | rq = task_rq_lock(p, &flags); | |
4655 | on_rq = p->on_rq; | |
4656 | running = task_current(rq, p); | |
4657 | ||
4658 | if (on_rq) | |
4659 | dequeue_task(rq, p, 0); | |
4660 | if (running) | |
4661 | p->sched_class->put_prev_task(rq, p); | |
4662 | ||
4663 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4664 | |
4665 | if (running) | |
4666 | p->sched_class->set_curr_task(rq); | |
4667 | if (on_rq) | |
4668 | enqueue_task(rq, p, 0); | |
4669 | task_rq_unlock(rq, p, &flags); | |
4670 | } | |
e6628d5b MG |
4671 | #endif |
4672 | ||
1da177e4 | 4673 | /* |
969c7921 TH |
4674 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4675 | * and performs thread migration by bumping thread off CPU then | |
4676 | * 'pushing' onto another runqueue. | |
1da177e4 | 4677 | */ |
969c7921 | 4678 | static int migration_cpu_stop(void *data) |
1da177e4 | 4679 | { |
969c7921 | 4680 | struct migration_arg *arg = data; |
f7b4cddc | 4681 | |
969c7921 TH |
4682 | /* |
4683 | * The original target cpu might have gone down and we might | |
4684 | * be on another cpu but it doesn't matter. | |
4685 | */ | |
f7b4cddc | 4686 | local_irq_disable(); |
969c7921 | 4687 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4688 | local_irq_enable(); |
1da177e4 | 4689 | return 0; |
f7b4cddc ON |
4690 | } |
4691 | ||
1da177e4 | 4692 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4693 | |
054b9108 | 4694 | /* |
48c5ccae PZ |
4695 | * Ensures that the idle task is using init_mm right before its cpu goes |
4696 | * offline. | |
054b9108 | 4697 | */ |
48c5ccae | 4698 | void idle_task_exit(void) |
1da177e4 | 4699 | { |
48c5ccae | 4700 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4701 | |
48c5ccae | 4702 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4703 | |
48c5ccae PZ |
4704 | if (mm != &init_mm) |
4705 | switch_mm(mm, &init_mm, current); | |
4706 | mmdrop(mm); | |
1da177e4 LT |
4707 | } |
4708 | ||
4709 | /* | |
5d180232 PZ |
4710 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4711 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4712 | * nr_active count is stable. | |
4713 | * | |
4714 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4715 | */ |
5d180232 | 4716 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4717 | { |
5d180232 PZ |
4718 | long delta = calc_load_fold_active(rq); |
4719 | if (delta) | |
4720 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4721 | } |
4722 | ||
3f1d2a31 PZ |
4723 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4724 | { | |
4725 | } | |
4726 | ||
4727 | static const struct sched_class fake_sched_class = { | |
4728 | .put_prev_task = put_prev_task_fake, | |
4729 | }; | |
4730 | ||
4731 | static struct task_struct fake_task = { | |
4732 | /* | |
4733 | * Avoid pull_{rt,dl}_task() | |
4734 | */ | |
4735 | .prio = MAX_PRIO + 1, | |
4736 | .sched_class = &fake_sched_class, | |
4737 | }; | |
4738 | ||
48f24c4d | 4739 | /* |
48c5ccae PZ |
4740 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4741 | * try_to_wake_up()->select_task_rq(). | |
4742 | * | |
4743 | * Called with rq->lock held even though we'er in stop_machine() and | |
4744 | * there's no concurrency possible, we hold the required locks anyway | |
4745 | * because of lock validation efforts. | |
1da177e4 | 4746 | */ |
48c5ccae | 4747 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4748 | { |
70b97a7f | 4749 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4750 | struct task_struct *next, *stop = rq->stop; |
4751 | int dest_cpu; | |
1da177e4 LT |
4752 | |
4753 | /* | |
48c5ccae PZ |
4754 | * Fudge the rq selection such that the below task selection loop |
4755 | * doesn't get stuck on the currently eligible stop task. | |
4756 | * | |
4757 | * We're currently inside stop_machine() and the rq is either stuck | |
4758 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4759 | * either way we should never end up calling schedule() until we're | |
4760 | * done here. | |
1da177e4 | 4761 | */ |
48c5ccae | 4762 | rq->stop = NULL; |
48f24c4d | 4763 | |
77bd3970 FW |
4764 | /* |
4765 | * put_prev_task() and pick_next_task() sched | |
4766 | * class method both need to have an up-to-date | |
4767 | * value of rq->clock[_task] | |
4768 | */ | |
4769 | update_rq_clock(rq); | |
4770 | ||
dd41f596 | 4771 | for ( ; ; ) { |
48c5ccae PZ |
4772 | /* |
4773 | * There's this thread running, bail when that's the only | |
4774 | * remaining thread. | |
4775 | */ | |
4776 | if (rq->nr_running == 1) | |
dd41f596 | 4777 | break; |
48c5ccae | 4778 | |
3f1d2a31 | 4779 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4780 | BUG_ON(!next); |
79c53799 | 4781 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4782 | |
48c5ccae PZ |
4783 | /* Find suitable destination for @next, with force if needed. */ |
4784 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4785 | raw_spin_unlock(&rq->lock); | |
4786 | ||
4787 | __migrate_task(next, dead_cpu, dest_cpu); | |
4788 | ||
4789 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4790 | } |
dce48a84 | 4791 | |
48c5ccae | 4792 | rq->stop = stop; |
dce48a84 | 4793 | } |
48c5ccae | 4794 | |
1da177e4 LT |
4795 | #endif /* CONFIG_HOTPLUG_CPU */ |
4796 | ||
e692ab53 NP |
4797 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4798 | ||
4799 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4800 | { |
4801 | .procname = "sched_domain", | |
c57baf1e | 4802 | .mode = 0555, |
e0361851 | 4803 | }, |
56992309 | 4804 | {} |
e692ab53 NP |
4805 | }; |
4806 | ||
4807 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4808 | { |
4809 | .procname = "kernel", | |
c57baf1e | 4810 | .mode = 0555, |
e0361851 AD |
4811 | .child = sd_ctl_dir, |
4812 | }, | |
56992309 | 4813 | {} |
e692ab53 NP |
4814 | }; |
4815 | ||
4816 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4817 | { | |
4818 | struct ctl_table *entry = | |
5cf9f062 | 4819 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4820 | |
e692ab53 NP |
4821 | return entry; |
4822 | } | |
4823 | ||
6382bc90 MM |
4824 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4825 | { | |
cd790076 | 4826 | struct ctl_table *entry; |
6382bc90 | 4827 | |
cd790076 MM |
4828 | /* |
4829 | * In the intermediate directories, both the child directory and | |
4830 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4831 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4832 | * static strings and all have proc handlers. |
4833 | */ | |
4834 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4835 | if (entry->child) |
4836 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4837 | if (entry->proc_handler == NULL) |
4838 | kfree(entry->procname); | |
4839 | } | |
6382bc90 MM |
4840 | |
4841 | kfree(*tablep); | |
4842 | *tablep = NULL; | |
4843 | } | |
4844 | ||
201c373e | 4845 | static int min_load_idx = 0; |
fd9b86d3 | 4846 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4847 | |
e692ab53 | 4848 | static void |
e0361851 | 4849 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4850 | const char *procname, void *data, int maxlen, |
201c373e NK |
4851 | umode_t mode, proc_handler *proc_handler, |
4852 | bool load_idx) | |
e692ab53 | 4853 | { |
e692ab53 NP |
4854 | entry->procname = procname; |
4855 | entry->data = data; | |
4856 | entry->maxlen = maxlen; | |
4857 | entry->mode = mode; | |
4858 | entry->proc_handler = proc_handler; | |
201c373e NK |
4859 | |
4860 | if (load_idx) { | |
4861 | entry->extra1 = &min_load_idx; | |
4862 | entry->extra2 = &max_load_idx; | |
4863 | } | |
e692ab53 NP |
4864 | } |
4865 | ||
4866 | static struct ctl_table * | |
4867 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4868 | { | |
37e6bae8 | 4869 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4870 | |
ad1cdc1d MM |
4871 | if (table == NULL) |
4872 | return NULL; | |
4873 | ||
e0361851 | 4874 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4875 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4876 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4877 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4878 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4879 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4880 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4881 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4882 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4883 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4884 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4885 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4886 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4887 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4888 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4889 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4890 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4891 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4892 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4893 | &sd->cache_nice_tries, |
201c373e | 4894 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4895 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4896 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4897 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4898 | &sd->max_newidle_lb_cost, | |
4899 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4900 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4901 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4902 | /* &table[13] is terminator */ |
e692ab53 NP |
4903 | |
4904 | return table; | |
4905 | } | |
4906 | ||
be7002e6 | 4907 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4908 | { |
4909 | struct ctl_table *entry, *table; | |
4910 | struct sched_domain *sd; | |
4911 | int domain_num = 0, i; | |
4912 | char buf[32]; | |
4913 | ||
4914 | for_each_domain(cpu, sd) | |
4915 | domain_num++; | |
4916 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4917 | if (table == NULL) |
4918 | return NULL; | |
e692ab53 NP |
4919 | |
4920 | i = 0; | |
4921 | for_each_domain(cpu, sd) { | |
4922 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4923 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4924 | entry->mode = 0555; |
e692ab53 NP |
4925 | entry->child = sd_alloc_ctl_domain_table(sd); |
4926 | entry++; | |
4927 | i++; | |
4928 | } | |
4929 | return table; | |
4930 | } | |
4931 | ||
4932 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4933 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4934 | { |
6ad4c188 | 4935 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4936 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4937 | char buf[32]; | |
4938 | ||
7378547f MM |
4939 | WARN_ON(sd_ctl_dir[0].child); |
4940 | sd_ctl_dir[0].child = entry; | |
4941 | ||
ad1cdc1d MM |
4942 | if (entry == NULL) |
4943 | return; | |
4944 | ||
6ad4c188 | 4945 | for_each_possible_cpu(i) { |
e692ab53 | 4946 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4947 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4948 | entry->mode = 0555; |
e692ab53 | 4949 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4950 | entry++; |
e692ab53 | 4951 | } |
7378547f MM |
4952 | |
4953 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4954 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4955 | } | |
6382bc90 | 4956 | |
7378547f | 4957 | /* may be called multiple times per register */ |
6382bc90 MM |
4958 | static void unregister_sched_domain_sysctl(void) |
4959 | { | |
7378547f MM |
4960 | if (sd_sysctl_header) |
4961 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4962 | sd_sysctl_header = NULL; |
7378547f MM |
4963 | if (sd_ctl_dir[0].child) |
4964 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4965 | } |
e692ab53 | 4966 | #else |
6382bc90 MM |
4967 | static void register_sched_domain_sysctl(void) |
4968 | { | |
4969 | } | |
4970 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4971 | { |
4972 | } | |
4973 | #endif | |
4974 | ||
1f11eb6a GH |
4975 | static void set_rq_online(struct rq *rq) |
4976 | { | |
4977 | if (!rq->online) { | |
4978 | const struct sched_class *class; | |
4979 | ||
c6c4927b | 4980 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4981 | rq->online = 1; |
4982 | ||
4983 | for_each_class(class) { | |
4984 | if (class->rq_online) | |
4985 | class->rq_online(rq); | |
4986 | } | |
4987 | } | |
4988 | } | |
4989 | ||
4990 | static void set_rq_offline(struct rq *rq) | |
4991 | { | |
4992 | if (rq->online) { | |
4993 | const struct sched_class *class; | |
4994 | ||
4995 | for_each_class(class) { | |
4996 | if (class->rq_offline) | |
4997 | class->rq_offline(rq); | |
4998 | } | |
4999 | ||
c6c4927b | 5000 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5001 | rq->online = 0; |
5002 | } | |
5003 | } | |
5004 | ||
1da177e4 LT |
5005 | /* |
5006 | * migration_call - callback that gets triggered when a CPU is added. | |
5007 | * Here we can start up the necessary migration thread for the new CPU. | |
5008 | */ | |
0db0628d | 5009 | static int |
48f24c4d | 5010 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5011 | { |
48f24c4d | 5012 | int cpu = (long)hcpu; |
1da177e4 | 5013 | unsigned long flags; |
969c7921 | 5014 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5015 | |
48c5ccae | 5016 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5017 | |
1da177e4 | 5018 | case CPU_UP_PREPARE: |
a468d389 | 5019 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5020 | break; |
48f24c4d | 5021 | |
1da177e4 | 5022 | case CPU_ONLINE: |
1f94ef59 | 5023 | /* Update our root-domain */ |
05fa785c | 5024 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5025 | if (rq->rd) { |
c6c4927b | 5026 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5027 | |
5028 | set_rq_online(rq); | |
1f94ef59 | 5029 | } |
05fa785c | 5030 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5031 | break; |
48f24c4d | 5032 | |
1da177e4 | 5033 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5034 | case CPU_DYING: |
317f3941 | 5035 | sched_ttwu_pending(); |
57d885fe | 5036 | /* Update our root-domain */ |
05fa785c | 5037 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5038 | if (rq->rd) { |
c6c4927b | 5039 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5040 | set_rq_offline(rq); |
57d885fe | 5041 | } |
48c5ccae PZ |
5042 | migrate_tasks(cpu); |
5043 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5044 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5045 | break; |
48c5ccae | 5046 | |
5d180232 | 5047 | case CPU_DEAD: |
f319da0c | 5048 | calc_load_migrate(rq); |
57d885fe | 5049 | break; |
1da177e4 LT |
5050 | #endif |
5051 | } | |
49c022e6 PZ |
5052 | |
5053 | update_max_interval(); | |
5054 | ||
1da177e4 LT |
5055 | return NOTIFY_OK; |
5056 | } | |
5057 | ||
f38b0820 PM |
5058 | /* |
5059 | * Register at high priority so that task migration (migrate_all_tasks) | |
5060 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5061 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5062 | */ |
0db0628d | 5063 | static struct notifier_block migration_notifier = { |
1da177e4 | 5064 | .notifier_call = migration_call, |
50a323b7 | 5065 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5066 | }; |
5067 | ||
0db0628d | 5068 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5069 | unsigned long action, void *hcpu) |
5070 | { | |
5071 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5072 | case CPU_STARTING: |
3a101d05 TH |
5073 | case CPU_DOWN_FAILED: |
5074 | set_cpu_active((long)hcpu, true); | |
5075 | return NOTIFY_OK; | |
5076 | default: | |
5077 | return NOTIFY_DONE; | |
5078 | } | |
5079 | } | |
5080 | ||
0db0628d | 5081 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5082 | unsigned long action, void *hcpu) |
5083 | { | |
de212f18 PZ |
5084 | unsigned long flags; |
5085 | long cpu = (long)hcpu; | |
5086 | ||
3a101d05 TH |
5087 | switch (action & ~CPU_TASKS_FROZEN) { |
5088 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5089 | set_cpu_active(cpu, false); |
5090 | ||
5091 | /* explicitly allow suspend */ | |
5092 | if (!(action & CPU_TASKS_FROZEN)) { | |
5093 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5094 | bool overflow; | |
5095 | int cpus; | |
5096 | ||
5097 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5098 | cpus = dl_bw_cpus(cpu); | |
5099 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5100 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5101 | ||
5102 | if (overflow) | |
5103 | return notifier_from_errno(-EBUSY); | |
5104 | } | |
3a101d05 | 5105 | return NOTIFY_OK; |
3a101d05 | 5106 | } |
de212f18 PZ |
5107 | |
5108 | return NOTIFY_DONE; | |
3a101d05 TH |
5109 | } |
5110 | ||
7babe8db | 5111 | static int __init migration_init(void) |
1da177e4 LT |
5112 | { |
5113 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5114 | int err; |
48f24c4d | 5115 | |
3a101d05 | 5116 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5117 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5118 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5119 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5120 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5121 | |
3a101d05 TH |
5122 | /* Register cpu active notifiers */ |
5123 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5124 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5125 | ||
a004cd42 | 5126 | return 0; |
1da177e4 | 5127 | } |
7babe8db | 5128 | early_initcall(migration_init); |
1da177e4 LT |
5129 | #endif |
5130 | ||
5131 | #ifdef CONFIG_SMP | |
476f3534 | 5132 | |
4cb98839 PZ |
5133 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5134 | ||
3e9830dc | 5135 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5136 | |
d039ac60 | 5137 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5138 | |
d039ac60 | 5139 | static int __init sched_debug_setup(char *str) |
f6630114 | 5140 | { |
d039ac60 | 5141 | sched_debug_enabled = 1; |
f6630114 MT |
5142 | |
5143 | return 0; | |
5144 | } | |
d039ac60 PZ |
5145 | early_param("sched_debug", sched_debug_setup); |
5146 | ||
5147 | static inline bool sched_debug(void) | |
5148 | { | |
5149 | return sched_debug_enabled; | |
5150 | } | |
f6630114 | 5151 | |
7c16ec58 | 5152 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5153 | struct cpumask *groupmask) |
1da177e4 | 5154 | { |
4dcf6aff | 5155 | struct sched_group *group = sd->groups; |
434d53b0 | 5156 | char str[256]; |
1da177e4 | 5157 | |
968ea6d8 | 5158 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5159 | cpumask_clear(groupmask); |
4dcf6aff IM |
5160 | |
5161 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5162 | ||
5163 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5164 | printk("does not load-balance\n"); |
4dcf6aff | 5165 | if (sd->parent) |
3df0fc5b PZ |
5166 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5167 | " has parent"); | |
4dcf6aff | 5168 | return -1; |
41c7ce9a NP |
5169 | } |
5170 | ||
3df0fc5b | 5171 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5172 | |
758b2cdc | 5173 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5174 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5175 | "CPU%d\n", cpu); | |
4dcf6aff | 5176 | } |
758b2cdc | 5177 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5178 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5179 | " CPU%d\n", cpu); | |
4dcf6aff | 5180 | } |
1da177e4 | 5181 | |
4dcf6aff | 5182 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5183 | do { |
4dcf6aff | 5184 | if (!group) { |
3df0fc5b PZ |
5185 | printk("\n"); |
5186 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5187 | break; |
5188 | } | |
5189 | ||
c3decf0d PZ |
5190 | /* |
5191 | * Even though we initialize ->power to something semi-sane, | |
5192 | * we leave power_orig unset. This allows us to detect if | |
5193 | * domain iteration is still funny without causing /0 traps. | |
5194 | */ | |
5195 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5196 | printk(KERN_CONT "\n"); |
5197 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5198 | "set\n"); | |
4dcf6aff IM |
5199 | break; |
5200 | } | |
1da177e4 | 5201 | |
758b2cdc | 5202 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5203 | printk(KERN_CONT "\n"); |
5204 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5205 | break; |
5206 | } | |
1da177e4 | 5207 | |
cb83b629 PZ |
5208 | if (!(sd->flags & SD_OVERLAP) && |
5209 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5210 | printk(KERN_CONT "\n"); |
5211 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5212 | break; |
5213 | } | |
1da177e4 | 5214 | |
758b2cdc | 5215 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5216 | |
968ea6d8 | 5217 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5218 | |
3df0fc5b | 5219 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5220 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5221 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5222 | group->sgp->power); |
381512cf | 5223 | } |
1da177e4 | 5224 | |
4dcf6aff IM |
5225 | group = group->next; |
5226 | } while (group != sd->groups); | |
3df0fc5b | 5227 | printk(KERN_CONT "\n"); |
1da177e4 | 5228 | |
758b2cdc | 5229 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5230 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5231 | |
758b2cdc RR |
5232 | if (sd->parent && |
5233 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5234 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5235 | "of domain->span\n"); | |
4dcf6aff IM |
5236 | return 0; |
5237 | } | |
1da177e4 | 5238 | |
4dcf6aff IM |
5239 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5240 | { | |
5241 | int level = 0; | |
1da177e4 | 5242 | |
d039ac60 | 5243 | if (!sched_debug_enabled) |
f6630114 MT |
5244 | return; |
5245 | ||
4dcf6aff IM |
5246 | if (!sd) { |
5247 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5248 | return; | |
5249 | } | |
1da177e4 | 5250 | |
4dcf6aff IM |
5251 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5252 | ||
5253 | for (;;) { | |
4cb98839 | 5254 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5255 | break; |
1da177e4 LT |
5256 | level++; |
5257 | sd = sd->parent; | |
33859f7f | 5258 | if (!sd) |
4dcf6aff IM |
5259 | break; |
5260 | } | |
1da177e4 | 5261 | } |
6d6bc0ad | 5262 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5263 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5264 | static inline bool sched_debug(void) |
5265 | { | |
5266 | return false; | |
5267 | } | |
6d6bc0ad | 5268 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5269 | |
1a20ff27 | 5270 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5271 | { |
758b2cdc | 5272 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5273 | return 1; |
5274 | ||
5275 | /* Following flags need at least 2 groups */ | |
5276 | if (sd->flags & (SD_LOAD_BALANCE | | |
5277 | SD_BALANCE_NEWIDLE | | |
5278 | SD_BALANCE_FORK | | |
89c4710e SS |
5279 | SD_BALANCE_EXEC | |
5280 | SD_SHARE_CPUPOWER | | |
5281 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5282 | if (sd->groups != sd->groups->next) |
5283 | return 0; | |
5284 | } | |
5285 | ||
5286 | /* Following flags don't use groups */ | |
c88d5910 | 5287 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5288 | return 0; |
5289 | ||
5290 | return 1; | |
5291 | } | |
5292 | ||
48f24c4d IM |
5293 | static int |
5294 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5295 | { |
5296 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5297 | ||
5298 | if (sd_degenerate(parent)) | |
5299 | return 1; | |
5300 | ||
758b2cdc | 5301 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5302 | return 0; |
5303 | ||
245af2c7 SS |
5304 | /* Flags needing groups don't count if only 1 group in parent */ |
5305 | if (parent->groups == parent->groups->next) { | |
5306 | pflags &= ~(SD_LOAD_BALANCE | | |
5307 | SD_BALANCE_NEWIDLE | | |
5308 | SD_BALANCE_FORK | | |
89c4710e SS |
5309 | SD_BALANCE_EXEC | |
5310 | SD_SHARE_CPUPOWER | | |
10866e62 PZ |
5311 | SD_SHARE_PKG_RESOURCES | |
5312 | SD_PREFER_SIBLING); | |
5436499e KC |
5313 | if (nr_node_ids == 1) |
5314 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5315 | } |
5316 | if (~cflags & pflags) | |
5317 | return 0; | |
5318 | ||
5319 | return 1; | |
5320 | } | |
5321 | ||
dce840a0 | 5322 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5323 | { |
dce840a0 | 5324 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5325 | |
68e74568 | 5326 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5327 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5328 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5329 | free_cpumask_var(rd->rto_mask); |
5330 | free_cpumask_var(rd->online); | |
5331 | free_cpumask_var(rd->span); | |
5332 | kfree(rd); | |
5333 | } | |
5334 | ||
57d885fe GH |
5335 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5336 | { | |
a0490fa3 | 5337 | struct root_domain *old_rd = NULL; |
57d885fe | 5338 | unsigned long flags; |
57d885fe | 5339 | |
05fa785c | 5340 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5341 | |
5342 | if (rq->rd) { | |
a0490fa3 | 5343 | old_rd = rq->rd; |
57d885fe | 5344 | |
c6c4927b | 5345 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5346 | set_rq_offline(rq); |
57d885fe | 5347 | |
c6c4927b | 5348 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5349 | |
a0490fa3 | 5350 | /* |
0515973f | 5351 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5352 | * set old_rd to NULL to skip the freeing later |
5353 | * in this function: | |
5354 | */ | |
5355 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5356 | old_rd = NULL; | |
57d885fe GH |
5357 | } |
5358 | ||
5359 | atomic_inc(&rd->refcount); | |
5360 | rq->rd = rd; | |
5361 | ||
c6c4927b | 5362 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5363 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5364 | set_rq_online(rq); |
57d885fe | 5365 | |
05fa785c | 5366 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5367 | |
5368 | if (old_rd) | |
dce840a0 | 5369 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5370 | } |
5371 | ||
68c38fc3 | 5372 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5373 | { |
5374 | memset(rd, 0, sizeof(*rd)); | |
5375 | ||
68c38fc3 | 5376 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5377 | goto out; |
68c38fc3 | 5378 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5379 | goto free_span; |
1baca4ce | 5380 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5381 | goto free_online; |
1baca4ce JL |
5382 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5383 | goto free_dlo_mask; | |
6e0534f2 | 5384 | |
332ac17e | 5385 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5386 | if (cpudl_init(&rd->cpudl) != 0) |
5387 | goto free_dlo_mask; | |
332ac17e | 5388 | |
68c38fc3 | 5389 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5390 | goto free_rto_mask; |
c6c4927b | 5391 | return 0; |
6e0534f2 | 5392 | |
68e74568 RR |
5393 | free_rto_mask: |
5394 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5395 | free_dlo_mask: |
5396 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5397 | free_online: |
5398 | free_cpumask_var(rd->online); | |
5399 | free_span: | |
5400 | free_cpumask_var(rd->span); | |
0c910d28 | 5401 | out: |
c6c4927b | 5402 | return -ENOMEM; |
57d885fe GH |
5403 | } |
5404 | ||
029632fb PZ |
5405 | /* |
5406 | * By default the system creates a single root-domain with all cpus as | |
5407 | * members (mimicking the global state we have today). | |
5408 | */ | |
5409 | struct root_domain def_root_domain; | |
5410 | ||
57d885fe GH |
5411 | static void init_defrootdomain(void) |
5412 | { | |
68c38fc3 | 5413 | init_rootdomain(&def_root_domain); |
c6c4927b | 5414 | |
57d885fe GH |
5415 | atomic_set(&def_root_domain.refcount, 1); |
5416 | } | |
5417 | ||
dc938520 | 5418 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5419 | { |
5420 | struct root_domain *rd; | |
5421 | ||
5422 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5423 | if (!rd) | |
5424 | return NULL; | |
5425 | ||
68c38fc3 | 5426 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5427 | kfree(rd); |
5428 | return NULL; | |
5429 | } | |
57d885fe GH |
5430 | |
5431 | return rd; | |
5432 | } | |
5433 | ||
e3589f6c PZ |
5434 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5435 | { | |
5436 | struct sched_group *tmp, *first; | |
5437 | ||
5438 | if (!sg) | |
5439 | return; | |
5440 | ||
5441 | first = sg; | |
5442 | do { | |
5443 | tmp = sg->next; | |
5444 | ||
5445 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5446 | kfree(sg->sgp); | |
5447 | ||
5448 | kfree(sg); | |
5449 | sg = tmp; | |
5450 | } while (sg != first); | |
5451 | } | |
5452 | ||
dce840a0 PZ |
5453 | static void free_sched_domain(struct rcu_head *rcu) |
5454 | { | |
5455 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5456 | |
5457 | /* | |
5458 | * If its an overlapping domain it has private groups, iterate and | |
5459 | * nuke them all. | |
5460 | */ | |
5461 | if (sd->flags & SD_OVERLAP) { | |
5462 | free_sched_groups(sd->groups, 1); | |
5463 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5464 | kfree(sd->groups->sgp); |
dce840a0 | 5465 | kfree(sd->groups); |
9c3f75cb | 5466 | } |
dce840a0 PZ |
5467 | kfree(sd); |
5468 | } | |
5469 | ||
5470 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5471 | { | |
5472 | call_rcu(&sd->rcu, free_sched_domain); | |
5473 | } | |
5474 | ||
5475 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5476 | { | |
5477 | for (; sd; sd = sd->parent) | |
5478 | destroy_sched_domain(sd, cpu); | |
5479 | } | |
5480 | ||
518cd623 PZ |
5481 | /* |
5482 | * Keep a special pointer to the highest sched_domain that has | |
5483 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5484 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5485 | * | |
5486 | * Also keep a unique ID per domain (we use the first cpu number in | |
5487 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5488 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5489 | */ |
5490 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5491 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5492 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5493 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5494 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5495 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5496 | |
5497 | static void update_top_cache_domain(int cpu) | |
5498 | { | |
5499 | struct sched_domain *sd; | |
5d4cf996 | 5500 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5501 | int id = cpu; |
7d9ffa89 | 5502 | int size = 1; |
518cd623 PZ |
5503 | |
5504 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5505 | if (sd) { |
518cd623 | 5506 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5507 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5508 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5509 | } |
5d4cf996 | 5510 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5511 | |
5512 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5513 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5514 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5515 | |
5516 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5517 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5518 | |
5519 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5520 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5521 | } |
5522 | ||
1da177e4 | 5523 | /* |
0eab9146 | 5524 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5525 | * hold the hotplug lock. |
5526 | */ | |
0eab9146 IM |
5527 | static void |
5528 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5529 | { |
70b97a7f | 5530 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5531 | struct sched_domain *tmp; |
5532 | ||
5533 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5534 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5535 | struct sched_domain *parent = tmp->parent; |
5536 | if (!parent) | |
5537 | break; | |
f29c9b1c | 5538 | |
1a848870 | 5539 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5540 | tmp->parent = parent->parent; |
1a848870 SS |
5541 | if (parent->parent) |
5542 | parent->parent->child = tmp; | |
10866e62 PZ |
5543 | /* |
5544 | * Transfer SD_PREFER_SIBLING down in case of a | |
5545 | * degenerate parent; the spans match for this | |
5546 | * so the property transfers. | |
5547 | */ | |
5548 | if (parent->flags & SD_PREFER_SIBLING) | |
5549 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5550 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5551 | } else |
5552 | tmp = tmp->parent; | |
245af2c7 SS |
5553 | } |
5554 | ||
1a848870 | 5555 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5556 | tmp = sd; |
245af2c7 | 5557 | sd = sd->parent; |
dce840a0 | 5558 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5559 | if (sd) |
5560 | sd->child = NULL; | |
5561 | } | |
1da177e4 | 5562 | |
4cb98839 | 5563 | sched_domain_debug(sd, cpu); |
1da177e4 | 5564 | |
57d885fe | 5565 | rq_attach_root(rq, rd); |
dce840a0 | 5566 | tmp = rq->sd; |
674311d5 | 5567 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5568 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5569 | |
5570 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5571 | } |
5572 | ||
5573 | /* cpus with isolated domains */ | |
dcc30a35 | 5574 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5575 | |
5576 | /* Setup the mask of cpus configured for isolated domains */ | |
5577 | static int __init isolated_cpu_setup(char *str) | |
5578 | { | |
bdddd296 | 5579 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5580 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5581 | return 1; |
5582 | } | |
5583 | ||
8927f494 | 5584 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5585 | |
d3081f52 PZ |
5586 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5587 | { | |
5588 | return cpumask_of_node(cpu_to_node(cpu)); | |
5589 | } | |
5590 | ||
dce840a0 PZ |
5591 | struct sd_data { |
5592 | struct sched_domain **__percpu sd; | |
5593 | struct sched_group **__percpu sg; | |
9c3f75cb | 5594 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5595 | }; |
5596 | ||
49a02c51 | 5597 | struct s_data { |
21d42ccf | 5598 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5599 | struct root_domain *rd; |
5600 | }; | |
5601 | ||
2109b99e | 5602 | enum s_alloc { |
2109b99e | 5603 | sa_rootdomain, |
21d42ccf | 5604 | sa_sd, |
dce840a0 | 5605 | sa_sd_storage, |
2109b99e AH |
5606 | sa_none, |
5607 | }; | |
5608 | ||
54ab4ff4 PZ |
5609 | struct sched_domain_topology_level; |
5610 | ||
5611 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5612 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5613 | ||
e3589f6c PZ |
5614 | #define SDTL_OVERLAP 0x01 |
5615 | ||
eb7a74e6 | 5616 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5617 | sched_domain_init_f init; |
5618 | sched_domain_mask_f mask; | |
e3589f6c | 5619 | int flags; |
cb83b629 | 5620 | int numa_level; |
54ab4ff4 | 5621 | struct sd_data data; |
eb7a74e6 PZ |
5622 | }; |
5623 | ||
c1174876 PZ |
5624 | /* |
5625 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5626 | * domain traversal. | |
5627 | * | |
5628 | * Asymmetric node setups can result in situations where the domain tree is of | |
5629 | * unequal depth, make sure to skip domains that already cover the entire | |
5630 | * range. | |
5631 | * | |
5632 | * In that case build_sched_domains() will have terminated the iteration early | |
5633 | * and our sibling sd spans will be empty. Domains should always include the | |
5634 | * cpu they're built on, so check that. | |
5635 | * | |
5636 | */ | |
5637 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5638 | { | |
5639 | const struct cpumask *span = sched_domain_span(sd); | |
5640 | struct sd_data *sdd = sd->private; | |
5641 | struct sched_domain *sibling; | |
5642 | int i; | |
5643 | ||
5644 | for_each_cpu(i, span) { | |
5645 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5646 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5647 | continue; | |
5648 | ||
5649 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5650 | } | |
5651 | } | |
5652 | ||
5653 | /* | |
5654 | * Return the canonical balance cpu for this group, this is the first cpu | |
5655 | * of this group that's also in the iteration mask. | |
5656 | */ | |
5657 | int group_balance_cpu(struct sched_group *sg) | |
5658 | { | |
5659 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5660 | } | |
5661 | ||
e3589f6c PZ |
5662 | static int |
5663 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5664 | { | |
5665 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5666 | const struct cpumask *span = sched_domain_span(sd); | |
5667 | struct cpumask *covered = sched_domains_tmpmask; | |
5668 | struct sd_data *sdd = sd->private; | |
5669 | struct sched_domain *child; | |
5670 | int i; | |
5671 | ||
5672 | cpumask_clear(covered); | |
5673 | ||
5674 | for_each_cpu(i, span) { | |
5675 | struct cpumask *sg_span; | |
5676 | ||
5677 | if (cpumask_test_cpu(i, covered)) | |
5678 | continue; | |
5679 | ||
c1174876 PZ |
5680 | child = *per_cpu_ptr(sdd->sd, i); |
5681 | ||
5682 | /* See the comment near build_group_mask(). */ | |
5683 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5684 | continue; | |
5685 | ||
e3589f6c | 5686 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5687 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5688 | |
5689 | if (!sg) | |
5690 | goto fail; | |
5691 | ||
5692 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5693 | if (child->child) { |
5694 | child = child->child; | |
5695 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5696 | } else | |
5697 | cpumask_set_cpu(i, sg_span); | |
5698 | ||
5699 | cpumask_or(covered, covered, sg_span); | |
5700 | ||
74a5ce20 | 5701 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5702 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5703 | build_group_mask(sd, sg); | |
5704 | ||
c3decf0d PZ |
5705 | /* |
5706 | * Initialize sgp->power such that even if we mess up the | |
5707 | * domains and no possible iteration will get us here, we won't | |
5708 | * die on a /0 trap. | |
5709 | */ | |
5710 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
8e8339a3 | 5711 | sg->sgp->power_orig = sg->sgp->power; |
e3589f6c | 5712 | |
c1174876 PZ |
5713 | /* |
5714 | * Make sure the first group of this domain contains the | |
5715 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5716 | * breaks. See update_sg_lb_stats(). | |
5717 | */ | |
74a5ce20 | 5718 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5719 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5720 | groups = sg; |
5721 | ||
5722 | if (!first) | |
5723 | first = sg; | |
5724 | if (last) | |
5725 | last->next = sg; | |
5726 | last = sg; | |
5727 | last->next = first; | |
5728 | } | |
5729 | sd->groups = groups; | |
5730 | ||
5731 | return 0; | |
5732 | ||
5733 | fail: | |
5734 | free_sched_groups(first, 0); | |
5735 | ||
5736 | return -ENOMEM; | |
5737 | } | |
5738 | ||
dce840a0 | 5739 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5740 | { |
dce840a0 PZ |
5741 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5742 | struct sched_domain *child = sd->child; | |
1da177e4 | 5743 | |
dce840a0 PZ |
5744 | if (child) |
5745 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5746 | |
9c3f75cb | 5747 | if (sg) { |
dce840a0 | 5748 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5749 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5750 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5751 | } |
dce840a0 PZ |
5752 | |
5753 | return cpu; | |
1e9f28fa | 5754 | } |
1e9f28fa | 5755 | |
01a08546 | 5756 | /* |
dce840a0 PZ |
5757 | * build_sched_groups will build a circular linked list of the groups |
5758 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5759 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5760 | * |
5761 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5762 | */ |
e3589f6c PZ |
5763 | static int |
5764 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5765 | { |
dce840a0 PZ |
5766 | struct sched_group *first = NULL, *last = NULL; |
5767 | struct sd_data *sdd = sd->private; | |
5768 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5769 | struct cpumask *covered; |
dce840a0 | 5770 | int i; |
9c1cfda2 | 5771 | |
e3589f6c PZ |
5772 | get_group(cpu, sdd, &sd->groups); |
5773 | atomic_inc(&sd->groups->ref); | |
5774 | ||
0936629f | 5775 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5776 | return 0; |
5777 | ||
f96225fd PZ |
5778 | lockdep_assert_held(&sched_domains_mutex); |
5779 | covered = sched_domains_tmpmask; | |
5780 | ||
dce840a0 | 5781 | cpumask_clear(covered); |
6711cab4 | 5782 | |
dce840a0 PZ |
5783 | for_each_cpu(i, span) { |
5784 | struct sched_group *sg; | |
cd08e923 | 5785 | int group, j; |
6711cab4 | 5786 | |
dce840a0 PZ |
5787 | if (cpumask_test_cpu(i, covered)) |
5788 | continue; | |
6711cab4 | 5789 | |
cd08e923 | 5790 | group = get_group(i, sdd, &sg); |
dce840a0 | 5791 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5792 | sg->sgp->power = 0; |
c1174876 | 5793 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5794 | |
dce840a0 PZ |
5795 | for_each_cpu(j, span) { |
5796 | if (get_group(j, sdd, NULL) != group) | |
5797 | continue; | |
0601a88d | 5798 | |
dce840a0 PZ |
5799 | cpumask_set_cpu(j, covered); |
5800 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5801 | } | |
0601a88d | 5802 | |
dce840a0 PZ |
5803 | if (!first) |
5804 | first = sg; | |
5805 | if (last) | |
5806 | last->next = sg; | |
5807 | last = sg; | |
5808 | } | |
5809 | last->next = first; | |
e3589f6c PZ |
5810 | |
5811 | return 0; | |
0601a88d | 5812 | } |
51888ca2 | 5813 | |
89c4710e SS |
5814 | /* |
5815 | * Initialize sched groups cpu_power. | |
5816 | * | |
5817 | * cpu_power indicates the capacity of sched group, which is used while | |
5818 | * distributing the load between different sched groups in a sched domain. | |
5819 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5820 | * there are asymmetries in the topology. If there are asymmetries, group | |
5821 | * having more cpu_power will pickup more load compared to the group having | |
5822 | * less cpu_power. | |
89c4710e SS |
5823 | */ |
5824 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5825 | { | |
e3589f6c | 5826 | struct sched_group *sg = sd->groups; |
89c4710e | 5827 | |
94c95ba6 | 5828 | WARN_ON(!sg); |
e3589f6c PZ |
5829 | |
5830 | do { | |
5831 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5832 | sg = sg->next; | |
5833 | } while (sg != sd->groups); | |
89c4710e | 5834 | |
c1174876 | 5835 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5836 | return; |
aae6d3dd | 5837 | |
d274cb30 | 5838 | update_group_power(sd, cpu); |
69e1e811 | 5839 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5840 | } |
5841 | ||
029632fb PZ |
5842 | int __weak arch_sd_sibling_asym_packing(void) |
5843 | { | |
5844 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
5845 | } |
5846 | ||
7c16ec58 MT |
5847 | /* |
5848 | * Initializers for schedule domains | |
5849 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5850 | */ | |
5851 | ||
a5d8c348 IM |
5852 | #ifdef CONFIG_SCHED_DEBUG |
5853 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
5854 | #else | |
5855 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
5856 | #endif | |
5857 | ||
54ab4ff4 PZ |
5858 | #define SD_INIT_FUNC(type) \ |
5859 | static noinline struct sched_domain * \ | |
5860 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
5861 | { \ | |
5862 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
5863 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
5864 | SD_INIT_NAME(sd, type); \ |
5865 | sd->private = &tl->data; \ | |
5866 | return sd; \ | |
7c16ec58 MT |
5867 | } |
5868 | ||
5869 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
5870 | #ifdef CONFIG_SCHED_SMT |
5871 | SD_INIT_FUNC(SIBLING) | |
5872 | #endif | |
5873 | #ifdef CONFIG_SCHED_MC | |
5874 | SD_INIT_FUNC(MC) | |
5875 | #endif | |
01a08546 HC |
5876 | #ifdef CONFIG_SCHED_BOOK |
5877 | SD_INIT_FUNC(BOOK) | |
5878 | #endif | |
7c16ec58 | 5879 | |
1d3504fc | 5880 | static int default_relax_domain_level = -1; |
60495e77 | 5881 | int sched_domain_level_max; |
1d3504fc HS |
5882 | |
5883 | static int __init setup_relax_domain_level(char *str) | |
5884 | { | |
a841f8ce DS |
5885 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5886 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5887 | |
1d3504fc HS |
5888 | return 1; |
5889 | } | |
5890 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5891 | ||
5892 | static void set_domain_attribute(struct sched_domain *sd, | |
5893 | struct sched_domain_attr *attr) | |
5894 | { | |
5895 | int request; | |
5896 | ||
5897 | if (!attr || attr->relax_domain_level < 0) { | |
5898 | if (default_relax_domain_level < 0) | |
5899 | return; | |
5900 | else | |
5901 | request = default_relax_domain_level; | |
5902 | } else | |
5903 | request = attr->relax_domain_level; | |
5904 | if (request < sd->level) { | |
5905 | /* turn off idle balance on this domain */ | |
c88d5910 | 5906 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5907 | } else { |
5908 | /* turn on idle balance on this domain */ | |
c88d5910 | 5909 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5910 | } |
5911 | } | |
5912 | ||
54ab4ff4 PZ |
5913 | static void __sdt_free(const struct cpumask *cpu_map); |
5914 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5915 | ||
2109b99e AH |
5916 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5917 | const struct cpumask *cpu_map) | |
5918 | { | |
5919 | switch (what) { | |
2109b99e | 5920 | case sa_rootdomain: |
822ff793 PZ |
5921 | if (!atomic_read(&d->rd->refcount)) |
5922 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5923 | case sa_sd: |
5924 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5925 | case sa_sd_storage: |
54ab4ff4 | 5926 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5927 | case sa_none: |
5928 | break; | |
5929 | } | |
5930 | } | |
3404c8d9 | 5931 | |
2109b99e AH |
5932 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5933 | const struct cpumask *cpu_map) | |
5934 | { | |
dce840a0 PZ |
5935 | memset(d, 0, sizeof(*d)); |
5936 | ||
54ab4ff4 PZ |
5937 | if (__sdt_alloc(cpu_map)) |
5938 | return sa_sd_storage; | |
dce840a0 PZ |
5939 | d->sd = alloc_percpu(struct sched_domain *); |
5940 | if (!d->sd) | |
5941 | return sa_sd_storage; | |
2109b99e | 5942 | d->rd = alloc_rootdomain(); |
dce840a0 | 5943 | if (!d->rd) |
21d42ccf | 5944 | return sa_sd; |
2109b99e AH |
5945 | return sa_rootdomain; |
5946 | } | |
57d885fe | 5947 | |
dce840a0 PZ |
5948 | /* |
5949 | * NULL the sd_data elements we've used to build the sched_domain and | |
5950 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5951 | * will not free the data we're using. | |
5952 | */ | |
5953 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5954 | { | |
5955 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5956 | |
5957 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5958 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5959 | ||
e3589f6c | 5960 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5961 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5962 | |
5963 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5964 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5965 | } |
5966 | ||
2c402dc3 PZ |
5967 | #ifdef CONFIG_SCHED_SMT |
5968 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 5969 | { |
2c402dc3 | 5970 | return topology_thread_cpumask(cpu); |
3bd65a80 | 5971 | } |
2c402dc3 | 5972 | #endif |
7f4588f3 | 5973 | |
d069b916 PZ |
5974 | /* |
5975 | * Topology list, bottom-up. | |
5976 | */ | |
2c402dc3 | 5977 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
5978 | #ifdef CONFIG_SCHED_SMT |
5979 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 5980 | #endif |
1e9f28fa | 5981 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 5982 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 5983 | #endif |
d069b916 PZ |
5984 | #ifdef CONFIG_SCHED_BOOK |
5985 | { sd_init_BOOK, cpu_book_mask, }, | |
5986 | #endif | |
5987 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
5988 | { NULL, }, |
5989 | }; | |
5990 | ||
5991 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
5992 | ||
27723a68 VK |
5993 | #define for_each_sd_topology(tl) \ |
5994 | for (tl = sched_domain_topology; tl->init; tl++) | |
5995 | ||
cb83b629 PZ |
5996 | #ifdef CONFIG_NUMA |
5997 | ||
5998 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
5999 | static int *sched_domains_numa_distance; |
6000 | static struct cpumask ***sched_domains_numa_masks; | |
6001 | static int sched_domains_curr_level; | |
6002 | ||
cb83b629 PZ |
6003 | static inline int sd_local_flags(int level) |
6004 | { | |
10717dcd | 6005 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
6006 | return 0; |
6007 | ||
6008 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6009 | } | |
6010 | ||
6011 | static struct sched_domain * | |
6012 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6013 | { | |
6014 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6015 | int level = tl->numa_level; | |
6016 | int sd_weight = cpumask_weight( | |
6017 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6018 | ||
6019 | *sd = (struct sched_domain){ | |
6020 | .min_interval = sd_weight, | |
6021 | .max_interval = 2*sd_weight, | |
6022 | .busy_factor = 32, | |
870a0bb5 | 6023 | .imbalance_pct = 125, |
cb83b629 PZ |
6024 | .cache_nice_tries = 2, |
6025 | .busy_idx = 3, | |
6026 | .idle_idx = 2, | |
6027 | .newidle_idx = 0, | |
6028 | .wake_idx = 0, | |
6029 | .forkexec_idx = 0, | |
6030 | ||
6031 | .flags = 1*SD_LOAD_BALANCE | |
6032 | | 1*SD_BALANCE_NEWIDLE | |
6033 | | 0*SD_BALANCE_EXEC | |
6034 | | 0*SD_BALANCE_FORK | |
6035 | | 0*SD_BALANCE_WAKE | |
6036 | | 0*SD_WAKE_AFFINE | |
cb83b629 | 6037 | | 0*SD_SHARE_CPUPOWER |
cb83b629 PZ |
6038 | | 0*SD_SHARE_PKG_RESOURCES |
6039 | | 1*SD_SERIALIZE | |
6040 | | 0*SD_PREFER_SIBLING | |
3a7053b3 | 6041 | | 1*SD_NUMA |
cb83b629 PZ |
6042 | | sd_local_flags(level) |
6043 | , | |
6044 | .last_balance = jiffies, | |
6045 | .balance_interval = sd_weight, | |
6046 | }; | |
6047 | SD_INIT_NAME(sd, NUMA); | |
6048 | sd->private = &tl->data; | |
6049 | ||
6050 | /* | |
6051 | * Ugly hack to pass state to sd_numa_mask()... | |
6052 | */ | |
6053 | sched_domains_curr_level = tl->numa_level; | |
6054 | ||
6055 | return sd; | |
6056 | } | |
6057 | ||
6058 | static const struct cpumask *sd_numa_mask(int cpu) | |
6059 | { | |
6060 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6061 | } | |
6062 | ||
d039ac60 PZ |
6063 | static void sched_numa_warn(const char *str) |
6064 | { | |
6065 | static int done = false; | |
6066 | int i,j; | |
6067 | ||
6068 | if (done) | |
6069 | return; | |
6070 | ||
6071 | done = true; | |
6072 | ||
6073 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6074 | ||
6075 | for (i = 0; i < nr_node_ids; i++) { | |
6076 | printk(KERN_WARNING " "); | |
6077 | for (j = 0; j < nr_node_ids; j++) | |
6078 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6079 | printk(KERN_CONT "\n"); | |
6080 | } | |
6081 | printk(KERN_WARNING "\n"); | |
6082 | } | |
6083 | ||
6084 | static bool find_numa_distance(int distance) | |
6085 | { | |
6086 | int i; | |
6087 | ||
6088 | if (distance == node_distance(0, 0)) | |
6089 | return true; | |
6090 | ||
6091 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6092 | if (sched_domains_numa_distance[i] == distance) | |
6093 | return true; | |
6094 | } | |
6095 | ||
6096 | return false; | |
6097 | } | |
6098 | ||
cb83b629 PZ |
6099 | static void sched_init_numa(void) |
6100 | { | |
6101 | int next_distance, curr_distance = node_distance(0, 0); | |
6102 | struct sched_domain_topology_level *tl; | |
6103 | int level = 0; | |
6104 | int i, j, k; | |
6105 | ||
cb83b629 PZ |
6106 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6107 | if (!sched_domains_numa_distance) | |
6108 | return; | |
6109 | ||
6110 | /* | |
6111 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6112 | * unique distances in the node_distance() table. | |
6113 | * | |
6114 | * Assumes node_distance(0,j) includes all distances in | |
6115 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6116 | */ |
6117 | next_distance = curr_distance; | |
6118 | for (i = 0; i < nr_node_ids; i++) { | |
6119 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6120 | for (k = 0; k < nr_node_ids; k++) { |
6121 | int distance = node_distance(i, k); | |
6122 | ||
6123 | if (distance > curr_distance && | |
6124 | (distance < next_distance || | |
6125 | next_distance == curr_distance)) | |
6126 | next_distance = distance; | |
6127 | ||
6128 | /* | |
6129 | * While not a strong assumption it would be nice to know | |
6130 | * about cases where if node A is connected to B, B is not | |
6131 | * equally connected to A. | |
6132 | */ | |
6133 | if (sched_debug() && node_distance(k, i) != distance) | |
6134 | sched_numa_warn("Node-distance not symmetric"); | |
6135 | ||
6136 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6137 | sched_numa_warn("Node-0 not representative"); | |
6138 | } | |
6139 | if (next_distance != curr_distance) { | |
6140 | sched_domains_numa_distance[level++] = next_distance; | |
6141 | sched_domains_numa_levels = level; | |
6142 | curr_distance = next_distance; | |
6143 | } else break; | |
cb83b629 | 6144 | } |
d039ac60 PZ |
6145 | |
6146 | /* | |
6147 | * In case of sched_debug() we verify the above assumption. | |
6148 | */ | |
6149 | if (!sched_debug()) | |
6150 | break; | |
cb83b629 PZ |
6151 | } |
6152 | /* | |
6153 | * 'level' contains the number of unique distances, excluding the | |
6154 | * identity distance node_distance(i,i). | |
6155 | * | |
28b4a521 | 6156 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6157 | * numbers. |
6158 | */ | |
6159 | ||
5f7865f3 TC |
6160 | /* |
6161 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6162 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6163 | * the array will contain less then 'level' members. This could be | |
6164 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6165 | * in other functions. | |
6166 | * | |
6167 | * We reset it to 'level' at the end of this function. | |
6168 | */ | |
6169 | sched_domains_numa_levels = 0; | |
6170 | ||
cb83b629 PZ |
6171 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6172 | if (!sched_domains_numa_masks) | |
6173 | return; | |
6174 | ||
6175 | /* | |
6176 | * Now for each level, construct a mask per node which contains all | |
6177 | * cpus of nodes that are that many hops away from us. | |
6178 | */ | |
6179 | for (i = 0; i < level; i++) { | |
6180 | sched_domains_numa_masks[i] = | |
6181 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6182 | if (!sched_domains_numa_masks[i]) | |
6183 | return; | |
6184 | ||
6185 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6186 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6187 | if (!mask) |
6188 | return; | |
6189 | ||
6190 | sched_domains_numa_masks[i][j] = mask; | |
6191 | ||
6192 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6193 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6194 | continue; |
6195 | ||
6196 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6197 | } | |
6198 | } | |
6199 | } | |
6200 | ||
6201 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6202 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6203 | if (!tl) | |
6204 | return; | |
6205 | ||
6206 | /* | |
6207 | * Copy the default topology bits.. | |
6208 | */ | |
6209 | for (i = 0; default_topology[i].init; i++) | |
6210 | tl[i] = default_topology[i]; | |
6211 | ||
6212 | /* | |
6213 | * .. and append 'j' levels of NUMA goodness. | |
6214 | */ | |
6215 | for (j = 0; j < level; i++, j++) { | |
6216 | tl[i] = (struct sched_domain_topology_level){ | |
6217 | .init = sd_numa_init, | |
6218 | .mask = sd_numa_mask, | |
6219 | .flags = SDTL_OVERLAP, | |
6220 | .numa_level = j, | |
6221 | }; | |
6222 | } | |
6223 | ||
6224 | sched_domain_topology = tl; | |
5f7865f3 TC |
6225 | |
6226 | sched_domains_numa_levels = level; | |
cb83b629 | 6227 | } |
301a5cba TC |
6228 | |
6229 | static void sched_domains_numa_masks_set(int cpu) | |
6230 | { | |
6231 | int i, j; | |
6232 | int node = cpu_to_node(cpu); | |
6233 | ||
6234 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6235 | for (j = 0; j < nr_node_ids; j++) { | |
6236 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6237 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6238 | } | |
6239 | } | |
6240 | } | |
6241 | ||
6242 | static void sched_domains_numa_masks_clear(int cpu) | |
6243 | { | |
6244 | int i, j; | |
6245 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6246 | for (j = 0; j < nr_node_ids; j++) | |
6247 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6248 | } | |
6249 | } | |
6250 | ||
6251 | /* | |
6252 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6253 | * are onlined. | |
6254 | */ | |
6255 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6256 | unsigned long action, | |
6257 | void *hcpu) | |
6258 | { | |
6259 | int cpu = (long)hcpu; | |
6260 | ||
6261 | switch (action & ~CPU_TASKS_FROZEN) { | |
6262 | case CPU_ONLINE: | |
6263 | sched_domains_numa_masks_set(cpu); | |
6264 | break; | |
6265 | ||
6266 | case CPU_DEAD: | |
6267 | sched_domains_numa_masks_clear(cpu); | |
6268 | break; | |
6269 | ||
6270 | default: | |
6271 | return NOTIFY_DONE; | |
6272 | } | |
6273 | ||
6274 | return NOTIFY_OK; | |
cb83b629 PZ |
6275 | } |
6276 | #else | |
6277 | static inline void sched_init_numa(void) | |
6278 | { | |
6279 | } | |
301a5cba TC |
6280 | |
6281 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6282 | unsigned long action, | |
6283 | void *hcpu) | |
6284 | { | |
6285 | return 0; | |
6286 | } | |
cb83b629 PZ |
6287 | #endif /* CONFIG_NUMA */ |
6288 | ||
54ab4ff4 PZ |
6289 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6290 | { | |
6291 | struct sched_domain_topology_level *tl; | |
6292 | int j; | |
6293 | ||
27723a68 | 6294 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6295 | struct sd_data *sdd = &tl->data; |
6296 | ||
6297 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6298 | if (!sdd->sd) | |
6299 | return -ENOMEM; | |
6300 | ||
6301 | sdd->sg = alloc_percpu(struct sched_group *); | |
6302 | if (!sdd->sg) | |
6303 | return -ENOMEM; | |
6304 | ||
9c3f75cb PZ |
6305 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6306 | if (!sdd->sgp) | |
6307 | return -ENOMEM; | |
6308 | ||
54ab4ff4 PZ |
6309 | for_each_cpu(j, cpu_map) { |
6310 | struct sched_domain *sd; | |
6311 | struct sched_group *sg; | |
9c3f75cb | 6312 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6313 | |
6314 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6315 | GFP_KERNEL, cpu_to_node(j)); | |
6316 | if (!sd) | |
6317 | return -ENOMEM; | |
6318 | ||
6319 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6320 | ||
6321 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6322 | GFP_KERNEL, cpu_to_node(j)); | |
6323 | if (!sg) | |
6324 | return -ENOMEM; | |
6325 | ||
30b4e9eb IM |
6326 | sg->next = sg; |
6327 | ||
54ab4ff4 | 6328 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6329 | |
c1174876 | 6330 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6331 | GFP_KERNEL, cpu_to_node(j)); |
6332 | if (!sgp) | |
6333 | return -ENOMEM; | |
6334 | ||
6335 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6336 | } |
6337 | } | |
6338 | ||
6339 | return 0; | |
6340 | } | |
6341 | ||
6342 | static void __sdt_free(const struct cpumask *cpu_map) | |
6343 | { | |
6344 | struct sched_domain_topology_level *tl; | |
6345 | int j; | |
6346 | ||
27723a68 | 6347 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6348 | struct sd_data *sdd = &tl->data; |
6349 | ||
6350 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6351 | struct sched_domain *sd; |
6352 | ||
6353 | if (sdd->sd) { | |
6354 | sd = *per_cpu_ptr(sdd->sd, j); | |
6355 | if (sd && (sd->flags & SD_OVERLAP)) | |
6356 | free_sched_groups(sd->groups, 0); | |
6357 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6358 | } | |
6359 | ||
6360 | if (sdd->sg) | |
6361 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6362 | if (sdd->sgp) | |
6363 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6364 | } |
6365 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6366 | sdd->sd = NULL; |
54ab4ff4 | 6367 | free_percpu(sdd->sg); |
fb2cf2c6 | 6368 | sdd->sg = NULL; |
9c3f75cb | 6369 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6370 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6371 | } |
6372 | } | |
6373 | ||
2c402dc3 | 6374 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6375 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6376 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6377 | { |
54ab4ff4 | 6378 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6379 | if (!sd) |
d069b916 | 6380 | return child; |
2c402dc3 | 6381 | |
2c402dc3 | 6382 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6383 | if (child) { |
6384 | sd->level = child->level + 1; | |
6385 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6386 | child->parent = sd; |
c75e0128 | 6387 | sd->child = child; |
60495e77 | 6388 | } |
a841f8ce | 6389 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6390 | |
6391 | return sd; | |
6392 | } | |
6393 | ||
2109b99e AH |
6394 | /* |
6395 | * Build sched domains for a given set of cpus and attach the sched domains | |
6396 | * to the individual cpus | |
6397 | */ | |
dce840a0 PZ |
6398 | static int build_sched_domains(const struct cpumask *cpu_map, |
6399 | struct sched_domain_attr *attr) | |
2109b99e | 6400 | { |
1c632169 | 6401 | enum s_alloc alloc_state; |
dce840a0 | 6402 | struct sched_domain *sd; |
2109b99e | 6403 | struct s_data d; |
822ff793 | 6404 | int i, ret = -ENOMEM; |
9c1cfda2 | 6405 | |
2109b99e AH |
6406 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6407 | if (alloc_state != sa_rootdomain) | |
6408 | goto error; | |
9c1cfda2 | 6409 | |
dce840a0 | 6410 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6411 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6412 | struct sched_domain_topology_level *tl; |
6413 | ||
3bd65a80 | 6414 | sd = NULL; |
27723a68 | 6415 | for_each_sd_topology(tl) { |
4a850cbe | 6416 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6417 | if (tl == sched_domain_topology) |
6418 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6419 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6420 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6421 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6422 | break; | |
e3589f6c | 6423 | } |
dce840a0 PZ |
6424 | } |
6425 | ||
6426 | /* Build the groups for the domains */ | |
6427 | for_each_cpu(i, cpu_map) { | |
6428 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6429 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6430 | if (sd->flags & SD_OVERLAP) { |
6431 | if (build_overlap_sched_groups(sd, i)) | |
6432 | goto error; | |
6433 | } else { | |
6434 | if (build_sched_groups(sd, i)) | |
6435 | goto error; | |
6436 | } | |
1cf51902 | 6437 | } |
a06dadbe | 6438 | } |
9c1cfda2 | 6439 | |
1da177e4 | 6440 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6441 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6442 | if (!cpumask_test_cpu(i, cpu_map)) | |
6443 | continue; | |
9c1cfda2 | 6444 | |
dce840a0 PZ |
6445 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6446 | claim_allocations(i, sd); | |
cd4ea6ae | 6447 | init_sched_groups_power(i, sd); |
dce840a0 | 6448 | } |
f712c0c7 | 6449 | } |
9c1cfda2 | 6450 | |
1da177e4 | 6451 | /* Attach the domains */ |
dce840a0 | 6452 | rcu_read_lock(); |
abcd083a | 6453 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6454 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6455 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6456 | } |
dce840a0 | 6457 | rcu_read_unlock(); |
51888ca2 | 6458 | |
822ff793 | 6459 | ret = 0; |
51888ca2 | 6460 | error: |
2109b99e | 6461 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6462 | return ret; |
1da177e4 | 6463 | } |
029190c5 | 6464 | |
acc3f5d7 | 6465 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6466 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6467 | static struct sched_domain_attr *dattr_cur; |
6468 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6469 | |
6470 | /* | |
6471 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6472 | * cpumask) fails, then fallback to a single sched domain, |
6473 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6474 | */ |
4212823f | 6475 | static cpumask_var_t fallback_doms; |
029190c5 | 6476 | |
ee79d1bd HC |
6477 | /* |
6478 | * arch_update_cpu_topology lets virtualized architectures update the | |
6479 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6480 | * or 0 if it stayed the same. | |
6481 | */ | |
6482 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6483 | { |
ee79d1bd | 6484 | return 0; |
22e52b07 HC |
6485 | } |
6486 | ||
acc3f5d7 RR |
6487 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6488 | { | |
6489 | int i; | |
6490 | cpumask_var_t *doms; | |
6491 | ||
6492 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6493 | if (!doms) | |
6494 | return NULL; | |
6495 | for (i = 0; i < ndoms; i++) { | |
6496 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6497 | free_sched_domains(doms, i); | |
6498 | return NULL; | |
6499 | } | |
6500 | } | |
6501 | return doms; | |
6502 | } | |
6503 | ||
6504 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6505 | { | |
6506 | unsigned int i; | |
6507 | for (i = 0; i < ndoms; i++) | |
6508 | free_cpumask_var(doms[i]); | |
6509 | kfree(doms); | |
6510 | } | |
6511 | ||
1a20ff27 | 6512 | /* |
41a2d6cf | 6513 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6514 | * For now this just excludes isolated cpus, but could be used to |
6515 | * exclude other special cases in the future. | |
1a20ff27 | 6516 | */ |
c4a8849a | 6517 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6518 | { |
7378547f MM |
6519 | int err; |
6520 | ||
22e52b07 | 6521 | arch_update_cpu_topology(); |
029190c5 | 6522 | ndoms_cur = 1; |
acc3f5d7 | 6523 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6524 | if (!doms_cur) |
acc3f5d7 RR |
6525 | doms_cur = &fallback_doms; |
6526 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6527 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6528 | register_sched_domain_sysctl(); |
7378547f MM |
6529 | |
6530 | return err; | |
1a20ff27 DG |
6531 | } |
6532 | ||
1a20ff27 DG |
6533 | /* |
6534 | * Detach sched domains from a group of cpus specified in cpu_map | |
6535 | * These cpus will now be attached to the NULL domain | |
6536 | */ | |
96f874e2 | 6537 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6538 | { |
6539 | int i; | |
6540 | ||
dce840a0 | 6541 | rcu_read_lock(); |
abcd083a | 6542 | for_each_cpu(i, cpu_map) |
57d885fe | 6543 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6544 | rcu_read_unlock(); |
1a20ff27 DG |
6545 | } |
6546 | ||
1d3504fc HS |
6547 | /* handle null as "default" */ |
6548 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6549 | struct sched_domain_attr *new, int idx_new) | |
6550 | { | |
6551 | struct sched_domain_attr tmp; | |
6552 | ||
6553 | /* fast path */ | |
6554 | if (!new && !cur) | |
6555 | return 1; | |
6556 | ||
6557 | tmp = SD_ATTR_INIT; | |
6558 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6559 | new ? (new + idx_new) : &tmp, | |
6560 | sizeof(struct sched_domain_attr)); | |
6561 | } | |
6562 | ||
029190c5 PJ |
6563 | /* |
6564 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6565 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6566 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6567 | * It destroys each deleted domain and builds each new domain. | |
6568 | * | |
acc3f5d7 | 6569 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6570 | * The masks don't intersect (don't overlap.) We should setup one |
6571 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6572 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6573 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6574 | * it as it is. | |
6575 | * | |
acc3f5d7 RR |
6576 | * The passed in 'doms_new' should be allocated using |
6577 | * alloc_sched_domains. This routine takes ownership of it and will | |
6578 | * free_sched_domains it when done with it. If the caller failed the | |
6579 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6580 | * and partition_sched_domains() will fallback to the single partition | |
6581 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6582 | * |
96f874e2 | 6583 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6584 | * ndoms_new == 0 is a special case for destroying existing domains, |
6585 | * and it will not create the default domain. | |
dfb512ec | 6586 | * |
029190c5 PJ |
6587 | * Call with hotplug lock held |
6588 | */ | |
acc3f5d7 | 6589 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6590 | struct sched_domain_attr *dattr_new) |
029190c5 | 6591 | { |
dfb512ec | 6592 | int i, j, n; |
d65bd5ec | 6593 | int new_topology; |
029190c5 | 6594 | |
712555ee | 6595 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6596 | |
7378547f MM |
6597 | /* always unregister in case we don't destroy any domains */ |
6598 | unregister_sched_domain_sysctl(); | |
6599 | ||
d65bd5ec HC |
6600 | /* Let architecture update cpu core mappings. */ |
6601 | new_topology = arch_update_cpu_topology(); | |
6602 | ||
dfb512ec | 6603 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6604 | |
6605 | /* Destroy deleted domains */ | |
6606 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6607 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6608 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6609 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6610 | goto match1; |
6611 | } | |
6612 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6613 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6614 | match1: |
6615 | ; | |
6616 | } | |
6617 | ||
c8d2d47a | 6618 | n = ndoms_cur; |
e761b772 | 6619 | if (doms_new == NULL) { |
c8d2d47a | 6620 | n = 0; |
acc3f5d7 | 6621 | doms_new = &fallback_doms; |
6ad4c188 | 6622 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6623 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6624 | } |
6625 | ||
029190c5 PJ |
6626 | /* Build new domains */ |
6627 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6628 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6629 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6630 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6631 | goto match2; |
6632 | } | |
6633 | /* no match - add a new doms_new */ | |
dce840a0 | 6634 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6635 | match2: |
6636 | ; | |
6637 | } | |
6638 | ||
6639 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6640 | if (doms_cur != &fallback_doms) |
6641 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6642 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6643 | doms_cur = doms_new; |
1d3504fc | 6644 | dattr_cur = dattr_new; |
029190c5 | 6645 | ndoms_cur = ndoms_new; |
7378547f MM |
6646 | |
6647 | register_sched_domain_sysctl(); | |
a1835615 | 6648 | |
712555ee | 6649 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6650 | } |
6651 | ||
d35be8ba SB |
6652 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6653 | ||
1da177e4 | 6654 | /* |
3a101d05 TH |
6655 | * Update cpusets according to cpu_active mask. If cpusets are |
6656 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6657 | * around partition_sched_domains(). | |
d35be8ba SB |
6658 | * |
6659 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6660 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6661 | */ |
0b2e918a TH |
6662 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6663 | void *hcpu) | |
e761b772 | 6664 | { |
d35be8ba SB |
6665 | switch (action) { |
6666 | case CPU_ONLINE_FROZEN: | |
6667 | case CPU_DOWN_FAILED_FROZEN: | |
6668 | ||
6669 | /* | |
6670 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6671 | * resume sequence. As long as this is not the last online | |
6672 | * operation in the resume sequence, just build a single sched | |
6673 | * domain, ignoring cpusets. | |
6674 | */ | |
6675 | num_cpus_frozen--; | |
6676 | if (likely(num_cpus_frozen)) { | |
6677 | partition_sched_domains(1, NULL, NULL); | |
6678 | break; | |
6679 | } | |
6680 | ||
6681 | /* | |
6682 | * This is the last CPU online operation. So fall through and | |
6683 | * restore the original sched domains by considering the | |
6684 | * cpuset configurations. | |
6685 | */ | |
6686 | ||
e761b772 | 6687 | case CPU_ONLINE: |
6ad4c188 | 6688 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6689 | cpuset_update_active_cpus(true); |
d35be8ba | 6690 | break; |
3a101d05 TH |
6691 | default: |
6692 | return NOTIFY_DONE; | |
6693 | } | |
d35be8ba | 6694 | return NOTIFY_OK; |
3a101d05 | 6695 | } |
e761b772 | 6696 | |
0b2e918a TH |
6697 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6698 | void *hcpu) | |
3a101d05 | 6699 | { |
d35be8ba | 6700 | switch (action) { |
3a101d05 | 6701 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6702 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6703 | break; |
6704 | case CPU_DOWN_PREPARE_FROZEN: | |
6705 | num_cpus_frozen++; | |
6706 | partition_sched_domains(1, NULL, NULL); | |
6707 | break; | |
e761b772 MK |
6708 | default: |
6709 | return NOTIFY_DONE; | |
6710 | } | |
d35be8ba | 6711 | return NOTIFY_OK; |
e761b772 | 6712 | } |
e761b772 | 6713 | |
1da177e4 LT |
6714 | void __init sched_init_smp(void) |
6715 | { | |
dcc30a35 RR |
6716 | cpumask_var_t non_isolated_cpus; |
6717 | ||
6718 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6719 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6720 | |
cb83b629 PZ |
6721 | sched_init_numa(); |
6722 | ||
6acce3ef PZ |
6723 | /* |
6724 | * There's no userspace yet to cause hotplug operations; hence all the | |
6725 | * cpu masks are stable and all blatant races in the below code cannot | |
6726 | * happen. | |
6727 | */ | |
712555ee | 6728 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6729 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6730 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6731 | if (cpumask_empty(non_isolated_cpus)) | |
6732 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6733 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6734 | |
301a5cba | 6735 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6736 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6737 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6738 | |
b328ca18 | 6739 | init_hrtick(); |
5c1e1767 NP |
6740 | |
6741 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6742 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6743 | BUG(); |
19978ca6 | 6744 | sched_init_granularity(); |
dcc30a35 | 6745 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6746 | |
0e3900e6 | 6747 | init_sched_rt_class(); |
1baca4ce | 6748 | init_sched_dl_class(); |
1da177e4 LT |
6749 | } |
6750 | #else | |
6751 | void __init sched_init_smp(void) | |
6752 | { | |
19978ca6 | 6753 | sched_init_granularity(); |
1da177e4 LT |
6754 | } |
6755 | #endif /* CONFIG_SMP */ | |
6756 | ||
cd1bb94b AB |
6757 | const_debug unsigned int sysctl_timer_migration = 1; |
6758 | ||
1da177e4 LT |
6759 | int in_sched_functions(unsigned long addr) |
6760 | { | |
1da177e4 LT |
6761 | return in_lock_functions(addr) || |
6762 | (addr >= (unsigned long)__sched_text_start | |
6763 | && addr < (unsigned long)__sched_text_end); | |
6764 | } | |
6765 | ||
029632fb | 6766 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6767 | /* |
6768 | * Default task group. | |
6769 | * Every task in system belongs to this group at bootup. | |
6770 | */ | |
029632fb | 6771 | struct task_group root_task_group; |
35cf4e50 | 6772 | LIST_HEAD(task_groups); |
052f1dc7 | 6773 | #endif |
6f505b16 | 6774 | |
e6252c3e | 6775 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6776 | |
1da177e4 LT |
6777 | void __init sched_init(void) |
6778 | { | |
dd41f596 | 6779 | int i, j; |
434d53b0 MT |
6780 | unsigned long alloc_size = 0, ptr; |
6781 | ||
6782 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6783 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6784 | #endif | |
6785 | #ifdef CONFIG_RT_GROUP_SCHED | |
6786 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6787 | #endif |
df7c8e84 | 6788 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6789 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6790 | #endif |
434d53b0 | 6791 | if (alloc_size) { |
36b7b6d4 | 6792 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6793 | |
6794 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6795 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6796 | ptr += nr_cpu_ids * sizeof(void **); |
6797 | ||
07e06b01 | 6798 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6799 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6800 | |
6d6bc0ad | 6801 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6802 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6803 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6804 | ptr += nr_cpu_ids * sizeof(void **); |
6805 | ||
07e06b01 | 6806 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6807 | ptr += nr_cpu_ids * sizeof(void **); |
6808 | ||
6d6bc0ad | 6809 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6810 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6811 | for_each_possible_cpu(i) { | |
e6252c3e | 6812 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6813 | ptr += cpumask_size(); |
6814 | } | |
6815 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6816 | } |
dd41f596 | 6817 | |
332ac17e DF |
6818 | init_rt_bandwidth(&def_rt_bandwidth, |
6819 | global_rt_period(), global_rt_runtime()); | |
6820 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6821 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6822 | |
57d885fe GH |
6823 | #ifdef CONFIG_SMP |
6824 | init_defrootdomain(); | |
6825 | #endif | |
6826 | ||
d0b27fa7 | 6827 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6828 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6829 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6830 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6831 | |
7c941438 | 6832 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6833 | list_add(&root_task_group.list, &task_groups); |
6834 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6835 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6836 | autogroup_init(&init_task); |
54c707e9 | 6837 | |
7c941438 | 6838 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6839 | |
0a945022 | 6840 | for_each_possible_cpu(i) { |
70b97a7f | 6841 | struct rq *rq; |
1da177e4 LT |
6842 | |
6843 | rq = cpu_rq(i); | |
05fa785c | 6844 | raw_spin_lock_init(&rq->lock); |
7897986b | 6845 | rq->nr_running = 0; |
dce48a84 TG |
6846 | rq->calc_load_active = 0; |
6847 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6848 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6849 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6850 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6851 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6852 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6853 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6854 | /* |
07e06b01 | 6855 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6856 | * |
6857 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6858 | * gets 100% of the cpu resources in the system. This overall | |
6859 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6860 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6861 | * based on each entity's (task or task-group's) weight |
6862 | * (se->load.weight). | |
6863 | * | |
07e06b01 | 6864 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6865 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6866 | * then A0's share of the cpu resource is: | |
6867 | * | |
0d905bca | 6868 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6869 | * |
07e06b01 YZ |
6870 | * We achieve this by letting root_task_group's tasks sit |
6871 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6872 | */ |
ab84d31e | 6873 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6874 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6875 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6876 | ||
6877 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6878 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6879 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6880 | #endif |
1da177e4 | 6881 | |
dd41f596 IM |
6882 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6883 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6884 | |
6885 | rq->last_load_update_tick = jiffies; | |
6886 | ||
1da177e4 | 6887 | #ifdef CONFIG_SMP |
41c7ce9a | 6888 | rq->sd = NULL; |
57d885fe | 6889 | rq->rd = NULL; |
1399fa78 | 6890 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6891 | rq->post_schedule = 0; |
1da177e4 | 6892 | rq->active_balance = 0; |
dd41f596 | 6893 | rq->next_balance = jiffies; |
1da177e4 | 6894 | rq->push_cpu = 0; |
0a2966b4 | 6895 | rq->cpu = i; |
1f11eb6a | 6896 | rq->online = 0; |
eae0c9df MG |
6897 | rq->idle_stamp = 0; |
6898 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6899 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6900 | |
6901 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6902 | ||
dc938520 | 6903 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6904 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6905 | rq->nohz_flags = 0; |
83cd4fe2 | 6906 | #endif |
265f22a9 FW |
6907 | #ifdef CONFIG_NO_HZ_FULL |
6908 | rq->last_sched_tick = 0; | |
6909 | #endif | |
1da177e4 | 6910 | #endif |
8f4d37ec | 6911 | init_rq_hrtick(rq); |
1da177e4 | 6912 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6913 | } |
6914 | ||
2dd73a4f | 6915 | set_load_weight(&init_task); |
b50f60ce | 6916 | |
e107be36 AK |
6917 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6918 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6919 | #endif | |
6920 | ||
1da177e4 LT |
6921 | /* |
6922 | * The boot idle thread does lazy MMU switching as well: | |
6923 | */ | |
6924 | atomic_inc(&init_mm.mm_count); | |
6925 | enter_lazy_tlb(&init_mm, current); | |
6926 | ||
6927 | /* | |
6928 | * Make us the idle thread. Technically, schedule() should not be | |
6929 | * called from this thread, however somewhere below it might be, | |
6930 | * but because we are the idle thread, we just pick up running again | |
6931 | * when this runqueue becomes "idle". | |
6932 | */ | |
6933 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6934 | |
6935 | calc_load_update = jiffies + LOAD_FREQ; | |
6936 | ||
dd41f596 IM |
6937 | /* |
6938 | * During early bootup we pretend to be a normal task: | |
6939 | */ | |
6940 | current->sched_class = &fair_sched_class; | |
6892b75e | 6941 | |
bf4d83f6 | 6942 | #ifdef CONFIG_SMP |
4cb98839 | 6943 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6944 | /* May be allocated at isolcpus cmdline parse time */ |
6945 | if (cpu_isolated_map == NULL) | |
6946 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6947 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6948 | #endif |
6949 | init_sched_fair_class(); | |
6a7b3dc3 | 6950 | |
6892b75e | 6951 | scheduler_running = 1; |
1da177e4 LT |
6952 | } |
6953 | ||
d902db1e | 6954 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6955 | static inline int preempt_count_equals(int preempt_offset) |
6956 | { | |
234da7bc | 6957 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6958 | |
4ba8216c | 6959 | return (nested == preempt_offset); |
e4aafea2 FW |
6960 | } |
6961 | ||
d894837f | 6962 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6963 | { |
1da177e4 LT |
6964 | static unsigned long prev_jiffy; /* ratelimiting */ |
6965 | ||
b3fbab05 | 6966 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
6967 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
6968 | !is_idle_task(current)) || | |
e4aafea2 | 6969 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
6970 | return; |
6971 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6972 | return; | |
6973 | prev_jiffy = jiffies; | |
6974 | ||
3df0fc5b PZ |
6975 | printk(KERN_ERR |
6976 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6977 | file, line); | |
6978 | printk(KERN_ERR | |
6979 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6980 | in_atomic(), irqs_disabled(), | |
6981 | current->pid, current->comm); | |
aef745fc IM |
6982 | |
6983 | debug_show_held_locks(current); | |
6984 | if (irqs_disabled()) | |
6985 | print_irqtrace_events(current); | |
8f47b187 TG |
6986 | #ifdef CONFIG_DEBUG_PREEMPT |
6987 | if (!preempt_count_equals(preempt_offset)) { | |
6988 | pr_err("Preemption disabled at:"); | |
6989 | print_ip_sym(current->preempt_disable_ip); | |
6990 | pr_cont("\n"); | |
6991 | } | |
6992 | #endif | |
aef745fc | 6993 | dump_stack(); |
1da177e4 LT |
6994 | } |
6995 | EXPORT_SYMBOL(__might_sleep); | |
6996 | #endif | |
6997 | ||
6998 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6999 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7000 | { | |
da7a735e | 7001 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7002 | struct sched_attr attr = { |
7003 | .sched_policy = SCHED_NORMAL, | |
7004 | }; | |
da7a735e | 7005 | int old_prio = p->prio; |
3a5e4dc1 | 7006 | int on_rq; |
3e51f33f | 7007 | |
fd2f4419 | 7008 | on_rq = p->on_rq; |
3a5e4dc1 | 7009 | if (on_rq) |
4ca9b72b | 7010 | dequeue_task(rq, p, 0); |
d50dde5a | 7011 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7012 | if (on_rq) { |
4ca9b72b | 7013 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7014 | resched_task(rq->curr); |
7015 | } | |
da7a735e PZ |
7016 | |
7017 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7018 | } |
7019 | ||
1da177e4 LT |
7020 | void normalize_rt_tasks(void) |
7021 | { | |
a0f98a1c | 7022 | struct task_struct *g, *p; |
1da177e4 | 7023 | unsigned long flags; |
70b97a7f | 7024 | struct rq *rq; |
1da177e4 | 7025 | |
4cf5d77a | 7026 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7027 | do_each_thread(g, p) { |
178be793 IM |
7028 | /* |
7029 | * Only normalize user tasks: | |
7030 | */ | |
7031 | if (!p->mm) | |
7032 | continue; | |
7033 | ||
6cfb0d5d | 7034 | p->se.exec_start = 0; |
6cfb0d5d | 7035 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7036 | p->se.statistics.wait_start = 0; |
7037 | p->se.statistics.sleep_start = 0; | |
7038 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7039 | #endif |
dd41f596 | 7040 | |
aab03e05 | 7041 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7042 | /* |
7043 | * Renice negative nice level userspace | |
7044 | * tasks back to 0: | |
7045 | */ | |
d0ea0268 | 7046 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7047 | set_user_nice(p, 0); |
1da177e4 | 7048 | continue; |
dd41f596 | 7049 | } |
1da177e4 | 7050 | |
1d615482 | 7051 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7052 | rq = __task_rq_lock(p); |
1da177e4 | 7053 | |
178be793 | 7054 | normalize_task(rq, p); |
3a5e4dc1 | 7055 | |
b29739f9 | 7056 | __task_rq_unlock(rq); |
1d615482 | 7057 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7058 | } while_each_thread(g, p); |
7059 | ||
4cf5d77a | 7060 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7061 | } |
7062 | ||
7063 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7064 | |
67fc4e0c | 7065 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7066 | /* |
67fc4e0c | 7067 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7068 | * |
7069 | * They can only be called when the whole system has been | |
7070 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7071 | * activity can take place. Using them for anything else would | |
7072 | * be a serious bug, and as a result, they aren't even visible | |
7073 | * under any other configuration. | |
7074 | */ | |
7075 | ||
7076 | /** | |
7077 | * curr_task - return the current task for a given cpu. | |
7078 | * @cpu: the processor in question. | |
7079 | * | |
7080 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7081 | * |
7082 | * Return: The current task for @cpu. | |
1df5c10a | 7083 | */ |
36c8b586 | 7084 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7085 | { |
7086 | return cpu_curr(cpu); | |
7087 | } | |
7088 | ||
67fc4e0c JW |
7089 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7090 | ||
7091 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7092 | /** |
7093 | * set_curr_task - set the current task for a given cpu. | |
7094 | * @cpu: the processor in question. | |
7095 | * @p: the task pointer to set. | |
7096 | * | |
7097 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7098 | * are serviced on a separate stack. It allows the architecture to switch the |
7099 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7100 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7101 | * and caller must save the original value of the current task (see | |
7102 | * curr_task() above) and restore that value before reenabling interrupts and | |
7103 | * re-starting the system. | |
7104 | * | |
7105 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7106 | */ | |
36c8b586 | 7107 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7108 | { |
7109 | cpu_curr(cpu) = p; | |
7110 | } | |
7111 | ||
7112 | #endif | |
29f59db3 | 7113 | |
7c941438 | 7114 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7115 | /* task_group_lock serializes the addition/removal of task groups */ |
7116 | static DEFINE_SPINLOCK(task_group_lock); | |
7117 | ||
bccbe08a PZ |
7118 | static void free_sched_group(struct task_group *tg) |
7119 | { | |
7120 | free_fair_sched_group(tg); | |
7121 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7122 | autogroup_free(tg); |
bccbe08a PZ |
7123 | kfree(tg); |
7124 | } | |
7125 | ||
7126 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7127 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7128 | { |
7129 | struct task_group *tg; | |
bccbe08a PZ |
7130 | |
7131 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7132 | if (!tg) | |
7133 | return ERR_PTR(-ENOMEM); | |
7134 | ||
ec7dc8ac | 7135 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7136 | goto err; |
7137 | ||
ec7dc8ac | 7138 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7139 | goto err; |
7140 | ||
ace783b9 LZ |
7141 | return tg; |
7142 | ||
7143 | err: | |
7144 | free_sched_group(tg); | |
7145 | return ERR_PTR(-ENOMEM); | |
7146 | } | |
7147 | ||
7148 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7149 | { | |
7150 | unsigned long flags; | |
7151 | ||
8ed36996 | 7152 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7153 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7154 | |
7155 | WARN_ON(!parent); /* root should already exist */ | |
7156 | ||
7157 | tg->parent = parent; | |
f473aa5e | 7158 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7159 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7160 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7161 | } |
7162 | ||
9b5b7751 | 7163 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7164 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7165 | { |
29f59db3 | 7166 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7167 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7168 | } |
7169 | ||
9b5b7751 | 7170 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7171 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7172 | { |
7173 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7174 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7175 | } | |
7176 | ||
7177 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7178 | { |
8ed36996 | 7179 | unsigned long flags; |
9b5b7751 | 7180 | int i; |
29f59db3 | 7181 | |
3d4b47b4 PZ |
7182 | /* end participation in shares distribution */ |
7183 | for_each_possible_cpu(i) | |
bccbe08a | 7184 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7185 | |
7186 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7187 | list_del_rcu(&tg->list); |
f473aa5e | 7188 | list_del_rcu(&tg->siblings); |
8ed36996 | 7189 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7190 | } |
7191 | ||
9b5b7751 | 7192 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7193 | * The caller of this function should have put the task in its new group |
7194 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7195 | * reflect its new group. | |
9b5b7751 SV |
7196 | */ |
7197 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7198 | { |
8323f26c | 7199 | struct task_group *tg; |
29f59db3 SV |
7200 | int on_rq, running; |
7201 | unsigned long flags; | |
7202 | struct rq *rq; | |
7203 | ||
7204 | rq = task_rq_lock(tsk, &flags); | |
7205 | ||
051a1d1a | 7206 | running = task_current(rq, tsk); |
fd2f4419 | 7207 | on_rq = tsk->on_rq; |
29f59db3 | 7208 | |
0e1f3483 | 7209 | if (on_rq) |
29f59db3 | 7210 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7211 | if (unlikely(running)) |
7212 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7213 | |
8af01f56 | 7214 | tg = container_of(task_css_check(tsk, cpu_cgroup_subsys_id, |
8323f26c PZ |
7215 | lockdep_is_held(&tsk->sighand->siglock)), |
7216 | struct task_group, css); | |
7217 | tg = autogroup_task_group(tsk, tg); | |
7218 | tsk->sched_task_group = tg; | |
7219 | ||
810b3817 | 7220 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7221 | if (tsk->sched_class->task_move_group) |
7222 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7223 | else | |
810b3817 | 7224 | #endif |
b2b5ce02 | 7225 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7226 | |
0e1f3483 HS |
7227 | if (unlikely(running)) |
7228 | tsk->sched_class->set_curr_task(rq); | |
7229 | if (on_rq) | |
371fd7e7 | 7230 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7231 | |
0122ec5b | 7232 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7233 | } |
7c941438 | 7234 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7235 | |
a790de99 PT |
7236 | #ifdef CONFIG_RT_GROUP_SCHED |
7237 | /* | |
7238 | * Ensure that the real time constraints are schedulable. | |
7239 | */ | |
7240 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7241 | |
9a7e0b18 PZ |
7242 | /* Must be called with tasklist_lock held */ |
7243 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7244 | { |
9a7e0b18 | 7245 | struct task_struct *g, *p; |
b40b2e8e | 7246 | |
9a7e0b18 | 7247 | do_each_thread(g, p) { |
029632fb | 7248 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7249 | return 1; |
7250 | } while_each_thread(g, p); | |
b40b2e8e | 7251 | |
9a7e0b18 PZ |
7252 | return 0; |
7253 | } | |
b40b2e8e | 7254 | |
9a7e0b18 PZ |
7255 | struct rt_schedulable_data { |
7256 | struct task_group *tg; | |
7257 | u64 rt_period; | |
7258 | u64 rt_runtime; | |
7259 | }; | |
b40b2e8e | 7260 | |
a790de99 | 7261 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7262 | { |
7263 | struct rt_schedulable_data *d = data; | |
7264 | struct task_group *child; | |
7265 | unsigned long total, sum = 0; | |
7266 | u64 period, runtime; | |
b40b2e8e | 7267 | |
9a7e0b18 PZ |
7268 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7269 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7270 | |
9a7e0b18 PZ |
7271 | if (tg == d->tg) { |
7272 | period = d->rt_period; | |
7273 | runtime = d->rt_runtime; | |
b40b2e8e | 7274 | } |
b40b2e8e | 7275 | |
4653f803 PZ |
7276 | /* |
7277 | * Cannot have more runtime than the period. | |
7278 | */ | |
7279 | if (runtime > period && runtime != RUNTIME_INF) | |
7280 | return -EINVAL; | |
6f505b16 | 7281 | |
4653f803 PZ |
7282 | /* |
7283 | * Ensure we don't starve existing RT tasks. | |
7284 | */ | |
9a7e0b18 PZ |
7285 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7286 | return -EBUSY; | |
6f505b16 | 7287 | |
9a7e0b18 | 7288 | total = to_ratio(period, runtime); |
6f505b16 | 7289 | |
4653f803 PZ |
7290 | /* |
7291 | * Nobody can have more than the global setting allows. | |
7292 | */ | |
7293 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7294 | return -EINVAL; | |
6f505b16 | 7295 | |
4653f803 PZ |
7296 | /* |
7297 | * The sum of our children's runtime should not exceed our own. | |
7298 | */ | |
9a7e0b18 PZ |
7299 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7300 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7301 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7302 | |
9a7e0b18 PZ |
7303 | if (child == d->tg) { |
7304 | period = d->rt_period; | |
7305 | runtime = d->rt_runtime; | |
7306 | } | |
6f505b16 | 7307 | |
9a7e0b18 | 7308 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7309 | } |
6f505b16 | 7310 | |
9a7e0b18 PZ |
7311 | if (sum > total) |
7312 | return -EINVAL; | |
7313 | ||
7314 | return 0; | |
6f505b16 PZ |
7315 | } |
7316 | ||
9a7e0b18 | 7317 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7318 | { |
8277434e PT |
7319 | int ret; |
7320 | ||
9a7e0b18 PZ |
7321 | struct rt_schedulable_data data = { |
7322 | .tg = tg, | |
7323 | .rt_period = period, | |
7324 | .rt_runtime = runtime, | |
7325 | }; | |
7326 | ||
8277434e PT |
7327 | rcu_read_lock(); |
7328 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7329 | rcu_read_unlock(); | |
7330 | ||
7331 | return ret; | |
521f1a24 DG |
7332 | } |
7333 | ||
ab84d31e | 7334 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7335 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7336 | { |
ac086bc2 | 7337 | int i, err = 0; |
9f0c1e56 | 7338 | |
9f0c1e56 | 7339 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7340 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7341 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7342 | if (err) | |
9f0c1e56 | 7343 | goto unlock; |
ac086bc2 | 7344 | |
0986b11b | 7345 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7346 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7347 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7348 | |
7349 | for_each_possible_cpu(i) { | |
7350 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7351 | ||
0986b11b | 7352 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7353 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7354 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7355 | } |
0986b11b | 7356 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7357 | unlock: |
521f1a24 | 7358 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7359 | mutex_unlock(&rt_constraints_mutex); |
7360 | ||
7361 | return err; | |
6f505b16 PZ |
7362 | } |
7363 | ||
25cc7da7 | 7364 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7365 | { |
7366 | u64 rt_runtime, rt_period; | |
7367 | ||
7368 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7369 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7370 | if (rt_runtime_us < 0) | |
7371 | rt_runtime = RUNTIME_INF; | |
7372 | ||
ab84d31e | 7373 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7374 | } |
7375 | ||
25cc7da7 | 7376 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7377 | { |
7378 | u64 rt_runtime_us; | |
7379 | ||
d0b27fa7 | 7380 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7381 | return -1; |
7382 | ||
d0b27fa7 | 7383 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7384 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7385 | return rt_runtime_us; | |
7386 | } | |
d0b27fa7 | 7387 | |
25cc7da7 | 7388 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7389 | { |
7390 | u64 rt_runtime, rt_period; | |
7391 | ||
7392 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7393 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7394 | ||
619b0488 R |
7395 | if (rt_period == 0) |
7396 | return -EINVAL; | |
7397 | ||
ab84d31e | 7398 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7399 | } |
7400 | ||
25cc7da7 | 7401 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7402 | { |
7403 | u64 rt_period_us; | |
7404 | ||
7405 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7406 | do_div(rt_period_us, NSEC_PER_USEC); | |
7407 | return rt_period_us; | |
7408 | } | |
332ac17e | 7409 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7410 | |
332ac17e | 7411 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7412 | static int sched_rt_global_constraints(void) |
7413 | { | |
7414 | int ret = 0; | |
7415 | ||
7416 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7417 | read_lock(&tasklist_lock); |
4653f803 | 7418 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7419 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7420 | mutex_unlock(&rt_constraints_mutex); |
7421 | ||
7422 | return ret; | |
7423 | } | |
54e99124 | 7424 | |
25cc7da7 | 7425 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7426 | { |
7427 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7428 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7429 | return 0; | |
7430 | ||
7431 | return 1; | |
7432 | } | |
7433 | ||
6d6bc0ad | 7434 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7435 | static int sched_rt_global_constraints(void) |
7436 | { | |
ac086bc2 | 7437 | unsigned long flags; |
332ac17e | 7438 | int i, ret = 0; |
ec5d4989 | 7439 | |
0986b11b | 7440 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7441 | for_each_possible_cpu(i) { |
7442 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7443 | ||
0986b11b | 7444 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7445 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7446 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7447 | } |
0986b11b | 7448 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7449 | |
332ac17e | 7450 | return ret; |
d0b27fa7 | 7451 | } |
6d6bc0ad | 7452 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7453 | |
332ac17e DF |
7454 | static int sched_dl_global_constraints(void) |
7455 | { | |
1724813d PZ |
7456 | u64 runtime = global_rt_runtime(); |
7457 | u64 period = global_rt_period(); | |
332ac17e | 7458 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7459 | int cpu, ret = 0; |
332ac17e DF |
7460 | |
7461 | /* | |
7462 | * Here we want to check the bandwidth not being set to some | |
7463 | * value smaller than the currently allocated bandwidth in | |
7464 | * any of the root_domains. | |
7465 | * | |
7466 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7467 | * cycling on root_domains... Discussion on different/better | |
7468 | * solutions is welcome! | |
7469 | */ | |
1724813d PZ |
7470 | for_each_possible_cpu(cpu) { |
7471 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e DF |
7472 | |
7473 | raw_spin_lock(&dl_b->lock); | |
1724813d PZ |
7474 | if (new_bw < dl_b->total_bw) |
7475 | ret = -EBUSY; | |
332ac17e | 7476 | raw_spin_unlock(&dl_b->lock); |
1724813d PZ |
7477 | |
7478 | if (ret) | |
7479 | break; | |
332ac17e DF |
7480 | } |
7481 | ||
1724813d | 7482 | return ret; |
332ac17e DF |
7483 | } |
7484 | ||
1724813d | 7485 | static void sched_dl_do_global(void) |
ce0dbbbb | 7486 | { |
1724813d PZ |
7487 | u64 new_bw = -1; |
7488 | int cpu; | |
ce0dbbbb | 7489 | |
1724813d PZ |
7490 | def_dl_bandwidth.dl_period = global_rt_period(); |
7491 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7492 | ||
7493 | if (global_rt_runtime() != RUNTIME_INF) | |
7494 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7495 | ||
7496 | /* | |
7497 | * FIXME: As above... | |
7498 | */ | |
7499 | for_each_possible_cpu(cpu) { | |
7500 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7501 | ||
7502 | raw_spin_lock(&dl_b->lock); | |
7503 | dl_b->bw = new_bw; | |
7504 | raw_spin_unlock(&dl_b->lock); | |
ce0dbbbb | 7505 | } |
1724813d PZ |
7506 | } |
7507 | ||
7508 | static int sched_rt_global_validate(void) | |
7509 | { | |
7510 | if (sysctl_sched_rt_period <= 0) | |
7511 | return -EINVAL; | |
7512 | ||
7513 | if (sysctl_sched_rt_runtime > sysctl_sched_rt_period) | |
7514 | return -EINVAL; | |
7515 | ||
7516 | return 0; | |
7517 | } | |
7518 | ||
7519 | static void sched_rt_do_global(void) | |
7520 | { | |
7521 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7522 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7523 | } |
7524 | ||
d0b27fa7 | 7525 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7526 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7527 | loff_t *ppos) |
7528 | { | |
d0b27fa7 PZ |
7529 | int old_period, old_runtime; |
7530 | static DEFINE_MUTEX(mutex); | |
1724813d | 7531 | int ret; |
d0b27fa7 PZ |
7532 | |
7533 | mutex_lock(&mutex); | |
7534 | old_period = sysctl_sched_rt_period; | |
7535 | old_runtime = sysctl_sched_rt_runtime; | |
7536 | ||
8d65af78 | 7537 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7538 | |
7539 | if (!ret && write) { | |
1724813d PZ |
7540 | ret = sched_rt_global_validate(); |
7541 | if (ret) | |
7542 | goto undo; | |
7543 | ||
d0b27fa7 | 7544 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7545 | if (ret) |
7546 | goto undo; | |
7547 | ||
7548 | ret = sched_dl_global_constraints(); | |
7549 | if (ret) | |
7550 | goto undo; | |
7551 | ||
7552 | sched_rt_do_global(); | |
7553 | sched_dl_do_global(); | |
7554 | } | |
7555 | if (0) { | |
7556 | undo: | |
7557 | sysctl_sched_rt_period = old_period; | |
7558 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7559 | } |
7560 | mutex_unlock(&mutex); | |
7561 | ||
7562 | return ret; | |
7563 | } | |
68318b8e | 7564 | |
1724813d | 7565 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7566 | void __user *buffer, size_t *lenp, |
7567 | loff_t *ppos) | |
7568 | { | |
7569 | int ret; | |
332ac17e | 7570 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7571 | |
7572 | mutex_lock(&mutex); | |
332ac17e | 7573 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7574 | /* make sure that internally we keep jiffies */ |
7575 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7576 | if (!ret && write) { |
1724813d PZ |
7577 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7578 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7579 | } |
7580 | mutex_unlock(&mutex); | |
332ac17e DF |
7581 | return ret; |
7582 | } | |
7583 | ||
052f1dc7 | 7584 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7585 | |
a7c6d554 | 7586 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7587 | { |
a7c6d554 | 7588 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7589 | } |
7590 | ||
eb95419b TH |
7591 | static struct cgroup_subsys_state * |
7592 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7593 | { |
eb95419b TH |
7594 | struct task_group *parent = css_tg(parent_css); |
7595 | struct task_group *tg; | |
68318b8e | 7596 | |
eb95419b | 7597 | if (!parent) { |
68318b8e | 7598 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7599 | return &root_task_group.css; |
68318b8e SV |
7600 | } |
7601 | ||
ec7dc8ac | 7602 | tg = sched_create_group(parent); |
68318b8e SV |
7603 | if (IS_ERR(tg)) |
7604 | return ERR_PTR(-ENOMEM); | |
7605 | ||
68318b8e SV |
7606 | return &tg->css; |
7607 | } | |
7608 | ||
eb95419b | 7609 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7610 | { |
eb95419b TH |
7611 | struct task_group *tg = css_tg(css); |
7612 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7613 | |
63876986 TH |
7614 | if (parent) |
7615 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7616 | return 0; |
7617 | } | |
7618 | ||
eb95419b | 7619 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7620 | { |
eb95419b | 7621 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7622 | |
7623 | sched_destroy_group(tg); | |
7624 | } | |
7625 | ||
eb95419b | 7626 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7627 | { |
eb95419b | 7628 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7629 | |
7630 | sched_offline_group(tg); | |
7631 | } | |
7632 | ||
eb95419b | 7633 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7634 | struct cgroup_taskset *tset) |
68318b8e | 7635 | { |
bb9d97b6 TH |
7636 | struct task_struct *task; |
7637 | ||
d99c8727 | 7638 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7639 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7640 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7641 | return -EINVAL; |
b68aa230 | 7642 | #else |
bb9d97b6 TH |
7643 | /* We don't support RT-tasks being in separate groups */ |
7644 | if (task->sched_class != &fair_sched_class) | |
7645 | return -EINVAL; | |
b68aa230 | 7646 | #endif |
bb9d97b6 | 7647 | } |
be367d09 BB |
7648 | return 0; |
7649 | } | |
68318b8e | 7650 | |
eb95419b | 7651 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7652 | struct cgroup_taskset *tset) |
68318b8e | 7653 | { |
bb9d97b6 TH |
7654 | struct task_struct *task; |
7655 | ||
d99c8727 | 7656 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7657 | sched_move_task(task); |
68318b8e SV |
7658 | } |
7659 | ||
eb95419b TH |
7660 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7661 | struct cgroup_subsys_state *old_css, | |
7662 | struct task_struct *task) | |
068c5cc5 PZ |
7663 | { |
7664 | /* | |
7665 | * cgroup_exit() is called in the copy_process() failure path. | |
7666 | * Ignore this case since the task hasn't ran yet, this avoids | |
7667 | * trying to poke a half freed task state from generic code. | |
7668 | */ | |
7669 | if (!(task->flags & PF_EXITING)) | |
7670 | return; | |
7671 | ||
7672 | sched_move_task(task); | |
7673 | } | |
7674 | ||
052f1dc7 | 7675 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7676 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7677 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7678 | { |
182446d0 | 7679 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7680 | } |
7681 | ||
182446d0 TH |
7682 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7683 | struct cftype *cft) | |
68318b8e | 7684 | { |
182446d0 | 7685 | struct task_group *tg = css_tg(css); |
68318b8e | 7686 | |
c8b28116 | 7687 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7688 | } |
ab84d31e PT |
7689 | |
7690 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7691 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7692 | ||
ab84d31e PT |
7693 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7694 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7695 | ||
a790de99 PT |
7696 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7697 | ||
ab84d31e PT |
7698 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7699 | { | |
56f570e5 | 7700 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7701 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7702 | |
7703 | if (tg == &root_task_group) | |
7704 | return -EINVAL; | |
7705 | ||
7706 | /* | |
7707 | * Ensure we have at some amount of bandwidth every period. This is | |
7708 | * to prevent reaching a state of large arrears when throttled via | |
7709 | * entity_tick() resulting in prolonged exit starvation. | |
7710 | */ | |
7711 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7712 | return -EINVAL; | |
7713 | ||
7714 | /* | |
7715 | * Likewise, bound things on the otherside by preventing insane quota | |
7716 | * periods. This also allows us to normalize in computing quota | |
7717 | * feasibility. | |
7718 | */ | |
7719 | if (period > max_cfs_quota_period) | |
7720 | return -EINVAL; | |
7721 | ||
a790de99 PT |
7722 | mutex_lock(&cfs_constraints_mutex); |
7723 | ret = __cfs_schedulable(tg, period, quota); | |
7724 | if (ret) | |
7725 | goto out_unlock; | |
7726 | ||
58088ad0 | 7727 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7728 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7729 | /* |
7730 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7731 | * before making related changes, and on->off must occur afterwards | |
7732 | */ | |
7733 | if (runtime_enabled && !runtime_was_enabled) | |
7734 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7735 | raw_spin_lock_irq(&cfs_b->lock); |
7736 | cfs_b->period = ns_to_ktime(period); | |
7737 | cfs_b->quota = quota; | |
58088ad0 | 7738 | |
a9cf55b2 | 7739 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7740 | /* restart the period timer (if active) to handle new period expiry */ |
7741 | if (runtime_enabled && cfs_b->timer_active) { | |
7742 | /* force a reprogram */ | |
7743 | cfs_b->timer_active = 0; | |
7744 | __start_cfs_bandwidth(cfs_b); | |
7745 | } | |
ab84d31e PT |
7746 | raw_spin_unlock_irq(&cfs_b->lock); |
7747 | ||
7748 | for_each_possible_cpu(i) { | |
7749 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7750 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7751 | |
7752 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7753 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7754 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7755 | |
029632fb | 7756 | if (cfs_rq->throttled) |
671fd9da | 7757 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7758 | raw_spin_unlock_irq(&rq->lock); |
7759 | } | |
1ee14e6c BS |
7760 | if (runtime_was_enabled && !runtime_enabled) |
7761 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7762 | out_unlock: |
7763 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7764 | |
a790de99 | 7765 | return ret; |
ab84d31e PT |
7766 | } |
7767 | ||
7768 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7769 | { | |
7770 | u64 quota, period; | |
7771 | ||
029632fb | 7772 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7773 | if (cfs_quota_us < 0) |
7774 | quota = RUNTIME_INF; | |
7775 | else | |
7776 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7777 | ||
7778 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7779 | } | |
7780 | ||
7781 | long tg_get_cfs_quota(struct task_group *tg) | |
7782 | { | |
7783 | u64 quota_us; | |
7784 | ||
029632fb | 7785 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7786 | return -1; |
7787 | ||
029632fb | 7788 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7789 | do_div(quota_us, NSEC_PER_USEC); |
7790 | ||
7791 | return quota_us; | |
7792 | } | |
7793 | ||
7794 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7795 | { | |
7796 | u64 quota, period; | |
7797 | ||
7798 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7799 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7800 | |
ab84d31e PT |
7801 | return tg_set_cfs_bandwidth(tg, period, quota); |
7802 | } | |
7803 | ||
7804 | long tg_get_cfs_period(struct task_group *tg) | |
7805 | { | |
7806 | u64 cfs_period_us; | |
7807 | ||
029632fb | 7808 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7809 | do_div(cfs_period_us, NSEC_PER_USEC); |
7810 | ||
7811 | return cfs_period_us; | |
7812 | } | |
7813 | ||
182446d0 TH |
7814 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7815 | struct cftype *cft) | |
ab84d31e | 7816 | { |
182446d0 | 7817 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7818 | } |
7819 | ||
182446d0 TH |
7820 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7821 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7822 | { |
182446d0 | 7823 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7824 | } |
7825 | ||
182446d0 TH |
7826 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7827 | struct cftype *cft) | |
ab84d31e | 7828 | { |
182446d0 | 7829 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7830 | } |
7831 | ||
182446d0 TH |
7832 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7833 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7834 | { |
182446d0 | 7835 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7836 | } |
7837 | ||
a790de99 PT |
7838 | struct cfs_schedulable_data { |
7839 | struct task_group *tg; | |
7840 | u64 period, quota; | |
7841 | }; | |
7842 | ||
7843 | /* | |
7844 | * normalize group quota/period to be quota/max_period | |
7845 | * note: units are usecs | |
7846 | */ | |
7847 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7848 | struct cfs_schedulable_data *d) | |
7849 | { | |
7850 | u64 quota, period; | |
7851 | ||
7852 | if (tg == d->tg) { | |
7853 | period = d->period; | |
7854 | quota = d->quota; | |
7855 | } else { | |
7856 | period = tg_get_cfs_period(tg); | |
7857 | quota = tg_get_cfs_quota(tg); | |
7858 | } | |
7859 | ||
7860 | /* note: these should typically be equivalent */ | |
7861 | if (quota == RUNTIME_INF || quota == -1) | |
7862 | return RUNTIME_INF; | |
7863 | ||
7864 | return to_ratio(period, quota); | |
7865 | } | |
7866 | ||
7867 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7868 | { | |
7869 | struct cfs_schedulable_data *d = data; | |
029632fb | 7870 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7871 | s64 quota = 0, parent_quota = -1; |
7872 | ||
7873 | if (!tg->parent) { | |
7874 | quota = RUNTIME_INF; | |
7875 | } else { | |
029632fb | 7876 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7877 | |
7878 | quota = normalize_cfs_quota(tg, d); | |
7879 | parent_quota = parent_b->hierarchal_quota; | |
7880 | ||
7881 | /* | |
7882 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7883 | * limit is set | |
7884 | */ | |
7885 | if (quota == RUNTIME_INF) | |
7886 | quota = parent_quota; | |
7887 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7888 | return -EINVAL; | |
7889 | } | |
7890 | cfs_b->hierarchal_quota = quota; | |
7891 | ||
7892 | return 0; | |
7893 | } | |
7894 | ||
7895 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7896 | { | |
8277434e | 7897 | int ret; |
a790de99 PT |
7898 | struct cfs_schedulable_data data = { |
7899 | .tg = tg, | |
7900 | .period = period, | |
7901 | .quota = quota, | |
7902 | }; | |
7903 | ||
7904 | if (quota != RUNTIME_INF) { | |
7905 | do_div(data.period, NSEC_PER_USEC); | |
7906 | do_div(data.quota, NSEC_PER_USEC); | |
7907 | } | |
7908 | ||
8277434e PT |
7909 | rcu_read_lock(); |
7910 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7911 | rcu_read_unlock(); | |
7912 | ||
7913 | return ret; | |
a790de99 | 7914 | } |
e8da1b18 | 7915 | |
2da8ca82 | 7916 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7917 | { |
2da8ca82 | 7918 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7919 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7920 | |
44ffc75b TH |
7921 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7922 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7923 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7924 | |
7925 | return 0; | |
7926 | } | |
ab84d31e | 7927 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7928 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7929 | |
052f1dc7 | 7930 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7931 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7932 | struct cftype *cft, s64 val) | |
6f505b16 | 7933 | { |
182446d0 | 7934 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7935 | } |
7936 | ||
182446d0 TH |
7937 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7938 | struct cftype *cft) | |
6f505b16 | 7939 | { |
182446d0 | 7940 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7941 | } |
d0b27fa7 | 7942 | |
182446d0 TH |
7943 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7944 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7945 | { |
182446d0 | 7946 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7947 | } |
7948 | ||
182446d0 TH |
7949 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7950 | struct cftype *cft) | |
d0b27fa7 | 7951 | { |
182446d0 | 7952 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7953 | } |
6d6bc0ad | 7954 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7955 | |
fe5c7cc2 | 7956 | static struct cftype cpu_files[] = { |
052f1dc7 | 7957 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7958 | { |
7959 | .name = "shares", | |
f4c753b7 PM |
7960 | .read_u64 = cpu_shares_read_u64, |
7961 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7962 | }, |
052f1dc7 | 7963 | #endif |
ab84d31e PT |
7964 | #ifdef CONFIG_CFS_BANDWIDTH |
7965 | { | |
7966 | .name = "cfs_quota_us", | |
7967 | .read_s64 = cpu_cfs_quota_read_s64, | |
7968 | .write_s64 = cpu_cfs_quota_write_s64, | |
7969 | }, | |
7970 | { | |
7971 | .name = "cfs_period_us", | |
7972 | .read_u64 = cpu_cfs_period_read_u64, | |
7973 | .write_u64 = cpu_cfs_period_write_u64, | |
7974 | }, | |
e8da1b18 NR |
7975 | { |
7976 | .name = "stat", | |
2da8ca82 | 7977 | .seq_show = cpu_stats_show, |
e8da1b18 | 7978 | }, |
ab84d31e | 7979 | #endif |
052f1dc7 | 7980 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7981 | { |
9f0c1e56 | 7982 | .name = "rt_runtime_us", |
06ecb27c PM |
7983 | .read_s64 = cpu_rt_runtime_read, |
7984 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7985 | }, |
d0b27fa7 PZ |
7986 | { |
7987 | .name = "rt_period_us", | |
f4c753b7 PM |
7988 | .read_u64 = cpu_rt_period_read_uint, |
7989 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7990 | }, |
052f1dc7 | 7991 | #endif |
4baf6e33 | 7992 | { } /* terminate */ |
68318b8e SV |
7993 | }; |
7994 | ||
68318b8e | 7995 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae | 7996 | .name = "cpu", |
92fb9748 TH |
7997 | .css_alloc = cpu_cgroup_css_alloc, |
7998 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
7999 | .css_online = cpu_cgroup_css_online, |
8000 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8001 | .can_attach = cpu_cgroup_can_attach, |
8002 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8003 | .exit = cpu_cgroup_exit, |
38605cae | 8004 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 8005 | .base_cftypes = cpu_files, |
68318b8e SV |
8006 | .early_init = 1, |
8007 | }; | |
8008 | ||
052f1dc7 | 8009 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8010 | |
b637a328 PM |
8011 | void dump_cpu_task(int cpu) |
8012 | { | |
8013 | pr_info("Task dump for CPU %d:\n", cpu); | |
8014 | sched_show_task(cpu_curr(cpu)); | |
8015 | } |